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 = NULL;
 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_acquire_dquot(struct dquot *dquot);
 525 static int ext3_release_dquot(struct dquot *dquot);
 526 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 527 static int ext3_write_info(struct super_block *sb, int type);
 528 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 529 static int ext3_quota_on_mount(struct super_block *sb, int type);
 530 static int ext3_quota_off_mount(struct super_block *sb, int type);
 531
 532 static struct dquot_operations ext3_quota_operations = {
 533         .initialize     = ext3_dquot_initialize,
 534         .drop           = ext3_dquot_drop,
 535         .alloc_space    = dquot_alloc_space,
 536         .alloc_inode    = dquot_alloc_inode,
 537         .free_space     = dquot_free_space,
 538         .free_inode     = dquot_free_inode,
 539         .transfer       = dquot_transfer,
 540         .write_dquot    = ext3_write_dquot,
 541         .acquire_dquot  = ext3_acquire_dquot,
 542         .release_dquot  = ext3_release_dquot,
 543         .mark_dirty     = ext3_mark_dquot_dirty,
 544         .write_info     = ext3_write_info
 545 };
 546
 547 static struct quotactl_ops ext3_qctl_operations = {
 548         .quota_on       = ext3_quota_on,
 549         .quota_off      = vfs_quota_off,
 550         .quota_sync     = vfs_quota_sync,
 551         .get_info       = vfs_get_dqinfo,
 552         .set_info       = vfs_set_dqinfo,
 553         .get_dqblk      = vfs_get_dqblk,
 554         .set_dqblk      = vfs_set_dqblk
 555 };
 556 #endif
 557
 558 static struct super_operations ext3_sops = {
 559         .alloc_inode    = ext3_alloc_inode,
 560         .destroy_inode  = ext3_destroy_inode,
 561         .read_inode     = ext3_read_inode,
 562         .write_inode    = ext3_write_inode,
 563         .dirty_inode    = ext3_dirty_inode,
 564         .put_inode      = ext3_put_inode,
 565         .delete_inode   = ext3_delete_inode,
 566         .put_super      = ext3_put_super,
 567         .write_super    = ext3_write_super,
 568         .sync_fs        = ext3_sync_fs,
 569         .write_super_lockfs = ext3_write_super_lockfs,
 570         .unlockfs       = ext3_unlockfs,
 571         .statfs         = ext3_statfs,
 572         .remount_fs     = ext3_remount,
 573         .clear_inode    = ext3_clear_inode,
 574 };
 575
 576 struct dentry *ext3_get_parent(struct dentry *child);
 577 static struct export_operations ext3_export_ops = {
 578         .get_parent = ext3_get_parent,
 579 };
 580
 581 enum {
 582         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 583         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 584         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 585         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, Opt_noload,
 586         Opt_commit, Opt_journal_update, Opt_journal_inum,
 587         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 588         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 589         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 590         Opt_ignore, Opt_err,
 591 };
 592
 593 static match_table_t tokens = {
 594         {Opt_bsd_df, "bsddf"},
 595         {Opt_minix_df, "minixdf"},
 596         {Opt_grpid, "grpid"},
 597         {Opt_grpid, "bsdgroups"},
 598         {Opt_nogrpid, "nogrpid"},
 599         {Opt_nogrpid, "sysvgroups"},
 600         {Opt_resgid, "resgid=%u"},
 601         {Opt_resuid, "resuid=%u"},
 602         {Opt_sb, "sb=%u"},
 603         {Opt_err_cont, "errors=continue"},
 604         {Opt_err_panic, "errors=panic"},
 605         {Opt_err_ro, "errors=remount-ro"},
 606         {Opt_nouid32, "nouid32"},
 607         {Opt_nocheck, "nocheck"},
 608         {Opt_nocheck, "check=none"},
 609         {Opt_check, "check"},
 610         {Opt_debug, "debug"},
 611         {Opt_oldalloc, "oldalloc"},
 612         {Opt_orlov, "orlov"},
 613         {Opt_user_xattr, "user_xattr"},
 614         {Opt_nouser_xattr, "nouser_xattr"},
 615         {Opt_acl, "acl"},
 616         {Opt_noacl, "noacl"},
 617         {Opt_noload, "noload"},
 618         {Opt_commit, "commit=%u"},
 619         {Opt_journal_update, "journal=update"},
 620         {Opt_journal_inum, "journal=%u"},
 621         {Opt_abort, "abort"},
 622         {Opt_data_journal, "data=journal"},
 623         {Opt_data_ordered, "data=ordered"},
 624         {Opt_data_writeback, "data=writeback"},
 625         {Opt_offusrjquota, "usrjquota="},
 626         {Opt_usrjquota, "usrjquota=%s"},
 627         {Opt_offgrpjquota, "grpjquota="},
 628         {Opt_grpjquota, "grpjquota=%s"},
 629         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 630         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 631         {Opt_ignore, "grpquota"},
 632         {Opt_ignore, "noquota"},
 633         {Opt_ignore, "quota"},
 634         {Opt_ignore, "usrquota"},
 635         {Opt_err, NULL}
 636 };
 637
 638 static unsigned long get_sb_block(void **data)
 639 {
 640         unsigned long   sb_block;
 641         char            *options = (char *) *data;
 642
 643         if (!options || strncmp(options, "sb=", 3) != 0)
 644                 return 1;       /* Default location */
 645         options += 3;
 646         sb_block = simple_strtoul(options, &options, 0);
 647         if (*options && *options != ',') {
 648                 printk("EXT3-fs: Invalid sb specification: %s\n",
 649                        (char *) *data);
 650                 return 1;
 651         }
 652         if (*options == ',')
 653                 options++;
 654         *data = (void *) options;
 655         return sb_block;
 656 }
 657
 658 static int parse_options (char * options, struct super_block *sb,
 659                           unsigned long * inum, int is_remount)
 660 {
 661         struct ext3_sb_info *sbi = EXT3_SB(sb);
 662         char * p;
 663         substring_t args[MAX_OPT_ARGS];
 664         int data_opt = 0;
 665         int option;
 666 #ifdef CONFIG_QUOTA
 667         int qtype;
 668 #endif
 669
 670         if (!options)
 671                 return 1;
 672
 673         while ((p = strsep (&options, ",")) != NULL) {
 674                 int token;
 675                 if (!*p)
 676                         continue;
 677
 678                 token = match_token(p, tokens, args);
 679                 switch (token) {
 680                 case Opt_bsd_df:
 681                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 682                         break;
 683                 case Opt_minix_df:
 684                         set_opt (sbi->s_mount_opt, MINIX_DF);
 685                         break;
 686                 case Opt_grpid:
 687                         set_opt (sbi->s_mount_opt, GRPID);
 688                         break;
 689                 case Opt_nogrpid:
 690                         clear_opt (sbi->s_mount_opt, GRPID);
 691                         break;
 692                 case Opt_resuid:
 693                         if (match_int(&args[0], &option))
 694                                 return 0;
 695                         sbi->s_resuid = option;
 696                         break;
 697                 case Opt_resgid:
 698                         if (match_int(&args[0], &option))
 699                                 return 0;
 700                         sbi->s_resgid = option;
 701                         break;
 702                 case Opt_sb:
 703                         /* handled by get_sb_block() instead of here */
 704                         /* *sb_block = match_int(&args[0]); */
 705                         break;
 706                 case Opt_err_panic:
 707                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 708                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 709                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 710                         break;
 711                 case Opt_err_ro:
 712                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 713                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 714                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 715                         break;
 716                 case Opt_err_cont:
 717                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 718                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 719                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 720                         break;
 721                 case Opt_nouid32:
 722                         set_opt (sbi->s_mount_opt, NO_UID32);
 723                         break;
 724                 case Opt_check:
 725 #ifdef CONFIG_EXT3_CHECK
 726                         set_opt (sbi->s_mount_opt, CHECK);
 727 #else
 728                         printk(KERN_ERR
 729                                "EXT3 Check option not supported\n");
 730 #endif
 731                         break;
 732                 case Opt_nocheck:
 733                         clear_opt (sbi->s_mount_opt, CHECK);
 734                         break;
 735                 case Opt_debug:
 736                         set_opt (sbi->s_mount_opt, DEBUG);
 737                         break;
 738                 case Opt_oldalloc:
 739                         set_opt (sbi->s_mount_opt, OLDALLOC);
 740                         break;
 741                 case Opt_orlov:
 742                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 743                         break;
 744 #ifdef CONFIG_EXT3_FS_XATTR
 745                 case Opt_user_xattr:
 746                         set_opt (sbi->s_mount_opt, XATTR_USER);
 747                         break;
 748                 case Opt_nouser_xattr:
 749                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 750                         break;
 751 #else
 752                 case Opt_user_xattr:
