fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53
  54 /*
  55  * Wrappers for journal_start/end.
  56  *
  57  * The only special thing we need to do here is to make sure that all
  58  * journal_end calls result in the superblock being marked dirty, so
  59  * that sync() will call the filesystem's write_super callback if
  60  * appropriate.
  61  */
  62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (inode->i_sb->s_flags & MS_RDONLY)
  67                 return ERR_PTR(-EROFS);
  68
  69         /* Special case here: if the journal has aborted behind our
  70          * backs (eg. EIO in the commit thread), then we still need to
  71          * take the FS itself readonly cleanly. */
  72         journal = EXT3_JOURNAL(inode);
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(inode->i_sb, __FUNCTION__,
  75                            "Detected aborted journal");
  76                 return ERR_PTR(-EROFS);
  77         }
  78
  79         return journal_start(journal, nblocks);
  80 }
  81
  82 /*
  83  * The only special thing we need to do here is to make sure that all
  84  * journal_stop calls result in the superblock being marked dirty, so
  85  * that sync() will call the filesystem's write_super callback if
  86  * appropriate.
  87  */
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         printk(KERN_ERR "%s: aborting transaction: %s in %s",
 112                caller, errstr, err_fn);
 113
 114         if (bh)
 115                 BUFFER_TRACE(bh, "abort");
 116         journal_abort_handle(handle);
 117         if (!handle->h_err)
 118                 handle->h_err = err;
 119 }
 120
 121 static char error_buf[1024];
 122
 123 /* Deal with the reporting of failure conditions on a filesystem such as
 124  * inconsistencies detected or read IO failures.
 125  *
 126  * On ext2, we can store the error state of the filesystem in the
 127  * superblock.  That is not possible on ext3, because we may have other
 128  * write ordering constraints on the superblock which prevent us from
 129  * writing it out straight away; and given that the journal is about to
 130  * be aborted, we can't rely on the current, or future, transactions to
 131  * write out the superblock safely.
 132  *
 133  * We'll just use the journal_abort() error code to record an error in
 134  * the journal instead.  On recovery, the journal will compain about
 135  * that error until we've noted it down and cleared it.
 136  */
 137
 138 static void ext3_handle_error(struct super_block *sb)
 139 {
 140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 141
 142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 144
 145         if (sb->s_flags & MS_RDONLY)
 146                 return;
 147
 148         if (test_opt (sb, ERRORS_PANIC))
 149                 panic ("EXT3-fs (device %s): panic forced after error\n",
 150                        sb->s_id);
 151         if (test_opt (sb, ERRORS_RO)) {
 152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 153                 sb->s_flags |= MS_RDONLY;
 154         } else {
 155                 journal_t *journal = EXT3_SB(sb)->s_journal;
 156
 157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 158                 if (journal)
 159                         journal_abort(journal, -EIO);
 160         }
 161         ext3_commit_super(sb, es, 1);
 162 }
 163
 164 void ext3_error (struct super_block * sb, const char * function,
 165                  const char * fmt, ...)
 166 {
 167         va_list args;
 168
 169         va_start (args, fmt);
 170         vsprintf (error_buf, fmt, args);
 171         va_end (args);
 172
 173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
 174                 sb->s_id, function, error_buf);
 175
 176         ext3_handle_error(sb);
 177 }
 178
 179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 180 {
 181         char *errstr = NULL;
 182
 183         switch (errno) {
 184         case -EIO:
 185                 errstr = "IO failure";
 186                 break;
 187         case -ENOMEM:
 188                 errstr = "Out of memory";
 189                 break;
 190         case -EROFS:
 191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 192                         errstr = "Journal has aborted";
 193                 else
 194                         errstr = "Readonly filesystem";
 195                 break;
 196         default:
 197                 /* If the caller passed in an extra buffer for unknown
 198                  * errors, textualise them now.  Else we just return
 199                  * NULL. */
 200                 if (nbuf) {
 201                         /* Check for truncated error codes... */
 202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 203                                 errstr = nbuf;
 204                 }
 205                 break;
 206         }
 207
 208         return errstr;
 209 }
 210
 211 /* __ext3_std_error decodes expected errors from journaling functions
 212  * automatically and invokes the appropriate error response.  */
 213
 214 void __ext3_std_error (struct super_block * sb, const char * function,
 215                        int errno)
 216 {
 217         char nbuf[16];
 218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 219
 220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 221                 sb->s_id, function, errstr);
 222
 223         ext3_handle_error(sb);
 224 }
 225
 226 /*
 227  * ext3_abort is a much stronger failure handler than ext3_error.  The
 228  * abort function may be used to deal with unrecoverable failures such
 229  * as journal IO errors or ENOMEM at a critical moment in log management.
 230  *
 231  * We unconditionally force the filesystem into an ABORT|READONLY state,
 232  * unless the error response on the fs has been set to panic in which
 233  * case we take the easy way out and panic immediately.
 234  */
 235
 236 void ext3_abort (struct super_block * sb, const char * function,
 237                  const char * fmt, ...)
 238 {
 239         va_list args;
 240
 241         printk (KERN_CRIT "ext3_abort called.\n");
 242
 243         va_start (args, fmt);
 244         vsprintf (error_buf, fmt, args);
 245         va_end (args);
 246
 247         if (test_opt (sb, ERRORS_PANIC))
 248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
 249                        sb->s_id, function, error_buf);
 250
 251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
 252                 sb->s_id, function, error_buf);
 253
 254         if (sb->s_flags & MS_RDONLY)
 255                 return;
 256
 257         printk (KERN_CRIT "Remounting filesystem read-only\n");
 258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 259         sb->s_flags |= MS_RDONLY;
 260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 262 }
 263
 264 /* Deal with the reporting of failure conditions while running, such as
 265  * inconsistencies in operation or invalid system states.
 266  *
 267  * Use ext3_error() for cases of invalid filesystem states, as that will
 268  * record an error on disk and force a filesystem check on the next boot.
 269  */
 270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 271                             const char * fmt, ...)
 272 {
 273         va_list args;
 274
 275         va_start (args, fmt);
 276         vsprintf (error_buf, fmt, args);
 277         va_end (args);
 278
 279         /* this is to prevent panic from syncing this filesystem */
 280         /* AKPM: is this sufficient? */
 281         sb->s_flags |= MS_RDONLY;
 282         panic ("EXT3-fs panic (device %s): %s: %s\n",
 283                sb->s_id, function, error_buf);
 284 }
 285
 286 void ext3_warning (struct super_block * sb, const char * function,
 287                    const char * fmt, ...)
