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 /* Deal with the reporting of failure conditions on a filesystem such as
 122  * inconsistencies detected or read IO failures.
 123  *
 124  * On ext2, we can store the error state of the filesystem in the
 125  * superblock.  That is not possible on ext3, because we may have other
 126  * write ordering constraints on the superblock which prevent us from
 127  * writing it out straight away; and given that the journal is about to
 128  * be aborted, we can't rely on the current, or future, transactions to
 129  * write out the superblock safely.
 130  *
 131  * We'll just use the journal_abort() error code to record an error in
 132  * the journal instead.  On recovery, the journal will compain about
 133  * that error until we've noted it down and cleared it.
 134  */
 135
 136 static void ext3_handle_error(struct super_block *sb)
 137 {
 138         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 139
 140         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 141         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 142
 143         if (sb->s_flags & MS_RDONLY)
 144                 return;
 145
 146         if (test_opt (sb, ERRORS_PANIC))
 147                 panic ("EXT3-fs (device %s): panic forced after error\n",
 148                        sb->s_id);
 149         if (test_opt (sb, ERRORS_RO)) {
 150                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 151                 sb->s_flags |= MS_RDONLY;
 152         } else {
 153                 journal_t *journal = EXT3_SB(sb)->s_journal;
 154
 155                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 156                 if (journal)
 157                         journal_abort(journal, -EIO);
 158         }
 159         ext3_commit_super(sb, es, 1);
 160 }
 161
 162 void ext3_error (struct super_block * sb, const char * function,
 163                  const char * fmt, ...)
 164 {
 165         va_list args;
 166
 167         va_start(args, fmt);
 168         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 169         vprintk(fmt, args);
 170         printk("\n");
 171         va_end(args);
 172
 173         ext3_handle_error(sb);
 174 }
 175
 176 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 177 {
 178         char *errstr = NULL;
 179
 180         switch (errno) {
 181         case -EIO:
 182                 errstr = "IO failure";
 183                 break;
 184         case -ENOMEM:
 185                 errstr = "Out of memory";
 186                 break;
 187         case -EROFS:
 188                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 189                         errstr = "Journal has aborted";
 190                 else
 191                         errstr = "Readonly filesystem";
 192                 break;
 193         default:
 194                 /* If the caller passed in an extra buffer for unknown
 195                  * errors, textualise them now.  Else we just return
 196                  * NULL. */
 197                 if (nbuf) {
 198                         /* Check for truncated error codes... */
 199                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 200                                 errstr = nbuf;
 201                 }
 202                 break;
 203         }
 204
 205         return errstr;
 206 }
 207
 208 /* __ext3_std_error decodes expected errors from journaling functions
 209  * automatically and invokes the appropriate error response.  */
 210
 211 void __ext3_std_error (struct super_block * sb, const char * function,
 212                        int errno)
 213 {
 214         char nbuf[16];
 215         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 216
 217         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 218                 sb->s_id, function, errstr);
 219
 220         ext3_handle_error(sb);
 221 }
 222
 223 /*
 224  * ext3_abort is a much stronger failure handler than ext3_error.  The
 225  * abort function may be used to deal with unrecoverable failures such
 226  * as journal IO errors or ENOMEM at a critical moment in log management.
 227  *
 228  * We unconditionally force the filesystem into an ABORT|READONLY state,
 229  * unless the error response on the fs has been set to panic in which
 230  * case we take the easy way out and panic immediately.
 231  */
 232
 233 void ext3_abort (struct super_block * sb, const char * function,
 234                  const char * fmt, ...)
 235 {
 236         va_list args;
 237
 238         printk (KERN_CRIT "ext3_abort called.\n");
 239
 240         va_start(args, fmt);
 241         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 242         vprintk(fmt, args);
 243         printk("\n");
 244         va_end(args);
 245
 246         if (test_opt(sb, ERRORS_PANIC))
 247                 panic("EXT3-fs panic from previous error\n");
 248
 249         if (sb->s_flags & MS_RDONLY)
 250                 return;
 251
 252         printk(KERN_CRIT "Remounting filesystem read-only\n");
 253         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 254         sb->s_flags |= MS_RDONLY;
 255         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 256         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 257 }
 258
 259 /* Deal with the reporting of failure conditions while running, such as
 260  * inconsistencies in operation or invalid system states.
 261  *
 262  * Use ext3_error() for cases of invalid filesystem states, as that will
 263  * record an error on disk and force a filesystem check on the next boot.
 264  */
 265 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 266                             const char * fmt, ...)
 267 {
 268         va_list args;
 269
 270         va_start(args, fmt);
 271         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 272         vprintk(fmt, args);
 273         printk("\n");
 274         va_end(args);
 275
 276         /* this is to prevent panic from syncing this filesystem */
 277         /* AKPM: is this sufficient? */
 278         sb->s_flags |= MS_RDONLY;
 279         panic ("EXT3-fs panic forced\n");
 280 }
 281
 282 void ext3_warning (struct super_block * sb, const char * function,
 283                    const char * fmt, ...)
 284 {
 285         va_list args;
 286
 287         va_start(args, fmt);
 288         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 289                sb->s_id, function);
 290         vprintk(fmt, args);
 291         printk("\n");
 292         va_end(args);
 293 }
 294
 295 void ext3_update_dynamic_rev(struct super_block *sb)
 296 {
 297         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 298
 299         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 300                 return;
 301
 302         ext3_warning(sb, __FUNCTION__,
 303                      "updating to rev %d because of new feature flag, "
 304                      "running e2fsck is recommended",
 305                      EXT3_DYNAMIC_REV);
 306
 307         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 308         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 309         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 310         /* leave es->s_feature_*compat flags alone */
 311         /* es->s_uuid will be set by e2fsck if empty */
 312
 313         /*
 314          * The rest of the superblock fields should be zero, and if not it
 315          * means they are likely already in use, so leave them alone.  We
 316          * can leave it up to e2fsck to clean up any inconsistencies there.