 753                 case Opt_nouser_xattr:
 754                         printk("EXT3 (no)user_xattr options not supported\n");
 755                         break;
 756 #endif
 757 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 758                 case Opt_acl:
 759                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 760                         break;
 761                 case Opt_noacl:
 762                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 763                         break;
 764 #else
 765                 case Opt_acl:
 766                 case Opt_noacl:
 767                         printk("EXT3 (no)acl options not supported\n");
 768                         break;
 769 #endif
 770                 case Opt_journal_update:
 771                         /* @@@ FIXME */
 772                         /* Eventually we will want to be able to create
 773                            a journal file here.  For now, only allow the
 774                            user to specify an existing inode to be the
 775                            journal file. */
 776                         if (is_remount) {
 777                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 778                                        "journal on remount\n");
 779                                 return 0;
 780                         }
 781                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 782                         break;
 783                 case Opt_journal_inum:
 784                         if (is_remount) {
 785                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 786                                        "journal on remount\n");
 787                                 return 0;
 788                         }
 789                         if (match_int(&args[0], &option))
 790                                 return 0;
 791                         *inum = option;
 792                         break;
 793                 case Opt_noload:
 794                         set_opt (sbi->s_mount_opt, NOLOAD);
 795                         break;
 796                 case Opt_commit:
 797                         if (match_int(&args[0], &option))
 798                                 return 0;
 799                         if (option < 0)
 800                                 return 0;
 801                         if (option == 0)
 802                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 803                         sbi->s_commit_interval = HZ * option;
 804                         break;
 805                 case Opt_data_journal:
 806                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 807                         goto datacheck;
 808                 case Opt_data_ordered:
 809                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 810                         goto datacheck;
 811                 case Opt_data_writeback:
 812                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 813                 datacheck:
 814                         if (is_remount) {
 815                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 816                                                 != data_opt) {
 817                                         printk(KERN_ERR
 818                                                 "EXT3-fs: cannot change data "
 819                                                 "mode on remount\n");
 820                                         return 0;
 821                                 }
 822                         } else {
 823                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 824                                 sbi->s_mount_opt |= data_opt;
 825                         }
 826                         break;
 827 #ifdef CONFIG_QUOTA
 828                 case Opt_usrjquota:
 829                         qtype = USRQUOTA;
 830                         goto set_qf_name;
 831                 case Opt_grpjquota:
 832                         qtype = GRPQUOTA;
 833 set_qf_name:
 834                         if (sb_any_quota_enabled(sb)) {
 835                                 printk(KERN_ERR
 836                                         "EXT3-fs: Cannot change journalled "
 837                                         "quota options when quota turned on.\n");
 838                                 return 0;
 839                         }
 840                         if (sbi->s_qf_names[qtype]) {
 841                                 printk(KERN_ERR
 842                                         "EXT3-fs: %s quota file already "
 843                                         "specified.\n", QTYPE2NAME(qtype));
 844                                 return 0;
 845                         }
 846                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 847                         if (!sbi->s_qf_names[qtype]) {
 848                                 printk(KERN_ERR
 849                                         "EXT3-fs: not enough memory for "
 850                                         "storing quotafile name.\n");
 851                                 return 0;
 852                         }
 853                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 854                                 printk(KERN_ERR
 855                                         "EXT3-fs: quotafile must be on "
 856                                         "filesystem root.\n");
 857                                 kfree(sbi->s_qf_names[qtype]);
 858                                 sbi->s_qf_names[qtype] = NULL;
 859                                 return 0;
 860                         }
 861                         break;
 862                 case Opt_offusrjquota:
 863                         qtype = USRQUOTA;
 864                         goto clear_qf_name;
 865                 case Opt_offgrpjquota:
 866                         qtype = GRPQUOTA;
 867 clear_qf_name:
 868                         if (sb_any_quota_enabled(sb)) {
 869                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 870                                         "journalled quota options when "
 871                                         "quota turned on.\n");
 872                                 return 0;
 873                         }
 874                         if (sbi->s_qf_names[qtype]) {
 875                                 kfree(sbi->s_qf_names[qtype]);
 876                                 sbi->s_qf_names[qtype] = NULL;
 877                         }
 878                         break;
 879                 case Opt_jqfmt_vfsold:
 880                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 881                         break;
 882                 case Opt_jqfmt_vfsv0:
 883                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 884                         break;
 885 #else
 886                 case Opt_usrjquota:
 887                 case Opt_grpjquota:
 888                 case Opt_offusrjquota:
 889                 case Opt_offgrpjquota:
 890                 case Opt_jqfmt_vfsold:
 891                 case Opt_jqfmt_vfsv0:
 892                         printk(KERN_ERR
 893                                 "EXT3-fs: journalled quota options not "
 894                                 "supported.\n");
 895                         break;
 896 #endif
 897                 case Opt_abort:
 898                         set_opt(sbi->s_mount_opt, ABORT);
 899                         break;
 900                 case Opt_ignore:
 901                         break;
 902                 default:
 903                         printk (KERN_ERR
 904                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 905                                 "or missing value\n", p);
 906                         return 0;
 907                 }
 908         }
 909 #ifdef CONFIG_QUOTA
 910         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 911             sbi->s_qf_names[GRPQUOTA])) {
 912                 printk(KERN_ERR
 913                         "EXT3-fs: journalled quota format not specified.\n");
 914                 return 0;
 915         }
 916 #endif
 917
 918         return 1;
 919 }
 920
 921 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 922                             int read_only)
 923 {
 924         struct ext3_sb_info *sbi = EXT3_SB(sb);
 925         int res = 0;
 926
 927         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 928                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 929                         "forcing read-only mode\n");
 930                 res = MS_RDONLY;
 931         }
 932         if (read_only)
 933                 return res;
 934         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 935                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 936                         "running e2fsck is recommended\n");
 937         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 938                 printk (KERN_WARNING
 939                         "EXT3-fs warning: mounting fs with errors, "
 940                         "running e2fsck is recommended\n");
 941         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 942                  le16_to_cpu(es->s_mnt_count) >=
 943                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 944                 printk (KERN_WARNING
 945                         "EXT3-fs warning: maximal mount count reached, "
 946                         "running e2fsck is recommended\n");
 947         else if (le32_to_cpu(es->s_checkinterval) &&
 948                 (le32_to_cpu(es->s_lastcheck) +
 949                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 950                 printk (KERN_WARNING
 951                         "EXT3-fs warning: checktime reached, "
 952                         "running e2fsck is recommended\n");
 953 #if 0