 288 {
 289         va_list args;
 290
 291         va_start (args, fmt);
 292         vsprintf (error_buf, fmt, args);
 293         va_end (args);
 294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
 295                 sb->s_id, function, error_buf);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = 0;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        le32_to_cpu(NEXT_ORPHAN(inode)));
 384         }
 385 }
 386
 387 void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = le16_to_cpu(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         kfree(sbi->s_debts);
 407         brelse(sbi->s_sbh);
 408 #ifdef CONFIG_QUOTA
 409         for (i = 0; i < MAXQUOTAS; i++) {
 410                 if (sbi->s_qf_names[i])
 411                         kfree(sbi->s_qf_names[i]);
 412         }
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 init_MUTEX(&ei->truncate_sem);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 484                                              init_once, NULL);
 485         if (ext3_inode_cachep == NULL)
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static void destroy_inodecache(void)
 491 {
 492         if (kmem_cache_destroy(ext3_inode_cachep))
 493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 494 }
 495
 496 static void ext3_clear_inode(struct inode *inode)
 497 {
 498 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 499        if (EXT3_I(inode)->i_acl &&
 500            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 501                posix_acl_release(EXT3_I(inode)->i_acl);
 502                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 503        }
 504        if (EXT3_I(inode)->i_default_acl &&
 505            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 506                posix_acl_release(EXT3_I(inode)->i_default_acl);
 507                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 508        }
 509 #endif
 510         if (!is_bad_inode(inode))
 511                 ext3_discard_reservation(inode);
 512 }
 513
 514 #ifdef CONFIG_QUOTA
 515
 516 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 517 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 518
 519 static int ext3_dquot_initialize(struct inode *inode, int type);
 520 static int ext3_dquot_drop(struct inode *inode);
 521 static int ext3_write_dquot(struct dquot *dquot);
 522 static int ext3_acquire_dquot(struct dquot *dquot);
 523 static int ext3_release_dquot(struct dquot *dquot);
 524 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 525 static int ext3_write_info(struct super_block *sb, int type);
 526 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 527 static int ext3_quota_on_mount(struct super_block *sb, int type);
 528 static int ext3_quota_off_mount(struct super_block *sb, int type);
 529
 530 static struct dquot_operations ext3_quota_operations = {
 531         .initialize     = ext3_dquot_initialize,
 532         .drop           = ext3_dquot_drop,
 533         .alloc_space    = dquot_alloc_space,
 534         .alloc_inode    = dquot_alloc_inode,
 535         .free_space     = dquot_free_space,
 536         .free_inode     = dquot_free_inode,
 537         .transfer       = dquot_transfer,
 538         .write_dquot    = ext3_write_dquot,
 539         .acquire_dquot  = ext3_acquire_dquot,
 540         .release_dquot  = ext3_release_dquot,
 541         .mark_dirty     = ext3_mark_dquot_dirty,
 542         .write_info     = ext3_write_info
 543 };
 544
 545 static struct quotactl_ops ext3_qctl_operations = {
 546         .quota_on       = ext3_quota_on,
 547         .quota_off      = vfs_quota_off,
 548         .quota_sync     = vfs_quota_sync,
 549         .get_info       = vfs_get_dqinfo,
 550         .set_info       = vfs_set_dqinfo,
 551         .get_dqblk      = vfs_get_dqblk,
 552         .set_dqblk      = vfs_set_dqblk
 553 };
 554 #endif
 555
 556 static struct super_operations ext3_sops = {
 557         .alloc_inode    = ext3_alloc_inode,
 558         .destroy_inode  = ext3_destroy_inode,
 559         .read_inode     = ext3_read_inode,
 560         .write_inode    = ext3_write_inode,
 561         .dirty_inode    = ext3_dirty_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,
 583         Opt_reservation, Opt_noreservation, Opt_noload,
 584         Opt_commit, Opt_journal_update, Opt_journal_inum,
 585         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 586         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 587         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 588         Opt_tagxid, Opt_ignore, Opt_err, Opt_resize,
 589 };
 590
 591 static match_table_t tokens = {
 592         {Opt_bsd_df, "bsddf"},
 593         {Opt_minix_df, "minixdf"},
 594         {Opt_grpid, "grpid"},
 595         {Opt_grpid, "bsdgroups"},
 596         {Opt_nogrpid, "nogrpid"},
 597         {Opt_nogrpid, "sysvgroups"},
 598         {Opt_resgid, "resgid=%u"},
 599         {Opt_resuid, "resuid=%u"},
 600         {Opt_sb, "sb=%u"},
 601         {Opt_err_cont, "errors=continue"},
 602         {Opt_err_panic, "errors=panic"},
 603         {Opt_err_ro, "errors=remount-ro"},
 604         {Opt_nouid32, "nouid32"},
 605         {Opt_nocheck, "nocheck"},
 606         {Opt_nocheck, "check=none"},
 607         {Opt_check, "check"},
 608         {Opt_debug, "debug"},
 609         {Opt_oldalloc, "oldalloc"},
 610         {Opt_orlov, "orlov"},
 611         {Opt_user_xattr, "user_xattr"},
 612         {Opt_nouser_xattr, "nouser_xattr"},
 613         {Opt_acl, "acl"},
 614         {Opt_noacl, "noacl"},
 615         {Opt_reservation, "reservation"},
 616         {Opt_noreservation, "noreservation"},
 617         {Opt_noload, "noload"},
 618         {Opt_commit, "commit=%u"},
 619         {Opt_journal_update, "journal=update"},
 620         {Opt_journal_inum, "journal=%u"},
 621         {Opt_abort, "abort"},
 622         {Opt_data_journal, "data=journal"},
 623         {Opt_data_ordered, "data=ordered"},
 624         {Opt_data_writeback, "data=writeback"},
 625         {Opt_offusrjquota, "usrjquota="},
 626         {Opt_usrjquota, "usrjquota=%s"},
 627         {Opt_offgrpjquota, "grpjquota="},
 628         {Opt_grpjquota, "grpjquota=%s"},
 629         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 630         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 631         {Opt_tagxid, "tagxid"},
 632         {Opt_ignore, "grpquota"},
 633         {Opt_ignore, "noquota"},
 634         {Opt_ignore, "quota"},
 635         {Opt_ignore, "usrquota"},
 636         {Opt_err, NULL},
 637         {Opt_resize, "resize"}
 638
 639 };
 640
 641 static unsigned long get_sb_block(void **data)
 642 {
 643         unsigned long   sb_block;
 644         char            *options = (char *) *data;
 645
 646         if (!options || strncmp(options, "sb=", 3) != 0)
 647                 return 1;       /* Default location */
 648         options += 3;
 649         sb_block = simple_strtoul(options, &options, 0);
 650         if (*options && *options != ',') {
 651                 printk("EXT3-fs: Invalid sb specification: %s\n",
 652                        (char *) *data);
 653                 return 1;
 654         }
 655         if (*options == ',')
 656                 options++;
 657         *data = (void *) options;
 658         return sb_block;
 659 }
 660
 661 static int parse_options (char * options, struct super_block *sb,
 662                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 663 {
 664         struct ext3_sb_info *sbi = EXT3_SB(sb);
 665         char * p;
 666         substring_t args[MAX_OPT_ARGS];
 667         int data_opt = 0;
 668         int option;
 669 #ifdef CONFIG_QUOTA
 670         int qtype;
 671 #endif
 672
 673         if (!options)
 674                 return 1;
 675
 676         while ((p = strsep (&options, ",")) != NULL) {
 677                 int token;
 678                 if (!*p)
 679                         continue;
 680
 681                 token = match_token(p, tokens, args);
 682                 switch (token) {
 683                 case Opt_bsd_df:
 684                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 685                         break;
 686                 case Opt_minix_df:
 687                         set_opt (sbi->s_mount_opt, MINIX_DF);
 688                         break;
 689                 case Opt_grpid:
 690                         set_opt (sbi->s_mount_opt, GRPID);
 691                         break;
 692                 case Opt_nogrpid:
 693                         clear_opt (sbi->s_mount_opt, GRPID);
 694                         break;
 695                 case Opt_resuid:
 696                         if (match_int(&args[0], &option))
 697                                 return 0;
 698                         sbi->s_resuid = option;
 699                         break;
 700                 case Opt_resgid:
 701                         if (match_int(&args[0], &option))
 702                                 return 0;
 703                         sbi->s_resgid = option;
 704                         break;
 705                 case Opt_sb:
 706                         /* handled by get_sb_block() instead of here */
 707                         /* *sb_block = match_int(&args[0]); */
 708                         break;
 709                 case Opt_err_panic:
 710                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 711                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 712                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 713                         break;
 714                 case Opt_err_ro:
 715                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 716                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 717                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 718                         break;
 719                 case Opt_err_cont:
 720                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 721                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 722                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 723                         break;
 724                 case Opt_nouid32:
 725                         set_opt (sbi->s_mount_opt, NO_UID32);
 726                         break;
 727 #ifndef CONFIG_INOXID_NONE
 728                 case Opt_tagxid:
 729                         set_opt (sbi->s_mount_opt, TAG_XID);
 730                         break;
 731 #endif
 732                 case Opt_check:
 733 #ifdef CONFIG_EXT3_CHECK
 734                         set_opt (sbi->s_mount_opt, CHECK);
 735 #else
 736                         printk(KERN_ERR
 737                                "EXT3 Check option not supported\n");
 738 #endif
 739                         break;
 740                 case Opt_nocheck:
 741                         clear_opt (sbi->s_mount_opt, CHECK);
 742                         break;
 743                 case Opt_debug:
 744                         set_opt (sbi->s_mount_opt, DEBUG);
 745                         break;
 746                 case Opt_oldalloc:
 747                         set_opt (sbi->s_mount_opt, OLDALLOC);
 748                         break;
 749                 case Opt_orlov:
 750                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 751                         break;
 752 #ifdef CONFIG_EXT3_FS_XATTR
 753                 case Opt_user_xattr:
 754                         set_opt (sbi->s_mount_opt, XATTR_USER);
 755                         break;
 756                 case Opt_nouser_xattr:
 757                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 758                         break;
 759 #else
 760                 case Opt_user_xattr:
 761                 case Opt_nouser_xattr:
 762                         printk("EXT3 (no)user_xattr options not supported\n");
 763                         break;
 764 #endif
 765 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 766                 case Opt_acl:
 767                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 768                         break;
 769                 case Opt_noacl:
 770                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 771                         break;
 772 #else
 773                 case Opt_acl:
 774                 case Opt_noacl:
 775                         printk("EXT3 (no)acl options not supported\n");
 776                         break;
 777 #endif
 778                 case Opt_reservation:
 779                         set_opt(sbi->s_mount_opt, RESERVATION);
 780                         break;
 781                 case Opt_noreservation:
 782                         clear_opt(sbi->s_mount_opt, RESERVATION);
 783                         break;
 784                 case Opt_journal_update:
 785                         /* @@@ FIXME */
 786                         /* Eventually we will want to be able to create
 787                            a journal file here.  For now, only allow the
 788                            user to specify an existing inode to be the
 789                            journal file. */
 790                         if (is_remount) {
 791                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 792                                        "journal on remount\n");
 793                                 return 0;
 794                         }
 795                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 796                         break;
 797                 case Opt_journal_inum:
 798                         if (is_remount) {
 799                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 800                                        "journal on remount\n");
