fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53
  54 /*
  55  * Wrappers for journal_start/end.
  56  *
  57  * The only special thing we need to do here is to make sure that all
  58  * journal_end calls result in the superblock being marked dirty, so
  59  * that sync() will call the filesystem's write_super callback if
  60  * appropriate.
  61  */
  62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (inode->i_sb->s_flags & MS_RDONLY)
  67                 return ERR_PTR(-EROFS);
  68
  69         /* Special case here: if the journal has aborted behind our
  70          * backs (eg. EIO in the commit thread), then we still need to
  71          * take the FS itself readonly cleanly. */
  72         journal = EXT3_JOURNAL(inode);
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(inode->i_sb, __FUNCTION__,
  75                            "Detected aborted journal");
  76                 return ERR_PTR(-EROFS);
  77         }
  78
  79         return journal_start(journal, nblocks);
  80 }
  81
  82 /*
  83  * The only special thing we need to do here is to make sure that all
  84  * journal_stop calls result in the superblock being marked dirty, so
  85  * that sync() will call the filesystem's write_super callback if
  86  * appropriate.
  87  */
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         printk(KERN_ERR "%s: aborting transaction: %s in %s",
 112                caller, errstr, err_fn);
 113
 114         if (bh)
 115                 BUFFER_TRACE(bh, "abort");
 116         journal_abort_handle(handle);
 117         if (!handle->h_err)
 118                 handle->h_err = err;
 119 }
 120
 121 static char error_buf[1024];
 122
 123 /* Deal with the reporting of failure conditions on a filesystem such as
 124  * inconsistencies detected or read IO failures.
 125  *
 126  * On ext2, we can store the error state of the filesystem in the
 127  * superblock.  That is not possible on ext3, because we may have other
 128  * write ordering constraints on the superblock which prevent us from
 129  * writing it out straight away; and given that the journal is about to
 130  * be aborted, we can't rely on the current, or future, transactions to
 131  * write out the superblock safely.
 132  *
 133  * We'll just use the journal_abort() error code to record an error in
 134  * the journal instead.  On recovery, the journal will compain about
 135  * that error until we've noted it down and cleared it.
 136  */
 137
 138 static void ext3_handle_error(struct super_block *sb)
 139 {
 140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 141
 142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 144
 145         if (sb->s_flags & MS_RDONLY)
 146                 return;
 147
 148         if (test_opt (sb, ERRORS_PANIC))
 149                 panic ("EXT3-fs (device %s): panic forced after error\n",
 150                        sb->s_id);
 151         if (test_opt (sb, ERRORS_RO)) {
 152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 153                 sb->s_flags |= MS_RDONLY;
 154         } else {
 155                 journal_t *journal = EXT3_SB(sb)->s_journal;
 156
 157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 158                 if (journal)
 159                         journal_abort(journal, -EIO);
 160         }
 161         ext3_commit_super(sb, es, 1);
 162 }
 163
 164 void ext3_error (struct super_block * sb, const char * function,
 165                  const char * fmt, ...)
 166 {
 167         va_list args;
 168
 169         va_start (args, fmt);
 170         vsprintf (error_buf, fmt, args);
 171         va_end (args);
 172
 173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
 174                 sb->s_id, function, error_buf);
 175
 176         ext3_handle_error(sb);
 177 }
 178
 179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 180 {
 181         char *errstr = NULL;
 182
 183         switch (errno) {
 184         case -EIO:
 185                 errstr = "IO failure";
 186                 break;
 187         case -ENOMEM:
 188                 errstr = "Out of memory";
 189                 break;
 190         case -EROFS:
 191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 192                         errstr = "Journal has aborted";
 193                 else
 194                         errstr = "Readonly filesystem";
 195                 break;
 196         default:
 197                 /* If the caller passed in an extra buffer for unknown
 198                  * errors, textualise them now.  Else we just return
 199                  * NULL. */
 200                 if (nbuf) {
 201                         /* Check for truncated error codes... */
 202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 203                                 errstr = nbuf;
 204                 }
 205                 break;
 206         }
 207
 208         return errstr;
 209 }
 210
 211 /* __ext3_std_error decodes expected errors from journaling functions
 212  * automatically and invokes the appropriate error response.  */
 213
 214 void __ext3_std_error (struct super_block * sb, const char * function,
 215                        int errno)
 216 {
 217         char nbuf[16];
 218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 219
 220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 221                 sb->s_id, function, errstr);
 222
 223         ext3_handle_error(sb);
 224 }
 225
 226 /*
 227  * ext3_abort is a much stronger failure handler than ext3_error.  The
 228  * abort function may be used to deal with unrecoverable failures such
 229  * as journal IO errors or ENOMEM at a critical moment in log management.
 230  *
 231  * We unconditionally force the filesystem into an ABORT|READONLY state,
 232  * unless the error response on the fs has been set to panic in which
 233  * case we take the easy way out and panic immediately.
 234  */
 235
 236 void ext3_abort (struct super_block * sb, const char * function,
 237                  const char * fmt, ...)
 238 {
 239         va_list args;
 240
 241         printk (KERN_CRIT "ext3_abort called.\n");
 242
 243         va_start (args, fmt);
 244         vsprintf (error_buf, fmt, args);
 245         va_end (args);
 246
 247         if (test_opt (sb, ERRORS_PANIC))
 248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
 249                        sb->s_id, function, error_buf);
 250
 251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
 252                 sb->s_id, function, error_buf);
 253
 254         if (sb->s_flags & MS_RDONLY)
 255                 return;
 256
 257         printk (KERN_CRIT "Remounting filesystem read-only\n");
 258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 259         sb->s_flags |= MS_RDONLY;
 260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 262 }
 263
 264 /* Deal with the reporting of failure conditions while running, such as
 265  * inconsistencies in operation or invalid system states.
 266  *
 267  * Use ext3_error() for cases of invalid filesystem states, as that will
 268  * record an error on disk and force a filesystem check on the next boot.
 269  */
 270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 271                             const char * fmt, ...)
 272 {
 273         va_list args;
 274
 275         va_start (args, fmt);
 276         vsprintf (error_buf, fmt, args);
 277         va_end (args);
 278
 279         /* this is to prevent panic from syncing this filesystem */
 280         /* AKPM: is this sufficient? */
 281         sb->s_flags |= MS_RDONLY;
 282         panic ("EXT3-fs panic (device %s): %s: %s\n",
 283                sb->s_id, function, error_buf);
 284 }
 285
 286 void ext3_warning (struct super_block * sb, const char * function,
 287                    const char * fmt, ...)
