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 static const char *ext3_decode_error(struct super_block * sb, int errno,
  54                                      char nbuf[16]);
  55 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  56 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  57 static void ext3_unlockfs(struct super_block *sb);
  58 static void ext3_write_super (struct super_block * sb);
  59 static void ext3_write_super_lockfs(struct super_block *sb);
  60
  61 /*
  62  * Wrappers for journal_start/end.
  63  *
  64  * The only special thing we need to do here is to make sure that all
  65  * journal_end calls result in the superblock being marked dirty, so
  66  * that sync() will call the filesystem's write_super callback if
  67  * appropriate.
  68  */
  69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  70 {
  71         journal_t *journal;
  72
  73         if (sb->s_flags & MS_RDONLY)
  74                 return ERR_PTR(-EROFS);
  75
  76         /* Special case here: if the journal has aborted behind our
  77          * backs (eg. EIO in the commit thread), then we still need to
  78          * take the FS itself readonly cleanly. */
  79         journal = EXT3_SB(sb)->s_journal;
  80         if (is_journal_aborted(journal)) {
  81                 ext3_abort(sb, __FUNCTION__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 /*
  90  * The only special thing we need to do here is to make sure that all
  91  * journal_stop calls result in the superblock being marked dirty, so
  92  * that sync() will call the filesystem's write_super callback if
  93  * appropriate.
  94  */
  95 int __ext3_journal_stop(const char *where, handle_t *handle)
  96 {
  97         struct super_block *sb;
  98         int err;
  99         int rc;
 100
 101         sb = handle->h_transaction->t_journal->j_private;
 102         err = handle->h_err;
 103         rc = journal_stop(handle);
 104
 105         if (!err)
 106                 err = rc;
 107         if (err)
 108                 __ext3_std_error(sb, where, err);
 109         return err;
 110 }
 111
 112 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 113                 struct buffer_head *bh, handle_t *handle, int err)
 114 {
 115         char nbuf[16];
 116         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 117
 118         if (bh)
 119                 BUFFER_TRACE(bh, "abort");
 120
 121         if (!handle->h_err)
 122                 handle->h_err = err;
 123
 124         if (is_handle_aborted(handle))
 125                 return;
 126
 127         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 128                caller, errstr, err_fn);
 129
 130         journal_abort_handle(handle);
 131 }
 132
 133 /* Deal with the reporting of failure conditions on a filesystem such as
 134  * inconsistencies detected or read IO failures.
 135  *
 136  * On ext2, we can store the error state of the filesystem in the
 137  * superblock.  That is not possible on ext3, because we may have other
 138  * write ordering constraints on the superblock which prevent us from
 139  * writing it out straight away; and given that the journal is about to
 140  * be aborted, we can't rely on the current, or future, transactions to
 141  * write out the superblock safely.
 142  *
 143  * We'll just use the journal_abort() error code to record an error in
 144  * the journal instead.  On recovery, the journal will compain about
 145  * that error until we've noted it down and cleared it.
 146  */
 147
 148 static void ext3_handle_error(struct super_block *sb)
 149 {
 150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 151
 152         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 153         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 154
 155         if (sb->s_flags & MS_RDONLY)
 156                 return;
 157
 158         if (test_opt (sb, ERRORS_RO)) {
 159                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 160                 sb->s_flags |= MS_RDONLY;
 161         } else {
 162                 journal_t *journal = EXT3_SB(sb)->s_journal;
 163
 164                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 165                 if (journal)
 166                         journal_abort(journal, -EIO);
 167         }
 168         if (test_opt(sb, ERRORS_PANIC))
 169                 panic("EXT3-fs (device %s): panic forced after error\n",
 170                         sb->s_id);
 171         ext3_commit_super(sb, es, 1);
 172 }
 173
 174 void ext3_error (struct super_block * sb, const char * function,
 175                  const char * fmt, ...)
 176 {
 177         va_list args;
 178
 179         va_start(args, fmt);
 180         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 181         vprintk(fmt, args);
 182         printk("\n");
 183         va_end(args);
 184
 185         ext3_handle_error(sb);
 186 }
 187
 188 static const char *ext3_decode_error(struct super_block * sb, int errno,
 189                                      char nbuf[16])
 190 {
 191         char *errstr = NULL;
 192
 193         switch (errno) {
 194         case -EIO:
 195                 errstr = "IO failure";
 196                 break;
 197         case -ENOMEM:
 198                 errstr = "Out of memory";
 199                 break;
 200         case -EROFS:
 201                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 202                         errstr = "Journal has aborted";
 203                 else
 204                         errstr = "Readonly filesystem";
 205                 break;
 206         default:
 207                 /* If the caller passed in an extra buffer for unknown
 208                  * errors, textualise them now.  Else we just return
 209                  * NULL. */
 210                 if (nbuf) {
 211                         /* Check for truncated error codes... */
 212                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 213                                 errstr = nbuf;
 214                 }
 215                 break;
 216         }
 217
 218         return errstr;
 219 }
 220
 221 /* __ext3_std_error decodes expected errors from journaling functions
 222  * automatically and invokes the appropriate error response.  */
 223
 224 void __ext3_std_error (struct super_block * sb, const char * function,
 225                        int errno)
 226 {
 227         char nbuf[16];
 228         const char *errstr;
 229
 230         /* Special case: if the error is EROFS, and we're not already
 231          * inside a transaction, then there's really no point in logging
 232          * an error. */
 233         if (errno == -EROFS && journal_current_handle() == NULL &&
 234             (sb->s_flags & MS_RDONLY))
 235                 return;
 236
 237         errstr = ext3_decode_error(sb, errno, nbuf);
 238         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 239                 sb->s_id, function, errstr);
 240
 241         ext3_handle_error(sb);
 242 }
 243
 244 /*
 245  * ext3_abort is a much stronger failure handler than ext3_error.  The
 246  * abort function may be used to deal with unrecoverable failures such
 247  * as journal IO errors or ENOMEM at a critical moment in log management.
 248  *
 249  * We unconditionally force the filesystem into an ABORT|READONLY state,
 250  * unless the error response on the fs has been set to panic in which
 251  * case we take the easy way out and panic immediately.
 252  */
 253
 254 void ext3_abort (struct super_block * sb, const char * function,
 255                  const char * fmt, ...)
 256 {
 257         va_list args;
 258
 259         printk (KERN_CRIT "ext3_abort called.\n");
 260
 261         va_start(args, fmt);
 262         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 263         vprintk(fmt, args);
 264         printk("\n");
 265         va_end(args);
 266
 267         if (test_opt(sb, ERRORS_PANIC))
 268                 panic("EXT3-fs panic from previous error\n");
 269
 270         if (sb->s_flags & MS_RDONLY)
 271                 return;
 272
 273         printk(KERN_CRIT "Remounting filesystem read-only\n");
 274         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 275         sb->s_flags |= MS_RDONLY;
 276         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 277         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 278 }
 279
 280 void ext3_warning (struct super_block * sb, const char * function,
 281                    const char * fmt, ...)
 282 {
 283         va_list args;
 284
 285         va_start(args, fmt);
 286         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 287                sb->s_id, function);
 288         vprintk(fmt, args);
 289         printk("\n");
 290         va_end(args);
 291 }
 292
 293 void ext3_update_dynamic_rev(struct super_block *sb)
 294 {
 295         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 296
 297         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 298                 return;
 299
 300         ext3_warning(sb, __FUNCTION__,
 301                      "updating to rev %d because of new feature flag, "
 302                      "running e2fsck is recommended",
 303                      EXT3_DYNAMIC_REV);
 304
 305         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 306         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 307         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 308         /* leave es->s_feature_*compat flags alone */
 309         /* es->s_uuid will be set by e2fsck if empty */
 310
 311         /*
 312          * The rest of the superblock fields should be zero, and if not it
 313          * means they are likely already in use, so leave them alone.  We
 314          * can leave it up to e2fsck to clean up any inconsistencies there.
 315          */
 316 }
 317
 318 /*
 319  * Open the external journal device
 320  */
 321 static struct block_device *ext3_blkdev_get(dev_t dev)
 322 {
 323         struct block_device *bdev;
 324         char b[BDEVNAME_SIZE];
 325
 326         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 327         if (IS_ERR(bdev))
 328                 goto fail;
 329         return bdev;
 330
 331 fail:
 332         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 333                         __bdevname(dev, b), PTR_ERR(bdev));
 334         return NULL;
 335 }
 336
 337 /*
 338  * Release the journal device
 339  */
 340 static int ext3_blkdev_put(struct block_device *bdev)
 341 {
 342         bd_release(bdev);
 343         return blkdev_put(bdev);
 344 }
 345
 346 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 347 {
 348         struct block_device *bdev;
 349         int ret = -ENODEV;
 350
 351         bdev = sbi->journal_bdev;
 352         if (bdev) {
 353                 ret = ext3_blkdev_put(bdev);
 354                 sbi->journal_bdev = NULL;
 355         }
 356         return ret;
 357 }
 358
 359 static inline struct inode *orphan_list_entry(struct list_head *l)
 360 {
 361         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 362 }
 363
 364 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 365 {
 366         struct list_head *l;
 367
 368         printk(KERN_ERR "sb orphan head is %d\n",
 369                le32_to_cpu(sbi->s_es->s_last_orphan));
 370
 371         printk(KERN_ERR "sb_info orphan list:\n");
 372         list_for_each(l, &sbi->s_orphan) {
 373                 struct inode *inode = orphan_list_entry(l);
 374                 printk(KERN_ERR "  "
 375                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 376                        inode->i_sb->s_id, inode->i_ino, inode,
 377                        inode->i_mode, inode->i_nlink,
 378                        NEXT_ORPHAN(inode));
 379         }
 380 }
 381
 382 static void ext3_put_super (struct super_block * sb)
 383 {
 384         struct ext3_sb_info *sbi = EXT3_SB(sb);
 385         struct ext3_super_block *es = sbi->s_es;
 386         int i;
 387
 388         ext3_xattr_put_super(sb);
 389         journal_destroy(sbi->s_journal);
 390         if (!(sb->s_flags & MS_RDONLY)) {
 391                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 392                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 393                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 394                 mark_buffer_dirty(sbi->s_sbh);
 395                 ext3_commit_super(sb, es, 1);
 396         }
 397
 398         for (i = 0; i < sbi->s_gdb_count; i++)
 399                 brelse(sbi->s_group_desc[i]);
 400         kfree(sbi->s_group_desc);
 401         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 402         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 403         percpu_counter_destroy(&sbi->s_dirs_counter);
 404         brelse(sbi->s_sbh);
 405 #ifdef CONFIG_QUOTA
 406         for (i = 0; i < MAXQUOTAS; i++)
 407                 kfree(sbi->s_qf_names[i]);
 408 #endif
 409
 410         /* Debugging code just in case the in-memory inode orphan list
 411          * isn't empty.  The on-disk one can be non-empty if we've
 412          * detected an error and taken the fs readonly, but the
 413          * in-memory list had better be clean by this point. */
 414         if (!list_empty(&sbi->s_orphan))
 415                 dump_orphan_list(sb, sbi);
 416         J_ASSERT(list_empty(&sbi->s_orphan));
 417
 418         invalidate_bdev(sb->s_bdev, 0);
 419         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 420                 /*
 421                  * Invalidate the journal device's buffers.  We don't want them
 422                  * floating about in memory - the physical journal device may
 423                  * hotswapped, and it breaks the `ro-after' testing code.
