fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53
  54 /*
  55  * Wrappers for journal_start/end.
  56  *
  57  * The only special thing we need to do here is to make sure that all
  58  * journal_end calls result in the superblock being marked dirty, so
  59  * that sync() will call the filesystem's write_super callback if
  60  * appropriate.
  61  */
  62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (inode->i_sb->s_flags & MS_RDONLY)
  67                 return ERR_PTR(-EROFS);
  68
  69         /* Special case here: if the journal has aborted behind our
  70          * backs (eg. EIO in the commit thread), then we still need to
  71          * take the FS itself readonly cleanly. */
  72         journal = EXT3_JOURNAL(inode);
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(inode->i_sb, __FUNCTION__,
  75                            "Detected aborted journal");
  76                 return ERR_PTR(-EROFS);
  77         }
  78
  79         return journal_start(journal, nblocks);
  80 }
  81
  82 /*
  83  * The only special thing we need to do here is to make sure that all
  84  * journal_stop calls result in the superblock being marked dirty, so
  85  * that sync() will call the filesystem's write_super callback if
  86  * appropriate.
  87  */
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         printk(KERN_ERR "%s: aborting transaction: %s in %s",
 112                caller, errstr, err_fn);
 113
 114         if (bh)
 115                 BUFFER_TRACE(bh, "abort");
 116         journal_abort_handle(handle);
 117         if (!handle->h_err)
 118                 handle->h_err = err;
 119 }
 120
 121 static char error_buf[1024];
 122
 123 /* Deal with the reporting of failure conditions on a filesystem such as
 124  * inconsistencies detected or read IO failures.
 125  *
 126  * On ext2, we can store the error state of the filesystem in the
 127  * superblock.  That is not possible on ext3, because we may have other
 128  * write ordering constraints on the superblock which prevent us from
 129  * writing it out straight away; and given that the journal is about to
 130  * be aborted, we can't rely on the current, or future, transactions to
 131  * write out the superblock safely.
 132  *
 133  * We'll just use the journal_abort() error code to record an error in
 134  * the journal instead.  On recovery, the journal will compain about
 135  * that error until we've noted it down and cleared it.
 136  */
 137
 138 static void ext3_handle_error(struct super_block *sb)
 139 {
 140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 141
 142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 144
 145         if (sb->s_flags & MS_RDONLY)
 146                 return;
 147
 148         if (test_opt (sb, ERRORS_PANIC))
 149                 panic ("EXT3-fs (device %s): panic forced after error\n",
 150                        sb->s_id);
 151         if (test_opt (sb, ERRORS_RO)) {
 152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 153                 sb->s_flags |= MS_RDONLY;
 154         } else {
 155                 journal_t *journal = EXT3_SB(sb)->s_journal;
 156
 157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 158                 if (journal)
 159                         journal_abort(journal, -EIO);
 160         }
 161         ext3_commit_super(sb, es, 1);
 162 }
 163
 164 void ext3_error (struct super_block * sb, const char * function,
 165                  const char * fmt, ...)
 166 {
 167         va_list args;
 168
 169         va_start (args, fmt);
 170         vsprintf (error_buf, fmt, args);
 171         va_end (args);
 172
 173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
 174                 sb->s_id, function, error_buf);
 175
 176         ext3_handle_error(sb);
 177 }
 178
 179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 180 {
 181         char *errstr = NULL;
 182
 183         switch (errno) {
 184         case -EIO:
 185                 errstr = "IO failure";
 186                 break;
 187         case -ENOMEM:
 188                 errstr = "Out of memory";
 189                 break;
 190         case -EROFS:
 191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 192                         errstr = "Journal has aborted";
 193                 else
 194                         errstr = "Readonly filesystem";
 195                 break;
 196         default:
 197                 /* If the caller passed in an extra buffer for unknown
 198                  * errors, textualise them now.  Else we just return
 199                  * NULL. */
 200                 if (nbuf) {
 201                         /* Check for truncated error codes... */
 202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 203                                 errstr = nbuf;
 204                 }
 205                 break;
 206         }
 207
 208         return errstr;
 209 }
 210
 211 /* __ext3_std_error decodes expected errors from journaling functions
 212  * automatically and invokes the appropriate error response.  */
 213
 214 void __ext3_std_error (struct super_block * sb, const char * function,
 215                        int errno)
 216 {
 217         char nbuf[16];
 218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 219
 220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 221                 sb->s_id, function, errstr);
 222
 223         ext3_handle_error(sb);
 224 }
 225
 226 /*
 227  * ext3_abort is a much stronger failure handler than ext3_error.  The
 228  * abort function may be used to deal with unrecoverable failures such
 229  * as journal IO errors or ENOMEM at a critical moment in log management.
 230  *
 231  * We unconditionally force the filesystem into an ABORT|READONLY state,
 232  * unless the error response on the fs has been set to panic in which
 233  * case we take the easy way out and panic immediately.
 234  */
 235
 236 void ext3_abort (struct super_block * sb, const char * function,
 237                  const char * fmt, ...)
 238 {
 239         va_list args;
 240
 241         printk (KERN_CRIT "ext3_abort called.\n");
 242
 243         va_start (args, fmt);
 244         vsprintf (error_buf, fmt, args);
 245         va_end (args);
 246
 247         if (test_opt (sb, ERRORS_PANIC))
 248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
 249                        sb->s_id, function, error_buf);
 250
 251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
 252                 sb->s_id, function, error_buf);
 253
 254         if (sb->s_flags & MS_RDONLY)
 255                 return;
 256
 257         printk (KERN_CRIT "Remounting filesystem read-only\n");
 258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 259         sb->s_flags |= MS_RDONLY;
 260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 262 }
 263
 264 /* Deal with the reporting of failure conditions while running, such as
 265  * inconsistencies in operation or invalid system states.
 266  *
 267  * Use ext3_error() for cases of invalid filesystem states, as that will
 268  * record an error on disk and force a filesystem check on the next boot.
 269  */
 270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 271                             const char * fmt, ...)
 272 {
 273         va_list args;
 274
 275         va_start (args, fmt);
 276         vsprintf (error_buf, fmt, args);
 277         va_end (args);
 278
 279         /* this is to prevent panic from syncing this filesystem */
 280         /* AKPM: is this sufficient? */
 281         sb->s_flags |= MS_RDONLY;
 282         panic ("EXT3-fs panic (device %s): %s: %s\n",
 283                sb->s_id, function, error_buf);
 284 }
 285
 286 void ext3_warning (struct super_block * sb, const char * function,
 287                    const char * fmt, ...)