 288 {
 289         va_list args;
 290
 291         va_start (args, fmt);
 292         vsprintf (error_buf, fmt, args);
 293         va_end (args);
 294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
 295                 sb->s_id, function, error_buf);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = 0;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        le32_to_cpu(NEXT_ORPHAN(inode)));
 384         }
 385 }
 386
 387 void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = le16_to_cpu(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         kfree(sbi->s_debts);
 407         brelse(sbi->s_sbh);
 408 #ifdef CONFIG_QUOTA
 409         for (i = 0; i < MAXQUOTAS; i++) {
 410                 if (sbi->s_qf_names[i])
 411                         kfree(sbi->s_qf_names[i]);
 412         }
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 init_MUTEX(&ei->truncate_sem);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 484                                              init_once, NULL);
 485         if (ext3_inode_cachep == NULL)
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static void destroy_inodecache(void)
 491 {
 492         if (kmem_cache_destroy(ext3_inode_cachep))
 493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 494 }
 495
 496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500        if (EXT3_I(inode)->i_acl &&
 501            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 502                posix_acl_release(EXT3_I(inode)->i_acl);
 503                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 504        }
 505        if (EXT3_I(inode)->i_default_acl &&
 506            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 507                posix_acl_release(EXT3_I(inode)->i_default_acl);
 508                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 509        }
 510 }
 511
 512 #else
 513 # define ext3_clear_inode NULL
 514 #endif
 515
 516 #ifdef CONFIG_QUOTA
 517
 518 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 519 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 520
 521 static int ext3_dquot_initialize(struct inode *inode, int type);
 522 static int ext3_dquot_drop(struct inode *inode);
 523 static int ext3_write_dquot(struct dquot *dquot);
 524 static int ext3_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_tagxid, 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_tagxid, "tagxid"},
 632         {Opt_ignore, "grpquota"},
 633         {Opt_ignore, "noquota"},
 634         {Opt_ignore, "quota"},
 635         {Opt_ignore, "usrquota"},
 636         {Opt_err, NULL}
 637 };
 638
 639 static unsigned long get_sb_block(void **data)
 640 {
 641         unsigned long   sb_block;
 642         char            *options = (char *) *data;
 643
 644         if (!options || strncmp(options, "sb=", 3) != 0)
 645                 return 1;       /* Default location */
 646         options += 3;
 647         sb_block = simple_strtoul(options, &options, 0);
 648         if (*options && *options != ',') {
 649                 printk("EXT3-fs: Invalid sb specification: %s\n",
 650                        (char *) *data);
 651                 return 1;
 652         }
 653         if (*options == ',')
 654                 options++;
 655         *data = (void *) options;
 656         return sb_block;
 657 }
 658
 659 static int parse_options (char * options, struct super_block *sb,
 660                           unsigned long * inum, int is_remount)
 661 {
 662         struct ext3_sb_info *sbi = EXT3_SB(sb);
 663         char * p;
 664         substring_t args[MAX_OPT_ARGS];
 665         int data_opt = 0;
 666         int option;
 667 #ifdef CONFIG_QUOTA
 668         int qtype;
 669 #endif
 670
 671         if (!options)
 672                 return 1;
 673
 674         while ((p = strsep (&options, ",")) != NULL) {
 675                 int token;
 676                 if (!*p)
 677                         continue;
 678
 679                 token = match_token(p, tokens, args);
 680                 switch (token) {
 681                 case Opt_bsd_df:
 682                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 683                         break;
 684                 case Opt_minix_df:
 685                         set_opt (sbi->s_mount_opt, MINIX_DF);
 686                         break;
 687                 case Opt_grpid:
 688                         set_opt (sbi->s_mount_opt, GRPID);
 689                         break;
 690                 case Opt_nogrpid:
 691                         clear_opt (sbi->s_mount_opt, GRPID);
 692                         break;
 693                 case Opt_resuid:
 694                         if (match_int(&args[0], &option))
 695                                 return 0;
 696                         sbi->s_resuid = option;
 697                         break;
 698                 case Opt_resgid:
 699                         if (match_int(&args[0], &option))
 700                                 return 0;
 701                         sbi->s_resgid = option;
 702                         break;
 703                 case Opt_sb:
 704                         /* handled by get_sb_block() instead of here */
 705                         /* *sb_block = match_int(&args[0]); */
 706                         break;
 707                 case Opt_err_panic:
 708                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 709                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 710                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 711                         break;
 712                 case Opt_err_ro:
 713                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 714                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 715                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 716                         break;
 717                 case Opt_err_cont:
 718                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 719                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 720                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 721                         break;
 722                 case Opt_nouid32:
 723                         set_opt (sbi->s_mount_opt, NO_UID32);
 724                         break;
 725 #ifndef CONFIG_INOXID_NONE
 726                 case Opt_tagxid:
 727                         set_opt (sbi->s_mount_opt, TAG_XID);
 728                         break;
 729 #endif
 730                 case Opt_check:
 731 #ifdef CONFIG_EXT3_CHECK
 732                         set_opt (sbi->s_mount_opt, CHECK);
 733 #else
 734                         printk(KERN_ERR
 735                                "EXT3 Check option not supported\n");
 736 #endif
 737                         break;
 738                 case Opt_nocheck:
 739                         clear_opt (sbi->s_mount_opt, CHECK);
 740                         break;
 741                 case Opt_debug:
 742                         set_opt (sbi->s_mount_opt, DEBUG);
 743                         break;
 744                 case Opt_oldalloc:
 745                         set_opt (sbi->s_mount_opt, OLDALLOC);
 746                         break;
 747                 case Opt_orlov:
 748                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 749                         break;
 750 #ifdef CONFIG_EXT3_FS_XATTR
 751                 case Opt_user_xattr:
 752                         set_opt (sbi->s_mount_opt, XATTR_USER);
 753                         break;
 754                 case Opt_nouser_xattr:
 755                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 756                         break;
 757 #else
 758                 case Opt_user_xattr:
 759                 case Opt_nouser_xattr:
 760                         printk("EXT3 (no)user_xattr options not supported\n");
 761                         break;
 762 #endif
 763 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 764                 case Opt_acl:
 765                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 766                         break;
 767                 case Opt_noacl:
 768                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 769                         break;
 770 #else
 771                 case Opt_acl:
 772                 case Opt_noacl:
 773                         printk("EXT3 (no)acl options not supported\n");
 774                         break;
 775 #endif
 776                 case Opt_journal_update:
 777                         /* @@@ FIXME */
 778                         /* Eventually we will want to be able to create
 779                            a journal file here.  For now, only allow the
 780                            user to specify an existing inode to be the
 781                            journal file. */
 782                         if (is_remount) {
 783                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 784                                        "journal on remount\n");
 785                                 return 0;
 786                         }
 787                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 788                         break;
 789                 case Opt_journal_inum:
 790                         if (is_remount) {
 791                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 792                                        "journal on remount\n");
 793                                 return 0;
 794                         }
 795                         if (match_int(&args[0], &option))
 796                                 return 0;
 797                         *inum = option;
 798                         break;
 799                 case Opt_noload:
 800                         set_opt (sbi->s_mount_opt, NOLOAD);
 801                         break;
 802                 case Opt_commit:
 803                         if (match_int(&args[0], &option))
 804                                 return 0;
 805                         if (option < 0)
 806                                 return 0;
 807                         if (option == 0)
 808                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 809                         sbi->s_commit_interval = HZ * option;
 810                         break;
 811                 case Opt_data_journal:
 812                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 813                         goto datacheck;
 814                 case Opt_data_ordered:
 815                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 816                         goto datacheck;
 817                 case Opt_data_writeback:
 818                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 819                 datacheck:
 820                         if (is_remount) {
 821                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 822                                                 != data_opt) {
 823                                         printk(KERN_ERR
 824                                                 "EXT3-fs: cannot change data "
 825                                                 "mode on remount\n");
 826                                         return 0;
 827                                 }
 828                         } else {
 829                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 830                                 sbi->s_mount_opt |= data_opt;
 831                         }
 832                         break;
 833 #ifdef CONFIG_QUOTA
 834                 case Opt_usrjquota:
 835                         qtype = USRQUOTA;
 836                         goto set_qf_name;
 837                 case Opt_grpjquota:
 838                         qtype = GRPQUOTA;
 839 set_qf_name:
 840                         if (sb_any_quota_enabled(sb)) {
 841                                 printk(KERN_ERR
 842                                         "EXT3-fs: Cannot change journalled "
 843                                         "quota options when quota turned on.\n");
 844                                 return 0;
 845                         }
 846                         if (sbi->s_qf_names[qtype]) {
 847                                 printk(KERN_ERR
 848                                         "EXT3-fs: %s quota file already "
 849                                         "specified.\n", QTYPE2NAME(qtype));
 850                                 return 0;
 851                         }
 852                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 853                         if (!sbi->s_qf_names[qtype]) {
 854                                 printk(KERN_ERR
 855                                         "EXT3-fs: not enough memory for "
 856                                         "storing quotafile name.\n");
 857                                 return 0;
 858                         }
 859                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 860                                 printk(KERN_ERR
 861                                         "EXT3-fs: quotafile must be on "
 862                                         "filesystem root.\n");
 863                                 kfree(sbi->s_qf_names[qtype]);
 864                                 sbi->s_qf_names[qtype] = NULL;
 865                                 return 0;
 866                         }
 867                         break;
 868                 case Opt_offusrjquota:
 869                         qtype = USRQUOTA;
 870                         goto clear_qf_name;
 871                 case Opt_offgrpjquota:
 872                         qtype = GRPQUOTA;
 873 clear_qf_name:
 874                         if (sb_any_quota_enabled(sb)) {
 875                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 876                                         "journalled quota options when "
 877                                         "quota turned on.\n");
 878                                 return 0;
 879                         }
 880                         if (sbi->s_qf_names[qtype]) {
 881                                 kfree(sbi->s_qf_names[qtype]);
 882                                 sbi->s_qf_names[qtype] = NULL;
 883                         }
 884                         break;
 885                 case Opt_jqfmt_vfsold:
 886                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 887                         break;
 888                 case Opt_jqfmt_vfsv0:
 889                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 890                         break;
 891 #else
 892                 case Opt_usrjquota:
 893                 case Opt_grpjquota:
 894                 case Opt_offusrjquota:
 895                 case Opt_offgrpjquota:
 896                 case Opt_jqfmt_vfsold:
 897                 case Opt_jqfmt_vfsv0:
 898                         printk(KERN_ERR
 899                                 "EXT3-fs: journalled quota options not "
 900                                 "supported.\n");
 901                         break;
 902 #endif
 903                 case Opt_abort:
 904                         set_opt(sbi->s_mount_opt, ABORT);
 905                         break;
 906                 case Opt_ignore:
 907                         break;
 908                 default:
 909                         printk (KERN_ERR
 910                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 911                                 "or missing value\n", p);
 912                         return 0;
 913                 }
 914         }
 915 #ifdef CONFIG_QUOTA
 916         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 917             sbi->s_qf_names[GRPQUOTA])) {
 918                 printk(KERN_ERR
 919                         "EXT3-fs: journalled quota format not specified.\n");
 920                 return 0;
 921         }
 922 #endif
 923
 924         return 1;
 925 }
 926
 927 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 928                             int read_only)
 929 {
 930         struct ext3_sb_info *sbi = EXT3_SB(sb);
 931         int res = 0;
 932
 933         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 934                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 935                         "forcing read-only mode\n");
 936                 res = MS_RDONLY;
 937         }
 938         if (read_only)
 939                 return res;
 940         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 941                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 942                         "running e2fsck is recommended\n");
 943         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 944                 printk (KERN_WARNING
 945                         "EXT3-fs warning: mounting fs with errors, "
 946                         "running e2fsck is recommended\n");
 947         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 948                  le16_to_cpu(es->s_mnt_count) >=
 949                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 950                 printk (KERN_WARNING
 951                         "EXT3-fs warning: maximal mount count reached, "
 952                         "running e2fsck is recommended\n");
 953         else if (le32_to_cpu(es->s_checkinterval) &&
 954                 (le32_to_cpu(es->s_lastcheck) +
 955                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 956                 printk (KERN_WARNING
 957                         "EXT3-fs warning: checktime reached, "
 958                         "running e2fsck is recommended\n");
 959 #if 0
 960                 /* @@@ We _will_ want to clear the valid bit if we find
 961                    inconsistencies, to force a fsck at reboot.  But for
 962                    a plain journaled filesystem we can keep it set as
 963                    valid forever! :) */
 964         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 965 #endif
 966         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 967                 es->s_max_mnt_count =
 968                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 969         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 970         es->s_mtime = cpu_to_le32(get_seconds());
 971         ext3_update_dynamic_rev(sb);
 972         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 973
 974         ext3_commit_super(sb, es, 1);
 975         if (test_opt(sb, DEBUG))
 976                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 977                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 978                         sb->s_blocksize,
 979                         sbi->s_groups_count,
 980                         EXT3_BLOCKS_PER_GROUP(sb),
 981                         EXT3_INODES_PER_GROUP(sb),
 982                         sbi->s_mount_opt);
 983
 984         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 985         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 986                 char b[BDEVNAME_SIZE];
 987
 988                 printk("external journal on %s\n",
 989                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
 990         } else {
 991                 printk("internal journal\n");
 992         }
 993 #ifdef CONFIG_EXT3_CHECK
 994         if (test_opt (sb, CHECK)) {
 995                 ext3_check_blocks_bitmap (sb);
 996                 ext3_check_inodes_bitmap (sb);
 997         }
 998 #endif
 999         return res;
1000 }
1001
1002 static int ext3_check_descriptors (struct super_block * sb)
1003 {
1004         struct ext3_sb_info *sbi = EXT3_SB(sb);
1005         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1006         struct ext3_group_desc * gdp = NULL;
1007         int desc_block = 0;
1008         int i;
1009
1010         ext3_debug ("Checking group descriptors");
1011
1012         for (i = 0; i < sbi->s_groups_count; i++)
1013         {
1014                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1015                         gdp = (struct ext3_group_desc *)
1016                                         sbi->s_group_desc[desc_block++]->b_data;
1017                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1018                     le32_to_cpu(gdp->bg_block_bitmap) >=
1019                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1020                 {
1021                         ext3_error (sb, "ext3_check_descriptors",
1022                                     "Block bitmap for group %d"
1023                                     " not in group (block %lu)!",
1024                                     i, (unsigned long)
1025                                         le32_to_cpu(gdp->bg_block_bitmap));
1026                         return 0;
1027                 }
1028                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1029                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1030                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1031                 {
1032                         ext3_error (sb, "ext3_check_descriptors",
1033                                     "Inode bitmap for group %d"
1034                                     " not in group (block %lu)!",
1035                                     i, (unsigned long)
1036                                         le32_to_cpu(gdp->bg_inode_bitmap));
1037                         return 0;
1038                 }
1039                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1040                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1041                     block + EXT3_BLOCKS_PER_GROUP(sb))
1042                 {
1043                         ext3_error (sb, "ext3_check_descriptors",
1044                                     "Inode table for group %d"
1045                                     " not in group (block %lu)!",
1046                                     i, (unsigned long)
1047                                         le32_to_cpu(gdp->bg_inode_table));
1048                         return 0;
1049                 }
1050                 block += EXT3_BLOCKS_PER_GROUP(sb);
1051                 gdp++;
1052         }
1053
1054         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1055         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1056         return 1;
1057 }
1058
1059
1060 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1061  * the superblock) which were deleted from all directories, but held open by
1062  * a process at the time of a crash.  We walk the list and try to delete these
1063  * inodes at recovery time (only with a read-write filesystem).
1064  *
1065  * In order to keep the orphan inode chain consistent during traversal (in
1066  * case of crash during recovery), we link each inode into the superblock
1067  * orphan list_head and handle it the same way as an inode deletion during
1068  * normal operation (which journals the operations for us).
1069  *
1070  * We only do an iget() and an iput() on each inode, which is very safe if we
1071  * accidentally point at an in-use or already deleted inode.  The worst that
1072  * can happen in this case is that we get a "bit already cleared" message from
1073  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1074  * e2fsck was run on this filesystem, and it must have already done the orphan
1075  * inode cleanup for us, so we can safely abort without any further action.