 424                  */
 425                 sync_blockdev(sbi->journal_bdev);
 426                 invalidate_bdev(sbi->journal_bdev, 0);
 427                 ext3_blkdev_remove(sbi);
 428         }
 429         sb->s_fs_info = NULL;
 430         kfree(sbi);
 431         return;
 432 }
 433
 434 static kmem_cache_t *ext3_inode_cachep;
 435
 436 /*
 437  * Called inside transaction, so use GFP_NOFS
 438  */
 439 static struct inode *ext3_alloc_inode(struct super_block *sb)
 440 {
 441         struct ext3_inode_info *ei;
 442
 443         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 444         if (!ei)
 445                 return NULL;
 446 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 447         ei->i_acl = EXT3_ACL_NOT_CACHED;
 448         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 449 #endif
 450         ei->i_block_alloc_info = NULL;
 451         ei->vfs_inode.i_version = 1;
 452         return &ei->vfs_inode;
 453 }
 454
 455 static void ext3_destroy_inode(struct inode *inode)
 456 {
 457         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 458 }
 459
 460 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 461 {
 462         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 463
 464         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 465             SLAB_CTOR_CONSTRUCTOR) {
 466                 INIT_LIST_HEAD(&ei->i_orphan);
 467 #ifdef CONFIG_EXT3_FS_XATTR
 468                 init_rwsem(&ei->xattr_sem);
 469 #endif
 470                 init_MUTEX(&ei->truncate_sem);
 471                 inode_init_once(&ei->vfs_inode);
 472         }
 473 }
 474
 475 static int init_inodecache(void)
 476 {
 477         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 478                                              sizeof(struct ext3_inode_info),
 479                                              0, SLAB_RECLAIM_ACCOUNT,
 480                                              init_once, NULL);
 481         if (ext3_inode_cachep == NULL)
 482                 return -ENOMEM;
 483         return 0;
 484 }
 485
 486 static void destroy_inodecache(void)
 487 {
 488         if (kmem_cache_destroy(ext3_inode_cachep))
 489                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 490 }
 491
 492 static void ext3_clear_inode(struct inode *inode)
 493 {
 494         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 495 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 496        if (EXT3_I(inode)->i_acl &&
 497            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 498                posix_acl_release(EXT3_I(inode)->i_acl);
 499                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 500        }
 501        if (EXT3_I(inode)->i_default_acl &&
 502            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 503                posix_acl_release(EXT3_I(inode)->i_default_acl);
 504                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 505        }
 506 #endif
 507         ext3_discard_reservation(inode);
 508         EXT3_I(inode)->i_block_alloc_info = NULL;
 509         kfree(rsv);
 510 }
 511
 512 #ifdef CONFIG_QUOTA
 513
 514 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 515 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 516
 517 static int ext3_dquot_initialize(struct inode *inode, int type);
 518 static int ext3_dquot_drop(struct inode *inode);
 519 static int ext3_write_dquot(struct dquot *dquot);
 520 static int ext3_acquire_dquot(struct dquot *dquot);
 521 static int ext3_release_dquot(struct dquot *dquot);
 522 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 523 static int ext3_write_info(struct super_block *sb, int type);
 524 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 525 static int ext3_quota_on_mount(struct super_block *sb, int type);
 526 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 527                                size_t len, loff_t off);
 528 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 529                                 const char *data, size_t len, loff_t off);
 530
 531 static struct dquot_operations ext3_quota_operations = {
 532         .initialize     = ext3_dquot_initialize,
 533         .drop           = ext3_dquot_drop,
 534         .alloc_space    = dquot_alloc_space,
 535         .alloc_inode    = dquot_alloc_inode,
 536         .free_space     = dquot_free_space,
 537         .free_inode     = dquot_free_inode,
 538         .transfer       = dquot_transfer,
 539         .write_dquot    = ext3_write_dquot,
 540         .acquire_dquot  = ext3_acquire_dquot,
 541         .release_dquot  = ext3_release_dquot,
 542         .mark_dirty     = ext3_mark_dquot_dirty,
 543         .write_info     = ext3_write_info
 544 };
 545
 546 static struct quotactl_ops ext3_qctl_operations = {
 547         .quota_on       = ext3_quota_on,
 548         .quota_off      = vfs_quota_off,
 549         .quota_sync     = vfs_quota_sync,
 550         .get_info       = vfs_get_dqinfo,
 551         .set_info       = vfs_set_dqinfo,
 552         .get_dqblk      = vfs_get_dqblk,
 553         .set_dqblk      = vfs_set_dqblk
 554 };
 555 #endif
 556
 557 static struct super_operations ext3_sops = {
 558         .alloc_inode    = ext3_alloc_inode,
 559         .destroy_inode  = ext3_destroy_inode,
 560         .read_inode     = ext3_read_inode,
 561         .write_inode    = ext3_write_inode,
 562         .dirty_inode    = ext3_dirty_inode,
 563         .delete_inode   = ext3_delete_inode,
 564         .put_super      = ext3_put_super,
 565         .write_super    = ext3_write_super,
 566         .sync_fs        = ext3_sync_fs,
 567         .write_super_lockfs = ext3_write_super_lockfs,
 568         .unlockfs       = ext3_unlockfs,
 569         .statfs         = ext3_statfs,
 570         .remount_fs     = ext3_remount,
 571         .clear_inode    = ext3_clear_inode,
 572 #ifdef CONFIG_QUOTA
 573         .quota_read     = ext3_quota_read,
 574         .quota_write    = ext3_quota_write,
 575 #endif
 576 };
 577
 578 struct dentry *ext3_get_parent(struct dentry *child);
 579 static struct export_operations ext3_export_ops = {
 580         .get_parent = ext3_get_parent,
 581 };
 582
 583 enum {
 584         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 585         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 586         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 587         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 588         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 589         Opt_commit, Opt_journal_update, Opt_journal_inum,
 590         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 591         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 592         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 593         Opt_tagxid, Opt_barrier, Opt_ignore, Opt_err, Opt_resize,
 594 };
 595
 596 static match_table_t tokens = {
 597         {Opt_bsd_df, "bsddf"},
 598         {Opt_minix_df, "minixdf"},
 599         {Opt_grpid, "grpid"},
 600         {Opt_grpid, "bsdgroups"},
 601         {Opt_nogrpid, "nogrpid"},
 602         {Opt_nogrpid, "sysvgroups"},
 603         {Opt_resgid, "resgid=%u"},
 604         {Opt_resuid, "resuid=%u"},
 605         {Opt_sb, "sb=%u"},
 606         {Opt_err_cont, "errors=continue"},
 607         {Opt_err_panic, "errors=panic"},
 608         {Opt_err_ro, "errors=remount-ro"},
 609         {Opt_nouid32, "nouid32"},
 610         {Opt_nocheck, "nocheck"},
 611         {Opt_nocheck, "check=none"},
 612         {Opt_check, "check"},
 613         {Opt_debug, "debug"},
 614         {Opt_oldalloc, "oldalloc"},
 615         {Opt_orlov, "orlov"},
 616         {Opt_user_xattr, "user_xattr"},
 617         {Opt_nouser_xattr, "nouser_xattr"},
 618         {Opt_acl, "acl"},
 619         {Opt_noacl, "noacl"},
 620         {Opt_reservation, "reservation"},
 621         {Opt_noreservation, "noreservation"},
 622         {Opt_noload, "noload"},
 623         {Opt_nobh, "nobh"},
 624         {Opt_commit, "commit=%u"},
 625         {Opt_journal_update, "journal=update"},
 626         {Opt_journal_inum, "journal=%u"},
 627         {Opt_abort, "abort"},
 628         {Opt_data_journal, "data=journal"},
 629         {Opt_data_ordered, "data=ordered"},
 630         {Opt_data_writeback, "data=writeback"},
 631         {Opt_offusrjquota, "usrjquota="},
 632         {Opt_usrjquota, "usrjquota=%s"},
 633         {Opt_offgrpjquota, "grpjquota="},
 634         {Opt_grpjquota, "grpjquota=%s"},
 635         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 636         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 637         {Opt_tagxid, "tagxid"},
 638         {Opt_ignore, "grpquota"},
 639         {Opt_ignore, "noquota"},
 640         {Opt_ignore, "quota"},
 641         {Opt_ignore, "usrquota"},
 642         {Opt_barrier, "barrier=%u"},
 643         {Opt_err, NULL},
 644         {Opt_resize, "resize"},
 645 };
 646
 647 static unsigned long get_sb_block(void **data)
 648 {
 649         unsigned long   sb_block;
 650         char            *options = (char *) *data;
 651
 652         if (!options || strncmp(options, "sb=", 3) != 0)
 653                 return 1;       /* Default location */
 654         options += 3;
 655         sb_block = simple_strtoul(options, &options, 0);
 656         if (*options && *options != ',') {
 657                 printk("EXT3-fs: Invalid sb specification: %s\n",
 658                        (char *) *data);
 659                 return 1;
 660         }
 661         if (*options == ',')
 662                 options++;
 663         *data = (void *) options;
 664         return sb_block;
 665 }
 666
 667 static int parse_options (char * options, struct super_block *sb,
 668                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 669 {
 670         struct ext3_sb_info *sbi = EXT3_SB(sb);
 671         char * p;
 672         substring_t args[MAX_OPT_ARGS];
 673         int data_opt = 0;
 674         int option;
 675 #ifdef CONFIG_QUOTA
 676         int qtype;
 677         char *qname;
 678 #endif
 679
 680         if (!options)
 681                 return 1;
 682
 683         while ((p = strsep (&options, ",")) != NULL) {
 684                 int token;
 685                 if (!*p)
 686                         continue;
 687
 688                 token = match_token(p, tokens, args);
 689                 switch (token) {
 690                 case Opt_bsd_df:
 691                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 692                         break;
 693                 case Opt_minix_df:
 694                         set_opt (sbi->s_mount_opt, MINIX_DF);
 695                         break;
 696                 case Opt_grpid:
 697                         set_opt (sbi->s_mount_opt, GRPID);
 698                         break;
 699                 case Opt_nogrpid:
 700                         clear_opt (sbi->s_mount_opt, GRPID);
 701                         break;
 702                 case Opt_resuid:
 703                         if (match_int(&args[0], &option))
 704                                 return 0;
 705                         sbi->s_resuid = option;
 706                         break;
 707                 case Opt_resgid:
 708                         if (match_int(&args[0], &option))
 709                                 return 0;
 710                         sbi->s_resgid = option;
 711                         break;
 712                 case Opt_sb:
 713                         /* handled by get_sb_block() instead of here */
 714                         /* *sb_block = match_int(&args[0]); */
 715                         break;
 716                 case Opt_err_panic:
 717                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 718                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 719                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 720                         break;
