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_le16(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                        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 = cpu_to_le16(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_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_tagxid, Opt_barrier, Opt_ignore, 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_tagxid, "tagxid"},
 629         {Opt_ignore, "grpquota"},
 630         {Opt_ignore, "noquota"},
 631         {Opt_ignore, "quota"},
 632         {Opt_ignore, "usrquota"},
 633         {Opt_barrier, "barrier=%u"},
 634         {Opt_err, NULL}
 635 };
 636
 637 static unsigned long get_sb_block(void **data)
 638 {
 639         unsigned long   sb_block;
 640         char            *options = (char *) *data;
 641
 642         if (!options || strncmp(options, "sb=", 3) != 0)
 643                 return 1;       /* Default location */
 644         options += 3;
 645         sb_block = simple_strtoul(options, &options, 0);
 646         if (*options && *options != ',') {
 647                 printk("EXT3-fs: Invalid sb specification: %s\n",
 648                        (char *) *data);
 649                 return 1;
 650         }
 651         if (*options == ',')
 652                 options++;
 653         *data = (void *) options;
 654         return sb_block;
 655 }
 656
 657 static int parse_options (char * options, struct super_block *sb,
 658                           unsigned long * inum, int is_remount)
 659 {
 660         struct ext3_sb_info *sbi = EXT3_SB(sb);
 661         char * p;
 662         substring_t args[MAX_OPT_ARGS];
 663         int data_opt = 0;
 664         int option;
 665 #ifdef CONFIG_QUOTA
 666         int qtype;
 667 #endif
 668
 669         if (!options)
 670                 return 1;
 671
 672         while ((p = strsep (&options, ",")) != NULL) {
 673                 int token;
 674                 if (!*p)
 675                         continue;
 676
 677                 token = match_token(p, tokens, args);
 678                 switch (token) {
 679                 case Opt_bsd_df:
 680                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 681                         break;
 682                 case Opt_minix_df:
 683                         set_opt (sbi->s_mount_opt, MINIX_DF);
 684                         break;
 685                 case Opt_grpid:
 686                         set_opt (sbi->s_mount_opt, GRPID);
 687                         break;
 688                 case Opt_nogrpid:
 689                         clear_opt (sbi->s_mount_opt, GRPID);
 690                         break;
 691                 case Opt_resuid:
 692                         if (match_int(&args[0], &option))
 693                                 return 0;
 694                         sbi->s_resuid = option;
 695                         break;
 696                 case Opt_resgid:
 697                         if (match_int(&args[0], &option))
 698                                 return 0;
 699                         sbi->s_resgid = option;
 700                         break;
 701                 case Opt_sb:
 702                         /* handled by get_sb_block() instead of here */
 703                         /* *sb_block = match_int(&args[0]); */
 704                         break;
 705                 case Opt_err_panic:
 706                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 707                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 708                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 709                         break;
 710                 case Opt_err_ro:
 711                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 712                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 713                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 714                         break;
 715                 case Opt_err_cont:
 716                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 717                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 718                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 719                         break;
 720                 case Opt_nouid32:
 721                         set_opt (sbi->s_mount_opt, NO_UID32);
 722                         break;
 723 #ifndef CONFIG_INOXID_NONE
 724                 case Opt_tagxid:
 725                         if (is_remount) {
 726                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 727                                        "tagxid on remount\n");
 728                                 return 0;
 729                         }
 730                         set_opt (sbi->s_mount_opt, TAG_XID);
 731                         break;
 732 #endif
 733                 case Opt_check:
 734 #ifdef CONFIG_EXT3_CHECK
 735                         set_opt (sbi->s_mount_opt, CHECK);
 736 #else
 737                         printk(KERN_ERR
 738                                "EXT3 Check option not supported\n");
 739 #endif
 740                         break;
 741                 case Opt_nocheck:
 742                         clear_opt (sbi->s_mount_opt, CHECK);
 743                         break;
 744                 case Opt_debug:
 745                         set_opt (sbi->s_mount_opt, DEBUG);
 746                         break;
 747                 case Opt_oldalloc:
 748                         set_opt (sbi->s_mount_opt, OLDALLOC);
 749                         break;
 750                 case Opt_orlov:
 751                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 752                         break;
 753 #ifdef CONFIG_EXT3_FS_XATTR
 754                 case Opt_user_xattr:
 755                         set_opt (sbi->s_mount_opt, XATTR_USER);
 756                         break;
 757                 case Opt_nouser_xattr:
 758                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 759                         break;
 760 #else
 761                 case Opt_user_xattr:
 762                 case Opt_nouser_xattr:
 763                         printk("EXT3 (no)user_xattr options not supported\n");
 764                         break;
 765 #endif
 766 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 767                 case Opt_acl:
 768                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 769                         break;
 770                 case Opt_noacl:
 771                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 772                         break;
 773 #else
 774                 case Opt_acl:
 775                 case Opt_noacl:
 776                         printk("EXT3 (no)acl options not supported\n");
 777                         break;
 778 #endif
 779                 case Opt_journal_update:
 780                         /* @@@ FIXME */
 781                         /* Eventually we will want to be able to create
 782                            a journal file here.  For now, only allow the
 783                            user to specify an existing inode to be the
 784                            journal file. */
 785                         if (is_remount) {
 786                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 787                                        "journal on remount\n");
 788                                 return 0;
 789                         }
 790                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 791                         break;
 792                 case Opt_journal_inum:
 793                         if (is_remount) {
 794                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 795                                        "journal on remount\n");
 796                                 return 0;
 797                         }
 798                         if (match_int(&args[0], &option))
 799                                 return 0;
 800                         *inum = option;
 801                         break;
 802                 case Opt_noload:
 803                         set_opt (sbi->s_mount_opt, NOLOAD);
 804                         break;
 805                 case Opt_commit:
 806                         if (match_int(&args[0], &option))
 807                                 return 0;
 808                         if (option < 0)
 809                                 return 0;
 810                         if (option == 0)
 811                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 812                         sbi->s_commit_interval = HZ * option;
 813                         break;
 814                 case Opt_data_journal:
 815                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 816                         goto datacheck;
 817                 case Opt_data_ordered:
 818                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 819                         goto datacheck;
 820                 case Opt_data_writeback:
 821                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 822                 datacheck:
 823                         if (is_remount) {
 824                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 825                                                 != data_opt) {
 826                                         printk(KERN_ERR
 827                                                 "EXT3-fs: cannot change data "
 828                                                 "mode on remount\n");
 829                                         return 0;
 830                                 }
 831                         } else {
 832                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 833                                 sbi->s_mount_opt |= data_opt;
 834                         }
 835                         break;
 836 #ifdef CONFIG_QUOTA
 837                 case Opt_usrjquota:
 838                         qtype = USRQUOTA;
 839                         goto set_qf_name;
 840                 case Opt_grpjquota:
 841                         qtype = GRPQUOTA;
 842 set_qf_name:
 843                         if (sb_any_quota_enabled(sb)) {
 844                                 printk(KERN_ERR
 845                                         "EXT3-fs: Cannot change journalled "
 846                                         "quota options when quota turned on.\n");
 847                                 return 0;
 848                         }
 849                         if (sbi->s_qf_names[qtype]) {
 850                                 printk(KERN_ERR
 851                                         "EXT3-fs: %s quota file already "