1076  */
1077 static void ext3_orphan_cleanup (struct super_block * sb,
1078                                  struct ext3_super_block * es)
1079 {
1080         unsigned int s_flags = sb->s_flags;
1081         int nr_orphans = 0, nr_truncates = 0;
1082 #ifdef CONFIG_QUOTA
1083         int i;
1084 #endif
1085         if (!es->s_last_orphan) {
1086                 jbd_debug(4, "no orphan inodes to clean up\n");
1087                 return;
1088         }
1089
1090         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1091                 if (es->s_last_orphan)
1092                         jbd_debug(1, "Errors on filesystem, "
1093                                   "clearing orphan list.\n");
1094                 es->s_last_orphan = 0;
1095                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1096                 return;
1097         }
1098
1099         if (s_flags & MS_RDONLY) {
1100                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1101                        sb->s_id);
1102                 sb->s_flags &= ~MS_RDONLY;
1103         }
1104 #ifdef CONFIG_QUOTA
1105         /* Needed for iput() to work correctly and not trash data */
1106         sb->s_flags |= MS_ACTIVE;
1107         /* Turn on quotas so that they are updated correctly */
1108         for (i = 0; i < MAXQUOTAS; i++) {
1109                 if (EXT3_SB(sb)->s_qf_names[i]) {
1110                         int ret = ext3_quota_on_mount(sb, i);
1111                         if (ret < 0)
1112                                 printk(KERN_ERR
1113                                         "EXT3-fs: Cannot turn on journalled "
1114                                         "quota: error %d\n", ret);
1115                 }
1116         }
1117 #endif
1118
1119         while (es->s_last_orphan) {
1120                 struct inode *inode;
1121
1122                 if (!(inode =
1123                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1124                         es->s_last_orphan = 0;
1125                         break;
1126                 }
1127
1128                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1129                 DQUOT_INIT(inode);
1130                 if (inode->i_nlink) {
1131                         printk(KERN_DEBUG
1132                                 "%s: truncating inode %ld to %Ld bytes\n",
1133                                 __FUNCTION__, inode->i_ino, inode->i_size);
1134                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1135                                   inode->i_ino, inode->i_size);
1136                         ext3_truncate(inode);
1137                         nr_truncates++;
1138                 } else {
1139                         printk(KERN_DEBUG
1140                                 "%s: deleting unreferenced inode %ld\n",
1141                                 __FUNCTION__, inode->i_ino);
1142                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1143                                   inode->i_ino);
1144                         nr_orphans++;
1145                 }
1146                 iput(inode);  /* The delete magic happens here! */
1147         }
1148
1149 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1150
1151         if (nr_orphans)
1152                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1153                        sb->s_id, PLURAL(nr_orphans));
1154         if (nr_truncates)
1155                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1156                        sb->s_id, PLURAL(nr_truncates));
1157 #ifdef CONFIG_QUOTA
1158         /* Turn quotas off */
1159         for (i = 0; i < MAXQUOTAS; i++) {
1160                 if (sb_dqopt(sb)->files[i])
1161                         ext3_quota_off_mount(sb, i);
1162         }
1163 #endif
1164         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1165 }
1166
1167 #define log2(n) ffz(~(n))
1168
1169 /*
1170  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1171  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1172  * We need to be 1 filesystem block less than the 2^32 sector limit.
1173  */
1174 static loff_t ext3_max_size(int bits)
1175 {
1176         loff_t res = EXT3_NDIR_BLOCKS;
1177         res += 1LL << (bits-2);
1178         res += 1LL << (2*(bits-2));
1179         res += 1LL << (3*(bits-2));
1180         res <<= bits;
1181         if (res > (512LL << 32) - (1 << bits))
1182                 res = (512LL << 32) - (1 << bits);
1183         return res;
1184 }
1185
1186 static unsigned long descriptor_loc(struct super_block *sb,
1187                                     unsigned long logic_sb_block,
1188                                     int nr)
1189 {
1190         struct ext3_sb_info *sbi = EXT3_SB(sb);
1191         unsigned long bg, first_data_block, first_meta_bg;
1192         int has_super = 0;
1193
1194         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1195         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1196
1197         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1198             nr < first_meta_bg)
1199                 return (logic_sb_block + nr + 1);
1200         bg = sbi->s_desc_per_block * nr;
1201         if (ext3_bg_has_super(sb, bg))
1202                 has_super = 1;
1203         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1204 }
1205
1206
1207 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1208 {
1209         struct buffer_head * bh;
1210         struct ext3_super_block *es = 0;
1211         struct ext3_sb_info *sbi;
1212         unsigned long block;
1213         unsigned long sb_block = get_sb_block(&data);
1214         unsigned long logic_sb_block;
1215         unsigned long offset = 0;
1216         unsigned long journal_inum = 0;
1217         unsigned long def_mount_opts;
1218         struct inode *root;
1219         int blocksize;
1220         int hblock;
1221         int db_count;
1222         int i;
1223         int needs_recovery;
1224
1225         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1226         if (!sbi)
1227                 return -ENOMEM;
1228         sb->s_fs_info = sbi;
1229         memset(sbi, 0, sizeof(*sbi));
1230         sbi->s_mount_opt = 0;
1231         sbi->s_resuid = EXT3_DEF_RESUID;
1232         sbi->s_resgid = EXT3_DEF_RESGID;
1233
1234         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1235         if (!blocksize) {
1236                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1237                 goto out_fail;
1238         }
1239
1240         /*
1241          * The ext3 superblock will not be buffer aligned for other than 1kB
1242          * block sizes.  We need to calculate the offset from buffer start.
1243          */
1244         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1245                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1246                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1247         } else {
1248                 logic_sb_block = sb_block;
1249         }
1250
1251         if (!(bh = sb_bread(sb, logic_sb_block))) {
1252                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1253                 goto out_fail;
1254         }
1255         /*
1256          * Note: s_es must be initialized as soon as possible because
1257          *       some ext3 macro-instructions depend on its value
1258          */
1259         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1260         sbi->s_es = es;
1261         sb->s_magic = le16_to_cpu(es->s_magic);
1262         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1263                 if (!silent)
1264                         printk(KERN_ERR
1265                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1266                                sb->s_id);
1267                 goto failed_mount;
1268         }
1269
1270         /* Set defaults before we parse the mount options */
1271         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1272         if (def_mount_opts & EXT3_DEFM_DEBUG)
1273                 set_opt(sbi->s_mount_opt, DEBUG);
1274         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1275                 set_opt(sbi->s_mount_opt, GRPID);
1276         if (def_mount_opts & EXT3_DEFM_UID16)
1277                 set_opt(sbi->s_mount_opt, NO_UID32);
1278         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1279                 set_opt(sbi->s_mount_opt, XATTR_USER);
1280         if (def_mount_opts & EXT3_DEFM_ACL)
1281                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1282         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1283                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1284         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1285                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1286         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1287                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1288
1289         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1290                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1291         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1292                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1293
1294         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1295         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1296
1297         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1298                 goto failed_mount;
1299
1300         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1301                 sb->s_flags |= MS_TAGXID;
1302         sb->s_flags |= MS_ONE_SECOND;
1303         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1304                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1305
1306         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1307             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1308              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1309              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1310                 printk(KERN_WARNING
1311                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1312                        "running e2fsck is recommended\n");
1313         /*
1314          * Check feature flags regardless of the revision level, since we
1315          * previously didn't change the revision level when setting the flags,
1316          * so there is a chance incompat flags are set on a rev 0 filesystem.
1317          */
1318         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1319                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1320                        "unsupported optional features (%x).\n",
1321                        sb->s_id, i);
1322                 goto failed_mount;
1323         }
1324         if (!(sb->s_flags & MS_RDONLY) &&
1325             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1326                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1327                        "unsupported optional features (%x).\n",
1328                        sb->s_id, i);
1329                 goto failed_mount;
1330         }
1331         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1332
1333         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1334             blocksize > EXT3_MAX_BLOCK_SIZE) {
1335                 printk(KERN_ERR
1336                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1337                        blocksize, sb->s_id);
1338                 goto failed_mount;
1339         }
1340
1341         hblock = bdev_hardsect_size(sb->s_bdev);
1342         if (sb->s_blocksize != blocksize) {
1343                 /*
1344                  * Make sure the blocksize for the filesystem is larger
1345                  * than the hardware sectorsize for the machine.
