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