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/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.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 <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static int ext3_commit_super(struct super_block *sb,
  52                                struct ext3_super_block *es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static int ext3_unfreeze(struct super_block *sb);
  64 static void ext3_write_super (struct super_block * sb);
  65 static int ext3_freeze(struct super_block *sb);
  66
  67 /*
  68  * Wrappers for journal_start/end.
  69  *
  70  * The only special thing we need to do here is to make sure that all
  71  * journal_end calls result in the superblock being marked dirty, so
  72  * that sync() will call the filesystem's write_super callback if
  73  * appropriate.
  74  */
  75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  76 {
  77         journal_t *journal;
  78
  79         if (sb->s_flags & MS_RDONLY)
  80                 return ERR_PTR(-EROFS);
  81
  82         /* Special case here: if the journal has aborted behind our
  83          * backs (eg. EIO in the commit thread), then we still need to
  84          * take the FS itself readonly cleanly. */
  85         journal = EXT3_SB(sb)->s_journal;
  86         if (is_journal_aborted(journal)) {
  87                 ext3_abort(sb, __func__,
  88                            "Detected aborted journal");
  89                 return ERR_PTR(-EROFS);
  90         }
  91
  92         return journal_start(journal, nblocks);
  93 }
  94
  95 /*
  96  * The only special thing we need to do here is to make sure that all
  97  * journal_stop calls result in the superblock being marked dirty, so
  98  * that sync() will call the filesystem's write_super callback if
  99  * appropriate.
 100  */
 101 int __ext3_journal_stop(const char *where, handle_t *handle)
 102 {
 103         struct super_block *sb;
 104         int err;
 105         int rc;
 106
 107         sb = handle->h_transaction->t_journal->j_private;
 108         err = handle->h_err;
 109         rc = journal_stop(handle);
 110
 111         if (!err)
 112                 err = rc;
 113         if (err)
 114                 __ext3_std_error(sb, where, err);
 115         return err;
 116 }
 117
 118 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 119                 struct buffer_head *bh, handle_t *handle, int err)
 120 {
 121         char nbuf[16];
 122         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 123
 124         if (bh)
 125                 BUFFER_TRACE(bh, "abort");
 126
 127         if (!handle->h_err)
 128                 handle->h_err = err;
 129
 130         if (is_handle_aborted(handle))
 131                 return;
 132
 133         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 134                caller, errstr, err_fn);
 135
 136         journal_abort_handle(handle);
 137 }
 138
 139 /* Deal with the reporting of failure conditions on a filesystem such as
 140  * inconsistencies detected or read IO failures.
 141  *
 142  * On ext2, we can store the error state of the filesystem in the
 143  * superblock.  That is not possible on ext3, because we may have other
 144  * write ordering constraints on the superblock which prevent us from
 145  * writing it out straight away; and given that the journal is about to
 146  * be aborted, we can't rely on the current, or future, transactions to
 147  * write out the superblock safely.
 148  *
 149  * We'll just use the journal_abort() error code to record an error in
 150  * the journal instead.  On recovery, the journal will compain about
 151  * that error until we've noted it down and cleared it.
 152  */
 153
 154 static void ext3_handle_error(struct super_block *sb)
 155 {
 156         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 157
 158         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 159         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 160
 161         if (sb->s_flags & MS_RDONLY)
 162                 return;
 163
 164         if (!test_opt (sb, ERRORS_CONT)) {
 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         if (test_opt (sb, ERRORS_RO)) {
 172                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 173                 sb->s_flags |= MS_RDONLY;
 174         }
 175         ext3_commit_super(sb, es, 1);
 176         if (test_opt(sb, ERRORS_PANIC))
 177                 panic("EXT3-fs (device %s): panic forced after error\n",
 178                         sb->s_id);
 179 }
 180
 181 void ext3_error (struct super_block * sb, const char * function,
 182                  const char * fmt, ...)
 183 {
 184         va_list args;
 185
 186         va_start(args, fmt);
 187         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 188         vprintk(fmt, args);
 189         printk("\n");
 190         va_end(args);
 191
 192         ext3_handle_error(sb);
 193 }
 194
 195 static const char *ext3_decode_error(struct super_block * sb, int errno,
 196                                      char nbuf[16])
 197 {
 198         char *errstr = NULL;
 199
 200         switch (errno) {
 201         case -EIO:
 202                 errstr = "IO failure";
 203                 break;
 204         case -ENOMEM:
 205                 errstr = "Out of memory";
 206                 break;
 207         case -EROFS:
 208                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 209                         errstr = "Journal has aborted";
 210                 else
 211                         errstr = "Readonly filesystem";
 212                 break;
 213         default:
 214                 /* If the caller passed in an extra buffer for unknown
 215                  * errors, textualise them now.  Else we just return
 216                  * NULL. */
 217                 if (nbuf) {
 218                         /* Check for truncated error codes... */
 219                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 220                                 errstr = nbuf;
 221                 }
 222                 break;
 223         }
 224
 225         return errstr;
 226 }
 227
 228 /* __ext3_std_error decodes expected errors from journaling functions
 229  * automatically and invokes the appropriate error response.  */
 230
 231 void __ext3_std_error (struct super_block * sb, const char * function,
 232                        int errno)
 233 {
 234         char nbuf[16];
 235         const char *errstr;
 236
 237         /* Special case: if the error is EROFS, and we're not already
 238          * inside a transaction, then there's really no point in logging
 239          * an error. */
 240         if (errno == -EROFS && journal_current_handle() == NULL &&
 241             (sb->s_flags & MS_RDONLY))
 242                 return;
 243
 244         errstr = ext3_decode_error(sb, errno, nbuf);
 245         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 246                 sb->s_id, function, errstr);
 247
 248         ext3_handle_error(sb);
 249 }
 250
 251 /*
 252  * ext3_abort is a much stronger failure handler than ext3_error.  The
 253  * abort function may be used to deal with unrecoverable failures such
 254  * as journal IO errors or ENOMEM at a critical moment in log management.
 255  *
 256  * We unconditionally force the filesystem into an ABORT|READONLY state,
 257  * unless the error response on the fs has been set to panic in which
 258  * case we take the easy way out and panic immediately.
 259  */
 260
 261 void ext3_abort (struct super_block * sb, const char * function,
 262                  const char * fmt, ...)
 263 {
 264         va_list args;
 265
 266         printk (KERN_CRIT "ext3_abort called.\n");
 267
 268         va_start(args, fmt);
 269         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 270         vprintk(fmt, args);
 271         printk("\n");
 272         va_end(args);
 273
 274         if (test_opt(sb, ERRORS_PANIC))
 275                 panic("EXT3-fs panic from previous error\n");
 276
 277         if (sb->s_flags & MS_RDONLY)
 278                 return;
 279
 280         printk(KERN_CRIT "Remounting filesystem read-only\n");
 281         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 282         sb->s_flags |= MS_RDONLY;
 283         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 284         if (EXT3_SB(sb)->s_journal)
 285                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 286 }
 287
 288 void ext3_warning (struct super_block * sb, const char * function,
 289                    const char * fmt, ...)
 290 {
 291         va_list args;
 292
 293         va_start(args, fmt);
 294         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 295                sb->s_id, function);
 296         vprintk(fmt, args);
 297         printk("\n");
 298         va_end(args);
 299 }
 300
 301 void ext3_update_dynamic_rev(struct super_block *sb)
 302 {
 303         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 304
 305         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 306                 return;
 307
 308         ext3_warning(sb, __func__,
 309                      "updating to rev %d because of new feature flag, "
 310                      "running e2fsck is recommended",
 311                      EXT3_DYNAMIC_REV);
 312
 313         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 314         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 315         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 316         /* leave es->s_feature_*compat flags alone */
 317         /* es->s_uuid will be set by e2fsck if empty */
 318
 319         /*
 320          * The rest of the superblock fields should be zero, and if not it
 321          * means they are likely already in use, so leave them alone.  We
 322          * can leave it up to e2fsck to clean up any inconsistencies there.
 323          */
 324 }
 325
 326 /*
 327  * Open the external journal device
 328  */
 329 static struct block_device *ext3_blkdev_get(dev_t dev)
 330 {
 331         struct block_device *bdev;
 332         char b[BDEVNAME_SIZE];
 333
 334         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 335         if (IS_ERR(bdev))
 336                 goto fail;
 337         return bdev;
 338
 339 fail:
 340         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 341                         __bdevname(dev, b), PTR_ERR(bdev));
 342         return NULL;
 343 }
 344
 345 /*
 346  * Release the journal device
 347  */
 348 static int ext3_blkdev_put(struct block_device *bdev)
 349 {
 350         bd_release(bdev);
 351         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 352 }
 353
 354 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 355 {
 356         struct block_device *bdev;
 357         int ret = -ENODEV;
 358
 359         bdev = sbi->journal_bdev;
 360         if (bdev) {
 361                 ret = ext3_blkdev_put(bdev);
 362                 sbi->journal_bdev = NULL;
 363         }
 364         return ret;
 365 }
 366
 367 static inline struct inode *orphan_list_entry(struct list_head *l)
 368 {
 369         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 370 }
 371
 372 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 373 {
 374         struct list_head *l;
 375
 376         printk(KERN_ERR "sb orphan head is %d\n",
 377                le32_to_cpu(sbi->s_es->s_last_orphan));
 378
 379         printk(KERN_ERR "sb_info orphan list:\n");
 380         list_for_each(l, &sbi->s_orphan) {
 381                 struct inode *inode = orphan_list_entry(l);
 382                 printk(KERN_ERR "  "
 383                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 384                        inode->i_sb->s_id, inode->i_ino, inode,
 385                        inode->i_mode, inode->i_nlink,
 386                        NEXT_ORPHAN(inode));
 387         }
 388 }
 389
 390 static void ext3_put_super (struct super_block * sb)
 391 {
 392         struct ext3_sb_info *sbi = EXT3_SB(sb);
 393         struct ext3_super_block *es = sbi->s_es;
 394         int i, err;
 395
 396         ext3_xattr_put_super(sb);
 397         err = journal_destroy(sbi->s_journal);
 398         sbi->s_journal = NULL;
 399         if (err < 0)
 400                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 401
 402         if (!(sb->s_flags & MS_RDONLY)) {
 403                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 404                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 405                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 406                 mark_buffer_dirty(sbi->s_sbh);
 407                 ext3_commit_super(sb, es, 1);
 408         }
 409
 410         for (i = 0; i < sbi->s_gdb_count; i++)
 411                 brelse(sbi->s_group_desc[i]);
 412         kfree(sbi->s_group_desc);
 413         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 414         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 415         percpu_counter_destroy(&sbi->s_dirs_counter);
 416         brelse(sbi->s_sbh);
 417 #ifdef CONFIG_QUOTA
 418         for (i = 0; i < MAXQUOTAS; i++)
 419                 kfree(sbi->s_qf_names[i]);
 420 #endif
 421
 422         /* Debugging code just in case the in-memory inode orphan list
 423          * isn't empty.  The on-disk one can be non-empty if we've
 424          * detected an error and taken the fs readonly, but the
 425          * in-memory list had better be clean by this point. */
 426         if (!list_empty(&sbi->s_orphan))
 427                 dump_orphan_list(sb, sbi);
 428         J_ASSERT(list_empty(&sbi->s_orphan));
 429
 430         invalidate_bdev(sb->s_bdev);
 431         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 432                 /*
 433                  * Invalidate the journal device's buffers.  We don't want them
 434                  * floating about in memory - the physical journal device may
 435                  * hotswapped, and it breaks the `ro-after' testing code.