 954                 /* @@@ We _will_ want to clear the valid bit if we find
 955                    inconsistencies, to force a fsck at reboot.  But for
 956                    a plain journaled filesystem we can keep it set as
 957                    valid forever! :) */
 958         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 959 #endif
 960         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 961                 es->s_max_mnt_count =
 962                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 963         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 964         es->s_mtime = cpu_to_le32(get_seconds());
 965         ext3_update_dynamic_rev(sb);
 966         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 967
 968         ext3_commit_super(sb, es, 1);
 969         if (test_opt(sb, DEBUG))
 970                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 971                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 972                         sb->s_blocksize,
 973                         sbi->s_groups_count,
 974                         EXT3_BLOCKS_PER_GROUP(sb),
 975                         EXT3_INODES_PER_GROUP(sb),
 976                         sbi->s_mount_opt);
 977
 978         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 979         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 980                 char b[BDEVNAME_SIZE];
 981
 982                 printk("external journal on %s\n",
 983                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
 984         } else {
 985                 printk("internal journal\n");
 986         }
 987 #ifdef CONFIG_EXT3_CHECK
 988         if (test_opt (sb, CHECK)) {
 989                 ext3_check_blocks_bitmap (sb);
 990                 ext3_check_inodes_bitmap (sb);
 991         }
 992 #endif
 993         return res;
 994 }
 995
 996 static int ext3_check_descriptors (struct super_block * sb)
 997 {
 998         struct ext3_sb_info *sbi = EXT3_SB(sb);
 999         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1000         struct ext3_group_desc * gdp = NULL;
1001         int desc_block = 0;
1002         int i;
1003
1004         ext3_debug ("Checking group descriptors");
1005
1006         for (i = 0; i < sbi->s_groups_count; i++)
1007         {
1008                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1009                         gdp = (struct ext3_group_desc *)
1010                                         sbi->s_group_desc[desc_block++]->b_data;
1011                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1012                     le32_to_cpu(gdp->bg_block_bitmap) >=
1013                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1014                 {
1015                         ext3_error (sb, "ext3_check_descriptors",
1016                                     "Block bitmap for group %d"
1017                                     " not in group (block %lu)!",
1018                                     i, (unsigned long)
1019                                         le32_to_cpu(gdp->bg_block_bitmap));
1020                         return 0;
1021                 }
1022                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1023                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1024                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1025                 {
1026                         ext3_error (sb, "ext3_check_descriptors",
1027                                     "Inode bitmap for group %d"
1028                                     " not in group (block %lu)!",
1029                                     i, (unsigned long)
1030                                         le32_to_cpu(gdp->bg_inode_bitmap));
1031                         return 0;
1032                 }
1033                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1034                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1035                     block + EXT3_BLOCKS_PER_GROUP(sb))
1036                 {
1037                         ext3_error (sb, "ext3_check_descriptors",
1038                                     "Inode table for group %d"
1039                                     " not in group (block %lu)!",
1040                                     i, (unsigned long)
1041                                         le32_to_cpu(gdp->bg_inode_table));
1042                         return 0;
1043                 }
1044                 block += EXT3_BLOCKS_PER_GROUP(sb);
1045                 gdp++;
1046         }
1047
1048         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1049         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1050         return 1;
1051 }
1052
1053
1054 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1055  * the superblock) which were deleted from all directories, but held open by
1056  * a process at the time of a crash.  We walk the list and try to delete these
1057  * inodes at recovery time (only with a read-write filesystem).
1058  *
1059  * In order to keep the orphan inode chain consistent during traversal (in
1060  * case of crash during recovery), we link each inode into the superblock
1061  * orphan list_head and handle it the same way as an inode deletion during
1062  * normal operation (which journals the operations for us).
1063  *
1064  * We only do an iget() and an iput() on each inode, which is very safe if we
1065  * accidentally point at an in-use or already deleted inode.  The worst that
1066  * can happen in this case is that we get a "bit already cleared" message from
1067  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1068  * e2fsck was run on this filesystem, and it must have already done the orphan
1069  * inode cleanup for us, so we can safely abort without any further action.
1070  */
1071 static void ext3_orphan_cleanup (struct super_block * sb,
1072                                  struct ext3_super_block * es)
1073 {
1074         unsigned int s_flags = sb->s_flags;
1075         int nr_orphans = 0, nr_truncates = 0;
1076 #ifdef CONFIG_QUOTA
1077         int i;
1078 #endif
1079         if (!es->s_last_orphan) {
1080                 jbd_debug(4, "no orphan inodes to clean up\n");
1081                 return;
1082         }
1083
1084         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1085                 if (es->s_last_orphan)
1086                         jbd_debug(1, "Errors on filesystem, "
1087                                   "clearing orphan list.\n");
1088                 es->s_last_orphan = 0;
1089                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1090                 return;
1091         }
1092
1093         if (s_flags & MS_RDONLY) {
1094                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1095                        sb->s_id);
1096                 sb->s_flags &= ~MS_RDONLY;
1097         }
1098 #ifdef CONFIG_QUOTA
1099         /* Needed for iput() to work correctly and not trash data */
1100         sb->s_flags |= MS_ACTIVE;
1101         /* Turn on quotas so that they are updated correctly */
1102         for (i = 0; i < MAXQUOTAS; i++) {
1103                 if (EXT3_SB(sb)->s_qf_names[i]) {
1104                         int ret = ext3_quota_on_mount(sb, i);
1105                         if (ret < 0)
1106                                 printk(KERN_ERR
1107                                         "EXT3-fs: Cannot turn on journalled "
1108                                         "quota: error %d\n", ret);
1109                 }
1110         }
1111 #endif
1112
1113         while (es->s_last_orphan) {
1114                 struct inode *inode;
1115
1116                 if (!(inode =
1117                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1118                         es->s_last_orphan = 0;
1119                         break;
1120                 }
1121
1122                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1123                 DQUOT_INIT(inode);
1124                 if (inode->i_nlink) {
1125                         printk(KERN_DEBUG
1126                                 "%s: truncating inode %ld to %Ld bytes\n",
1127                                 __FUNCTION__, inode->i_ino, inode->i_size);
1128                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1129                                   inode->i_ino, inode->i_size);
1130                         ext3_truncate(inode);
1131                         nr_truncates++;
1132                 } else {
1133                         printk(KERN_DEBUG
1134                                 "%s: deleting unreferenced inode %ld\n",
1135                                 __FUNCTION__, inode->i_ino);
1136                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1137                                   inode->i_ino);
1138                         nr_orphans++;
1139                 }
1140                 iput(inode);  /* The delete magic happens here! */
1141         }
1142
1143 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1144
1145         if (nr_orphans)
1146                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1147                        sb->s_id, PLURAL(nr_orphans));
1148         if (nr_truncates)
1149                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1150                        sb->s_id, PLURAL(nr_truncates));
1151 #ifdef CONFIG_QUOTA
1152         /* Turn quotas off */
1153         for (i = 0; i < MAXQUOTAS; i++) {
1154                 if (sb_dqopt(sb)->files[i])
1155                         ext3_quota_off_mount(sb, i);
1156         }
1157 #endif
1158         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1159 }
1160
1161 #define log2(n) ffz(~(n))
1162
1163 /*
1164  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1165  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1166  * We need to be 1 filesystem block less than the 2^32 sector limit.
