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_sb(struct super_block *sb, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (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_SB(sb)->s_journal;
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(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         if (bh)
 112                 BUFFER_TRACE(bh, "abort");
 113
 114         if (!handle->h_err)
 115                 handle->h_err = err;
 116
 117         if (is_handle_aborted(handle))
 118                 return;
 119
 120         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 121                caller, errstr, err_fn);
 122
 123         journal_abort_handle(handle);
 124 }
 125
 126 /* Deal with the reporting of failure conditions on a filesystem such as
 127  * inconsistencies detected or read IO failures.
 128  *
 129  * On ext2, we can store the error state of the filesystem in the
 130  * superblock.  That is not possible on ext3, because we may have other
 131  * write ordering constraints on the superblock which prevent us from
 132  * writing it out straight away; and given that the journal is about to
 133  * be aborted, we can't rely on the current, or future, transactions to
 134  * write out the superblock safely.
 135  *
 136  * We'll just use the journal_abort() error code to record an error in
 137  * the journal instead.  On recovery, the journal will compain about
 138  * that error until we've noted it down and cleared it.
 139  */
 140
 141 static void ext3_handle_error(struct super_block *sb)
 142 {
 143         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 144
 145         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 146         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 147
 148         if (sb->s_flags & MS_RDONLY)
 149                 return;
 150
 151         if (test_opt (sb, ERRORS_PANIC))
 152                 panic ("EXT3-fs (device %s): panic forced after error\n",
 153                        sb->s_id);
 154         if (test_opt (sb, ERRORS_RO)) {
 155                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 156                 sb->s_flags |= MS_RDONLY;
 157         } else {
 158                 journal_t *journal = EXT3_SB(sb)->s_journal;
 159
 160                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 161                 if (journal)
 162                         journal_abort(journal, -EIO);
 163         }
 164         ext3_commit_super(sb, es, 1);
 165 }
 166
 167 void ext3_error (struct super_block * sb, const char * function,
 168                  const char * fmt, ...)
 169 {
 170         va_list args;
 171
 172         va_start(args, fmt);
 173         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 174         vprintk(fmt, args);
 175         printk("\n");
 176         va_end(args);
 177
 178         ext3_handle_error(sb);
 179 }
 180
 181 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 182 {
 183         char *errstr = NULL;
 184
 185         switch (errno) {
 186         case -EIO:
 187                 errstr = "IO failure";
 188                 break;
 189         case -ENOMEM:
 190                 errstr = "Out of memory";
 191                 break;
 192         case -EROFS:
 193                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 194                         errstr = "Journal has aborted";
 195                 else
 196                         errstr = "Readonly filesystem";
 197                 break;
 198         default:
 199                 /* If the caller passed in an extra buffer for unknown
 200                  * errors, textualise them now.  Else we just return
 201                  * NULL. */
 202                 if (nbuf) {
 203                         /* Check for truncated error codes... */
 204                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 205                                 errstr = nbuf;
 206                 }
 207                 break;
 208         }
 209
 210         return errstr;
 211 }
 212
 213 /* __ext3_std_error decodes expected errors from journaling functions
 214  * automatically and invokes the appropriate error response.  */
 215
 216 void __ext3_std_error (struct super_block * sb, const char * function,
 217                        int errno)
 218 {
 219         char nbuf[16];
 220         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 221
 222         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 223                 sb->s_id, function, errstr);
 224
 225         ext3_handle_error(sb);
 226 }
 227
 228 /*
 229  * ext3_abort is a much stronger failure handler than ext3_error.  The
 230  * abort function may be used to deal with unrecoverable failures such
 231  * as journal IO errors or ENOMEM at a critical moment in log management.
 232  *
 233  * We unconditionally force the filesystem into an ABORT|READONLY state,
 234  * unless the error response on the fs has been set to panic in which
 235  * case we take the easy way out and panic immediately.
 236  */
 237
 238 void ext3_abort (struct super_block * sb, const char * function,
 239                  const char * fmt, ...)
 240 {
 241         va_list args;
 242
 243         printk (KERN_CRIT "ext3_abort called.\n");
 244
 245         va_start(args, fmt);
 246         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 247         vprintk(fmt, args);
 248         printk("\n");
 249         va_end(args);
 250
 251         if (test_opt(sb, ERRORS_PANIC))
 252                 panic("EXT3-fs panic from previous error\n");
 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         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 277         vprintk(fmt, args);
 278         printk("\n");
 279         va_end(args);
 280
 281         /* this is to prevent panic from syncing this filesystem */
 282         /* AKPM: is this sufficient? */
 283         sb->s_flags |= MS_RDONLY;
 284         panic ("EXT3-fs panic forced\n");
 285 }
 286
 287 void ext3_warning (struct super_block * sb, const char * function,
 288                    const char * fmt, ...)
 289 {
 290         va_list args;
 291
 292         va_start(args, fmt);
 293         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 294                sb->s_id, function);
 295         vprintk(fmt, args);
 296         printk("\n");
 297         va_end(args);
 298 }
 299
 300 void ext3_update_dynamic_rev(struct super_block *sb)
 301 {
 302         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 303
 304         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 305                 return;
 306
 307         ext3_warning(sb, __FUNCTION__,
 308                      "updating to rev %d because of new feature flag, "
 309                      "running e2fsck is recommended",
 310                      EXT3_DYNAMIC_REV);
 311
 312         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 313         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 314         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 315         /* leave es->s_feature_*compat flags alone */
 316         /* es->s_uuid will be set by e2fsck if empty */
 317
 318         /*
 319          * The rest of the superblock fields should be zero, and if not it
 320          * means they are likely already in use, so leave them alone.  We
 321          * can leave it up to e2fsck to clean up any inconsistencies there.
 322          */
 323 }
 324
 325 /*
 326  * Open the external journal device
 327  */
 328 static struct block_device *ext3_blkdev_get(dev_t dev)
 329 {
 330         struct block_device *bdev;
 331         char b[BDEVNAME_SIZE];
 332
 333         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 334         if (IS_ERR(bdev))
 335                 goto fail;
 336         return bdev;
 337
 338 fail:
 339         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 340                         __bdevname(dev, b), PTR_ERR(bdev));
 341         return NULL;
 342 }
 343
 344 /*
 345  * Release the journal device
 346  */
 347 static int ext3_blkdev_put(struct block_device *bdev)
 348 {
 349         bd_release(bdev);
 350         return blkdev_put(bdev);
 351 }
 352
 353 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 354 {
 355         struct block_device *bdev;
 356         int ret = -ENODEV;
 357
 358         bdev = sbi->journal_bdev;
 359         if (bdev) {
 360                 ret = ext3_blkdev_put(bdev);
 361                 sbi->journal_bdev = NULL;
 362         }
 363         return ret;
 364 }
 365
 366 static inline struct inode *orphan_list_entry(struct list_head *l)
 367 {
 368         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 369 }
 370
 371 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 372 {
 373         struct list_head *l;
 374
 375         printk(KERN_ERR "sb orphan head is %d\n",
 376                le32_to_cpu(sbi->s_es->s_last_orphan));
 377
 378         printk(KERN_ERR "sb_info orphan list:\n");
 379         list_for_each(l, &sbi->s_orphan) {
 380                 struct inode *inode = orphan_list_entry(l);
 381                 printk(KERN_ERR "  "
 382                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 383                        inode->i_sb->s_id, inode->i_ino, inode,
 384                        inode->i_mode, inode->i_nlink,
 385                        NEXT_ORPHAN(inode));
 386         }
 387 }
 388
 389 void ext3_put_super (struct super_block * sb)
 390 {
 391         struct ext3_sb_info *sbi = EXT3_SB(sb);
 392         struct ext3_super_block *es = sbi->s_es;
 393         int i;
 394
 395         ext3_xattr_put_super(sb);
 396         journal_destroy(sbi->s_journal);
 397         if (!(sb->s_flags & MS_RDONLY)) {
 398                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 399                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 400                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 401                 mark_buffer_dirty(sbi->s_sbh);
 402                 ext3_commit_super(sb, es, 1);
 403         }
 404
 405         for (i = 0; i < sbi->s_gdb_count; i++)
 406                 brelse(sbi->s_group_desc[i]);
 407         kfree(sbi->s_group_desc);
 408         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 409         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 410         percpu_counter_destroy(&sbi->s_dirs_counter);
 411         brelse(sbi->s_sbh);
 412 #ifdef CONFIG_QUOTA
 413         for (i = 0; i < MAXQUOTAS; i++) {
 414                 if (sbi->s_qf_names[i])
 415                         kfree(sbi->s_qf_names[i]);
 416         }
 417 #endif
 418
 419         /* Debugging code just in case the in-memory inode orphan list
 420          * isn't empty.  The on-disk one can be non-empty if we've
 421          * detected an error and taken the fs readonly, but the
 422          * in-memory list had better be clean by this point. */
 423         if (!list_empty(&sbi->s_orphan))
 424                 dump_orphan_list(sb, sbi);
 425         J_ASSERT(list_empty(&sbi->s_orphan));
 426
 427         invalidate_bdev(sb->s_bdev, 0);
 428         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 429                 /*
 430                  * Invalidate the journal device's buffers.  We don't want them
 431                  * floating about in memory - the physical journal device may
 432                  * hotswapped, and it breaks the `ro-after' testing code.