 317          */
 318 }
 319
 320 /*
 321  * Open the external journal device
 322  */
 323 static struct block_device *ext3_blkdev_get(dev_t dev)
 324 {
 325         struct block_device *bdev;
 326         char b[BDEVNAME_SIZE];
 327
 328         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 329         if (IS_ERR(bdev))
 330                 goto fail;
 331         return bdev;
 332
 333 fail:
 334         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 335                         __bdevname(dev, b), PTR_ERR(bdev));
 336         return NULL;
 337 }
 338
 339 /*
 340  * Release the journal device
 341  */
 342 static int ext3_blkdev_put(struct block_device *bdev)
 343 {
 344         bd_release(bdev);
 345         return blkdev_put(bdev);
 346 }
 347
 348 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 349 {
 350         struct block_device *bdev;
 351         int ret = -ENODEV;
 352
 353         bdev = sbi->journal_bdev;
 354         if (bdev) {
 355                 ret = ext3_blkdev_put(bdev);
 356                 sbi->journal_bdev = NULL;
 357         }
 358         return ret;
 359 }
 360
 361 static inline struct inode *orphan_list_entry(struct list_head *l)
 362 {
 363         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 364 }
 365
 366 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 367 {
 368         struct list_head *l;
 369
 370         printk(KERN_ERR "sb orphan head is %d\n",
 371                le32_to_cpu(sbi->s_es->s_last_orphan));
 372
 373         printk(KERN_ERR "sb_info orphan list:\n");
 374         list_for_each(l, &sbi->s_orphan) {
 375                 struct inode *inode = orphan_list_entry(l);
 376                 printk(KERN_ERR "  "
 377                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 378                        inode->i_sb->s_id, inode->i_ino, inode,
 379                        inode->i_mode, inode->i_nlink,
 380                        le32_to_cpu(NEXT_ORPHAN(inode)));
 381         }
 382 }
 383
 384 void ext3_put_super (struct super_block * sb)
 385 {
 386         struct ext3_sb_info *sbi = EXT3_SB(sb);
 387         struct ext3_super_block *es = sbi->s_es;
 388         int i;
 389
 390         ext3_xattr_put_super(sb);
 391         journal_destroy(sbi->s_journal);
 392         if (!(sb->s_flags & MS_RDONLY)) {
 393                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 394                 es->s_state = le16_to_cpu(sbi->s_mount_state);
 395                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 396                 mark_buffer_dirty(sbi->s_sbh);
 397                 ext3_commit_super(sb, es, 1);
 398         }
 399
 400         for (i = 0; i < sbi->s_gdb_count; i++)
 401                 brelse(sbi->s_group_desc[i]);
 402         kfree(sbi->s_group_desc);
 403         kfree(sbi->s_debts);
 404         brelse(sbi->s_sbh);
 405 #ifdef CONFIG_QUOTA
 406         for (i = 0; i < MAXQUOTAS; i++) {
 407                 if (sbi->s_qf_names[i])
 408                         kfree(sbi->s_qf_names[i]);
 409         }
 410 #endif
 411
 412         /* Debugging code just in case the in-memory inode orphan list
 413          * isn't empty.  The on-disk one can be non-empty if we've
 414          * detected an error and taken the fs readonly, but the
 415          * in-memory list had better be clean by this point. */
 416         if (!list_empty(&sbi->s_orphan))
 417                 dump_orphan_list(sb, sbi);
 418         J_ASSERT(list_empty(&sbi->s_orphan));
 419
 420         invalidate_bdev(sb->s_bdev, 0);
 421         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 422                 /*
 423                  * Invalidate the journal device's buffers.  We don't want them
 424                  * floating about in memory - the physical journal device may
 425                  * hotswapped, and it breaks the `ro-after' testing code.
 426                  */
 427                 sync_blockdev(sbi->journal_bdev);
 428                 invalidate_bdev(sbi->journal_bdev, 0);
 429                 ext3_blkdev_remove(sbi);
 430         }
 431         sb->s_fs_info = NULL;
 432         kfree(sbi);
 433         return;
 434 }
 435
 436 static kmem_cache_t *ext3_inode_cachep;
 437
 438 /*
 439  * Called inside transaction, so use GFP_NOFS
 440  */
 441 static struct inode *ext3_alloc_inode(struct super_block *sb)
 442 {
 443         struct ext3_inode_info *ei;
 444
 445         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 446         if (!ei)
 447                 return NULL;
 448 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 449         ei->i_acl = EXT3_ACL_NOT_CACHED;
 450         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 451 #endif
 452         ei->vfs_inode.i_version = 1;
 453         return &ei->vfs_inode;
 454 }
 455
 456 static void ext3_destroy_inode(struct inode *inode)
 457 {
 458         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 459 }
 460
 461 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 462 {
 463         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 464
 465         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 466             SLAB_CTOR_CONSTRUCTOR) {
 467                 INIT_LIST_HEAD(&ei->i_orphan);
 468 #ifdef CONFIG_EXT3_FS_XATTR
 469                 init_rwsem(&ei->xattr_sem);
 470 #endif
 471                 init_MUTEX(&ei->truncate_sem);
 472                 inode_init_once(&ei->vfs_inode);
 473         }
 474 }
 475
 476 static int init_inodecache(void)
 477 {
 478         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 479                                              sizeof(struct ext3_inode_info),
 480                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 481                                              init_once, NULL);
 482         if (ext3_inode_cachep == NULL)
 483                 return -ENOMEM;
 484         return 0;
 485 }
 486
 487 static void destroy_inodecache(void)
 488 {
 489         if (kmem_cache_destroy(ext3_inode_cachep))
 490                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 491 }
 492
 493 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 494
 495 static void ext3_clear_inode(struct inode *inode)
 496 {
 497        if (EXT3_I(inode)->i_acl &&
 498            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 499                posix_acl_release(EXT3_I(inode)->i_acl);
 500                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 501        }
 502        if (EXT3_I(inode)->i_default_acl &&
 503            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 504                posix_acl_release(EXT3_I(inode)->i_default_acl);
 505                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 506        }
 507 }
 508
 509 #else
 510 # define ext3_clear_inode NULL
 511 #endif
 512
 513 #ifdef CONFIG_QUOTA
 514
 515 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 516 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 517
 518 static int ext3_dquot_initialize(struct inode *inode, int type);
 519 static int ext3_dquot_drop(struct inode *inode);
 520 static int ext3_write_dquot(struct dquot *dquot);
 521 static int ext3_acquire_dquot(struct dquot *dquot);
 522 static int ext3_release_dquot(struct dquot *dquot);
 523 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 524 static int ext3_write_info(struct super_block *sb, int type);
 525 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 526 static int ext3_quota_on_mount(struct super_block *sb, int type);
 527 static int ext3_quota_off_mount(struct super_block *sb, int type);
 528
 529 static struct dquot_operations ext3_quota_operations = {
 530         .initialize     = ext3_dquot_initialize,
 531         .drop           = ext3_dquot_drop,
 532         .alloc_space    = dquot_alloc_space,
 533         .alloc_inode    = dquot_alloc_inode,
 534         .free_space     = dquot_free_space,
 535         .free_inode     = dquot_free_inode,
 536         .transfer       = dquot_transfer,
 537         .write_dquot    = ext3_write_dquot,
 538         .acquire_dquot  = ext3_acquire_dquot,
 539         .release_dquot  = ext3_release_dquot,
 540         .mark_dirty     = ext3_mark_dquot_dirty,
 541         .write_info     = ext3_write_info
 542 };
 543
 544 static struct quotactl_ops ext3_qctl_operations = {
 545         .quota_on       = ext3_quota_on,
 546         .quota_off      = vfs_quota_off,
 547         .quota_sync     = vfs_quota_sync,
 548         .get_info       = vfs_get_dqinfo,
 549         .set_info       = vfs_set_dqinfo,
 550         .get_dqblk      = vfs_get_dqblk,
 551         .set_dqblk      = vfs_set_dqblk
 552 };
 553 #endif
 554
 555 static struct super_operations ext3_sops = {
 556         .alloc_inode    = ext3_alloc_inode,
 557         .destroy_inode  = ext3_destroy_inode,
 558         .read_inode     = ext3_read_inode,
 559         .write_inode    = ext3_write_inode,
 560         .dirty_inode    = ext3_dirty_inode,
 561         .put_inode      = ext3_put_inode,
 562         .delete_inode   = ext3_delete_inode,
 563         .put_super      = ext3_put_super,