 852                                         "specified.\n", QTYPE2NAME(qtype));
 853                                 return 0;
 854                         }
 855                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 856                         if (!sbi->s_qf_names[qtype]) {
 857                                 printk(KERN_ERR
 858                                         "EXT3-fs: not enough memory for "
 859                                         "storing quotafile name.\n");
 860                                 return 0;
 861                         }
 862                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 863                                 printk(KERN_ERR
 864                                         "EXT3-fs: quotafile must be on "
 865                                         "filesystem root.\n");
 866                                 kfree(sbi->s_qf_names[qtype]);
 867                                 sbi->s_qf_names[qtype] = NULL;
 868                                 return 0;
 869                         }
 870                         break;
 871                 case Opt_offusrjquota:
 872                         qtype = USRQUOTA;
 873                         goto clear_qf_name;
 874                 case Opt_offgrpjquota:
 875                         qtype = GRPQUOTA;
 876 clear_qf_name:
 877                         if (sb_any_quota_enabled(sb)) {
 878                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 879                                         "journalled quota options when "
 880                                         "quota turned on.\n");
 881                                 return 0;
 882                         }
 883                         if (sbi->s_qf_names[qtype]) {
 884                                 kfree(sbi->s_qf_names[qtype]);
 885                                 sbi->s_qf_names[qtype] = NULL;
 886                         }
 887                         break;
 888                 case Opt_jqfmt_vfsold:
 889                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 890                         break;
 891                 case Opt_jqfmt_vfsv0:
 892                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 893                         break;
 894 #else
 895                 case Opt_usrjquota:
 896                 case Opt_grpjquota:
 897                 case Opt_offusrjquota:
 898                 case Opt_offgrpjquota:
 899                 case Opt_jqfmt_vfsold:
 900                 case Opt_jqfmt_vfsv0:
 901                         printk(KERN_ERR
 902                                 "EXT3-fs: journalled quota options not "
 903                                 "supported.\n");
 904                         break;
 905 #endif
 906                 case Opt_abort:
 907                         set_opt(sbi->s_mount_opt, ABORT);
 908                         break;
 909                 case Opt_barrier:
 910                         if (match_int(&args[0], &option))
 911                                 return 0;
 912                         if (option)
 913                                 set_opt(sbi->s_mount_opt, BARRIER);
 914                         else
 915                                 clear_opt(sbi->s_mount_opt, BARRIER);
 916                         break;
 917                 case Opt_ignore:
 918                         break;
 919                 default:
 920                         printk (KERN_ERR
 921                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 922                                 "or missing value\n", p);
 923                         return 0;
 924                 }
 925         }
 926 #ifdef CONFIG_QUOTA
 927         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 928             sbi->s_qf_names[GRPQUOTA])) {
 929                 printk(KERN_ERR
 930                         "EXT3-fs: journalled quota format not specified.\n");
 931                 return 0;
 932         }
 933 #endif
 934
 935         return 1;
 936 }
 937
 938 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 939                             int read_only)
 940 {
 941         struct ext3_sb_info *sbi = EXT3_SB(sb);
 942         int res = 0;
 943
 944         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 945                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 946                         "forcing read-only mode\n");
 947                 res = MS_RDONLY;
 948         }
 949         if (read_only)
 950                 return res;
 951         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 952                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 953                         "running e2fsck is recommended\n");
 954         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 955                 printk (KERN_WARNING
 956                         "EXT3-fs warning: mounting fs with errors, "
 957                         "running e2fsck is recommended\n");
 958         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 959                  le16_to_cpu(es->s_mnt_count) >=
 960                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 961                 printk (KERN_WARNING
 962                         "EXT3-fs warning: maximal mount count reached, "
 963                         "running e2fsck is recommended\n");
 964         else if (le32_to_cpu(es->s_checkinterval) &&
 965                 (le32_to_cpu(es->s_lastcheck) +
 966                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 967                 printk (KERN_WARNING
 968                         "EXT3-fs warning: checktime reached, "
 969                         "running e2fsck is recommended\n");
 970 #if 0
 971                 /* @@@ We _will_ want to clear the valid bit if we find
 972                    inconsistencies, to force a fsck at reboot.  But for
 973                    a plain journaled filesystem we can keep it set as
 974                    valid forever! :) */
 975         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 976 #endif
 977         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 978                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 979         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 980         es->s_mtime = cpu_to_le32(get_seconds());
 981         ext3_update_dynamic_rev(sb);
 982         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 983
 984         ext3_commit_super(sb, es, 1);
 985         if (test_opt(sb, DEBUG))
 986                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 987                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 988                         sb->s_blocksize,
 989                         sbi->s_groups_count,
 990                         EXT3_BLOCKS_PER_GROUP(sb),
 991                         EXT3_INODES_PER_GROUP(sb),
 992                         sbi->s_mount_opt);
 993
 994         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 995         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 996                 char b[BDEVNAME_SIZE];
 997
 998                 printk("external journal on %s\n",
 999                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1000         } else {
1001                 printk("internal journal\n");
1002         }
1003 #ifdef CONFIG_EXT3_CHECK
1004         if (test_opt (sb, CHECK)) {
1005                 ext3_check_blocks_bitmap (sb);
1006                 ext3_check_inodes_bitmap (sb);
1007         }
1008 #endif
1009         return res;
1010 }
1011
1012 static int ext3_check_descriptors (struct super_block * sb)
1013 {
1014         struct ext3_sb_info *sbi = EXT3_SB(sb);
1015         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1016         struct ext3_group_desc * gdp = NULL;
1017         int desc_block = 0;
1018         int i;
1019
1020         ext3_debug ("Checking group descriptors");
1021
1022         for (i = 0; i < sbi->s_groups_count; i++)
1023         {
1024                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1025                         gdp = (struct ext3_group_desc *)
1026                                         sbi->s_group_desc[desc_block++]->b_data;
1027                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1028                     le32_to_cpu(gdp->bg_block_bitmap) >=
1029                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1030                 {
1031                         ext3_error (sb, "ext3_check_descriptors",
1032                                     "Block bitmap for group %d"
1033                                     " not in group (block %lu)!",
1034                                     i, (unsigned long)
1035                                         le32_to_cpu(gdp->bg_block_bitmap));
1036                         return 0;
1037                 }
1038                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1039                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1040                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1041                 {
1042                         ext3_error (sb, "ext3_check_descriptors",
1043                                     "Inode bitmap for group %d"
1044                                     " not in group (block %lu)!",
1045                                     i, (unsigned long)
1046                                         le32_to_cpu(gdp->bg_inode_bitmap));
1047                         return 0;
1048                 }
1049                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1050                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1051                     block + EXT3_BLOCKS_PER_GROUP(sb))
1052                 {
1053                         ext3_error (sb, "ext3_check_descriptors",
1054                                     "Inode table for group %d"
1055                                     " not in group (block %lu)!",
1056                                     i, (unsigned long)
1057                                         le32_to_cpu(gdp->bg_inode_table));
1058                         return 0;
1059                 }
1060                 block += EXT3_BLOCKS_PER_GROUP(sb);
1061                 gdp++;
1062         }
1063
1064         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1065         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1066         return 1;
1067 }
1068
1069
1070 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1071  * the superblock) which were deleted from all directories, but held open by
1072  * a process at the time of a crash.  We walk the list and try to delete these
1073  * inodes at recovery time (only with a read-write filesystem).