 288 {
 289         va_list args;
 290
 291         va_start (args, fmt);
 292         vsprintf (error_buf, fmt, args);
 293         va_end (args);
 294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
 295                 sb->s_id, function, error_buf);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        le32_to_cpu(NEXT_ORPHAN(inode)));
 384         }
 385 }
 386
 387 void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = le16_to_cpu(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         kfree(sbi->s_debts);
 407         brelse(sbi->s_sbh);
 408 #ifdef CONFIG_QUOTA
 409         for (i = 0; i < MAXQUOTAS; i++) {
 410                 if (sbi->s_qf_names[i])
 411                         kfree(sbi->s_qf_names[i]);
 412         }
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 init_MUTEX(&ei->truncate_sem);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 484                                              init_once, NULL);
 485         if (ext3_inode_cachep == NULL)
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static void destroy_inodecache(void)
 491 {
 492         if (kmem_cache_destroy(ext3_inode_cachep))
 493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 494 }
 495
 496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500        if (EXT3_I(inode)->i_acl &&
 501            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 502                posix_acl_release(EXT3_I(inode)->i_acl);
 503                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 504        }
 505        if (EXT3_I(inode)->i_default_acl &&
 506            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 507                posix_acl_release(EXT3_I(inode)->i_default_acl);
 508                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 509        }
 510 }
 511
 512 #else
 513 # define ext3_clear_inode NULL
 514 #endif
 515
 516 #ifdef CONFIG_QUOTA
 517
 518 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 519 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 520
 521 static int ext3_dquot_initialize(struct inode *inode, int type);
 522 static int ext3_dquot_drop(struct inode *inode);
 523 static int ext3_write_dquot(struct dquot *dquot);
 524 static int ext3_acquire_dquot(struct dquot *dquot);
 525 static int ext3_release_dquot(struct dquot *dquot);
 526 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 527 static int ext3_write_info(struct super_block *sb, int type);
 528 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 529 static int ext3_quota_on_mount(struct super_block *sb, int type);
 530 static int ext3_quota_off_mount(struct super_block *sb, int type);
 531
 532 static struct dquot_operations ext3_quota_operations = {
 533         .initialize     = ext3_dquot_initialize,
 534         .drop           = ext3_dquot_drop,
 535         .alloc_space    = dquot_alloc_space,
 536         .alloc_inode    = dquot_alloc_inode,
 537         .free_space     = dquot_free_space,
 538         .free_inode     = dquot_free_inode,
 539         .transfer       = dquot_transfer,
 540         .write_dquot    = ext3_write_dquot,
 541         .acquire_dquot  = ext3_acquire_dquot,
 542         .release_dquot  = ext3_release_dquot,
 543         .mark_dirty     = ext3_mark_dquot_dirty,
 544         .write_info     = ext3_write_info
 545 };
 546
 547 static struct quotactl_ops ext3_qctl_operations = {
 548         .quota_on       = ext3_quota_on,
 549         .quota_off      = vfs_quota_off,
 550         .quota_sync     = vfs_quota_sync,
 551         .get_info       = vfs_get_dqinfo,
 552         .set_info       = vfs_set_dqinfo,
 553         .get_dqblk      = vfs_get_dqblk,
 554         .set_dqblk      = vfs_set_dqblk
 555 };
 556 #endif
 557
 558 static struct super_operations ext3_sops = {
 559         .alloc_inode    = ext3_alloc_inode,
 560         .destroy_inode  = ext3_destroy_inode,
 561         .read_inode     = ext3_read_inode,
 562         .write_inode    = ext3_write_inode,
 563         .dirty_inode    = ext3_dirty_inode,
 564         .put_inode      = ext3_put_inode,
 565         .delete_inode   = ext3_delete_inode,
 566         .put_super      = ext3_put_super,
 567         .write_super    = ext3_write_super,
 568         .sync_fs        = ext3_sync_fs,
 569         .write_super_lockfs = ext3_write_super_lockfs,
 570         .unlockfs       = ext3_unlockfs,
 571         .statfs         = ext3_statfs,
 572         .remount_fs     = ext3_remount,
 573         .clear_inode    = ext3_clear_inode,
 574 };
 575
 576 struct dentry *ext3_get_parent(struct dentry *child);
 577 static struct export_operations ext3_export_ops = {
 578         .get_parent = ext3_get_parent,
 579 };
 580
 581 enum {
 582         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 583         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 584         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 585         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, Opt_noload,
 586         Opt_commit, Opt_journal_update, Opt_journal_inum,
 587         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 588         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 589         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 590         Opt_tagxid, Opt_ignore, Opt_err,
 591 };
 592
 593 static match_table_t tokens = {
 594         {Opt_bsd_df, "bsddf"},
 595         {Opt_minix_df, "minixdf"},
 596         {Opt_grpid, "grpid"},
 597         {Opt_grpid, "bsdgroups"},
 598         {Opt_nogrpid, "nogrpid"},
 599         {Opt_nogrpid, "sysvgroups"},
 600         {Opt_resgid, "resgid=%u"},
 601         {Opt_resuid, "resuid=%u"},
 602         {Opt_sb, "sb=%u"},
 603         {Opt_err_cont, "errors=continue"},
 604         {Opt_err_panic, "errors=panic"},
 605         {Opt_err_ro, "errors=remount-ro"},
 606         {Opt_nouid32, "nouid32"},
 607         {Opt_nocheck, "nocheck"},
 608         {Opt_nocheck, "check=none"},
 609         {Opt_check, "check"},
 610         {Opt_debug, "debug"},
 611         {Opt_oldalloc, "oldalloc"},
 612         {Opt_orlov, "orlov"},
 613         {Opt_user_xattr, "user_xattr"},
 614         {Opt_nouser_xattr, "nouser_xattr"},
 615         {Opt_acl, "acl"},
 616         {Opt_noacl, "noacl"},
 617         {Opt_noload, "noload"},
 618         {Opt_commit, "commit=%u"},
 619         {Opt_journal_update, "journal=update"},
 620         {Opt_journal_inum, "journal=%u"},
 621         {Opt_abort, "abort"},
 622         {Opt_data_journal, "data=journal"},
 623         {Opt_data_ordered, "data=ordered"},
 624         {Opt_data_writeback, "data=writeback"},
 625         {Opt_offusrjquota, "usrjquota="},
 626         {Opt_usrjquota, "usrjquota=%s"},
 627         {Opt_offgrpjquota, "grpjquota="},
 628         {Opt_grpjquota, "grpjquota=%s"},
 629         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 630         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 631         {Opt_tagxid, "tagxid"},
 632         {Opt_ignore, "grpquota"},
 633         {Opt_ignore, "noquota"},
 634         {Opt_ignore, "quota"},
 635         {Opt_ignore, "usrquota"},
 636         {Opt_err, NULL}
 637 };
 638
 639 static unsigned long get_sb_block(void **data)
 640 {
 641         unsigned long   sb_block;
 642         char            *options = (char *) *data;
 643
 644         if (!options || strncmp(options, "sb=", 3) != 0)
 645                 return 1;       /* Default location */
 646         options += 3;
 647         sb_block = simple_strtoul(options, &options, 0);
 648         if (*options && *options != ',') {
 649                 printk("EXT3-fs: Invalid sb specification: %s\n",
 650                        (char *) *data);
 651                 return 1;
 652         }
 653         if (*options == ',')
 654                 options++;
 655         *data = (void *) options;
 656         return sb_block;
 657 }
 658
 659 static int parse_options (char * options, struct super_block *sb,
 660                           unsigned long * inum, int is_remount)
 661 {
 662         struct ext3_sb_info *sbi = EXT3_SB(sb);
 663         char * p;
 664         substring_t args[MAX_OPT_ARGS];
 665         int data_opt = 0;
 666         int option;
 667 #ifdef CONFIG_QUOTA
 668         int qtype;
 669 #endif
 670
 671         if (!options)
 672                 return 1;
 673
 674         while ((p = strsep (&options, ",")) != NULL) {
 675                 int token;
 676                 if (!*p)
 677                         continue;
 678
 679                 token = match_token(p, tokens, args);
 680                 switch (token) {
 681                 case Opt_bsd_df:
 682                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 683                         break;
 684                 case Opt_minix_df:
 685                         set_opt (sbi->s_mount_opt, MINIX_DF);
 686                         break;
 687                 case Opt_grpid:
 688                         set_opt (sbi->s_mount_opt, GRPID);
 689                         break;
 690                 case Opt_nogrpid:
 691                         clear_opt (sbi->s_mount_opt, GRPID);
 692                         break;
 693                 case Opt_resuid:
 694                         if (match_int(&args[0], &option))
 695                                 return 0;
 696                         sbi->s_resuid = option;
 697                         break;
 698                 case Opt_resgid:
 699                         if (match_int(&args[0], &option))
 700                                 return 0;
 701                         sbi->s_resgid = option;
 702                         break;
 703                 case Opt_sb:
 704                         /* handled by get_sb_block() instead of here */
 705                         /* *sb_block = match_int(&args[0]); */
 706                         break;
 707                 case Opt_err_panic:
 708                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 709                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 710                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 711                         break;
 712                 case Opt_err_ro:
 713                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 714                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 715                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 716                         break;
 717                 case Opt_err_cont:
 718                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 719                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 720                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 721                         break;
 722                 case Opt_nouid32:
 723                         set_opt (sbi->s_mount_opt, NO_UID32);
 724                         break;
 725 #ifndef CONFIG_INOXID_NONE
 726                 case Opt_tagxid:
 727                         if (is_remount) {
 728                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 729                                        "tagxid on remount\n");
 730                                 return 0;
 731                         }
 732                         set_opt (sbi->s_mount_opt, TAG_XID);
 733                         break;
 734 #endif
 735                 case Opt_check:
 736 #ifdef CONFIG_EXT3_CHECK
 737                         set_opt (sbi->s_mount_opt, CHECK);
 738 #else
 739                         printk(KERN_ERR
 740                                "EXT3 Check option not supported\n");
 741 #endif
 742                         break;
 743                 case Opt_nocheck:
 744                         clear_opt (sbi->s_mount_opt, CHECK);
 745                         break;
 746                 case Opt_debug:
 747                         set_opt (sbi->s_mount_opt, DEBUG);
 748                         break;
 749                 case Opt_oldalloc:
 750                         set_opt (sbi->s_mount_opt, OLDALLOC);
 751                         break;
 752                 case Opt_orlov:
 753                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 754                         break;
 755 #ifdef CONFIG_EXT3_FS_XATTR
 756                 case Opt_user_xattr:
 757                         set_opt (sbi->s_mount_opt, XATTR_USER);
 758                         break;
 759                 case Opt_nouser_xattr:
 760                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 761                         break;
 762 #else
 763                 case Opt_user_xattr:
 764                 case Opt_nouser_xattr:
 765                         printk("EXT3 (no)user_xattr options not supported\n");
 766                         break;
 767 #endif
 768 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 769                 case Opt_acl:
 770                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 771                         break;
 772                 case Opt_noacl:
 773                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 774                         break;
 775 #else
 776                 case Opt_acl:
 777                 case Opt_noacl:
 778                         printk("EXT3 (no)acl options not supported\n");
 779                         break;
 780 #endif
 781                 case Opt_journal_update:
 782                         /* @@@ FIXME */
 783                         /* Eventually we will want to be able to create
 784                            a journal file here.  For now, only allow the
 785                            user to specify an existing inode to be the
 786                            journal file. */
 787                         if (is_remount) {