 721                 case Opt_err_ro:
 722                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 723                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 724                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 725                         break;
 726                 case Opt_err_cont:
 727                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 728                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 729                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 730                         break;
 731                 case Opt_nouid32:
 732                         set_opt (sbi->s_mount_opt, NO_UID32);
 733                         break;
 734 #ifndef CONFIG_INOXID_NONE
 735                 case Opt_tagxid:
 736                         if (is_remount) {
 737                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 738                                        "tagxid on remount\n");
 739                                 return 0;
 740                         }
 741                         set_opt (sbi->s_mount_opt, TAG_XID);
 742                         break;
 743 #endif
 744                 case Opt_check:
 745 #ifdef CONFIG_EXT3_CHECK
 746                         set_opt (sbi->s_mount_opt, CHECK);
 747 #else
 748                         printk(KERN_ERR
 749                                "EXT3 Check option not supported\n");
 750 #endif
 751                         break;
 752                 case Opt_nocheck:
 753                         clear_opt (sbi->s_mount_opt, CHECK);
 754                         break;
 755                 case Opt_debug:
 756                         set_opt (sbi->s_mount_opt, DEBUG);
 757                         break;
 758                 case Opt_oldalloc:
 759                         set_opt (sbi->s_mount_opt, OLDALLOC);
 760                         break;
 761                 case Opt_orlov:
 762                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 763                         break;
 764 #ifdef CONFIG_EXT3_FS_XATTR
 765                 case Opt_user_xattr:
 766                         set_opt (sbi->s_mount_opt, XATTR_USER);
 767                         break;
 768                 case Opt_nouser_xattr:
 769                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 770                         break;
 771 #else
 772                 case Opt_user_xattr:
 773                 case Opt_nouser_xattr:
 774                         printk("EXT3 (no)user_xattr options not supported\n");
 775                         break;
 776 #endif
 777 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 778                 case Opt_acl:
 779                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 780                         break;
 781                 case Opt_noacl:
 782                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 783                         break;
 784 #else
 785                 case Opt_acl:
 786                 case Opt_noacl:
 787                         printk("EXT3 (no)acl options not supported\n");
 788                         break;
 789 #endif
 790                 case Opt_reservation:
 791                         set_opt(sbi->s_mount_opt, RESERVATION);
 792                         break;
 793                 case Opt_noreservation:
 794                         clear_opt(sbi->s_mount_opt, RESERVATION);
 795                         break;
 796                 case Opt_journal_update:
 797                         /* @@@ FIXME */
 798                         /* Eventually we will want to be able to create
 799                            a journal file here.  For now, only allow the
 800                            user to specify an existing inode to be the
 801                            journal file. */
 802                         if (is_remount) {
 803                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 804                                        "journal on remount\n");
 805                                 return 0;
 806                         }
 807                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 808                         break;
 809                 case Opt_journal_inum:
 810                         if (is_remount) {
 811                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 812                                        "journal on remount\n");
 813                                 return 0;
 814                         }
 815                         if (match_int(&args[0], &option))
 816                                 return 0;
 817                         *inum = option;
 818                         break;
 819                 case Opt_noload:
 820                         set_opt (sbi->s_mount_opt, NOLOAD);
 821                         break;
 822                 case Opt_commit:
 823                         if (match_int(&args[0], &option))
 824                                 return 0;
 825                         if (option < 0)
 826                                 return 0;
 827                         if (option == 0)
 828                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 829                         sbi->s_commit_interval = HZ * option;
 830                         break;
 831                 case Opt_data_journal:
 832                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 833                         goto datacheck;
 834                 case Opt_data_ordered:
 835                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 836                         goto datacheck;
 837                 case Opt_data_writeback:
 838                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 839                 datacheck:
 840                         if (is_remount) {
 841                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 842                                                 != data_opt) {
 843                                         printk(KERN_ERR
 844                                                 "EXT3-fs: cannot change data "
 845                                                 "mode on remount\n");
 846                                         return 0;
 847                                 }
 848                         } else {
 849                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 850                                 sbi->s_mount_opt |= data_opt;
 851                         }
 852                         break;
 853 #ifdef CONFIG_QUOTA
 854                 case Opt_usrjquota:
 855                         qtype = USRQUOTA;
 856                         goto set_qf_name;
 857                 case Opt_grpjquota:
 858                         qtype = GRPQUOTA;
 859 set_qf_name:
 860                         if (sb_any_quota_enabled(sb)) {
 861                                 printk(KERN_ERR
 862                                         "EXT3-fs: Cannot change journalled "
 863                                         "quota options when quota turned on.\n");
 864                                 return 0;
 865                         }
 866                         qname = match_strdup(&args[0]);
 867                         if (!qname) {
 868                                 printk(KERN_ERR
 869                                         "EXT3-fs: not enough memory for "
 870                                         "storing quotafile name.\n");
 871                                 return 0;
 872                         }
 873                         if (sbi->s_qf_names[qtype] &&
 874                             strcmp(sbi->s_qf_names[qtype], qname)) {
 875                                 printk(KERN_ERR
 876                                         "EXT3-fs: %s quota file already "
 877                                         "specified.\n", QTYPE2NAME(qtype));
 878                                 kfree(qname);
 879                                 return 0;
 880                         }
 881                         sbi->s_qf_names[qtype] = qname;
 882                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 883                                 printk(KERN_ERR
 884                                         "EXT3-fs: quotafile must be on "
 885                                         "filesystem root.\n");
 886                                 kfree(sbi->s_qf_names[qtype]);
 887                                 sbi->s_qf_names[qtype] = NULL;
 888                                 return 0;
 889                         }
 890                         break;
 891                 case Opt_offusrjquota:
 892                         qtype = USRQUOTA;
 893                         goto clear_qf_name;
 894                 case Opt_offgrpjquota:
 895                         qtype = GRPQUOTA;
 896 clear_qf_name:
 897                         if (sb_any_quota_enabled(sb)) {
 898                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 899                                         "journalled quota options when "
 900                                         "quota turned on.\n");
 901                                 return 0;
 902                         }
 903                         kfree(sbi->s_qf_names[qtype]);
 904                         sbi->s_qf_names[qtype] = NULL;
 905                         break;
 906                 case Opt_jqfmt_vfsold:
 907                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 908                         break;
 909                 case Opt_jqfmt_vfsv0:
 910                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 911                         break;
 912 #else
 913                 case Opt_usrjquota:
 914                 case Opt_grpjquota:
 915                 case Opt_offusrjquota:
 916                 case Opt_offgrpjquota:
 917                 case Opt_jqfmt_vfsold:
 918                 case Opt_jqfmt_vfsv0:
 919                         printk(KERN_ERR
 920                                 "EXT3-fs: journalled quota options not "
 921                                 "supported.\n");
 922                         break;
 923 #endif
 924                 case Opt_abort:
 925                         set_opt(sbi->s_mount_opt, ABORT);
 926                         break;
 927                 case Opt_barrier:
 928                         if (match_int(&args[0], &option))
 929                                 return 0;
 930                         if (option)
 931                                 set_opt(sbi->s_mount_opt, BARRIER);
 932                         else
 933                                 clear_opt(sbi->s_mount_opt, BARRIER);
 934                         break;
 935                 case Opt_ignore:
 936                         break;
 937                 case Opt_resize:
 938                         if (!n_blocks_count) {
 939                                 printk("EXT3-fs: resize option only available "
 940                                         "for remount\n");
 941                                 return 0;
 942                         }
 943                         match_int(&args[0], &option);
 944                         *n_blocks_count = option;
 945                         break;
 946                 case Opt_nobh:
 947                         set_opt(sbi->s_mount_opt, NOBH);
 948                         break;
 949                 default:
 950                         printk (KERN_ERR
 951                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 952                                 "or missing value\n", p);
 953                         return 0;
 954                 }
 955         }
 956 #ifdef CONFIG_QUOTA
 957         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 958             sbi->s_qf_names[GRPQUOTA])) {