 564         .write_super    = ext3_write_super,
 565         .sync_fs        = ext3_sync_fs,
 566         .write_super_lockfs = ext3_write_super_lockfs,
 567         .unlockfs       = ext3_unlockfs,
 568         .statfs         = ext3_statfs,
 569         .remount_fs     = ext3_remount,
 570         .clear_inode    = ext3_clear_inode,
 571 };
 572
 573 struct dentry *ext3_get_parent(struct dentry *child);
 574 static struct export_operations ext3_export_ops = {
 575         .get_parent = ext3_get_parent,
 576 };
 577
 578 enum {
 579         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 580         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 581         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 582         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, Opt_noload,
 583         Opt_commit, Opt_journal_update, Opt_journal_inum,
 584         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 585         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 586         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 587         Opt_ignore, Opt_barrier, Opt_err,
 588 };
 589
 590 static match_table_t tokens = {
 591         {Opt_bsd_df, "bsddf"},
 592         {Opt_minix_df, "minixdf"},
 593         {Opt_grpid, "grpid"},
 594         {Opt_grpid, "bsdgroups"},
 595         {Opt_nogrpid, "nogrpid"},
 596         {Opt_nogrpid, "sysvgroups"},
 597         {Opt_resgid, "resgid=%u"},
 598         {Opt_resuid, "resuid=%u"},
 599         {Opt_sb, "sb=%u"},
 600         {Opt_err_cont, "errors=continue"},
 601         {Opt_err_panic, "errors=panic"},
 602         {Opt_err_ro, "errors=remount-ro"},
 603         {Opt_nouid32, "nouid32"},
 604         {Opt_nocheck, "nocheck"},
 605         {Opt_nocheck, "check=none"},
 606         {Opt_check, "check"},
 607         {Opt_debug, "debug"},
 608         {Opt_oldalloc, "oldalloc"},
 609         {Opt_orlov, "orlov"},
 610         {Opt_user_xattr, "user_xattr"},
 611         {Opt_nouser_xattr, "nouser_xattr"},
 612         {Opt_acl, "acl"},
 613         {Opt_noacl, "noacl"},
 614         {Opt_noload, "noload"},
 615         {Opt_commit, "commit=%u"},
 616         {Opt_journal_update, "journal=update"},
 617         {Opt_journal_inum, "journal=%u"},
 618         {Opt_abort, "abort"},
 619         {Opt_data_journal, "data=journal"},
 620         {Opt_data_ordered, "data=ordered"},
 621         {Opt_data_writeback, "data=writeback"},
 622         {Opt_offusrjquota, "usrjquota="},
 623         {Opt_usrjquota, "usrjquota=%s"},
 624         {Opt_offgrpjquota, "grpjquota="},
 625         {Opt_grpjquota, "grpjquota=%s"},
 626         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 627         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 628         {Opt_ignore, "grpquota"},
 629         {Opt_ignore, "noquota"},
 630         {Opt_ignore, "quota"},
 631         {Opt_ignore, "usrquota"},
 632         {Opt_barrier, "barrier=%u"},
 633         {Opt_err, NULL}
 634 };
 635
 636 static unsigned long get_sb_block(void **data)
 637 {
 638         unsigned long   sb_block;
 639         char            *options = (char *) *data;
 640
 641         if (!options || strncmp(options, "sb=", 3) != 0)
 642                 return 1;       /* Default location */
 643         options += 3;
 644         sb_block = simple_strtoul(options, &options, 0);
 645         if (*options && *options != ',') {
 646                 printk("EXT3-fs: Invalid sb specification: %s\n",
 647                        (char *) *data);
 648                 return 1;
 649         }
 650         if (*options == ',')
 651                 options++;
 652         *data = (void *) options;
 653         return sb_block;
 654 }
 655
 656 static int parse_options (char * options, struct super_block *sb,
 657                           unsigned long * inum, int is_remount)
 658 {
 659         struct ext3_sb_info *sbi = EXT3_SB(sb);
 660         char * p;
 661         substring_t args[MAX_OPT_ARGS];
 662         int data_opt = 0;
 663         int option;
 664 #ifdef CONFIG_QUOTA
 665         int qtype;
 666 #endif
 667
 668         if (!options)
 669                 return 1;
 670
 671         while ((p = strsep (&options, ",")) != NULL) {
 672                 int token;
 673                 if (!*p)
 674                         continue;
 675
 676                 token = match_token(p, tokens, args);
 677                 switch (token) {
 678                 case Opt_bsd_df:
 679                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 680                         break;
 681                 case Opt_minix_df:
 682                         set_opt (sbi->s_mount_opt, MINIX_DF);
 683                         break;
 684                 case Opt_grpid:
 685                         set_opt (sbi->s_mount_opt, GRPID);
 686                         break;
 687                 case Opt_nogrpid:
 688                         clear_opt (sbi->s_mount_opt, GRPID);
 689                         break;
 690                 case Opt_resuid:
 691                         if (match_int(&args[0], &option))
 692                                 return 0;
 693                         sbi->s_resuid = option;
 694                         break;
 695                 case Opt_resgid:
 696                         if (match_int(&args[0], &option))
 697                                 return 0;
 698                         sbi->s_resgid = option;
 699                         break;
 700                 case Opt_sb:
 701                         /* handled by get_sb_block() instead of here */
 702                         /* *sb_block = match_int(&args[0]); */
 703                         break;
 704                 case Opt_err_panic:
 705                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 706                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 707                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 708                         break;
 709                 case Opt_err_ro:
 710                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 711                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 712                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 713                         break;
 714                 case Opt_err_cont:
 715                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 716                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 717                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 718                         break;
 719                 case Opt_nouid32:
 720                         set_opt (sbi->s_mount_opt, NO_UID32);
 721                         break;
 722                 case Opt_check:
 723 #ifdef CONFIG_EXT3_CHECK
 724                         set_opt (sbi->s_mount_opt, CHECK);
 725 #else
 726                         printk(KERN_ERR
 727                                "EXT3 Check option not supported\n");
 728 #endif
 729                         break;
 730                 case Opt_nocheck:
 731                         clear_opt (sbi->s_mount_opt, CHECK);
 732                         break;
 733                 case Opt_debug:
 734                         set_opt (sbi->s_mount_opt, DEBUG);
 735                         break;
 736                 case Opt_oldalloc:
 737                         set_opt (sbi->s_mount_opt, OLDALLOC);
 738                         break;
 739                 case Opt_orlov:
 740                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 741                         break;
 742 #ifdef CONFIG_EXT3_FS_XATTR
 743                 case Opt_user_xattr:
 744                         set_opt (sbi->s_mount_opt, XATTR_USER);
 745                         break;
 746                 case Opt_nouser_xattr:
 747                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 748                         break;
 749 #else
 750                 case Opt_user_xattr:
 751                 case Opt_nouser_xattr:
 752                         printk("EXT3 (no)user_xattr options not supported\n");
 753                         break;
 754 #endif
 755 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 756                 case Opt_acl:
 757                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 758                         break;
 759                 case Opt_noacl:
 760                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 761                         break;
 762 #else
 763                 case Opt_acl:
 764                 case Opt_noacl:
 765                         printk("EXT3 (no)acl options not supported\n");
 766                         break;
 767 #endif
 768                 case Opt_journal_update:
 769                         /* @@@ FIXME */
 770                         /* Eventually we will want to be able to create
 771                            a journal file here.  For now, only allow the
 772                            user to specify an existing inode to be the
 773                            journal file. */
 774                         if (is_remount) {