 436                  */
 437                 sync_blockdev(sbi->journal_bdev);
 438                 invalidate_bdev(sbi->journal_bdev);
 439                 ext3_blkdev_remove(sbi);
 440         }
 441         sb->s_fs_info = NULL;
 442         kfree(sbi->s_blockgroup_lock);
 443         kfree(sbi);
 444         return;
 445 }
 446
 447 static struct kmem_cache *ext3_inode_cachep;
 448
 449 /*
 450  * Called inside transaction, so use GFP_NOFS
 451  */
 452 static struct inode *ext3_alloc_inode(struct super_block *sb)
 453 {
 454         struct ext3_inode_info *ei;
 455
 456         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 457         if (!ei)
 458                 return NULL;
 459 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 460         ei->i_acl = EXT3_ACL_NOT_CACHED;
 461         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 462 #endif
 463         ei->i_block_alloc_info = NULL;
 464         ei->vfs_inode.i_version = 1;
 465         return &ei->vfs_inode;
 466 }
 467
 468 static void ext3_destroy_inode(struct inode *inode)
 469 {
 470         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 471                 printk("EXT3 Inode %p: orphan list check failed!\n",
 472                         EXT3_I(inode));
 473                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 474                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 475                                 false);
 476                 dump_stack();
 477         }
 478         remove_extent_mappings(&EXT3_I(inode)->extent_tree, 0, (u64) -1);
 479         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 480 }
 481
 482 static void init_once(void *foo)
 483 {
 484         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 485
 486         INIT_LIST_HEAD(&ei->i_orphan);
 487 #ifdef CONFIG_EXT3_FS_XATTR
 488         init_rwsem(&ei->xattr_sem);
 489 #endif
 490         mutex_init(&ei->truncate_mutex);
 491         extent_map_tree_init(&ei->extent_tree);
 492         inode_init_once(&ei->vfs_inode);
 493 }
 494
 495 static int init_inodecache(void)
 496 {
 497         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 498                                              sizeof(struct ext3_inode_info),
 499                                              0, (SLAB_RECLAIM_ACCOUNT|
 500                                                 SLAB_MEM_SPREAD),
 501                                              init_once);
 502         if (ext3_inode_cachep == NULL)
 503                 return -ENOMEM;
 504         return 0;
 505 }
 506
 507 static void destroy_inodecache(void)
 508 {
 509         kmem_cache_destroy(ext3_inode_cachep);
 510 }
 511
 512 static void ext3_clear_inode(struct inode *inode)
 513 {
 514         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 515 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 516         if (EXT3_I(inode)->i_acl &&
 517                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 518                 posix_acl_release(EXT3_I(inode)->i_acl);
 519                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 520         }
 521         if (EXT3_I(inode)->i_default_acl &&
 522                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 523                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 524                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 525         }
 526 #endif
 527         ext3_discard_reservation(inode);
 528         EXT3_I(inode)->i_block_alloc_info = NULL;
 529         if (unlikely(rsv))
 530                 kfree(rsv);
 531 }
 532
 533 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 534 {
 535 #if defined(CONFIG_QUOTA)
 536         struct ext3_sb_info *sbi = EXT3_SB(sb);
 537
 538         if (sbi->s_jquota_fmt)
 539                 seq_printf(seq, ",jqfmt=%s",
 540                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 541
 542         if (sbi->s_qf_names[USRQUOTA])
 543                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 544
 545         if (sbi->s_qf_names[GRPQUOTA])
 546                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 547
 548         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 549                 seq_puts(seq, ",usrquota");
 550
 551         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 552                 seq_puts(seq, ",grpquota");
 553 #endif
 554 }
 555
 556 /*
 557  * Show an option if
 558  *  - it's set to a non-default value OR
 559  *  - if the per-sb default is different from the global default
 560  */
 561 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 562 {
 563         struct super_block *sb = vfs->mnt_sb;
 564         struct ext3_sb_info *sbi = EXT3_SB(sb);
 565         struct ext3_super_block *es = sbi->s_es;
 566         unsigned long def_mount_opts;
 567
 568         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 569
 570         if (sbi->s_sb_block != 1)
 571                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 572         if (test_opt(sb, MINIX_DF))
 573                 seq_puts(seq, ",minixdf");
 574         if (test_opt(sb, GRPID))
 575                 seq_puts(seq, ",grpid");
 576         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 577                 seq_puts(seq, ",nogrpid");
 578         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 579             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 580                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 581         }
 582         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 583             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 584                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 585         }
 586         if (test_opt(sb, ERRORS_RO)) {
 587                 int def_errors = le16_to_cpu(es->s_errors);
 588
 589                 if (def_errors == EXT3_ERRORS_PANIC ||
 590                     def_errors == EXT3_ERRORS_CONTINUE) {
 591                         seq_puts(seq, ",errors=remount-ro");
 592                 }
 593         }
 594         if (test_opt(sb, ERRORS_CONT))
 595                 seq_puts(seq, ",errors=continue");
 596         if (test_opt(sb, ERRORS_PANIC))
 597                 seq_puts(seq, ",errors=panic");
 598         if (test_opt(sb, NO_UID32))
 599                 seq_puts(seq, ",nouid32");
 600         if (test_opt(sb, DEBUG))
 601                 seq_puts(seq, ",debug");
 602         if (test_opt(sb, OLDALLOC))
 603                 seq_puts(seq, ",oldalloc");
 604 #ifdef CONFIG_EXT3_FS_XATTR
 605         if (test_opt(sb, XATTR_USER))
 606                 seq_puts(seq, ",user_xattr");
 607         if (!test_opt(sb, XATTR_USER) &&
 608             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 609                 seq_puts(seq, ",nouser_xattr");
 610         }
 611 #endif
 612 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 613         if (test_opt(sb, POSIX_ACL))
 614                 seq_puts(seq, ",acl");
 615         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 616                 seq_puts(seq, ",noacl");
 617 #endif
 618         if (!test_opt(sb, RESERVATION))
 619                 seq_puts(seq, ",noreservation");
 620         if (sbi->s_commit_interval) {
 621                 seq_printf(seq, ",commit=%u",
 622                            (unsigned) (sbi->s_commit_interval / HZ));
 623         }
 624         if (test_opt(sb, BARRIER))
 625                 seq_puts(seq, ",barrier=1");
 626         if (test_opt(sb, NOBH))
 627                 seq_puts(seq, ",nobh");
 628
 629         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 630                 seq_puts(seq, ",data=journal");
 631         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 632                 seq_puts(seq, ",data=ordered");
 633         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 634                 seq_puts(seq, ",data=writeback");
 635
 636         if (test_opt(sb, DATA_ERR_ABORT))
 637                 seq_puts(seq, ",data_err=abort");
 638
 639         ext3_show_quota_options(seq, sb);
 640
 641         return 0;
 642 }
 643
 644
 645 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 646                 u64 ino, u32 generation)
 647 {
 648         struct inode *inode;
 649
 650         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 651                 return ERR_PTR(-ESTALE);
 652         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 653                 return ERR_PTR(-ESTALE);
 654
 655         /* iget isn't really right if the inode is currently unallocated!!
 656          *
 657          * ext3_read_inode will return a bad_inode if the inode had been
 658          * deleted, so we should be safe.
 659          *
 660          * Currently we don't know the generation for parent directory, so
 661          * a generation of 0 means "accept any"
 662          */
 663         inode = ext3_iget(sb, ino);
 664         if (IS_ERR(inode))
 665                 return ERR_CAST(inode);
 666         if (generation && inode->i_generation != generation) {
 667                 iput(inode);
 668                 return ERR_PTR(-ESTALE);
 669         }
 670
 671         return inode;
 672 }
 673
 674 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 675                 int fh_len, int fh_type)
 676 {
 677         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 678                                     ext3_nfs_get_inode);
 679 }
 680
 681 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 682                 int fh_len, int fh_type)
 683 {
 684         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 685                                     ext3_nfs_get_inode);
 686 }
 687
 688 /*
 689  * Try to release metadata pages (indirect blocks, directories) which are
 690  * mapped via the block device.  Since these pages could have journal heads
 691  * which would prevent try_to_free_buffers() from freeing them, we must use
 692  * jbd layer's try_to_free_buffers() function to release them.
 693  */
 694 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 695                                  gfp_t wait)
 696 {
 697         journal_t *journal = EXT3_SB(sb)->s_journal;
 698
 699         WARN_ON(PageChecked(page));
 700         if (!page_has_buffers(page))
 701                 return 0;
 702         if (journal)
 703                 return journal_try_to_free_buffers(journal, page,
 704                                                    wait & ~__GFP_WAIT);
 705         return try_to_free_buffers(page);
 706 }
 707
 708 #ifdef CONFIG_QUOTA
 709 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 710 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 711
 712 static int ext3_dquot_initialize(struct inode *inode, int type);
 713 static int ext3_dquot_drop(struct inode *inode);
 714 static int ext3_write_dquot(struct dquot *dquot);
 715 static int ext3_acquire_dquot(struct dquot *dquot);
 716 static int ext3_release_dquot(struct dquot *dquot);
 717 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 718 static int ext3_write_info(struct super_block *sb, int type);
 719 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 720                                 char *path, int remount);
 721 static int ext3_quota_on_mount(struct super_block *sb, int type);
 722 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 723                                size_t len, loff_t off);
 724 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 725                                 const char *data, size_t len, loff_t off);
 726
 727 static struct dquot_operations ext3_quota_operations = {
 728         .initialize     = ext3_dquot_initialize,
 729         .drop           = ext3_dquot_drop,
 730         .alloc_space    = dquot_alloc_space,
 731         .alloc_inode    = dquot_alloc_inode,
 732         .free_space     = dquot_free_space,
 733         .free_inode     = dquot_free_inode,
 734         .transfer       = dquot_transfer,
 735         .write_dquot    = ext3_write_dquot,
 736         .acquire_dquot  = ext3_acquire_dquot,
 737         .release_dquot  = ext3_release_dquot,
 738         .mark_dirty     = ext3_mark_dquot_dirty,
 739         .write_info     = ext3_write_info,
 740         .alloc_dquot    = dquot_alloc,
 741         .destroy_dquot  = dquot_destroy,
 742 };
 743
 744 static struct quotactl_ops ext3_qctl_operations = {
 745         .quota_on       = ext3_quota_on,
 746         .quota_off      = vfs_quota_off,
 747         .quota_sync     = vfs_quota_sync,
 748         .get_info       = vfs_get_dqinfo,
 749         .set_info       = vfs_set_dqinfo,
 750         .get_dqblk      = vfs_get_dqblk,
 751         .set_dqblk      = vfs_set_dqblk
 752 };
 753 #endif
 754
 755 static const struct super_operations ext3_sops = {
 756         .alloc_inode    = ext3_alloc_inode,
 757         .destroy_inode  = ext3_destroy_inode,
 758         .write_inode    = ext3_write_inode,
 759         .dirty_inode    = ext3_dirty_inode,
 760         .delete_inode   = ext3_delete_inode,
 761         .put_super      = ext3_put_super,
 762         .write_super    = ext3_write_super,
 763         .sync_fs        = ext3_sync_fs,
 764         .freeze_fs      = ext3_freeze,
 765         .unfreeze_fs    = ext3_unfreeze,
 766         .statfs         = ext3_statfs,
 767         .remount_fs     = ext3_remount,
 768         .clear_inode    = ext3_clear_inode,
 769         .show_options   = ext3_show_options,
 770 #ifdef CONFIG_QUOTA
 771         .quota_read     = ext3_quota_read,
 772         .quota_write    = ext3_quota_write,
 773 #endif
 774         .bdev_try_to_free_page = bdev_try_to_free_page,
 775 };
 776
 777 static const struct export_operations ext3_export_ops = {
 778         .fh_to_dentry = ext3_fh_to_dentry,
 779         .fh_to_parent = ext3_fh_to_parent,
 780         .get_parent = ext3_get_parent,
 781 };
 782
 783 enum {
 784         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 785         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 786         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 787         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 788         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 789         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 790         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 791         Opt_data_err_abort, Opt_data_err_ignore,
 792         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 793         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 794         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 795         Opt_grpquota
 796 };
 797
 798 static const match_table_t tokens = {
 799         {Opt_bsd_df, "bsddf"},
 800         {Opt_minix_df, "minixdf"},
 801         {Opt_grpid, "grpid"},
 802         {Opt_grpid, "bsdgroups"},
 803         {Opt_nogrpid, "nogrpid"},
 804         {Opt_nogrpid, "sysvgroups"},
 805         {Opt_resgid, "resgid=%u"},
 806         {Opt_resuid, "resuid=%u"},
 807         {Opt_sb, "sb=%u"},
 808         {Opt_err_cont, "errors=continue"},
 809         {Opt_err_panic, "errors=panic"},
 810         {Opt_err_ro, "errors=remount-ro"},
 811         {Opt_nouid32, "nouid32"},
 812         {Opt_nocheck, "nocheck"},
 813         {Opt_nocheck, "check=none"},
 814         {Opt_debug, "debug"},
 815         {Opt_oldalloc, "oldalloc"},
 816         {Opt_orlov, "orlov"},
 817         {Opt_user_xattr, "user_xattr"},
 818         {Opt_nouser_xattr, "nouser_xattr"},
 819         {Opt_acl, "acl"},
 820         {Opt_noacl, "noacl"},
 821         {Opt_reservation, "reservation"},
 822         {Opt_noreservation, "noreservation"},
 823         {Opt_noload, "noload"},