 959                 printk(KERN_ERR
 960                         "EXT3-fs: journalled quota format not specified.\n");
 961                 return 0;
 962         }
 963 #endif
 964
 965         return 1;
 966 }
 967
 968 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 969                             int read_only)
 970 {
 971         struct ext3_sb_info *sbi = EXT3_SB(sb);
 972         int res = 0;
 973
 974         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 975                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 976                         "forcing read-only mode\n");
 977                 res = MS_RDONLY;
 978         }
 979         if (read_only)
 980                 return res;
 981         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 982                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 983                         "running e2fsck is recommended\n");
 984         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 985                 printk (KERN_WARNING
 986                         "EXT3-fs warning: mounting fs with errors, "
 987                         "running e2fsck is recommended\n");
 988         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 989                  le16_to_cpu(es->s_mnt_count) >=
 990                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 991                 printk (KERN_WARNING
 992                         "EXT3-fs warning: maximal mount count reached, "
 993                         "running e2fsck is recommended\n");
 994         else if (le32_to_cpu(es->s_checkinterval) &&
 995                 (le32_to_cpu(es->s_lastcheck) +
 996                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 997                 printk (KERN_WARNING
 998                         "EXT3-fs warning: checktime reached, "
 999                         "running e2fsck is recommended\n");
1000 #if 0
1001                 /* @@@ We _will_ want to clear the valid bit if we find
1002                    inconsistencies, to force a fsck at reboot.  But for
1003                    a plain journaled filesystem we can keep it set as
1004                    valid forever! :) */
1005         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1006 #endif
1007         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1008                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1009         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1010         es->s_mtime = cpu_to_le32(get_seconds());
1011         ext3_update_dynamic_rev(sb);
1012         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1013
1014         ext3_commit_super(sb, es, 1);
1015         if (test_opt(sb, DEBUG))
1016                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1017                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1018                         sb->s_blocksize,
1019                         sbi->s_groups_count,
1020                         EXT3_BLOCKS_PER_GROUP(sb),
1021                         EXT3_INODES_PER_GROUP(sb),
1022                         sbi->s_mount_opt);
1023
1024         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1025         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1026                 char b[BDEVNAME_SIZE];
1027
1028                 printk("external journal on %s\n",
1029                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1030         } else {
1031                 printk("internal journal\n");
1032         }
1033 #ifdef CONFIG_EXT3_CHECK
1034         if (test_opt (sb, CHECK)) {
1035                 ext3_check_blocks_bitmap (sb);
1036                 ext3_check_inodes_bitmap (sb);
1037         }
1038 #endif
1039         return res;
1040 }
1041
1042 /* Called at mount-time, super-block is locked */
1043 static int ext3_check_descriptors (struct super_block * sb)
1044 {
1045         struct ext3_sb_info *sbi = EXT3_SB(sb);
1046         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1047         struct ext3_group_desc * gdp = NULL;
1048         int desc_block = 0;
1049         int i;
1050
1051         ext3_debug ("Checking group descriptors");
1052
1053         for (i = 0; i < sbi->s_groups_count; i++)
1054         {
1055                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1056                         gdp = (struct ext3_group_desc *)
1057                                         sbi->s_group_desc[desc_block++]->b_data;
1058                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1059                     le32_to_cpu(gdp->bg_block_bitmap) >=
1060                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1061                 {
1062                         ext3_error (sb, "ext3_check_descriptors",
1063                                     "Block bitmap for group %d"
1064                                     " not in group (block %lu)!",
1065                                     i, (unsigned long)
1066                                         le32_to_cpu(gdp->bg_block_bitmap));
1067                         return 0;
1068                 }
1069                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1070                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1071                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1072                 {
1073                         ext3_error (sb, "ext3_check_descriptors",
1074                                     "Inode bitmap for group %d"
1075                                     " not in group (block %lu)!",
1076                                     i, (unsigned long)
1077                                         le32_to_cpu(gdp->bg_inode_bitmap));
1078                         return 0;
1079                 }
1080                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1081                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1082                     block + EXT3_BLOCKS_PER_GROUP(sb))
1083                 {
1084                         ext3_error (sb, "ext3_check_descriptors",
1085                                     "Inode table for group %d"
1086                                     " not in group (block %lu)!",
1087                                     i, (unsigned long)
1088                                         le32_to_cpu(gdp->bg_inode_table));
1089                         return 0;
1090                 }
1091                 block += EXT3_BLOCKS_PER_GROUP(sb);
1092                 gdp++;
1093         }
1094
1095         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1096         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1097         return 1;
1098 }
1099
1100
1101 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1102  * the superblock) which were deleted from all directories, but held open by
1103  * a process at the time of a crash.  We walk the list and try to delete these
1104  * inodes at recovery time (only with a read-write filesystem).
1105  *
1106  * In order to keep the orphan inode chain consistent during traversal (in
1107  * case of crash during recovery), we link each inode into the superblock
1108  * orphan list_head and handle it the same way as an inode deletion during
1109  * normal operation (which journals the operations for us).
1110  *
1111  * We only do an iget() and an iput() on each inode, which is very safe if we
1112  * accidentally point at an in-use or already deleted inode.  The worst that
1113  * can happen in this case is that we get a "bit already cleared" message from
1114  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1115  * e2fsck was run on this filesystem, and it must have already done the orphan
1116  * inode cleanup for us, so we can safely abort without any further action.
1117  */
1118 static void ext3_orphan_cleanup (struct super_block * sb,
1119                                  struct ext3_super_block * es)
1120 {
1121         unsigned int s_flags = sb->s_flags;
1122         int nr_orphans = 0, nr_truncates = 0;
1123 #ifdef CONFIG_QUOTA
1124         int i;
1125 #endif
1126         if (!es->s_last_orphan) {
1127                 jbd_debug(4, "no orphan inodes to clean up\n");
1128                 return;
1129         }
1130
1131         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1132                 if (es->s_last_orphan)
1133                         jbd_debug(1, "Errors on filesystem, "
1134                                   "clearing orphan list.\n");
1135                 es->s_last_orphan = 0;
1136                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1137                 return;
1138         }
1139
1140         if (s_flags & MS_RDONLY) {
1141                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1142                        sb->s_id);
1143                 sb->s_flags &= ~MS_RDONLY;
1144         }
1145 #ifdef CONFIG_QUOTA
1146         /* Needed for iput() to work correctly and not trash data */
1147         sb->s_flags |= MS_ACTIVE;
1148         /* Turn on quotas so that they are updated correctly */
1149         for (i = 0; i < MAXQUOTAS; i++) {
1150                 if (EXT3_SB(sb)->s_qf_names[i]) {
1151                         int ret = ext3_quota_on_mount(sb, i);
1152                         if (ret < 0)
1153                                 printk(KERN_ERR
1154                                         "EXT3-fs: Cannot turn on journalled "
1155                                         "quota: error %d\n", ret);
1156                 }
1157         }
1158 #endif
1159
1160         while (es->s_last_orphan) {
1161                 struct inode *inode;
1162
1163                 if (!(inode =
1164                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1165                         es->s_last_orphan = 0;
1166                         break;
1167                 }
1168
1169                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1170                 DQUOT_INIT(inode);
1171                 if (inode->i_nlink) {
1172                         printk(KERN_DEBUG
1173                                 "%s: truncating inode %ld to %Ld bytes\n",
1174                                 __FUNCTION__, inode->i_ino, inode->i_size);
1175                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1176                                   inode->i_ino, inode->i_size);
1177                         ext3_truncate(inode);
1178                         nr_truncates++;
1179                 } else {
1180                         printk(KERN_DEBUG
1181                                 "%s: deleting unreferenced inode %ld\n",
1182                                 __FUNCTION__, inode->i_ino);
1183                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1184                                   inode->i_ino);
1185                         nr_orphans++;
1186                 }
1187                 iput(inode);  /* The delete magic happens here! */
1188         }
1189
1190 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1191
1192         if (nr_orphans)
1193                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1194                        sb->s_id, PLURAL(nr_orphans));
1195         if (nr_truncates)
1196                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1197                        sb->s_id, PLURAL(nr_truncates));
1198 #ifdef CONFIG_QUOTA
1199         /* Turn quotas off */
1200         for (i = 0; i < MAXQUOTAS; i++) {
1201                 if (sb_dqopt(sb)->files[i])
1202                         vfs_quota_off(sb, i);
1203         }
1204 #endif
1205         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1206 }
1207
1208 #define log2(n) ffz(~(n))
1209
1210 /*
1211  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1212  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1213  * We need to be 1 filesystem block less than the 2^32 sector limit.