1074  *
1075  * In order to keep the orphan inode chain consistent during traversal (in
1076  * case of crash during recovery), we link each inode into the superblock
1077  * orphan list_head and handle it the same way as an inode deletion during
1078  * normal operation (which journals the operations for us).
1079  *
1080  * We only do an iget() and an iput() on each inode, which is very safe if we
1081  * accidentally point at an in-use or already deleted inode.  The worst that
1082  * can happen in this case is that we get a "bit already cleared" message from
1083  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1084  * e2fsck was run on this filesystem, and it must have already done the orphan
1085  * inode cleanup for us, so we can safely abort without any further action.
1086  */
1087 static void ext3_orphan_cleanup (struct super_block * sb,
1088                                  struct ext3_super_block * es)
1089 {
1090         unsigned int s_flags = sb->s_flags;
1091         int nr_orphans = 0, nr_truncates = 0;
1092 #ifdef CONFIG_QUOTA
1093         int i;
1094 #endif
1095         if (!es->s_last_orphan) {
1096                 jbd_debug(4, "no orphan inodes to clean up\n");
1097                 return;
1098         }
1099
1100         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1101                 if (es->s_last_orphan)
1102                         jbd_debug(1, "Errors on filesystem, "
1103                                   "clearing orphan list.\n");
1104                 es->s_last_orphan = 0;
1105                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1106                 return;
1107         }
1108
1109         if (s_flags & MS_RDONLY) {
1110                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1111                        sb->s_id);
1112                 sb->s_flags &= ~MS_RDONLY;
1113         }
1114 #ifdef CONFIG_QUOTA
1115         /* Needed for iput() to work correctly and not trash data */
1116         sb->s_flags |= MS_ACTIVE;
1117         /* Turn on quotas so that they are updated correctly */
1118         for (i = 0; i < MAXQUOTAS; i++) {
1119                 if (EXT3_SB(sb)->s_qf_names[i]) {
1120                         int ret = ext3_quota_on_mount(sb, i);
1121                         if (ret < 0)
1122                                 printk(KERN_ERR
1123                                         "EXT3-fs: Cannot turn on journalled "
1124                                         "quota: error %d\n", ret);
1125                 }
1126         }
1127 #endif
1128
1129         while (es->s_last_orphan) {
1130                 struct inode *inode;
1131
1132                 if (!(inode =
1133                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1134                         es->s_last_orphan = 0;
1135                         break;
1136                 }
1137
1138                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1139                 DQUOT_INIT(inode);
1140                 if (inode->i_nlink) {
1141                         printk(KERN_DEBUG
1142                                 "%s: truncating inode %ld to %Ld bytes\n",
1143                                 __FUNCTION__, inode->i_ino, inode->i_size);
1144                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1145                                   inode->i_ino, inode->i_size);
1146                         ext3_truncate(inode);
1147                         nr_truncates++;
1148                 } else {
1149                         printk(KERN_DEBUG
1150                                 "%s: deleting unreferenced inode %ld\n",
1151                                 __FUNCTION__, inode->i_ino);
1152                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1153                                   inode->i_ino);
1154                         nr_orphans++;
1155                 }
1156                 iput(inode);  /* The delete magic happens here! */
1157         }
1158
1159 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1160
1161         if (nr_orphans)
1162                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1163                        sb->s_id, PLURAL(nr_orphans));
1164         if (nr_truncates)
1165                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1166                        sb->s_id, PLURAL(nr_truncates));
1167 #ifdef CONFIG_QUOTA
1168         /* Turn quotas off */
1169         for (i = 0; i < MAXQUOTAS; i++) {
1170                 if (sb_dqopt(sb)->files[i])
1171                         ext3_quota_off_mount(sb, i);
1172         }
1173 #endif
1174         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1175 }
1176
1177 #define log2(n) ffz(~(n))
1178
1179 /*
1180  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1181  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1182  * We need to be 1 filesystem block less than the 2^32 sector limit.
1183  */
1184 static loff_t ext3_max_size(int bits)
1185 {
1186         loff_t res = EXT3_NDIR_BLOCKS;
1187         res += 1LL << (bits-2);
1188         res += 1LL << (2*(bits-2));
1189         res += 1LL << (3*(bits-2));
1190         res <<= bits;
1191         if (res > (512LL << 32) - (1 << bits))
1192                 res = (512LL << 32) - (1 << bits);
1193         return res;
1194 }
1195
1196 static unsigned long descriptor_loc(struct super_block *sb,
1197                                     unsigned long logic_sb_block,
1198                                     int nr)
1199 {
1200         struct ext3_sb_info *sbi = EXT3_SB(sb);
1201         unsigned long bg, first_data_block, first_meta_bg;
1202         int has_super = 0;
1203
1204         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1205         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1206
1207         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1208             nr < first_meta_bg)
1209                 return (logic_sb_block + nr + 1);
1210         bg = sbi->s_desc_per_block * nr;
1211         if (ext3_bg_has_super(sb, bg))
1212                 has_super = 1;
1213         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1214 }
1215
1216
1217 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1218 {
1219         struct buffer_head * bh;
1220         struct ext3_super_block *es = NULL;
1221         struct ext3_sb_info *sbi;
1222         unsigned long block;
1223         unsigned long sb_block = get_sb_block(&data);
1224         unsigned long logic_sb_block;
1225         unsigned long offset = 0;
1226         unsigned long journal_inum = 0;
1227         unsigned long def_mount_opts;
1228         struct inode *root;
1229         int blocksize;
1230         int hblock;
1231         int db_count;
1232         int i;
1233         int needs_recovery;
1234         __le32 features;
1235
1236         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1237         if (!sbi)
1238                 return -ENOMEM;
1239         sb->s_fs_info = sbi;
1240         memset(sbi, 0, sizeof(*sbi));
1241         sbi->s_mount_opt = 0;
1242         sbi->s_resuid = EXT3_DEF_RESUID;
1243         sbi->s_resgid = EXT3_DEF_RESGID;
1244
1245         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1246         if (!blocksize) {
1247                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1248                 goto out_fail;
1249         }
1250
1251         /*
1252          * The ext3 superblock will not be buffer aligned for other than 1kB
1253          * block sizes.  We need to calculate the offset from buffer start.