 775                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 776                                        "journal on remount\n");
 777                                 return 0;
 778                         }
 779                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 780                         break;
 781                 case Opt_journal_inum:
 782                         if (is_remount) {
 783                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 784                                        "journal on remount\n");
 785                                 return 0;
 786                         }
 787                         if (match_int(&args[0], &option))
 788                                 return 0;
 789                         *inum = option;
 790                         break;
 791                 case Opt_noload:
 792                         set_opt (sbi->s_mount_opt, NOLOAD);
 793                         break;
 794                 case Opt_commit:
 795                         if (match_int(&args[0], &option))
 796                                 return 0;
 797                         if (option < 0)
 798                                 return 0;
 799                         if (option == 0)
 800                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 801                         sbi->s_commit_interval = HZ * option;
 802                         break;
 803                 case Opt_data_journal:
 804                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 805                         goto datacheck;
 806                 case Opt_data_ordered:
 807                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 808                         goto datacheck;
 809                 case Opt_data_writeback:
 810                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 811                 datacheck:
 812                         if (is_remount) {
 813                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 814                                                 != data_opt) {
 815                                         printk(KERN_ERR
 816                                                 "EXT3-fs: cannot change data "
 817                                                 "mode on remount\n");
 818                                         return 0;
 819                                 }
 820                         } else {
 821                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 822                                 sbi->s_mount_opt |= data_opt;
 823                         }
 824                         break;
 825 #ifdef CONFIG_QUOTA
 826                 case Opt_usrjquota:
 827                         qtype = USRQUOTA;
 828                         goto set_qf_name;
 829                 case Opt_grpjquota:
 830                         qtype = GRPQUOTA;
 831 set_qf_name:
 832                         if (sb_any_quota_enabled(sb)) {
 833                                 printk(KERN_ERR
 834                                         "EXT3-fs: Cannot change journalled "
 835                                         "quota options when quota turned on.\n");
 836                                 return 0;
 837                         }
 838                         if (sbi->s_qf_names[qtype]) {
 839                                 printk(KERN_ERR
 840                                         "EXT3-fs: %s quota file already "
 841                                         "specified.\n", QTYPE2NAME(qtype));
 842                                 return 0;
 843                         }
 844                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 845                         if (!sbi->s_qf_names[qtype]) {
 846                                 printk(KERN_ERR
 847                                         "EXT3-fs: not enough memory for "
 848                                         "storing quotafile name.\n");
 849                                 return 0;
 850                         }
 851                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 852                                 printk(KERN_ERR
 853                                         "EXT3-fs: quotafile must be on "
 854                                         "filesystem root.\n");
 855                                 kfree(sbi->s_qf_names[qtype]);
 856                                 sbi->s_qf_names[qtype] = NULL;
 857                                 return 0;
 858                         }
 859                         break;
 860                 case Opt_offusrjquota:
 861                         qtype = USRQUOTA;
 862                         goto clear_qf_name;
 863                 case Opt_offgrpjquota:
 864                         qtype = GRPQUOTA;
 865 clear_qf_name:
 866                         if (sb_any_quota_enabled(sb)) {
 867                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 868                                         "journalled quota options when "
 869                                         "quota turned on.\n");
 870                                 return 0;
 871                         }
 872                         if (sbi->s_qf_names[qtype]) {
 873                                 kfree(sbi->s_qf_names[qtype]);
 874                                 sbi->s_qf_names[qtype] = NULL;
 875                         }
 876                         break;
 877                 case Opt_jqfmt_vfsold:
 878                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 879                         break;
 880                 case Opt_jqfmt_vfsv0:
 881                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 882                         break;
 883 #else
 884                 case Opt_usrjquota:
 885                 case Opt_grpjquota:
 886                 case Opt_offusrjquota:
 887                 case Opt_offgrpjquota:
 888                 case Opt_jqfmt_vfsold:
 889                 case Opt_jqfmt_vfsv0:
 890                         printk(KERN_ERR
 891                                 "EXT3-fs: journalled quota options not "
 892                                 "supported.\n");
 893                         break;
 894 #endif
 895                 case Opt_abort:
 896                         set_opt(sbi->s_mount_opt, ABORT);
 897                         break;
 898                 case Opt_barrier:
 899                         if (match_int(&args[0], &option))
 900                                 return 0;
 901                         if (option)
 902                                 set_opt(sbi->s_mount_opt, BARRIER);
 903                         else
 904                                 clear_opt(sbi->s_mount_opt, BARRIER);
 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 = NULL;
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         sb->s_flags |= MS_ONE_SECOND;
1301         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1302                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1303
1304         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1305             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1306              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1307              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1308                 printk(KERN_WARNING
1309                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1310                        "running e2fsck is recommended\n");
1311         /*
1312          * Check feature flags regardless of the revision level, since we
1313          * previously didn't change the revision level when setting the flags,
1314          * so there is a chance incompat flags are set on a rev 0 filesystem.
1315          */
1316         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1317                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1318                        "unsupported optional features (%x).\n",
1319                        sb->s_id, i);
1320                 goto failed_mount;
1321         }
1322         if (!(sb->s_flags & MS_RDONLY) &&
1323             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1324                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1325                        "unsupported optional features (%x).\n",
1326                        sb->s_id, i);
1327                 goto failed_mount;
1328         }
1329         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1330
1331         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1332             blocksize > EXT3_MAX_BLOCK_SIZE) {
1333                 printk(KERN_ERR
1334                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1335                        blocksize, sb->s_id);
1336                 goto failed_mount;
1337         }
1338
1339         hblock = bdev_hardsect_size(sb->s_bdev);
1340         if (sb->s_blocksize != blocksize) {
1341                 /*
1342                  * Make sure the blocksize for the filesystem is larger
1343                  * than the hardware sectorsize for the machine.