1214  */
1215 static loff_t ext3_max_size(int bits)
1216 {
1217         loff_t res = EXT3_NDIR_BLOCKS;
1218         /* This constant is calculated to be the largest file size for a
1219          * dense, 4k-blocksize file such that the total number of
1220          * sectors in the file, including data and all indirect blocks,
1221          * does not exceed 2^32. */
1222         const loff_t upper_limit = 0x1ff7fffd000LL;
1223
1224         res += 1LL << (bits-2);
1225         res += 1LL << (2*(bits-2));
1226         res += 1LL << (3*(bits-2));
1227         res <<= bits;
1228         if (res > upper_limit)
1229                 res = upper_limit;
1230         return res;
1231 }
1232
1233 static unsigned long descriptor_loc(struct super_block *sb,
1234                                     unsigned long logic_sb_block,
1235                                     int nr)
1236 {
1237         struct ext3_sb_info *sbi = EXT3_SB(sb);
1238         unsigned long bg, first_data_block, first_meta_bg;
1239         int has_super = 0;
1240
1241         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1242         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1243
1244         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1245             nr < first_meta_bg)
1246                 return (logic_sb_block + nr + 1);
1247         bg = sbi->s_desc_per_block * nr;
1248         if (ext3_bg_has_super(sb, bg))
1249                 has_super = 1;
1250         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1251 }
1252
1253
1254 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1255 {
1256         struct buffer_head * bh;
1257         struct ext3_super_block *es = NULL;
1258         struct ext3_sb_info *sbi;
1259         unsigned long block;
1260         unsigned long sb_block = get_sb_block(&data);
1261         unsigned long logic_sb_block;
1262         unsigned long offset = 0;
1263         unsigned long journal_inum = 0;
1264         unsigned long def_mount_opts;
1265         struct inode *root;
1266         int blocksize;
1267         int hblock;
1268         int db_count;
1269         int i;
1270         int needs_recovery;
1271         __le32 features;
1272
1273         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1274         if (!sbi)
1275                 return -ENOMEM;
1276         sb->s_fs_info = sbi;
1277         memset(sbi, 0, sizeof(*sbi));
1278         sbi->s_mount_opt = 0;
1279         sbi->s_resuid = EXT3_DEF_RESUID;
1280         sbi->s_resgid = EXT3_DEF_RESGID;
1281
1282         unlock_kernel();
1283
1284         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1285         if (!blocksize) {
1286                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1287                 goto out_fail;
1288         }
1289
1290         /*
1291          * The ext3 superblock will not be buffer aligned for other than 1kB
1292          * block sizes.  We need to calculate the offset from buffer start.
1293          */
1294         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1295                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1296                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1297         } else {
1298                 logic_sb_block = sb_block;
1299         }
1300
1301         if (!(bh = sb_bread(sb, logic_sb_block))) {
1302                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1303                 goto out_fail;
1304         }
1305         /*
1306          * Note: s_es must be initialized as soon as possible because
1307          *       some ext3 macro-instructions depend on its value
1308          */
1309         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1310         sbi->s_es = es;
1311         sb->s_magic = le16_to_cpu(es->s_magic);
1312         if (sb->s_magic != EXT3_SUPER_MAGIC)
1313                 goto cantfind_ext3;
1314
1315         /* Set defaults before we parse the mount options */
1316         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1317         if (def_mount_opts & EXT3_DEFM_DEBUG)
1318                 set_opt(sbi->s_mount_opt, DEBUG);
1319         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1320                 set_opt(sbi->s_mount_opt, GRPID);
1321         if (def_mount_opts & EXT3_DEFM_UID16)
1322                 set_opt(sbi->s_mount_opt, NO_UID32);
1323         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1324                 set_opt(sbi->s_mount_opt, XATTR_USER);
1325         if (def_mount_opts & EXT3_DEFM_ACL)
1326                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1327         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1328                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1329         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1330                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1331         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1332                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1333
1334         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1335                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1336         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1337                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1338
1339         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1340         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1341
1342         set_opt(sbi->s_mount_opt, RESERVATION);
1343
1344         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1345                 goto failed_mount;
1346
1347         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1348                 sb->s_flags |= MS_TAGXID;
1349
1350         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1351                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1352
1353         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1354             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1355              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1356              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1357                 printk(KERN_WARNING
1358                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1359                        "running e2fsck is recommended\n");
1360         /*
1361          * Check feature flags regardless of the revision level, since we
1362          * previously didn't change the revision level when setting the flags,
1363          * so there is a chance incompat flags are set on a rev 0 filesystem.
1364          */
1365         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1366         if (features) {
1367                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1368                        "unsupported optional features (%x).\n",
1369                        sb->s_id, le32_to_cpu(features));
1370                 goto failed_mount;
1371         }
1372         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1373         if (!(sb->s_flags & MS_RDONLY) && features) {
1374                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1375                        "unsupported optional features (%x).\n",
1376                        sb->s_id, le32_to_cpu(features));
1377                 goto failed_mount;
1378         }
1379         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1380
1381         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1382             blocksize > EXT3_MAX_BLOCK_SIZE) {
1383                 printk(KERN_ERR
1384                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1385                        blocksize, sb->s_id);
1386                 goto failed_mount;
1387         }
1388
1389         hblock = bdev_hardsect_size(sb->s_bdev);
1390         if (sb->s_blocksize != blocksize) {
1391                 /*
1392                  * Make sure the blocksize for the filesystem is larger
1393                  * than the hardware sectorsize for the machine.
1394                  */
1395                 if (blocksize < hblock) {
1396                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1397                                "device blocksize %d.\n", blocksize, hblock);
1398                         goto failed_mount;
1399                 }
1400
1401                 brelse (bh);
1402                 sb_set_blocksize(sb, blocksize);
1403                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1404                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1405                 bh = sb_bread(sb, logic_sb_block);
1406                 if (!bh) {
1407                         printk(KERN_ERR
1408                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1409                         goto failed_mount;
1410                 }
1411                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1412                 sbi->s_es = es;
1413                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1414                         printk (KERN_ERR
1415                                 "EXT3-fs: Magic mismatch, very weird !\n");
1416                         goto failed_mount;
1417                 }
1418         }
1419
1420         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1421
1422         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1423                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1424                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1425         } else {
1426                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1427                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1428                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1429                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1430                     (sbi->s_inode_size > blocksize)) {
1431                         printk (KERN_ERR
1432                                 "EXT3-fs: unsupported inode size: %d\n",
1433                                 sbi->s_inode_size);
1434                         goto failed_mount;
1435                 }
1436         }
1437         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1438                                    le32_to_cpu(es->s_log_frag_size);
1439         if (blocksize != sbi->s_frag_size) {
1440                 printk(KERN_ERR
1441                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1442                        sbi->s_frag_size, blocksize);
1443                 goto failed_mount;
1444         }
1445         sbi->s_frags_per_block = 1;
1446         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1447         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1448         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1449         if (EXT3_INODE_SIZE(sb) == 0)
1450                 goto cantfind_ext3;
1451         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1452         if (sbi->s_inodes_per_block == 0)
1453                 goto cantfind_ext3;
1454         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1455                                         sbi->s_inodes_per_block;
1456         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1457         sbi->s_sbh = bh;
1458         sbi->s_mount_state = le16_to_cpu(es->s_state);
1459         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1460         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1461         for (i=0; i < 4; i++)
1462                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1463         sbi->s_def_hash_version = es->s_def_hash_version;
1464
1465         if (sbi->s_blocks_per_group > blocksize * 8) {
1466                 printk (KERN_ERR
1467                         "EXT3-fs: #blocks per group too big: %lu\n",
1468                         sbi->s_blocks_per_group);
1469                 goto failed_mount;
1470         }
1471         if (sbi->s_frags_per_group > blocksize * 8) {
1472                 printk (KERN_ERR
1473                         "EXT3-fs: #fragments per group too big: %lu\n",
1474                         sbi->s_frags_per_group);
1475                 goto failed_mount;
1476         }
1477         if (sbi->s_inodes_per_group > blocksize * 8) {
1478                 printk (KERN_ERR
1479                         "EXT3-fs: #inodes per group too big: %lu\n",
1480                         sbi->s_inodes_per_group);
1481                 goto failed_mount;
1482         }
1483
1484         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1485                 goto cantfind_ext3;
1486         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1487                                le32_to_cpu(es->s_first_data_block) +
1488                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1489                               EXT3_BLOCKS_PER_GROUP(sb);
1490         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1491                    EXT3_DESC_PER_BLOCK(sb);
1492         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1493                                     GFP_KERNEL);
1494         if (sbi->s_group_desc == NULL) {
1495                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1496                 goto failed_mount;
1497         }
1498
1499         percpu_counter_init(&sbi->s_freeblocks_counter);
1500         percpu_counter_init(&sbi->s_freeinodes_counter);
1501         percpu_counter_init(&sbi->s_dirs_counter);
1502         bgl_lock_init(&sbi->s_blockgroup_lock);
1503
1504         for (i = 0; i < db_count; i++) {
1505                 block = descriptor_loc(sb, logic_sb_block, i);
1506                 sbi->s_group_desc[i] = sb_bread(sb, block);
1507                 if (!sbi->s_group_desc[i]) {
1508                         printk (KERN_ERR "EXT3-fs: "
1509                                 "can't read group descriptor %d\n", i);
1510                         db_count = i;
1511                         goto failed_mount2;
1512                 }
1513         }
1514         if (!ext3_check_descriptors (sb)) {
1515                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1516                 goto failed_mount2;
1517         }
1518         sbi->s_gdb_count = db_count;
1519         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1520         spin_lock_init(&sbi->s_next_gen_lock);
1521         /* per fileystem reservation list head & lock */
1522         spin_lock_init(&sbi->s_rsv_window_lock);
1523         sbi->s_rsv_window_root = RB_ROOT;
1524         /* Add a single, static dummy reservation to the start of the
1525          * reservation window list --- it gives us a placeholder for
1526          * append-at-start-of-list which makes the allocation logic
1527          * _much_ simpler. */
1528         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1529         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1530         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1531         sbi->s_rsv_window_head.rsv_goal_size = 0;
1532         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1533
1534         /*
1535          * set up enough so that it can read an inode
1536          */
1537         sb->s_op = &ext3_sops;
1538         sb->s_export_op = &ext3_export_ops;
1539         sb->s_xattr = ext3_xattr_handlers;
1540 #ifdef CONFIG_QUOTA
1541         sb->s_qcop = &ext3_qctl_operations;
1542         sb->dq_op = &ext3_quota_operations;
1543 #endif
1544         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1545
1546         sb->s_root = NULL;
1547
1548         needs_recovery = (es->s_last_orphan != 0 ||
1549                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1550                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1551
1552         /*
1553          * The first inode we look at is the journal inode.  Don't try
1554          * root first: it may be modified in the journal!