 801                                 return 0;
 802                         }
 803                         if (match_int(&args[0], &option))
 804                                 return 0;
 805                         *inum = option;
 806                         break;
 807                 case Opt_noload:
 808                         set_opt (sbi->s_mount_opt, NOLOAD);
 809                         break;
 810                 case Opt_commit:
 811                         if (match_int(&args[0], &option))
 812                                 return 0;
 813                         if (option < 0)
 814                                 return 0;
 815                         if (option == 0)
 816                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 817                         sbi->s_commit_interval = HZ * option;
 818                         break;
 819                 case Opt_data_journal:
 820                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 821                         goto datacheck;
 822                 case Opt_data_ordered:
 823                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 824                         goto datacheck;
 825                 case Opt_data_writeback:
 826                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 827                 datacheck:
 828                         if (is_remount) {
 829                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 830                                                 != data_opt) {
 831                                         printk(KERN_ERR
 832                                                 "EXT3-fs: cannot change data "
 833                                                 "mode on remount\n");
 834                                         return 0;
 835                                 }
 836                         } else {
 837                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 838                                 sbi->s_mount_opt |= data_opt;
 839                         }
 840                         break;
 841 #ifdef CONFIG_QUOTA
 842                 case Opt_usrjquota:
 843                         qtype = USRQUOTA;
 844                         goto set_qf_name;
 845                 case Opt_grpjquota:
 846                         qtype = GRPQUOTA;
 847 set_qf_name:
 848                         if (sb_any_quota_enabled(sb)) {
 849                                 printk(KERN_ERR
 850                                         "EXT3-fs: Cannot change journalled "
 851                                         "quota options when quota turned on.\n");
 852                                 return 0;
 853                         }
 854                         if (sbi->s_qf_names[qtype]) {
 855                                 printk(KERN_ERR
 856                                         "EXT3-fs: %s quota file already "
 857                                         "specified.\n", QTYPE2NAME(qtype));
 858                                 return 0;
 859                         }
 860                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 861                         if (!sbi->s_qf_names[qtype]) {
 862                                 printk(KERN_ERR
 863                                         "EXT3-fs: not enough memory for "
 864                                         "storing quotafile name.\n");
 865                                 return 0;
 866                         }
 867                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 868                                 printk(KERN_ERR
 869                                         "EXT3-fs: quotafile must be on "
 870                                         "filesystem root.\n");
 871                                 kfree(sbi->s_qf_names[qtype]);
 872                                 sbi->s_qf_names[qtype] = NULL;
 873                                 return 0;
 874                         }
 875                         break;
 876                 case Opt_offusrjquota:
 877                         qtype = USRQUOTA;
 878                         goto clear_qf_name;
 879                 case Opt_offgrpjquota:
 880                         qtype = GRPQUOTA;
 881 clear_qf_name:
 882                         if (sb_any_quota_enabled(sb)) {
 883                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 884                                         "journalled quota options when "
 885                                         "quota turned on.\n");
 886                                 return 0;
 887                         }
 888                         if (sbi->s_qf_names[qtype]) {
 889                                 kfree(sbi->s_qf_names[qtype]);
 890                                 sbi->s_qf_names[qtype] = NULL;
 891                         }
 892                         break;
 893                 case Opt_jqfmt_vfsold:
 894                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 895                         break;
 896                 case Opt_jqfmt_vfsv0:
 897                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 898                         break;
 899 #else
 900                 case Opt_usrjquota:
 901                 case Opt_grpjquota:
 902                 case Opt_offusrjquota:
 903                 case Opt_offgrpjquota:
 904                 case Opt_jqfmt_vfsold:
 905                 case Opt_jqfmt_vfsv0:
 906                         printk(KERN_ERR
 907                                 "EXT3-fs: journalled quota options not "
 908                                 "supported.\n");
 909                         break;
 910 #endif
 911                 case Opt_abort:
 912                         set_opt(sbi->s_mount_opt, ABORT);
 913                         break;
 914                 case Opt_ignore:
 915                         break;
 916                 case Opt_resize:
 917                         if (!n_blocks_count) {
 918                                 printk("EXT3-fs: resize option only available "
 919                                         "for remount\n");
 920                                 return 0;
 921                         }
 922                         match_int(&args[0], &option);
 923                         *n_blocks_count = option;
 924                         break;
 925                 default:
 926                         printk (KERN_ERR
 927                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 928                                 "or missing value\n", p);
 929                         return 0;
 930                 }
 931         }
 932 #ifdef CONFIG_QUOTA
 933         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 934             sbi->s_qf_names[GRPQUOTA])) {
 935                 printk(KERN_ERR
 936                         "EXT3-fs: journalled quota format not specified.\n");
 937                 return 0;
 938         }
 939 #endif
 940
 941         return 1;
 942 }
 943
 944 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 945                             int read_only)
 946 {
 947         struct ext3_sb_info *sbi = EXT3_SB(sb);
 948         int res = 0;
 949
 950         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 951                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 952                         "forcing read-only mode\n");
 953                 res = MS_RDONLY;
 954         }
 955         if (read_only)
 956                 return res;
 957         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 958                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 959                         "running e2fsck is recommended\n");
 960         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 961                 printk (KERN_WARNING
 962                         "EXT3-fs warning: mounting fs with errors, "
 963                         "running e2fsck is recommended\n");
 964         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 965                  le16_to_cpu(es->s_mnt_count) >=
 966                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 967                 printk (KERN_WARNING
 968                         "EXT3-fs warning: maximal mount count reached, "
 969                         "running e2fsck is recommended\n");
 970         else if (le32_to_cpu(es->s_checkinterval) &&
 971                 (le32_to_cpu(es->s_lastcheck) +
 972                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 973                 printk (KERN_WARNING
 974                         "EXT3-fs warning: checktime reached, "
 975                         "running e2fsck is recommended\n");
 976 #if 0
 977                 /* @@@ We _will_ want to clear the valid bit if we find
 978                    inconsistencies, to force a fsck at reboot.  But for
 979                    a plain journaled filesystem we can keep it set as
 980                    valid forever! :) */
 981         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 982 #endif
 983         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 984                 es->s_max_mnt_count =
 985                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 986         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 987         es->s_mtime = cpu_to_le32(get_seconds());
 988         ext3_update_dynamic_rev(sb);
 989         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 990
 991         ext3_commit_super(sb, es, 1);
 992         if (test_opt(sb, DEBUG))
 993                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 994                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 995                         sb->s_blocksize,
 996                         sbi->s_groups_count,
 997                         EXT3_BLOCKS_PER_GROUP(sb),
 998                         EXT3_INODES_PER_GROUP(sb),
 999                         sbi->s_mount_opt);
1000
1001         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1002         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1003                 char b[BDEVNAME_SIZE];
1004
1005                 printk("external journal on %s\n",
1006                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1007         } else {
1008                 printk("internal journal\n");
1009         }
1010 #ifdef CONFIG_EXT3_CHECK
1011         if (test_opt (sb, CHECK)) {
1012                 ext3_check_blocks_bitmap (sb);
1013                 ext3_check_inodes_bitmap (sb);
1014         }
1015 #endif
1016         return res;
1017 }
1018
1019 static int ext3_check_descriptors (struct super_block * sb)
1020 {
1021         struct ext3_sb_info *sbi = EXT3_SB(sb);
1022         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1023         struct ext3_group_desc * gdp = NULL;
1024         int desc_block = 0;
1025         int i;
1026
1027         ext3_debug ("Checking group descriptors");
1028
1029         for (i = 0; i < sbi->s_groups_count; i++)
1030         {
1031                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1032                         gdp = (struct ext3_group_desc *)
1033                                         sbi->s_group_desc[desc_block++]->b_data;
1034                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1035                     le32_to_cpu(gdp->bg_block_bitmap) >=
1036                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1037                 {
1038                         ext3_error (sb, "ext3_check_descriptors",
1039                                     "Block bitmap for group %d"
1040                                     " not in group (block %lu)!",
1041                                     i, (unsigned long)
1042                                         le32_to_cpu(gdp->bg_block_bitmap));
1043                         return 0;
1044                 }
1045                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1046                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1047                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1048                 {
1049                         ext3_error (sb, "ext3_check_descriptors",
1050                                     "Inode bitmap for group %d"
1051                                     " not in group (block %lu)!",
1052                                     i, (unsigned long)
1053                                         le32_to_cpu(gdp->bg_inode_bitmap));
1054                         return 0;
1055                 }
1056                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1057                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1058                     block + EXT3_BLOCKS_PER_GROUP(sb))
1059                 {
1060                         ext3_error (sb, "ext3_check_descriptors",
1061                                     "Inode table for group %d"
1062                                     " not in group (block %lu)!",
1063                                     i, (unsigned long)
1064                                         le32_to_cpu(gdp->bg_inode_table));
1065                         return 0;
1066                 }
1067                 block += EXT3_BLOCKS_PER_GROUP(sb);
1068                 gdp++;
1069         }
1070
1071         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1072         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1073         return 1;
1074 }
1075
1076
1077 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1078  * the superblock) which were deleted from all directories, but held open by
1079  * a process at the time of a crash.  We walk the list and try to delete these
1080  * inodes at recovery time (only with a read-write filesystem).