1344                  */
1345                 if (blocksize < hblock) {
1346                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1347                                "device blocksize %d.\n", blocksize, hblock);
1348                         goto failed_mount;
1349                 }
1350
1351                 brelse (bh);
1352                 sb_set_blocksize(sb, blocksize);
1353                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1354                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1355                 bh = sb_bread(sb, logic_sb_block);
1356                 if (!bh) {
1357                         printk(KERN_ERR
1358                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1359                         goto failed_mount;
1360                 }
1361                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1362                 sbi->s_es = es;
1363                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1364                         printk (KERN_ERR
1365                                 "EXT3-fs: Magic mismatch, very weird !\n");
1366                         goto failed_mount;
1367                 }
1368         }
1369
1370         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1371
1372         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1373                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1374                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1375         } else {
1376                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1377                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1378                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1379                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1380                     (sbi->s_inode_size > blocksize)) {
1381                         printk (KERN_ERR
1382                                 "EXT3-fs: unsupported inode size: %d\n",
1383                                 sbi->s_inode_size);
1384                         goto failed_mount;
1385                 }
1386         }
1387         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1388                                    le32_to_cpu(es->s_log_frag_size);
1389         if (blocksize != sbi->s_frag_size) {
1390                 printk(KERN_ERR
1391                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1392                        sbi->s_frag_size, blocksize);
1393                 goto failed_mount;
1394         }
1395         sbi->s_frags_per_block = 1;
1396         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1397         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1398         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1399         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1400         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1401         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1402         sbi->s_sbh = bh;
1403         sbi->s_mount_state = le16_to_cpu(es->s_state);
1404         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1405         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1406         for (i=0; i < 4; i++)
1407                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1408         sbi->s_def_hash_version = es->s_def_hash_version;
1409
1410         if (sbi->s_blocks_per_group > blocksize * 8) {
1411                 printk (KERN_ERR
1412                         "EXT3-fs: #blocks per group too big: %lu\n",
1413                         sbi->s_blocks_per_group);
1414                 goto failed_mount;
1415         }
1416         if (sbi->s_frags_per_group > blocksize * 8) {
1417                 printk (KERN_ERR
1418                         "EXT3-fs: #fragments per group too big: %lu\n",
1419                         sbi->s_frags_per_group);
1420                 goto failed_mount;
1421         }
1422         if (sbi->s_inodes_per_group > blocksize * 8) {
1423                 printk (KERN_ERR
1424                         "EXT3-fs: #inodes per group too big: %lu\n",
1425                         sbi->s_inodes_per_group);
1426                 goto failed_mount;
1427         }
1428
1429         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1430                                le32_to_cpu(es->s_first_data_block) +
1431                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1432                               EXT3_BLOCKS_PER_GROUP(sb);
1433         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1434                    EXT3_DESC_PER_BLOCK(sb);
1435         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1436                                     GFP_KERNEL);
1437         if (sbi->s_group_desc == NULL) {
1438                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1439                 goto failed_mount;
1440         }
1441         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1442                         GFP_KERNEL);
1443         if (!sbi->s_debts) {
1444                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1445                 goto failed_mount2;
1446         }
1447         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1448
1449         percpu_counter_init(&sbi->s_freeblocks_counter);
1450         percpu_counter_init(&sbi->s_freeinodes_counter);
1451         percpu_counter_init(&sbi->s_dirs_counter);
1452         bgl_lock_init(&sbi->s_blockgroup_lock);
1453
1454         for (i = 0; i < db_count; i++) {
1455                 block = descriptor_loc(sb, logic_sb_block, i);
1456                 sbi->s_group_desc[i] = sb_bread(sb, block);
1457                 if (!sbi->s_group_desc[i]) {
1458                         printk (KERN_ERR "EXT3-fs: "
1459                                 "can't read group descriptor %d\n", i);
1460                         db_count = i;
1461                         goto failed_mount2;
1462                 }
1463         }
1464         if (!ext3_check_descriptors (sb)) {
1465                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1466                 goto failed_mount2;
1467         }
1468         sbi->s_gdb_count = db_count;
1469         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1470         spin_lock_init(&sbi->s_next_gen_lock);
1471         /*
1472          * set up enough so that it can read an inode
1473          */
1474         sb->s_op = &ext3_sops;
1475         sb->s_export_op = &ext3_export_ops;
1476 #ifdef CONFIG_QUOTA
1477         sb->s_qcop = &ext3_qctl_operations;
1478         sb->dq_op = &ext3_quota_operations;
1479 #endif
1480         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1481
1482         sb->s_root = NULL;
1483
1484         needs_recovery = (es->s_last_orphan != 0 ||
1485                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1486                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1487
1488         /*
1489          * The first inode we look at is the journal inode.  Don't try
1490          * root first: it may be modified in the journal!
1491          */
1492         if (!test_opt(sb, NOLOAD) &&
1493             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1494                 if (ext3_load_journal(sb, es))
1495                         goto failed_mount2;
1496         } else if (journal_inum) {
1497                 if (ext3_create_journal(sb, es, journal_inum))
1498                         goto failed_mount2;
1499         } else {
1500                 if (!silent)
1501                         printk (KERN_ERR
1502                                 "ext3: No journal on filesystem on %s\n",
1503                                 sb->s_id);
1504                 goto failed_mount2;
1505         }
1506
1507         /* We have now updated the journal if required, so we can
1508          * validate the data journaling mode. */
1509         switch (test_opt(sb, DATA_FLAGS)) {
1510         case 0:
1511                 /* No mode set, assume a default based on the journal
1512                    capabilities: ORDERED_DATA if the journal can
1513                    cope, else JOURNAL_DATA */
1514                 if (journal_check_available_features
1515                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1516                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1517                 else
1518                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1519                 break;
1520
1521         case EXT3_MOUNT_ORDERED_DATA:
1522         case EXT3_MOUNT_WRITEBACK_DATA:
1523                 if (!journal_check_available_features
1524                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1525                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1526                                "requested data journaling mode\n");
1527                         goto failed_mount3;
1528                 }
1529         default:
1530                 break;
1531         }
1532
1533         /*
1534          * The journal_load will have done any necessary log recovery,
1535          * so we can safely mount the rest of the filesystem now.
1536          */
1537
1538         root = iget(sb, EXT3_ROOT_INO);
1539         sb->s_root = d_alloc_root(root);
1540         if (!sb->s_root) {
1541                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1542                 iput(root);
1543                 goto failed_mount3;
1544         }
1545         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1546                 dput(sb->s_root);
1547                 sb->s_root = NULL;
1548                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1549                 goto failed_mount3;
1550         }
1551
1552         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1553         /*
1554          * akpm: core read_super() calls in here with the superblock locked.
1555          * That deadlocks, because orphan cleanup needs to lock the superblock
1556          * in numerous places.  Here we just pop the lock - it's relatively
1557          * harmless, because we are now ready to accept write_super() requests,
1558          * and aviro says that's the only reason for hanging onto the
1559          * superblock lock.
1560          */
1561         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1562         ext3_orphan_cleanup(sb, es);
1563         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1564         if (needs_recovery)
1565                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1566         ext3_mark_recovery_complete(sb, es);
1567         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1568                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1569                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1570                 "writeback");
1571
1572         percpu_counter_mod(&sbi->s_freeblocks_counter,
1573                 ext3_count_free_blocks(sb));
1574         percpu_counter_mod(&sbi->s_freeinodes_counter,
1575                 ext3_count_free_inodes(sb));
1576         percpu_counter_mod(&sbi->s_dirs_counter,
1577                 ext3_count_dirs(sb));
1578
1579         return 0;
1580
1581 failed_mount3:
1582         journal_destroy(sbi->s_journal);
1583 failed_mount2:
1584         kfree(sbi->s_debts);
1585         for (i = 0; i < db_count; i++)
1586                 brelse(sbi->s_group_desc[i]);
1587         kfree(sbi->s_group_desc);
1588 failed_mount:
1589 #ifdef CONFIG_QUOTA
1590         for (i = 0; i < MAXQUOTAS; i++) {
1591                 if (sbi->s_qf_names[i])
1592                         kfree(sbi->s_qf_names[i]);
1593         }
1594 #endif
1595         ext3_blkdev_remove(sbi);
1596         brelse(bh);
1597 out_fail:
1598         sb->s_fs_info = NULL;
1599         kfree(sbi);
1600         return -EINVAL;
1601 }
1602
1603 /*
1604  * Setup any per-fs journal parameters now.  We'll do this both on
1605  * initial mount, once the journal has been initialised but before we've
1606  * done any recovery; and again on any subsequent remount.