1555          */
1556         if (!test_opt(sb, NOLOAD) &&
1557             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1558                 if (ext3_load_journal(sb, es))
1559                         goto failed_mount2;
1560         } else if (journal_inum) {
1561                 if (ext3_create_journal(sb, es, journal_inum))
1562                         goto failed_mount2;
1563         } else {
1564                 if (!silent)
1565                         printk (KERN_ERR
1566                                 "ext3: No journal on filesystem on %s\n",
1567                                 sb->s_id);
1568                 goto failed_mount2;
1569         }
1570
1571         /* We have now updated the journal if required, so we can
1572          * validate the data journaling mode. */
1573         switch (test_opt(sb, DATA_FLAGS)) {
1574         case 0:
1575                 /* No mode set, assume a default based on the journal
1576                    capabilities: ORDERED_DATA if the journal can
1577                    cope, else JOURNAL_DATA */
1578                 if (journal_check_available_features
1579                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1580                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1581                 else
1582                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1583                 break;
1584
1585         case EXT3_MOUNT_ORDERED_DATA:
1586         case EXT3_MOUNT_WRITEBACK_DATA:
1587                 if (!journal_check_available_features
1588                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1589                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1590                                "requested data journaling mode\n");
1591                         goto failed_mount3;
1592                 }
1593         default:
1594                 break;
1595         }
1596
1597         if (test_opt(sb, NOBH)) {
1598                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1599                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1600                                 "since filesystem blocksize doesn't match "
1601                                 "pagesize\n");
1602                         clear_opt(sbi->s_mount_opt, NOBH);
1603                 }
1604                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1605                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1606                                 "its supported only with writeback mode\n");
1607                         clear_opt(sbi->s_mount_opt, NOBH);
1608                 }
1609         }
1610         /*
1611          * The journal_load will have done any necessary log recovery,
1612          * so we can safely mount the rest of the filesystem now.
1613          */
1614
1615         root = iget(sb, EXT3_ROOT_INO);
1616         sb->s_root = d_alloc_root(root);
1617         if (!sb->s_root) {
1618                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1619                 iput(root);
1620                 goto failed_mount3;
1621         }
1622         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1623                 dput(sb->s_root);
1624                 sb->s_root = NULL;
1625                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1626                 goto failed_mount3;
1627         }
1628
1629         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1630         /*
1631          * akpm: core read_super() calls in here with the superblock locked.
1632          * That deadlocks, because orphan cleanup needs to lock the superblock
1633          * in numerous places.  Here we just pop the lock - it's relatively
1634          * harmless, because we are now ready to accept write_super() requests,
1635          * and aviro says that's the only reason for hanging onto the
1636          * superblock lock.
1637          */
1638         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1639         ext3_orphan_cleanup(sb, es);
1640         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1641         if (needs_recovery)
1642                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1643         ext3_mark_recovery_complete(sb, es);
1644         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1645                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1646                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1647                 "writeback");
1648
1649         percpu_counter_mod(&sbi->s_freeblocks_counter,
1650                 ext3_count_free_blocks(sb));
1651         percpu_counter_mod(&sbi->s_freeinodes_counter,
1652                 ext3_count_free_inodes(sb));
1653         percpu_counter_mod(&sbi->s_dirs_counter,
1654                 ext3_count_dirs(sb));
1655
1656         lock_kernel();
1657         return 0;
1658
1659 cantfind_ext3:
1660         if (!silent)
1661                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1662                        sb->s_id);
1663         goto failed_mount;
1664
1665 failed_mount3:
1666         journal_destroy(sbi->s_journal);
1667 failed_mount2:
1668         for (i = 0; i < db_count; i++)
1669                 brelse(sbi->s_group_desc[i]);
1670         kfree(sbi->s_group_desc);
1671 failed_mount:
1672 #ifdef CONFIG_QUOTA
1673         for (i = 0; i < MAXQUOTAS; i++)
1674                 kfree(sbi->s_qf_names[i]);
1675 #endif
1676         ext3_blkdev_remove(sbi);
1677         brelse(bh);
1678 out_fail:
1679         sb->s_fs_info = NULL;
1680         kfree(sbi);
1681         lock_kernel();
1682         return -EINVAL;
1683 }
1684
1685 /*
1686  * Setup any per-fs journal parameters now.  We'll do this both on
1687  * initial mount, once the journal has been initialised but before we've
1688  * done any recovery; and again on any subsequent remount.
1689  */
1690 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1691 {
1692         struct ext3_sb_info *sbi = EXT3_SB(sb);
1693
1694         if (sbi->s_commit_interval)
1695                 journal->j_commit_interval = sbi->s_commit_interval;
1696         /* We could also set up an ext3-specific default for the commit
1697          * interval here, but for now we'll just fall back to the jbd
1698          * default. */
1699
1700         spin_lock(&journal->j_state_lock);
1701         if (test_opt(sb, BARRIER))
1702                 journal->j_flags |= JFS_BARRIER;
1703         else
1704                 journal->j_flags &= ~JFS_BARRIER;
1705         spin_unlock(&journal->j_state_lock);
1706 }
1707
1708 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1709 {
1710         struct inode *journal_inode;
1711         journal_t *journal;
1712
1713         /* First, test for the existence of a valid inode on disk.  Bad
1714          * things happen if we iget() an unused inode, as the subsequent
1715          * iput() will try to delete it. */
1716
1717         journal_inode = iget(sb, journal_inum);
1718         if (!journal_inode) {
1719                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1720                 return NULL;
1721         }
1722         if (!journal_inode->i_nlink) {
1723                 make_bad_inode(journal_inode);
1724                 iput(journal_inode);
1725                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1726                 return NULL;
1727         }
1728
1729         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1730                   journal_inode, journal_inode->i_size);
1731         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1732                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1733                 iput(journal_inode);
1734                 return NULL;
1735         }
1736
1737         journal = journal_init_inode(journal_inode);
1738         if (!journal) {
1739                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1740                 iput(journal_inode);
1741                 return NULL;
1742         }
1743         journal->j_private = sb;
1744         ext3_init_journal_params(sb, journal);
1745         return journal;
1746 }
1747
1748 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1749                                        dev_t j_dev)
1750 {
1751         struct buffer_head * bh;
1752         journal_t *journal;
1753         int start;
1754         int len;
1755         int hblock, blocksize;
1756         unsigned long sb_block;
1757         unsigned long offset;
1758         struct ext3_super_block * es;
1759         struct block_device *bdev;
1760
1761         bdev = ext3_blkdev_get(j_dev);
1762         if (bdev == NULL)
1763                 return NULL;
1764
1765         if (bd_claim(bdev, sb)) {
1766                 printk(KERN_ERR
1767                         "EXT3: failed to claim external journal device.\n");
1768                 blkdev_put(bdev);
1769                 return NULL;
1770         }
1771
1772         blocksize = sb->s_blocksize;
1773         hblock = bdev_hardsect_size(bdev);
1774         if (blocksize < hblock) {
1775                 printk(KERN_ERR
1776                         "EXT3-fs: blocksize too small for journal device.\n");
1777                 goto out_bdev;
1778         }
1779
1780         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1781         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1782         set_blocksize(bdev, blocksize);
1783         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1784                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1785                        "external journal\n");
1786                 goto out_bdev;
1787         }
1788
1789         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1790         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1791             !(le32_to_cpu(es->s_feature_incompat) &
1792               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1793                 printk(KERN_ERR "EXT3-fs: external journal has "
1794                                         "bad superblock\n");
1795                 brelse(bh);
1796                 goto out_bdev;
1797         }
1798
1799         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1800                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1801                 brelse(bh);
1802                 goto out_bdev;
1803         }
1804
1805         len = le32_to_cpu(es->s_blocks_count);
1806         start = sb_block + 1;
1807         brelse(bh);     /* we're done with the superblock */
1808
1809         journal = journal_init_dev(bdev, sb->s_bdev,
1810                                         start, len, blocksize);
1811         if (!journal) {
1812                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1813                 goto out_bdev;
1814         }
1815         journal->j_private = sb;
1816         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1817         wait_on_buffer(journal->j_sb_buffer);
1818         if (!buffer_uptodate(journal->j_sb_buffer)) {
1819                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1820                 goto out_journal;
1821         }
1822         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1823                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1824                                         "user (unsupported) - %d\n",
1825                         be32_to_cpu(journal->j_superblock->s_nr_users));
1826                 goto out_journal;
1827         }
1828         EXT3_SB(sb)->journal_bdev = bdev;
1829         ext3_init_journal_params(sb, journal);
1830         return journal;
1831 out_journal:
1832         journal_destroy(journal);
1833 out_bdev:
1834         ext3_blkdev_put(bdev);
1835         return NULL;
1836 }
1837
1838 static int ext3_load_journal(struct super_block * sb,
1839                              struct ext3_super_block * es)
1840 {
1841         journal_t *journal;
1842         int journal_inum = le32_to_cpu(es->s_journal_inum);
1843         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1844         int err = 0;
1845         int really_read_only;
1846
1847         really_read_only = bdev_read_only(sb->s_bdev);
1848
1849         /*
1850          * Are we loading a blank journal or performing recovery after a
1851          * crash?  For recovery, we need to check in advance whether we
1852          * can get read-write access to the device.