1081  *
1082  * In order to keep the orphan inode chain consistent during traversal (in
1083  * case of crash during recovery), we link each inode into the superblock
1084  * orphan list_head and handle it the same way as an inode deletion during
1085  * normal operation (which journals the operations for us).
1086  *
1087  * We only do an iget() and an iput() on each inode, which is very safe if we
1088  * accidentally point at an in-use or already deleted inode.  The worst that
1089  * can happen in this case is that we get a "bit already cleared" message from
1090  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1091  * e2fsck was run on this filesystem, and it must have already done the orphan
1092  * inode cleanup for us, so we can safely abort without any further action.
1093  */
1094 static void ext3_orphan_cleanup (struct super_block * sb,
1095                                  struct ext3_super_block * es)
1096 {
1097         unsigned int s_flags = sb->s_flags;
1098         int nr_orphans = 0, nr_truncates = 0;
1099 #ifdef CONFIG_QUOTA
1100         int i;
1101 #endif
1102         if (!es->s_last_orphan) {
1103                 jbd_debug(4, "no orphan inodes to clean up\n");
1104                 return;
1105         }
1106
1107         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1108                 if (es->s_last_orphan)
1109                         jbd_debug(1, "Errors on filesystem, "
1110                                   "clearing orphan list.\n");
1111                 es->s_last_orphan = 0;
1112                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1113                 return;
1114         }
1115
1116         if (s_flags & MS_RDONLY) {
1117                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1118                        sb->s_id);
1119                 sb->s_flags &= ~MS_RDONLY;
1120         }
1121 #ifdef CONFIG_QUOTA
1122         /* Needed for iput() to work correctly and not trash data */
1123         sb->s_flags |= MS_ACTIVE;
1124         /* Turn on quotas so that they are updated correctly */
1125         for (i = 0; i < MAXQUOTAS; i++) {
1126                 if (EXT3_SB(sb)->s_qf_names[i]) {
1127                         int ret = ext3_quota_on_mount(sb, i);
1128                         if (ret < 0)
1129                                 printk(KERN_ERR
1130                                         "EXT3-fs: Cannot turn on journalled "
1131                                         "quota: error %d\n", ret);
1132                 }
1133         }
1134 #endif
1135
1136         while (es->s_last_orphan) {
1137                 struct inode *inode;
1138
1139                 if (!(inode =
1140                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1141                         es->s_last_orphan = 0;
1142                         break;
1143                 }
1144
1145                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1146                 DQUOT_INIT(inode);
1147                 if (inode->i_nlink) {
1148                         printk(KERN_DEBUG
1149                                 "%s: truncating inode %ld to %Ld bytes\n",
1150                                 __FUNCTION__, inode->i_ino, inode->i_size);
1151                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1152                                   inode->i_ino, inode->i_size);
1153                         ext3_truncate(inode);
1154                         nr_truncates++;
1155                 } else {
1156                         printk(KERN_DEBUG
1157                                 "%s: deleting unreferenced inode %ld\n",
1158                                 __FUNCTION__, inode->i_ino);
1159                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1160                                   inode->i_ino);
1161                         nr_orphans++;
1162                 }
1163                 iput(inode);  /* The delete magic happens here! */
1164         }
1165
1166 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1167
1168         if (nr_orphans)
1169                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1170                        sb->s_id, PLURAL(nr_orphans));
1171         if (nr_truncates)
1172                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1173                        sb->s_id, PLURAL(nr_truncates));
1174 #ifdef CONFIG_QUOTA
1175         /* Turn quotas off */
1176         for (i = 0; i < MAXQUOTAS; i++) {
1177                 if (sb_dqopt(sb)->files[i])
1178                         ext3_quota_off_mount(sb, i);
1179         }
1180 #endif
1181         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1182 }
1183
1184 #define log2(n) ffz(~(n))
1185
1186 /*
1187  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1188  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1189  * We need to be 1 filesystem block less than the 2^32 sector limit.
1190  */
1191 static loff_t ext3_max_size(int bits)
1192 {
1193         loff_t res = EXT3_NDIR_BLOCKS;
1194         res += 1LL << (bits-2);
1195         res += 1LL << (2*(bits-2));
1196         res += 1LL << (3*(bits-2));
1197         res <<= bits;
1198         if (res > (512LL << 32) - (1 << bits))
1199                 res = (512LL << 32) - (1 << bits);
1200         return res;
1201 }
1202
1203 static unsigned long descriptor_loc(struct super_block *sb,
1204                                     unsigned long logic_sb_block,
1205                                     int nr)
1206 {
1207         struct ext3_sb_info *sbi = EXT3_SB(sb);
1208         unsigned long bg, first_data_block, first_meta_bg;
1209         int has_super = 0;
1210
1211         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1212         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1213
1214         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1215             nr < first_meta_bg)
1216                 return (logic_sb_block + nr + 1);
1217         bg = sbi->s_desc_per_block * nr;
1218         if (ext3_bg_has_super(sb, bg))
1219                 has_super = 1;
1220         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1221 }
1222
1223
1224 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1225 {
1226         struct buffer_head * bh;
1227         struct ext3_super_block *es = 0;
1228         struct ext3_sb_info *sbi;
1229         unsigned long block;
1230         unsigned long sb_block = get_sb_block(&data);
1231         unsigned long logic_sb_block;
1232         unsigned long offset = 0;
1233         unsigned long journal_inum = 0;
1234         unsigned long def_mount_opts;
1235         struct inode *root;
1236         int blocksize;
1237         int hblock;
1238         int db_count;
1239         int i;
1240         int needs_recovery;
1241
1242         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1243         if (!sbi)
1244                 return -ENOMEM;
1245         sb->s_fs_info = sbi;
1246         memset(sbi, 0, sizeof(*sbi));
1247         sbi->s_mount_opt = 0;
1248         sbi->s_resuid = EXT3_DEF_RESUID;
1249         sbi->s_resgid = EXT3_DEF_RESGID;
1250
1251         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1252         if (!blocksize) {
1253                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1254                 goto out_fail;
1255         }
1256
1257         /*
1258          * The ext3 superblock will not be buffer aligned for other than 1kB
1259          * block sizes.  We need to calculate the offset from buffer start.