 788                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 789                                        "journal on remount\n");
 790                                 return 0;
 791                         }
 792                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 793                         break;
 794                 case Opt_journal_inum:
 795                         if (is_remount) {
 796                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 797                                        "journal on remount\n");
 798                                 return 0;
 799                         }
 800                         if (match_int(&args[0], &option))
 801                                 return 0;
 802                         *inum = option;
 803                         break;
 804                 case Opt_noload:
 805                         set_opt (sbi->s_mount_opt, NOLOAD);
 806                         break;
 807                 case Opt_commit:
 808                         if (match_int(&args[0], &option))
 809                                 return 0;
 810                         if (option < 0)
 811                                 return 0;
 812                         if (option == 0)
 813                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 814                         sbi->s_commit_interval = HZ * option;
 815                         break;
 816                 case Opt_data_journal:
 817                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 818                         goto datacheck;
 819                 case Opt_data_ordered:
 820                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 821                         goto datacheck;
 822                 case Opt_data_writeback:
 823                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 824                 datacheck:
 825                         if (is_remount) {
 826                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 827                                                 != data_opt) {
 828                                         printk(KERN_ERR
 829                                                 "EXT3-fs: cannot change data "
 830                                                 "mode on remount\n");
 831                                         return 0;
 832                                 }
 833                         } else {
 834                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 835                                 sbi->s_mount_opt |= data_opt;
 836                         }
 837                         break;
 838 #ifdef CONFIG_QUOTA
 839                 case Opt_usrjquota:
 840                         qtype = USRQUOTA;
 841                         goto set_qf_name;
 842                 case Opt_grpjquota:
 843                         qtype = GRPQUOTA;
 844 set_qf_name:
 845                         if (sb_any_quota_enabled(sb)) {
 846                                 printk(KERN_ERR
 847                                         "EXT3-fs: Cannot change journalled "
 848                                         "quota options when quota turned on.\n");
 849                                 return 0;
 850                         }
 851                         if (sbi->s_qf_names[qtype]) {
 852                                 printk(KERN_ERR
 853                                         "EXT3-fs: %s quota file already "
 854                                         "specified.\n", QTYPE2NAME(qtype));
 855                                 return 0;
 856                         }
 857                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 858                         if (!sbi->s_qf_names[qtype]) {
 859                                 printk(KERN_ERR
 860                                         "EXT3-fs: not enough memory for "
 861                                         "storing quotafile name.\n");
 862                                 return 0;
 863                         }
 864                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 865                                 printk(KERN_ERR
 866                                         "EXT3-fs: quotafile must be on "
 867                                         "filesystem root.\n");
 868                                 kfree(sbi->s_qf_names[qtype]);
 869                                 sbi->s_qf_names[qtype] = NULL;
 870                                 return 0;
 871                         }
 872                         break;
 873                 case Opt_offusrjquota:
 874                         qtype = USRQUOTA;
 875                         goto clear_qf_name;
 876                 case Opt_offgrpjquota:
 877                         qtype = GRPQUOTA;
 878 clear_qf_name:
 879                         if (sb_any_quota_enabled(sb)) {
 880                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 881                                         "journalled quota options when "
 882                                         "quota turned on.\n");
 883                                 return 0;
 884                         }
 885                         if (sbi->s_qf_names[qtype]) {
 886                                 kfree(sbi->s_qf_names[qtype]);
 887                                 sbi->s_qf_names[qtype] = NULL;
 888                         }
 889                         break;
 890                 case Opt_jqfmt_vfsold:
 891                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 892                         break;
 893                 case Opt_jqfmt_vfsv0:
 894                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 895                         break;
 896 #else
 897                 case Opt_usrjquota:
 898                 case Opt_grpjquota:
 899                 case Opt_offusrjquota:
 900                 case Opt_offgrpjquota:
 901                 case Opt_jqfmt_vfsold:
 902                 case Opt_jqfmt_vfsv0:
 903                         printk(KERN_ERR
 904                                 "EXT3-fs: journalled quota options not "
 905                                 "supported.\n");
 906                         break;
 907 #endif
 908                 case Opt_abort:
 909                         set_opt(sbi->s_mount_opt, ABORT);
 910                         break;
 911                 case Opt_ignore:
 912                         break;
 913                 default:
 914                         printk (KERN_ERR
 915                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 916                                 "or missing value\n", p);
 917                         return 0;
 918                 }
 919         }
 920 #ifdef CONFIG_QUOTA
 921         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 922             sbi->s_qf_names[GRPQUOTA])) {
 923                 printk(KERN_ERR
 924                         "EXT3-fs: journalled quota format not specified.\n");
 925                 return 0;
 926         }
 927 #endif
 928
 929         return 1;
 930 }
 931
 932 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 933                             int read_only)
 934 {
 935         struct ext3_sb_info *sbi = EXT3_SB(sb);
 936         int res = 0;
 937
 938         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 939                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 940                         "forcing read-only mode\n");
 941                 res = MS_RDONLY;
 942         }
 943         if (read_only)
 944                 return res;
 945         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 946                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 947                         "running e2fsck is recommended\n");
 948         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 949                 printk (KERN_WARNING
 950                         "EXT3-fs warning: mounting fs with errors, "
 951                         "running e2fsck is recommended\n");
 952         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 953                  le16_to_cpu(es->s_mnt_count) >=
 954                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 955                 printk (KERN_WARNING
 956                         "EXT3-fs warning: maximal mount count reached, "
 957                         "running e2fsck is recommended\n");
 958         else if (le32_to_cpu(es->s_checkinterval) &&
 959                 (le32_to_cpu(es->s_lastcheck) +
 960                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 961                 printk (KERN_WARNING
 962                         "EXT3-fs warning: checktime reached, "
 963                         "running e2fsck is recommended\n");
 964 #if 0
 965                 /* @@@ We _will_ want to clear the valid bit if we find
 966                    inconsistencies, to force a fsck at reboot.  But for
 967                    a plain journaled filesystem we can keep it set as
 968                    valid forever! :) */
 969         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 970 #endif
 971         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 972                 es->s_max_mnt_count =
 973                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 974         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 975         es->s_mtime = cpu_to_le32(get_seconds());
 976         ext3_update_dynamic_rev(sb);
 977         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 978
 979         ext3_commit_super(sb, es, 1);
 980         if (test_opt(sb, DEBUG))
 981                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 982                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 983                         sb->s_blocksize,
 984                         sbi->s_groups_count,
 985                         EXT3_BLOCKS_PER_GROUP(sb),
 986                         EXT3_INODES_PER_GROUP(sb),
 987                         sbi->s_mount_opt);
 988
 989         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 990         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 991                 char b[BDEVNAME_SIZE];
 992
 993                 printk("external journal on %s\n",
 994                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
 995         } else {
 996                 printk("internal journal\n");
 997         }
 998 #ifdef CONFIG_EXT3_CHECK
 999         if (test_opt (sb, CHECK)) {
1000                 ext3_check_blocks_bitmap (sb);
1001                 ext3_check_inodes_bitmap (sb);
1002         }
1003 #endif
1004         return res;
1005 }
1006
1007 static int ext3_check_descriptors (struct super_block * sb)
1008 {
1009         struct ext3_sb_info *sbi = EXT3_SB(sb);
1010         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1011         struct ext3_group_desc * gdp = NULL;
1012         int desc_block = 0;
1013         int i;
1014
1015         ext3_debug ("Checking group descriptors");
1016
1017         for (i = 0; i < sbi->s_groups_count; i++)
1018         {
1019                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1020                         gdp = (struct ext3_group_desc *)
1021                                         sbi->s_group_desc[desc_block++]->b_data;
1022                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1023                     le32_to_cpu(gdp->bg_block_bitmap) >=
1024                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1025                 {
1026                         ext3_error (sb, "ext3_check_descriptors",
1027                                     "Block bitmap for group %d"
1028                                     " not in group (block %lu)!",
1029                                     i, (unsigned long)
1030                                         le32_to_cpu(gdp->bg_block_bitmap));
1031                         return 0;
1032                 }
1033                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1034                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1035                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1036                 {
1037                         ext3_error (sb, "ext3_check_descriptors",
1038                                     "Inode bitmap for group %d"
1039                                     " not in group (block %lu)!",
1040                                     i, (unsigned long)
1041                                         le32_to_cpu(gdp->bg_inode_bitmap));
1042                         return 0;
1043                 }
1044                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1045                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1046                     block + EXT3_BLOCKS_PER_GROUP(sb))
1047                 {
1048                         ext3_error (sb, "ext3_check_descriptors",
1049                                     "Inode table for group %d"
1050                                     " not in group (block %lu)!",
1051                                     i, (unsigned long)
1052                                         le32_to_cpu(gdp->bg_inode_table));
1053                         return 0;
1054                 }
1055                 block += EXT3_BLOCKS_PER_GROUP(sb);
1056                 gdp++;
1057         }
1058
1059         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1060         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1061         return 1;
1062 }
1063
1064
1065 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1066  * the superblock) which were deleted from all directories, but held open by
1067  * a process at the time of a crash.  We walk the list and try to delete these
1068  * inodes at recovery time (only with a read-write filesystem).