 433                  */
 434                 sync_blockdev(sbi->journal_bdev);
 435                 invalidate_bdev(sbi->journal_bdev, 0);
 436                 ext3_blkdev_remove(sbi);
 437         }
 438         sb->s_fs_info = NULL;
 439         kfree(sbi);
 440         return;
 441 }
 442
 443 static kmem_cache_t *ext3_inode_cachep;
 444
 445 /*
 446  * Called inside transaction, so use GFP_NOFS
 447  */
 448 static struct inode *ext3_alloc_inode(struct super_block *sb)
 449 {
 450         struct ext3_inode_info *ei;
 451
 452         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 453         if (!ei)
 454                 return NULL;
 455 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 456         ei->i_acl = EXT3_ACL_NOT_CACHED;
 457         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 458 #endif
 459         ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 460         ei->vfs_inode.i_version = 1;
 461         return &ei->vfs_inode;
 462 }
 463
 464 static void ext3_destroy_inode(struct inode *inode)
 465 {
 466         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 467 }
 468
 469 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 470 {
 471         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 472
 473         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 474             SLAB_CTOR_CONSTRUCTOR) {
 475                 INIT_LIST_HEAD(&ei->i_orphan);
 476 #ifdef CONFIG_EXT3_FS_XATTR
 477                 init_rwsem(&ei->xattr_sem);
 478 #endif
 479                 init_MUTEX(&ei->truncate_sem);
 480                 inode_init_once(&ei->vfs_inode);
 481         }
 482 }
 483
 484 static int init_inodecache(void)
 485 {
 486         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 487                                              sizeof(struct ext3_inode_info),
 488                                              0, SLAB_RECLAIM_ACCOUNT,
 489                                              init_once, NULL);
 490         if (ext3_inode_cachep == NULL)
 491                 return -ENOMEM;
 492         return 0;
 493 }
 494
 495 static void destroy_inodecache(void)
 496 {
 497         if (kmem_cache_destroy(ext3_inode_cachep))
 498                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 499 }
 500
 501 static void ext3_clear_inode(struct inode *inode)
 502 {
 503 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 504        if (EXT3_I(inode)->i_acl &&
 505            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 506                posix_acl_release(EXT3_I(inode)->i_acl);
 507                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 508        }
 509        if (EXT3_I(inode)->i_default_acl &&
 510            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 511                posix_acl_release(EXT3_I(inode)->i_default_acl);
 512                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 513        }
 514 #endif
 515         ext3_discard_reservation(inode);
 516 }
 517
 518 #ifdef CONFIG_QUOTA
 519
 520 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 521 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 522
 523 static int ext3_dquot_initialize(struct inode *inode, int type);
 524 static int ext3_dquot_drop(struct inode *inode);
 525 static int ext3_write_dquot(struct dquot *dquot);
 526 static int ext3_acquire_dquot(struct dquot *dquot);
 527 static int ext3_release_dquot(struct dquot *dquot);
 528 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 529 static int ext3_write_info(struct super_block *sb, int type);
 530 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 531 static int ext3_quota_on_mount(struct super_block *sb, int type);
 532 static int ext3_quota_off_mount(struct super_block *sb, int type);
 533
 534 static struct dquot_operations ext3_quota_operations = {
 535         .initialize     = ext3_dquot_initialize,
 536         .drop           = ext3_dquot_drop,
 537         .alloc_space    = dquot_alloc_space,
 538         .alloc_inode    = dquot_alloc_inode,
 539         .free_space     = dquot_free_space,
 540         .free_inode     = dquot_free_inode,
 541         .transfer       = dquot_transfer,
 542         .write_dquot    = ext3_write_dquot,
 543         .acquire_dquot  = ext3_acquire_dquot,
 544         .release_dquot  = ext3_release_dquot,
 545         .mark_dirty     = ext3_mark_dquot_dirty,
 546         .write_info     = ext3_write_info
 547 };
 548
 549 static struct quotactl_ops ext3_qctl_operations = {
 550         .quota_on       = ext3_quota_on,
 551         .quota_off      = vfs_quota_off,
 552         .quota_sync     = vfs_quota_sync,
 553         .get_info       = vfs_get_dqinfo,
 554         .set_info       = vfs_set_dqinfo,
 555         .get_dqblk      = vfs_get_dqblk,
 556         .set_dqblk      = vfs_set_dqblk
 557 };
 558 #endif
 559
 560 static struct super_operations ext3_sops = {
 561         .alloc_inode    = ext3_alloc_inode,
 562         .destroy_inode  = ext3_destroy_inode,
 563         .read_inode     = ext3_read_inode,
 564         .write_inode    = ext3_write_inode,
 565         .dirty_inode    = ext3_dirty_inode,
 566         .delete_inode   = ext3_delete_inode,
 567         .put_super      = ext3_put_super,
 568         .write_super    = ext3_write_super,
 569         .sync_fs        = ext3_sync_fs,
 570         .write_super_lockfs = ext3_write_super_lockfs,
 571         .unlockfs       = ext3_unlockfs,
 572         .statfs         = ext3_statfs,
 573         .remount_fs     = ext3_remount,
 574         .clear_inode    = ext3_clear_inode,
 575 };
 576
 577 struct dentry *ext3_get_parent(struct dentry *child);
 578 static struct export_operations ext3_export_ops = {
 579         .get_parent = ext3_get_parent,
 580 };
 581
 582 enum {
 583         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 584         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 585         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 586         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 587         Opt_reservation, Opt_noreservation, Opt_noload,
 588         Opt_commit, Opt_journal_update, Opt_journal_inum,
 589         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 590         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 591         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 592         Opt_tagxid, Opt_barrier, Opt_ignore, Opt_err, Opt_resize,
 593 };
 594
 595 static match_table_t tokens = {
 596         {Opt_bsd_df, "bsddf"},
 597         {Opt_minix_df, "minixdf"},
 598         {Opt_grpid, "grpid"},
 599         {Opt_grpid, "bsdgroups"},
 600         {Opt_nogrpid, "nogrpid"},
 601         {Opt_nogrpid, "sysvgroups"},
 602         {Opt_resgid, "resgid=%u"},
 603         {Opt_resuid, "resuid=%u"},
 604         {Opt_sb, "sb=%u"},
 605         {Opt_err_cont, "errors=continue"},
 606         {Opt_err_panic, "errors=panic"},
 607         {Opt_err_ro, "errors=remount-ro"},
 608         {Opt_nouid32, "nouid32"},
 609         {Opt_nocheck, "nocheck"},
 610         {Opt_nocheck, "check=none"},
 611         {Opt_check, "check"},
 612         {Opt_debug, "debug"},
 613         {Opt_oldalloc, "oldalloc"},
 614         {Opt_orlov, "orlov"},
 615         {Opt_user_xattr, "user_xattr"},
 616         {Opt_nouser_xattr, "nouser_xattr"},
 617         {Opt_acl, "acl"},
 618         {Opt_noacl, "noacl"},
 619         {Opt_reservation, "reservation"},
 620         {Opt_noreservation, "noreservation"},
 621         {Opt_noload, "noload"},
 622         {Opt_commit, "commit=%u"},
 623         {Opt_journal_update, "journal=update"},
 624         {Opt_journal_inum, "journal=%u"},
 625         {Opt_abort, "abort"},
 626         {Opt_data_journal, "data=journal"},
 627         {Opt_data_ordered, "data=ordered"},
 628         {Opt_data_writeback, "data=writeback"},
 629         {Opt_offusrjquota, "usrjquota="},
 630         {Opt_usrjquota, "usrjquota=%s"},
 631         {Opt_offgrpjquota, "grpjquota="},
 632         {Opt_grpjquota, "grpjquota=%s"},
 633         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 634         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 635         {Opt_tagxid, "tagxid"},
 636         {Opt_ignore, "grpquota"},
 637         {Opt_ignore, "noquota"},
 638         {Opt_ignore, "quota"},
 639         {Opt_ignore, "usrquota"},
 640         {Opt_barrier, "barrier=%u"},
 641         {Opt_err, NULL},
 642         {Opt_resize, "resize"},
 643 };
 644
 645 static unsigned long get_sb_block(void **data)
 646 {
 647         unsigned long   sb_block;
 648         char            *options = (char *) *data;
 649
 650         if (!options || strncmp(options, "sb=", 3) != 0)
 651                 return 1;       /* Default location */
 652         options += 3;
 653         sb_block = simple_strtoul(options, &options, 0);
 654         if (*options && *options != ',') {
 655                 printk("EXT3-fs: Invalid sb specification: %s\n",
 656                        (char *) *data);
 657                 return 1;
 658         }
 659         if (*options == ',')
 660                 options++;
 661         *data = (void *) options;
 662         return sb_block;
 663 }
 664
 665 static int parse_options (char * options, struct super_block *sb,
 666                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 667 {
 668         struct ext3_sb_info *sbi = EXT3_SB(sb);
 669         char * p;
 670         substring_t args[MAX_OPT_ARGS];
 671         int data_opt = 0;
 672         int option;
 673 #ifdef CONFIG_QUOTA
 674         int qtype;
 675 #endif
 676
 677         if (!options)
 678                 return 1;
 679
 680         while ((p = strsep (&options, ",")) != NULL) {
 681                 int token;
 682                 if (!*p)
 683                         continue;
 684
 685                 token = match_token(p, tokens, args);
 686                 switch (token) {
 687                 case Opt_bsd_df:
 688                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 689                         break;
 690                 case Opt_minix_df:
 691                         set_opt (sbi->s_mount_opt, MINIX_DF);
 692                         break;
 693                 case Opt_grpid:
 694                         set_opt (sbi->s_mount_opt, GRPID);
 695                         break;
 696                 case Opt_nogrpid:
 697                         clear_opt (sbi->s_mount_opt, GRPID);
 698                         break;
 699                 case Opt_resuid:
 700                         if (match_int(&args[0], &option))
 701                                 return 0;
 702                         sbi->s_resuid = option;
 703                         break;
 704                 case Opt_resgid:
 705                         if (match_int(&args[0], &option))
 706                                 return 0;
 707                         sbi->s_resgid = option;
 708                         break;
 709                 case Opt_sb:
 710                         /* handled by get_sb_block() instead of here */
 711                         /* *sb_block = match_int(&args[0]); */
 712                         break;
 713                 case Opt_err_panic:
 714                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 715                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 716                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 717                         break;
 718                 case Opt_err_ro:
 719                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 720                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 721                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 722                         break;