1260          */
1261         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1262                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1263                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1264         } else {
1265                 logic_sb_block = sb_block;
1266         }
1267
1268         if (!(bh = sb_bread(sb, logic_sb_block))) {
1269                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1270                 goto out_fail;
1271         }
1272         /*
1273          * Note: s_es must be initialized as soon as possible because
1274          *       some ext3 macro-instructions depend on its value
1275          */
1276         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1277         sbi->s_es = es;
1278         sb->s_magic = le16_to_cpu(es->s_magic);
1279         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1280                 if (!silent)
1281                         printk(KERN_ERR
1282                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1283                                sb->s_id);
1284                 goto failed_mount;
1285         }
1286
1287         /* Set defaults before we parse the mount options */
1288         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1289         if (def_mount_opts & EXT3_DEFM_DEBUG)
1290                 set_opt(sbi->s_mount_opt, DEBUG);
1291         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1292                 set_opt(sbi->s_mount_opt, GRPID);
1293         if (def_mount_opts & EXT3_DEFM_UID16)
1294                 set_opt(sbi->s_mount_opt, NO_UID32);
1295         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1296                 set_opt(sbi->s_mount_opt, XATTR_USER);
1297         if (def_mount_opts & EXT3_DEFM_ACL)
1298                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1299         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1300                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1301         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1302                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1303         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1304                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1305
1306         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1307                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1308         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1309                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1310
1311         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1312         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1313
1314         set_opt(sbi->s_mount_opt, RESERVATION);
1315
1316         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1317                 goto failed_mount;
1318
1319         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1320                 sb->s_flags |= MS_TAGXID;
1321         sb->s_flags |= MS_ONE_SECOND;
1322         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1323                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1324
1325         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1326             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1327              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1328              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1329                 printk(KERN_WARNING
1330                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1331                        "running e2fsck is recommended\n");
1332         /*
1333          * Check feature flags regardless of the revision level, since we
1334          * previously didn't change the revision level when setting the flags,
1335          * so there is a chance incompat flags are set on a rev 0 filesystem.
1336          */
1337         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1338                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1339                        "unsupported optional features (%x).\n",
1340                        sb->s_id, i);
1341                 goto failed_mount;
1342         }
1343         if (!(sb->s_flags & MS_RDONLY) &&
1344             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1345                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1346                        "unsupported optional features (%x).\n",
1347                        sb->s_id, i);
1348                 goto failed_mount;
1349         }
1350         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1351
1352         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1353             blocksize > EXT3_MAX_BLOCK_SIZE) {
1354                 printk(KERN_ERR
1355                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1356                        blocksize, sb->s_id);
1357                 goto failed_mount;
1358         }
1359
1360         hblock = bdev_hardsect_size(sb->s_bdev);
1361         if (sb->s_blocksize != blocksize) {
1362                 /*
1363                  * Make sure the blocksize for the filesystem is larger
1364                  * than the hardware sectorsize for the machine.
1365                  */
1366                 if (blocksize < hblock) {
1367                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1368                                "device blocksize %d.\n", blocksize, hblock);
1369                         goto failed_mount;
1370                 }
1371
1372                 brelse (bh);
1373                 sb_set_blocksize(sb, blocksize);
1374                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1375                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1376                 bh = sb_bread(sb, logic_sb_block);
1377                 if (!bh) {
1378                         printk(KERN_ERR
1379                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1380                         goto failed_mount;
1381                 }
1382                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1383                 sbi->s_es = es;
1384                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1385                         printk (KERN_ERR
1386                                 "EXT3-fs: Magic mismatch, very weird !\n");
1387                         goto failed_mount;
1388                 }
1389         }
1390
1391         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1392
1393         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1394                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1395                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1396         } else {
1397                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1398                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1399                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1400                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1401                     (sbi->s_inode_size > blocksize)) {
1402                         printk (KERN_ERR
1403                                 "EXT3-fs: unsupported inode size: %d\n",
1404                                 sbi->s_inode_size);
1405                         goto failed_mount;
1406                 }
1407         }
1408         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1409                                    le32_to_cpu(es->s_log_frag_size);
1410         if (blocksize != sbi->s_frag_size) {
1411                 printk(KERN_ERR
1412                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1413                        sbi->s_frag_size, blocksize);
1414                 goto failed_mount;
1415         }
1416         sbi->s_frags_per_block = 1;
1417         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1418         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1419         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1420         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1421         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1422         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1423         sbi->s_sbh = bh;
1424         sbi->s_mount_state = le16_to_cpu(es->s_state);
1425         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1426         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1427         for (i=0; i < 4; i++)
1428                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1429         sbi->s_def_hash_version = es->s_def_hash_version;
1430
1431         if (sbi->s_blocks_per_group > blocksize * 8) {
1432                 printk (KERN_ERR
1433                         "EXT3-fs: #blocks per group too big: %lu\n",
1434                         sbi->s_blocks_per_group);
1435                 goto failed_mount;
1436         }
1437         if (sbi->s_frags_per_group > blocksize * 8) {
1438                 printk (KERN_ERR
1439                         "EXT3-fs: #fragments per group too big: %lu\n",
1440                         sbi->s_frags_per_group);
1441                 goto failed_mount;
1442         }
1443         if (sbi->s_inodes_per_group > blocksize * 8) {
1444                 printk (KERN_ERR
1445                         "EXT3-fs: #inodes per group too big: %lu\n",
1446                         sbi->s_inodes_per_group);
1447                 goto failed_mount;
1448         }
1449
1450         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1451                                le32_to_cpu(es->s_first_data_block) +
1452                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1453                               EXT3_BLOCKS_PER_GROUP(sb);
1454         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1455                    EXT3_DESC_PER_BLOCK(sb);
1456         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1457                                     GFP_KERNEL);
1458         if (sbi->s_group_desc == NULL) {
1459                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1460                 goto failed_mount;
1461         }
1462         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1463                         GFP_KERNEL);
1464         if (!sbi->s_debts) {
1465                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1466                 goto failed_mount2;
1467         }
1468         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1469
1470         percpu_counter_init(&sbi->s_freeblocks_counter);
1471         percpu_counter_init(&sbi->s_freeinodes_counter);
1472         percpu_counter_init(&sbi->s_dirs_counter);
1473         bgl_lock_init(&sbi->s_blockgroup_lock);
1474
1475         for (i = 0; i < db_count; i++) {
1476                 block = descriptor_loc(sb, logic_sb_block, i);
1477                 sbi->s_group_desc[i] = sb_bread(sb, block);
1478                 if (!sbi->s_group_desc[i]) {
1479                         printk (KERN_ERR "EXT3-fs: "
1480                                 "can't read group descriptor %d\n", i);
1481                         db_count = i;
1482                         goto failed_mount2;
1483                 }
1484         }
1485         if (!ext3_check_descriptors (sb)) {
1486                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1487                 goto failed_mount2;
1488         }
1489         sbi->s_gdb_count = db_count;
1490         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1491         spin_lock_init(&sbi->s_next_gen_lock);
1492         /* per fileystem reservation list head & lock */
1493         spin_lock_init(&sbi->s_rsv_window_lock);
1494         INIT_LIST_HEAD(&sbi->s_rsv_window_head.rsv_list);
1495         sbi->s_rsv_window_head.rsv_start = 0;
1496         sbi->s_rsv_window_head.rsv_end = 0;
1497         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1498         atomic_set(&sbi->s_rsv_window_head.rsv_goal_size, 0);
1499
1500         /*
1501          * set up enough so that it can read an inode
1502          */
1503         sb->s_op = &ext3_sops;
1504         sb->s_export_op = &ext3_export_ops;
1505 #ifdef CONFIG_QUOTA
1506         sb->s_qcop = &ext3_qctl_operations;
1507         sb->dq_op = &ext3_quota_operations;
1508 #endif
1509         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1510
1511         sb->s_root = 0;
1512
1513         needs_recovery = (es->s_last_orphan != 0 ||
1514                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1515                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1516
1517         /*
1518          * The first inode we look at is the journal inode.  Don't try
1519          * root first: it may be modified in the journal!
1520          */
1521         if (!test_opt(sb, NOLOAD) &&
1522             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1523                 if (ext3_load_journal(sb, es))
1524                         goto failed_mount2;
1525         } else if (journal_inum) {
1526                 if (ext3_create_journal(sb, es, journal_inum))
1527                         goto failed_mount2;
1528         } else {
1529                 if (!silent)
1530                         printk (KERN_ERR
1531                                 "ext3: No journal on filesystem on %s\n",
1532                                 sb->s_id);
1533                 goto failed_mount2;
1534         }
1535
1536         /* We have now updated the journal if required, so we can
1537          * validate the data journaling mode. */
1538         switch (test_opt(sb, DATA_FLAGS)) {
1539         case 0:
1540                 /* No mode set, assume a default based on the journal
1541                    capabilities: ORDERED_DATA if the journal can
1542                    cope, else JOURNAL_DATA */
1543                 if (journal_check_available_features
1544                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1545                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1546                 else
1547                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1548                 break;
1549
1550         case EXT3_MOUNT_ORDERED_DATA:
1551         case EXT3_MOUNT_WRITEBACK_DATA:
1552                 if (!journal_check_available_features
1553                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1554                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1555                                "requested data journaling mode\n");
1556                         goto failed_mount3;
1557                 }
1558         default:
1559                 break;
1560         }
1561
1562         /*
1563          * The journal_load will have done any necessary log recovery,
1564          * so we can safely mount the rest of the filesystem now.