1069  *
1070  * In order to keep the orphan inode chain consistent during traversal (in
1071  * case of crash during recovery), we link each inode into the superblock
1072  * orphan list_head and handle it the same way as an inode deletion during
1073  * normal operation (which journals the operations for us).
1074  *
1075  * We only do an iget() and an iput() on each inode, which is very safe if we
1076  * accidentally point at an in-use or already deleted inode.  The worst that
1077  * can happen in this case is that we get a "bit already cleared" message from
1078  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1079  * e2fsck was run on this filesystem, and it must have already done the orphan
1080  * inode cleanup for us, so we can safely abort without any further action.
1081  */
1082 static void ext3_orphan_cleanup (struct super_block * sb,
1083                                  struct ext3_super_block * es)
1084 {
1085         unsigned int s_flags = sb->s_flags;
1086         int nr_orphans = 0, nr_truncates = 0;
1087 #ifdef CONFIG_QUOTA
1088         int i;
1089 #endif
1090         if (!es->s_last_orphan) {
1091                 jbd_debug(4, "no orphan inodes to clean up\n");
1092                 return;
1093         }
1094
1095         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1096                 if (es->s_last_orphan)
1097                         jbd_debug(1, "Errors on filesystem, "
1098                                   "clearing orphan list.\n");
1099                 es->s_last_orphan = 0;
1100                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1101                 return;
1102         }
1103
1104         if (s_flags & MS_RDONLY) {
1105                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1106                        sb->s_id);
1107                 sb->s_flags &= ~MS_RDONLY;
1108         }
1109 #ifdef CONFIG_QUOTA
1110         /* Needed for iput() to work correctly and not trash data */
1111         sb->s_flags |= MS_ACTIVE;
1112         /* Turn on quotas so that they are updated correctly */
1113         for (i = 0; i < MAXQUOTAS; i++) {
1114                 if (EXT3_SB(sb)->s_qf_names[i]) {
1115                         int ret = ext3_quota_on_mount(sb, i);
1116                         if (ret < 0)
1117                                 printk(KERN_ERR
1118                                         "EXT3-fs: Cannot turn on journalled "
1119                                         "quota: error %d\n", ret);
1120                 }
1121         }
1122 #endif
1123
1124         while (es->s_last_orphan) {
1125                 struct inode *inode;
1126
1127                 if (!(inode =
1128                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1129                         es->s_last_orphan = 0;
1130                         break;
1131                 }
1132
1133                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1134                 DQUOT_INIT(inode);
1135                 if (inode->i_nlink) {
1136                         printk(KERN_DEBUG
1137                                 "%s: truncating inode %ld to %Ld bytes\n",
1138                                 __FUNCTION__, inode->i_ino, inode->i_size);
1139                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1140                                   inode->i_ino, inode->i_size);
1141                         ext3_truncate(inode);
1142                         nr_truncates++;
1143                 } else {
1144                         printk(KERN_DEBUG
1145                                 "%s: deleting unreferenced inode %ld\n",
1146                                 __FUNCTION__, inode->i_ino);
1147                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1148                                   inode->i_ino);
1149                         nr_orphans++;
1150                 }
1151                 iput(inode);  /* The delete magic happens here! */
1152         }
1153
1154 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1155
1156         if (nr_orphans)
1157                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1158                        sb->s_id, PLURAL(nr_orphans));
1159         if (nr_truncates)
1160                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1161                        sb->s_id, PLURAL(nr_truncates));
1162 #ifdef CONFIG_QUOTA
1163         /* Turn quotas off */
1164         for (i = 0; i < MAXQUOTAS; i++) {
1165                 if (sb_dqopt(sb)->files[i])
1166                         ext3_quota_off_mount(sb, i);
1167         }
1168 #endif
1169         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1170 }
1171
1172 #define log2(n) ffz(~(n))
1173
1174 /*
1175  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1176  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1177  * We need to be 1 filesystem block less than the 2^32 sector limit.
1178  */
1179 static loff_t ext3_max_size(int bits)
1180 {
1181         loff_t res = EXT3_NDIR_BLOCKS;
1182         res += 1LL << (bits-2);
1183         res += 1LL << (2*(bits-2));
1184         res += 1LL << (3*(bits-2));
1185         res <<= bits;
1186         if (res > (512LL << 32) - (1 << bits))
1187                 res = (512LL << 32) - (1 << bits);
1188         return res;
1189 }
1190
1191 static unsigned long descriptor_loc(struct super_block *sb,
1192                                     unsigned long logic_sb_block,
1193                                     int nr)
1194 {
1195         struct ext3_sb_info *sbi = EXT3_SB(sb);
1196         unsigned long bg, first_data_block, first_meta_bg;
1197         int has_super = 0;
1198
1199         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1200         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1201
1202         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1203             nr < first_meta_bg)
1204                 return (logic_sb_block + nr + 1);
1205         bg = sbi->s_desc_per_block * nr;
1206         if (ext3_bg_has_super(sb, bg))
1207                 has_super = 1;
1208         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1209 }
1210
1211
1212 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1213 {
1214         struct buffer_head * bh;
1215         struct ext3_super_block *es = NULL;
1216         struct ext3_sb_info *sbi;
1217         unsigned long block;
1218         unsigned long sb_block = get_sb_block(&data);
1219         unsigned long logic_sb_block;
1220         unsigned long offset = 0;
1221         unsigned long journal_inum = 0;
1222         unsigned long def_mount_opts;
1223         struct inode *root;
1224         int blocksize;
1225         int hblock;
1226         int db_count;
1227         int i;
1228         int needs_recovery;
1229
1230         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1231         if (!sbi)
1232                 return -ENOMEM;
1233         sb->s_fs_info = sbi;
1234         memset(sbi, 0, sizeof(*sbi));
1235         sbi->s_mount_opt = 0;
1236         sbi->s_resuid = EXT3_DEF_RESUID;
1237         sbi->s_resgid = EXT3_DEF_RESGID;
1238
1239         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1240         if (!blocksize) {
1241                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1242                 goto out_fail;
1243         }
1244
1245         /*
1246          * The ext3 superblock will not be buffer aligned for other than 1kB
1247          * block sizes.  We need to calculate the offset from buffer start.