 824         {Opt_nobh, "nobh"},
 825         {Opt_bh, "bh"},
 826         {Opt_commit, "commit=%u"},
 827         {Opt_journal_update, "journal=update"},
 828         {Opt_journal_inum, "journal=%u"},
 829         {Opt_journal_dev, "journal_dev=%u"},
 830         {Opt_abort, "abort"},
 831         {Opt_data_journal, "data=journal"},
 832         {Opt_data_ordered, "data=ordered"},
 833         {Opt_data_writeback, "data=writeback"},
 834         {Opt_data_err_abort, "data_err=abort"},
 835         {Opt_data_err_ignore, "data_err=ignore"},
 836         {Opt_offusrjquota, "usrjquota="},
 837         {Opt_usrjquota, "usrjquota=%s"},
 838         {Opt_offgrpjquota, "grpjquota="},
 839         {Opt_grpjquota, "grpjquota=%s"},
 840         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 841         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 842         {Opt_grpquota, "grpquota"},
 843         {Opt_noquota, "noquota"},
 844         {Opt_quota, "quota"},
 845         {Opt_usrquota, "usrquota"},
 846         {Opt_barrier, "barrier=%u"},
 847         {Opt_resize, "resize"},
 848         {Opt_err, NULL},
 849 };
 850
 851 static ext3_fsblk_t get_sb_block(void **data)
 852 {
 853         ext3_fsblk_t    sb_block;
 854         char            *options = (char *) *data;
 855
 856         if (!options || strncmp(options, "sb=", 3) != 0)
 857                 return 1;       /* Default location */
 858         options += 3;
 859         /*todo: use simple_strtoll with >32bit ext3 */
 860         sb_block = simple_strtoul(options, &options, 0);
 861         if (*options && *options != ',') {
 862                 printk("EXT3-fs: Invalid sb specification: %s\n",
 863                        (char *) *data);
 864                 return 1;
 865         }
 866         if (*options == ',')
 867                 options++;
 868         *data = (void *) options;
 869         return sb_block;
 870 }
 871
 872 static int parse_options (char *options, struct super_block *sb,
 873                           unsigned int *inum, unsigned long *journal_devnum,
 874                           ext3_fsblk_t *n_blocks_count, int is_remount)
 875 {
 876         struct ext3_sb_info *sbi = EXT3_SB(sb);
 877         char * p;
 878         substring_t args[MAX_OPT_ARGS];
 879         int data_opt = 0;
 880         int option;
 881 #ifdef CONFIG_QUOTA
 882         int qtype, qfmt;
 883         char *qname;
 884 #endif
 885
 886         if (!options)
 887                 return 1;
 888
 889         while ((p = strsep (&options, ",")) != NULL) {
 890                 int token;
 891                 if (!*p)
 892                         continue;
 893
 894                 token = match_token(p, tokens, args);
 895                 switch (token) {
 896                 case Opt_bsd_df:
 897                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 898                         break;
 899                 case Opt_minix_df:
 900                         set_opt (sbi->s_mount_opt, MINIX_DF);
 901                         break;
 902                 case Opt_grpid:
 903                         set_opt (sbi->s_mount_opt, GRPID);
 904                         break;
 905                 case Opt_nogrpid:
 906                         clear_opt (sbi->s_mount_opt, GRPID);
 907                         break;
 908                 case Opt_resuid:
 909                         if (match_int(&args[0], &option))
 910                                 return 0;
 911                         sbi->s_resuid = option;
 912                         break;
 913                 case Opt_resgid:
 914                         if (match_int(&args[0], &option))
 915                                 return 0;
 916                         sbi->s_resgid = option;
 917                         break;
 918                 case Opt_sb:
 919                         /* handled by get_sb_block() instead of here */
 920                         /* *sb_block = match_int(&args[0]); */
 921                         break;
 922                 case Opt_err_panic:
 923                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 924                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 925                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 926                         break;
 927                 case Opt_err_ro:
 928                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 929                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 930                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 931                         break;
 932                 case Opt_err_cont:
 933                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 934                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 935                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 936                         break;
 937                 case Opt_nouid32:
 938                         set_opt (sbi->s_mount_opt, NO_UID32);
 939                         break;
 940                 case Opt_nocheck:
 941                         clear_opt (sbi->s_mount_opt, CHECK);
 942                         break;
 943                 case Opt_debug:
 944                         set_opt (sbi->s_mount_opt, DEBUG);
 945                         break;
 946                 case Opt_oldalloc:
 947                         set_opt (sbi->s_mount_opt, OLDALLOC);
 948                         break;
 949                 case Opt_orlov:
 950                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 951                         break;
 952 #ifdef CONFIG_EXT3_FS_XATTR
 953                 case Opt_user_xattr:
 954                         set_opt (sbi->s_mount_opt, XATTR_USER);
 955                         break;
 956                 case Opt_nouser_xattr:
 957                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 958                         break;
 959 #else
 960                 case Opt_user_xattr:
 961                 case Opt_nouser_xattr:
 962                         printk("EXT3 (no)user_xattr options not supported\n");
 963                         break;
 964 #endif
 965 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 966                 case Opt_acl:
 967                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 968                         break;
 969                 case Opt_noacl:
 970                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 971                         break;
 972 #else
 973                 case Opt_acl:
 974                 case Opt_noacl:
 975                         printk("EXT3 (no)acl options not supported\n");
 976                         break;
 977 #endif
 978                 case Opt_reservation:
 979                         set_opt(sbi->s_mount_opt, RESERVATION);
 980                         break;
 981                 case Opt_noreservation:
 982                         clear_opt(sbi->s_mount_opt, RESERVATION);
 983                         break;
 984                 case Opt_journal_update:
 985                         /* @@@ FIXME */
 986                         /* Eventually we will want to be able to create
 987                            a journal file here.  For now, only allow the
 988                            user to specify an existing inode to be the
 989                            journal file. */
 990                         if (is_remount) {
 991                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 992                                        "journal on remount\n");
 993                                 return 0;
 994                         }
 995                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 996                         break;
 997                 case Opt_journal_inum:
 998                         if (is_remount) {
 999                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1000                                        "journal on remount\n");
1001                                 return 0;
1002                         }
1003                         if (match_int(&args[0], &option))
1004                                 return 0;
1005                         *inum = option;
1006                         break;
1007                 case Opt_journal_dev:
1008                         if (is_remount) {
1009                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1010                                        "journal on remount\n");
1011                                 return 0;
1012                         }
1013                         if (match_int(&args[0], &option))
1014                                 return 0;
1015                         *journal_devnum = option;
1016                         break;
1017                 case Opt_noload:
1018                         set_opt (sbi->s_mount_opt, NOLOAD);
1019                         break;
1020                 case Opt_commit:
1021                         if (match_int(&args[0], &option))
1022                                 return 0;
1023                         if (option < 0)
1024                                 return 0;
1025                         if (option == 0)
1026                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1027                         sbi->s_commit_interval = HZ * option;
1028                         break;
1029                 case Opt_data_journal:
1030                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1031                         goto datacheck;
1032                 case Opt_data_ordered:
1033                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1034                         goto datacheck;
1035                 case Opt_data_writeback:
1036                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1037                 datacheck:
1038                         if (is_remount) {
1039                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1040                                                 != data_opt) {
1041                                         printk(KERN_ERR
1042                                                 "EXT3-fs: cannot change data "
1043                                                 "mode on remount\n");
1044                                         return 0;
1045                                 }
1046                         } else {
1047                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1048                                 sbi->s_mount_opt |= data_opt;
1049                         }
1050                         break;
1051                 case Opt_data_err_abort:
1052                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1053                         break;
1054                 case Opt_data_err_ignore:
1055                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1056                         break;
1057 #ifdef CONFIG_QUOTA
1058                 case Opt_usrjquota:
1059                         qtype = USRQUOTA;
1060                         goto set_qf_name;
1061                 case Opt_grpjquota:
1062                         qtype = GRPQUOTA;
1063 set_qf_name:
1064                         if (sb_any_quota_loaded(sb) &&
1065                             !sbi->s_qf_names[qtype]) {
1066                                 printk(KERN_ERR
1067                                         "EXT3-fs: Cannot change journaled "
1068                                         "quota options when quota turned on.\n");
1069                                 return 0;
1070                         }
1071                         qname = match_strdup(&args[0]);
1072                         if (!qname) {
1073                                 printk(KERN_ERR
1074                                         "EXT3-fs: not enough memory for "
1075                                         "storing quotafile name.\n");
1076                                 return 0;
1077                         }
1078                         if (sbi->s_qf_names[qtype] &&
1079                             strcmp(sbi->s_qf_names[qtype], qname)) {
1080                                 printk(KERN_ERR
1081                                         "EXT3-fs: %s quota file already "
1082                                         "specified.\n", QTYPE2NAME(qtype));
1083                                 kfree(qname);
1084                                 return 0;
1085                         }
1086                         sbi->s_qf_names[qtype] = qname;
1087                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1088                                 printk(KERN_ERR
1089                                         "EXT3-fs: quotafile must be on "
1090                                         "filesystem root.\n");
1091                                 kfree(sbi->s_qf_names[qtype]);
1092                                 sbi->s_qf_names[qtype] = NULL;
1093                                 return 0;
1094                         }
1095                         set_opt(sbi->s_mount_opt, QUOTA);
1096                         break;
1097                 case Opt_offusrjquota:
1098                         qtype = USRQUOTA;
1099                         goto clear_qf_name;
1100                 case Opt_offgrpjquota:
1101                         qtype = GRPQUOTA;
1102 clear_qf_name:
1103                         if (sb_any_quota_loaded(sb) &&
1104                             sbi->s_qf_names[qtype]) {
1105                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1106                                         "journaled quota options when "
1107                                         "quota turned on.\n");
1108                                 return 0;
1109                         }
1110                         /*
1111                          * The space will be released later when all options
1112                          * are confirmed to be correct
1113                          */
1114                         sbi->s_qf_names[qtype] = NULL;
1115                         break;
1116                 case Opt_jqfmt_vfsold:
1117                         qfmt = QFMT_VFS_OLD;
1118                         goto set_qf_format;
1119                 case Opt_jqfmt_vfsv0:
1120                         qfmt = QFMT_VFS_V0;
1121 set_qf_format:
1122                         if (sb_any_quota_loaded(sb) &&
1123                             sbi->s_jquota_fmt != qfmt) {
1124                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1125                                         "journaled quota options when "
1126                                         "quota turned on.\n");
1127                                 return 0;
1128                         }
1129                         sbi->s_jquota_fmt = qfmt;
1130                         break;
1131                 case Opt_quota:
1132                 case Opt_usrquota:
1133                         set_opt(sbi->s_mount_opt, QUOTA);
1134                         set_opt(sbi->s_mount_opt, USRQUOTA);
1135                         break;
1136                 case Opt_grpquota:
1137                         set_opt(sbi->s_mount_opt, QUOTA);
1138                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1139                         break;
1140                 case Opt_noquota:
1141                         if (sb_any_quota_loaded(sb)) {
1142                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1143                                         "options when quota turned on.\n");
1144                                 return 0;
1145                         }
1146                         clear_opt(sbi->s_mount_opt, QUOTA);