1607  */
1608 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1609 {
1610         struct ext3_sb_info *sbi = EXT3_SB(sb);
1611
1612         if (sbi->s_commit_interval)
1613                 journal->j_commit_interval = sbi->s_commit_interval;
1614         /* We could also set up an ext3-specific default for the commit
1615          * interval here, but for now we'll just fall back to the jbd
1616          * default. */
1617
1618         spin_lock(&journal->j_state_lock);
1619         if (test_opt(sb, BARRIER))
1620                 journal->j_flags |= JFS_BARRIER;
1621         else
1622                 journal->j_flags &= ~JFS_BARRIER;
1623         spin_unlock(&journal->j_state_lock);
1624 }
1625
1626 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1627 {
1628         struct inode *journal_inode;
1629         journal_t *journal;
1630
1631         /* First, test for the existence of a valid inode on disk.  Bad
1632          * things happen if we iget() an unused inode, as the subsequent
1633          * iput() will try to delete it. */
1634
1635         journal_inode = iget(sb, journal_inum);
1636         if (!journal_inode) {
1637                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1638                 return NULL;
1639         }
1640         if (!journal_inode->i_nlink) {
1641                 make_bad_inode(journal_inode);
1642                 iput(journal_inode);
1643                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1644                 return NULL;
1645         }
1646
1647         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1648                   journal_inode, journal_inode->i_size);
1649         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1650                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1651                 iput(journal_inode);
1652                 return NULL;
1653         }
1654
1655         journal = journal_init_inode(journal_inode);
1656         if (!journal) {
1657                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1658                 iput(journal_inode);
1659                 return NULL;
1660         }
1661         journal->j_private = sb;
1662         ext3_init_journal_params(sb, journal);
1663         return journal;
1664 }
1665
1666 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1667                                        dev_t j_dev)
1668 {
1669         struct buffer_head * bh;
1670         journal_t *journal;
1671         int start;
1672         int len;
1673         int hblock, blocksize;
1674         unsigned long sb_block;
1675         unsigned long offset;
1676         struct ext3_super_block * es;
1677         struct block_device *bdev;
1678
1679         bdev = ext3_blkdev_get(j_dev);
1680         if (bdev == NULL)
1681                 return NULL;
1682
1683         if (bd_claim(bdev, sb)) {
1684                 printk(KERN_ERR
1685                         "EXT3: failed to claim external journal device.\n");
1686                 blkdev_put(bdev);
1687                 return NULL;
1688         }
1689
1690         blocksize = sb->s_blocksize;
1691         hblock = bdev_hardsect_size(bdev);
1692         if (blocksize < hblock) {
1693                 printk(KERN_ERR
1694                         "EXT3-fs: blocksize too small for journal device.\n");
1695                 goto out_bdev;
1696         }
1697
1698         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1699         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1700         set_blocksize(bdev, blocksize);
1701         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1702                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1703                        "external journal\n");
1704                 goto out_bdev;
1705         }
1706
1707         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1708         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1709             !(le32_to_cpu(es->s_feature_incompat) &
1710               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1711                 printk(KERN_ERR "EXT3-fs: external journal has "
1712                                         "bad superblock\n");
1713                 brelse(bh);
1714                 goto out_bdev;
1715         }
1716
1717         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1718                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1719                 brelse(bh);
1720                 goto out_bdev;
1721         }
1722
1723         len = le32_to_cpu(es->s_blocks_count);
1724         start = sb_block + 1;
1725         brelse(bh);     /* we're done with the superblock */
1726
1727         journal = journal_init_dev(bdev, sb->s_bdev,
1728                                         start, len, blocksize);
1729         if (!journal) {
1730                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1731                 goto out_bdev;
1732         }
1733         journal->j_private = sb;
1734         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1735         wait_on_buffer(journal->j_sb_buffer);
1736         if (!buffer_uptodate(journal->j_sb_buffer)) {
1737                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1738                 goto out_journal;
1739         }
1740         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1741                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1742                                         "user (unsupported) - %d\n",
1743                         ntohl(journal->j_superblock->s_nr_users));
1744                 goto out_journal;
1745         }
1746         EXT3_SB(sb)->journal_bdev = bdev;
1747         ext3_init_journal_params(sb, journal);
1748         return journal;
1749 out_journal:
1750         journal_destroy(journal);
1751 out_bdev:
1752         ext3_blkdev_put(bdev);
1753         return NULL;
1754 }
1755
1756 static int ext3_load_journal(struct super_block * sb,
1757                              struct ext3_super_block * es)
1758 {
1759         journal_t *journal;
1760         int journal_inum = le32_to_cpu(es->s_journal_inum);
1761         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1762         int err = 0;
1763         int really_read_only;
1764
1765         really_read_only = bdev_read_only(sb->s_bdev);
1766
1767         /*
1768          * Are we loading a blank journal or performing recovery after a
1769          * crash?  For recovery, we need to check in advance whether we
1770          * can get read-write access to the device.
1771          */
1772
1773         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1774                 if (sb->s_flags & MS_RDONLY) {
1775                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1776                                         "required on readonly filesystem.\n");
1777                         if (really_read_only) {
1778                                 printk(KERN_ERR "EXT3-fs: write access "
1779                                         "unavailable, cannot proceed.\n");
1780                                 return -EROFS;
1781                         }
1782                         printk (KERN_INFO "EXT3-fs: write access will "
1783                                         "be enabled during recovery.\n");
1784                 }
1785         }
1786
1787         if (journal_inum && journal_dev) {
1788                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1789                        "and inode journals!\n");
1790                 return -EINVAL;
1791         }
1792
1793         if (journal_inum) {
1794                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1795                         return -EINVAL;
1796         } else {
1797                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1798                         return -EINVAL;
1799         }
1800
1801         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1802                 err = journal_update_format(journal);
1803                 if (err)  {
1804                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1805                         journal_destroy(journal);
1806                         return err;
1807                 }
1808         }
1809
1810         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1811                 err = journal_wipe(journal, !really_read_only);
1812         if (!err)
1813                 err = journal_load(journal);
1814
1815         if (err) {
1816                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1817                 journal_destroy(journal);
1818                 return err;
1819         }
1820
1821         EXT3_SB(sb)->s_journal = journal;
1822         ext3_clear_journal_err(sb, es);
1823         return 0;
1824 }
1825
1826 static int ext3_create_journal(struct super_block * sb,
1827                                struct ext3_super_block * es,
1828                                int journal_inum)
1829 {
1830         journal_t *journal;
1831
1832         if (sb->s_flags & MS_RDONLY) {
1833                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1834                                 "create journal.\n");
1835                 return -EROFS;
1836         }
1837
1838         if (!(journal = ext3_get_journal(sb, journal_inum)))
1839                 return -EINVAL;
1840
1841         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1842                journal_inum);
1843
1844         if (journal_create(journal)) {
1845                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1846                 journal_destroy(journal);
1847                 return -EIO;
1848         }
1849
1850         EXT3_SB(sb)->s_journal = journal;
1851
1852         ext3_update_dynamic_rev(sb);
1853         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1854         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1855
1856         es->s_journal_inum = cpu_to_le32(journal_inum);
1857         sb->s_dirt = 1;
1858
1859         /* Make sure we flush the recovery flag to disk. */
1860         ext3_commit_super(sb, es, 1);
1861
1862         return 0;
1863 }
1864
1865 static void ext3_commit_super (struct super_block * sb,
1866                                struct ext3_super_block * es,
1867                                int sync)
1868 {
1869         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1870
1871         if (!sbh)
1872                 return;
1873         es->s_wtime = cpu_to_le32(get_seconds());
1874         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1875         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1876         BUFFER_TRACE(sbh, "marking dirty");
1877         mark_buffer_dirty(sbh);
1878         if (sync)
1879                 sync_dirty_buffer(sbh);
1880 }
1881
1882
1883 /*
1884  * Have we just finished recovery?  If so, and if we are mounting (or
1885  * remounting) the filesystem readonly, then we will end up with a
1886  * consistent fs on disk.  Record that fact.