1167  */
1168 static loff_t ext3_max_size(int bits)
1169 {
1170         loff_t res = EXT3_NDIR_BLOCKS;
1171         res += 1LL << (bits-2);
1172         res += 1LL << (2*(bits-2));
1173         res += 1LL << (3*(bits-2));
1174         res <<= bits;
1175         if (res > (512LL << 32) - (1 << bits))
1176                 res = (512LL << 32) - (1 << bits);
1177         return res;
1178 }
1179
1180 static unsigned long descriptor_loc(struct super_block *sb,
1181                                     unsigned long logic_sb_block,
1182                                     int nr)
1183 {
1184         struct ext3_sb_info *sbi = EXT3_SB(sb);
1185         unsigned long bg, first_data_block, first_meta_bg;
1186         int has_super = 0;
1187
1188         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1189         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1190
1191         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1192             nr < first_meta_bg)
1193                 return (logic_sb_block + nr + 1);
1194         bg = sbi->s_desc_per_block * nr;
1195         if (ext3_bg_has_super(sb, bg))
1196                 has_super = 1;
1197         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1198 }
1199
1200
1201 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1202 {
1203         struct buffer_head * bh;
1204         struct ext3_super_block *es = NULL;
1205         struct ext3_sb_info *sbi;
1206         unsigned long block;
1207         unsigned long sb_block = get_sb_block(&data);
1208         unsigned long logic_sb_block;
1209         unsigned long offset = 0;
1210         unsigned long journal_inum = 0;
1211         unsigned long def_mount_opts;
1212         struct inode *root;
1213         int blocksize;
1214         int hblock;
1215         int db_count;
1216         int i;
1217         int needs_recovery;
1218
1219         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1220         if (!sbi)
1221                 return -ENOMEM;
1222         sb->s_fs_info = sbi;
1223         memset(sbi, 0, sizeof(*sbi));
1224         sbi->s_mount_opt = 0;
1225         sbi->s_resuid = EXT3_DEF_RESUID;
1226         sbi->s_resgid = EXT3_DEF_RESGID;
1227
1228         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1229         if (!blocksize) {
1230                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1231                 goto out_fail;
1232         }
1233
1234         /*
1235          * The ext3 superblock will not be buffer aligned for other than 1kB
1236          * block sizes.  We need to calculate the offset from buffer start.
1237          */
1238         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1239                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1240                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1241         } else {
1242                 logic_sb_block = sb_block;
1243         }
1244
1245         if (!(bh = sb_bread(sb, logic_sb_block))) {
1246                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1247                 goto out_fail;
1248         }
1249         /*
1250          * Note: s_es must be initialized as soon as possible because
1251          *       some ext3 macro-instructions depend on its value
1252          */
1253         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1254         sbi->s_es = es;
1255         sb->s_magic = le16_to_cpu(es->s_magic);
1256         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1257                 if (!silent)
1258                         printk(KERN_ERR
1259                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1260                                sb->s_id);
1261                 goto failed_mount;
1262         }
1263
1264         /* Set defaults before we parse the mount options */
1265         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1266         if (def_mount_opts & EXT3_DEFM_DEBUG)
1267                 set_opt(sbi->s_mount_opt, DEBUG);
1268         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1269                 set_opt(sbi->s_mount_opt, GRPID);
1270         if (def_mount_opts & EXT3_DEFM_UID16)
1271                 set_opt(sbi->s_mount_opt, NO_UID32);
1272         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1273                 set_opt(sbi->s_mount_opt, XATTR_USER);
1274         if (def_mount_opts & EXT3_DEFM_ACL)
1275                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1276         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1277                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1278         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1279                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1280         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1281                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1282
1283         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1284                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1285         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1286                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1287
1288         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1289         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1290
1291         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1292                 goto failed_mount;
1293
1294         sb->s_flags |= MS_ONE_SECOND;
1295         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1296                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1297
1298         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1299             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1300              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1301              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1302                 printk(KERN_WARNING
1303                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1304                        "running e2fsck is recommended\n");
1305         /*
1306          * Check feature flags regardless of the revision level, since we
1307          * previously didn't change the revision level when setting the flags,
1308          * so there is a chance incompat flags are set on a rev 0 filesystem.
1309          */
1310         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1311                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1312                        "unsupported optional features (%x).\n",
1313                        sb->s_id, i);
1314                 goto failed_mount;
1315         }
1316         if (!(sb->s_flags & MS_RDONLY) &&
1317             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1318                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1319                        "unsupported optional features (%x).\n",
1320                        sb->s_id, i);
1321                 goto failed_mount;
1322         }
1323         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1324
1325         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1326             blocksize > EXT3_MAX_BLOCK_SIZE) {
1327                 printk(KERN_ERR
1328                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1329                        blocksize, sb->s_id);
1330                 goto failed_mount;
1331         }
1332
1333         hblock = bdev_hardsect_size(sb->s_bdev);
1334         if (sb->s_blocksize != blocksize) {
1335                 /*
1336                  * Make sure the blocksize for the filesystem is larger
1337                  * than the hardware sectorsize for the machine.