1346                  */
1347                 if (blocksize < hblock) {
1348                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1349                                "device blocksize %d.\n", blocksize, hblock);
1350                         goto failed_mount;
1351                 }
1352
1353                 brelse (bh);
1354                 sb_set_blocksize(sb, blocksize);
1355                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1356                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1357                 bh = sb_bread(sb, logic_sb_block);
1358                 if (!bh) {
1359                         printk(KERN_ERR
1360                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1361                         goto failed_mount;
1362                 }
1363                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1364                 sbi->s_es = es;
1365                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1366                         printk (KERN_ERR
1367                                 "EXT3-fs: Magic mismatch, very weird !\n");
1368                         goto failed_mount;
1369                 }
1370         }
1371
1372         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1373
1374         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1375                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1376                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1377         } else {
1378                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1379                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1380                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1381                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1382                     (sbi->s_inode_size > blocksize)) {
1383                         printk (KERN_ERR
1384                                 "EXT3-fs: unsupported inode size: %d\n",
1385                                 sbi->s_inode_size);
1386                         goto failed_mount;
1387                 }
1388         }
1389         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1390                                    le32_to_cpu(es->s_log_frag_size);
1391         if (blocksize != sbi->s_frag_size) {
1392                 printk(KERN_ERR
1393                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1394                        sbi->s_frag_size, blocksize);
1395                 goto failed_mount;
1396         }
1397         sbi->s_frags_per_block = 1;
1398         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1399         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1400         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1401         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1402         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1403         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1404         sbi->s_sbh = bh;
1405         sbi->s_mount_state = le16_to_cpu(es->s_state);
1406         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1407         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1408         for (i=0; i < 4; i++)
1409                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1410         sbi->s_def_hash_version = es->s_def_hash_version;
1411
1412         if (sbi->s_blocks_per_group > blocksize * 8) {
1413                 printk (KERN_ERR
1414                         "EXT3-fs: #blocks per group too big: %lu\n",
1415                         sbi->s_blocks_per_group);
1416                 goto failed_mount;
1417         }
1418         if (sbi->s_frags_per_group > blocksize * 8) {
1419                 printk (KERN_ERR
1420                         "EXT3-fs: #fragments per group too big: %lu\n",
1421                         sbi->s_frags_per_group);
1422                 goto failed_mount;
1423         }
1424         if (sbi->s_inodes_per_group > blocksize * 8) {
1425                 printk (KERN_ERR
1426                         "EXT3-fs: #inodes per group too big: %lu\n",
1427                         sbi->s_inodes_per_group);
1428                 goto failed_mount;
1429         }
1430
1431         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1432                                le32_to_cpu(es->s_first_data_block) +
1433                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1434                               EXT3_BLOCKS_PER_GROUP(sb);
1435         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1436                    EXT3_DESC_PER_BLOCK(sb);
1437         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1438                                     GFP_KERNEL);
1439         if (sbi->s_group_desc == NULL) {
1440                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1441                 goto failed_mount;
1442         }
1443         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1444                         GFP_KERNEL);
1445         if (!sbi->s_debts) {
1446                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1447                 goto failed_mount2;
1448         }
1449         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1450
1451         percpu_counter_init(&sbi->s_freeblocks_counter);
1452         percpu_counter_init(&sbi->s_freeinodes_counter);
1453         percpu_counter_init(&sbi->s_dirs_counter);
1454         bgl_lock_init(&sbi->s_blockgroup_lock);
1455
1456         for (i = 0; i < db_count; i++) {
1457                 block = descriptor_loc(sb, logic_sb_block, i);
1458                 sbi->s_group_desc[i] = sb_bread(sb, block);
1459                 if (!sbi->s_group_desc[i]) {
1460                         printk (KERN_ERR "EXT3-fs: "
1461                                 "can't read group descriptor %d\n", i);
1462                         db_count = i;
1463                         goto failed_mount2;
1464                 }
1465         }
1466         if (!ext3_check_descriptors (sb)) {
1467                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1468                 goto failed_mount2;
1469         }
1470         sbi->s_gdb_count = db_count;
1471         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1472         spin_lock_init(&sbi->s_next_gen_lock);
1473         /*
1474          * set up enough so that it can read an inode
1475          */
1476         sb->s_op = &ext3_sops;
1477         sb->s_export_op = &ext3_export_ops;
1478 #ifdef CONFIG_QUOTA
1479         sb->s_qcop = &ext3_qctl_operations;
1480         sb->dq_op = &ext3_quota_operations;
1481 #endif
1482         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1483
1484         sb->s_root = 0;
1485
1486         needs_recovery = (es->s_last_orphan != 0 ||
1487                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1488                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1489
1490         /*
1491          * The first inode we look at is the journal inode.  Don't try
1492          * root first: it may be modified in the journal!
1493          */
1494         if (!test_opt(sb, NOLOAD) &&
1495             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1496                 if (ext3_load_journal(sb, es))
1497                         goto failed_mount2;
1498         } else if (journal_inum) {
1499                 if (ext3_create_journal(sb, es, journal_inum))
1500                         goto failed_mount2;
1501         } else {
1502                 if (!silent)
1503                         printk (KERN_ERR
1504                                 "ext3: No journal on filesystem on %s\n",
1505                                 sb->s_id);
1506                 goto failed_mount2;
1507         }
1508
1509         /* We have now updated the journal if required, so we can
1510          * validate the data journaling mode. */
1511         switch (test_opt(sb, DATA_FLAGS)) {
1512         case 0:
1513                 /* No mode set, assume a default based on the journal
1514                    capabilities: ORDERED_DATA if the journal can
1515                    cope, else JOURNAL_DATA */
1516                 if (journal_check_available_features
1517                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1518                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1519                 else
1520                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1521                 break;
1522
1523         case EXT3_MOUNT_ORDERED_DATA:
1524         case EXT3_MOUNT_WRITEBACK_DATA:
1525                 if (!journal_check_available_features
1526                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1527                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1528                                "requested data journaling mode\n");
1529                         goto failed_mount3;
1530                 }
1531         default:
1532                 break;
1533         }
1534
1535         /*
1536          * The journal_load will have done any necessary log recovery,
1537          * so we can safely mount the rest of the filesystem now.
1538          */
1539
1540         root = iget(sb, EXT3_ROOT_INO);
1541         sb->s_root = d_alloc_root(root);
1542         if (!sb->s_root) {
1543                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1544                 iput(root);
1545                 goto failed_mount3;
1546         }
1547         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1548                 dput(sb->s_root);
1549                 sb->s_root = NULL;
1550                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1551                 goto failed_mount3;
1552         }
1553
1554         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1555         /*
1556          * akpm: core read_super() calls in here with the superblock locked.
1557          * That deadlocks, because orphan cleanup needs to lock the superblock
1558          * in numerous places.  Here we just pop the lock - it's relatively
1559          * harmless, because we are now ready to accept write_super() requests,
1560          * and aviro says that's the only reason for hanging onto the
1561          * superblock lock.