1248          */
1249         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1250                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1251                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1252         } else {
1253                 logic_sb_block = sb_block;
1254         }
1255
1256         if (!(bh = sb_bread(sb, logic_sb_block))) {
1257                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1258                 goto out_fail;
1259         }
1260         /*
1261          * Note: s_es must be initialized as soon as possible because
1262          *       some ext3 macro-instructions depend on its value
1263          */
1264         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1265         sbi->s_es = es;
1266         sb->s_magic = le16_to_cpu(es->s_magic);
1267         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1268                 if (!silent)
1269                         printk(KERN_ERR
1270                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1271                                sb->s_id);
1272                 goto failed_mount;
1273         }
1274
1275         /* Set defaults before we parse the mount options */
1276         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1277         if (def_mount_opts & EXT3_DEFM_DEBUG)
1278                 set_opt(sbi->s_mount_opt, DEBUG);
1279         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1280                 set_opt(sbi->s_mount_opt, GRPID);
1281         if (def_mount_opts & EXT3_DEFM_UID16)
1282                 set_opt(sbi->s_mount_opt, NO_UID32);
1283         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1284                 set_opt(sbi->s_mount_opt, XATTR_USER);
1285         if (def_mount_opts & EXT3_DEFM_ACL)
1286                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1287         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1288                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1289         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1290                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1291         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1292                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1293
1294         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1295                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1296         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1297                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1298
1299         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1300         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1301
1302         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1303                 goto failed_mount;
1304
1305         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1306                 sb->s_flags |= MS_TAGXID;
1307         sb->s_flags |= MS_ONE_SECOND;
1308         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1309                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1310
1311         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1312             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1313              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1314              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1315                 printk(KERN_WARNING
1316                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1317                        "running e2fsck is recommended\n");
1318         /*
1319          * Check feature flags regardless of the revision level, since we
1320          * previously didn't change the revision level when setting the flags,
1321          * so there is a chance incompat flags are set on a rev 0 filesystem.
1322          */
1323         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1324                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1325                        "unsupported optional features (%x).\n",
1326                        sb->s_id, i);
1327                 goto failed_mount;
1328         }
1329         if (!(sb->s_flags & MS_RDONLY) &&
1330             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1331                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1332                        "unsupported optional features (%x).\n",
1333                        sb->s_id, i);
1334                 goto failed_mount;
1335         }
1336         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1337
1338         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1339             blocksize > EXT3_MAX_BLOCK_SIZE) {
1340                 printk(KERN_ERR
1341                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1342                        blocksize, sb->s_id);
1343                 goto failed_mount;
1344         }
1345
1346         hblock = bdev_hardsect_size(sb->s_bdev);
1347         if (sb->s_blocksize != blocksize) {
1348                 /*
1349                  * Make sure the blocksize for the filesystem is larger
1350                  * than the hardware sectorsize for the machine.
1351                  */
1352                 if (blocksize < hblock) {
1353                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1354                                "device blocksize %d.\n", blocksize, hblock);
1355                         goto failed_mount;
1356                 }
1357
1358                 brelse (bh);
1359                 sb_set_blocksize(sb, blocksize);
1360                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1361                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1362                 bh = sb_bread(sb, logic_sb_block);
1363                 if (!bh) {
1364                         printk(KERN_ERR
1365                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1366                         goto failed_mount;
1367                 }
1368                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1369                 sbi->s_es = es;
1370                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1371                         printk (KERN_ERR
1372                                 "EXT3-fs: Magic mismatch, very weird !\n");
1373                         goto failed_mount;
1374                 }
1375         }
1376
1377         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1378
1379         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1380                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1381                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1382         } else {
1383                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1384                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1385                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1386                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1387                     (sbi->s_inode_size > blocksize)) {
1388                         printk (KERN_ERR
1389                                 "EXT3-fs: unsupported inode size: %d\n",
1390                                 sbi->s_inode_size);
1391                         goto failed_mount;
1392                 }
1393         }
1394         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1395                                    le32_to_cpu(es->s_log_frag_size);
1396         if (blocksize != sbi->s_frag_size) {
1397                 printk(KERN_ERR
1398                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1399                        sbi->s_frag_size, blocksize);
1400                 goto failed_mount;
1401         }
1402         sbi->s_frags_per_block = 1;
1403         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1404         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1405         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1406         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1407         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1408         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1409         sbi->s_sbh = bh;
1410         sbi->s_mount_state = le16_to_cpu(es->s_state);
1411         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1412         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1413         for (i=0; i < 4; i++)
1414                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1415         sbi->s_def_hash_version = es->s_def_hash_version;
1416
1417         if (sbi->s_blocks_per_group > blocksize * 8) {
1418                 printk (KERN_ERR
1419                         "EXT3-fs: #blocks per group too big: %lu\n",
1420                         sbi->s_blocks_per_group);
1421                 goto failed_mount;
1422         }
1423         if (sbi->s_frags_per_group > blocksize * 8) {
1424                 printk (KERN_ERR
1425                         "EXT3-fs: #fragments per group too big: %lu\n",
1426                         sbi->s_frags_per_group);
1427                 goto failed_mount;
1428         }
1429         if (sbi->s_inodes_per_group > blocksize * 8) {
1430                 printk (KERN_ERR
1431                         "EXT3-fs: #inodes per group too big: %lu\n",
1432                         sbi->s_inodes_per_group);
1433                 goto failed_mount;
1434         }
1435
1436         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1437                                le32_to_cpu(es->s_first_data_block) +
1438                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1439                               EXT3_BLOCKS_PER_GROUP(sb);
1440         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1441                    EXT3_DESC_PER_BLOCK(sb);
1442         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1443                                     GFP_KERNEL);
1444         if (sbi->s_group_desc == NULL) {
1445                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1446                 goto failed_mount;
1447         }
1448         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1449                         GFP_KERNEL);
1450         if (!sbi->s_debts) {
1451                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1452                 goto failed_mount2;
1453         }
1454         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1455
1456         percpu_counter_init(&sbi->s_freeblocks_counter);
1457         percpu_counter_init(&sbi->s_freeinodes_counter);
1458         percpu_counter_init(&sbi->s_dirs_counter);
1459         bgl_lock_init(&sbi->s_blockgroup_lock);
1460
1461         for (i = 0; i < db_count; i++) {
1462                 block = descriptor_loc(sb, logic_sb_block, i);
1463                 sbi->s_group_desc[i] = sb_bread(sb, block);
1464                 if (!sbi->s_group_desc[i]) {
1465                         printk (KERN_ERR "EXT3-fs: "
1466                                 "can't read group descriptor %d\n", i);
1467                         db_count = i;
1468                         goto failed_mount2;
1469                 }
1470         }
1471         if (!ext3_check_descriptors (sb)) {
1472                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1473                 goto failed_mount2;
1474         }
1475         sbi->s_gdb_count = db_count;
1476         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1477         spin_lock_init(&sbi->s_next_gen_lock);
1478         /*
1479          * set up enough so that it can read an inode
1480          */
1481         sb->s_op = &ext3_sops;
1482         sb->s_export_op = &ext3_export_ops;
1483 #ifdef CONFIG_QUOTA
1484         sb->s_qcop = &ext3_qctl_operations;
1485         sb->dq_op = &ext3_quota_operations;
1486 #endif
1487         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1488
1489         sb->s_root = NULL;
1490
1491         needs_recovery = (es->s_last_orphan != 0 ||
1492                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1493                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1494
1495         /*
1496          * The first inode we look at is the journal inode.  Don't try
1497          * root first: it may be modified in the journal!
1498          */
1499         if (!test_opt(sb, NOLOAD) &&
1500             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1501                 if (ext3_load_journal(sb, es))
1502                         goto failed_mount2;
1503         } else if (journal_inum) {
1504                 if (ext3_create_journal(sb, es, journal_inum))
1505                         goto failed_mount2;
1506         } else {
1507                 if (!silent)
1508                         printk (KERN_ERR
1509                                 "ext3: No journal on filesystem on %s\n",
1510                                 sb->s_id);
1511                 goto failed_mount2;
1512         }
1513
1514         /* We have now updated the journal if required, so we can
1515          * validate the data journaling mode. */
1516         switch (test_opt(sb, DATA_FLAGS)) {
1517         case 0:
1518                 /* No mode set, assume a default based on the journal
1519                    capabilities: ORDERED_DATA if the journal can
1520                    cope, else JOURNAL_DATA */
1521                 if (journal_check_available_features
1522                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1523                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1524                 else
1525                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1526                 break;
1527
1528         case EXT3_MOUNT_ORDERED_DATA:
1529         case EXT3_MOUNT_WRITEBACK_DATA:
1530                 if (!journal_check_available_features
1531                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1532                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1533                                "requested data journaling mode\n");
1534                         goto failed_mount3;
1535                 }
1536         default:
1537                 break;
1538         }
1539
1540         /*
1541          * The journal_load will have done any necessary log recovery,
1542          * so we can safely mount the rest of the filesystem now.