1254          */
1255         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1256                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1257                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1258         } else {
1259                 logic_sb_block = sb_block;
1260         }
1261
1262         if (!(bh = sb_bread(sb, logic_sb_block))) {
1263                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1264                 goto out_fail;
1265         }
1266         /*
1267          * Note: s_es must be initialized as soon as possible because
1268          *       some ext3 macro-instructions depend on its value
1269          */
1270         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1271         sbi->s_es = es;
1272         sb->s_magic = le16_to_cpu(es->s_magic);
1273         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1274                 if (!silent)
1275                         printk(KERN_ERR
1276                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1277                                sb->s_id);
1278                 goto failed_mount;
1279         }
1280
1281         /* Set defaults before we parse the mount options */
1282         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1283         if (def_mount_opts & EXT3_DEFM_DEBUG)
1284                 set_opt(sbi->s_mount_opt, DEBUG);
1285         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1286                 set_opt(sbi->s_mount_opt, GRPID);
1287         if (def_mount_opts & EXT3_DEFM_UID16)
1288                 set_opt(sbi->s_mount_opt, NO_UID32);
1289         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1290                 set_opt(sbi->s_mount_opt, XATTR_USER);
1291         if (def_mount_opts & EXT3_DEFM_ACL)
1292                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1293         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1294                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1295         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1296                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1297         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1298                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1299
1300         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1301                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1302         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1303                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1304
1305         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1306         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1307
1308         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1309                 goto failed_mount;
1310
1311         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1312                 sb->s_flags |= MS_TAGXID;
1313         sb->s_flags |= MS_ONE_SECOND;
1314         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1315                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1316
1317         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1318             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1319              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1320              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1321                 printk(KERN_WARNING
1322                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1323                        "running e2fsck is recommended\n");
1324         /*
1325          * Check feature flags regardless of the revision level, since we
1326          * previously didn't change the revision level when setting the flags,
1327          * so there is a chance incompat flags are set on a rev 0 filesystem.
1328          */
1329         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1330         if (features) {
1331                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1332                        "unsupported optional features (%x).\n",
1333                        sb->s_id, le32_to_cpu(features));
1334                 goto failed_mount;
1335         }
1336         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1337         if (!(sb->s_flags & MS_RDONLY) && features) {
1338                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1339                        "unsupported optional features (%x).\n",
1340                        sb->s_id, le32_to_cpu(features));
1341                 goto failed_mount;
1342         }
1343         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1344
1345         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1346             blocksize > EXT3_MAX_BLOCK_SIZE) {
1347                 printk(KERN_ERR
1348                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1349                        blocksize, sb->s_id);
1350                 goto failed_mount;
1351         }
1352
1353         hblock = bdev_hardsect_size(sb->s_bdev);
1354         if (sb->s_blocksize != blocksize) {
1355                 /*
1356                  * Make sure the blocksize for the filesystem is larger
1357                  * than the hardware sectorsize for the machine.
1358                  */
1359                 if (blocksize < hblock) {
1360                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1361                                "device blocksize %d.\n", blocksize, hblock);
1362                         goto failed_mount;
1363                 }
1364
1365                 brelse (bh);
1366                 sb_set_blocksize(sb, blocksize);
1367                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1368                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1369                 bh = sb_bread(sb, logic_sb_block);
1370                 if (!bh) {
1371                         printk(KERN_ERR
1372                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1373                         goto failed_mount;
1374                 }
1375                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1376                 sbi->s_es = es;
1377                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1378                         printk (KERN_ERR
1379                                 "EXT3-fs: Magic mismatch, very weird !\n");
1380                         goto failed_mount;
1381                 }
1382         }
1383
1384         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1385
1386         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1387                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1388                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1389         } else {
1390                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1391                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1392                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1393                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1394                     (sbi->s_inode_size > blocksize)) {
1395                         printk (KERN_ERR
1396                                 "EXT3-fs: unsupported inode size: %d\n",
1397                                 sbi->s_inode_size);
1398                         goto failed_mount;
1399                 }
1400         }
1401         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1402                                    le32_to_cpu(es->s_log_frag_size);
1403         if (blocksize != sbi->s_frag_size) {
1404                 printk(KERN_ERR
1405                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1406                        sbi->s_frag_size, blocksize);
1407                 goto failed_mount;
1408         }
1409         sbi->s_frags_per_block = 1;
1410         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1411         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1412         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1413         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1414         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1415         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1416         sbi->s_sbh = bh;
1417         sbi->s_mount_state = le16_to_cpu(es->s_state);
1418         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1419         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1420         for (i=0; i < 4; i++)
1421                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1422         sbi->s_def_hash_version = es->s_def_hash_version;
1423
1424         if (sbi->s_blocks_per_group > blocksize * 8) {
1425                 printk (KERN_ERR
1426                         "EXT3-fs: #blocks per group too big: %lu\n",
1427                         sbi->s_blocks_per_group);
1428                 goto failed_mount;
1429         }
1430         if (sbi->s_frags_per_group > blocksize * 8) {
1431                 printk (KERN_ERR
1432                         "EXT3-fs: #fragments per group too big: %lu\n",
1433                         sbi->s_frags_per_group);
1434                 goto failed_mount;
1435         }
1436         if (sbi->s_inodes_per_group > blocksize * 8) {
1437                 printk (KERN_ERR
1438                         "EXT3-fs: #inodes per group too big: %lu\n",
1439                         sbi->s_inodes_per_group);
1440                 goto failed_mount;
1441         }
1442
1443         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1444                                le32_to_cpu(es->s_first_data_block) +
1445                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1446                               EXT3_BLOCKS_PER_GROUP(sb);
1447         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1448                    EXT3_DESC_PER_BLOCK(sb);
1449         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1450                                     GFP_KERNEL);
1451         if (sbi->s_group_desc == NULL) {
1452                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1453                 goto failed_mount;
1454         }
1455         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1456                         GFP_KERNEL);
1457         if (!sbi->s_debts) {
1458                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1459                 goto failed_mount2;
1460         }
1461         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1462
1463         percpu_counter_init(&sbi->s_freeblocks_counter);
1464         percpu_counter_init(&sbi->s_freeinodes_counter);
1465         percpu_counter_init(&sbi->s_dirs_counter);
1466         bgl_lock_init(&sbi->s_blockgroup_lock);
1467
1468         for (i = 0; i < db_count; i++) {
1469                 block = descriptor_loc(sb, logic_sb_block, i);
1470                 sbi->s_group_desc[i] = sb_bread(sb, block);
1471                 if (!sbi->s_group_desc[i]) {
1472                         printk (KERN_ERR "EXT3-fs: "
1473                                 "can't read group descriptor %d\n", i);
1474                         db_count = i;
1475                         goto failed_mount2;
1476                 }
1477         }
1478         if (!ext3_check_descriptors (sb)) {
1479                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1480                 goto failed_mount2;
1481         }
1482         sbi->s_gdb_count = db_count;
1483         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1484         spin_lock_init(&sbi->s_next_gen_lock);
1485         /*
1486          * set up enough so that it can read an inode
1487          */
1488         sb->s_op = &ext3_sops;
1489         sb->s_export_op = &ext3_export_ops;
1490 #ifdef CONFIG_QUOTA
1491         sb->s_qcop = &ext3_qctl_operations;
1492         sb->dq_op = &ext3_quota_operations;
1493 #endif
1494         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1495
1496         sb->s_root = NULL;
1497
1498         needs_recovery = (es->s_last_orphan != 0 ||
1499                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1500                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1501
1502         /*
1503          * The first inode we look at is the journal inode.  Don't try
1504          * root first: it may be modified in the journal!