1562          */
1563         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1564         ext3_orphan_cleanup(sb, es);
1565         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1566         if (needs_recovery)
1567                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1568         ext3_mark_recovery_complete(sb, es);
1569         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1570                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1571                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1572                 "writeback");
1573
1574         percpu_counter_mod(&sbi->s_freeblocks_counter,
1575                 ext3_count_free_blocks(sb));
1576         percpu_counter_mod(&sbi->s_freeinodes_counter,
1577                 ext3_count_free_inodes(sb));
1578         percpu_counter_mod(&sbi->s_dirs_counter,
1579                 ext3_count_dirs(sb));
1580
1581         return 0;
1582
1583 failed_mount3:
1584         journal_destroy(sbi->s_journal);
1585 failed_mount2:
1586         kfree(sbi->s_debts);
1587         for (i = 0; i < db_count; i++)
1588                 brelse(sbi->s_group_desc[i]);
1589         kfree(sbi->s_group_desc);
1590 failed_mount:
1591 #ifdef CONFIG_QUOTA
1592         for (i = 0; i < MAXQUOTAS; i++) {
1593                 if (sbi->s_qf_names[i])
1594                         kfree(sbi->s_qf_names[i]);
1595         }
1596 #endif
1597         ext3_blkdev_remove(sbi);
1598         brelse(bh);
1599 out_fail:
1600         sb->s_fs_info = NULL;
1601         kfree(sbi);
1602         return -EINVAL;
1603 }
1604
1605 /*
1606  * Setup any per-fs journal parameters now.  We'll do this both on
1607  * initial mount, once the journal has been initialised but before we've
1608  * done any recovery; and again on any subsequent remount.
1609  */
1610 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1611                                      journal_t *journal)
1612 {
1613         if (sbi->s_commit_interval)
1614                 journal->j_commit_interval = sbi->s_commit_interval;
1615         /* We could also set up an ext3-specific default for the commit
1616          * interval here, but for now we'll just fall back to the jbd
1617          * default. */
1618 }
1619
1620
1621 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1622 {
1623         struct inode *journal_inode;
1624         journal_t *journal;
1625
1626         /* First, test for the existence of a valid inode on disk.  Bad
1627          * things happen if we iget() an unused inode, as the subsequent
1628          * iput() will try to delete it. */
1629
1630         journal_inode = iget(sb, journal_inum);
1631         if (!journal_inode) {
1632                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1633                 return NULL;
1634         }
1635         if (!journal_inode->i_nlink) {
1636                 make_bad_inode(journal_inode);
1637                 iput(journal_inode);
1638                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1639                 return NULL;
1640         }
1641
1642         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1643                   journal_inode, journal_inode->i_size);
1644         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1645                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1646                 iput(journal_inode);
1647                 return NULL;
1648         }
1649
1650         journal = journal_init_inode(journal_inode);
1651         if (!journal) {
1652                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1653                 iput(journal_inode);
1654                 return NULL;
1655         }
1656         journal->j_private = sb;
1657         ext3_init_journal_params(EXT3_SB(sb), journal);
1658         return journal;
1659 }
1660
1661 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1662                                        dev_t j_dev)
1663 {
1664         struct buffer_head * bh;
1665         journal_t *journal;
1666         int start;
1667         int len;
1668         int hblock, blocksize;
1669         unsigned long sb_block;
1670         unsigned long offset;
1671         struct ext3_super_block * es;
1672         struct block_device *bdev;
1673
1674         bdev = ext3_blkdev_get(j_dev);
1675         if (bdev == NULL)
1676                 return NULL;
1677
1678         if (bd_claim(bdev, sb)) {
1679                 printk(KERN_ERR
1680                         "EXT3: failed to claim external journal device.\n");
1681                 blkdev_put(bdev);
1682                 return NULL;
1683         }
1684
1685         blocksize = sb->s_blocksize;
1686         hblock = bdev_hardsect_size(bdev);
1687         if (blocksize < hblock) {
1688                 printk(KERN_ERR
1689                         "EXT3-fs: blocksize too small for journal device.\n");
1690                 goto out_bdev;
1691         }
1692
1693         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1694         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1695         set_blocksize(bdev, blocksize);
1696         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1697                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1698                        "external journal\n");
1699                 goto out_bdev;
1700         }
1701
1702         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1703         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1704             !(le32_to_cpu(es->s_feature_incompat) &
1705               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1706                 printk(KERN_ERR "EXT3-fs: external journal has "
1707                                         "bad superblock\n");
1708                 brelse(bh);
1709                 goto out_bdev;
1710         }
1711
1712         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1713                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1714                 brelse(bh);
1715                 goto out_bdev;
1716         }
1717
1718         len = le32_to_cpu(es->s_blocks_count);
1719         start = sb_block + 1;
1720         brelse(bh);     /* we're done with the superblock */
1721
1722         journal = journal_init_dev(bdev, sb->s_bdev,
1723                                         start, len, blocksize);
1724         if (!journal) {
1725                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1726                 goto out_bdev;
1727         }
1728         journal->j_private = sb;
1729         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1730         wait_on_buffer(journal->j_sb_buffer);
1731         if (!buffer_uptodate(journal->j_sb_buffer)) {
1732                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1733                 goto out_journal;
1734         }
1735         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1736                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1737                                         "user (unsupported) - %d\n",
1738                         ntohl(journal->j_superblock->s_nr_users));
1739                 goto out_journal;
1740         }
1741         EXT3_SB(sb)->journal_bdev = bdev;
1742         ext3_init_journal_params(EXT3_SB(sb), journal);
1743         return journal;
1744 out_journal:
1745         journal_destroy(journal);
1746 out_bdev:
1747         ext3_blkdev_put(bdev);
1748         return NULL;
1749 }
1750
1751 static int ext3_load_journal(struct super_block * sb,
1752                              struct ext3_super_block * es)
1753 {
1754         journal_t *journal;
1755         int journal_inum = le32_to_cpu(es->s_journal_inum);
1756         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1757         int err = 0;
1758         int really_read_only;
1759
1760         really_read_only = bdev_read_only(sb->s_bdev);
1761
1762         /*
1763          * Are we loading a blank journal or performing recovery after a
1764          * crash?  For recovery, we need to check in advance whether we
1765          * can get read-write access to the device.
1766          */
1767
1768         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1769                 if (sb->s_flags & MS_RDONLY) {
1770                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1771                                         "required on readonly filesystem.\n");
1772                         if (really_read_only) {
1773                                 printk(KERN_ERR "EXT3-fs: write access "
1774                                         "unavailable, cannot proceed.\n");
1775                                 return -EROFS;
1776                         }
1777                         printk (KERN_INFO "EXT3-fs: write access will "
1778                                         "be enabled during recovery.\n");
1779                 }
1780         }
1781
1782         if (journal_inum && journal_dev) {
1783                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1784                        "and inode journals!\n");
1785                 return -EINVAL;
1786         }
1787
1788         if (journal_inum) {
1789                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1790                         return -EINVAL;
1791         } else {
1792                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1793                         return -EINVAL;
1794         }
1795
1796         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1797                 err = journal_update_format(journal);
1798                 if (err)  {
1799                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1800                         journal_destroy(journal);
1801                         return err;
1802                 }
1803         }
1804
1805         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1806                 err = journal_wipe(journal, !really_read_only);
1807         if (!err)
1808                 err = journal_load(journal);
1809
1810         if (err) {
1811                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1812                 journal_destroy(journal);
1813                 return err;
1814         }
1815
1816         EXT3_SB(sb)->s_journal = journal;
1817         ext3_clear_journal_err(sb, es);
1818         return 0;
1819 }
1820
1821 static int ext3_create_journal(struct super_block * sb,
1822                                struct ext3_super_block * es,
1823                                int journal_inum)
1824 {
1825         journal_t *journal;
1826
1827         if (sb->s_flags & MS_RDONLY) {
1828                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1829                                 "create journal.\n");
1830                 return -EROFS;
1831         }
1832
1833         if (!(journal = ext3_get_journal(sb, journal_inum)))
1834                 return -EINVAL;
1835
1836         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1837                journal_inum);
1838
1839         if (journal_create(journal)) {
1840                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1841                 journal_destroy(journal);
1842                 return -EIO;
1843         }
1844
1845         EXT3_SB(sb)->s_journal = journal;
1846
1847         ext3_update_dynamic_rev(sb);
1848         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1849         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1850
1851         es->s_journal_inum = cpu_to_le32(journal_inum);
1852         sb->s_dirt = 1;
1853
1854         /* Make sure we flush the recovery flag to disk. */
1855         ext3_commit_super(sb, es, 1);
1856
1857         return 0;
1858 }
1859
1860 static void ext3_commit_super (struct super_block * sb,
1861                                struct ext3_super_block * es,
1862                                int sync)
1863 {
1864         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1865
1866         if (!sbh)
1867                 return;
1868         es->s_wtime = cpu_to_le32(get_seconds());
1869         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1870         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1871         BUFFER_TRACE(sbh, "marking dirty");
1872         mark_buffer_dirty(sbh);
1873         if (sync)
1874                 sync_dirty_buffer(sbh);
1875 }
1876
1877
1878 /*
1879  * Have we just finished recovery?  If so, and if we are mounting (or
1880  * remounting) the filesystem readonly, then we will end up with a
1881  * consistent fs on disk.  Record that fact.