1338                  */
1339                 if (blocksize < hblock) {
1340                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1341                                "device blocksize %d.\n", blocksize, hblock);
1342                         goto failed_mount;
1343                 }
1344
1345                 brelse (bh);
1346                 sb_set_blocksize(sb, blocksize);
1347                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1348                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1349                 bh = sb_bread(sb, logic_sb_block);
1350                 if (!bh) {
1351                         printk(KERN_ERR
1352                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1353                         goto failed_mount;
1354                 }
1355                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1356                 sbi->s_es = es;
1357                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1358                         printk (KERN_ERR
1359                                 "EXT3-fs: Magic mismatch, very weird !\n");
1360                         goto failed_mount;
1361                 }
1362         }
1363
1364         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1365
1366         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1367                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1368                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1369         } else {
1370                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1371                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1372                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1373                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1374                     (sbi->s_inode_size > blocksize)) {
1375                         printk (KERN_ERR
1376                                 "EXT3-fs: unsupported inode size: %d\n",
1377                                 sbi->s_inode_size);
1378                         goto failed_mount;
1379                 }
1380         }
1381         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1382                                    le32_to_cpu(es->s_log_frag_size);
1383         if (blocksize != sbi->s_frag_size) {
1384                 printk(KERN_ERR
1385                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1386                        sbi->s_frag_size, blocksize);
1387                 goto failed_mount;
1388         }
1389         sbi->s_frags_per_block = 1;
1390         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1391         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1392         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1393         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1394         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1395         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1396         sbi->s_sbh = bh;
1397         sbi->s_mount_state = le16_to_cpu(es->s_state);
1398         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1399         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1400         for (i=0; i < 4; i++)
1401                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1402         sbi->s_def_hash_version = es->s_def_hash_version;
1403
1404         if (sbi->s_blocks_per_group > blocksize * 8) {
1405                 printk (KERN_ERR
1406                         "EXT3-fs: #blocks per group too big: %lu\n",
1407                         sbi->s_blocks_per_group);
1408                 goto failed_mount;
1409         }
1410         if (sbi->s_frags_per_group > blocksize * 8) {
1411                 printk (KERN_ERR
1412                         "EXT3-fs: #fragments per group too big: %lu\n",
1413                         sbi->s_frags_per_group);
1414                 goto failed_mount;
1415         }
1416         if (sbi->s_inodes_per_group > blocksize * 8) {
1417                 printk (KERN_ERR
1418                         "EXT3-fs: #inodes per group too big: %lu\n",
1419                         sbi->s_inodes_per_group);
1420                 goto failed_mount;
1421         }
1422
1423         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1424                                le32_to_cpu(es->s_first_data_block) +
1425                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1426                               EXT3_BLOCKS_PER_GROUP(sb);
1427         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1428                    EXT3_DESC_PER_BLOCK(sb);
1429         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1430                                     GFP_KERNEL);
1431         if (sbi->s_group_desc == NULL) {
1432                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1433                 goto failed_mount;
1434         }
1435         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1436                         GFP_KERNEL);
1437         if (!sbi->s_debts) {
1438                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1439                 goto failed_mount2;
1440         }
1441         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1442
1443         percpu_counter_init(&sbi->s_freeblocks_counter);
1444         percpu_counter_init(&sbi->s_freeinodes_counter);
1445         percpu_counter_init(&sbi->s_dirs_counter);
1446         bgl_lock_init(&sbi->s_blockgroup_lock);
1447
1448         for (i = 0; i < db_count; i++) {
1449                 block = descriptor_loc(sb, logic_sb_block, i);
1450                 sbi->s_group_desc[i] = sb_bread(sb, block);
1451                 if (!sbi->s_group_desc[i]) {
1452                         printk (KERN_ERR "EXT3-fs: "
1453                                 "can't read group descriptor %d\n", i);
1454                         db_count = i;
1455                         goto failed_mount2;
1456                 }
1457         }
1458         if (!ext3_check_descriptors (sb)) {
1459                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1460                 goto failed_mount2;
1461         }
1462         sbi->s_gdb_count = db_count;
1463         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1464         spin_lock_init(&sbi->s_next_gen_lock);
1465         /*
1466          * set up enough so that it can read an inode
1467          */
1468         sb->s_op = &ext3_sops;
1469         sb->s_export_op = &ext3_export_ops;
1470 #ifdef CONFIG_QUOTA
1471         sb->s_qcop = &ext3_qctl_operations;
1472         sb->dq_op = &ext3_quota_operations;
1473 #endif
1474         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1475
1476         sb->s_root = NULL;
1477
1478         needs_recovery = (es->s_last_orphan != 0 ||
1479                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1480                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1481
1482         /*
1483          * The first inode we look at is the journal inode.  Don't try
1484          * root first: it may be modified in the journal!
1485          */
1486         if (!test_opt(sb, NOLOAD) &&
1487             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1488                 if (ext3_load_journal(sb, es))
1489                         goto failed_mount2;
1490         } else if (journal_inum) {
1491                 if (ext3_create_journal(sb, es, journal_inum))
1492                         goto failed_mount2;
1493         } else {
1494                 if (!silent)
1495                         printk (KERN_ERR
1496                                 "ext3: No journal on filesystem on %s\n",
1497                                 sb->s_id);
1498                 goto failed_mount2;
1499         }
1500
1501         /* We have now updated the journal if required, so we can
1502          * validate the data journaling mode. */
1503         switch (test_opt(sb, DATA_FLAGS)) {
1504         case 0:
1505                 /* No mode set, assume a default based on the journal
1506                    capabilities: ORDERED_DATA if the journal can
1507                    cope, else JOURNAL_DATA */
1508                 if (journal_check_available_features
1509                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1510                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1511                 else
1512                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1513                 break;
1514
1515         case EXT3_MOUNT_ORDERED_DATA:
1516         case EXT3_MOUNT_WRITEBACK_DATA:
1517                 if (!journal_check_available_features
1518                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1519                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1520                                "requested data journaling mode\n");
1521                         goto failed_mount3;
1522                 }
1523         default:
1524                 break;
1525         }
1526
1527         /*
1528          * The journal_load will have done any necessary log recovery,
1529          * so we can safely mount the rest of the filesystem now.
1530          */
1531
1532         root = iget(sb, EXT3_ROOT_INO);
1533         sb->s_root = d_alloc_root(root);
1534         if (!sb->s_root) {
1535                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1536                 iput(root);
1537                 goto failed_mount3;
1538         }
1539         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1540                 dput(sb->s_root);
1541                 sb->s_root = NULL;
1542                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1543                 goto failed_mount3;
1544         }
1545
1546         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1547         /*
1548          * akpm: core read_super() calls in here with the superblock locked.
1549          * That deadlocks, because orphan cleanup needs to lock the superblock
1550          * in numerous places.  Here we just pop the lock - it's relatively
1551          * harmless, because we are now ready to accept write_super() requests,
1552          * and aviro says that's the only reason for hanging onto the
1553          * superblock lock.
1554          */
1555         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1556         ext3_orphan_cleanup(sb, es);
1557         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1558         if (needs_recovery)
1559                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1560         ext3_mark_recovery_complete(sb, es);
1561         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1562                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1563                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1564                 "writeback");
1565
1566         percpu_counter_mod(&sbi->s_freeblocks_counter,
1567                 ext3_count_free_blocks(sb));
1568         percpu_counter_mod(&sbi->s_freeinodes_counter,
1569                 ext3_count_free_inodes(sb));
1570         percpu_counter_mod(&sbi->s_dirs_counter,
1571                 ext3_count_dirs(sb));
1572
1573         return 0;
1574
1575 failed_mount3:
1576         journal_destroy(sbi->s_journal);
1577 failed_mount2:
1578         kfree(sbi->s_debts);
1579         for (i = 0; i < db_count; i++)
1580                 brelse(sbi->s_group_desc[i]);
1581         kfree(sbi->s_group_desc);
1582 failed_mount:
1583 #ifdef CONFIG_QUOTA
1584         for (i = 0; i < MAXQUOTAS; i++) {
1585                 if (sbi->s_qf_names[i])
1586                         kfree(sbi->s_qf_names[i]);
1587         }
1588 #endif
1589         ext3_blkdev_remove(sbi);
1590         brelse(bh);
1591 out_fail:
1592         sb->s_fs_info = NULL;
1593         kfree(sbi);
1594         return -EINVAL;
1595 }
1596
1597 /*
1598  * Setup any per-fs journal parameters now.  We'll do this both on
1599  * initial mount, once the journal has been initialised but before we've
1600  * done any recovery; and again on any subsequent remount.