 723                 case Opt_err_cont:
 724                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 725                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 726                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 727                         break;
 728                 case Opt_nouid32:
 729                         set_opt (sbi->s_mount_opt, NO_UID32);
 730                         break;
 731 #ifndef CONFIG_INOXID_NONE
 732                 case Opt_tagxid:
 733                         if (is_remount) {
 734                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 735                                        "tagxid on remount\n");
 736                                 return 0;
 737                         }
 738                         set_opt (sbi->s_mount_opt, TAG_XID);
 739                         break;
 740 #endif
 741                 case Opt_check:
 742 #ifdef CONFIG_EXT3_CHECK
 743                         set_opt (sbi->s_mount_opt, CHECK);
 744 #else
 745                         printk(KERN_ERR
 746                                "EXT3 Check option not supported\n");
 747 #endif
 748                         break;
 749                 case Opt_nocheck:
 750                         clear_opt (sbi->s_mount_opt, CHECK);
 751                         break;
 752                 case Opt_debug:
 753                         set_opt (sbi->s_mount_opt, DEBUG);
 754                         break;
 755                 case Opt_oldalloc:
 756                         set_opt (sbi->s_mount_opt, OLDALLOC);
 757                         break;
 758                 case Opt_orlov:
 759                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 760                         break;
 761 #ifdef CONFIG_EXT3_FS_XATTR
 762                 case Opt_user_xattr:
 763                         set_opt (sbi->s_mount_opt, XATTR_USER);
 764                         break;
 765                 case Opt_nouser_xattr:
 766                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 767                         break;
 768 #else
 769                 case Opt_user_xattr:
 770                 case Opt_nouser_xattr:
 771                         printk("EXT3 (no)user_xattr options not supported\n");
 772                         break;
 773 #endif
 774 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 775                 case Opt_acl:
 776                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 777                         break;
 778                 case Opt_noacl:
 779                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 780                         break;
 781 #else
 782                 case Opt_acl:
 783                 case Opt_noacl:
 784                         printk("EXT3 (no)acl options not supported\n");
 785                         break;
 786 #endif
 787                 case Opt_reservation:
 788                         set_opt(sbi->s_mount_opt, RESERVATION);
 789                         break;
 790                 case Opt_noreservation:
 791                         clear_opt(sbi->s_mount_opt, RESERVATION);
 792                         break;
 793                 case Opt_journal_update:
 794                         /* @@@ FIXME */
 795                         /* Eventually we will want to be able to create
 796                            a journal file here.  For now, only allow the
 797                            user to specify an existing inode to be the
 798                            journal file. */
 799                         if (is_remount) {
 800                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 801                                        "journal on remount\n");
 802                                 return 0;
 803                         }
 804                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 805                         break;
 806                 case Opt_journal_inum:
 807                         if (is_remount) {
 808                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 809                                        "journal on remount\n");
 810                                 return 0;
 811                         }
 812                         if (match_int(&args[0], &option))
 813                                 return 0;
 814                         *inum = option;
 815                         break;
 816                 case Opt_noload:
 817                         set_opt (sbi->s_mount_opt, NOLOAD);
 818                         break;
 819                 case Opt_commit:
 820                         if (match_int(&args[0], &option))
 821                                 return 0;
 822                         if (option < 0)
 823                                 return 0;
 824                         if (option == 0)
 825                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 826                         sbi->s_commit_interval = HZ * option;
 827                         break;
 828                 case Opt_data_journal:
 829                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 830                         goto datacheck;
 831                 case Opt_data_ordered:
 832                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 833                         goto datacheck;
 834                 case Opt_data_writeback:
 835                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 836                 datacheck:
 837                         if (is_remount) {
 838                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 839                                                 != data_opt) {
 840                                         printk(KERN_ERR
 841                                                 "EXT3-fs: cannot change data "
 842                                                 "mode on remount\n");
 843                                         return 0;
 844                                 }
 845                         } else {
 846                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 847                                 sbi->s_mount_opt |= data_opt;
 848                         }
 849                         break;
 850 #ifdef CONFIG_QUOTA
 851                 case Opt_usrjquota:
 852                         qtype = USRQUOTA;
 853                         goto set_qf_name;
 854                 case Opt_grpjquota:
 855                         qtype = GRPQUOTA;
 856 set_qf_name:
 857                         if (sb_any_quota_enabled(sb)) {
 858                                 printk(KERN_ERR
 859                                         "EXT3-fs: Cannot change journalled "
 860                                         "quota options when quota turned on.\n");
 861                                 return 0;
 862                         }
 863                         if (sbi->s_qf_names[qtype]) {
 864                                 printk(KERN_ERR
 865                                         "EXT3-fs: %s quota file already "
 866                                         "specified.\n", QTYPE2NAME(qtype));
 867                                 return 0;
 868                         }
 869                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 870                         if (!sbi->s_qf_names[qtype]) {
 871                                 printk(KERN_ERR
 872                                         "EXT3-fs: not enough memory for "
 873                                         "storing quotafile name.\n");
 874                                 return 0;
 875                         }
 876                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 877                                 printk(KERN_ERR
 878                                         "EXT3-fs: quotafile must be on "
 879                                         "filesystem root.\n");
 880                                 kfree(sbi->s_qf_names[qtype]);
 881                                 sbi->s_qf_names[qtype] = NULL;
 882                                 return 0;
 883                         }
 884                         break;
 885                 case Opt_offusrjquota:
 886                         qtype = USRQUOTA;
 887                         goto clear_qf_name;
 888                 case Opt_offgrpjquota:
 889                         qtype = GRPQUOTA;
 890 clear_qf_name:
 891                         if (sb_any_quota_enabled(sb)) {
 892                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 893                                         "journalled quota options when "
 894                                         "quota turned on.\n");
 895                                 return 0;
 896                         }
 897                         if (sbi->s_qf_names[qtype]) {
 898                                 kfree(sbi->s_qf_names[qtype]);
 899                                 sbi->s_qf_names[qtype] = NULL;
 900                         }
 901                         break;
 902                 case Opt_jqfmt_vfsold:
 903                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 904                         break;
 905                 case Opt_jqfmt_vfsv0:
 906                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 907                         break;
 908 #else
 909                 case Opt_usrjquota:
 910                 case Opt_grpjquota:
 911                 case Opt_offusrjquota:
 912                 case Opt_offgrpjquota:
 913                 case Opt_jqfmt_vfsold:
 914                 case Opt_jqfmt_vfsv0:
 915                         printk(KERN_ERR
 916                                 "EXT3-fs: journalled quota options not "
 917                                 "supported.\n");
 918                         break;
 919 #endif
 920                 case Opt_abort:
 921                         set_opt(sbi->s_mount_opt, ABORT);
 922                         break;
 923                 case Opt_barrier:
 924                         if (match_int(&args[0], &option))
 925                                 return 0;
 926                         if (option)
 927                                 set_opt(sbi->s_mount_opt, BARRIER);
 928                         else
 929                                 clear_opt(sbi->s_mount_opt, BARRIER);
 930                         break;
 931                 case Opt_ignore:
 932                         break;
 933                 case Opt_resize:
 934                         if (!n_blocks_count) {
 935                                 printk("EXT3-fs: resize option only available "
 936                                         "for remount\n");
 937                                 return 0;
 938                         }
 939                         match_int(&args[0], &option);
 940                         *n_blocks_count = option;
 941                         break;
 942                 default:
 943                         printk (KERN_ERR
 944                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 945                                 "or missing value\n", p);
 946                         return 0;
 947                 }
 948         }
 949 #ifdef CONFIG_QUOTA
 950         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 951             sbi->s_qf_names[GRPQUOTA])) {
 952                 printk(KERN_ERR
 953                         "EXT3-fs: journalled quota format not specified.\n");
 954                 return 0;
 955         }
 956 #endif
 957
 958         return 1;
 959 }
 960
