fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/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/jbd2.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/log2.h>
  38 #include <linux/crc16.h>
  39 #include <asm/uaccess.h>
  40
  41 #include "ext4.h"
  42 #include "ext4_jbd2.h"
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46 #include "group.h"
  47
  48 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  49                              unsigned long journal_devnum);
  50 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  51                                unsigned int);
  52 static void ext4_commit_super (struct super_block * sb,
  53                                struct ext4_super_block * es,
  54                                int sync);
  55 static void ext4_mark_recovery_complete(struct super_block * sb,
  56                                         struct ext4_super_block * es);
  57 static void ext4_clear_journal_err(struct super_block * sb,
  58                                    struct ext4_super_block * es);
  59 static int ext4_sync_fs(struct super_block *sb, int wait);
  60 static const char *ext4_decode_error(struct super_block * sb, int errno,
  61                                      char nbuf[16]);
  62 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  63 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  64 static void ext4_unlockfs(struct super_block *sb);
  65 static void ext4_write_super (struct super_block * sb);
  66 static void ext4_write_super_lockfs(struct super_block *sb);
  67
  68
  69 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  70                                struct ext4_group_desc *bg)
  71 {
  72         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  73                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  74                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  75 }
  76
  77 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  78                                struct ext4_group_desc *bg)
  79 {
  80         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  81                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  82                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  83 }
  84
  85 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  86                               struct ext4_group_desc *bg)
  87 {
  88         return le32_to_cpu(bg->bg_inode_table_lo) |
  89                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  90                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  91 }
  92
  93 void ext4_block_bitmap_set(struct super_block *sb,
  94                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  95 {
  96         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  97         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  98                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  99 }
 100
 101 void ext4_inode_bitmap_set(struct super_block *sb,
 102                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 103 {
 104         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 105         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 106                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 107 }
 108
 109 void ext4_inode_table_set(struct super_block *sb,
 110                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 111 {
 112         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 113         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 114                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 115 }
 116
 117 /*
 118  * Wrappers for jbd2_journal_start/end.
 119  *
 120  * The only special thing we need to do here is to make sure that all
 121  * journal_end calls result in the superblock being marked dirty, so
 122  * that sync() will call the filesystem's write_super callback if
 123  * appropriate.
 124  */
 125 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 126 {
 127         journal_t *journal;
 128
 129         if (sb->s_flags & MS_RDONLY)
 130                 return ERR_PTR(-EROFS);
 131
 132         /* Special case here: if the journal has aborted behind our
 133          * backs (eg. EIO in the commit thread), then we still need to
 134          * take the FS itself readonly cleanly. */
 135         journal = EXT4_SB(sb)->s_journal;
 136         if (is_journal_aborted(journal)) {
 137                 ext4_abort(sb, __func__,
 138                            "Detected aborted journal");
 139                 return ERR_PTR(-EROFS);
 140         }
 141
 142         return jbd2_journal_start(journal, nblocks);
 143 }
 144
 145 /*
 146  * The only special thing we need to do here is to make sure that all
 147  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 148  * that sync() will call the filesystem's write_super callback if
 149  * appropriate.
 150  */
 151 int __ext4_journal_stop(const char *where, handle_t *handle)
 152 {
 153         struct super_block *sb;
 154         int err;
 155         int rc;
 156
 157         sb = handle->h_transaction->t_journal->j_private;
 158         err = handle->h_err;
 159         rc = jbd2_journal_stop(handle);
 160
 161         if (!err)
 162                 err = rc;
 163         if (err)
 164                 __ext4_std_error(sb, where, err);
 165         return err;
 166 }
 167
 168 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 169                 struct buffer_head *bh, handle_t *handle, int err)
 170 {
 171         char nbuf[16];
 172         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 173
 174         if (bh)
 175                 BUFFER_TRACE(bh, "abort");
 176
 177         if (!handle->h_err)
 178                 handle->h_err = err;
 179
 180         if (is_handle_aborted(handle))
 181                 return;
 182
 183         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 184                caller, errstr, err_fn);
 185
 186         jbd2_journal_abort_handle(handle);
 187 }
 188
 189 /* Deal with the reporting of failure conditions on a filesystem such as
 190  * inconsistencies detected or read IO failures.
 191  *
 192  * On ext2, we can store the error state of the filesystem in the
 193  * superblock.  That is not possible on ext4, because we may have other
 194  * write ordering constraints on the superblock which prevent us from
 195  * writing it out straight away; and given that the journal is about to
 196  * be aborted, we can't rely on the current, or future, transactions to
 197  * write out the superblock safely.