1543          */
1544
1545         root = iget(sb, EXT3_ROOT_INO);
1546         sb->s_root = d_alloc_root(root);
1547         if (!sb->s_root) {
1548                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1549                 iput(root);
1550                 goto failed_mount3;
1551         }
1552         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1553                 dput(sb->s_root);
1554                 sb->s_root = NULL;
1555                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1556                 goto failed_mount3;
1557         }
1558
1559         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1560         /*
1561          * akpm: core read_super() calls in here with the superblock locked.
1562          * That deadlocks, because orphan cleanup needs to lock the superblock
1563          * in numerous places.  Here we just pop the lock - it's relatively
1564          * harmless, because we are now ready to accept write_super() requests,
1565          * and aviro says that's the only reason for hanging onto the
1566          * superblock lock.
1567          */
1568         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1569         ext3_orphan_cleanup(sb, es);
1570         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1571         if (needs_recovery)
1572                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1573         ext3_mark_recovery_complete(sb, es);
1574         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1575                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1576                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1577                 "writeback");
1578
1579         percpu_counter_mod(&sbi->s_freeblocks_counter,
1580                 ext3_count_free_blocks(sb));
1581         percpu_counter_mod(&sbi->s_freeinodes_counter,
1582                 ext3_count_free_inodes(sb));
1583         percpu_counter_mod(&sbi->s_dirs_counter,
1584                 ext3_count_dirs(sb));
1585
1586         return 0;
1587
1588 failed_mount3:
1589         journal_destroy(sbi->s_journal);
1590 failed_mount2:
1591         kfree(sbi->s_debts);
1592         for (i = 0; i < db_count; i++)
1593                 brelse(sbi->s_group_desc[i]);
1594         kfree(sbi->s_group_desc);
1595 failed_mount:
1596 #ifdef CONFIG_QUOTA
1597         for (i = 0; i < MAXQUOTAS; i++) {
1598                 if (sbi->s_qf_names[i])
1599                         kfree(sbi->s_qf_names[i]);
1600         }
1601 #endif
1602         ext3_blkdev_remove(sbi);
1603         brelse(bh);
1604 out_fail:
1605         sb->s_fs_info = NULL;
1606         kfree(sbi);
1607         return -EINVAL;
1608 }
1609
1610 /*
1611  * Setup any per-fs journal parameters now.  We'll do this both on
1612  * initial mount, once the journal has been initialised but before we've
1613  * done any recovery; and again on any subsequent remount.
1614  */
1615 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1616                                      journal_t *journal)
1617 {
1618         if (sbi->s_commit_interval)
1619                 journal->j_commit_interval = sbi->s_commit_interval;
1620         /* We could also set up an ext3-specific default for the commit
1621          * interval here, but for now we'll just fall back to the jbd
1622          * default. */
1623 }
1624
1625
1626 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1627 {
1628         struct inode *journal_inode;
1629         journal_t *journal;
1630
1631         /* First, test for the existence of a valid inode on disk.  Bad
1632          * things happen if we iget() an unused inode, as the subsequent
1633          * iput() will try to delete it. */
1634
1635         journal_inode = iget(sb, journal_inum);
1636         if (!journal_inode) {
1637                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1638                 return NULL;
1639         }
1640         if (!journal_inode->i_nlink) {
1641                 make_bad_inode(journal_inode);
1642                 iput(journal_inode);
1643                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1644                 return NULL;
1645         }
1646
1647         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1648                   journal_inode, journal_inode->i_size);
1649         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1650                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1651                 iput(journal_inode);
1652                 return NULL;
1653         }
1654
1655         journal = journal_init_inode(journal_inode);
1656         if (!journal) {
1657                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1658                 iput(journal_inode);
1659                 return NULL;
1660         }
1661         journal->j_private = sb;
1662         ext3_init_journal_params(EXT3_SB(sb), journal);
1663         return journal;
1664 }
1665
1666 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1667                                        dev_t j_dev)
1668 {
1669         struct buffer_head * bh;
1670         journal_t *journal;
1671         int start;
1672         int len;
1673         int hblock, blocksize;
1674         unsigned long sb_block;
1675         unsigned long offset;
1676         struct ext3_super_block * es;
1677         struct block_device *bdev;
1678
1679         bdev = ext3_blkdev_get(j_dev);
1680         if (bdev == NULL)
1681                 return NULL;
1682
1683         if (bd_claim(bdev, sb)) {
1684                 printk(KERN_ERR
1685                         "EXT3: failed to claim external journal device.\n");
1686                 blkdev_put(bdev);
1687                 return NULL;
1688         }
1689
1690         blocksize = sb->s_blocksize;
1691         hblock = bdev_hardsect_size(bdev);
1692         if (blocksize < hblock) {
1693                 printk(KERN_ERR
1694                         "EXT3-fs: blocksize too small for journal device.\n");
1695                 goto out_bdev;
1696         }
1697
1698         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1699         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1700         set_blocksize(bdev, blocksize);
1701         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1702                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1703                        "external journal\n");
1704                 goto out_bdev;
1705         }
1706
1707         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1708         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1709             !(le32_to_cpu(es->s_feature_incompat) &
1710               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1711                 printk(KERN_ERR "EXT3-fs: external journal has "
1712                                         "bad superblock\n");
1713                 brelse(bh);
1714                 goto out_bdev;
1715         }
1716
1717         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1718                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1719                 brelse(bh);
1720                 goto out_bdev;
1721         }
1722
1723         len = le32_to_cpu(es->s_blocks_count);
1724         start = sb_block + 1;
1725         brelse(bh);     /* we're done with the superblock */
1726
1727         journal = journal_init_dev(bdev, sb->s_bdev,
1728                                         start, len, blocksize);
1729         if (!journal) {
1730                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1731                 goto out_bdev;
1732         }
1733         journal->j_private = sb;
1734         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1735         wait_on_buffer(journal->j_sb_buffer);
1736         if (!buffer_uptodate(journal->j_sb_buffer)) {
1737                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1738                 goto out_journal;
1739         }
1740         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1741                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1742                                         "user (unsupported) - %d\n",
1743                         ntohl(journal->j_superblock->s_nr_users));
1744                 goto out_journal;
1745         }
1746         EXT3_SB(sb)->journal_bdev = bdev;
1747         ext3_init_journal_params(EXT3_SB(sb), journal);
1748         return journal;
1749 out_journal:
1750         journal_destroy(journal);
1751 out_bdev:
1752         ext3_blkdev_put(bdev);
1753         return NULL;
1754 }
1755
1756 static int ext3_load_journal(struct super_block * sb,
1757                              struct ext3_super_block * es)
1758 {
1759         journal_t *journal;
1760         int journal_inum = le32_to_cpu(es->s_journal_inum);
1761         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1762         int err = 0;
1763         int really_read_only;
1764
1765         really_read_only = bdev_read_only(sb->s_bdev);
1766
1767         /*
1768          * Are we loading a blank journal or performing recovery after a
1769          * crash?  For recovery, we need to check in advance whether we
1770          * can get read-write access to the device.