 961 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 962                             int read_only)
 963 {
 964         struct ext3_sb_info *sbi = EXT3_SB(sb);
 965         int res = 0;
 966
 967         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 968                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 969                         "forcing read-only mode\n");
 970                 res = MS_RDONLY;
 971         }
 972         if (read_only)
 973                 return res;
 974         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 975                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 976                         "running e2fsck is recommended\n");
 977         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 978                 printk (KERN_WARNING
 979                         "EXT3-fs warning: mounting fs with errors, "
 980                         "running e2fsck is recommended\n");
 981         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 982                  le16_to_cpu(es->s_mnt_count) >=
 983                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 984                 printk (KERN_WARNING
 985                         "EXT3-fs warning: maximal mount count reached, "
 986                         "running e2fsck is recommended\n");
 987         else if (le32_to_cpu(es->s_checkinterval) &&
 988                 (le32_to_cpu(es->s_lastcheck) +
 989                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 990                 printk (KERN_WARNING
 991                         "EXT3-fs warning: checktime reached, "
 992                         "running e2fsck is recommended\n");
 993 #if 0
 994                 /* @@@ We _will_ want to clear the valid bit if we find
 995                    inconsistencies, to force a fsck at reboot.  But for
 996                    a plain journaled filesystem we can keep it set as
 997                    valid forever! :) */
 998         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 999 #endif
1000         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1001                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1002         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1003         es->s_mtime = cpu_to_le32(get_seconds());
1004         ext3_update_dynamic_rev(sb);
1005         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1006
1007         ext3_commit_super(sb, es, 1);
1008         if (test_opt(sb, DEBUG))
1009                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1010                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1011                         sb->s_blocksize,
1012                         sbi->s_groups_count,
1013                         EXT3_BLOCKS_PER_GROUP(sb),
1014                         EXT3_INODES_PER_GROUP(sb),
1015                         sbi->s_mount_opt);
1016
1017         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1018         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1019                 char b[BDEVNAME_SIZE];
1020
1021                 printk("external journal on %s\n",
1022                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1023         } else {
1024                 printk("internal journal\n");
1025         }
1026 #ifdef CONFIG_EXT3_CHECK
1027         if (test_opt (sb, CHECK)) {
1028                 ext3_check_blocks_bitmap (sb);
1029                 ext3_check_inodes_bitmap (sb);
1030         }
1031 #endif
1032         return res;
1033 }
1034
1035 /* Called at mount-time, super-block is locked */
1036 static int ext3_check_descriptors (struct super_block * sb)
1037 {
1038         struct ext3_sb_info *sbi = EXT3_SB(sb);
1039         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1040         struct ext3_group_desc * gdp = NULL;
1041         int desc_block = 0;
1042         int i;
1043
1044         ext3_debug ("Checking group descriptors");
1045
1046         for (i = 0; i < sbi->s_groups_count; i++)
1047         {
1048                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1049                         gdp = (struct ext3_group_desc *)
1050                                         sbi->s_group_desc[desc_block++]->b_data;
1051                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1052                     le32_to_cpu(gdp->bg_block_bitmap) >=
1053                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1054                 {
1055                         ext3_error (sb, "ext3_check_descriptors",
1056                                     "Block bitmap for group %d"
1057                                     " not in group (block %lu)!",
1058                                     i, (unsigned long)
1059                                         le32_to_cpu(gdp->bg_block_bitmap));
1060                         return 0;
1061                 }
1062                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1063                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1064                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1065                 {
1066                         ext3_error (sb, "ext3_check_descriptors",
1067                                     "Inode bitmap for group %d"
1068                                     " not in group (block %lu)!",
1069                                     i, (unsigned long)
1070                                         le32_to_cpu(gdp->bg_inode_bitmap));
1071                         return 0;
1072                 }
1073                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1074                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1075                     block + EXT3_BLOCKS_PER_GROUP(sb))
1076                 {
1077                         ext3_error (sb, "ext3_check_descriptors",
1078                                     "Inode table for group %d"
1079                                     " not in group (block %lu)!",
1080                                     i, (unsigned long)
1081                                         le32_to_cpu(gdp->bg_inode_table));
1082                         return 0;
1083                 }
1084                 block += EXT3_BLOCKS_PER_GROUP(sb);
1085                 gdp++;
1086         }
1087
1088         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1089         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1090         return 1;
1091 }
1092
1093
1094 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1095  * the superblock) which were deleted from all directories, but held open by
1096  * a process at the time of a crash.  We walk the list and try to delete these
1097  * inodes at recovery time (only with a read-write filesystem).
1098  *
1099  * In order to keep the orphan inode chain consistent during traversal (in
1100  * case of crash during recovery), we link each inode into the superblock
1101  * orphan list_head and handle it the same way as an inode deletion during
1102  * normal operation (which journals the operations for us).
1103  *
1104  * We only do an iget() and an iput() on each inode, which is very safe if we
1105  * accidentally point at an in-use or already deleted inode.  The worst that
1106  * can happen in this case is that we get a "bit already cleared" message from
1107  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1108  * e2fsck was run on this filesystem, and it must have already done the orphan
1109  * inode cleanup for us, so we can safely abort without any further action.
1110  */
1111 static void ext3_orphan_cleanup (struct super_block * sb,
1112                                  struct ext3_super_block * es)
1113 {
1114         unsigned int s_flags = sb->s_flags;
1115         int nr_orphans = 0, nr_truncates = 0;
1116 #ifdef CONFIG_QUOTA
1117         int i;
1118 #endif
1119         if (!es->s_last_orphan) {
1120                 jbd_debug(4, "no orphan inodes to clean up\n");
1121                 return;
1122         }
1123
1124         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1125                 if (es->s_last_orphan)
1126                         jbd_debug(1, "Errors on filesystem, "
1127                                   "clearing orphan list.\n");
1128                 es->s_last_orphan = 0;
1129                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1130                 return;
1131         }
1132
1133         if (s_flags & MS_RDONLY) {
1134                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1135                        sb->s_id);
1136                 sb->s_flags &= ~MS_RDONLY;
1137         }
1138 #ifdef CONFIG_QUOTA
1139         /* Needed for iput() to work correctly and not trash data */
1140         sb->s_flags |= MS_ACTIVE;
1141         /* Turn on quotas so that they are updated correctly */
1142         for (i = 0; i < MAXQUOTAS; i++) {
1143                 if (EXT3_SB(sb)->s_qf_names[i]) {
1144                         int ret = ext3_quota_on_mount(sb, i);
1145                         if (ret < 0)
1146                                 printk(KERN_ERR
1147                                         "EXT3-fs: Cannot turn on journalled "
1148                                         "quota: error %d\n", ret);
1149                 }
1150         }
1151 #endif
1152
1153         while (es->s_last_orphan) {
1154                 struct inode *inode;
1155
1156                 if (!(inode =
1157                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1158                         es->s_last_orphan = 0;
1159                         break;
1160                 }
1161
1162                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1163                 DQUOT_INIT(inode);
1164                 if (inode->i_nlink) {
1165                         printk(KERN_DEBUG
1166                                 "%s: truncating inode %ld to %Ld bytes\n",
1167                                 __FUNCTION__, inode->i_ino, inode->i_size);
1168                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1169                                   inode->i_ino, inode->i_size);
1170                         ext3_truncate(inode);
1171                         nr_truncates++;
1172                 } else {
1173                         printk(KERN_DEBUG
1174                                 "%s: deleting unreferenced inode %ld\n",
1175                                 __FUNCTION__, inode->i_ino);
1176                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1177                                   inode->i_ino);
1178                         nr_orphans++;
1179                 }
1180                 iput(inode);  /* The delete magic happens here! */
1181         }
1182
1183 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1184
1185         if (nr_orphans)
1186                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1187                        sb->s_id, PLURAL(nr_orphans));
1188         if (nr_truncates)
1189                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1190                        sb->s_id, PLURAL(nr_truncates));
1191 #ifdef CONFIG_QUOTA
1192         /* Turn quotas off */
1193         for (i = 0; i < MAXQUOTAS; i++) {
1194                 if (sb_dqopt(sb)->files[i])
1195                         ext3_quota_off_mount(sb, i);
1196         }
1197 #endif
1198         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1199 }
1200
1201 #define log2(n) ffz(~(n))
1202
1203 /*
1204  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1205  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1206  * We need to be 1 filesystem block less than the 2^32 sector limit.