1505          */
1506         if (!test_opt(sb, NOLOAD) &&
1507             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1508                 if (ext3_load_journal(sb, es))
1509                         goto failed_mount2;
1510         } else if (journal_inum) {
1511                 if (ext3_create_journal(sb, es, journal_inum))
1512                         goto failed_mount2;
1513         } else {
1514                 if (!silent)
1515                         printk (KERN_ERR
1516                                 "ext3: No journal on filesystem on %s\n",
1517                                 sb->s_id);
1518                 goto failed_mount2;
1519         }
1520
1521         /* We have now updated the journal if required, so we can
1522          * validate the data journaling mode. */
1523         switch (test_opt(sb, DATA_FLAGS)) {
1524         case 0:
1525                 /* No mode set, assume a default based on the journal
1526                    capabilities: ORDERED_DATA if the journal can
1527                    cope, else JOURNAL_DATA */
1528                 if (journal_check_available_features
1529                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1530                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1531                 else
1532                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1533                 break;
1534
1535         case EXT3_MOUNT_ORDERED_DATA:
1536         case EXT3_MOUNT_WRITEBACK_DATA:
1537                 if (!journal_check_available_features
1538                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1539                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1540                                "requested data journaling mode\n");
1541                         goto failed_mount3;
1542                 }
1543         default:
1544                 break;
1545         }
1546
1547         /*
1548          * The journal_load will have done any necessary log recovery,
1549          * so we can safely mount the rest of the filesystem now.
1550          */
1551
1552         root = iget(sb, EXT3_ROOT_INO);
1553         sb->s_root = d_alloc_root(root);
1554         if (!sb->s_root) {
1555                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1556                 iput(root);
1557                 goto failed_mount3;
1558         }
1559         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1560                 dput(sb->s_root);
1561                 sb->s_root = NULL;
1562                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1563                 goto failed_mount3;
1564         }
1565
1566         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1567         /*
1568          * akpm: core read_super() calls in here with the superblock locked.
1569          * That deadlocks, because orphan cleanup needs to lock the superblock
1570          * in numerous places.  Here we just pop the lock - it's relatively
1571          * harmless, because we are now ready to accept write_super() requests,
1572          * and aviro says that's the only reason for hanging onto the
1573          * superblock lock.
1574          */
1575         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1576         ext3_orphan_cleanup(sb, es);
1577         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1578         if (needs_recovery)
1579                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1580         ext3_mark_recovery_complete(sb, es);
1581         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1582                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1583                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1584                 "writeback");
1585
1586         percpu_counter_mod(&sbi->s_freeblocks_counter,
1587                 ext3_count_free_blocks(sb));
1588         percpu_counter_mod(&sbi->s_freeinodes_counter,
1589                 ext3_count_free_inodes(sb));
1590         percpu_counter_mod(&sbi->s_dirs_counter,
1591                 ext3_count_dirs(sb));
1592
1593         return 0;
1594
1595 failed_mount3:
1596         journal_destroy(sbi->s_journal);
1597 failed_mount2:
1598         kfree(sbi->s_debts);
1599         for (i = 0; i < db_count; i++)
1600                 brelse(sbi->s_group_desc[i]);
1601         kfree(sbi->s_group_desc);
1602 failed_mount:
1603 #ifdef CONFIG_QUOTA
1604         for (i = 0; i < MAXQUOTAS; i++) {
1605                 if (sbi->s_qf_names[i])
1606                         kfree(sbi->s_qf_names[i]);
1607         }
1608 #endif
1609         ext3_blkdev_remove(sbi);
1610         brelse(bh);
1611 out_fail:
1612         sb->s_fs_info = NULL;
1613         kfree(sbi);
1614         return -EINVAL;
1615 }
1616
1617 /*
1618  * Setup any per-fs journal parameters now.  We'll do this both on
1619  * initial mount, once the journal has been initialised but before we've
1620  * done any recovery; and again on any subsequent remount.
1621  */
1622 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1623 {
1624         struct ext3_sb_info *sbi = EXT3_SB(sb);
1625
1626         if (sbi->s_commit_interval)
1627                 journal->j_commit_interval = sbi->s_commit_interval;
1628         /* We could also set up an ext3-specific default for the commit
1629          * interval here, but for now we'll just fall back to the jbd
1630          * default. */
1631
1632         spin_lock(&journal->j_state_lock);
1633         if (test_opt(sb, BARRIER))
1634                 journal->j_flags |= JFS_BARRIER;
1635         else
1636                 journal->j_flags &= ~JFS_BARRIER;
1637         spin_unlock(&journal->j_state_lock);
1638 }
1639
1640 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1641 {
1642         struct inode *journal_inode;
1643         journal_t *journal;
1644
1645         /* First, test for the existence of a valid inode on disk.  Bad
1646          * things happen if we iget() an unused inode, as the subsequent
1647          * iput() will try to delete it. */
1648
1649         journal_inode = iget(sb, journal_inum);
1650         if (!journal_inode) {
1651                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1652                 return NULL;
1653         }
1654         if (!journal_inode->i_nlink) {
1655                 make_bad_inode(journal_inode);
1656                 iput(journal_inode);
1657                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1658                 return NULL;
1659         }
1660
1661         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1662                   journal_inode, journal_inode->i_size);
1663         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1664                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1665                 iput(journal_inode);
1666                 return NULL;
1667         }
1668
1669         journal = journal_init_inode(journal_inode);
1670         if (!journal) {
1671                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1672                 iput(journal_inode);
1673                 return NULL;
1674         }
1675         journal->j_private = sb;
1676         ext3_init_journal_params(sb, journal);
1677         return journal;
1678 }
1679
1680 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1681                                        dev_t j_dev)
1682 {
1683         struct buffer_head * bh;
1684         journal_t *journal;
1685         int start;
1686         int len;
1687         int hblock, blocksize;
1688         unsigned long sb_block;
1689         unsigned long offset;
1690         struct ext3_super_block * es;
1691         struct block_device *bdev;
1692
1693         bdev = ext3_blkdev_get(j_dev);
1694         if (bdev == NULL)
1695                 return NULL;
1696
1697         if (bd_claim(bdev, sb)) {
1698                 printk(KERN_ERR
1699                         "EXT3: failed to claim external journal device.\n");
1700                 blkdev_put(bdev);
1701                 return NULL;
1702         }
1703
1704         blocksize = sb->s_blocksize;
1705         hblock = bdev_hardsect_size(bdev);
1706         if (blocksize < hblock) {
1707                 printk(KERN_ERR
1708                         "EXT3-fs: blocksize too small for journal device.\n");
1709                 goto out_bdev;
1710         }
1711