 198  *
 199  * We'll just use the jbd2_journal_abort() error code to record an error in
 200  * the journal instead.  On recovery, the journal will compain about
 201  * that error until we've noted it down and cleared it.
 202  */
 203
 204 static void ext4_handle_error(struct super_block *sb)
 205 {
 206         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 207
 208         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 209         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 210
 211         if (sb->s_flags & MS_RDONLY)
 212                 return;
 213
 214         if (!test_opt (sb, ERRORS_CONT)) {
 215                 journal_t *journal = EXT4_SB(sb)->s_journal;
 216
 217                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 218                 if (journal)
 219                         jbd2_journal_abort(journal, -EIO);
 220         }
 221         if (test_opt (sb, ERRORS_RO)) {
 222                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 223                 sb->s_flags |= MS_RDONLY;
 224         }
 225         ext4_commit_super(sb, es, 1);
 226         if (test_opt(sb, ERRORS_PANIC))
 227                 panic("EXT4-fs (device %s): panic forced after error\n",
 228                         sb->s_id);
 229 }
 230
 231 void ext4_error (struct super_block * sb, const char * function,
 232                  const char * fmt, ...)
 233 {
 234         va_list args;
 235
 236         va_start(args, fmt);
 237         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 238         vprintk(fmt, args);
 239         printk("\n");
 240         va_end(args);
 241
 242         ext4_handle_error(sb);
 243 }
 244
 245 static const char *ext4_decode_error(struct super_block * sb, int errno,
 246                                      char nbuf[16])
 247 {
 248         char *errstr = NULL;
 249
 250         switch (errno) {
 251         case -EIO:
 252                 errstr = "IO failure";
 253                 break;
 254         case -ENOMEM:
 255                 errstr = "Out of memory";
 256                 break;
 257         case -EROFS:
 258                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 259                         errstr = "Journal has aborted";
 260                 else
 261                         errstr = "Readonly filesystem";
 262                 break;
 263         default:
 264                 /* If the caller passed in an extra buffer for unknown
 265                  * errors, textualise them now.  Else we just return
 266                  * NULL. */
 267                 if (nbuf) {
 268                         /* Check for truncated error codes... */
 269                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 270                                 errstr = nbuf;
 271                 }
 272                 break;
 273         }
 274
 275         return errstr;
 276 }
 277
 278 /* __ext4_std_error decodes expected errors from journaling functions
 279  * automatically and invokes the appropriate error response.  */
 280
 281 void __ext4_std_error (struct super_block * sb, const char * function,
 282                        int errno)
 283 {
 284         char nbuf[16];
 285         const char *errstr;
 286
 287         /* Special case: if the error is EROFS, and we're not already
 288          * inside a transaction, then there's really no point in logging
 289          * an error. */
 290         if (errno == -EROFS && journal_current_handle() == NULL &&
 291             (sb->s_flags & MS_RDONLY))
 292                 return;
 293
 294         errstr = ext4_decode_error(sb, errno, nbuf);
 295         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 296                 sb->s_id, function, errstr);
 297
 298         ext4_handle_error(sb);
 299 }
 300
 301 /*
 302  * ext4_abort is a much stronger failure handler than ext4_error.  The
 303  * abort function may be used to deal with unrecoverable failures such
 304  * as journal IO errors or ENOMEM at a critical moment in log management.
 305  *
 306  * We unconditionally force the filesystem into an ABORT|READONLY state,
 307  * unless the error response on the fs has been set to panic in which
 308  * case we take the easy way out and panic immediately.
 309  */
 310
 311 void ext4_abort (struct super_block * sb, const char * function,
 312                  const char * fmt, ...)
 313 {
 314         va_list args;
 315
 316         printk (KERN_CRIT "ext4_abort called.\n");
 317
 318         va_start(args, fmt);
 319         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 320         vprintk(fmt, args);
 321         printk("\n");
 322         va_end(args);
 323
 324         if (test_opt(sb, ERRORS_PANIC))
 325                 panic("EXT4-fs panic from previous error\n");
 326
 327         if (sb->s_flags & MS_RDONLY)
 328                 return;
 329
 330         printk(KERN_CRIT "Remounting filesystem read-only\n");
 331         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 332         sb->s_flags |= MS_RDONLY;
 333         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 334         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 335 }
 336
 337 void ext4_warning (struct super_block * sb, const char * function,
 338                    const char * fmt, ...)