1147                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1148                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1149                         break;
1150 #else
1151                 case Opt_quota:
1152                 case Opt_usrquota:
1153                 case Opt_grpquota:
1154                         printk(KERN_ERR
1155                                 "EXT3-fs: quota options not supported.\n");
1156                         break;
1157                 case Opt_usrjquota:
1158                 case Opt_grpjquota:
1159                 case Opt_offusrjquota:
1160                 case Opt_offgrpjquota:
1161                 case Opt_jqfmt_vfsold:
1162                 case Opt_jqfmt_vfsv0:
1163                         printk(KERN_ERR
1164                                 "EXT3-fs: journaled quota options not "
1165                                 "supported.\n");
1166                         break;
1167                 case Opt_noquota:
1168                         break;
1169 #endif
1170                 case Opt_abort:
1171                         set_opt(sbi->s_mount_opt, ABORT);
1172                         break;
1173                 case Opt_barrier:
1174                         if (match_int(&args[0], &option))
1175                                 return 0;
1176                         if (option)
1177                                 set_opt(sbi->s_mount_opt, BARRIER);
1178                         else
1179                                 clear_opt(sbi->s_mount_opt, BARRIER);
1180                         break;
1181                 case Opt_ignore:
1182                         break;
1183                 case Opt_resize:
1184                         if (!is_remount) {
1185                                 printk("EXT3-fs: resize option only available "
1186                                         "for remount\n");
1187                                 return 0;
1188                         }
1189                         if (match_int(&args[0], &option) != 0)
1190                                 return 0;
1191                         *n_blocks_count = option;
1192                         break;
1193                 case Opt_nobh:
1194                         set_opt(sbi->s_mount_opt, NOBH);
1195                         break;
1196                 case Opt_bh:
1197                         clear_opt(sbi->s_mount_opt, NOBH);
1198                         break;
1199                 default:
1200                         printk (KERN_ERR
1201                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1202                                 "or missing value\n", p);
1203                         return 0;
1204                 }
1205         }
1206 #ifdef CONFIG_QUOTA
1207         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1208                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1209                      sbi->s_qf_names[USRQUOTA])
1210                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1211
1212                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1213                      sbi->s_qf_names[GRPQUOTA])
1214                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1215
1216                 if ((sbi->s_qf_names[USRQUOTA] &&
1217                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1218                     (sbi->s_qf_names[GRPQUOTA] &&
1219                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1220                         printk(KERN_ERR "EXT3-fs: old and new quota "
1221                                         "format mixing.\n");
1222                         return 0;
1223                 }
1224
1225                 if (!sbi->s_jquota_fmt) {
1226                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1227                                         "not specified.\n");
1228                         return 0;
1229                 }
1230         } else {
1231                 if (sbi->s_jquota_fmt) {
1232                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1233                                         "specified with no journaling "
1234                                         "enabled.\n");
1235                         return 0;
1236                 }
1237         }
1238 #endif
1239         return 1;
1240 }
1241
1242 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1243                             int read_only)
1244 {
1245         struct ext3_sb_info *sbi = EXT3_SB(sb);
1246         int res = 0;
1247
1248         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1249                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1250                         "forcing read-only mode\n");
1251                 res = MS_RDONLY;
1252         }
1253         if (read_only)
1254                 return res;
1255         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1256                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1257                         "running e2fsck is recommended\n");
1258         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1259                 printk (KERN_WARNING
1260                         "EXT3-fs warning: mounting fs with errors, "
1261                         "running e2fsck is recommended\n");
1262         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1263                  le16_to_cpu(es->s_mnt_count) >=
1264                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1265                 printk (KERN_WARNING
1266                         "EXT3-fs warning: maximal mount count reached, "
1267                         "running e2fsck is recommended\n");
1268         else if (le32_to_cpu(es->s_checkinterval) &&
1269                 (le32_to_cpu(es->s_lastcheck) +
1270                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1271                 printk (KERN_WARNING
1272                         "EXT3-fs warning: checktime reached, "
1273                         "running e2fsck is recommended\n");
1274 #if 0
1275                 /* @@@ We _will_ want to clear the valid bit if we find
1276                    inconsistencies, to force a fsck at reboot.  But for
1277                    a plain journaled filesystem we can keep it set as
1278                    valid forever! :) */
1279         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1280 #endif
1281         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1282                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1283         le16_add_cpu(&es->s_mnt_count, 1);
1284         es->s_mtime = cpu_to_le32(get_seconds());
1285         ext3_update_dynamic_rev(sb);
1286         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1287
1288         ext3_commit_super(sb, es, 1);
1289         if (test_opt(sb, DEBUG))
1290                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1291                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1292                         sb->s_blocksize,
1293                         sbi->s_groups_count,
1294                         EXT3_BLOCKS_PER_GROUP(sb),
1295                         EXT3_INODES_PER_GROUP(sb),
1296                         sbi->s_mount_opt);
1297
1298         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1299         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1300                 char b[BDEVNAME_SIZE];
1301
1302                 printk("external journal on %s\n",
1303                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1304         } else {
1305                 printk("internal journal\n");
1306         }
1307         return res;
1308 }
1309
1310 /* Called at mount-time, super-block is locked */
1311 static int ext3_check_descriptors(struct super_block *sb)
1312 {
1313         struct ext3_sb_info *sbi = EXT3_SB(sb);
1314         int i;
1315
1316         ext3_debug ("Checking group descriptors");
1317
1318         for (i = 0; i < sbi->s_groups_count; i++) {
1319                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1320                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1321                 ext3_fsblk_t last_block;
1322
1323                 if (i == sbi->s_groups_count - 1)
1324                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1325                 else
1326                         last_block = first_block +
1327                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1328
1329                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1330                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1331                 {
1332                         ext3_error (sb, "ext3_check_descriptors",
1333                                     "Block bitmap for group %d"
1334                                     " not in group (block %lu)!",
1335                                     i, (unsigned long)
1336                                         le32_to_cpu(gdp->bg_block_bitmap));
1337                         return 0;
1338                 }
1339                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1340                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1341                 {
1342                         ext3_error (sb, "ext3_check_descriptors",
1343                                     "Inode bitmap for group %d"
1344                                     " not in group (block %lu)!",
1345                                     i, (unsigned long)
1346                                         le32_to_cpu(gdp->bg_inode_bitmap));
1347                         return 0;
1348                 }
1349                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1350                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1351                     last_block)
1352                 {
1353                         ext3_error (sb, "ext3_check_descriptors",
1354                                     "Inode table for group %d"
1355                                     " not in group (block %lu)!",
1356                                     i, (unsigned long)
1357                                         le32_to_cpu(gdp->bg_inode_table));
1358                         return 0;
1359                 }
1360         }
1361
1362         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1363         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1364         return 1;
1365 }
1366
1367
1368 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1369  * the superblock) which were deleted from all directories, but held open by
1370  * a process at the time of a crash.  We walk the list and try to delete these
1371  * inodes at recovery time (only with a read-write filesystem).
1372  *
1373  * In order to keep the orphan inode chain consistent during traversal (in
1374  * case of crash during recovery), we link each inode into the superblock
1375  * orphan list_head and handle it the same way as an inode deletion during
1376  * normal operation (which journals the operations for us).
1377  *
1378  * We only do an iget() and an iput() on each inode, which is very safe if we
1379  * accidentally point at an in-use or already deleted inode.  The worst that
1380  * can happen in this case is that we get a "bit already cleared" message from
1381  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1382  * e2fsck was run on this filesystem, and it must have already done the orphan
1383  * inode cleanup for us, so we can safely abort without any further action.
1384  */
1385 static void ext3_orphan_cleanup (struct super_block * sb,
1386                                  struct ext3_super_block * es)
1387 {
1388         unsigned int s_flags = sb->s_flags;
1389         int nr_orphans = 0, nr_truncates = 0;
1390 #ifdef CONFIG_QUOTA
1391         int i;
1392 #endif
1393         if (!es->s_last_orphan) {
1394                 jbd_debug(4, "no orphan inodes to clean up\n");
1395                 return;
1396         }
1397
1398         if (bdev_read_only(sb->s_bdev)) {
1399                 printk(KERN_ERR "EXT3-fs: write access "
1400                         "unavailable, skipping orphan cleanup.\n");
1401                 return;
1402         }
1403
1404         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1405                 if (es->s_last_orphan)
1406                         jbd_debug(1, "Errors on filesystem, "
1407                                   "clearing orphan list.\n");
1408                 es->s_last_orphan = 0;
1409                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1410                 return;
1411         }
1412
1413         if (s_flags & MS_RDONLY) {
1414                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1415                        sb->s_id);
1416                 sb->s_flags &= ~MS_RDONLY;
1417         }
1418 #ifdef CONFIG_QUOTA
1419         /* Needed for iput() to work correctly and not trash data */
1420         sb->s_flags |= MS_ACTIVE;
1421         /* Turn on quotas so that they are updated correctly */
1422         for (i = 0; i < MAXQUOTAS; i++) {
1423                 if (EXT3_SB(sb)->s_qf_names[i]) {
1424                         int ret = ext3_quota_on_mount(sb, i);
1425                         if (ret < 0)
1426                                 printk(KERN_ERR
1427                                         "EXT3-fs: Cannot turn on journaled "
1428                                         "quota: error %d\n", ret);
1429                 }
1430         }
1431 #endif
1432
1433         while (es->s_last_orphan) {
1434                 struct inode *inode;
1435
1436                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1437                 if (IS_ERR(inode)) {
1438                         es->s_last_orphan = 0;
1439                         break;
1440                 }
1441
1442                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1443                 DQUOT_INIT(inode);
1444                 if (inode->i_nlink) {
1445                         printk(KERN_DEBUG
1446                                 "%s: truncating inode %lu to %Ld bytes\n",
1447                                 __func__, inode->i_ino, inode->i_size);
1448                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1449                                   inode->i_ino, inode->i_size);
1450                         ext3_truncate(inode);
1451                         nr_truncates++;
1452                 } else {
1453                         printk(KERN_DEBUG
1454                                 "%s: deleting unreferenced inode %lu\n",
1455                                 __func__, inode->i_ino);
1456                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1457                                   inode->i_ino);
1458                         nr_orphans++;
1459                 }
1460                 iput(inode);  /* The delete magic happens here! */
1461         }
1462
1463 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1464
1465         if (nr_orphans)
1466                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1467                        sb->s_id, PLURAL(nr_orphans));
1468         if (nr_truncates)
1469                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1470                        sb->s_id, PLURAL(nr_truncates));
1471 #ifdef CONFIG_QUOTA
1472         /* Turn quotas off */
1473         for (i = 0; i < MAXQUOTAS; i++) {
1474                 if (sb_dqopt(sb)->files[i])
1475                         vfs_quota_off(sb, i, 0);
1476         }
1477 #endif
1478         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1479 }
1480
1481 /*
1482  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1483  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1484  * We need to be 1 filesystem block less than the 2^32 sector limit.