1712         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1713         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1714         set_blocksize(bdev, blocksize);
1715         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1716                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1717                        "external journal\n");
1718                 goto out_bdev;
1719         }
1720
1721         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1722         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1723             !(le32_to_cpu(es->s_feature_incompat) &
1724               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1725                 printk(KERN_ERR "EXT3-fs: external journal has "
1726                                         "bad superblock\n");
1727                 brelse(bh);
1728                 goto out_bdev;
1729         }
1730
1731         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1732                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1733                 brelse(bh);
1734                 goto out_bdev;
1735         }
1736
1737         len = le32_to_cpu(es->s_blocks_count);
1738         start = sb_block + 1;
1739         brelse(bh);     /* we're done with the superblock */
1740
1741         journal = journal_init_dev(bdev, sb->s_bdev,
1742                                         start, len, blocksize);
1743         if (!journal) {
1744                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1745                 goto out_bdev;
1746         }
1747         journal->j_private = sb;
1748         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1749         wait_on_buffer(journal->j_sb_buffer);
1750         if (!buffer_uptodate(journal->j_sb_buffer)) {
1751                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1752                 goto out_journal;
1753         }
1754         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1755                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1756                                         "user (unsupported) - %d\n",
1757                         be32_to_cpu(journal->j_superblock->s_nr_users));
1758                 goto out_journal;
1759         }
1760         EXT3_SB(sb)->journal_bdev = bdev;
1761         ext3_init_journal_params(sb, journal);
1762         return journal;
1763 out_journal:
1764         journal_destroy(journal);
1765 out_bdev:
1766         ext3_blkdev_put(bdev);
1767         return NULL;
1768 }
1769
1770 static int ext3_load_journal(struct super_block * sb,
1771                              struct ext3_super_block * es)
1772 {
1773         journal_t *journal;
1774         int journal_inum = le32_to_cpu(es->s_journal_inum);
1775         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1776         int err = 0;
1777         int really_read_only;
1778
1779         really_read_only = bdev_read_only(sb->s_bdev);
1780
1781         /*
1782          * Are we loading a blank journal or performing recovery after a
1783          * crash?  For recovery, we need to check in advance whether we
1784          * can get read-write access to the device.
1785          */
1786
1787         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1788                 if (sb->s_flags & MS_RDONLY) {
1789                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1790                                         "required on readonly filesystem.\n");
1791                         if (really_read_only) {
1792                                 printk(KERN_ERR "EXT3-fs: write access "
1793                                         "unavailable, cannot proceed.\n");
1794                                 return -EROFS;
1795                         }
1796                         printk (KERN_INFO "EXT3-fs: write access will "
1797                                         "be enabled during recovery.\n");
1798                 }
1799         }
1800
1801         if (journal_inum && journal_dev) {
1802                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1803                        "and inode journals!\n");
1804                 return -EINVAL;
1805         }
1806
1807         if (journal_inum) {
1808                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1809                         return -EINVAL;
1810         } else {
1811                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1812                         return -EINVAL;
1813         }
1814
1815         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1816                 err = journal_update_format(journal);
1817                 if (err)  {
1818                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1819                         journal_destroy(journal);
1820                         return err;
1821                 }
1822         }
1823
1824         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1825                 err = journal_wipe(journal, !really_read_only);
1826         if (!err)
1827                 err = journal_load(journal);
1828
1829         if (err) {
1830                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1831                 journal_destroy(journal);
1832                 return err;
1833         }
1834
1835         EXT3_SB(sb)->s_journal = journal;
1836         ext3_clear_journal_err(sb, es);
1837         return 0;
1838 }
1839
1840 static int ext3_create_journal(struct super_block * sb,
1841                                struct ext3_super_block * es,
1842                                int journal_inum)
1843 {
1844         journal_t *journal;
1845
1846         if (sb->s_flags & MS_RDONLY) {
1847                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1848                                 "create journal.\n");
1849                 return -EROFS;
1850         }
1851
1852         if (!(journal = ext3_get_journal(sb, journal_inum)))
1853                 return -EINVAL;
1854
1855         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1856                journal_inum);
1857
1858         if (journal_create(journal)) {
1859                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1860                 journal_destroy(journal);
1861                 return -EIO;
1862         }
1863
1864         EXT3_SB(sb)->s_journal = journal;
1865
1866         ext3_update_dynamic_rev(sb);
1867         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1868         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1869
1870         es->s_journal_inum = cpu_to_le32(journal_inum);
1871         sb->s_dirt = 1;
1872
1873         /* Make sure we flush the recovery flag to disk. */
1874         ext3_commit_super(sb, es, 1);
1875
1876         return 0;
1877 }
1878
1879 static void ext3_commit_super (struct super_block * sb,
1880                                struct ext3_super_block * es,
1881                                int sync)
1882 {
1883         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1884
1885         if (!sbh)
1886                 return;
1887         es->s_wtime = cpu_to_le32(get_seconds());
1888         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1889         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1890         BUFFER_TRACE(sbh, "marking dirty");
1891         mark_buffer_dirty(sbh);
1892         if (sync)
1893                 sync_dirty_buffer(sbh);
1894 }
1895
1896
1897 /*
1898  * Have we just finished recovery?  If so, and if we are mounting (or
1899  * remounting) the filesystem readonly, then we will end up with a
1900  * consistent fs on disk.  Record that fact.
1901  */
1902 static void ext3_mark_recovery_complete(struct super_block * sb,
1903                                         struct ext3_super_block * es)
1904 {
1905         journal_t *journal = EXT3_SB(sb)->s_journal;
1906
1907         journal_lock_updates(journal);
1908         journal_flush(journal);
1909         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1910             sb->s_flags & MS_RDONLY) {
1911                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1912                 sb->s_dirt = 0;
1913                 ext3_commit_super(sb, es, 1);
1914         }
1915         journal_unlock_updates(journal);
1916 }
1917
1918 /*
1919  * If we are mounting (or read-write remounting) a filesystem whose journal
1920  * has recorded an error from a previous lifetime, move that error to the
1921  * main filesystem now.