1207  */
1208 static loff_t ext3_max_size(int bits)
1209 {
1210         loff_t res = EXT3_NDIR_BLOCKS;
1211         res += 1LL << (bits-2);
1212         res += 1LL << (2*(bits-2));
1213         res += 1LL << (3*(bits-2));
1214         res <<= bits;
1215         if (res > (512LL << 32) - (1 << bits))
1216                 res = (512LL << 32) - (1 << bits);
1217         return res;
1218 }
1219
1220 static unsigned long descriptor_loc(struct super_block *sb,
1221                                     unsigned long logic_sb_block,
1222                                     int nr)
1223 {
1224         struct ext3_sb_info *sbi = EXT3_SB(sb);
1225         unsigned long bg, first_data_block, first_meta_bg;
1226         int has_super = 0;
1227
1228         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1229         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1230
1231         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1232             nr < first_meta_bg)
1233                 return (logic_sb_block + nr + 1);
1234         bg = sbi->s_desc_per_block * nr;
1235         if (ext3_bg_has_super(sb, bg))
1236                 has_super = 1;
1237         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1238 }
1239
1240
1241 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1242 {
1243         struct buffer_head * bh;
1244         struct ext3_super_block *es = NULL;
1245         struct ext3_sb_info *sbi;
1246         unsigned long block;
1247         unsigned long sb_block = get_sb_block(&data);
1248         unsigned long logic_sb_block;
1249         unsigned long offset = 0;
1250         unsigned long journal_inum = 0;
1251         unsigned long def_mount_opts;
1252         struct inode *root;
1253         int blocksize;
1254         int hblock;
1255         int db_count;
1256         int i;
1257         int needs_recovery;
1258         __le32 features;
1259
1260         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1261         if (!sbi)
1262                 return -ENOMEM;
1263         sb->s_fs_info = sbi;
1264         memset(sbi, 0, sizeof(*sbi));
1265         sbi->s_mount_opt = 0;
1266         sbi->s_resuid = EXT3_DEF_RESUID;
1267         sbi->s_resgid = EXT3_DEF_RESGID;
1268
1269         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1270         if (!blocksize) {
1271                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1272                 goto out_fail;
1273         }
1274
1275         /*
1276          * The ext3 superblock will not be buffer aligned for other than 1kB
1277          * block sizes.  We need to calculate the offset from buffer start.
1278          */
1279         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1280                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1281                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1282         } else {
1283                 logic_sb_block = sb_block;
1284         }
1285
1286         if (!(bh = sb_bread(sb, logic_sb_block))) {
1287                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1288                 goto out_fail;
1289         }
1290         /*
1291          * Note: s_es must be initialized as soon as possible because
1292          *       some ext3 macro-instructions depend on its value
1293          */
1294         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1295         sbi->s_es = es;
1296         sb->s_magic = le16_to_cpu(es->s_magic);
1297         if (sb->s_magic != EXT3_SUPER_MAGIC)
1298                 goto cantfind_ext3;
1299
1300         /* Set defaults before we parse the mount options */
1301         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1302         if (def_mount_opts & EXT3_DEFM_DEBUG)
1303                 set_opt(sbi->s_mount_opt, DEBUG);
1304         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1305                 set_opt(sbi->s_mount_opt, GRPID);
1306         if (def_mount_opts & EXT3_DEFM_UID16)
1307                 set_opt(sbi->s_mount_opt, NO_UID32);
1308         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1309                 set_opt(sbi->s_mount_opt, XATTR_USER);
1310         if (def_mount_opts & EXT3_DEFM_ACL)
1311                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1312         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1313                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1314         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1315                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1316         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1317                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1318
1319         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1320                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1321         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1322                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1323
1324         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1325         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1326
1327         set_opt(sbi->s_mount_opt, RESERVATION);
1328
1329         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1330                 goto failed_mount;
1331
1332         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1333                 sb->s_flags |= MS_TAGXID;
1334         sb->s_flags |= MS_ONE_SECOND;
1335         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1336                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1337
1338         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1339             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1340              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1341              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1342                 printk(KERN_WARNING
1343                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1344                        "running e2fsck is recommended\n");
1345         /*
1346          * Check feature flags regardless of the revision level, since we
1347          * previously didn't change the revision level when setting the flags,
1348          * so there is a chance incompat flags are set on a rev 0 filesystem.
1349          */
1350         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1351         if (features) {
1352                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1353                        "unsupported optional features (%x).\n",
1354                        sb->s_id, le32_to_cpu(features));
1355                 goto failed_mount;
1356         }
1357         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1358         if (!(sb->s_flags & MS_RDONLY) && features) {
1359                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1360                        "unsupported optional features (%x).\n",
1361                        sb->s_id, le32_to_cpu(features));
1362                 goto failed_mount;
1363         }
1364         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1365
1366         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1367             blocksize > EXT3_MAX_BLOCK_SIZE) {
1368                 printk(KERN_ERR
1369                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1370                        blocksize, sb->s_id);
1371                 goto failed_mount;
1372         }
1373
1374         hblock = bdev_hardsect_size(sb->s_bdev);
1375         if (sb->s_blocksize != blocksize) {
1376                 /*
1377                  * Make sure the blocksize for the filesystem is larger
1378                  * than the hardware sectorsize for the machine.
1379                  */
1380                 if (blocksize < hblock) {
1381                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1382                                "device blocksize %d.\n", blocksize, hblock);
1383                         goto failed_mount;
1384                 }
1385
1386                 brelse (bh);
1387                 sb_set_blocksize(sb, blocksize);
1388                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1389                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1390                 bh = sb_bread(sb, logic_sb_block);
1391                 if (!bh) {
1392                         printk(KERN_ERR
1393                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1394                         goto failed_mount;
1395                 }
1396                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1397                 sbi->s_es = es;
1398                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1399                         printk (KERN_ERR
1400                                 "EXT3-fs: Magic mismatch, very weird !\n");
1401                         goto failed_mount;
1402                 }
1403         }
1404
1405         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1406
1407         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1408                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1409                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1410         } else {
1411                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1412                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1413                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1414                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1415                     (sbi->s_inode_size > blocksize)) {
1416                         printk (KERN_ERR
1417                                 "EXT3-fs: unsupported inode size: %d\n",
1418                                 sbi->s_inode_size);
1419                         goto failed_mount;
1420                 }
1421         }
1422         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1423                                    le32_to_cpu(es->s_log_frag_size);
1424         if (blocksize != sbi->s_frag_size) {
1425                 printk(KERN_ERR
1426                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1427                        sbi->s_frag_size, blocksize);
1428                 goto failed_mount;
1429         }
1430         sbi->s_frags_per_block = 1;
1431         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1432         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1433         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1434         if (EXT3_INODE_SIZE(sb) == 0)
1435                 goto cantfind_ext3;
1436         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1437         if (sbi->s_inodes_per_block == 0)
1438                 goto cantfind_ext3;
1439         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1440                                         sbi->s_inodes_per_block;
1441         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1442         sbi->s_sbh = bh;
1443         sbi->s_mount_state = le16_to_cpu(es->s_state);
1444         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1445         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1446         for (i=0; i < 4; i++)
1447                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1448         sbi->s_def_hash_version = es->s_def_hash_version;
1449
1450         if (sbi->s_blocks_per_group > blocksize * 8) {
1451                 printk (KERN_ERR
1452                         "EXT3-fs: #blocks per group too big: %lu\n",
1453                         sbi->s_blocks_per_group);
1454                 goto failed_mount;
1455         }
1456         if (sbi->s_frags_per_group > blocksize * 8) {
1457                 printk (KERN_ERR
1458                         "EXT3-fs: #fragments per group too big: %lu\n",
1459                         sbi->s_frags_per_group);
1460                 goto failed_mount;
1461         }
1462         if (sbi->s_inodes_per_group > blocksize * 8) {
1463                 printk (KERN_ERR
1464                         "EXT3-fs: #inodes per group too big: %lu\n",
1465                         sbi->s_inodes_per_group);
1466                 goto failed_mount;
1467         }
1468
1469         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1470                 goto cantfind_ext3;
1471         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1472                                le32_to_cpu(es->s_first_data_block) +
1473                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1474                               EXT3_BLOCKS_PER_GROUP(sb);
1475         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1476                    EXT3_DESC_PER_BLOCK(sb);
1477         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1478                                     GFP_KERNEL);
1479         if (sbi->s_group_desc == NULL) {
1480                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1481                 goto failed_mount;
1482         }
1483
1484         percpu_counter_init(&sbi->s_freeblocks_counter);
1485         percpu_counter_init(&sbi->s_freeinodes_counter);
1486         percpu_counter_init(&sbi->s_dirs_counter);
1487         bgl_lock_init(&sbi->s_blockgroup_lock);
1488
1489         for (i = 0; i < db_count; i++) {
1490                 block = descriptor_loc(sb, logic_sb_block, i);
1491                 sbi->s_group_desc[i] = sb_bread(sb, block);
1492                 if (!sbi->s_group_desc[i]) {
1493                         printk (KERN_ERR "EXT3-fs: "
1494                                 "can't read group descriptor %d\n", i);
1495                         db_count = i;
1496                         goto failed_mount2;
1497                 }
1498         }
1499         if (!ext3_check_descriptors (sb)) {
1500                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1501                 goto failed_mount2;
1502         }
1503         sbi->s_gdb_count = db_count;
1504         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1505         spin_lock_init(&sbi->s_next_gen_lock);
1506         /* per fileystem reservation list head & lock */
1507         spin_lock_init(&sbi->s_rsv_window_lock);
1508         sbi->s_rsv_window_root = RB_ROOT;
1509         /* Add a single, static dummy reservation to the start of the
1510          * reservation window list --- it gives us a placeholder for
1511          * append-at-start-of-list which makes the allocation logic
1512          * _much_ simpler. */
1513         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1514         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1515         atomic_set(&sbi->s_rsv_window_head.rsv_alloc_hit, 0);
1516         atomic_set(&sbi->s_rsv_window_head.rsv_goal_size, 0);
1517         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1518
1519         /*
1520          * set up enough so that it can read an inode
1521          */
1522         sb->s_op = &ext3_sops;
1523         sb->s_export_op = &ext3_export_ops;
1524         sb->s_xattr = ext3_xattr_handlers;
1525 #ifdef CONFIG_QUOTA
1526         sb->s_qcop = &ext3_qctl_operations;
1527         sb->dq_op = &ext3_quota_operations;
1528 #endif
1529         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1530
1531         sb->s_root = NULL;
1532
1533         needs_recovery = (es->s_last_orphan != 0 ||
1534                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1535                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1536
1537         /*
1538          * The first inode we look at is the journal inode.  Don't try
1539          * root first: it may be modified in the journal!