1485  */
1486 static loff_t ext3_max_size(int bits)
1487 {
1488         loff_t res = EXT3_NDIR_BLOCKS;
1489         int meta_blocks;
1490         loff_t upper_limit;
1491
1492         /* This is calculated to be the largest file size for a
1493          * dense, file such that the total number of
1494          * sectors in the file, including data and all indirect blocks,
1495          * does not exceed 2^32 -1
1496          * __u32 i_blocks representing the total number of
1497          * 512 bytes blocks of the file
1498          */
1499         upper_limit = (1LL << 32) - 1;
1500
1501         /* total blocks in file system block size */
1502         upper_limit >>= (bits - 9);
1503
1504
1505         /* indirect blocks */
1506         meta_blocks = 1;
1507         /* double indirect blocks */
1508         meta_blocks += 1 + (1LL << (bits-2));
1509         /* tripple indirect blocks */
1510         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1511
1512         upper_limit -= meta_blocks;
1513         upper_limit <<= bits;
1514
1515         res += 1LL << (bits-2);
1516         res += 1LL << (2*(bits-2));
1517         res += 1LL << (3*(bits-2));
1518         res <<= bits;
1519         if (res > upper_limit)
1520                 res = upper_limit;
1521
1522         if (res > MAX_LFS_FILESIZE)
1523                 res = MAX_LFS_FILESIZE;
1524
1525         return res;
1526 }
1527
1528 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1529                                     ext3_fsblk_t logic_sb_block,
1530                                     int nr)
1531 {
1532         struct ext3_sb_info *sbi = EXT3_SB(sb);
1533         unsigned long bg, first_meta_bg;
1534         int has_super = 0;
1535
1536         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1537
1538         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1539             nr < first_meta_bg)
1540                 return (logic_sb_block + nr + 1);
1541         bg = sbi->s_desc_per_block * nr;
1542         if (ext3_bg_has_super(sb, bg))
1543                 has_super = 1;
1544         return (has_super + ext3_group_first_block_no(sb, bg));
1545 }
1546
1547
1548 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1549 {
1550         struct buffer_head * bh;
1551         struct ext3_super_block *es = NULL;
1552         struct ext3_sb_info *sbi;
1553         ext3_fsblk_t block;
1554         ext3_fsblk_t sb_block = get_sb_block(&data);
1555         ext3_fsblk_t logic_sb_block;
1556         unsigned long offset = 0;
1557         unsigned int journal_inum = 0;
1558         unsigned long journal_devnum = 0;
1559         unsigned long def_mount_opts;
1560         struct inode *root;
1561         int blocksize;
1562         int hblock;
1563         int db_count;
1564         int i;
1565         int needs_recovery;
1566         int ret = -EINVAL;
1567         __le32 features;
1568         int err;
1569
1570         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1571         if (!sbi)
1572                 return -ENOMEM;
1573
1574         sbi->s_blockgroup_lock =
1575                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1576         if (!sbi->s_blockgroup_lock) {
1577                 kfree(sbi);
1578                 return -ENOMEM;
1579         }
1580         sb->s_fs_info = sbi;
1581         sbi->s_mount_opt = 0;
1582         sbi->s_resuid = EXT3_DEF_RESUID;
1583         sbi->s_resgid = EXT3_DEF_RESGID;
1584         sbi->s_sb_block = sb_block;
1585
1586         unlock_kernel();
1587
1588         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1589         if (!blocksize) {
1590                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1591                 goto out_fail;
1592         }
1593
1594         /*
1595          * The ext3 superblock will not be buffer aligned for other than 1kB
1596          * block sizes.  We need to calculate the offset from buffer start.
1597          */
1598         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1599                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1600                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1601         } else {
1602                 logic_sb_block = sb_block;
1603         }
1604
1605         if (!(bh = sb_bread(sb, logic_sb_block))) {
1606                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1607                 goto out_fail;
1608         }
1609         /*
1610          * Note: s_es must be initialized as soon as possible because
1611          *       some ext3 macro-instructions depend on its value
1612          */
1613         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1614         sbi->s_es = es;
1615         sb->s_magic = le16_to_cpu(es->s_magic);
1616         if (sb->s_magic != EXT3_SUPER_MAGIC)
1617                 goto cantfind_ext3;
1618
1619         /* Set defaults before we parse the mount options */
1620         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1621         if (def_mount_opts & EXT3_DEFM_DEBUG)
1622                 set_opt(sbi->s_mount_opt, DEBUG);
1623         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1624                 set_opt(sbi->s_mount_opt, GRPID);
1625         if (def_mount_opts & EXT3_DEFM_UID16)
1626                 set_opt(sbi->s_mount_opt, NO_UID32);
1627 #ifdef CONFIG_EXT3_FS_XATTR
1628         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1629                 set_opt(sbi->s_mount_opt, XATTR_USER);
1630 #endif
1631 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1632         if (def_mount_opts & EXT3_DEFM_ACL)
1633                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1634 #endif
1635         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1636                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1637         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1638                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1639         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1640                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1641
1642         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1643                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1644         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1645                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1646         else
1647                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1648
1649         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1650         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1651
1652         set_opt(sbi->s_mount_opt, RESERVATION);
1653
1654         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1655                             NULL, 0))
1656                 goto failed_mount;
1657
1658         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1659                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1660
1661         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1662             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1663              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1664              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1665                 printk(KERN_WARNING
1666                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1667                        "running e2fsck is recommended\n");
1668         /*
1669          * Check feature flags regardless of the revision level, since we
1670          * previously didn't change the revision level when setting the flags,
1671          * so there is a chance incompat flags are set on a rev 0 filesystem.
1672          */
1673         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1674         if (features) {
1675                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1676                        "unsupported optional features (%x).\n",
1677                        sb->s_id, le32_to_cpu(features));
1678                 goto failed_mount;
1679         }
1680         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1681         if (!(sb->s_flags & MS_RDONLY) && features) {
1682                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1683                        "unsupported optional features (%x).\n",
1684                        sb->s_id, le32_to_cpu(features));
1685                 goto failed_mount;
1686         }
1687         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1688
1689         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1690             blocksize > EXT3_MAX_BLOCK_SIZE) {
1691                 printk(KERN_ERR
1692                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1693                        blocksize, sb->s_id);
1694                 goto failed_mount;
1695         }
1696
1697         hblock = bdev_hardsect_size(sb->s_bdev);
1698         if (sb->s_blocksize != blocksize) {
1699                 /*
1700                  * Make sure the blocksize for the filesystem is larger
1701                  * than the hardware sectorsize for the machine.
1702                  */
1703                 if (blocksize < hblock) {
1704                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1705                                "device blocksize %d.\n", blocksize, hblock);
1706                         goto failed_mount;
1707                 }
1708
1709                 brelse (bh);
1710                 if (!sb_set_blocksize(sb, blocksize)) {
1711                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1712                                 blocksize);
1713                         goto out_fail;
1714                 }
1715                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1716                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1717                 bh = sb_bread(sb, logic_sb_block);
1718                 if (!bh) {
1719                         printk(KERN_ERR
1720                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1721                         goto failed_mount;
1722                 }
1723                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1724                 sbi->s_es = es;
1725                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1726                         printk (KERN_ERR
1727                                 "EXT3-fs: Magic mismatch, very weird !\n");
1728                         goto failed_mount;
1729                 }
1730         }
1731
1732         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1733
1734         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1735                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1736                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1737         } else {
1738                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1739                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1740                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1741                     (!is_power_of_2(sbi->s_inode_size)) ||
1742                     (sbi->s_inode_size > blocksize)) {
1743                         printk (KERN_ERR
1744                                 "EXT3-fs: unsupported inode size: %d\n",
1745                                 sbi->s_inode_size);
1746                         goto failed_mount;
1747                 }
1748         }
1749         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1750                                    le32_to_cpu(es->s_log_frag_size);
1751         if (blocksize != sbi->s_frag_size) {
1752                 printk(KERN_ERR
1753                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1754                        sbi->s_frag_size, blocksize);
1755                 goto failed_mount;
1756         }
1757         sbi->s_frags_per_block = 1;
1758         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1759         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1760         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1761         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1762                 goto cantfind_ext3;
1763         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1764         if (sbi->s_inodes_per_block == 0)
1765                 goto cantfind_ext3;
1766         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1767                                         sbi->s_inodes_per_block;
1768         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1769         sbi->s_sbh = bh;
1770         sbi->s_mount_state = le16_to_cpu(es->s_state);
1771         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1772         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1773         for (i=0; i < 4; i++)
1774                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1775         sbi->s_def_hash_version = es->s_def_hash_version;
1776         i = le32_to_cpu(es->s_flags);
1777         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1778                 sbi->s_hash_unsigned = 3;
1779         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1780 #ifdef __CHAR_UNSIGNED__
1781                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1782                 sbi->s_hash_unsigned = 3;
1783 #else
1784                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1785 #endif
1786                 sb->s_dirt = 1;
1787         }
1788
1789         if (sbi->s_blocks_per_group > blocksize * 8) {
1790                 printk (KERN_ERR
1791                         "EXT3-fs: #blocks per group too big: %lu\n",
1792                         sbi->s_blocks_per_group);
1793                 goto failed_mount;
1794         }
1795         if (sbi->s_frags_per_group > blocksize * 8) {
1796                 printk (KERN_ERR
1797                         "EXT3-fs: #fragments per group too big: %lu\n",
1798                         sbi->s_frags_per_group);
1799                 goto failed_mount;
1800         }
1801         if (sbi->s_inodes_per_group > blocksize * 8) {
1802                 printk (KERN_ERR
1803                         "EXT3-fs: #inodes per group too big: %lu\n",
1804                         sbi->s_inodes_per_group);
1805                 goto failed_mount;
1806         }
1807
1808         if (le32_to_cpu(es->s_blocks_count) >
1809                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1810                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1811                         " too large to mount safely\n", sb->s_id);
1812                 if (sizeof(sector_t) < 8)
1813                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1814                                         "enabled\n");
1815                 goto failed_mount;
1816         }
1817
1818         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1819                 goto cantfind_ext3;
1820         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1821                                le32_to_cpu(es->s_first_data_block) - 1)
1822                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1823         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1824                    EXT3_DESC_PER_BLOCK(sb);
1825         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1826                                     GFP_KERNEL);
1827         if (sbi->s_group_desc == NULL) {
1828                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1829                 goto failed_mount;
1830         }
1831
1832         bgl_lock_init(sbi->s_blockgroup_lock);
1833
1834         for (i = 0; i < db_count; i++) {
1835                 block = descriptor_loc(sb, logic_sb_block, i);
1836                 sbi->s_group_desc[i] = sb_bread(sb, block);
1837                 if (!sbi->s_group_desc[i]) {
1838                         printk (KERN_ERR "EXT3-fs: "
1839                                 "can't read group descriptor %d\n", i);
1840                         db_count = i;
1841                         goto failed_mount2;
1842                 }
1843         }
1844         if (!ext3_check_descriptors (sb)) {
1845                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1846                 goto failed_mount2;
1847         }
1848         sbi->s_gdb_count = db_count;
1849         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1850         spin_lock_init(&sbi->s_next_gen_lock);
1851
1852         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1853                         ext3_count_free_blocks(sb));
1854         if (!err) {
1855                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1856                                 ext3_count_free_inodes(sb));
1857         }
1858         if (!err) {
1859                 err = percpu_counter_init(&sbi->s_dirs_counter,
1860                                 ext3_count_dirs(sb));
1861         }
1862         if (err) {
1863                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1864                 goto failed_mount3;
1865         }
1866
1867         /* per fileystem reservation list head & lock */
1868         spin_lock_init(&sbi->s_rsv_window_lock);
1869         sbi->s_rsv_window_root = RB_ROOT;
1870         /* Add a single, static dummy reservation to the start of the
1871          * reservation window list --- it gives us a placeholder for
1872          * append-at-start-of-list which makes the allocation logic
1873          * _much_ simpler. */
1874         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1875         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1876         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1877         sbi->s_rsv_window_head.rsv_goal_size = 0;
1878         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1879
1880         /*
1881          * set up enough so that it can read an inode
1882          */
1883         sb->s_op = &ext3_sops;
1884         sb->s_export_op = &ext3_export_ops;
1885         sb->s_xattr = ext3_xattr_handlers;
1886 #ifdef CONFIG_QUOTA
1887         sb->s_qcop = &ext3_qctl_operations;
1888         sb->dq_op = &ext3_quota_operations;
1889 #endif
1890         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1891
1892         sb->s_root = NULL;
1893
1894         needs_recovery = (es->s_last_orphan != 0 ||
1895                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1896                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1897
1898         /*
1899          * The first inode we look at is the journal inode.  Don't try
1900          * root first: it may be modified in the journal!