1601  */
1602 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1603                                      journal_t *journal)
1604 {
1605         if (sbi->s_commit_interval)
1606                 journal->j_commit_interval = sbi->s_commit_interval;
1607         /* We could also set up an ext3-specific default for the commit
1608          * interval here, but for now we'll just fall back to the jbd
1609          * default. */
1610 }
1611
1612
1613 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1614 {
1615         struct inode *journal_inode;
1616         journal_t *journal;
1617
1618         /* First, test for the existence of a valid inode on disk.  Bad
1619          * things happen if we iget() an unused inode, as the subsequent
1620          * iput() will try to delete it. */
1621
1622         journal_inode = iget(sb, journal_inum);
1623         if (!journal_inode) {
1624                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1625                 return NULL;
1626         }
1627         if (!journal_inode->i_nlink) {
1628                 make_bad_inode(journal_inode);
1629                 iput(journal_inode);
1630                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1631                 return NULL;
1632         }
1633
1634         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1635                   journal_inode, journal_inode->i_size);
1636         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1637                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1638                 iput(journal_inode);
1639                 return NULL;
1640         }
1641
1642         journal = journal_init_inode(journal_inode);
1643         if (!journal) {
1644                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1645                 iput(journal_inode);
1646                 return NULL;
1647         }
1648         journal->j_private = sb;
1649         ext3_init_journal_params(EXT3_SB(sb), journal);
1650         return journal;
1651 }
1652
1653 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1654                                        dev_t j_dev)
1655 {
1656         struct buffer_head * bh;
1657         journal_t *journal;
1658         int start;
1659         int len;
1660         int hblock, blocksize;
1661         unsigned long sb_block;
1662         unsigned long offset;
1663         struct ext3_super_block * es;
1664         struct block_device *bdev;
1665
1666         bdev = ext3_blkdev_get(j_dev);
1667         if (bdev == NULL)
1668                 return NULL;
1669
1670         if (bd_claim(bdev, sb)) {
1671                 printk(KERN_ERR
1672                         "EXT3: failed to claim external journal device.\n");
1673                 blkdev_put(bdev);
1674                 return NULL;
1675         }
1676
1677         blocksize = sb->s_blocksize;
1678         hblock = bdev_hardsect_size(bdev);
1679         if (blocksize < hblock) {
1680                 printk(KERN_ERR
1681                         "EXT3-fs: blocksize too small for journal device.\n");
1682                 goto out_bdev;
1683         }
1684
1685         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1686         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1687         set_blocksize(bdev, blocksize);
1688         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1689                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1690                        "external journal\n");
1691                 goto out_bdev;
1692         }
1693
1694         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1695         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1696             !(le32_to_cpu(es->s_feature_incompat) &
1697               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1698                 printk(KERN_ERR "EXT3-fs: external journal has "
1699                                         "bad superblock\n");
1700                 brelse(bh);
1701                 goto out_bdev;
1702         }
1703
1704         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1705                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1706                 brelse(bh);
1707                 goto out_bdev;
1708         }
1709
1710         len = le32_to_cpu(es->s_blocks_count);
1711         start = sb_block + 1;
1712         brelse(bh);     /* we're done with the superblock */
1713
1714         journal = journal_init_dev(bdev, sb->s_bdev,
1715                                         start, len, blocksize);
1716         if (!journal) {
1717                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1718                 goto out_bdev;
1719         }
1720         journal->j_private = sb;
1721         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1722         wait_on_buffer(journal->j_sb_buffer);
1723         if (!buffer_uptodate(journal->j_sb_buffer)) {
1724                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1725                 goto out_journal;
1726         }
1727         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1728                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1729                                         "user (unsupported) - %d\n",
1730                         ntohl(journal->j_superblock->s_nr_users));
1731                 goto out_journal;
1732         }
1733         EXT3_SB(sb)->journal_bdev = bdev;
1734         ext3_init_journal_params(EXT3_SB(sb), journal);
1735         return journal;
1736 out_journal:
1737         journal_destroy(journal);
1738 out_bdev:
1739         ext3_blkdev_put(bdev);
1740         return NULL;
1741 }
1742
1743 static int ext3_load_journal(struct super_block * sb,
1744                              struct ext3_super_block * es)
1745 {
1746         journal_t *journal;
1747         int journal_inum = le32_to_cpu(es->s_journal_inum);
1748         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1749         int err = 0;
1750         int really_read_only;
1751
1752         really_read_only = bdev_read_only(sb->s_bdev);
1753
1754         /*
1755          * Are we loading a blank journal or performing recovery after a
1756          * crash?  For recovery, we need to check in advance whether we
1757          * can get read-write access to the device.
1758          */
1759
1760         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1761                 if (sb->s_flags & MS_RDONLY) {
1762                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1763                                         "required on readonly filesystem.\n");
1764                         if (really_read_only) {
1765                                 printk(KERN_ERR "EXT3-fs: write access "
1766                                         "unavailable, cannot proceed.\n");
1767                                 return -EROFS;
1768                         }
1769                         printk (KERN_INFO "EXT3-fs: write access will "
1770                                         "be enabled during recovery.\n");
1771                 }
1772         }
1773
1774         if (journal_inum && journal_dev) {
1775                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1776                        "and inode journals!\n");
1777                 return -EINVAL;
1778         }
1779
1780         if (journal_inum) {
1781                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1782                         return -EINVAL;
1783         } else {
1784                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1785                         return -EINVAL;
1786         }
1787
1788         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1789                 err = journal_update_format(journal);
1790                 if (err)  {
1791                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1792                         journal_destroy(journal);
1793                         return err;
1794                 }
1795         }
1796
1797         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1798                 err = journal_wipe(journal, !really_read_only);
1799         if (!err)
1800                 err = journal_load(journal);
1801
1802         if (err) {
1803                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1804                 journal_destroy(journal);
1805                 return err;
1806         }
1807
1808         EXT3_SB(sb)->s_journal = journal;
1809         ext3_clear_journal_err(sb, es);
1810         return 0;
1811 }
1812
1813 static int ext3_create_journal(struct super_block * sb,
1814                                struct ext3_super_block * es,
1815                                int journal_inum)
1816 {
1817         journal_t *journal;
1818
1819         if (sb->s_flags & MS_RDONLY) {
1820                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1821                                 "create journal.\n");
1822                 return -EROFS;
1823         }
1824
1825         if (!(journal = ext3_get_journal(sb, journal_inum)))
1826                 return -EINVAL;
1827
1828         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1829                journal_inum);
1830
1831         if (journal_create(journal)) {
1832                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1833                 journal_destroy(journal);
1834                 return -EIO;
1835         }
1836
1837         EXT3_SB(sb)->s_journal = journal;
1838
1839         ext3_update_dynamic_rev(sb);
1840         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1841         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1842
1843         es->s_journal_inum = cpu_to_le32(journal_inum);
1844         sb->s_dirt = 1;
1845
1846         /* Make sure we flush the recovery flag to disk. */
1847         ext3_commit_super(sb, es, 1);
1848
1849         return 0;
1850 }
1851
1852 static void ext3_commit_super (struct super_block * sb,
1853                                struct ext3_super_block * es,
1854                                int sync)
1855 {
1856         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1857
1858         if (!sbh)
1859                 return;
1860         es->s_wtime = cpu_to_le32(get_seconds());
1861         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1862         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1863         BUFFER_TRACE(sbh, "marking dirty");
1864         mark_buffer_dirty(sbh);
1865         if (sync)
1866                 sync_dirty_buffer(sbh);
1867 }
1868
1869
1870 /*
1871  * Have we just finished recovery?  If so, and if we are mounting (or
1872  * remounting) the filesystem readonly, then we will end up with a
1873  * consistent fs on disk.  Record that fact.