1771          */
1772
1773         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1774                 if (sb->s_flags & MS_RDONLY) {
1775                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1776                                         "required on readonly filesystem.\n");
1777                         if (really_read_only) {
1778                                 printk(KERN_ERR "EXT3-fs: write access "
1779                                         "unavailable, cannot proceed.\n");
1780                                 return -EROFS;
1781                         }
1782                         printk (KERN_INFO "EXT3-fs: write access will "
1783                                         "be enabled during recovery.\n");
1784                 }
1785         }
1786
1787         if (journal_inum && journal_dev) {
1788                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1789                        "and inode journals!\n");
1790                 return -EINVAL;
1791         }
1792
1793         if (journal_inum) {
1794                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1795                         return -EINVAL;
1796         } else {
1797                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1798                         return -EINVAL;
1799         }
1800
1801         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1802                 err = journal_update_format(journal);
1803                 if (err)  {
1804                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1805                         journal_destroy(journal);
1806                         return err;
1807                 }
1808         }
1809
1810         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1811                 err = journal_wipe(journal, !really_read_only);
1812         if (!err)
1813                 err = journal_load(journal);
1814
1815         if (err) {
1816                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1817                 journal_destroy(journal);
1818                 return err;
1819         }
1820
1821         EXT3_SB(sb)->s_journal = journal;
1822         ext3_clear_journal_err(sb, es);
1823         return 0;
1824 }
1825
1826 static int ext3_create_journal(struct super_block * sb,
1827                                struct ext3_super_block * es,
1828                                int journal_inum)
1829 {
1830         journal_t *journal;
1831
1832         if (sb->s_flags & MS_RDONLY) {
1833                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1834                                 "create journal.\n");
1835                 return -EROFS;
1836         }
1837
1838         if (!(journal = ext3_get_journal(sb, journal_inum)))
1839                 return -EINVAL;
1840
1841         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1842                journal_inum);
1843
1844         if (journal_create(journal)) {
1845                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1846                 journal_destroy(journal);
1847                 return -EIO;
1848         }
1849
1850         EXT3_SB(sb)->s_journal = journal;
1851
1852         ext3_update_dynamic_rev(sb);
1853         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1854         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1855
1856         es->s_journal_inum = cpu_to_le32(journal_inum);
1857         sb->s_dirt = 1;
1858
1859         /* Make sure we flush the recovery flag to disk. */
1860         ext3_commit_super(sb, es, 1);
1861
1862         return 0;
1863 }
1864
1865 static void ext3_commit_super (struct super_block * sb,
1866                                struct ext3_super_block * es,
1867                                int sync)
1868 {
1869         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1870
1871         if (!sbh)
1872                 return;
1873         es->s_wtime = cpu_to_le32(get_seconds());
1874         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1875         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1876         BUFFER_TRACE(sbh, "marking dirty");
1877         mark_buffer_dirty(sbh);
1878         if (sync)
1879                 sync_dirty_buffer(sbh);
1880 }
1881
1882
1883 /*
1884  * Have we just finished recovery?  If so, and if we are mounting (or
1885  * remounting) the filesystem readonly, then we will end up with a
1886  * consistent fs on disk.  Record that fact.
1887  */
1888 static void ext3_mark_recovery_complete(struct super_block * sb,
1889                                         struct ext3_super_block * es)
1890 {
1891         journal_t *journal = EXT3_SB(sb)->s_journal;
1892
1893         journal_lock_updates(journal);
1894         journal_flush(journal);
1895         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1896             sb->s_flags & MS_RDONLY) {
1897                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1898                 sb->s_dirt = 0;
1899                 ext3_commit_super(sb, es, 1);
1900         }
1901         journal_unlock_updates(journal);
1902 }
1903
1904 /*
1905  * If we are mounting (or read-write remounting) a filesystem whose journal
1906  * has recorded an error from a previous lifetime, move that error to the
1907  * main filesystem now.
1908  */
1909 static void ext3_clear_journal_err(struct super_block * sb,
1910                                    struct ext3_super_block * es)
1911 {
1912         journal_t *journal;
1913         int j_errno;
1914         const char *errstr;
1915
1916         journal = EXT3_SB(sb)->s_journal;
1917
1918         /*
1919          * Now check for any error status which may have been recorded in the
1920          * journal by a prior ext3_error() or ext3_abort()
1921          */
1922
1923         j_errno = journal_errno(journal);
1924         if (j_errno) {
1925                 char nbuf[16];
1926
1927                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1928                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1929                              "from previous mount: %s", errstr);
1930                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1931                              "filesystem check.");
1932
1933                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1934                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1935                 ext3_commit_super (sb, es, 1);
1936
1937                 journal_clear_err(journal);
1938         }
1939 }
1940
1941 /*
1942  * Force the running and committing transactions to commit,
1943  * and wait on the commit.
1944  */
1945 int ext3_force_commit(struct super_block *sb)
1946 {
1947         journal_t *journal;
1948         int ret;
1949
1950         if (sb->s_flags & MS_RDONLY)
1951                 return 0;
1952
1953         journal = EXT3_SB(sb)->s_journal;
1954         sb->s_dirt = 0;
1955         ret = ext3_journal_force_commit(journal);
1956         return ret;
1957 }
1958
1959 /*
1960  * Ext3 always journals updates to the superblock itself, so we don't
1961  * have to propagate any other updates to the superblock on disk at this
1962  * point.  Just start an async writeback to get the buffers on their way
1963  * to the disk.
1964  *
1965  * This implicitly triggers the writebehind on sync().
1966  */
1967
1968 void ext3_write_super (struct super_block * sb)
1969 {
1970         if (down_trylock(&sb->s_lock) == 0)
1971                 BUG();
1972         sb->s_dirt = 0;
1973 }
1974
1975 static int ext3_sync_fs(struct super_block *sb, int wait)
1976 {
1977         tid_t target;
1978
1979         sb->s_dirt = 0;
1980         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1981                 if (wait)
1982                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1983         }
1984         return 0;
1985 }
1986
1987 /*
1988  * LVM calls this function before a (read-only) snapshot is created.  This
1989  * gives us a chance to flush the journal completely and mark the fs clean.
1990  */
1991 void ext3_write_super_lockfs(struct super_block *sb)
1992 {
1993         sb->s_dirt = 0;
1994
1995         if (!(sb->s_flags & MS_RDONLY)) {
1996                 journal_t *journal = EXT3_SB(sb)->s_journal;
1997
1998                 /* Now we set up the journal barrier. */
1999                 journal_lock_updates(journal);
2000                 journal_flush(journal);
2001
2002                 /* Journal blocked and flushed, clear needs_recovery flag. */
2003                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2004                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2005         }
2006 }
2007
2008 /*
2009  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2010  * flag here, even though the filesystem is not technically dirty yet.
2011  */
2012 void ext3_unlockfs(struct super_block *sb)
2013 {
2014         if (!(sb->s_flags & MS_RDONLY)) {
2015                 lock_super(sb);
2016                 /* Reser the needs_recovery flag before the fs is unlocked. */
2017                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2018                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2019                 unlock_super(sb);
2020                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2021         }
2022 }
2023
2024 int ext3_remount (struct super_block * sb, int * flags, char * data)
2025 {
2026         struct ext3_super_block * es;
2027         struct ext3_sb_info *sbi = EXT3_SB(sb);
2028         unsigned long tmp;
2029
2030         /*
2031          * Allow the "check" option to be passed as a remount option.
2032          */
2033         if (!parse_options(data, sb, &tmp, 1))
2034                 return -EINVAL;
2035
2036         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2037                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2038
2039         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2040                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2041
2042         es = sbi->s_es;
2043
2044         ext3_init_journal_params(sbi, sbi->s_journal);
2045
2046         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2047                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2048                         return -EROFS;
2049
2050                 if (*flags & MS_RDONLY) {
2051                         /*
2052                          * First of all, the unconditional stuff we have to do
2053                          * to disable replay of the journal when we next remount
2054                          */
2055                         sb->s_flags |= MS_RDONLY;
2056
2057                         /*
2058                          * OK, test if we are remounting a valid rw partition
2059                          * readonly, and if so set the rdonly flag and then
2060                          * mark the partition as valid again.
2061                          */
2062                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2063                             (sbi->s_mount_state & EXT3_VALID_FS))
2064                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2065
2066                         ext3_mark_recovery_complete(sb, es);
2067                 } else {
2068                         int ret;
2069                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2070                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2071                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2072                                        "remount RDWR because of unsupported "
2073                                        "optional features (%x).\n",
2074                                        sb->s_id, ret);
2075                                 return -EROFS;
2076                         }
2077                         /*
2078                          * Mounting a RDONLY partition read-write, so reread
2079                          * and store the current valid flag.  (It may have
2080                          * been changed by e2fsck since we originally mounted
2081                          * the partition.)
2082                          */
2083                         ext3_clear_journal_err(sb, es);
2084                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2085                         if (!ext3_setup_super (sb, es, 0))
2086                                 sb->s_flags &= ~MS_RDONLY;
2087                 }
2088         }
2089         return 0;
2090 }
2091
2092 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2093 {
2094         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2095         unsigned long overhead;
2096         int i;
2097
2098         if (test_opt (sb, MINIX_DF))
2099                 overhead = 0;
2100         else {
2101                 /*
2102                  * Compute the overhead (FS structures)
2103                  */
2104
2105                 /*
2106                  * All of the blocks before first_data_block are
2107                  * overhead
2108                  */
2109                 overhead = le32_to_cpu(es->s_first_data_block);
2110
2111                 /*
2112                  * Add the overhead attributed to the superblock and
2113                  * block group descriptors.  If the sparse superblocks
2114                  * feature is turned on, then not all groups have this.