1901          */
1902         if (!test_opt(sb, NOLOAD) &&
1903             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1904                 if (ext3_load_journal(sb, es, journal_devnum))
1905                         goto failed_mount3;
1906         } else if (journal_inum) {
1907                 if (ext3_create_journal(sb, es, journal_inum))
1908                         goto failed_mount3;
1909         } else {
1910                 if (!silent)
1911                         printk (KERN_ERR
1912                                 "ext3: No journal on filesystem on %s\n",
1913                                 sb->s_id);
1914                 goto failed_mount3;
1915         }
1916
1917         /* We have now updated the journal if required, so we can
1918          * validate the data journaling mode. */
1919         switch (test_opt(sb, DATA_FLAGS)) {
1920         case 0:
1921                 /* No mode set, assume a default based on the journal
1922                    capabilities: ORDERED_DATA if the journal can
1923                    cope, else JOURNAL_DATA */
1924                 if (journal_check_available_features
1925                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1926                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1927                 else
1928                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1929                 break;
1930
1931         case EXT3_MOUNT_ORDERED_DATA:
1932         case EXT3_MOUNT_WRITEBACK_DATA:
1933                 if (!journal_check_available_features
1934                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1935                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1936                                "requested data journaling mode\n");
1937                         goto failed_mount4;
1938                 }
1939         default:
1940                 break;
1941         }
1942
1943         if (test_opt(sb, NOBH)) {
1944                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1945                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1946                                 "its supported only with writeback mode\n");
1947                         clear_opt(sbi->s_mount_opt, NOBH);
1948                 }
1949         }
1950         /*
1951          * The journal_load will have done any necessary log recovery,
1952          * so we can safely mount the rest of the filesystem now.
1953          */
1954
1955         root = ext3_iget(sb, EXT3_ROOT_INO);
1956         if (IS_ERR(root)) {
1957                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1958                 ret = PTR_ERR(root);
1959                 goto failed_mount4;
1960         }
1961         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1962                 iput(root);
1963                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1964                 goto failed_mount4;
1965         }
1966         sb->s_root = d_alloc_root(root);
1967         if (!sb->s_root) {
1968                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1969                 iput(root);
1970                 ret = -ENOMEM;
1971                 goto failed_mount4;
1972         }
1973
1974         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1975         /*
1976          * akpm: core read_super() calls in here with the superblock locked.
1977          * That deadlocks, because orphan cleanup needs to lock the superblock
1978          * in numerous places.  Here we just pop the lock - it's relatively
1979          * harmless, because we are now ready to accept write_super() requests,
1980          * and aviro says that's the only reason for hanging onto the
1981          * superblock lock.
1982          */
1983         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1984         ext3_orphan_cleanup(sb, es);
1985         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1986         if (needs_recovery)
1987                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1988         ext3_mark_recovery_complete(sb, es);
1989         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1990                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1991                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1992                 "writeback");
1993
1994         lock_kernel();
1995         return 0;
1996
1997 cantfind_ext3:
1998         if (!silent)
1999                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
2000                        sb->s_id);
2001         goto failed_mount;
2002
2003 failed_mount4:
2004         journal_destroy(sbi->s_journal);
2005 failed_mount3:
2006         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2007         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2008         percpu_counter_destroy(&sbi->s_dirs_counter);
2009 failed_mount2:
2010         for (i = 0; i < db_count; i++)
2011                 brelse(sbi->s_group_desc[i]);
2012         kfree(sbi->s_group_desc);
2013 failed_mount:
2014 #ifdef CONFIG_QUOTA
2015         for (i = 0; i < MAXQUOTAS; i++)
2016                 kfree(sbi->s_qf_names[i]);
2017 #endif
2018         ext3_blkdev_remove(sbi);
2019         brelse(bh);
2020 out_fail:
2021         sb->s_fs_info = NULL;
2022         kfree(sbi);
2023         lock_kernel();
2024         return ret;
2025 }
2026
2027 /*
2028  * Setup any per-fs journal parameters now.  We'll do this both on
2029  * initial mount, once the journal has been initialised but before we've
2030  * done any recovery; and again on any subsequent remount.
2031  */
2032 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2033 {
2034         struct ext3_sb_info *sbi = EXT3_SB(sb);
2035
2036         if (sbi->s_commit_interval)
2037                 journal->j_commit_interval = sbi->s_commit_interval;
2038         /* We could also set up an ext3-specific default for the commit
2039          * interval here, but for now we'll just fall back to the jbd
2040          * default. */
2041
2042         spin_lock(&journal->j_state_lock);
2043         if (test_opt(sb, BARRIER))
2044                 journal->j_flags |= JFS_BARRIER;
2045         else
2046                 journal->j_flags &= ~JFS_BARRIER;
2047         if (test_opt(sb, DATA_ERR_ABORT))
2048                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2049         else
2050                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2051         spin_unlock(&journal->j_state_lock);
2052 }
2053
2054 static journal_t *ext3_get_journal(struct super_block *sb,
2055                                    unsigned int journal_inum)
2056 {
2057         struct inode *journal_inode;
2058         journal_t *journal;
2059
2060         /* First, test for the existence of a valid inode on disk.  Bad
2061          * things happen if we iget() an unused inode, as the subsequent
2062          * iput() will try to delete it. */
2063
2064         journal_inode = ext3_iget(sb, journal_inum);
2065         if (IS_ERR(journal_inode)) {
2066                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2067                 return NULL;
2068         }
2069         if (!journal_inode->i_nlink) {
2070                 make_bad_inode(journal_inode);
2071                 iput(journal_inode);
2072                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2073                 return NULL;
2074         }
2075
2076         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2077                   journal_inode, journal_inode->i_size);
2078         if (!S_ISREG(journal_inode->i_mode)) {
2079                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2080                 iput(journal_inode);
2081                 return NULL;
2082         }
2083
2084         journal = journal_init_inode(journal_inode);
2085         if (!journal) {
2086                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2087                 iput(journal_inode);
2088                 return NULL;
2089         }
2090         journal->j_private = sb;
2091         ext3_init_journal_params(sb, journal);
2092         return journal;
2093 }
2094
2095 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2096                                        dev_t j_dev)
2097 {
2098         struct buffer_head * bh;
2099         journal_t *journal;
2100         ext3_fsblk_t start;
2101         ext3_fsblk_t len;
2102         int hblock, blocksize;
2103         ext3_fsblk_t sb_block;
2104         unsigned long offset;
2105         struct ext3_super_block * es;
2106         struct block_device *bdev;
2107
2108         bdev = ext3_blkdev_get(j_dev);
2109         if (bdev == NULL)
2110                 return NULL;
2111
2112         if (bd_claim(bdev, sb)) {
2113                 printk(KERN_ERR
2114                         "EXT3: failed to claim external journal device.\n");
2115                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2116                 return NULL;
2117         }
2118
2119         blocksize = sb->s_blocksize;
2120         hblock = bdev_hardsect_size(bdev);
2121         if (blocksize < hblock) {
2122                 printk(KERN_ERR
2123                         "EXT3-fs: blocksize too small for journal device.\n");
2124                 goto out_bdev;
2125         }
2126
2127         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2128         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2129         set_blocksize(bdev, blocksize);
2130         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2131                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2132                        "external journal\n");
2133                 goto out_bdev;
2134         }
2135
2136         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2137         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2138             !(le32_to_cpu(es->s_feature_incompat) &
2139               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2140                 printk(KERN_ERR "EXT3-fs: external journal has "
2141                                         "bad superblock\n");
2142                 brelse(bh);
2143                 goto out_bdev;
2144         }
2145
2146         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2147                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2148                 brelse(bh);
2149                 goto out_bdev;
2150         }
2151
2152         len = le32_to_cpu(es->s_blocks_count);
2153         start = sb_block + 1;
2154         brelse(bh);     /* we're done with the superblock */
2155
2156         journal = journal_init_dev(bdev, sb->s_bdev,
2157                                         start, len, blocksize);
2158         if (!journal) {
2159                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2160                 goto out_bdev;
2161         }
2162         journal->j_private = sb;
2163         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2164         wait_on_buffer(journal->j_sb_buffer);
2165         if (!buffer_uptodate(journal->j_sb_buffer)) {
2166                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2167                 goto out_journal;
2168         }
2169         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2170                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2171                                         "user (unsupported) - %d\n",
2172                         be32_to_cpu(journal->j_superblock->s_nr_users));
2173                 goto out_journal;
2174         }
2175         EXT3_SB(sb)->journal_bdev = bdev;
2176         ext3_init_journal_params(sb, journal);
2177         return journal;
2178 out_journal:
2179         journal_destroy(journal);
2180 out_bdev:
2181         ext3_blkdev_put(bdev);
2182         return NULL;
2183 }
2184
2185 static int ext3_load_journal(struct super_block *sb,
2186                              struct ext3_super_block *es,
2187                              unsigned long journal_devnum)
2188 {
2189         journal_t *journal;
2190         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2191         dev_t journal_dev;
2192         int err = 0;
2193         int really_read_only;
2194
2195         if (journal_devnum &&
2196             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2197                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2198                         "numbers have changed\n");
2199                 journal_dev = new_decode_dev(journal_devnum);
2200         } else
2201                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2202
2203         really_read_only = bdev_read_only(sb->s_bdev);
2204
2205         /*
2206          * Are we loading a blank journal or performing recovery after a
2207          * crash?  For recovery, we need to check in advance whether we
2208          * can get read-write access to the device.
2209          */
2210
2211         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2212                 if (sb->s_flags & MS_RDONLY) {
2213                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2214                                         "required on readonly filesystem.\n");
2215                         if (really_read_only) {
2216                                 printk(KERN_ERR "EXT3-fs: write access "
2217                                         "unavailable, cannot proceed.\n");
2218                                 return -EROFS;
2219                         }
2220                         printk (KERN_INFO "EXT3-fs: write access will "
2221                                         "be enabled during recovery.\n");
2222                 }
2223         }
2224
2225         if (journal_inum && journal_dev) {
2226                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2227                        "and inode journals!\n");
2228                 return -EINVAL;
2229         }
2230
2231         if (journal_inum) {
2232                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2233                         return -EINVAL;
2234         } else {
2235                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2236                         return -EINVAL;
2237         }
2238
2239         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2240                 err = journal_update_format(journal);
2241                 if (err)  {
2242                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2243                         journal_destroy(journal);
2244                         return err;
2245                 }
2246         }
2247
2248         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2249                 err = journal_wipe(journal, !really_read_only);
2250         if (!err)
2251                 err = journal_load(journal);
2252
2253         if (err) {
2254                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2255                 journal_destroy(journal);
2256                 return err;
2257         }
2258
2259         EXT3_SB(sb)->s_journal = journal;
2260         ext3_clear_journal_err(sb, es);
2261
2262         if (journal_devnum &&
2263             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2264                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2265                 sb->s_dirt = 1;
2266
2267                 /* Make sure we flush the recovery flag to disk. */
2268                 ext3_commit_super(sb, es, 1);
2269         }
2270
2271         return 0;
2272 }
2273
2274 static int ext3_create_journal(struct super_block * sb,
2275                                struct ext3_super_block * es,
2276                                unsigned int journal_inum)
2277 {
2278         journal_t *journal;
2279         int err;
2280
2281         if (sb->s_flags & MS_RDONLY) {
2282                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2283                                 "create journal.\n");
2284                 return -EROFS;
2285         }
2286
2287         journal = ext3_get_journal(sb, journal_inum);
2288         if (!journal)
2289                 return -EINVAL;
2290
2291         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2292                journal_inum);
2293
2294         err = journal_create(journal);
2295         if (err) {
2296                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2297                 journal_destroy(journal);
2298                 return -EIO;
2299         }
2300
2301         EXT3_SB(sb)->s_journal = journal;
2302
2303         ext3_update_dynamic_rev(sb);
2304         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2305         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2306
2307         es->s_journal_inum = cpu_to_le32(journal_inum);
2308         sb->s_dirt = 1;
2309
2310         /* Make sure we flush the recovery flag to disk. */
2311         ext3_commit_super(sb, es, 1);
2312
2313         return 0;
2314 }
2315
2316 static int ext3_commit_super(struct super_block *sb,
2317                                struct ext3_super_block *es,
2318                                int sync)
2319 {
2320         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2321         int error = 0;
2322
2323         if (!sbh)
2324                 return error;
2325         es->s_wtime = cpu_to_le32(get_seconds());
2326         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2327         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2328         BUFFER_TRACE(sbh, "marking dirty");
2329         mark_buffer_dirty(sbh);
2330         if (sync)
2331                 error = sync_dirty_buffer(sbh);
2332         return error;
2333 }
2334
2335
2336 /*
2337  * Have we just finished recovery?  If so, and if we are mounting (or
2338  * remounting) the filesystem readonly, then we will end up with a
2339  * consistent fs on disk.  Record that fact.