 339 {
 340         va_list args;
 341
 342         va_start(args, fmt);
 343         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 344                sb->s_id, function);
 345         vprintk(fmt, args);
 346         printk("\n");
 347         va_end(args);
 348 }
 349
 350 void ext4_update_dynamic_rev(struct super_block *sb)
 351 {
 352         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 353
 354         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 355                 return;
 356
 357         ext4_warning(sb, __func__,
 358                      "updating to rev %d because of new feature flag, "
 359                      "running e2fsck is recommended",
 360                      EXT4_DYNAMIC_REV);
 361
 362         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 363         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 364         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 365         /* leave es->s_feature_*compat flags alone */
 366         /* es->s_uuid will be set by e2fsck if empty */
 367
 368         /*
 369          * The rest of the superblock fields should be zero, and if not it
 370          * means they are likely already in use, so leave them alone.  We
 371          * can leave it up to e2fsck to clean up any inconsistencies there.
 372          */
 373 }
 374
 375 int ext4_update_compat_feature(handle_t *handle,
 376                                         struct super_block *sb, __u32 compat)
 377 {
 378         int err = 0;
 379         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 380                 err = ext4_journal_get_write_access(handle,
 381                                 EXT4_SB(sb)->s_sbh);
 382                 if (err)
 383                         return err;
 384                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 385                 sb->s_dirt = 1;
 386                 handle->h_sync = 1;
 387                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 388                                         "call ext4_journal_dirty_met adata");
 389                 err = ext4_journal_dirty_metadata(handle,
 390                                 EXT4_SB(sb)->s_sbh);
 391         }
 392         return err;
 393 }
 394
 395 int ext4_update_rocompat_feature(handle_t *handle,
 396                                         struct super_block *sb, __u32 rocompat)
 397 {
 398         int err = 0;
 399         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 400                 err = ext4_journal_get_write_access(handle,
 401                                 EXT4_SB(sb)->s_sbh);
 402                 if (err)
 403                         return err;
 404                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 405                 sb->s_dirt = 1;
 406                 handle->h_sync = 1;
 407                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 408                                         "call ext4_journal_dirty_met adata");
 409                 err = ext4_journal_dirty_metadata(handle,
 410                                 EXT4_SB(sb)->s_sbh);
 411         }
 412         return err;
 413 }
 414
 415 int ext4_update_incompat_feature(handle_t *handle,
 416                                         struct super_block *sb, __u32 incompat)
 417 {
 418         int err = 0;
 419         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 420                 err = ext4_journal_get_write_access(handle,
 421                                 EXT4_SB(sb)->s_sbh);
 422                 if (err)
 423                         return err;
 424                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 425                 sb->s_dirt = 1;
 426                 handle->h_sync = 1;
 427                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 428                                         "call ext4_journal_dirty_met adata");
 429                 err = ext4_journal_dirty_metadata(handle,
 430                                 EXT4_SB(sb)->s_sbh);
 431         }
 432         return err;
 433 }
 434
 435 /*
 436  * Open the external journal device
 437  */
 438 static struct block_device *ext4_blkdev_get(dev_t dev)
 439 {
 440         struct block_device *bdev;
 441         char b[BDEVNAME_SIZE];
 442
 443         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 444         if (IS_ERR(bdev))
 445                 goto fail;
 446         return bdev;
 447
 448 fail:
 449         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 450                         __bdevname(dev, b), PTR_ERR(bdev));
 451         return NULL;
 452 }
 453
 454 /*
 455  * Release the journal device
 456  */
 457 static int ext4_blkdev_put(struct block_device *bdev)
 458 {
 459         bd_release(bdev);
 460         return blkdev_put(bdev);
 461 }
 462
 463 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 464 {
 465         struct block_device *bdev;
 466         int ret = -ENODEV;
 467
 468         bdev = sbi->journal_bdev;
 469         if (bdev) {
 470                 ret = ext4_blkdev_put(bdev);
 471                 sbi->journal_bdev = NULL;
 472         }
 473         return ret;
 474 }
 475
 476 static inline struct inode *orphan_list_entry(struct list_head *l)
 477 {
 478         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 479 }
 480
 481 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 482 {
 483         struct list_head *l;
 484
 485         printk(KERN_ERR "sb orphan head is %d\n",
 486                le32_to_cpu(sbi->s_es->s_last_orphan));
 487
 488         printk(KERN_ERR "sb_info orphan list:\n");
 489         list_for_each(l, &sbi->s_orphan) {
 490                 struct inode *inode = orphan_list_entry(l);
 491                 printk(KERN_ERR "  "
 492                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 493                        inode->i_sb->s_id, inode->i_ino, inode,