1887  */
1888 static void ext3_mark_recovery_complete(struct super_block * sb,
1889                                         struct ext3_super_block * es)
1890 {
1891         journal_t *journal = EXT3_SB(sb)->s_journal;
1892
1893         journal_lock_updates(journal);
1894         journal_flush(journal);
1895         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1896             sb->s_flags & MS_RDONLY) {
1897                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1898                 sb->s_dirt = 0;
1899                 ext3_commit_super(sb, es, 1);
1900         }
1901         journal_unlock_updates(journal);
1902 }
1903
1904 /*
1905  * If we are mounting (or read-write remounting) a filesystem whose journal
1906  * has recorded an error from a previous lifetime, move that error to the
1907  * main filesystem now.
1908  */
1909 static void ext3_clear_journal_err(struct super_block * sb,
1910                                    struct ext3_super_block * es)
1911 {
1912         journal_t *journal;
1913         int j_errno;
1914         const char *errstr;
1915
1916         journal = EXT3_SB(sb)->s_journal;
1917
1918         /*
1919          * Now check for any error status which may have been recorded in the
1920          * journal by a prior ext3_error() or ext3_abort()
1921          */
1922
1923         j_errno = journal_errno(journal);
1924         if (j_errno) {
1925                 char nbuf[16];
1926
1927                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1928                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1929                              "from previous mount: %s", errstr);
1930                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1931                              "filesystem check.");
1932
1933                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1934                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1935                 ext3_commit_super (sb, es, 1);
1936
1937                 journal_clear_err(journal);
1938         }
1939 }
1940
1941 /*
1942  * Force the running and committing transactions to commit,
1943  * and wait on the commit.
1944  */
1945 int ext3_force_commit(struct super_block *sb)
1946 {
1947         journal_t *journal;
1948         int ret;
1949
1950         if (sb->s_flags & MS_RDONLY)
1951                 return 0;
1952
1953         journal = EXT3_SB(sb)->s_journal;
1954         sb->s_dirt = 0;
1955         ret = ext3_journal_force_commit(journal);
1956         return ret;
1957 }
1958
1959 /*
1960  * Ext3 always journals updates to the superblock itself, so we don't
1961  * have to propagate any other updates to the superblock on disk at this
1962  * point.  Just start an async writeback to get the buffers on their way
1963  * to the disk.
1964  *
1965  * This implicitly triggers the writebehind on sync().
1966  */
1967
1968 void ext3_write_super (struct super_block * sb)
1969 {
1970         if (down_trylock(&sb->s_lock) == 0)
1971                 BUG();
1972         sb->s_dirt = 0;
1973 }
1974
1975 static int ext3_sync_fs(struct super_block *sb, int wait)
1976 {
1977         tid_t target;
1978
1979         sb->s_dirt = 0;
1980         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1981                 if (wait)
1982                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1983         }
1984         return 0;
1985 }
1986
1987 /*
1988  * LVM calls this function before a (read-only) snapshot is created.  This
1989  * gives us a chance to flush the journal completely and mark the fs clean.
1990  */
1991 void ext3_write_super_lockfs(struct super_block *sb)
1992 {
1993         sb->s_dirt = 0;
1994
1995         if (!(sb->s_flags & MS_RDONLY)) {
1996                 journal_t *journal = EXT3_SB(sb)->s_journal;
1997
1998                 /* Now we set up the journal barrier. */
1999                 journal_lock_updates(journal);
2000                 journal_flush(journal);
2001
2002                 /* Journal blocked and flushed, clear needs_recovery flag. */
2003                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2004                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2005         }
2006 }
2007
2008 /*
2009  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2010  * flag here, even though the filesystem is not technically dirty yet.
2011  */
2012 void ext3_unlockfs(struct super_block *sb)
2013 {
2014         if (!(sb->s_flags & MS_RDONLY)) {
2015                 lock_super(sb);
2016                 /* Reser the needs_recovery flag before the fs is unlocked. */
2017                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2018                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2019                 unlock_super(sb);
2020                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2021         }
2022 }
2023
2024 int ext3_remount (struct super_block * sb, int * flags, char * data)
2025 {
2026         struct ext3_super_block * es;
2027         struct ext3_sb_info *sbi = EXT3_SB(sb);
2028         unsigned long tmp;
2029
2030         /*
2031          * Allow the "check" option to be passed as a remount option.
2032          */
2033         if (!parse_options(data, sb, &tmp, 1))
2034                 return -EINVAL;
2035
2036         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2037                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2038
2039         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2040                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2041
2042         es = sbi->s_es;
2043
2044         ext3_init_journal_params(sb, sbi->s_journal);
2045
2046         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2047                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2048                         return -EROFS;
2049
2050                 if (*flags & MS_RDONLY) {
2051                         /*
2052                          * First of all, the unconditional stuff we have to do
2053                          * to disable replay of the journal when we next remount
2054                          */
2055                         sb->s_flags |= MS_RDONLY;
2056
2057                         /*
2058                          * OK, test if we are remounting a valid rw partition
2059                          * readonly, and if so set the rdonly flag and then
2060                          * mark the partition as valid again.
2061                          */
2062                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2063                             (sbi->s_mount_state & EXT3_VALID_FS))
2064                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2065
2066                         ext3_mark_recovery_complete(sb, es);
2067                 } else {
2068                         int ret;
2069                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2070                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2071                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2072                                        "remount RDWR because of unsupported "
2073                                        "optional features (%x).\n",
2074                                        sb->s_id, ret);
2075                                 return -EROFS;
2076                         }
2077                         /*
2078                          * Mounting a RDONLY partition read-write, so reread
2079                          * and store the current valid flag.  (It may have
2080                          * been changed by e2fsck since we originally mounted
2081                          * the partition.)
2082                          */
2083                         ext3_clear_journal_err(sb, es);
2084                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2085                         if (!ext3_setup_super (sb, es, 0))
2086                                 sb->s_flags &= ~MS_RDONLY;
2087                 }
2088         }
2089         return 0;
2090 }
2091
2092 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2093 {
2094         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2095         unsigned long overhead;
2096         int i;
2097
2098         if (test_opt (sb, MINIX_DF))
2099                 overhead = 0;
2100         else {
2101                 /*
2102                  * Compute the overhead (FS structures)
2103                  */
2104
2105                 /*
2106                  * All of the blocks before first_data_block are
2107                  * overhead
2108                  */
2109                 overhead = le32_to_cpu(es->s_first_data_block);
2110
2111                 /*
2112                  * Add the overhead attributed to the superblock and
2113                  * block group descriptors.  If the sparse superblocks
2114                  * feature is turned on, then not all groups have this.
2115                  */
2116                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2117                         overhead += ext3_bg_has_super(sb, i) +
2118                                 ext3_bg_num_gdb(sb, i);
2119
2120                 /*
2121                  * Every block group has an inode bitmap, a block
2122                  * bitmap, and an inode table.