1565          */
1566
1567         root = iget(sb, EXT3_ROOT_INO);
1568         sb->s_root = d_alloc_root(root);
1569         if (!sb->s_root) {
1570                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1571                 iput(root);
1572                 goto failed_mount3;
1573         }
1574         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1575                 dput(sb->s_root);
1576                 sb->s_root = NULL;
1577                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1578                 goto failed_mount3;
1579         }
1580
1581         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1582         /*
1583          * akpm: core read_super() calls in here with the superblock locked.
1584          * That deadlocks, because orphan cleanup needs to lock the superblock
1585          * in numerous places.  Here we just pop the lock - it's relatively
1586          * harmless, because we are now ready to accept write_super() requests,
1587          * and aviro says that's the only reason for hanging onto the
1588          * superblock lock.
1589          */
1590         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1591         ext3_orphan_cleanup(sb, es);
1592         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1593         if (needs_recovery)
1594                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1595         ext3_mark_recovery_complete(sb, es);
1596         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1597                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1598                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1599                 "writeback");
1600
1601         percpu_counter_mod(&sbi->s_freeblocks_counter,
1602                 ext3_count_free_blocks(sb));
1603         percpu_counter_mod(&sbi->s_freeinodes_counter,
1604                 ext3_count_free_inodes(sb));
1605         percpu_counter_mod(&sbi->s_dirs_counter,
1606                 ext3_count_dirs(sb));
1607
1608         return 0;
1609
1610 failed_mount3:
1611         journal_destroy(sbi->s_journal);
1612 failed_mount2:
1613         kfree(sbi->s_debts);
1614         for (i = 0; i < db_count; i++)
1615                 brelse(sbi->s_group_desc[i]);
1616         kfree(sbi->s_group_desc);
1617 failed_mount:
1618 #ifdef CONFIG_QUOTA
1619         for (i = 0; i < MAXQUOTAS; i++) {
1620                 if (sbi->s_qf_names[i])
1621                         kfree(sbi->s_qf_names[i]);
1622         }
1623 #endif
1624         ext3_blkdev_remove(sbi);
1625         brelse(bh);
1626 out_fail:
1627         sb->s_fs_info = NULL;
1628         kfree(sbi);
1629         return -EINVAL;
1630 }
1631
1632 /*
1633  * Setup any per-fs journal parameters now.  We'll do this both on
1634  * initial mount, once the journal has been initialised but before we've
1635  * done any recovery; and again on any subsequent remount.
1636  */
1637 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1638                                      journal_t *journal)
1639 {
1640         if (sbi->s_commit_interval)
1641                 journal->j_commit_interval = sbi->s_commit_interval;
1642         /* We could also set up an ext3-specific default for the commit
1643          * interval here, but for now we'll just fall back to the jbd
1644          * default. */
1645 }
1646
1647
1648 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1649 {
1650         struct inode *journal_inode;
1651         journal_t *journal;
1652
1653         /* First, test for the existence of a valid inode on disk.  Bad
1654          * things happen if we iget() an unused inode, as the subsequent
1655          * iput() will try to delete it. */
1656
1657         journal_inode = iget(sb, journal_inum);
1658         if (!journal_inode) {
1659                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1660                 return NULL;
1661         }
1662         if (!journal_inode->i_nlink) {
1663                 make_bad_inode(journal_inode);
1664                 iput(journal_inode);
1665                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1666                 return NULL;
1667         }
1668
1669         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1670                   journal_inode, journal_inode->i_size);
1671         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1672                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1673                 iput(journal_inode);
1674                 return NULL;
1675         }
1676
1677         journal = journal_init_inode(journal_inode);
1678         if (!journal) {
1679                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1680                 iput(journal_inode);
1681         }
1682         journal->j_private = sb;
1683         ext3_init_journal_params(EXT3_SB(sb), journal);
1684         return journal;
1685 }
1686
1687 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1688                                        dev_t j_dev)
1689 {
1690         struct buffer_head * bh;
1691         journal_t *journal;
1692         int start;
1693         int len;
1694         int hblock, blocksize;
1695         unsigned long sb_block;
1696         unsigned long offset;
1697         struct ext3_super_block * es;
1698         struct block_device *bdev;
1699
1700         bdev = ext3_blkdev_get(j_dev);
1701         if (bdev == NULL)
1702                 return NULL;
1703
1704         if (bd_claim(bdev, sb)) {
1705                 printk(KERN_ERR
1706                         "EXT3: failed to claim external journal device.\n");
1707                 blkdev_put(bdev);
1708                 return NULL;
1709         }
1710
1711         blocksize = sb->s_blocksize;
1712         hblock = bdev_hardsect_size(bdev);
1713         if (blocksize < hblock) {
1714                 printk(KERN_ERR
1715                         "EXT3-fs: blocksize too small for journal device.\n");
1716                 goto out_bdev;
1717         }
1718
1719         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1720         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1721         set_blocksize(bdev, blocksize);
1722         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1723                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1724                        "external journal\n");
1725                 goto out_bdev;
1726         }
1727
1728         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1729         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1730             !(le32_to_cpu(es->s_feature_incompat) &
1731               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1732                 printk(KERN_ERR "EXT3-fs: external journal has "
1733                                         "bad superblock\n");
1734                 brelse(bh);
1735                 goto out_bdev;
1736         }
1737
1738         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1739                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1740                 brelse(bh);
1741                 goto out_bdev;
1742         }
1743
1744         len = le32_to_cpu(es->s_blocks_count);
1745         start = sb_block + 1;
1746         brelse(bh);     /* we're done with the superblock */
1747
1748         journal = journal_init_dev(bdev, sb->s_bdev,
1749                                         start, len, blocksize);
1750         if (!journal) {
1751                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1752                 goto out_bdev;
1753         }
1754         journal->j_private = sb;
1755         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1756         wait_on_buffer(journal->j_sb_buffer);
1757         if (!buffer_uptodate(journal->j_sb_buffer)) {
1758                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1759                 goto out_journal;
1760         }
1761         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1762                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1763                                         "user (unsupported) - %d\n",
1764                         ntohl(journal->j_superblock->s_nr_users));
1765                 goto out_journal;
1766         }
1767         EXT3_SB(sb)->journal_bdev = bdev;
1768         ext3_init_journal_params(EXT3_SB(sb), journal);
1769         return journal;
1770 out_journal:
1771         journal_destroy(journal);
1772 out_bdev:
1773         ext3_blkdev_put(bdev);
1774         return NULL;
1775 }
1776
1777 static int ext3_load_journal(struct super_block * sb,
1778                              struct ext3_super_block * es)
1779 {
1780         journal_t *journal;
1781         int journal_inum = le32_to_cpu(es->s_journal_inum);
1782         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1783         int err = 0;
1784         int really_read_only;
1785
1786         really_read_only = bdev_read_only(sb->s_bdev);
1787
1788         /*
1789          * Are we loading a blank journal or performing recovery after a
1790          * crash?  For recovery, we need to check in advance whether we
1791          * can get read-write access to the device.