 494                        inode->i_mode, inode->i_nlink,
 495                        NEXT_ORPHAN(inode));
 496         }
 497 }
 498
 499 static void ext4_put_super (struct super_block * sb)
 500 {
 501         struct ext4_sb_info *sbi = EXT4_SB(sb);
 502         struct ext4_super_block *es = sbi->s_es;
 503         int i;
 504
 505         ext4_mb_release(sb);
 506         ext4_ext_release(sb);
 507         ext4_xattr_put_super(sb);
 508         jbd2_journal_destroy(sbi->s_journal);
 509         if (!(sb->s_flags & MS_RDONLY)) {
 510                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 511                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 512                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 513                 mark_buffer_dirty(sbi->s_sbh);
 514                 ext4_commit_super(sb, es, 1);
 515         }
 516
 517         for (i = 0; i < sbi->s_gdb_count; i++)
 518                 brelse(sbi->s_group_desc[i]);
 519         kfree(sbi->s_group_desc);
 520         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 521         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 522         percpu_counter_destroy(&sbi->s_dirs_counter);
 523         brelse(sbi->s_sbh);
 524 #ifdef CONFIG_QUOTA
 525         for (i = 0; i < MAXQUOTAS; i++)
 526                 kfree(sbi->s_qf_names[i]);
 527 #endif
 528
 529         /* Debugging code just in case the in-memory inode orphan list
 530          * isn't empty.  The on-disk one can be non-empty if we've
 531          * detected an error and taken the fs readonly, but the
 532          * in-memory list had better be clean by this point. */
 533         if (!list_empty(&sbi->s_orphan))
 534                 dump_orphan_list(sb, sbi);
 535         J_ASSERT(list_empty(&sbi->s_orphan));
 536
 537         invalidate_bdev(sb->s_bdev);
 538         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 539                 /*
 540                  * Invalidate the journal device's buffers.  We don't want them
 541                  * floating about in memory - the physical journal device may
 542                  * hotswapped, and it breaks the `ro-after' testing code.
 543                  */
 544                 sync_blockdev(sbi->journal_bdev);
 545                 invalidate_bdev(sbi->journal_bdev);
 546                 ext4_blkdev_remove(sbi);
 547         }
 548         sb->s_fs_info = NULL;
 549         kfree(sbi);
 550         return;
 551 }
 552
 553 static struct kmem_cache *ext4_inode_cachep;
 554
 555 /*
 556  * Called inside transaction, so use GFP_NOFS
 557  */
 558 static struct inode *ext4_alloc_inode(struct super_block *sb)
 559 {
 560         struct ext4_inode_info *ei;
 561
 562         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 563         if (!ei)
 564                 return NULL;
 565 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 566         ei->i_acl = EXT4_ACL_NOT_CACHED;
 567         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 568 #endif
 569         ei->i_block_alloc_info = NULL;
 570         ei->vfs_inode.i_version = 1;
 571         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 572         INIT_LIST_HEAD(&ei->i_prealloc_list);
 573         spin_lock_init(&ei->i_prealloc_lock);
 574         return &ei->vfs_inode;
 575 }
 576
 577 static void ext4_destroy_inode(struct inode *inode)
 578 {
 579         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 580                 printk("EXT4 Inode %p: orphan list check failed!\n",
 581                         EXT4_I(inode));
 582                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 583                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 584                                 true);
 585                 dump_stack();
 586         }
 587         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 588 }
 589
 590 static void init_once(struct kmem_cache *cachep, void *foo)
 591 {
 592         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 593
 594         INIT_LIST_HEAD(&ei->i_orphan);
 595 #ifdef CONFIG_EXT4DEV_FS_XATTR
 596         init_rwsem(&ei->xattr_sem);
 597 #endif
 598         init_rwsem(&ei->i_data_sem);
 599         inode_init_once(&ei->vfs_inode);
 600 }
 601
 602 static int init_inodecache(void)
 603 {
 604         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 605                                              sizeof(struct ext4_inode_info),
 606                                              0, (SLAB_RECLAIM_ACCOUNT|
 607                                                 SLAB_MEM_SPREAD),
 608                                              init_once);
 609         if (ext4_inode_cachep == NULL)
 610                 return -ENOMEM;
 611         return 0;
 612 }
 613
 614 static void destroy_inodecache(void)
 615 {
 616         kmem_cache_destroy(ext4_inode_cachep);
 617 }
 618
 619 static void ext4_clear_inode(struct inode *inode)
 620 {
 621         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 622 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 623         if (EXT4_I(inode)->i_acl &&
 624                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 625                 posix_acl_release(EXT4_I(inode)->i_acl);
 626                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 627         }
 628         if (EXT4_I(inode)->i_default_acl &&
 629                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 630                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 631                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 632         }
 633 #endif
 634         ext4_discard_reservation(inode);
 635         EXT4_I(inode)->i_block_alloc_info = NULL;
 636         if (unlikely(rsv))
 637                 kfree(rsv);