1853          */
1854
1855         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1856                 if (sb->s_flags & MS_RDONLY) {
1857                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1858                                         "required on readonly filesystem.\n");
1859                         if (really_read_only) {
1860                                 printk(KERN_ERR "EXT3-fs: write access "
1861                                         "unavailable, cannot proceed.\n");
1862                                 return -EROFS;
1863                         }
1864                         printk (KERN_INFO "EXT3-fs: write access will "
1865                                         "be enabled during recovery.\n");
1866                 }
1867         }
1868
1869         if (journal_inum && journal_dev) {
1870                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1871                        "and inode journals!\n");
1872                 return -EINVAL;
1873         }
1874
1875         if (journal_inum) {
1876                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1877                         return -EINVAL;
1878         } else {
1879                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1880                         return -EINVAL;
1881         }
1882
1883         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1884                 err = journal_update_format(journal);
1885                 if (err)  {
1886                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1887                         journal_destroy(journal);
1888                         return err;
1889                 }
1890         }
1891
1892         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1893                 err = journal_wipe(journal, !really_read_only);
1894         if (!err)
1895                 err = journal_load(journal);
1896
1897         if (err) {
1898                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1899                 journal_destroy(journal);
1900                 return err;
1901         }
1902
1903         EXT3_SB(sb)->s_journal = journal;
1904         ext3_clear_journal_err(sb, es);
1905         return 0;
1906 }
1907
1908 static int ext3_create_journal(struct super_block * sb,
1909                                struct ext3_super_block * es,
1910                                int journal_inum)
1911 {
1912         journal_t *journal;
1913
1914         if (sb->s_flags & MS_RDONLY) {
1915                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1916                                 "create journal.\n");
1917                 return -EROFS;
1918         }
1919
1920         if (!(journal = ext3_get_journal(sb, journal_inum)))
1921                 return -EINVAL;
1922
1923         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1924                journal_inum);
1925
1926         if (journal_create(journal)) {
1927                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1928                 journal_destroy(journal);
1929                 return -EIO;
1930         }
1931
1932         EXT3_SB(sb)->s_journal = journal;
1933
1934         ext3_update_dynamic_rev(sb);
1935         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1936         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1937
1938         es->s_journal_inum = cpu_to_le32(journal_inum);
1939         sb->s_dirt = 1;
1940
1941         /* Make sure we flush the recovery flag to disk. */
1942         ext3_commit_super(sb, es, 1);
1943
1944         return 0;
1945 }
1946
1947 static void ext3_commit_super (struct super_block * sb,
1948                                struct ext3_super_block * es,
1949                                int sync)
1950 {
1951         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1952
1953         if (!sbh)
1954                 return;
1955         es->s_wtime = cpu_to_le32(get_seconds());
1956         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1957         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1958         BUFFER_TRACE(sbh, "marking dirty");
1959         mark_buffer_dirty(sbh);
1960         if (sync)
1961                 sync_dirty_buffer(sbh);
1962 }
1963
1964
1965 /*
1966  * Have we just finished recovery?  If so, and if we are mounting (or
1967  * remounting) the filesystem readonly, then we will end up with a
1968  * consistent fs on disk.  Record that fact.
1969  */
1970 static void ext3_mark_recovery_complete(struct super_block * sb,
1971                                         struct ext3_super_block * es)
1972 {
1973         journal_t *journal = EXT3_SB(sb)->s_journal;
1974
1975         journal_lock_updates(journal);
1976         journal_flush(journal);
1977         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1978             sb->s_flags & MS_RDONLY) {
1979                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1980                 sb->s_dirt = 0;
1981                 ext3_commit_super(sb, es, 1);
1982         }
1983         journal_unlock_updates(journal);
1984 }
1985
1986 /*
1987  * If we are mounting (or read-write remounting) a filesystem whose journal
1988  * has recorded an error from a previous lifetime, move that error to the
1989  * main filesystem now.
1990  */
1991 static void ext3_clear_journal_err(struct super_block * sb,
1992                                    struct ext3_super_block * es)
1993 {
1994         journal_t *journal;
1995         int j_errno;
1996         const char *errstr;
1997
1998         journal = EXT3_SB(sb)->s_journal;
1999
2000         /*
2001          * Now check for any error status which may have been recorded in the
2002          * journal by a prior ext3_error() or ext3_abort()
2003          */
2004
2005         j_errno = journal_errno(journal);
2006         if (j_errno) {
2007                 char nbuf[16];
2008
2009                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2010                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2011                              "from previous mount: %s", errstr);
2012                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2013                              "filesystem check.");
2014
2015                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2016                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2017                 ext3_commit_super (sb, es, 1);
2018
2019                 journal_clear_err(journal);
2020         }
2021 }
2022
2023 /*
2024  * Force the running and committing transactions to commit,
2025  * and wait on the commit.
2026  */
2027 int ext3_force_commit(struct super_block *sb)
2028 {
2029         journal_t *journal;
2030         int ret;
2031
2032         if (sb->s_flags & MS_RDONLY)
2033                 return 0;
2034
2035         journal = EXT3_SB(sb)->s_journal;
2036         sb->s_dirt = 0;
2037         ret = ext3_journal_force_commit(journal);
2038         return ret;
2039 }
2040
2041 /*
2042  * Ext3 always journals updates to the superblock itself, so we don't
2043  * have to propagate any other updates to the superblock on disk at this
2044  * point.  Just start an async writeback to get the buffers on their way
2045  * to the disk.
2046  *
2047  * This implicitly triggers the writebehind on sync().
2048  */
2049
2050 static void ext3_write_super (struct super_block * sb)
2051 {
2052         if (down_trylock(&sb->s_lock) == 0)
2053                 BUG();
2054         sb->s_dirt = 0;
2055 }
2056
2057 static int ext3_sync_fs(struct super_block *sb, int wait)
2058 {
2059         tid_t target;
2060
2061         sb->s_dirt = 0;
2062         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2063                 if (wait)
2064                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2065         }
2066         return 0;
2067 }
2068
2069 /*
2070  * LVM calls this function before a (read-only) snapshot is created.  This
2071  * gives us a chance to flush the journal completely and mark the fs clean.
2072  */
2073 static void ext3_write_super_lockfs(struct super_block *sb)
2074 {
2075         sb->s_dirt = 0;
2076
2077         if (!(sb->s_flags & MS_RDONLY)) {
2078                 journal_t *journal = EXT3_SB(sb)->s_journal;
2079
2080                 /* Now we set up the journal barrier. */
2081                 journal_lock_updates(journal);
2082                 journal_flush(journal);
2083
2084                 /* Journal blocked and flushed, clear needs_recovery flag. */
2085                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2086                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2087         }
2088 }
2089
2090 /*
2091  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2092  * flag here, even though the filesystem is not technically dirty yet.
2093  */
2094 static void ext3_unlockfs(struct super_block *sb)
2095 {
2096         if (!(sb->s_flags & MS_RDONLY)) {
2097                 lock_super(sb);
2098                 /* Reser the needs_recovery flag before the fs is unlocked. */
2099                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2100                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2101                 unlock_super(sb);
2102                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2103         }
2104 }
2105
2106 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2107 {
2108         struct ext3_super_block * es;
2109         struct ext3_sb_info *sbi = EXT3_SB(sb);
2110         unsigned long tmp;
2111         unsigned long n_blocks_count = 0;
2112
2113         /*
2114          * Allow the "check" option to be passed as a remount option.
2115          */
2116         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2117                 return -EINVAL;
2118
2119         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2120                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2121
2122         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2123                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2124
2125         es = sbi->s_es;
2126
2127         ext3_init_journal_params(sb, sbi->s_journal);
2128
2129         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2130                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2131                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2132                         return -EROFS;
2133
2134                 if (*flags & MS_RDONLY) {
2135                         /*
2136                          * First of all, the unconditional stuff we have to do
2137                          * to disable replay of the journal when we next remount
2138                          */
2139                         sb->s_flags |= MS_RDONLY;
2140
2141                         /*
2142                          * OK, test if we are remounting a valid rw partition
2143                          * readonly, and if so set the rdonly flag and then
2144                          * mark the partition as valid again.
2145                          */
2146                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2147                             (sbi->s_mount_state & EXT3_VALID_FS))
2148                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2149
2150                         ext3_mark_recovery_complete(sb, es);
2151                 } else {
2152                         __le32 ret;
2153                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2154                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2155                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2156                                        "remount RDWR because of unsupported "
2157                                        "optional features (%x).\n",
2158                                        sb->s_id, le32_to_cpu(ret));
2159                                 return -EROFS;
2160                         }
2161                         /*
2162                          * Mounting a RDONLY partition read-write, so reread
2163                          * and store the current valid flag.  (It may have
2164                          * been changed by e2fsck since we originally mounted
2165                          * the partition.)
2166                          */
2167                         ext3_clear_journal_err(sb, es);
2168                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2169                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2170                                 return ret;
2171                         if (!ext3_setup_super (sb, es, 0))
2172                                 sb->s_flags &= ~MS_RDONLY;
2173                 }
2174         }
2175         return 0;
2176 }
2177
2178 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2179 {
2180         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2181         unsigned long overhead;
2182         int i;
2183
2184         if (test_opt (sb, MINIX_DF))
2185                 overhead = 0;
2186         else {
2187                 unsigned long ngroups;
2188                 ngroups = EXT3_SB(sb)->s_groups_count;
2189                 smp_rmb();
2190
2191                 /*
2192                  * Compute the overhead (FS structures)
2193                  */
2194
2195                 /*
2196                  * All of the blocks before first_data_block are
2197                  * overhead
2198                  */
2199                 overhead = le32_to_cpu(es->s_first_data_block);
2200
2201                 /*
2202                  * Add the overhead attributed to the superblock and
2203                  * block group descriptors.  If the sparse superblocks
2204                  * feature is turned on, then not all groups have this.
2205                  */
2206                 for (i = 0; i < ngroups; i++) {
2207                         overhead += ext3_bg_has_super(sb, i) +
2208                                 ext3_bg_num_gdb(sb, i);
2209                         cond_resched();
2210                 }
2211
2212                 /*
2213                  * Every block group has an inode bitmap, a block
2214                  * bitmap, and an inode table.