1792          */
1793
1794         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1795                 if (sb->s_flags & MS_RDONLY) {
1796                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1797                                         "required on readonly filesystem.\n");
1798                         if (really_read_only) {
1799                                 printk(KERN_ERR "EXT3-fs: write access "
1800                                         "unavailable, cannot proceed.\n");
1801                                 return -EROFS;
1802                         }
1803                         printk (KERN_INFO "EXT3-fs: write access will "
1804                                         "be enabled during recovery.\n");
1805                 }
1806         }
1807
1808         if (journal_inum && journal_dev) {
1809                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1810                        "and inode journals!\n");
1811                 return -EINVAL;
1812         }
1813
1814         if (journal_inum) {
1815                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1816                         return -EINVAL;
1817         } else {
1818                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1819                         return -EINVAL;
1820         }
1821
1822         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1823                 err = journal_update_format(journal);
1824                 if (err)  {
1825                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1826                         journal_destroy(journal);
1827                         return err;
1828                 }
1829         }
1830
1831         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1832                 err = journal_wipe(journal, !really_read_only);
1833         if (!err)
1834                 err = journal_load(journal);
1835
1836         if (err) {
1837                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1838                 journal_destroy(journal);
1839                 return err;
1840         }
1841
1842         EXT3_SB(sb)->s_journal = journal;
1843         ext3_clear_journal_err(sb, es);
1844         return 0;
1845 }
1846
1847 static int ext3_create_journal(struct super_block * sb,
1848                                struct ext3_super_block * es,
1849                                int journal_inum)
1850 {
1851         journal_t *journal;
1852
1853         if (sb->s_flags & MS_RDONLY) {
1854                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1855                                 "create journal.\n");
1856                 return -EROFS;
1857         }
1858
1859         if (!(journal = ext3_get_journal(sb, journal_inum)))
1860                 return -EINVAL;
1861
1862         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1863                journal_inum);
1864
1865         if (journal_create(journal)) {
1866                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1867                 journal_destroy(journal);
1868                 return -EIO;
1869         }
1870
1871         EXT3_SB(sb)->s_journal = journal;
1872
1873         ext3_update_dynamic_rev(sb);
1874         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1875         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1876
1877         es->s_journal_inum = cpu_to_le32(journal_inum);
1878         sb->s_dirt = 1;
1879
1880         /* Make sure we flush the recovery flag to disk. */
1881         ext3_commit_super(sb, es, 1);
1882
1883         return 0;
1884 }
1885
1886 static void ext3_commit_super (struct super_block * sb,
1887                                struct ext3_super_block * es,
1888                                int sync)
1889 {
1890         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1891
1892         if (!sbh)
1893                 return;
1894         es->s_wtime = cpu_to_le32(get_seconds());
1895         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1896         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1897         BUFFER_TRACE(sbh, "marking dirty");
1898         mark_buffer_dirty(sbh);
1899         if (sync)
1900                 sync_dirty_buffer(sbh);
1901 }
1902
1903
1904 /*
1905  * Have we just finished recovery?  If so, and if we are mounting (or
1906  * remounting) the filesystem readonly, then we will end up with a
1907  * consistent fs on disk.  Record that fact.
1908  */
1909 static void ext3_mark_recovery_complete(struct super_block * sb,
1910                                         struct ext3_super_block * es)
1911 {
1912         journal_flush(EXT3_SB(sb)->s_journal);
1913         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1914             sb->s_flags & MS_RDONLY) {
1915                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1916                 sb->s_dirt = 0;
1917                 ext3_commit_super(sb, es, 1);
1918         }
1919 }
1920
1921 /*
1922  * If we are mounting (or read-write remounting) a filesystem whose journal
1923  * has recorded an error from a previous lifetime, move that error to the
1924  * main filesystem now.
1925  */
1926 static void ext3_clear_journal_err(struct super_block * sb,
1927                                    struct ext3_super_block * es)
1928 {
1929         journal_t *journal;
1930         int j_errno;
1931         const char *errstr;
1932
1933         journal = EXT3_SB(sb)->s_journal;
1934
1935         /*
1936          * Now check for any error status which may have been recorded in the
1937          * journal by a prior ext3_error() or ext3_abort()
1938          */
1939
1940         j_errno = journal_errno(journal);
1941         if (j_errno) {
1942                 char nbuf[16];
1943
1944                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1945                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1946                              "from previous mount: %s", errstr);
1947                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1948                              "filesystem check.");
1949
1950                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1951                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1952                 ext3_commit_super (sb, es, 1);
1953
1954                 journal_clear_err(journal);
1955         }
1956 }
1957
1958 /*
1959  * Force the running and committing transactions to commit,
1960  * and wait on the commit.
1961  */
1962 int ext3_force_commit(struct super_block *sb)
1963 {
1964         journal_t *journal;
1965         int ret;
1966
1967         if (sb->s_flags & MS_RDONLY)
1968                 return 0;
1969
1970         journal = EXT3_SB(sb)->s_journal;
1971         sb->s_dirt = 0;
1972         ret = ext3_journal_force_commit(journal);
1973         return ret;
1974 }
1975
1976 /*
1977  * Ext3 always journals updates to the superblock itself, so we don't
1978  * have to propagate any other updates to the superblock on disk at this
1979  * point.  Just start an async writeback to get the buffers on their way
1980  * to the disk.
1981  *
1982  * This implicitly triggers the writebehind on sync().
1983  */
1984
1985 void ext3_write_super (struct super_block * sb)
1986 {
1987         if (down_trylock(&sb->s_lock) == 0)
1988                 BUG();
1989         sb->s_dirt = 0;
1990 }
1991
1992 static int ext3_sync_fs(struct super_block *sb, int wait)
1993 {
1994         tid_t target;
1995
1996         sb->s_dirt = 0;
1997         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1998                 if (wait)
1999                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2000         }
2001         return 0;
2002 }
2003
2004 /*
2005  * LVM calls this function before a (read-only) snapshot is created.  This
2006  * gives us a chance to flush the journal completely and mark the fs clean.
2007  */
2008 void ext3_write_super_lockfs(struct super_block *sb)
2009 {
2010         sb->s_dirt = 0;
2011
2012         if (!(sb->s_flags & MS_RDONLY)) {
2013                 journal_t *journal = EXT3_SB(sb)->s_journal;
2014
2015                 /* Now we set up the journal barrier. */
2016                 journal_lock_updates(journal);
2017                 journal_flush(journal);
2018
2019                 /* Journal blocked and flushed, clear needs_recovery flag. */
2020                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2021                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2022         }
2023 }
2024
2025 /*
2026  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2027  * flag here, even though the filesystem is not technically dirty yet.
2028  */
2029 void ext3_unlockfs(struct super_block *sb)
2030 {
2031         if (!(sb->s_flags & MS_RDONLY)) {
2032                 lock_super(sb);
2033                 /* Reser the needs_recovery flag before the fs is unlocked. */
2034                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2035                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2036                 unlock_super(sb);
2037                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2038         }
2039 }
2040
2041 int ext3_remount (struct super_block * sb, int * flags, char * data)
2042 {
2043         struct ext3_super_block * es;
2044         struct ext3_sb_info *sbi = EXT3_SB(sb);
2045         unsigned long tmp;
2046         unsigned long n_blocks_count = 0;
2047
2048         /*
2049          * Allow the "check" option to be passed as a remount option.
2050          */
2051         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2052                 return -EINVAL;
2053
2054         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2055                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2056
2057         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2058                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2059
2060         es = sbi->s_es;
2061
2062         ext3_init_journal_params(sbi, sbi->s_journal);
2063
2064         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2065                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2066                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2067                         return -EROFS;
2068
2069                 if (*flags & MS_RDONLY) {
2070                         /*
2071                          * First of all, the unconditional stuff we have to do
2072                          * to disable replay of the journal when we next remount
2073                          */
2074                         sb->s_flags |= MS_RDONLY;
2075
2076                         /*
2077                          * OK, test if we are remounting a valid rw partition
2078                          * readonly, and if so set the rdonly flag and then
2079                          * mark the partition as valid again.
2080                          */
2081                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2082                             (sbi->s_mount_state & EXT3_VALID_FS))
2083                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2084
2085                         ext3_mark_recovery_complete(sb, es);
2086                 } else {
2087                         int ret;
2088                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2089                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2090                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2091                                        "remount RDWR because of unsupported "
2092                                        "optional features (%x).\n",
2093                                        sb->s_id, ret);
2094                                 return -EROFS;
2095                         }
2096                         /*
2097                          * Mounting a RDONLY partition read-write, so reread
2098                          * and store the current valid flag.  (It may have
2099                          * been changed by e2fsck since we originally mounted
2100                          * the partition.)
2101                          */
2102                         ext3_clear_journal_err(sb, es);
2103                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2104                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2105                                 return ret;
2106                         if (!ext3_setup_super (sb, es, 0))
2107                                 sb->s_flags &= ~MS_RDONLY;
2108                 }
2109         }
2110         return 0;
2111 }
2112
2113 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2114 {
2115         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2116         unsigned long overhead;
2117         int i;
2118
2119         if (test_opt (sb, MINIX_DF))
2120                 overhead = 0;
2121         else {
2122                 /*
2123                  * Compute the overhead (FS structures)
2124                  */
2125
2126                 /*
2127                  * All of the blocks before first_data_block are
2128                  * overhead
2129                  */
2130                 overhead = le32_to_cpu(es->s_first_data_block);
2131
2132                 /*
2133                  * Add the overhead attributed to the superblock and
2134                  * block group descriptors.  If the sparse superblocks
2135                  * feature is turned on, then not all groups have this.