 638 }
 639
 640 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 641 {
 642 #if defined(CONFIG_QUOTA)
 643         struct ext4_sb_info *sbi = EXT4_SB(sb);
 644
 645         if (sbi->s_jquota_fmt)
 646                 seq_printf(seq, ",jqfmt=%s",
 647                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 648
 649         if (sbi->s_qf_names[USRQUOTA])
 650                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 651
 652         if (sbi->s_qf_names[GRPQUOTA])
 653                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 654
 655         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 656                 seq_puts(seq, ",usrquota");
 657
 658         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 659                 seq_puts(seq, ",grpquota");
 660 #endif
 661 }
 662
 663 /*
 664  * Show an option if
 665  *  - it's set to a non-default value OR
 666  *  - if the per-sb default is different from the global default
 667  */
 668 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 669 {
 670         int def_errors;
 671         unsigned long def_mount_opts;
 672         struct super_block *sb = vfs->mnt_sb;
 673         struct ext4_sb_info *sbi = EXT4_SB(sb);
 674         struct ext4_super_block *es = sbi->s_es;
 675
 676         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 677         def_errors     = le16_to_cpu(es->s_errors);
 678
 679         if (sbi->s_sb_block != 1)
 680                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 681         if (test_opt(sb, MINIX_DF))
 682                 seq_puts(seq, ",minixdf");
 683         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 684                 seq_puts(seq, ",grpid");
 685         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 686                 seq_puts(seq, ",nogrpid");
 687         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 688             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 689                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 690         }
 691         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 692             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 693                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 694         }
 695         if (test_opt(sb, ERRORS_RO)) {
 696                 if (def_errors == EXT4_ERRORS_PANIC ||
 697                     def_errors == EXT4_ERRORS_CONTINUE) {
 698                         seq_puts(seq, ",errors=remount-ro");
 699                 }
 700         }
 701         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 702                 seq_puts(seq, ",errors=continue");
 703         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 704                 seq_puts(seq, ",errors=panic");
 705         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 706                 seq_puts(seq, ",nouid32");
 707         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 708                 seq_puts(seq, ",debug");
 709         if (test_opt(sb, OLDALLOC))
 710                 seq_puts(seq, ",oldalloc");
 711 #ifdef CONFIG_EXT4DEV_FS_XATTR
 712         if (test_opt(sb, XATTR_USER) &&
 713                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 714                 seq_puts(seq, ",user_xattr");
 715         if (!test_opt(sb, XATTR_USER) &&
 716             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 717                 seq_puts(seq, ",nouser_xattr");
 718         }
 719 #endif
 720 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 721         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 722                 seq_puts(seq, ",acl");
 723         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 724                 seq_puts(seq, ",noacl");
 725 #endif
 726         if (!test_opt(sb, RESERVATION))
 727                 seq_puts(seq, ",noreservation");
 728         if (sbi->s_commit_interval) {
 729                 seq_printf(seq, ",commit=%u",
 730                            (unsigned) (sbi->s_commit_interval / HZ));
 731         }
 732         if (test_opt(sb, BARRIER))
 733                 seq_puts(seq, ",barrier=1");
 734         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 735                 seq_puts(seq, ",journal_async_commit");
 736         if (test_opt(sb, NOBH))
 737                 seq_puts(seq, ",nobh");
 738         if (!test_opt(sb, EXTENTS))
 739                 seq_puts(seq, ",noextents");
 740         if (!test_opt(sb, MBALLOC))
 741                 seq_puts(seq, ",nomballoc");
 742         if (test_opt(sb, I_VERSION))
 743                 seq_puts(seq, ",i_version");
 744
 745         if (sbi->s_stripe)
 746                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 747         /*
 748          * journal mode get enabled in different ways
 749          * So just print the value even if we didn't specify it
 750          */
 751         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 752                 seq_puts(seq, ",data=journal");
 753         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 754                 seq_puts(seq, ",data=ordered");
 755         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 756                 seq_puts(seq, ",data=writeback");
 757
 758         ext4_show_quota_options(seq, sb);
 759         return 0;
 760 }
 761
 762
 763 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 764                 u64 ino, u32 generation)
 765 {
 766         struct inode *inode;
 767
 768         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 769                 return ERR_PTR(-ESTALE);
 770         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 771                 return ERR_PTR(-ESTALE);
 772
 773         /* iget isn't really right if the inode is currently unallocated!!