1882  */
1883 static void ext3_mark_recovery_complete(struct super_block * sb,
1884                                         struct ext3_super_block * es)
1885 {
1886         journal_t *journal = EXT3_SB(sb)->s_journal;
1887
1888         journal_lock_updates(journal);
1889         journal_flush(journal);
1890         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1891             sb->s_flags & MS_RDONLY) {
1892                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1893                 sb->s_dirt = 0;
1894                 ext3_commit_super(sb, es, 1);
1895         }
1896         journal_unlock_updates(journal);
1897 }
1898
1899 /*
1900  * If we are mounting (or read-write remounting) a filesystem whose journal
1901  * has recorded an error from a previous lifetime, move that error to the
1902  * main filesystem now.
1903  */
1904 static void ext3_clear_journal_err(struct super_block * sb,
1905                                    struct ext3_super_block * es)
1906 {
1907         journal_t *journal;
1908         int j_errno;
1909         const char *errstr;
1910
1911         journal = EXT3_SB(sb)->s_journal;
1912
1913         /*
1914          * Now check for any error status which may have been recorded in the
1915          * journal by a prior ext3_error() or ext3_abort()
1916          */
1917
1918         j_errno = journal_errno(journal);
1919         if (j_errno) {
1920                 char nbuf[16];
1921
1922                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1923                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1924                              "from previous mount: %s", errstr);
1925                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1926                              "filesystem check.");
1927
1928                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1929                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1930                 ext3_commit_super (sb, es, 1);
1931
1932                 journal_clear_err(journal);
1933         }
1934 }
1935
1936 /*
1937  * Force the running and committing transactions to commit,
1938  * and wait on the commit.
1939  */
1940 int ext3_force_commit(struct super_block *sb)
1941 {
1942         journal_t *journal;
1943         int ret;
1944
1945         if (sb->s_flags & MS_RDONLY)
1946                 return 0;
1947
1948         journal = EXT3_SB(sb)->s_journal;
1949         sb->s_dirt = 0;
1950         ret = ext3_journal_force_commit(journal);
1951         return ret;
1952 }
1953
1954 /*
1955  * Ext3 always journals updates to the superblock itself, so we don't
1956  * have to propagate any other updates to the superblock on disk at this
1957  * point.  Just start an async writeback to get the buffers on their way
1958  * to the disk.
1959  *
1960  * This implicitly triggers the writebehind on sync().
1961  */
1962
1963 void ext3_write_super (struct super_block * sb)
1964 {
1965         if (down_trylock(&sb->s_lock) == 0)
1966                 BUG();
1967         sb->s_dirt = 0;
1968 }
1969
1970 static int ext3_sync_fs(struct super_block *sb, int wait)
1971 {
1972         tid_t target;
1973
1974         sb->s_dirt = 0;
1975         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1976                 if (wait)
1977                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1978         }
1979         return 0;
1980 }
1981
1982 /*
1983  * LVM calls this function before a (read-only) snapshot is created.  This
1984  * gives us a chance to flush the journal completely and mark the fs clean.
1985  */
1986 void ext3_write_super_lockfs(struct super_block *sb)
1987 {
1988         sb->s_dirt = 0;
1989
1990         if (!(sb->s_flags & MS_RDONLY)) {
1991                 journal_t *journal = EXT3_SB(sb)->s_journal;
1992
1993                 /* Now we set up the journal barrier. */
1994                 journal_lock_updates(journal);
1995                 journal_flush(journal);
1996
1997                 /* Journal blocked and flushed, clear needs_recovery flag. */
1998                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1999                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2000         }
2001 }
2002
2003 /*
2004  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2005  * flag here, even though the filesystem is not technically dirty yet.
2006  */
2007 void ext3_unlockfs(struct super_block *sb)
2008 {
2009         if (!(sb->s_flags & MS_RDONLY)) {
2010                 lock_super(sb);
2011                 /* Reser the needs_recovery flag before the fs is unlocked. */
2012                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2013                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2014                 unlock_super(sb);
2015                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2016         }
2017 }
2018
2019 int ext3_remount (struct super_block * sb, int * flags, char * data)
2020 {
2021         struct ext3_super_block * es;
2022         struct ext3_sb_info *sbi = EXT3_SB(sb);
2023         unsigned long tmp;
2024
2025         /*
2026          * Allow the "check" option to be passed as a remount option.
2027          */
2028         if (!parse_options(data, sb, &tmp, 1))
2029                 return -EINVAL;
2030
2031         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2032                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2033
2034         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2035                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2036
2037         es = sbi->s_es;
2038
2039         ext3_init_journal_params(sbi, sbi->s_journal);
2040
2041         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2042                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2043                         return -EROFS;
2044
2045                 if (*flags & MS_RDONLY) {
2046                         /*
2047                          * First of all, the unconditional stuff we have to do
2048                          * to disable replay of the journal when we next remount
2049                          */
2050                         sb->s_flags |= MS_RDONLY;
2051
2052                         /*
2053                          * OK, test if we are remounting a valid rw partition
2054                          * readonly, and if so set the rdonly flag and then
2055                          * mark the partition as valid again.
2056                          */
2057                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2058                             (sbi->s_mount_state & EXT3_VALID_FS))
2059                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2060
2061                         ext3_mark_recovery_complete(sb, es);
2062                 } else {
2063                         int ret;
2064                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2065                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2066                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2067                                        "remount RDWR because of unsupported "
2068                                        "optional features (%x).\n",
2069                                        sb->s_id, ret);
2070                                 return -EROFS;
2071                         }
2072                         /*
2073                          * Mounting a RDONLY partition read-write, so reread
2074                          * and store the current valid flag.  (It may have
2075                          * been changed by e2fsck since we originally mounted
2076                          * the partition.)
2077                          */
2078                         ext3_clear_journal_err(sb, es);
2079                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2080                         if (!ext3_setup_super (sb, es, 0))
2081                                 sb->s_flags &= ~MS_RDONLY;
2082                 }
2083         }
2084         return 0;
2085 }
2086
2087 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2088 {
2089         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2090         unsigned long overhead;
2091         int i;
2092
2093         if (test_opt (sb, MINIX_DF))
2094                 overhead = 0;
2095         else {
2096                 /*
2097                  * Compute the overhead (FS structures)
2098                  */
2099
2100                 /*
2101                  * All of the blocks before first_data_block are
2102                  * overhead
2103                  */
2104                 overhead = le32_to_cpu(es->s_first_data_block);
2105
2106                 /*
2107                  * Add the overhead attributed to the superblock and
2108                  * block group descriptors.  If the sparse superblocks
2109                  * feature is turned on, then not all groups have this.
2110                  */
2111                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2112                         overhead += ext3_bg_has_super(sb, i) +
2113                                 ext3_bg_num_gdb(sb, i);
2114
2115                 /*
2116                  * Every block group has an inode bitmap, a block
2117                  * bitmap, and an inode table.