2136                  */
2137                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2138                         overhead += ext3_bg_has_super(sb, i) +
2139                                 ext3_bg_num_gdb(sb, i);
2140
2141                 /*
2142                  * Every block group has an inode bitmap, a block
2143                  * bitmap, and an inode table.
2144                  */
2145                 overhead += (EXT3_SB(sb)->s_groups_count *
2146                              (2 + EXT3_SB(sb)->s_itb_per_group));
2147         }
2148
2149         buf->f_type = EXT3_SUPER_MAGIC;
2150         buf->f_bsize = sb->s_blocksize;
2151         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2152         buf->f_bfree = ext3_count_free_blocks (sb);
2153         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2154         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2155                 buf->f_bavail = 0;
2156         buf->f_files = le32_to_cpu(es->s_inodes_count);
2157         buf->f_ffree = ext3_count_free_inodes (sb);
2158         buf->f_namelen = EXT3_NAME_LEN;
2159         return 0;
2160 }
2161
2162 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2163  * is locked for write. Otherwise the are possible deadlocks:
2164  * Process 1                         Process 2
2165  * ext3_create()                     quota_sync()
2166  *   journal_start()                   write_dquot()
2167  *   DQUOT_INIT()                        down(dqio_sem)
2168  *     down(dqio_sem)                    journal_start()
2169  *
2170  */
2171
2172 #ifdef CONFIG_QUOTA
2173
2174 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2175 {
2176         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2177 }
2178
2179 static int ext3_dquot_initialize(struct inode *inode, int type)
2180 {
2181         handle_t *handle;
2182         int ret, err;
2183
2184         /* We may create quota structure so we need to reserve enough blocks */
2185         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2186         if (IS_ERR(handle))
2187                 return PTR_ERR(handle);
2188         ret = dquot_initialize(inode, type);
2189         err = ext3_journal_stop(handle);
2190         if (!ret)
2191                 ret = err;
2192         return ret;
2193 }
2194
2195 static int ext3_dquot_drop(struct inode *inode)
2196 {
2197         handle_t *handle;
2198         int ret, err;
2199
2200         /* We may delete quota structure so we need to reserve enough blocks */
2201         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2202         if (IS_ERR(handle))
2203                 return PTR_ERR(handle);
2204         ret = dquot_drop(inode);
2205         err = ext3_journal_stop(handle);
2206         if (!ret)
2207                 ret = err;
2208         return ret;
2209 }
2210
2211 static int ext3_write_dquot(struct dquot *dquot)
2212 {
2213         int ret, err;
2214         handle_t *handle;
2215
2216         handle = ext3_journal_start(dquot_to_inode(dquot),
2217                                         EXT3_QUOTA_TRANS_BLOCKS);
2218         if (IS_ERR(handle))
2219                 return PTR_ERR(handle);
2220         ret = dquot_commit(dquot);
2221         err = ext3_journal_stop(handle);
2222         if (!ret)
2223                 ret = err;
2224         return ret;
2225 }
2226
2227 static int ext3_acquire_dquot(struct dquot *dquot)
2228 {
2229         int ret, err;
2230         handle_t *handle;
2231
2232         handle = ext3_journal_start(dquot_to_inode(dquot),
2233                                         EXT3_QUOTA_INIT_BLOCKS);
2234         if (IS_ERR(handle))
2235                 return PTR_ERR(handle);
2236         ret = dquot_acquire(dquot);
2237         err = ext3_journal_stop(handle);
2238         if (!ret)
2239                 ret = err;
2240         return ret;
2241 }
2242
2243 static int ext3_release_dquot(struct dquot *dquot)
2244 {
2245         int ret, err;
2246         handle_t *handle;
2247
2248         handle = ext3_journal_start(dquot_to_inode(dquot),
2249                                         EXT3_QUOTA_INIT_BLOCKS);
2250         if (IS_ERR(handle))
2251                 return PTR_ERR(handle);
2252         ret = dquot_release(dquot);
2253         err = ext3_journal_stop(handle);
2254         if (!ret)
2255                 ret = err;
2256         return ret;
2257 }
2258
2259 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2260 {
2261         /* Are we journalling quotas? */
2262         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2263             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2264                 dquot_mark_dquot_dirty(dquot);
2265                 return ext3_write_dquot(dquot);
2266         } else {
2267                 return dquot_mark_dquot_dirty(dquot);
2268         }
2269 }
2270
2271 static int ext3_write_info(struct super_block *sb, int type)
2272 {
2273         int ret, err;
2274         handle_t *handle;
2275
2276         /* Data block + inode block */
2277         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2278         if (IS_ERR(handle))
2279                 return PTR_ERR(handle);
2280         ret = dquot_commit_info(sb, type);
2281         err = ext3_journal_stop(handle);
2282         if (!ret)
2283                 ret = err;
2284         return ret;
2285 }
2286
2287 /*
2288  * Turn on quotas during mount time - we need to find
2289  * the quota file and such...
2290  */
2291 static int ext3_quota_on_mount(struct super_block *sb, int type)
2292 {
2293         int err;
2294         struct dentry *dentry;
2295         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2296                              .hash = 0,
2297                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2298
2299         dentry = lookup_hash(&name, sb->s_root);
2300         if (IS_ERR(dentry))
2301                 return PTR_ERR(dentry);
2302         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2303         if (err)
2304                 dput(dentry);
2305         /* We keep the dentry reference if everything went ok - we drop it
2306          * on quota_off time */
2307         return err;
2308 }
2309
2310 /* Turn quotas off during mount time */
2311 static int ext3_quota_off_mount(struct super_block *sb, int type)
2312 {
2313         int err;
2314         struct dentry *dentry;
2315
2316         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2317         err = vfs_quota_off_mount(sb, type);
2318         /* We invalidate dentry - it has at least wrong hash... */
2319         d_invalidate(dentry);
2320         dput(dentry);
2321         return err;
2322 }
2323
2324 /*
2325  * Standard function to be called on quota_on
2326  */
2327 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2328                          char *path)
2329 {
2330         int err;
2331         struct nameidata nd;
2332
2333         /* Not journalling quota? */
2334         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2335             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2336                 return vfs_quota_on(sb, type, format_id, path);
2337         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2338         if (err)
2339                 return err;
2340         /* Quotafile not on the same filesystem? */
2341         if (nd.mnt->mnt_sb != sb)
2342                 return -EXDEV;
2343         /* Quotafile not of fs root? */
2344         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2345                 printk(KERN_WARNING
2346                         "EXT3-fs: Quota file not on filesystem root. "
2347                         "Journalled quota will not work.\n");
2348         if (!ext3_should_journal_data(nd.dentry->d_inode))
2349                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2350                         "data-journalling. Journalled quota will not work.\n");
2351         path_release(&nd);
2352         return vfs_quota_on(sb, type, format_id, path);
2353 }
2354
2355 #endif
2356
2357 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2358         int flags, const char *dev_name, void *data)
2359 {
2360         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2361 }
2362
2363 static struct file_system_type ext3_fs_type = {
2364         .owner          = THIS_MODULE,
2365         .name           = "ext3",
2366         .get_sb         = ext3_get_sb,
2367         .kill_sb        = kill_block_super,
2368         .fs_flags       = FS_REQUIRES_DEV,
2369 };
2370
2371 static int __init init_ext3_fs(void)
2372 {
2373         int err = init_ext3_xattr();
2374         if (err)
2375                 return err;
2376         err = init_inodecache();
2377         if (err)
2378                 goto out1;
2379         err = register_filesystem(&ext3_fs_type);
2380         if (err)
2381                 goto out;
2382         return 0;
2383 out:
2384         destroy_inodecache();
2385 out1:
2386         exit_ext3_xattr();
2387         return err;
2388 }
2389
2390 static void __exit exit_ext3_fs(void)
2391 {
2392         unregister_filesystem(&ext3_fs_type);
2393         destroy_inodecache();
2394         exit_ext3_xattr();
2395 }
2396
2397 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2398 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2399 MODULE_LICENSE("GPL");
2400 module_init(init_ext3_fs)
2401 module_exit(exit_ext3_fs)