 774          *
 775          * ext4_read_inode will return a bad_inode if the inode had been
 776          * deleted, so we should be safe.
 777          *
 778          * Currently we don't know the generation for parent directory, so
 779          * a generation of 0 means "accept any"
 780          */
 781         inode = ext4_iget(sb, ino);
 782         if (IS_ERR(inode))
 783                 return ERR_CAST(inode);
 784         if (generation && inode->i_generation != generation) {
 785                 iput(inode);
 786                 return ERR_PTR(-ESTALE);
 787         }
 788
 789         return inode;
 790 }
 791
 792 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 793                 int fh_len, int fh_type)
 794 {
 795         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 796                                     ext4_nfs_get_inode);
 797 }
 798
 799 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 800                 int fh_len, int fh_type)
 801 {
 802         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 803                                     ext4_nfs_get_inode);
 804 }
 805
 806 #ifdef CONFIG_QUOTA
 807 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 808 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 809
 810 static int ext4_dquot_initialize(struct inode *inode, int type);
 811 static int ext4_dquot_drop(struct inode *inode);
 812 static int ext4_write_dquot(struct dquot *dquot);
 813 static int ext4_acquire_dquot(struct dquot *dquot);
 814 static int ext4_release_dquot(struct dquot *dquot);
 815 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 816 static int ext4_write_info(struct super_block *sb, int type);
 817 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 818                                 char *path, int remount);
 819 static int ext4_quota_on_mount(struct super_block *sb, int type);
 820 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 821                                size_t len, loff_t off);
 822 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 823                                 const char *data, size_t len, loff_t off);
 824
 825 static struct dquot_operations ext4_quota_operations = {
 826         .initialize     = ext4_dquot_initialize,
 827         .drop           = ext4_dquot_drop,
 828         .alloc_space    = dquot_alloc_space,
 829         .alloc_inode    = dquot_alloc_inode,
 830         .free_space     = dquot_free_space,
 831         .free_inode     = dquot_free_inode,
 832         .transfer       = dquot_transfer,
 833         .write_dquot    = ext4_write_dquot,
 834         .acquire_dquot  = ext4_acquire_dquot,
 835         .release_dquot  = ext4_release_dquot,
 836         .mark_dirty     = ext4_mark_dquot_dirty,
 837         .write_info     = ext4_write_info
 838 };
 839
 840 static struct quotactl_ops ext4_qctl_operations = {
 841         .quota_on       = ext4_quota_on,
 842         .quota_off      = vfs_quota_off,
 843         .quota_sync     = vfs_quota_sync,
 844         .get_info       = vfs_get_dqinfo,
 845         .set_info       = vfs_set_dqinfo,
 846         .get_dqblk      = vfs_get_dqblk,
 847         .set_dqblk      = vfs_set_dqblk
 848 };
 849 #endif
 850
 851 static const struct super_operations ext4_sops = {
 852         .alloc_inode    = ext4_alloc_inode,
 853         .destroy_inode  = ext4_destroy_inode,
 854         .write_inode    = ext4_write_inode,
 855         .dirty_inode    = ext4_dirty_inode,
 856         .delete_inode   = ext4_delete_inode,
 857         .put_super      = ext4_put_super,
 858         .write_super    = ext4_write_super,
 859         .sync_fs        = ext4_sync_fs,
 860         .write_super_lockfs = ext4_write_super_lockfs,
 861         .unlockfs       = ext4_unlockfs,
 862         .statfs         = ext4_statfs,
 863         .remount_fs     = ext4_remount,
 864         .clear_inode    = ext4_clear_inode,
 865         .show_options   = ext4_show_options,
 866 #ifdef CONFIG_QUOTA
 867         .quota_read     = ext4_quota_read,
 868         .quota_write    = ext4_quota_write,
 869 #endif
 870 };
 871
 872 static const struct export_operations ext4_export_ops = {
 873         .fh_to_dentry = ext4_fh_to_dentry,
 874         .fh_to_parent = ext4_fh_to_parent,
 875         .get_parent = ext4_get_parent,
 876 };
 877
 878 enum {
 879         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 880         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 881         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 882         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 883         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 884         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 885         Opt_journal_checksum, Opt_journal_async_commit,
 886         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 887         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 888         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 889         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 890         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 891         Opt_mballoc, Opt_nomballoc, Opt_stripe,
 892 };
 893
 894 static match_table_t tokens = {
 895         {Opt_bsd_df, "bsddf"},
 896         {Opt_minix_df, "minixdf"},
 897         {Opt_grpid, "grpid"},
 898         {Opt_grpid, "bsdgroups"},
 899         {Opt_nogrpid, "nogrpid"},
 900         {Opt_nogrpid, "sysvgroups"},
 901         {Opt_resgid, "resgid=%u"},
 902         {Opt_resuid, "resuid=%u"},
 903         {Opt_sb, "sb=%u"},
 904         {Opt_err_cont, "errors=continue"},
 905         {Opt_err_panic, "errors=panic"},
 906         {Opt_err_ro, "errors=remount-ro"},