2340  */
2341 static void ext3_mark_recovery_complete(struct super_block * sb,
2342                                         struct ext3_super_block * es)
2343 {
2344         journal_t *journal = EXT3_SB(sb)->s_journal;
2345
2346         journal_lock_updates(journal);
2347         if (journal_flush(journal) < 0)
2348                 goto out;
2349
2350         lock_super(sb);
2351         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2352             sb->s_flags & MS_RDONLY) {
2353                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2354                 sb->s_dirt = 0;
2355                 ext3_commit_super(sb, es, 1);
2356         }
2357         unlock_super(sb);
2358
2359 out:
2360         journal_unlock_updates(journal);
2361 }
2362
2363 /*
2364  * If we are mounting (or read-write remounting) a filesystem whose journal
2365  * has recorded an error from a previous lifetime, move that error to the
2366  * main filesystem now.
2367  */
2368 static void ext3_clear_journal_err(struct super_block * sb,
2369                                    struct ext3_super_block * es)
2370 {
2371         journal_t *journal;
2372         int j_errno;
2373         const char *errstr;
2374
2375         journal = EXT3_SB(sb)->s_journal;
2376
2377         /*
2378          * Now check for any error status which may have been recorded in the
2379          * journal by a prior ext3_error() or ext3_abort()
2380          */
2381
2382         j_errno = journal_errno(journal);
2383         if (j_errno) {
2384                 char nbuf[16];
2385
2386                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2387                 ext3_warning(sb, __func__, "Filesystem error recorded "
2388                              "from previous mount: %s", errstr);
2389                 ext3_warning(sb, __func__, "Marking fs in need of "
2390                              "filesystem check.");
2391
2392                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2393                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2394                 ext3_commit_super (sb, es, 1);
2395
2396                 journal_clear_err(journal);
2397         }
2398 }
2399
2400 /*
2401  * Force the running and committing transactions to commit,
2402  * and wait on the commit.
2403  */
2404 int ext3_force_commit(struct super_block *sb)
2405 {
2406         journal_t *journal;
2407         int ret;
2408
2409         if (sb->s_flags & MS_RDONLY)
2410                 return 0;
2411
2412         journal = EXT3_SB(sb)->s_journal;
2413         sb->s_dirt = 0;
2414         ret = ext3_journal_force_commit(journal);
2415         return ret;
2416 }
2417
2418 /*
2419  * Ext3 always journals updates to the superblock itself, so we don't
2420  * have to propagate any other updates to the superblock on disk at this
2421  * point.  (We can probably nuke this function altogether, and remove
2422  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2423  */
2424 static void ext3_write_super (struct super_block * sb)
2425 {
2426         if (mutex_trylock(&sb->s_lock) != 0)
2427                 BUG();
2428         sb->s_dirt = 0;
2429 }
2430
2431 static int ext3_sync_fs(struct super_block *sb, int wait)
2432 {
2433         tid_t target;
2434
2435         sb->s_dirt = 0;
2436         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2437                 if (wait)
2438                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2439         }
2440         return 0;
2441 }
2442
2443 /*
2444  * LVM calls this function before a (read-only) snapshot is created.  This
2445  * gives us a chance to flush the journal completely and mark the fs clean.
2446  */
2447 static int ext3_freeze(struct super_block *sb)
2448 {
2449         int error = 0;
2450         journal_t *journal;
2451         sb->s_dirt = 0;
2452
2453         if (!(sb->s_flags & MS_RDONLY)) {
2454                 journal = EXT3_SB(sb)->s_journal;
2455
2456                 /* Now we set up the journal barrier. */
2457                 journal_lock_updates(journal);
2458
2459                 /*
2460                  * We don't want to clear needs_recovery flag when we failed
2461                  * to flush the journal.
2462                  */
2463                 error = journal_flush(journal);
2464                 if (error < 0)
2465                         goto out;
2466
2467                 /* Journal blocked and flushed, clear needs_recovery flag. */
2468                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2469                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2470                 if (error)
2471                         goto out;
2472         }
2473         return 0;
2474
2475 out:
2476         journal_unlock_updates(journal);
2477         return error;
2478 }
2479
2480 /*
2481  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2482  * flag here, even though the filesystem is not technically dirty yet.
2483  */
2484 static int ext3_unfreeze(struct super_block *sb)
2485 {
2486         if (!(sb->s_flags & MS_RDONLY)) {
2487                 lock_super(sb);
2488                 /* Reser the needs_recovery flag before the fs is unlocked. */
2489                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2490                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2491                 unlock_super(sb);
2492                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2493         }
2494         return 0;
2495 }
2496
2497 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2498 {
2499         struct ext3_super_block * es;
2500         struct ext3_sb_info *sbi = EXT3_SB(sb);
2501         ext3_fsblk_t n_blocks_count = 0;
2502         unsigned long old_sb_flags;
2503         struct ext3_mount_options old_opts;
2504         int err;
2505 #ifdef CONFIG_QUOTA
2506         int i;
2507 #endif
2508
2509         /* Store the original options */
2510         old_sb_flags = sb->s_flags;
2511         old_opts.s_mount_opt = sbi->s_mount_opt;
2512         old_opts.s_resuid = sbi->s_resuid;
2513         old_opts.s_resgid = sbi->s_resgid;
2514         old_opts.s_commit_interval = sbi->s_commit_interval;
2515 #ifdef CONFIG_QUOTA
2516         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2517         for (i = 0; i < MAXQUOTAS; i++)
2518                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2519 #endif
2520
2521         /*
2522          * Allow the "check" option to be passed as a remount option.
2523          */
2524         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2525                 err = -EINVAL;
2526                 goto restore_opts;
2527         }
2528
2529         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2530                 ext3_abort(sb, __func__, "Abort forced by user");
2531
2532         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2533                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2534
2535         es = sbi->s_es;
2536
2537         ext3_init_journal_params(sb, sbi->s_journal);
2538
2539         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2540                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2541                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2542                         err = -EROFS;
2543                         goto restore_opts;
2544                 }
2545
2546                 if (*flags & MS_RDONLY) {
2547                         /*
2548                          * First of all, the unconditional stuff we have to do
2549                          * to disable replay of the journal when we next remount
2550                          */
2551                         sb->s_flags |= MS_RDONLY;
2552
2553                         /*
2554                          * OK, test if we are remounting a valid rw partition
2555                          * readonly, and if so set the rdonly flag and then
2556                          * mark the partition as valid again.
2557                          */
2558                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2559                             (sbi->s_mount_state & EXT3_VALID_FS))
2560                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2561
2562                         /*
2563                          * We have to unlock super so that we can wait for
2564                          * transactions.
2565                          */
2566                         unlock_super(sb);
2567                         ext3_mark_recovery_complete(sb, es);
2568                         lock_super(sb);
2569                 } else {
2570                         __le32 ret;
2571                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2572                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2573                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2574                                        "remount RDWR because of unsupported "
2575                                        "optional features (%x).\n",
2576                                        sb->s_id, le32_to_cpu(ret));
2577                                 err = -EROFS;
2578                                 goto restore_opts;
2579                         }
2580
2581                         /*
2582                          * If we have an unprocessed orphan list hanging
2583                          * around from a previously readonly bdev mount,
2584                          * require a full umount/remount for now.
2585                          */
2586                         if (es->s_last_orphan) {
2587                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2588                                        "remount RDWR because of unprocessed "
2589                                        "orphan inode list.  Please "
2590                                        "umount/remount instead.\n",
2591                                        sb->s_id);
2592                                 err = -EINVAL;
2593                                 goto restore_opts;
2594                         }
2595
2596                         /*
2597                          * Mounting a RDONLY partition read-write, so reread
2598                          * and store the current valid flag.  (It may have
2599                          * been changed by e2fsck since we originally mounted
2600                          * the partition.)
2601                          */
2602                         ext3_clear_journal_err(sb, es);
2603                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2604                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2605                                 goto restore_opts;
2606                         if (!ext3_setup_super (sb, es, 0))
2607                                 sb->s_flags &= ~MS_RDONLY;
2608                 }
2609         }
2610 #ifdef CONFIG_QUOTA
2611         /* Release old quota file names */
2612         for (i = 0; i < MAXQUOTAS; i++)
2613                 if (old_opts.s_qf_names[i] &&
2614                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2615                         kfree(old_opts.s_qf_names[i]);
2616 #endif
2617         return 0;
2618 restore_opts:
2619         sb->s_flags = old_sb_flags;
2620         sbi->s_mount_opt = old_opts.s_mount_opt;
2621         sbi->s_resuid = old_opts.s_resuid;
2622         sbi->s_resgid = old_opts.s_resgid;
2623         sbi->s_commit_interval = old_opts.s_commit_interval;
2624 #ifdef CONFIG_QUOTA
2625         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2626         for (i = 0; i < MAXQUOTAS; i++) {
2627                 if (sbi->s_qf_names[i] &&
2628                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2629                         kfree(sbi->s_qf_names[i]);
2630                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2631         }
2632 #endif
2633         return err;
2634 }
2635
2636 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2637 {
2638         struct super_block *sb = dentry->d_sb;
2639         struct ext3_sb_info *sbi = EXT3_SB(sb);
2640         struct ext3_super_block *es = sbi->s_es;
2641         u64 fsid;
2642
2643         if (test_opt(sb, MINIX_DF)) {
2644                 sbi->s_overhead_last = 0;
2645         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2646                 unsigned long ngroups = sbi->s_groups_count, i;
2647                 ext3_fsblk_t overhead = 0;
2648                 smp_rmb();
2649
2650                 /*
2651                  * Compute the overhead (FS structures).  This is constant
2652                  * for a given filesystem unless the number of block groups
2653                  * changes so we cache the previous value until it does.
2654                  */
2655
2656                 /*
2657                  * All of the blocks before first_data_block are
2658                  * overhead
2659                  */
2660                 overhead = le32_to_cpu(es->s_first_data_block);
2661
2662                 /*
2663                  * Add the overhead attributed to the superblock and
2664                  * block group descriptors.  If the sparse superblocks
2665                  * feature is turned on, then not all groups have this.
2666                  */
2667                 for (i = 0; i < ngroups; i++) {
2668                         overhead += ext3_bg_has_super(sb, i) +
2669                                 ext3_bg_num_gdb(sb, i);
2670                         cond_resched();
2671                 }
2672
2673                 /*
2674                  * Every block group has an inode bitmap, a block
2675                  * bitmap, and an inode table.