1540          */
1541         if (!test_opt(sb, NOLOAD) &&
1542             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1543                 if (ext3_load_journal(sb, es))
1544                         goto failed_mount2;
1545         } else if (journal_inum) {
1546                 if (ext3_create_journal(sb, es, journal_inum))
1547                         goto failed_mount2;
1548         } else {
1549                 if (!silent)
1550                         printk (KERN_ERR
1551                                 "ext3: No journal on filesystem on %s\n",
1552                                 sb->s_id);
1553                 goto failed_mount2;
1554         }
1555
1556         /* We have now updated the journal if required, so we can
1557          * validate the data journaling mode. */
1558         switch (test_opt(sb, DATA_FLAGS)) {
1559         case 0:
1560                 /* No mode set, assume a default based on the journal
1561                    capabilities: ORDERED_DATA if the journal can
1562                    cope, else JOURNAL_DATA */
1563                 if (journal_check_available_features
1564                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1565                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1566                 else
1567                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1568                 break;
1569
1570         case EXT3_MOUNT_ORDERED_DATA:
1571         case EXT3_MOUNT_WRITEBACK_DATA:
1572                 if (!journal_check_available_features
1573                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1574                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1575                                "requested data journaling mode\n");
1576                         goto failed_mount3;
1577                 }
1578         default:
1579                 break;
1580         }
1581
1582         /*
1583          * The journal_load will have done any necessary log recovery,
1584          * so we can safely mount the rest of the filesystem now.
1585          */
1586
1587         root = iget(sb, EXT3_ROOT_INO);
1588         sb->s_root = d_alloc_root(root);
1589         if (!sb->s_root) {
1590                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1591                 iput(root);
1592                 goto failed_mount3;
1593         }
1594         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1595                 dput(sb->s_root);
1596                 sb->s_root = NULL;
1597                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1598                 goto failed_mount3;
1599         }
1600
1601         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1602         /*
1603          * akpm: core read_super() calls in here with the superblock locked.
1604          * That deadlocks, because orphan cleanup needs to lock the superblock
1605          * in numerous places.  Here we just pop the lock - it's relatively
1606          * harmless, because we are now ready to accept write_super() requests,
1607          * and aviro says that's the only reason for hanging onto the
1608          * superblock lock.
1609          */
1610         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1611         ext3_orphan_cleanup(sb, es);
1612         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1613         if (needs_recovery)
1614                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1615         ext3_mark_recovery_complete(sb, es);
1616         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1617                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1618                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1619                 "writeback");
1620
1621         percpu_counter_mod(&sbi->s_freeblocks_counter,
1622                 ext3_count_free_blocks(sb));
1623         percpu_counter_mod(&sbi->s_freeinodes_counter,
1624                 ext3_count_free_inodes(sb));
1625         percpu_counter_mod(&sbi->s_dirs_counter,
1626                 ext3_count_dirs(sb));
1627
1628         return 0;
1629
1630 cantfind_ext3:
1631         if (!silent)
1632                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1633                        sb->s_id);
1634         goto failed_mount;
1635
1636 failed_mount3:
1637         journal_destroy(sbi->s_journal);
1638 failed_mount2:
1639         for (i = 0; i < db_count; i++)
1640                 brelse(sbi->s_group_desc[i]);
1641         kfree(sbi->s_group_desc);
1642 failed_mount:
1643 #ifdef CONFIG_QUOTA
1644         for (i = 0; i < MAXQUOTAS; i++)
1645                 kfree(sbi->s_qf_names[i]);
1646 #endif
1647         ext3_blkdev_remove(sbi);
1648         brelse(bh);
1649 out_fail:
1650         sb->s_fs_info = NULL;
1651         kfree(sbi);
1652         return -EINVAL;
1653 }
1654
1655 /*
1656  * Setup any per-fs journal parameters now.  We'll do this both on
1657  * initial mount, once the journal has been initialised but before we've
1658  * done any recovery; and again on any subsequent remount.
1659  */
1660 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1661 {
1662         struct ext3_sb_info *sbi = EXT3_SB(sb);
1663
1664         if (sbi->s_commit_interval)
1665                 journal->j_commit_interval = sbi->s_commit_interval;
1666         /* We could also set up an ext3-specific default for the commit
1667          * interval here, but for now we'll just fall back to the jbd
1668          * default. */
1669
1670         spin_lock(&journal->j_state_lock);
1671         if (test_opt(sb, BARRIER))
1672                 journal->j_flags |= JFS_BARRIER;
1673         else
1674                 journal->j_flags &= ~JFS_BARRIER;
1675         spin_unlock(&journal->j_state_lock);
1676 }
1677
1678 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1679 {
1680         struct inode *journal_inode;
1681         journal_t *journal;
1682
1683         /* First, test for the existence of a valid inode on disk.  Bad
1684          * things happen if we iget() an unused inode, as the subsequent
1685          * iput() will try to delete it. */
1686
1687         journal_inode = iget(sb, journal_inum);
1688         if (!journal_inode) {
1689                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1690                 return NULL;
1691         }
1692         if (!journal_inode->i_nlink) {
1693                 make_bad_inode(journal_inode);
1694                 iput(journal_inode);
1695                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1696                 return NULL;
1697         }
1698
1699         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1700                   journal_inode, journal_inode->i_size);
1701         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1702                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1703                 iput(journal_inode);
1704                 return NULL;
1705         }
1706
1707         journal = journal_init_inode(journal_inode);
1708         if (!journal) {
1709                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1710                 iput(journal_inode);
1711                 return NULL;
1712         }
1713         journal->j_private = sb;
1714         ext3_init_journal_params(sb, journal);
1715         return journal;
1716 }
1717
1718 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1719                                        dev_t j_dev)
1720 {
1721         struct buffer_head * bh;
1722         journal_t *journal;
1723         int start;
1724         int len;
1725         int hblock, blocksize;
1726         unsigned long sb_block;
1727         unsigned long offset;
1728         struct ext3_super_block * es;
1729         struct block_device *bdev;
1730
1731         bdev = ext3_blkdev_get(j_dev);
1732         if (bdev == NULL)
1733                 return NULL;
1734
1735         if (bd_claim(bdev, sb)) {
1736                 printk(KERN_ERR
1737                         "EXT3: failed to claim external journal device.\n");
1738                 blkdev_put(bdev);
1739                 return NULL;
1740         }
1741
1742         blocksize = sb->s_blocksize;
1743         hblock = bdev_hardsect_size(bdev);
1744         if (blocksize < hblock) {
1745                 printk(KERN_ERR
1746                         "EXT3-fs: blocksize too small for journal device.\n");
1747                 goto out_bdev;
1748         }
1749
1750         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1751         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1752         set_blocksize(bdev, blocksize);
1753         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1754                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1755                        "external journal\n");
1756                 goto out_bdev;
1757         }
1758
1759         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1760         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1761             !(le32_to_cpu(es->s_feature_incompat) &
1762               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1763                 printk(KERN_ERR "EXT3-fs: external journal has "
1764                                         "bad superblock\n");
1765                 brelse(bh);
1766                 goto out_bdev;
1767         }
1768
1769         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1770                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1771                 brelse(bh);
1772                 goto out_bdev;
1773         }
1774
1775         len = le32_to_cpu(es->s_blocks_count);
1776         start = sb_block + 1;
1777         brelse(bh);     /* we're done with the superblock */
1778
1779         journal = journal_init_dev(bdev, sb->s_bdev,
1780                                         start, len, blocksize);
1781         if (!journal) {
1782                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1783                 goto out_bdev;
1784         }
1785         journal->j_private = sb;
1786         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1787         wait_on_buffer(journal->j_sb_buffer);
1788         if (!buffer_uptodate(journal->j_sb_buffer)) {
1789                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1790                 goto out_journal;
1791         }
1792         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1793                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1794                                         "user (unsupported) - %d\n",
1795                         be32_to_cpu(journal->j_superblock->s_nr_users));
1796                 goto out_journal;
1797         }
1798         EXT3_SB(sb)->journal_bdev = bdev;
1799         ext3_init_journal_params(sb, journal);
1800         return journal;
1801 out_journal:
1802         journal_destroy(journal);
1803 out_bdev:
1804         ext3_blkdev_put(bdev);
1805         return NULL;
1806 }
1807
1808 static int ext3_load_journal(struct super_block * sb,
1809                              struct ext3_super_block * es)
1810 {
1811         journal_t *journal;
1812         int journal_inum = le32_to_cpu(es->s_journal_inum);
1813         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1814         int err = 0;
1815         int really_read_only;
1816
1817         really_read_only = bdev_read_only(sb->s_bdev);
1818
1819         /*
1820          * Are we loading a blank journal or performing recovery after a
1821          * crash?  For recovery, we need to check in advance whether we
1822          * can get read-write access to the device.