2215                  */
2216                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2217         }
2218
2219         buf->f_type = EXT3_SUPER_MAGIC;
2220         buf->f_bsize = sb->s_blocksize;
2221         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2222         buf->f_bfree = ext3_count_free_blocks (sb);
2223         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2224         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2225                 buf->f_bavail = 0;
2226         buf->f_files = le32_to_cpu(es->s_inodes_count);
2227         buf->f_ffree = ext3_count_free_inodes (sb);
2228         buf->f_namelen = EXT3_NAME_LEN;
2229         return 0;
2230 }
2231
2232 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2233  * is locked for write. Otherwise the are possible deadlocks:
2234  * Process 1                         Process 2
2235  * ext3_create()                     quota_sync()
2236  *   journal_start()                   write_dquot()
2237  *   DQUOT_INIT()                        down(dqio_sem)
2238  *     down(dqio_sem)                    journal_start()
2239  *
2240  */
2241
2242 #ifdef CONFIG_QUOTA
2243
2244 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2245 {
2246         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2247 }
2248
2249 static int ext3_dquot_initialize(struct inode *inode, int type)
2250 {
2251         handle_t *handle;
2252         int ret, err;
2253
2254         /* We may create quota structure so we need to reserve enough blocks */
2255         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2256         if (IS_ERR(handle))
2257                 return PTR_ERR(handle);
2258         ret = dquot_initialize(inode, type);
2259         err = ext3_journal_stop(handle);
2260         if (!ret)
2261                 ret = err;
2262         return ret;
2263 }
2264
2265 static int ext3_dquot_drop(struct inode *inode)
2266 {
2267         handle_t *handle;
2268         int ret, err;
2269
2270         /* We may delete quota structure so we need to reserve enough blocks */
2271         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2272         if (IS_ERR(handle))
2273                 return PTR_ERR(handle);
2274         ret = dquot_drop(inode);
2275         err = ext3_journal_stop(handle);
2276         if (!ret)
2277                 ret = err;
2278         return ret;
2279 }
2280
2281 static int ext3_write_dquot(struct dquot *dquot)
2282 {
2283         int ret, err;
2284         handle_t *handle;
2285         struct inode *inode;
2286
2287         inode = dquot_to_inode(dquot);
2288         handle = ext3_journal_start(inode,
2289                                         EXT3_QUOTA_TRANS_BLOCKS);
2290         if (IS_ERR(handle))
2291                 return PTR_ERR(handle);
2292         ret = dquot_commit(dquot);
2293         err = ext3_journal_stop(handle);
2294         if (!ret)
2295                 ret = err;
2296         return ret;
2297 }
2298
2299 static int ext3_acquire_dquot(struct dquot *dquot)
2300 {
2301         int ret, err;
2302         handle_t *handle;
2303
2304         handle = ext3_journal_start(dquot_to_inode(dquot),
2305                                         EXT3_QUOTA_INIT_BLOCKS);
2306         if (IS_ERR(handle))
2307                 return PTR_ERR(handle);
2308         ret = dquot_acquire(dquot);
2309         err = ext3_journal_stop(handle);
2310         if (!ret)
2311                 ret = err;
2312         return ret;
2313 }
2314
2315 static int ext3_release_dquot(struct dquot *dquot)
2316 {
2317         int ret, err;
2318         handle_t *handle;
2319
2320         handle = ext3_journal_start(dquot_to_inode(dquot),
2321                                         EXT3_QUOTA_INIT_BLOCKS);
2322         if (IS_ERR(handle))
2323                 return PTR_ERR(handle);
2324         ret = dquot_release(dquot);
2325         err = ext3_journal_stop(handle);
2326         if (!ret)
2327                 ret = err;
2328         return ret;
2329 }
2330
2331 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2332 {
2333         /* Are we journalling quotas? */
2334         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2335             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2336                 dquot_mark_dquot_dirty(dquot);
2337                 return ext3_write_dquot(dquot);
2338         } else {
2339                 return dquot_mark_dquot_dirty(dquot);
2340         }
2341 }
2342
2343 static int ext3_write_info(struct super_block *sb, int type)
2344 {
2345         int ret, err;
2346         handle_t *handle;
2347
2348         /* Data block + inode block */
2349         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2350         if (IS_ERR(handle))
2351                 return PTR_ERR(handle);
2352         ret = dquot_commit_info(sb, type);
2353         err = ext3_journal_stop(handle);
2354         if (!ret)
2355                 ret = err;
2356         return ret;
2357 }
2358
2359 /*
2360  * Turn on quotas during mount time - we need to find
2361  * the quota file and such...
2362  */
2363 static int ext3_quota_on_mount(struct super_block *sb, int type)
2364 {
2365         int err;
2366         struct dentry *dentry;
2367         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2368                              .hash = 0,
2369                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2370
2371         dentry = lookup_hash(&name, sb->s_root);
2372         if (IS_ERR(dentry))
2373                 return PTR_ERR(dentry);
2374         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2375         /* Now invalidate and put the dentry - quota got its own reference
2376          * to inode and dentry has at least wrong hash so we had better
2377          * throw it away */
2378         d_invalidate(dentry);
2379         dput(dentry);
2380         return err;
2381 }
2382
2383 /*
2384  * Standard function to be called on quota_on
2385  */
2386 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2387                          char *path)
2388 {
2389         int err;
2390         struct nameidata nd;
2391
2392         /* Not journalling quota? */
2393         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2394             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2395                 return vfs_quota_on(sb, type, format_id, path);
2396         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2397         if (err)
2398                 return err;
2399         /* Quotafile not on the same filesystem? */
2400         if (nd.mnt->mnt_sb != sb) {
2401                 path_release(&nd);
2402                 return -EXDEV;
2403         }
2404         /* Quotafile not of fs root? */
2405         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2406                 printk(KERN_WARNING
2407                         "EXT3-fs: Quota file not on filesystem root. "
2408                         "Journalled quota will not work.\n");
2409         path_release(&nd);
2410         return vfs_quota_on(sb, type, format_id, path);
2411 }
2412
2413 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2414  * acquiring the locks... As quota files are never truncated and quota code
2415  * itself serializes the operations (and noone else should touch the files)
2416  * we don't have to be afraid of races */
2417 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2418                                size_t len, loff_t off)
2419 {
2420         struct inode *inode = sb_dqopt(sb)->files[type];
2421         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2422         int err = 0;
2423         int offset = off & (sb->s_blocksize - 1);
2424         int tocopy;
2425         size_t toread;
2426         struct buffer_head *bh;
2427         loff_t i_size = i_size_read(inode);
2428
2429         if (off > i_size)
2430                 return 0;
2431         if (off+len > i_size)
2432                 len = i_size-off;
2433         toread = len;
2434         while (toread > 0) {
2435                 tocopy = sb->s_blocksize - offset < toread ?
2436                                 sb->s_blocksize - offset : toread;
2437                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2438                 if (err)
2439                         return err;
2440                 if (!bh)        /* A hole? */
2441                         memset(data, 0, tocopy);
2442                 else
2443                         memcpy(data, bh->b_data+offset, tocopy);
2444                 brelse(bh);
2445                 offset = 0;
2446                 toread -= tocopy;
2447                 data += tocopy;
2448                 blk++;
2449         }
2450         return len;
2451 }
2452
2453 /* Write to quotafile (we know the transaction is already started and has
2454  * enough credits) */
2455 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2456                                 const char *data, size_t len, loff_t off)
2457 {
2458         struct inode *inode = sb_dqopt(sb)->files[type];
2459         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2460         int err = 0;
2461         int offset = off & (sb->s_blocksize - 1);
2462         int tocopy;
2463         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2464         size_t towrite = len;
2465         struct buffer_head *bh;
2466         handle_t *handle = journal_current_handle();
2467
2468         down(&inode->i_sem);
2469         while (towrite > 0) {
2470                 tocopy = sb->s_blocksize - offset < towrite ?
2471                                 sb->s_blocksize - offset : towrite;
2472                 bh = ext3_bread(handle, inode, blk, 1, &err);
2473                 if (!bh)
2474                         goto out;
2475                 if (journal_quota) {
2476                         err = ext3_journal_get_write_access(handle, bh);
2477                         if (err) {
2478                                 brelse(bh);
2479                                 goto out;
2480                         }
2481                 }
2482                 lock_buffer(bh);
2483                 memcpy(bh->b_data+offset, data, tocopy);
2484                 flush_dcache_page(bh->b_page);
2485                 unlock_buffer(bh);
2486                 if (journal_quota)
2487                         err = ext3_journal_dirty_metadata(handle, bh);
2488                 else {
2489                         /* Always do at least ordered writes for quotas */
2490                         err = ext3_journal_dirty_data(handle, bh);
2491                         mark_buffer_dirty(bh);
2492                 }
2493                 brelse(bh);
2494                 if (err)
2495                         goto out;
2496                 offset = 0;
2497                 towrite -= tocopy;
2498                 data += tocopy;
2499                 blk++;
2500         }
2501 out:
2502         if (len == towrite)
2503                 return err;
2504         if (inode->i_size < off+len-towrite) {
2505                 i_size_write(inode, off+len-towrite);
2506                 EXT3_I(inode)->i_disksize = inode->i_size;
2507         }
2508         inode->i_version++;
2509         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2510         ext3_mark_inode_dirty(handle, inode);
2511         up(&inode->i_sem);
2512         return len - towrite;
2513 }
2514
2515 #endif
2516
2517 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2518         int flags, const char *dev_name, void *data)
2519 {
2520         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2521 }
2522
2523 static struct file_system_type ext3_fs_type = {
2524         .owner          = THIS_MODULE,
2525         .name           = "ext3",
2526         .get_sb         = ext3_get_sb,
2527         .kill_sb        = kill_block_super,
2528         .fs_flags       = FS_REQUIRES_DEV,
2529 };
2530
2531 static int __init init_ext3_fs(void)
2532 {
2533         int err = init_ext3_xattr();
2534         if (err)
2535                 return err;
2536         err = init_inodecache();
2537         if (err)
2538                 goto out1;
2539         err = register_filesystem(&ext3_fs_type);
2540         if (err)
2541                 goto out;
2542         return 0;
2543 out:
2544         destroy_inodecache();
2545 out1:
2546         exit_ext3_xattr();
2547         return err;
2548 }
2549
2550 static void __exit exit_ext3_fs(void)
2551 {
2552         unregister_filesystem(&ext3_fs_type);
2553         destroy_inodecache();
2554         exit_ext3_xattr();
2555 }
2556
2557 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2558 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2559 MODULE_LICENSE("GPL");
2560 module_init(init_ext3_fs)
2561 module_exit(exit_ext3_fs)