1874  */
1875 static void ext3_mark_recovery_complete(struct super_block * sb,
1876                                         struct ext3_super_block * es)
1877 {
1878         journal_t *journal = EXT3_SB(sb)->s_journal;
1879
1880         journal_lock_updates(journal);
1881         journal_flush(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         journal_unlock_updates(journal);
1889 }
1890
1891 /*
1892  * If we are mounting (or read-write remounting) a filesystem whose journal
1893  * has recorded an error from a previous lifetime, move that error to the
1894  * main filesystem now.
1895  */
1896 static void ext3_clear_journal_err(struct super_block * sb,
1897                                    struct ext3_super_block * es)
1898 {
1899         journal_t *journal;
1900         int j_errno;
1901         const char *errstr;
1902
1903         journal = EXT3_SB(sb)->s_journal;
1904
1905         /*
1906          * Now check for any error status which may have been recorded in the
1907          * journal by a prior ext3_error() or ext3_abort()
1908          */
1909
1910         j_errno = journal_errno(journal);
1911         if (j_errno) {
1912                 char nbuf[16];
1913
1914                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1915                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1916                              "from previous mount: %s", errstr);
1917                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1918                              "filesystem check.");
1919
1920                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1921                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1922                 ext3_commit_super (sb, es, 1);
1923
1924                 journal_clear_err(journal);
1925         }
1926 }
1927
1928 /*
1929  * Force the running and committing transactions to commit,
1930  * and wait on the commit.
1931  */
1932 int ext3_force_commit(struct super_block *sb)
1933 {
1934         journal_t *journal;
1935         int ret;
1936
1937         if (sb->s_flags & MS_RDONLY)
1938                 return 0;
1939
1940         journal = EXT3_SB(sb)->s_journal;
1941         sb->s_dirt = 0;
1942         ret = ext3_journal_force_commit(journal);
1943         return ret;
1944 }
1945
1946 /*
1947  * Ext3 always journals updates to the superblock itself, so we don't
1948  * have to propagate any other updates to the superblock on disk at this
1949  * point.  Just start an async writeback to get the buffers on their way
1950  * to the disk.
1951  *
1952  * This implicitly triggers the writebehind on sync().
1953  */
1954
1955 void ext3_write_super (struct super_block * sb)
1956 {
1957         if (down_trylock(&sb->s_lock) == 0)
1958                 BUG();
1959         sb->s_dirt = 0;
1960 }
1961
1962 static int ext3_sync_fs(struct super_block *sb, int wait)
1963 {
1964         tid_t target;
1965
1966         sb->s_dirt = 0;
1967         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1968                 if (wait)
1969                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1970         }
1971         return 0;
1972 }
1973
1974 /*
1975  * LVM calls this function before a (read-only) snapshot is created.  This
1976  * gives us a chance to flush the journal completely and mark the fs clean.
1977  */
1978 void ext3_write_super_lockfs(struct super_block *sb)
1979 {
1980         sb->s_dirt = 0;
1981
1982         if (!(sb->s_flags & MS_RDONLY)) {
1983                 journal_t *journal = EXT3_SB(sb)->s_journal;
1984
1985                 /* Now we set up the journal barrier. */
1986                 journal_lock_updates(journal);
1987                 journal_flush(journal);
1988
1989                 /* Journal blocked and flushed, clear needs_recovery flag. */
1990                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1991                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
1992         }
1993 }
1994
1995 /*
1996  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
1997  * flag here, even though the filesystem is not technically dirty yet.
1998  */
1999 void ext3_unlockfs(struct super_block *sb)
2000 {
2001         if (!(sb->s_flags & MS_RDONLY)) {
2002                 lock_super(sb);
2003                 /* Reser the needs_recovery flag before the fs is unlocked. */
2004                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2005                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2006                 unlock_super(sb);
2007                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2008         }
2009 }
2010
2011 int ext3_remount (struct super_block * sb, int * flags, char * data)
2012 {
2013         struct ext3_super_block * es;
2014         struct ext3_sb_info *sbi = EXT3_SB(sb);
2015         unsigned long tmp;
2016
2017         /*
2018          * Allow the "check" option to be passed as a remount option.
2019          */
2020         if (!parse_options(data, sb, &tmp, 1))
2021                 return -EINVAL;
2022
2023         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2024                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2025
2026         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2027                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2028
2029         es = sbi->s_es;
2030
2031         ext3_init_journal_params(sbi, sbi->s_journal);
2032
2033         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2034                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2035                         return -EROFS;
2036
2037                 if (*flags & MS_RDONLY) {
2038                         /*
2039                          * First of all, the unconditional stuff we have to do
2040                          * to disable replay of the journal when we next remount
2041                          */
2042                         sb->s_flags |= MS_RDONLY;
2043
2044                         /*
2045                          * OK, test if we are remounting a valid rw partition
2046                          * readonly, and if so set the rdonly flag and then
2047                          * mark the partition as valid again.
2048                          */
2049                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2050                             (sbi->s_mount_state & EXT3_VALID_FS))
2051                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2052
2053                         ext3_mark_recovery_complete(sb, es);
2054                 } else {
2055                         int ret;
2056                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2057                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2058                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2059                                        "remount RDWR because of unsupported "
2060                                        "optional features (%x).\n",
2061                                        sb->s_id, ret);
2062                                 return -EROFS;
2063                         }
2064                         /*
2065                          * Mounting a RDONLY partition read-write, so reread
2066                          * and store the current valid flag.  (It may have
2067                          * been changed by e2fsck since we originally mounted
2068                          * the partition.)
2069                          */
2070                         ext3_clear_journal_err(sb, es);
2071                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2072                         if (!ext3_setup_super (sb, es, 0))
2073                                 sb->s_flags &= ~MS_RDONLY;
2074                 }
2075         }
2076         return 0;
2077 }
2078
2079 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2080 {
2081         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2082         unsigned long overhead;
2083         int i;
2084
2085         if (test_opt (sb, MINIX_DF))
2086                 overhead = 0;
2087         else {
2088                 /*
2089                  * Compute the overhead (FS structures)
2090                  */
2091
2092                 /*
2093                  * All of the blocks before first_data_block are
2094                  * overhead
2095                  */
2096                 overhead = le32_to_cpu(es->s_first_data_block);
2097
2098                 /*
2099                  * Add the overhead attributed to the superblock and
2100                  * block group descriptors.  If the sparse superblocks
2101                  * feature is turned on, then not all groups have this.
2102                  */
2103                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2104                         overhead += ext3_bg_has_super(sb, i) +
2105                                 ext3_bg_num_gdb(sb, i);
2106
2107                 /*
2108                  * Every block group has an inode bitmap, a block
2109                  * bitmap, and an inode table.