1922  */
1923 static void ext3_clear_journal_err(struct super_block * sb,
1924                                    struct ext3_super_block * es)
1925 {
1926         journal_t *journal;
1927         int j_errno;
1928         const char *errstr;
1929
1930         journal = EXT3_SB(sb)->s_journal;
1931
1932         /*
1933          * Now check for any error status which may have been recorded in the
1934          * journal by a prior ext3_error() or ext3_abort()
1935          */
1936
1937         j_errno = journal_errno(journal);
1938         if (j_errno) {
1939                 char nbuf[16];
1940
1941                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1942                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1943                              "from previous mount: %s", errstr);
1944                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1945                              "filesystem check.");
1946
1947                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1948                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1949                 ext3_commit_super (sb, es, 1);
1950
1951                 journal_clear_err(journal);
1952         }
1953 }
1954
1955 /*
1956  * Force the running and committing transactions to commit,
1957  * and wait on the commit.
1958  */
1959 int ext3_force_commit(struct super_block *sb)
1960 {
1961         journal_t *journal;
1962         int ret;
1963
1964         if (sb->s_flags & MS_RDONLY)
1965                 return 0;
1966
1967         journal = EXT3_SB(sb)->s_journal;
1968         sb->s_dirt = 0;
1969         ret = ext3_journal_force_commit(journal);
1970         return ret;
1971 }
1972
1973 /*
1974  * Ext3 always journals updates to the superblock itself, so we don't
1975  * have to propagate any other updates to the superblock on disk at this
1976  * point.  Just start an async writeback to get the buffers on their way
1977  * to the disk.
1978  *
1979  * This implicitly triggers the writebehind on sync().
1980  */
1981
1982 void ext3_write_super (struct super_block * sb)
1983 {
1984         if (down_trylock(&sb->s_lock) == 0)
1985                 BUG();
1986         sb->s_dirt = 0;
1987 }
1988
1989 static int ext3_sync_fs(struct super_block *sb, int wait)
1990 {
1991         tid_t target;
1992
1993         sb->s_dirt = 0;
1994         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1995                 if (wait)
1996                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1997         }
1998         return 0;
1999 }
2000
2001 /*
2002  * LVM calls this function before a (read-only) snapshot is created.  This
2003  * gives us a chance to flush the journal completely and mark the fs clean.
2004  */
2005 void ext3_write_super_lockfs(struct super_block *sb)
2006 {
2007         sb->s_dirt = 0;
2008
2009         if (!(sb->s_flags & MS_RDONLY)) {
2010                 journal_t *journal = EXT3_SB(sb)->s_journal;
2011
2012                 /* Now we set up the journal barrier. */
2013                 journal_lock_updates(journal);
2014                 journal_flush(journal);
2015
2016                 /* Journal blocked and flushed, clear needs_recovery flag. */
2017                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2018                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2019         }
2020 }
2021
2022 /*
2023  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2024  * flag here, even though the filesystem is not technically dirty yet.
2025  */
2026 void ext3_unlockfs(struct super_block *sb)
2027 {
2028         if (!(sb->s_flags & MS_RDONLY)) {
2029                 lock_super(sb);
2030                 /* Reser the needs_recovery flag before the fs is unlocked. */
2031                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2032                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2033                 unlock_super(sb);
2034                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2035         }
2036 }
2037
2038 int ext3_remount (struct super_block * sb, int * flags, char * data)
2039 {
2040         struct ext3_super_block * es;
2041         struct ext3_sb_info *sbi = EXT3_SB(sb);
2042         unsigned long tmp;
2043
2044         /*
2045          * Allow the "check" option to be passed as a remount option.
2046          */
2047         if (!parse_options(data, sb, &tmp, 1))
2048                 return -EINVAL;
2049
2050         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2051                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2052
2053         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2054                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2055
2056         es = sbi->s_es;
2057
2058         ext3_init_journal_params(sb, sbi->s_journal);
2059
2060         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2061                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2062                         return -EROFS;
2063
2064                 if (*flags & MS_RDONLY) {
2065                         /*
2066                          * First of all, the unconditional stuff we have to do
2067                          * to disable replay of the journal when we next remount
2068                          */
2069                         sb->s_flags |= MS_RDONLY;
2070
2071                         /*
2072                          * OK, test if we are remounting a valid rw partition
2073                          * readonly, and if so set the rdonly flag and then
2074                          * mark the partition as valid again.
2075                          */
2076                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2077                             (sbi->s_mount_state & EXT3_VALID_FS))
2078                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2079
2080                         ext3_mark_recovery_complete(sb, es);
2081                 } else {
2082                         __le32 ret;
2083                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2084                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2085                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2086                                        "remount RDWR because of unsupported "
2087                                        "optional features (%x).\n",
2088                                        sb->s_id, le32_to_cpu(ret));
2089                                 return -EROFS;
2090                         }
2091                         /*
2092                          * Mounting a RDONLY partition read-write, so reread
2093                          * and store the current valid flag.  (It may have
2094                          * been changed by e2fsck since we originally mounted
2095                          * the partition.)
2096                          */
2097                         ext3_clear_journal_err(sb, es);
2098                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2099                         if (!ext3_setup_super (sb, es, 0))
2100                                 sb->s_flags &= ~MS_RDONLY;
2101                 }
2102         }
2103         return 0;
2104 }
2105
2106 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2107 {
2108         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2109         unsigned long overhead;
2110         int i;
2111
2112         if (test_opt (sb, MINIX_DF))
2113                 overhead = 0;
2114         else {
2115                 /*
2116                  * Compute the overhead (FS structures)
2117                  */
2118
2119                 /*
2120                  * All of the blocks before first_data_block are
2121                  * overhead
2122                  */
2123                 overhead = le32_to_cpu(es->s_first_data_block);
2124
2125                 /*
2126                  * Add the overhead attributed to the superblock and
2127                  * block group descriptors.  If the sparse superblocks
2128                  * feature is turned on, then not all groups have this.
2129                  */
2130                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2131                         overhead += ext3_bg_has_super(sb, i) +
2132                                 ext3_bg_num_gdb(sb, i);
2133
2134                 /*
2135                  * Every block group has an inode bitmap, a block
2136                  * bitmap, and an inode table.