2115                  */
2116                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2117                         overhead += ext3_bg_has_super(sb, i) +
2118                                 ext3_bg_num_gdb(sb, i);
2119
2120                 /*
2121                  * Every block group has an inode bitmap, a block
2122                  * bitmap, and an inode table.
2123                  */
2124                 overhead += (EXT3_SB(sb)->s_groups_count *
2125                              (2 + EXT3_SB(sb)->s_itb_per_group));
2126         }
2127
2128         buf->f_type = EXT3_SUPER_MAGIC;
2129         buf->f_bsize = sb->s_blocksize;
2130         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2131         buf->f_bfree = ext3_count_free_blocks (sb);
2132         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2133         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2134                 buf->f_bavail = 0;
2135         buf->f_files = le32_to_cpu(es->s_inodes_count);
2136         buf->f_ffree = ext3_count_free_inodes (sb);
2137         buf->f_namelen = EXT3_NAME_LEN;
2138         return 0;
2139 }
2140
2141 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2142  * is locked for write. Otherwise the are possible deadlocks:
2143  * Process 1                         Process 2
2144  * ext3_create()                     quota_sync()
2145  *   journal_start()                   write_dquot()
2146  *   DQUOT_INIT()                        down(dqio_sem)
2147  *     down(dqio_sem)                    journal_start()
2148  *
2149  */
2150
2151 #ifdef CONFIG_QUOTA
2152
2153 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2154 {
2155         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2156 }
2157
2158 static int ext3_dquot_initialize(struct inode *inode, int type)
2159 {
2160         handle_t *handle;
2161         int ret, err;
2162
2163         /* We may create quota structure so we need to reserve enough blocks */
2164         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2165         if (IS_ERR(handle))
2166                 return PTR_ERR(handle);
2167         ret = dquot_initialize(inode, type);
2168         err = ext3_journal_stop(handle);
2169         if (!ret)
2170                 ret = err;
2171         return ret;
2172 }
2173
2174 static int ext3_dquot_drop(struct inode *inode)
2175 {
2176         handle_t *handle;
2177         int ret, err;
2178
2179         /* We may delete quota structure so we need to reserve enough blocks */
2180         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2181         if (IS_ERR(handle))
2182                 return PTR_ERR(handle);
2183         ret = dquot_drop(inode);
2184         err = ext3_journal_stop(handle);
2185         if (!ret)
2186                 ret = err;
2187         return ret;
2188 }
2189
2190 static int ext3_write_dquot(struct dquot *dquot)
2191 {
2192         int ret, err;
2193         handle_t *handle;
2194
2195         handle = ext3_journal_start(dquot_to_inode(dquot),
2196                                         EXT3_QUOTA_TRANS_BLOCKS);
2197         if (IS_ERR(handle))
2198                 return PTR_ERR(handle);
2199         ret = dquot_commit(dquot);
2200         err = ext3_journal_stop(handle);
2201         if (!ret)
2202                 ret = err;
2203         return ret;
2204 }
2205
2206 static int ext3_acquire_dquot(struct dquot *dquot)
2207 {
2208         int ret, err;
2209         handle_t *handle;
2210
2211         handle = ext3_journal_start(dquot_to_inode(dquot),
2212                                         EXT3_QUOTA_INIT_BLOCKS);
2213         if (IS_ERR(handle))
2214                 return PTR_ERR(handle);
2215         ret = dquot_acquire(dquot);
2216         err = ext3_journal_stop(handle);
2217         if (!ret)
2218                 ret = err;
2219         return ret;
2220 }
2221
2222 static int ext3_release_dquot(struct dquot *dquot)
2223 {
2224         int ret, err;
2225         handle_t *handle;
2226
2227         handle = ext3_journal_start(dquot_to_inode(dquot),
2228                                         EXT3_QUOTA_INIT_BLOCKS);
2229         if (IS_ERR(handle))
2230                 return PTR_ERR(handle);
2231         ret = dquot_release(dquot);
2232         err = ext3_journal_stop(handle);
2233         if (!ret)
2234                 ret = err;
2235         return ret;
2236 }
2237
2238 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2239 {
2240         /* Are we journalling quotas? */
2241         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2242             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2243                 dquot_mark_dquot_dirty(dquot);
2244                 return ext3_write_dquot(dquot);
2245         } else {
2246                 return dquot_mark_dquot_dirty(dquot);
2247         }
2248 }
2249
2250 static int ext3_write_info(struct super_block *sb, int type)
2251 {
2252         int ret, err;
2253         handle_t *handle;
2254
2255         /* Data block + inode block */
2256         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2257         if (IS_ERR(handle))
2258                 return PTR_ERR(handle);
2259         ret = dquot_commit_info(sb, type);
2260         err = ext3_journal_stop(handle);
2261         if (!ret)
2262                 ret = err;
2263         return ret;
2264 }
2265
2266 /*
2267  * Turn on quotas during mount time - we need to find
2268  * the quota file and such...
2269  */
2270 static int ext3_quota_on_mount(struct super_block *sb, int type)
2271 {
2272         int err;
2273         struct dentry *dentry;
2274         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2275                              .hash = 0,
2276                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2277
2278         dentry = lookup_hash(&name, sb->s_root);
2279         if (IS_ERR(dentry))
2280                 return PTR_ERR(dentry);
2281         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2282         if (err)
2283                 dput(dentry);
2284         /* We keep the dentry reference if everything went ok - we drop it
2285          * on quota_off time */
2286         return err;
2287 }
2288
2289 /* Turn quotas off during mount time */
2290 static int ext3_quota_off_mount(struct super_block *sb, int type)
2291 {
2292         int err;
2293         struct dentry *dentry;
2294
2295         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2296         err = vfs_quota_off_mount(sb, type);
2297         /* We invalidate dentry - it has at least wrong hash... */
2298         d_invalidate(dentry);
2299         dput(dentry);
2300         return err;
2301 }
2302
2303 /*
2304  * Standard function to be called on quota_on
2305  */
2306 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2307                          char *path)
2308 {
2309         int err;
2310         struct nameidata nd;
2311
2312         /* Not journalling quota? */
2313         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2314             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2315                 return vfs_quota_on(sb, type, format_id, path);
2316         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2317         if (err)
2318                 return err;
2319         /* Quotafile not on the same filesystem? */
2320         if (nd.mnt->mnt_sb != sb)
2321                 return -EXDEV;
2322         /* Quotafile not of fs root? */
2323         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2324                 printk(KERN_WARNING
2325                         "EXT3-fs: Quota file not on filesystem root. "
2326                         "Journalled quota will not work.\n");
2327         if (!ext3_should_journal_data(nd.dentry->d_inode))
2328                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2329                         "data-journalling. Journalled quota will not work.\n");
2330         path_release(&nd);
2331         return vfs_quota_on(sb, type, format_id, path);
2332 }
2333
2334 #endif
2335
2336 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2337         int flags, const char *dev_name, void *data)
2338 {
2339         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2340 }
2341
2342 static struct file_system_type ext3_fs_type = {
2343         .owner          = THIS_MODULE,
2344         .name           = "ext3",
2345         .get_sb         = ext3_get_sb,
2346         .kill_sb        = kill_block_super,
2347         .fs_flags       = FS_REQUIRES_DEV,
2348 };
2349
2350 static int __init init_ext3_fs(void)
2351 {
2352         int err = init_ext3_xattr();
2353         if (err)
2354                 return err;
2355         err = init_inodecache();
2356         if (err)
2357                 goto out1;
2358         err = register_filesystem(&ext3_fs_type);
2359         if (err)
2360                 goto out;
2361         return 0;
2362 out:
2363         destroy_inodecache();
2364 out1:
2365         exit_ext3_xattr();
2366         return err;
2367 }
2368
2369 static void __exit exit_ext3_fs(void)
2370 {
2371         unregister_filesystem(&ext3_fs_type);
2372         destroy_inodecache();
2373         exit_ext3_xattr();
2374 }
2375
2376 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2377 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2378 MODULE_LICENSE("GPL");
2379 module_init(init_ext3_fs)
2380 module_exit(exit_ext3_fs)