2118                  */
2119                 overhead += (EXT3_SB(sb)->s_groups_count *
2120                              (2 + EXT3_SB(sb)->s_itb_per_group));
2121         }
2122
2123         buf->f_type = EXT3_SUPER_MAGIC;
2124         buf->f_bsize = sb->s_blocksize;
2125         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2126         buf->f_bfree = ext3_count_free_blocks (sb);
2127         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2128         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2129                 buf->f_bavail = 0;
2130         buf->f_files = le32_to_cpu(es->s_inodes_count);
2131         buf->f_ffree = ext3_count_free_inodes (sb);
2132         buf->f_namelen = EXT3_NAME_LEN;
2133         return 0;
2134 }
2135
2136 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2137  * is locked for write. Otherwise the are possible deadlocks:
2138  * Process 1                         Process 2
2139  * ext3_create()                     quota_sync()
2140  *   journal_start()                   write_dquot()
2141  *   DQUOT_INIT()                        down(dqio_sem)
2142  *     down(dqio_sem)                    journal_start()
2143  *
2144  */
2145
2146 #ifdef CONFIG_QUOTA
2147
2148 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2149 {
2150         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2151 }
2152
2153 static int ext3_dquot_initialize(struct inode *inode, int type)
2154 {
2155         handle_t *handle;
2156         int ret, err;
2157
2158         /* We may create quota structure so we need to reserve enough blocks */
2159         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2160         if (IS_ERR(handle))
2161                 return PTR_ERR(handle);
2162         ret = dquot_initialize(inode, type);
2163         err = ext3_journal_stop(handle);
2164         if (!ret)
2165                 ret = err;
2166         return ret;
2167 }
2168
2169 static int ext3_dquot_drop(struct inode *inode)
2170 {
2171         handle_t *handle;
2172         int ret, err;
2173
2174         /* We may delete quota structure so we need to reserve enough blocks */
2175         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2176         if (IS_ERR(handle))
2177                 return PTR_ERR(handle);
2178         ret = dquot_drop(inode);
2179         err = ext3_journal_stop(handle);
2180         if (!ret)
2181                 ret = err;
2182         return ret;
2183 }
2184
2185 static int ext3_write_dquot(struct dquot *dquot)
2186 {
2187         int ret, err;
2188         handle_t *handle;
2189
2190         handle = ext3_journal_start(dquot_to_inode(dquot),
2191                                         EXT3_QUOTA_TRANS_BLOCKS);
2192         if (IS_ERR(handle))
2193                 return PTR_ERR(handle);
2194         ret = dquot_commit(dquot);
2195         err = ext3_journal_stop(handle);
2196         if (!ret)
2197                 ret = err;
2198         return ret;
2199 }
2200
2201 static int ext3_acquire_dquot(struct dquot *dquot)
2202 {
2203         int ret, err;
2204         handle_t *handle;
2205
2206         handle = ext3_journal_start(dquot_to_inode(dquot),
2207                                         EXT3_QUOTA_INIT_BLOCKS);
2208         if (IS_ERR(handle))
2209                 return PTR_ERR(handle);
2210         ret = dquot_acquire(dquot);
2211         err = ext3_journal_stop(handle);
2212         if (!ret)
2213                 ret = err;
2214         return ret;
2215 }
2216
2217 static int ext3_release_dquot(struct dquot *dquot)
2218 {
2219         int ret, err;
2220         handle_t *handle;
2221
2222         handle = ext3_journal_start(dquot_to_inode(dquot),
2223                                         EXT3_QUOTA_INIT_BLOCKS);
2224         if (IS_ERR(handle))
2225                 return PTR_ERR(handle);
2226         ret = dquot_release(dquot);
2227         err = ext3_journal_stop(handle);
2228         if (!ret)
2229                 ret = err;
2230         return ret;
2231 }
2232
2233 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2234 {
2235         /* Are we journalling quotas? */
2236         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2237             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2238                 dquot_mark_dquot_dirty(dquot);
2239                 return ext3_write_dquot(dquot);
2240         } else {
2241                 return dquot_mark_dquot_dirty(dquot);
2242         }
2243 }
2244
2245 static int ext3_write_info(struct super_block *sb, int type)
2246 {
2247         int ret, err;
2248         handle_t *handle;
2249
2250         /* Data block + inode block */
2251         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2252         if (IS_ERR(handle))
2253                 return PTR_ERR(handle);
2254         ret = dquot_commit_info(sb, type);
2255         err = ext3_journal_stop(handle);
2256         if (!ret)
2257                 ret = err;
2258         return ret;
2259 }
2260
2261 /*
2262  * Turn on quotas during mount time - we need to find
2263  * the quota file and such...
2264  */
2265 static int ext3_quota_on_mount(struct super_block *sb, int type)
2266 {
2267         int err;
2268         struct dentry *dentry;
2269         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2270                              .hash = 0,
2271                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2272
2273         dentry = lookup_hash(&name, sb->s_root);
2274         if (IS_ERR(dentry))
2275                 return PTR_ERR(dentry);
2276         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2277         if (err)
2278                 dput(dentry);
2279         /* We keep the dentry reference if everything went ok - we drop it
2280          * on quota_off time */
2281         return err;
2282 }
2283
2284 /* Turn quotas off during mount time */
2285 static int ext3_quota_off_mount(struct super_block *sb, int type)
2286 {
2287         int err;
2288         struct dentry *dentry;
2289
2290         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2291         err = vfs_quota_off_mount(sb, type);
2292         /* We invalidate dentry - it has at least wrong hash... */
2293         d_invalidate(dentry);
2294         dput(dentry);
2295         return err;
2296 }
2297
2298 /*
2299  * Standard function to be called on quota_on
2300  */
2301 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2302                          char *path)
2303 {
2304         int err;
2305         struct nameidata nd;
2306
2307         /* Not journalling quota? */
2308         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2309             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2310                 return vfs_quota_on(sb, type, format_id, path);
2311         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2312         if (err)
2313                 return err;
2314         /* Quotafile not on the same filesystem? */
2315         if (nd.mnt->mnt_sb != sb)
2316                 return -EXDEV;
2317         /* Quotafile not of fs root? */
2318         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2319                 printk(KERN_WARNING
2320                         "EXT3-fs: Quota file not on filesystem root. "
2321                         "Journalled quota will not work.\n");
2322         if (!ext3_should_journal_data(nd.dentry->d_inode))
2323                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2324                         "data-journalling. Journalled quota will not work.\n");
2325         path_release(&nd);
2326         return vfs_quota_on(sb, type, format_id, path);
2327 }
2328
2329 #endif
2330
2331 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2332         int flags, const char *dev_name, void *data)
2333 {
2334         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2335 }
2336
2337 static struct file_system_type ext3_fs_type = {
2338         .owner          = THIS_MODULE,
2339         .name           = "ext3",
2340         .get_sb         = ext3_get_sb,
2341         .kill_sb        = kill_block_super,
2342         .fs_flags       = FS_REQUIRES_DEV,
2343 };
2344
2345 static int __init init_ext3_fs(void)
2346 {
2347         int err = init_ext3_xattr();
2348         if (err)
2349                 return err;
2350         err = init_inodecache();
2351         if (err)
2352                 goto out1;
2353         err = register_filesystem(&ext3_fs_type);
2354         if (err)
2355                 goto out;
2356         return 0;
2357 out:
2358         destroy_inodecache();
2359 out1:
2360         exit_ext3_xattr();
2361         return err;
2362 }
2363
2364 static void __exit exit_ext3_fs(void)
2365 {
2366         unregister_filesystem(&ext3_fs_type);
2367         destroy_inodecache();
2368         exit_ext3_xattr();
2369 }
2370
2371 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2372 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2373 MODULE_LICENSE("GPL");
2374 module_init(init_ext3_fs)
2375 module_exit(exit_ext3_fs)