2123                  */
2124                 overhead += (EXT3_SB(sb)->s_groups_count *
2125                              (2 + EXT3_SB(sb)->s_itb_per_group));
2126         }
2127
2128         buf->f_type = EXT3_SUPER_MAGIC;
2129         buf->f_bsize = sb->s_blocksize;
2130         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2131         buf->f_bfree = ext3_count_free_blocks (sb);
2132         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2133         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2134                 buf->f_bavail = 0;
2135         buf->f_files = le32_to_cpu(es->s_inodes_count);
2136         buf->f_ffree = ext3_count_free_inodes (sb);
2137         buf->f_namelen = EXT3_NAME_LEN;
2138         return 0;
2139 }
2140
2141 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2142  * is locked for write. Otherwise the are possible deadlocks:
2143  * Process 1                         Process 2
2144  * ext3_create()                     quota_sync()
2145  *   journal_start()                   write_dquot()
2146  *   DQUOT_INIT()                        down(dqio_sem)
2147  *     down(dqio_sem)                    journal_start()
2148  *
2149  */
2150
2151 #ifdef CONFIG_QUOTA
2152
2153 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2154 {
2155         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2156 }
2157
2158 static int ext3_dquot_initialize(struct inode *inode, int type)
2159 {
2160         handle_t *handle;
2161         int ret, err;
2162
2163         /* We may create quota structure so we need to reserve enough blocks */
2164         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2165         if (IS_ERR(handle))
2166                 return PTR_ERR(handle);
2167         ret = dquot_initialize(inode, type);
2168         err = ext3_journal_stop(handle);
2169         if (!ret)
2170                 ret = err;
2171         return ret;
2172 }
2173
2174 static int ext3_dquot_drop(struct inode *inode)
2175 {
2176         handle_t *handle;
2177         int ret, err;
2178
2179         /* We may delete quota structure so we need to reserve enough blocks */
2180         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2181         if (IS_ERR(handle))
2182                 return PTR_ERR(handle);
2183         ret = dquot_drop(inode);
2184         err = ext3_journal_stop(handle);
2185         if (!ret)
2186                 ret = err;
2187         return ret;
2188 }
2189
2190 static int ext3_write_dquot(struct dquot *dquot)
2191 {
2192         int ret, err;
2193         handle_t *handle;
2194
2195         handle = ext3_journal_start(dquot_to_inode(dquot),
2196                                         EXT3_QUOTA_TRANS_BLOCKS);
2197         if (IS_ERR(handle))
2198                 return PTR_ERR(handle);
2199         ret = dquot_commit(dquot);
2200         err = ext3_journal_stop(handle);
2201         if (!ret)
2202                 ret = err;
2203         return ret;
2204 }
2205
2206 static int ext3_acquire_dquot(struct dquot *dquot)
2207 {
2208         int ret, err;
2209         handle_t *handle;
2210
2211         handle = ext3_journal_start(dquot_to_inode(dquot),
2212                                         EXT3_QUOTA_INIT_BLOCKS);
2213         if (IS_ERR(handle))
2214                 return PTR_ERR(handle);
2215         ret = dquot_acquire(dquot);
2216         err = ext3_journal_stop(handle);
2217         if (!ret)
2218                 ret = err;
2219         return ret;
2220 }
2221
2222 static int ext3_release_dquot(struct dquot *dquot)
2223 {
2224         int ret, err;
2225         handle_t *handle;
2226
2227         handle = ext3_journal_start(dquot_to_inode(dquot),
2228                                         EXT3_QUOTA_INIT_BLOCKS);
2229         if (IS_ERR(handle))
2230                 return PTR_ERR(handle);
2231         ret = dquot_release(dquot);
2232         err = ext3_journal_stop(handle);
2233         if (!ret)
2234                 ret = err;
2235         return ret;
2236 }
2237
2238 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2239 {
2240         /* Are we journalling quotas? */
2241         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2242             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2243                 dquot_mark_dquot_dirty(dquot);
2244                 return ext3_write_dquot(dquot);
2245         } else {
2246                 return dquot_mark_dquot_dirty(dquot);
2247         }
2248 }
2249
2250 static int ext3_write_info(struct super_block *sb, int type)
2251 {
2252         int ret, err;
2253         handle_t *handle;
2254
2255         /* Data block + inode block */
2256         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2257         if (IS_ERR(handle))
2258                 return PTR_ERR(handle);
2259         ret = dquot_commit_info(sb, type);
2260         err = ext3_journal_stop(handle);
2261         if (!ret)
2262                 ret = err;
2263         return ret;
2264 }
2265
2266 /*
2267  * Turn on quotas during mount time - we need to find
2268  * the quota file and such...
2269  */
2270 static int ext3_quota_on_mount(struct super_block *sb, int type)
2271 {
2272         int err;
2273         struct dentry *dentry;
2274         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2275                              .hash = 0,
2276                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2277
2278         dentry = lookup_hash(&name, sb->s_root);
2279         if (IS_ERR(dentry))
2280                 return PTR_ERR(dentry);
2281         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2282         if (err)
2283                 dput(dentry);
2284         /* We keep the dentry reference if everything went ok - we drop it
2285          * on quota_off time */
2286         return err;
2287 }
2288
2289 /* Turn quotas off during mount time */
2290 static int ext3_quota_off_mount(struct super_block *sb, int type)
2291 {
2292         int err;
2293         struct dentry *dentry;
2294
2295         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2296         err = vfs_quota_off_mount(sb, type);
2297         /* We invalidate dentry - it has at least wrong hash... */
2298         d_invalidate(dentry);
2299         dput(dentry);
2300         return err;
2301 }
2302
2303 /*
2304  * Standard function to be called on quota_on
2305  */
2306 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2307                          char *path)
2308 {
2309         int err;
2310         struct nameidata nd;
2311
2312         /* Not journalling quota? */
2313         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2314             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2315                 return vfs_quota_on(sb, type, format_id, path);
2316         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2317         if (err)
2318                 return err;
2319         /* Quotafile not on the same filesystem? */
2320         if (nd.mnt->mnt_sb != sb)
2321                 return -EXDEV;
2322         /* Quotafile not of fs root? */
2323         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2324                 printk(KERN_WARNING
2325                         "EXT3-fs: Quota file not on filesystem root. "
2326                         "Journalled quota will not work.\n");
2327         if (!ext3_should_journal_data(nd.dentry->d_inode))
2328                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2329                         "data-journalling. Journalled quota will not work.\n");
2330         path_release(&nd);
2331         return vfs_quota_on(sb, type, format_id, path);
2332 }
2333
2334 #endif
2335
2336 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2337         int flags, const char *dev_name, void *data)
2338 {
2339         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2340 }
2341
2342 static struct file_system_type ext3_fs_type = {
2343         .owner          = THIS_MODULE,
2344         .name           = "ext3",
2345         .get_sb         = ext3_get_sb,
2346         .kill_sb        = kill_block_super,
2347         .fs_flags       = FS_REQUIRES_DEV,
2348 };
2349
2350 static int __init init_ext3_fs(void)
2351 {
2352         int err = init_ext3_xattr();
2353         if (err)
2354                 return err;
2355         err = init_inodecache();
2356         if (err)
2357                 goto out1;
2358         err = register_filesystem_lifo(&ext3_fs_type);
2359         if (err)
2360                 goto out;
2361         return 0;
2362 out:
2363         destroy_inodecache();
2364 out1:
2365         exit_ext3_xattr();
2366         return err;
2367 }
2368
2369 static void __exit exit_ext3_fs(void)
2370 {
2371         unregister_filesystem(&ext3_fs_type);
2372         destroy_inodecache();
2373         exit_ext3_xattr();
2374 }
2375
2376 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2377 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2378 MODULE_LICENSE("GPL");
2379 module_init(init_ext3_fs)
2380 module_exit(exit_ext3_fs)