2110                  */
2111                 overhead += (EXT3_SB(sb)->s_groups_count *
2112                              (2 + EXT3_SB(sb)->s_itb_per_group));
2113         }
2114
2115         buf->f_type = EXT3_SUPER_MAGIC;
2116         buf->f_bsize = sb->s_blocksize;
2117         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2118         buf->f_bfree = ext3_count_free_blocks (sb);
2119         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2120         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2121                 buf->f_bavail = 0;
2122         buf->f_files = le32_to_cpu(es->s_inodes_count);
2123         buf->f_ffree = ext3_count_free_inodes (sb);
2124         buf->f_namelen = EXT3_NAME_LEN;
2125         return 0;
2126 }
2127
2128 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2129  * is locked for write. Otherwise the are possible deadlocks:
2130  * Process 1                         Process 2
2131  * ext3_create()                     quota_sync()
2132  *   journal_start()                   write_dquot()
2133  *   DQUOT_INIT()                        down(dqio_sem)
2134  *     down(dqio_sem)                    journal_start()
2135  *
2136  */
2137
2138 #ifdef CONFIG_QUOTA
2139
2140 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2141 {
2142         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2143 }
2144
2145 static int ext3_dquot_initialize(struct inode *inode, int type)
2146 {
2147         handle_t *handle;
2148         int ret, err;
2149
2150         /* We may create quota structure so we need to reserve enough blocks */
2151         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2152         if (IS_ERR(handle))
2153                 return PTR_ERR(handle);
2154         ret = dquot_initialize(inode, type);
2155         err = ext3_journal_stop(handle);
2156         if (!ret)
2157                 ret = err;
2158         return ret;
2159 }
2160
2161 static int ext3_dquot_drop(struct inode *inode)
2162 {
2163         handle_t *handle;
2164         int ret, err;
2165
2166         /* We may delete quota structure so we need to reserve enough blocks */
2167         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2168         if (IS_ERR(handle))
2169                 return PTR_ERR(handle);
2170         ret = dquot_drop(inode);
2171         err = ext3_journal_stop(handle);
2172         if (!ret)
2173                 ret = err;
2174         return ret;
2175 }
2176
2177 static int ext3_write_dquot(struct dquot *dquot)
2178 {
2179         int ret, err;
2180         handle_t *handle;
2181
2182         handle = ext3_journal_start(dquot_to_inode(dquot),
2183                                         EXT3_QUOTA_TRANS_BLOCKS);
2184         if (IS_ERR(handle))
2185                 return PTR_ERR(handle);
2186         ret = dquot_commit(dquot);
2187         err = ext3_journal_stop(handle);
2188         if (!ret)
2189                 ret = err;
2190         return ret;
2191 }
2192
2193 static int ext3_acquire_dquot(struct dquot *dquot)
2194 {
2195         int ret, err;
2196         handle_t *handle;
2197
2198         handle = ext3_journal_start(dquot_to_inode(dquot),
2199                                         EXT3_QUOTA_INIT_BLOCKS);
2200         if (IS_ERR(handle))
2201                 return PTR_ERR(handle);
2202         ret = dquot_acquire(dquot);
2203         err = ext3_journal_stop(handle);
2204         if (!ret)
2205                 ret = err;
2206         return ret;
2207 }
2208
2209 static int ext3_release_dquot(struct dquot *dquot)
2210 {
2211         int ret, err;
2212         handle_t *handle;
2213
2214         handle = ext3_journal_start(dquot_to_inode(dquot),
2215                                         EXT3_QUOTA_INIT_BLOCKS);
2216         if (IS_ERR(handle))
2217                 return PTR_ERR(handle);
2218         ret = dquot_release(dquot);
2219         err = ext3_journal_stop(handle);
2220         if (!ret)
2221                 ret = err;
2222         return ret;
2223 }
2224
2225 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2226 {
2227         /* Are we journalling quotas? */
2228         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2229             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2230                 dquot_mark_dquot_dirty(dquot);
2231                 return ext3_write_dquot(dquot);
2232         } else {
2233                 return dquot_mark_dquot_dirty(dquot);
2234         }
2235 }
2236
2237 static int ext3_write_info(struct super_block *sb, int type)
2238 {
2239         int ret, err;
2240         handle_t *handle;
2241
2242         /* Data block + inode block */
2243         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2244         if (IS_ERR(handle))
2245                 return PTR_ERR(handle);
2246         ret = dquot_commit_info(sb, type);
2247         err = ext3_journal_stop(handle);
2248         if (!ret)
2249                 ret = err;
2250         return ret;
2251 }
2252
2253 /*
2254  * Turn on quotas during mount time - we need to find
2255  * the quota file and such...
2256  */
2257 static int ext3_quota_on_mount(struct super_block *sb, int type)
2258 {
2259         int err;
2260         struct dentry *dentry;
2261         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2262                              .hash = 0,
2263                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2264
2265         dentry = lookup_hash(&name, sb->s_root);
2266         if (IS_ERR(dentry))
2267                 return PTR_ERR(dentry);
2268         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2269         if (err)
2270                 dput(dentry);
2271         /* We keep the dentry reference if everything went ok - we drop it
2272          * on quota_off time */
2273         return err;
2274 }
2275
2276 /* Turn quotas off during mount time */
2277 static int ext3_quota_off_mount(struct super_block *sb, int type)
2278 {
2279         int err;
2280         struct dentry *dentry;
2281
2282         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2283         err = vfs_quota_off_mount(sb, type);
2284         /* We invalidate dentry - it has at least wrong hash... */
2285         d_invalidate(dentry);
2286         dput(dentry);
2287         return err;
2288 }
2289
2290 /*
2291  * Standard function to be called on quota_on
2292  */
2293 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2294                          char *path)
2295 {
2296         int err;
2297         struct nameidata nd;
2298
2299         /* Not journalling quota? */
2300         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2301             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2302                 return vfs_quota_on(sb, type, format_id, path);
2303         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2304         if (err)
2305                 return err;
2306         /* Quotafile not on the same filesystem? */
2307         if (nd.mnt->mnt_sb != sb)
2308                 return -EXDEV;
2309         /* Quotafile not of fs root? */
2310         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2311                 printk(KERN_WARNING
2312                         "EXT3-fs: Quota file not on filesystem root. "
2313                         "Journalled quota will not work.\n");
2314         if (!ext3_should_journal_data(nd.dentry->d_inode))
2315                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2316                         "data-journalling. Journalled quota will not work.\n");
2317         path_release(&nd);
2318         return vfs_quota_on(sb, type, format_id, path);
2319 }
2320
2321 #endif
2322
2323 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2324         int flags, const char *dev_name, void *data)
2325 {
2326         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2327 }
2328
2329 static struct file_system_type ext3_fs_type = {
2330         .owner          = THIS_MODULE,
2331         .name           = "ext3",
2332         .get_sb         = ext3_get_sb,
2333         .kill_sb        = kill_block_super,
2334         .fs_flags       = FS_REQUIRES_DEV,
2335 };
2336
2337 static int __init init_ext3_fs(void)
2338 {
2339         int err = init_ext3_xattr();
2340         if (err)
2341                 return err;
2342         err = init_inodecache();
2343         if (err)
2344                 goto out1;
2345         err = register_filesystem(&ext3_fs_type);
2346         if (err)
2347                 goto out;
2348         return 0;
2349 out:
2350         destroy_inodecache();
2351 out1:
2352         exit_ext3_xattr();
2353         return err;
2354 }
2355
2356 static void __exit exit_ext3_fs(void)
2357 {
2358         unregister_filesystem(&ext3_fs_type);
2359         destroy_inodecache();
2360         exit_ext3_xattr();
2361 }
2362
2363 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2364 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2365 MODULE_LICENSE("GPL");
2366 module_init(init_ext3_fs)
2367 module_exit(exit_ext3_fs)