2676                  */
2677                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2678                 sbi->s_overhead_last = overhead;
2679                 smp_wmb();
2680                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2681         }
2682
2683         buf->f_type = EXT3_SUPER_MAGIC;
2684         buf->f_bsize = sb->s_blocksize;
2685         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2686         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2687         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2688         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2689         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2690                 buf->f_bavail = 0;
2691         buf->f_files = le32_to_cpu(es->s_inodes_count);
2692         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2693         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2694         buf->f_namelen = EXT3_NAME_LEN;
2695         fsid = le64_to_cpup((void *)es->s_uuid) ^
2696                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2697         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2698         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2699         return 0;
2700 }
2701
2702 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2703  * is locked for write. Otherwise the are possible deadlocks:
2704  * Process 1                         Process 2
2705  * ext3_create()                     quota_sync()
2706  *   journal_start()                   write_dquot()
2707  *   DQUOT_INIT()                        down(dqio_mutex)
2708  *     down(dqio_mutex)                    journal_start()
2709  *
2710  */
2711
2712 #ifdef CONFIG_QUOTA
2713
2714 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2715 {
2716         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2717 }
2718
2719 static int ext3_dquot_initialize(struct inode *inode, int type)
2720 {
2721         handle_t *handle;
2722         int ret, err;
2723
2724         /* We may create quota structure so we need to reserve enough blocks */
2725         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2726         if (IS_ERR(handle))
2727                 return PTR_ERR(handle);
2728         ret = dquot_initialize(inode, type);
2729         err = ext3_journal_stop(handle);
2730         if (!ret)
2731                 ret = err;
2732         return ret;
2733 }
2734
2735 static int ext3_dquot_drop(struct inode *inode)
2736 {
2737         handle_t *handle;
2738         int ret, err;
2739
2740         /* We may delete quota structure so we need to reserve enough blocks */
2741         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2742         if (IS_ERR(handle)) {
2743                 /*
2744                  * We call dquot_drop() anyway to at least release references
2745                  * to quota structures so that umount does not hang.
2746                  */
2747                 dquot_drop(inode);
2748                 return PTR_ERR(handle);
2749         }
2750         ret = dquot_drop(inode);
2751         err = ext3_journal_stop(handle);
2752         if (!ret)
2753                 ret = err;
2754         return ret;
2755 }
2756
2757 static int ext3_write_dquot(struct dquot *dquot)
2758 {
2759         int ret, err;
2760         handle_t *handle;
2761         struct inode *inode;
2762
2763         inode = dquot_to_inode(dquot);
2764         handle = ext3_journal_start(inode,
2765                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2766         if (IS_ERR(handle))
2767                 return PTR_ERR(handle);
2768         ret = dquot_commit(dquot);
2769         err = ext3_journal_stop(handle);
2770         if (!ret)
2771                 ret = err;
2772         return ret;
2773 }
2774
2775 static int ext3_acquire_dquot(struct dquot *dquot)
2776 {
2777         int ret, err;
2778         handle_t *handle;
2779
2780         handle = ext3_journal_start(dquot_to_inode(dquot),
2781                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2782         if (IS_ERR(handle))
2783                 return PTR_ERR(handle);
2784         ret = dquot_acquire(dquot);
2785         err = ext3_journal_stop(handle);
2786         if (!ret)
2787                 ret = err;
2788         return ret;
2789 }
2790
2791 static int ext3_release_dquot(struct dquot *dquot)
2792 {
2793         int ret, err;
2794         handle_t *handle;
2795
2796         handle = ext3_journal_start(dquot_to_inode(dquot),
2797                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2798         if (IS_ERR(handle)) {
2799                 /* Release dquot anyway to avoid endless cycle in dqput() */
2800                 dquot_release(dquot);
2801                 return PTR_ERR(handle);
2802         }
2803         ret = dquot_release(dquot);
2804         err = ext3_journal_stop(handle);
2805         if (!ret)
2806                 ret = err;
2807         return ret;
2808 }
2809
2810 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2811 {
2812         /* Are we journaling quotas? */
2813         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2814             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2815                 dquot_mark_dquot_dirty(dquot);
2816                 return ext3_write_dquot(dquot);
2817         } else {
2818                 return dquot_mark_dquot_dirty(dquot);
2819         }
2820 }
2821
2822 static int ext3_write_info(struct super_block *sb, int type)
2823 {
2824         int ret, err;
2825         handle_t *handle;
2826
2827         /* Data block + inode block */
2828         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2829         if (IS_ERR(handle))
2830                 return PTR_ERR(handle);
2831         ret = dquot_commit_info(sb, type);
2832         err = ext3_journal_stop(handle);
2833         if (!ret)
2834                 ret = err;
2835         return ret;
2836 }
2837
2838 /*
2839  * Turn on quotas during mount time - we need to find
2840  * the quota file and such...
2841  */
2842 static int ext3_quota_on_mount(struct super_block *sb, int type)
2843 {
2844         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2845                         EXT3_SB(sb)->s_jquota_fmt, type);
2846 }
2847
2848 /*
2849  * Standard function to be called on quota_on
2850  */
2851 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2852                          char *name, int remount)
2853 {
2854         int err;
2855         struct path path;
2856
2857         if (!test_opt(sb, QUOTA))
2858                 return -EINVAL;
2859         /* When remounting, no checks are needed and in fact, name is NULL */
2860         if (remount)
2861                 return vfs_quota_on(sb, type, format_id, name, remount);
2862
2863         err = kern_path(name, LOOKUP_FOLLOW, &path);
2864         if (err)
2865                 return err;
2866
2867         /* Quotafile not on the same filesystem? */
2868         if (path.mnt->mnt_sb != sb) {
2869                 path_put(&path);
2870                 return -EXDEV;
2871         }
2872         /* Journaling quota? */
2873         if (EXT3_SB(sb)->s_qf_names[type]) {
2874                 /* Quotafile not of fs root? */
2875                 if (path.dentry->d_parent != sb->s_root)
2876                         printk(KERN_WARNING
2877                                 "EXT3-fs: Quota file not on filesystem root. "
2878                                 "Journaled quota will not work.\n");
2879         }
2880
2881         /*
2882          * When we journal data on quota file, we have to flush journal to see
2883          * all updates to the file when we bypass pagecache...
2884          */
2885         if (ext3_should_journal_data(path.dentry->d_inode)) {
2886                 /*
2887                  * We don't need to lock updates but journal_flush() could
2888                  * otherwise be livelocked...
2889                  */
2890                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2891                 err = journal_flush(EXT3_SB(sb)->s_journal);
2892                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2893                 if (err) {
2894                         path_put(&path);
2895                         return err;
2896                 }
2897         }
2898
2899         err = vfs_quota_on_path(sb, type, format_id, &path);
2900         path_put(&path);
2901         return err;
2902 }
2903
2904 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2905  * acquiring the locks... As quota files are never truncated and quota code
2906  * itself serializes the operations (and noone else should touch the files)
2907  * we don't have to be afraid of races */
2908 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2909                                size_t len, loff_t off)
2910 {
2911         struct inode *inode = sb_dqopt(sb)->files[type];
2912         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2913         int err = 0;
2914         int offset = off & (sb->s_blocksize - 1);
2915         int tocopy;
2916         size_t toread;
2917         struct buffer_head *bh;
2918         loff_t i_size = i_size_read(inode);
2919
2920         if (off > i_size)
2921                 return 0;
2922         if (off+len > i_size)
2923                 len = i_size-off;
2924         toread = len;
2925         while (toread > 0) {
2926                 tocopy = sb->s_blocksize - offset < toread ?
2927                                 sb->s_blocksize - offset : toread;
2928                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2929                 if (err)
2930                         return err;
2931                 if (!bh)        /* A hole? */
2932                         memset(data, 0, tocopy);
2933                 else
2934                         memcpy(data, bh->b_data+offset, tocopy);
2935                 brelse(bh);
2936                 offset = 0;
2937                 toread -= tocopy;
2938                 data += tocopy;
2939                 blk++;
2940         }
2941         return len;
2942 }
2943
2944 /* Write to quotafile (we know the transaction is already started and has
2945  * enough credits) */
2946 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2947                                 const char *data, size_t len, loff_t off)
2948 {
2949         struct inode *inode = sb_dqopt(sb)->files[type];
2950         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2951         int err = 0;
2952         int offset = off & (sb->s_blocksize - 1);
2953         int tocopy;
2954         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2955         size_t towrite = len;
2956         struct buffer_head *bh;
2957         handle_t *handle = journal_current_handle();
2958
2959         if (!handle) {
2960                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2961                         " cancelled because transaction is not started.\n",
2962                         (unsigned long long)off, (unsigned long long)len);
2963                 return -EIO;
2964         }
2965         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2966         while (towrite > 0) {
2967                 tocopy = sb->s_blocksize - offset < towrite ?
2968                                 sb->s_blocksize - offset : towrite;
2969                 bh = ext3_bread(handle, inode, blk, 1, &err);
2970                 if (!bh)
2971                         goto out;
2972                 if (journal_quota) {
2973                         err = ext3_journal_get_write_access(handle, bh);
2974                         if (err) {
2975                                 brelse(bh);
2976                                 goto out;
2977                         }
2978                 }
2979                 lock_buffer(bh);
2980                 memcpy(bh->b_data+offset, data, tocopy);
2981                 flush_dcache_page(bh->b_page);
2982                 unlock_buffer(bh);
2983                 if (journal_quota)
2984                         err = ext3_journal_dirty_metadata(handle, bh);
2985                 else {
2986                         /* Always do at least ordered writes for quotas */
2987                         err = ext3_journal_dirty_data(handle, bh);
2988                         mark_buffer_dirty(bh);
2989                 }
2990                 brelse(bh);
2991                 if (err)
2992                         goto out;
2993                 offset = 0;
2994                 towrite -= tocopy;
2995                 data += tocopy;
2996                 blk++;
2997         }
2998 out:
2999         if (len == towrite) {
3000                 mutex_unlock(&inode->i_mutex);
3001                 return err;
3002         }
3003         if (inode->i_size < off+len-towrite) {
3004                 i_size_write(inode, off+len-towrite);
3005                 EXT3_I(inode)->i_disksize = inode->i_size;
3006         }
3007         inode->i_version++;
3008         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3009         ext3_mark_inode_dirty(handle, inode);
3010         mutex_unlock(&inode->i_mutex);
3011         return len - towrite;
3012 }
3013
3014 #endif
3015
3016 static int ext3_get_sb(struct file_system_type *fs_type,
3017         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3018 {
3019         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3020 }
3021
3022 static struct file_system_type ext3_fs_type = {
3023         .owner          = THIS_MODULE,
3024         .name           = "ext3",
3025         .get_sb         = ext3_get_sb,
3026         .kill_sb        = kill_block_super,
3027         .fs_flags       = FS_REQUIRES_DEV,
3028 };
3029
3030 static int __init init_ext3_fs(void)
3031 {
3032         int err = init_ext3_xattr();
3033         if (err)
3034                 return err;
3035         err = init_inodecache();
3036         if (err)
3037                 goto out1;
3038         err = register_filesystem(&ext3_fs_type);
3039         if (err)
3040                 goto out;
3041         return 0;
3042 out:
3043         destroy_inodecache();
3044 out1:
3045         exit_ext3_xattr();
3046         return err;
3047 }
3048
3049 static void __exit exit_ext3_fs(void)
3050 {
3051         unregister_filesystem(&ext3_fs_type);
3052         destroy_inodecache();
3053         exit_ext3_xattr();
3054 }
3055
3056 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3057 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3058 MODULE_LICENSE("GPL");
3059 module_init(init_ext3_fs)
3060 module_exit(exit_ext3_fs)