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