1823          */
1824
1825         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1826                 if (sb->s_flags & MS_RDONLY) {
1827                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1828                                         "required on readonly filesystem.\n");
1829                         if (really_read_only) {
1830                                 printk(KERN_ERR "EXT3-fs: write access "
1831                                         "unavailable, cannot proceed.\n");
1832                                 return -EROFS;
1833                         }
1834                         printk (KERN_INFO "EXT3-fs: write access will "
1835                                         "be enabled during recovery.\n");
1836                 }
1837         }
1838
1839         if (journal_inum && journal_dev) {
1840                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1841                        "and inode journals!\n");
1842                 return -EINVAL;
1843         }
1844
1845         if (journal_inum) {
1846                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1847                         return -EINVAL;
1848         } else {
1849                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1850                         return -EINVAL;
1851         }
1852
1853         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1854                 err = journal_update_format(journal);
1855                 if (err)  {
1856                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1857                         journal_destroy(journal);
1858                         return err;
1859                 }
1860         }
1861
1862         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1863                 err = journal_wipe(journal, !really_read_only);
1864         if (!err)
1865                 err = journal_load(journal);
1866
1867         if (err) {
1868                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1869                 journal_destroy(journal);
1870                 return err;
1871         }
1872
1873         EXT3_SB(sb)->s_journal = journal;
1874         ext3_clear_journal_err(sb, es);
1875         return 0;
1876 }
1877
1878 static int ext3_create_journal(struct super_block * sb,
1879                                struct ext3_super_block * es,
1880                                int journal_inum)
1881 {
1882         journal_t *journal;
1883
1884         if (sb->s_flags & MS_RDONLY) {
1885                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1886                                 "create journal.\n");
1887                 return -EROFS;
1888         }
1889
1890         if (!(journal = ext3_get_journal(sb, journal_inum)))
1891                 return -EINVAL;
1892
1893         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1894                journal_inum);
1895
1896         if (journal_create(journal)) {
1897                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1898                 journal_destroy(journal);
1899                 return -EIO;
1900         }
1901
1902         EXT3_SB(sb)->s_journal = journal;
1903
1904         ext3_update_dynamic_rev(sb);
1905         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1906         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1907
1908         es->s_journal_inum = cpu_to_le32(journal_inum);
1909         sb->s_dirt = 1;
1910
1911         /* Make sure we flush the recovery flag to disk. */
1912         ext3_commit_super(sb, es, 1);
1913
1914         return 0;
1915 }
1916
1917 static void ext3_commit_super (struct super_block * sb,
1918                                struct ext3_super_block * es,
1919                                int sync)
1920 {
1921         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1922
1923         if (!sbh)
1924                 return;
1925         es->s_wtime = cpu_to_le32(get_seconds());
1926         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1927         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1928         BUFFER_TRACE(sbh, "marking dirty");
1929         mark_buffer_dirty(sbh);
1930         if (sync)
1931                 sync_dirty_buffer(sbh);
1932 }
1933
1934
1935 /*
1936  * Have we just finished recovery?  If so, and if we are mounting (or
1937  * remounting) the filesystem readonly, then we will end up with a
1938  * consistent fs on disk.  Record that fact.
1939  */
1940 static void ext3_mark_recovery_complete(struct super_block * sb,
1941                                         struct ext3_super_block * es)
1942 {
1943         journal_t *journal = EXT3_SB(sb)->s_journal;
1944
1945         journal_lock_updates(journal);
1946         journal_flush(journal);
1947         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1948             sb->s_flags & MS_RDONLY) {
1949                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1950                 sb->s_dirt = 0;
1951                 ext3_commit_super(sb, es, 1);
1952         }
1953         journal_unlock_updates(journal);
1954 }
1955
1956 /*
1957  * If we are mounting (or read-write remounting) a filesystem whose journal
1958  * has recorded an error from a previous lifetime, move that error to the
1959  * main filesystem now.
1960  */
1961 static void ext3_clear_journal_err(struct super_block * sb,
1962                                    struct ext3_super_block * es)
1963 {
1964         journal_t *journal;
1965         int j_errno;
1966         const char *errstr;
1967
1968         journal = EXT3_SB(sb)->s_journal;
1969
1970         /*
1971          * Now check for any error status which may have been recorded in the
1972          * journal by a prior ext3_error() or ext3_abort()
1973          */
1974
1975         j_errno = journal_errno(journal);
1976         if (j_errno) {
1977                 char nbuf[16];
1978
1979                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1980                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1981                              "from previous mount: %s", errstr);
1982                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1983                              "filesystem check.");
1984
1985                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1986                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1987                 ext3_commit_super (sb, es, 1);
1988
1989                 journal_clear_err(journal);
1990         }
1991 }
1992
1993 /*
1994  * Force the running and committing transactions to commit,
1995  * and wait on the commit.
1996  */
1997 int ext3_force_commit(struct super_block *sb)
1998 {
1999         journal_t *journal;
2000         int ret;
2001
2002         if (sb->s_flags & MS_RDONLY)
2003                 return 0;
2004
2005         journal = EXT3_SB(sb)->s_journal;
2006         sb->s_dirt = 0;
2007         ret = ext3_journal_force_commit(journal);
2008         return ret;
2009 }
2010
2011 /*
2012  * Ext3 always journals updates to the superblock itself, so we don't
2013  * have to propagate any other updates to the superblock on disk at this
2014  * point.  Just start an async writeback to get the buffers on their way
2015  * to the disk.
2016  *
2017  * This implicitly triggers the writebehind on sync().
2018  */
2019
2020 void ext3_write_super (struct super_block * sb)
2021 {
2022         if (down_trylock(&sb->s_lock) == 0)
2023                 BUG();
2024         sb->s_dirt = 0;
2025 }
2026
2027 static int ext3_sync_fs(struct super_block *sb, int wait)
2028 {
2029         tid_t target;
2030
2031         sb->s_dirt = 0;
2032         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2033                 if (wait)
2034                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2035         }
2036         return 0;
2037 }
2038
2039 /*
2040  * LVM calls this function before a (read-only) snapshot is created.  This
2041  * gives us a chance to flush the journal completely and mark the fs clean.
2042  */
2043 void ext3_write_super_lockfs(struct super_block *sb)
2044 {
2045         sb->s_dirt = 0;
2046
2047         if (!(sb->s_flags & MS_RDONLY)) {
2048                 journal_t *journal = EXT3_SB(sb)->s_journal;
2049
2050                 /* Now we set up the journal barrier. */
2051                 journal_lock_updates(journal);
2052                 journal_flush(journal);
2053
2054                 /* Journal blocked and flushed, clear needs_recovery flag. */
2055                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2056                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2057         }
2058 }
2059
2060 /*
2061  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2062  * flag here, even though the filesystem is not technically dirty yet.
2063  */
2064 void ext3_unlockfs(struct super_block *sb)
2065 {
2066         if (!(sb->s_flags & MS_RDONLY)) {
2067                 lock_super(sb);
2068                 /* Reser the needs_recovery flag before the fs is unlocked. */
2069                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2070                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2071                 unlock_super(sb);
2072                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2073         }
2074 }
2075
2076 int ext3_remount (struct super_block * sb, int * flags, char * data)
2077 {
2078         struct ext3_super_block * es;
2079         struct ext3_sb_info *sbi = EXT3_SB(sb);
2080         unsigned long tmp;
2081         unsigned long n_blocks_count = 0;
2082
2083         /*
2084          * Allow the "check" option to be passed as a remount option.
2085          */
2086         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2087                 return -EINVAL;
2088
2089         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2090                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2091
2092         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2093                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2094
2095         es = sbi->s_es;
2096
2097         ext3_init_journal_params(sb, sbi->s_journal);
2098
2099         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2100                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2101                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2102                         return -EROFS;
2103
2104                 if (*flags & MS_RDONLY) {
2105                         /*
2106                          * First of all, the unconditional stuff we have to do
2107                          * to disable replay of the journal when we next remount
2108                          */
2109                         sb->s_flags |= MS_RDONLY;
2110
2111                         /*
2112                          * OK, test if we are remounting a valid rw partition
2113                          * readonly, and if so set the rdonly flag and then
2114                          * mark the partition as valid again.
2115                          */
2116                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2117                             (sbi->s_mount_state & EXT3_VALID_FS))
2118                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2119
2120                         ext3_mark_recovery_complete(sb, es);
2121                 } else {
2122                         __le32 ret;
2123                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2124                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2125                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2126                                        "remount RDWR because of unsupported "
2127                                        "optional features (%x).\n",
2128                                        sb->s_id, le32_to_cpu(ret));
2129                                 return -EROFS;
2130                         }
2131                         /*
2132                          * Mounting a RDONLY partition read-write, so reread
2133                          * and store the current valid flag.  (It may have
2134                          * been changed by e2fsck since we originally mounted
2135                          * the partition.)