2137                  */
2138                 overhead += (EXT3_SB(sb)->s_groups_count *
2139                              (2 + EXT3_SB(sb)->s_itb_per_group));
2140         }
2141
2142         buf->f_type = EXT3_SUPER_MAGIC;
2143         buf->f_bsize = sb->s_blocksize;
2144         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2145         buf->f_bfree = ext3_count_free_blocks (sb);
2146         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2147         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2148                 buf->f_bavail = 0;
2149         buf->f_files = le32_to_cpu(es->s_inodes_count);
2150         buf->f_ffree = ext3_count_free_inodes (sb);
2151         buf->f_namelen = EXT3_NAME_LEN;
2152         return 0;
2153 }
2154
2155 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2156  * is locked for write. Otherwise the are possible deadlocks:
2157  * Process 1                         Process 2
2158  * ext3_create()                     quota_sync()
2159  *   journal_start()                   write_dquot()
2160  *   DQUOT_INIT()                        down(dqio_sem)
2161  *     down(dqio_sem)                    journal_start()
2162  *
2163  */
2164
2165 #ifdef CONFIG_QUOTA
2166
2167 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2168 {
2169         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2170 }
2171
2172 static int ext3_dquot_initialize(struct inode *inode, int type)
2173 {
2174         handle_t *handle;
2175         int ret, err;
2176
2177         /* We may create quota structure so we need to reserve enough blocks */
2178         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2179         if (IS_ERR(handle))
2180                 return PTR_ERR(handle);
2181         ret = dquot_initialize(inode, type);
2182         err = ext3_journal_stop(handle);
2183         if (!ret)
2184                 ret = err;
2185         return ret;
2186 }
2187
2188 static int ext3_dquot_drop(struct inode *inode)
2189 {
2190         handle_t *handle;
2191         int ret, err;
2192
2193         /* We may delete quota structure so we need to reserve enough blocks */
2194         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2195         if (IS_ERR(handle))
2196                 return PTR_ERR(handle);
2197         ret = dquot_drop(inode);
2198         err = ext3_journal_stop(handle);
2199         if (!ret)
2200                 ret = err;
2201         return ret;
2202 }
2203
2204 static int ext3_write_dquot(struct dquot *dquot)
2205 {
2206         int ret, err;
2207         handle_t *handle;
2208
2209         handle = ext3_journal_start(dquot_to_inode(dquot),
2210                                         EXT3_QUOTA_TRANS_BLOCKS);
2211         if (IS_ERR(handle))
2212                 return PTR_ERR(handle);
2213         ret = dquot_commit(dquot);
2214         err = ext3_journal_stop(handle);
2215         if (!ret)
2216                 ret = err;
2217         return ret;
2218 }
2219
2220 static int ext3_acquire_dquot(struct dquot *dquot)
2221 {
2222         int ret, err;
2223         handle_t *handle;
2224
2225         handle = ext3_journal_start(dquot_to_inode(dquot),
2226                                         EXT3_QUOTA_INIT_BLOCKS);
2227         if (IS_ERR(handle))
2228                 return PTR_ERR(handle);
2229         ret = dquot_acquire(dquot);
2230         err = ext3_journal_stop(handle);
2231         if (!ret)
2232                 ret = err;
2233         return ret;
2234 }
2235
2236 static int ext3_release_dquot(struct dquot *dquot)
2237 {
2238         int ret, err;
2239         handle_t *handle;
2240
2241         handle = ext3_journal_start(dquot_to_inode(dquot),
2242                                         EXT3_QUOTA_INIT_BLOCKS);
2243         if (IS_ERR(handle))
2244                 return PTR_ERR(handle);
2245         ret = dquot_release(dquot);
2246         err = ext3_journal_stop(handle);
2247         if (!ret)
2248                 ret = err;
2249         return ret;
2250 }
2251
2252 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2253 {
2254         /* Are we journalling quotas? */
2255         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2256             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2257                 dquot_mark_dquot_dirty(dquot);
2258                 return ext3_write_dquot(dquot);
2259         } else {
2260                 return dquot_mark_dquot_dirty(dquot);
2261         }
2262 }
2263
2264 static int ext3_write_info(struct super_block *sb, int type)
2265 {
2266         int ret, err;
2267         handle_t *handle;
2268
2269         /* Data block + inode block */
2270         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2271         if (IS_ERR(handle))
2272                 return PTR_ERR(handle);
2273         ret = dquot_commit_info(sb, type);
2274         err = ext3_journal_stop(handle);
2275         if (!ret)
2276                 ret = err;
2277         return ret;
2278 }
2279
2280 /*
2281  * Turn on quotas during mount time - we need to find
2282  * the quota file and such...
2283  */
2284 static int ext3_quota_on_mount(struct super_block *sb, int type)
2285 {
2286         int err;
2287         struct dentry *dentry;
2288         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2289                              .hash = 0,
2290                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2291
2292         dentry = lookup_hash(&name, sb->s_root);
2293         if (IS_ERR(dentry))
2294                 return PTR_ERR(dentry);
2295         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2296         if (err)
2297                 dput(dentry);
2298         /* We keep the dentry reference if everything went ok - we drop it
2299          * on quota_off time */
2300         return err;
2301 }
2302
2303 /* Turn quotas off during mount time */
2304 static int ext3_quota_off_mount(struct super_block *sb, int type)
2305 {
2306         int err;
2307         struct dentry *dentry;
2308
2309         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2310         err = vfs_quota_off_mount(sb, type);
2311         /* We invalidate dentry - it has at least wrong hash... */
2312         d_invalidate(dentry);
2313         dput(dentry);
2314         return err;
2315 }
2316
2317 /*
2318  * Standard function to be called on quota_on
2319  */
2320 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2321                          char *path)
2322 {
2323         int err;
2324         struct nameidata nd;
2325
2326         /* Not journalling quota? */
2327         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2328             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2329                 return vfs_quota_on(sb, type, format_id, path);
2330         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2331         if (err)
2332                 return err;
2333         /* Quotafile not on the same filesystem? */
2334         if (nd.mnt->mnt_sb != sb)
2335                 return -EXDEV;
2336         /* Quotafile not of fs root? */
2337         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2338                 printk(KERN_WARNING
2339                         "EXT3-fs: Quota file not on filesystem root. "
2340                         "Journalled quota will not work.\n");
2341         if (!ext3_should_journal_data(nd.dentry->d_inode))
2342                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2343                         "data-journalling. Journalled quota will not work.\n");
2344         path_release(&nd);
2345         return vfs_quota_on(sb, type, format_id, path);
2346 }
2347
2348 #endif
2349
2350 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2351         int flags, const char *dev_name, void *data)
2352 {
2353         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2354 }
2355
2356 static struct file_system_type ext3_fs_type = {
2357         .owner          = THIS_MODULE,
2358         .name           = "ext3",
2359         .get_sb         = ext3_get_sb,
2360         .kill_sb        = kill_block_super,
2361         .fs_flags       = FS_REQUIRES_DEV,
2362 };
2363
2364 static int __init init_ext3_fs(void)
2365 {
2366         int err = init_ext3_xattr();
2367         if (err)
2368                 return err;
2369         err = init_inodecache();
2370         if (err)
2371                 goto out1;
2372         err = register_filesystem(&ext3_fs_type);
2373         if (err)
2374                 goto out;
2375         return 0;
2376 out:
2377         destroy_inodecache();
2378 out1:
2379         exit_ext3_xattr();
2380         return err;
2381 }
2382
2383 static void __exit exit_ext3_fs(void)
2384 {
2385         unregister_filesystem(&ext3_fs_type);
2386         destroy_inodecache();
2387         exit_ext3_xattr();
2388 }
2389
2390 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2391 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2392 MODULE_LICENSE("GPL");
2393 module_init(init_ext3_fs)
2394 module_exit(exit_ext3_fs)