2136                          */
2137                         ext3_clear_journal_err(sb, es);
2138                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2139                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2140                                 return ret;
2141                         if (!ext3_setup_super (sb, es, 0))
2142                                 sb->s_flags &= ~MS_RDONLY;
2143                 }
2144         }
2145         return 0;
2146 }
2147
2148 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2149 {
2150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2151         unsigned long overhead;
2152         int i;
2153
2154         if (test_opt (sb, MINIX_DF))
2155                 overhead = 0;
2156         else {
2157                 unsigned long ngroups;
2158                 ngroups = EXT3_SB(sb)->s_groups_count;
2159                 smp_rmb();
2160
2161                 /*
2162                  * Compute the overhead (FS structures)
2163                  */
2164
2165                 /*
2166                  * All of the blocks before first_data_block are
2167                  * overhead
2168                  */
2169                 overhead = le32_to_cpu(es->s_first_data_block);
2170
2171                 /*
2172                  * Add the overhead attributed to the superblock and
2173                  * block group descriptors.  If the sparse superblocks
2174                  * feature is turned on, then not all groups have this.
2175                  */
2176                 for (i = 0; i < ngroups; i++)
2177                         overhead += ext3_bg_has_super(sb, i) +
2178                                 ext3_bg_num_gdb(sb, i);
2179
2180                 /*
2181                  * Every block group has an inode bitmap, a block
2182                  * bitmap, and an inode table.
2183                  */
2184                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2185         }
2186
2187         buf->f_type = EXT3_SUPER_MAGIC;
2188         buf->f_bsize = sb->s_blocksize;
2189         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2190         buf->f_bfree = ext3_count_free_blocks (sb);
2191         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2192         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2193                 buf->f_bavail = 0;
2194         buf->f_files = le32_to_cpu(es->s_inodes_count);
2195         buf->f_ffree = ext3_count_free_inodes (sb);
2196         buf->f_namelen = EXT3_NAME_LEN;
2197         return 0;
2198 }
2199
2200 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2201  * is locked for write. Otherwise the are possible deadlocks:
2202  * Process 1                         Process 2
2203  * ext3_create()                     quota_sync()
2204  *   journal_start()                   write_dquot()
2205  *   DQUOT_INIT()                        down(dqio_sem)
2206  *     down(dqio_sem)                    journal_start()
2207  *
2208  */
2209
2210 #ifdef CONFIG_QUOTA
2211
2212 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2213 {
2214         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2215 }
2216
2217 static int ext3_dquot_initialize(struct inode *inode, int type)
2218 {
2219         handle_t *handle;
2220         int ret, err;
2221
2222         /* We may create quota structure so we need to reserve enough blocks */
2223         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2224         if (IS_ERR(handle))
2225                 return PTR_ERR(handle);
2226         ret = dquot_initialize(inode, type);
2227         err = ext3_journal_stop(handle);
2228         if (!ret)
2229                 ret = err;
2230         return ret;
2231 }
2232
2233 static int ext3_dquot_drop(struct inode *inode)
2234 {
2235         handle_t *handle;
2236         int ret, err;
2237
2238         /* We may delete quota structure so we need to reserve enough blocks */
2239         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2240         if (IS_ERR(handle))
2241                 return PTR_ERR(handle);
2242         ret = dquot_drop(inode);
2243         err = ext3_journal_stop(handle);
2244         if (!ret)
2245                 ret = err;
2246         return ret;
2247 }
2248
2249 static int ext3_write_dquot(struct dquot *dquot)
2250 {
2251         int ret, err;
2252         handle_t *handle;
2253
2254         handle = ext3_journal_start(dquot_to_inode(dquot),
2255                                         EXT3_QUOTA_TRANS_BLOCKS);
2256         if (IS_ERR(handle))
2257                 return PTR_ERR(handle);
2258         ret = dquot_commit(dquot);
2259         err = ext3_journal_stop(handle);
2260         if (!ret)
2261                 ret = err;
2262         return ret;
2263 }
2264
2265 static int ext3_acquire_dquot(struct dquot *dquot)
2266 {
2267         int ret, err;
2268         handle_t *handle;
2269
2270         handle = ext3_journal_start(dquot_to_inode(dquot),
2271                                         EXT3_QUOTA_INIT_BLOCKS);
2272         if (IS_ERR(handle))
2273                 return PTR_ERR(handle);
2274         ret = dquot_acquire(dquot);
2275         err = ext3_journal_stop(handle);
2276         if (!ret)
2277                 ret = err;
2278         return ret;
2279 }
2280
2281 static int ext3_release_dquot(struct dquot *dquot)
2282 {
2283         int ret, err;
2284         handle_t *handle;
2285
2286         handle = ext3_journal_start(dquot_to_inode(dquot),
2287                                         EXT3_QUOTA_INIT_BLOCKS);
2288         if (IS_ERR(handle))
2289                 return PTR_ERR(handle);
2290         ret = dquot_release(dquot);
2291         err = ext3_journal_stop(handle);
2292         if (!ret)
2293                 ret = err;
2294         return ret;
2295 }
2296
2297 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2298 {
2299         /* Are we journalling quotas? */
2300         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2301             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2302                 dquot_mark_dquot_dirty(dquot);
2303                 return ext3_write_dquot(dquot);
2304         } else {
2305                 return dquot_mark_dquot_dirty(dquot);
2306         }
2307 }
2308
2309 static int ext3_write_info(struct super_block *sb, int type)
2310 {
2311         int ret, err;
2312         handle_t *handle;
2313
2314         /* Data block + inode block */
2315         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2316         if (IS_ERR(handle))
2317                 return PTR_ERR(handle);
2318         ret = dquot_commit_info(sb, type);
2319         err = ext3_journal_stop(handle);
2320         if (!ret)
2321                 ret = err;
2322         return ret;
2323 }
2324
2325 /*
2326  * Turn on quotas during mount time - we need to find
2327  * the quota file and such...
2328  */
2329 static int ext3_quota_on_mount(struct super_block *sb, int type)
2330 {
2331         int err;
2332         struct dentry *dentry;
2333         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2334                              .hash = 0,
2335                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2336
2337         dentry = lookup_hash(&name, sb->s_root);
2338         if (IS_ERR(dentry))
2339                 return PTR_ERR(dentry);
2340         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2341         if (err)
2342                 dput(dentry);
2343         /* We keep the dentry reference if everything went ok - we drop it
2344          * on quota_off time */
2345         return err;
2346 }
2347
2348 /* Turn quotas off during mount time */
2349 static int ext3_quota_off_mount(struct super_block *sb, int type)
2350 {
2351         int err;
2352         struct dentry *dentry;
2353
2354         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2355         err = vfs_quota_off_mount(sb, type);
2356         /* We invalidate dentry - it has at least wrong hash... */
2357         d_invalidate(dentry);
2358         dput(dentry);
2359         return err;
2360 }
2361
2362 /*
2363  * Standard function to be called on quota_on
2364  */
2365 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2366                          char *path)
2367 {
2368         int err;
2369         struct nameidata nd;
2370
2371         /* Not journalling quota? */
2372         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2373             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2374                 return vfs_quota_on(sb, type, format_id, path);
2375         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2376         if (err)
2377                 return err;
2378         /* Quotafile not on the same filesystem? */
2379         if (nd.mnt->mnt_sb != sb)
2380                 return -EXDEV;
2381         /* Quotafile not of fs root? */
2382         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2383                 printk(KERN_WARNING
2384                         "EXT3-fs: Quota file not on filesystem root. "
2385                         "Journalled quota will not work.\n");
2386         if (!ext3_should_journal_data(nd.dentry->d_inode))
2387                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2388                         "data-journalling. Journalled quota will not work.\n");
2389         path_release(&nd);
2390         return vfs_quota_on(sb, type, format_id, path);
2391 }
2392
2393 #endif
2394
2395 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2396         int flags, const char *dev_name, void *data)
2397 {
2398         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2399 }
2400
2401 static struct file_system_type ext3_fs_type = {
2402         .owner          = THIS_MODULE,
2403         .name           = "ext3",
2404         .get_sb         = ext3_get_sb,
2405         .kill_sb        = kill_block_super,
2406         .fs_flags       = FS_REQUIRES_DEV,
2407 };
2408
2409 static int __init init_ext3_fs(void)
2410 {
2411         int err = init_ext3_xattr();
2412         if (err)
2413                 return err;
2414         err = init_inodecache();
2415         if (err)
2416                 goto out1;
2417         err = register_filesystem(&ext3_fs_type);
2418         if (err)
2419                 goto out;
2420         return 0;
2421 out:
2422         destroy_inodecache();
2423 out1:
2424         exit_ext3_xattr();
2425         return err;
2426 }
2427
2428 static void __exit exit_ext3_fs(void)
2429 {
2430         unregister_filesystem(&ext3_fs_type);
2431         destroy_inodecache();
2432         exit_ext3_xattr();
2433 }
2434
2435 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2436 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2437 MODULE_LICENSE("GPL");
2438 module_init(init_ext3_fs)
2439 module_exit(exit_ext3_fs)