 907         {Opt_nouid32, "nouid32"},
 908         {Opt_nocheck, "nocheck"},
 909         {Opt_nocheck, "check=none"},
 910         {Opt_debug, "debug"},
 911         {Opt_oldalloc, "oldalloc"},
 912         {Opt_orlov, "orlov"},
 913         {Opt_user_xattr, "user_xattr"},
 914         {Opt_nouser_xattr, "nouser_xattr"},
 915         {Opt_acl, "acl"},
 916         {Opt_noacl, "noacl"},
 917         {Opt_reservation, "reservation"},
 918         {Opt_noreservation, "noreservation"},
 919         {Opt_noload, "noload"},
 920         {Opt_nobh, "nobh"},
 921         {Opt_bh, "bh"},
 922         {Opt_commit, "commit=%u"},
 923         {Opt_journal_update, "journal=update"},
 924         {Opt_journal_inum, "journal=%u"},
 925         {Opt_journal_dev, "journal_dev=%u"},
 926         {Opt_journal_checksum, "journal_checksum"},
 927         {Opt_journal_async_commit, "journal_async_commit"},
 928         {Opt_abort, "abort"},
 929         {Opt_data_journal, "data=journal"},
 930         {Opt_data_ordered, "data=ordered"},
 931         {Opt_data_writeback, "data=writeback"},
 932         {Opt_offusrjquota, "usrjquota="},
 933         {Opt_usrjquota, "usrjquota=%s"},
 934         {Opt_offgrpjquota, "grpjquota="},
 935         {Opt_grpjquota, "grpjquota=%s"},
 936         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 937         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 938         {Opt_grpquota, "grpquota"},
 939         {Opt_noquota, "noquota"},
 940         {Opt_quota, "quota"},
 941         {Opt_usrquota, "usrquota"},
 942         {Opt_barrier, "barrier=%u"},
 943         {Opt_extents, "extents"},
 944         {Opt_noextents, "noextents"},
 945         {Opt_i_version, "i_version"},
 946         {Opt_mballoc, "mballoc"},
 947         {Opt_nomballoc, "nomballoc"},
 948         {Opt_stripe, "stripe=%u"},
 949         {Opt_resize, "resize"},
 950         {Opt_err, NULL},
 951 };
 952
 953 static ext4_fsblk_t get_sb_block(void **data)
 954 {
 955         ext4_fsblk_t    sb_block;
 956         char            *options = (char *) *data;
 957
 958         if (!options || strncmp(options, "sb=", 3) != 0)
 959                 return 1;       /* Default location */
 960         options += 3;
 961         /*todo: use simple_strtoll with >32bit ext4 */
 962         sb_block = simple_strtoul(options, &options, 0);
 963         if (*options && *options != ',') {
 964                 printk("EXT4-fs: Invalid sb specification: %s\n",
 965                        (char *) *data);
 966                 return 1;
 967         }
 968         if (*options == ',')
 969                 options++;
 970         *data = (void *) options;
 971         return sb_block;
 972 }
 973
 974 static int parse_options (char *options, struct super_block *sb,
 975                           unsigned int *inum, unsigned long *journal_devnum,
 976                           ext4_fsblk_t *n_blocks_count, int is_remount)
 977 {
 978         struct ext4_sb_info *sbi = EXT4_SB(sb);
 979         char * p;
 980         substring_t args[MAX_OPT_ARGS];
 981         int data_opt = 0;
 982         int option;
 983 #ifdef CONFIG_QUOTA
 984         int qtype, qfmt;
 985         char *qname;
 986 #endif
 987
 988         if (!options)
 989                 return 1;
 990
 991         while ((p = strsep (&options, ",")) != NULL) {
 992                 int token;
 993                 if (!*p)
 994                         continue;
 995
 996                 token = match_token(p, tokens, args);
 997                 switch (token) {
 998                 case Opt_bsd_df:
 999                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1000                         break;
1001                 case Opt_minix_df:
1002                         set_opt (sbi->s_mount_opt, MINIX_DF);
1003                         break;
1004                 case Opt_grpid:
1005                         set_opt (sbi->s_mount_opt, GRPID);
1006                         break;
1007                 case Opt_nogrpid:
1008                         clear_opt (sbi->s_mount_opt, GRPID);
1009                         break;
1010                 case Opt_resuid:
1011                         if (match_int(&args[0], &option))
1012                                 return 0;
1013                         sbi->s_resuid = option;
1014                         break;
1015                 case Opt_resgid:
1016                         if (match_int(&args[0], &option))
1017                                 return 0;
1018                         sbi->s_resgid = option;
1019                         break;
1020                 case Opt_sb:
1021                         /* handled by get_sb_block() instead of here */
1022                         /* *sb_block = match_int(&args[0]); */
1023                         break;
1024                 case Opt_err_panic:
1025                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1026                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1027                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1028                         break;
1029                 case Opt_err_ro:
1030                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1031                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1032                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1033                         break;
1034                 case Opt_err_cont:
1035                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1036                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1037                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1038                         break;
1039                 case Opt_nouid32:
1040                         set_opt (sbi->s_mount_opt, NO_UID32);
1041                         break;
1042                 case Opt_nocheck:
1043                         clear_opt (sbi->s_mount_opt, CHECK);
1044                         break;
1045                 case Opt_debug:
1046                         set_opt (sbi->s_mount_opt, DEBUG);
1047                         break;
1048                 case Opt_oldalloc:
1049                         set_opt (sbi->s_mount_opt, OLDALLOC);
1050                         break;
1051                 case Opt_orlov:
1052                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1053                         break;
1054 #ifdef CONFIG_EXT4DEV_FS_XATTR
1055                 case Opt_user_xattr:
1056                         set_opt (sbi->s_mount_opt, XATTR_USER);
1057                         break;
1058                 case Opt_nouser_xattr:
1059                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1060                         break;
1061 #else
1062                 case Opt_user_xattr:
1063                 case Opt_nouser_xattr:
1064                         printk("EXT4 (no)user_xattr options not supported\n");
1065                         break;
1066 #endif
1067 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1068                 case Opt_acl:
1069                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1070                         break;
1071                 case Opt_noacl:
1072                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1073                         break;
1074 #else
1075                 case Opt_acl:
1076                 case Opt_noacl:
1077                         printk("EXT4 (no)acl options not supported\n");
1078                         break;
1079 #endif
1080                 case Opt_reservation:
1081                         set_opt(sbi->s_mount_opt, RESERVATION);
1082                         break;
1083                 case Opt_noreservation:
1084                         clear_opt(sbi->s_mount_opt, RESERVATION);
1085                         break;
1086                 case Opt_journal_update:
1087                         /* @@@ FIXME */
1088                         /* Eventually we will want to be able to create
1089                            a journal file here.  For now, only allow the
1090                            user to specify an existing inode to be the
1091                            journal file. */
1092                         if (is_remount) {
1093                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1094                                        "journal on remount\n");
1095                                 return 0;
1096                         }
1097                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1098                         break;
1099                 case Opt_journal_inum:
1100                         if (is_remount) {
1101                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1102                                        "journal on remount\n");
1103                                 return 0;
1104                         }
1105                         if (match_int(&args[0], &option))
1106                                 return 0;
1107                         *inum = option;
1108                         break;
1109                 case Opt_journal_dev:
1110                         if (is_remount) {
1111                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1112                                        "journal on remount\n");
1113                                 return 0;
1114                         }
1115                         if (match_int(&args[0], &option))
1116                                 return 0;
1117                         *journal_devnum = option;
1118                         break;
1119                 case Opt_journal_checksum:
1120                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1121                         break;
1122                 case Opt_journal_async_commit:
1123                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1124                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1125                         break;
1126                 case Opt_noload:
1127                         set_opt (sbi->s_mount_opt, NOLOAD);
1128                         break;
1129                 case Opt_commit:
1130                         if (match_int(&args[0], &option))
1131                                 return 0;
1132                         if (option < 0)
1133                                 return 0;
1134                         if (option == 0)
1135                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1136                         sbi->s_commit_interval = HZ * option;
1137                         break;
1138                 case Opt_data_journal:
1139                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1140                         goto datacheck;
1141                 case Opt_data_ordered:
1142                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1143                         goto datacheck;
1144                 case Opt_data_writeback:
1145                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1146                 datacheck:
1147                         if (is_remount) {
1148                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1149                                                 != data_opt) {
1150                                         printk(KERN_ERR
1151                                                 "EXT4-fs: cannot change data "
1152                                                 "mode on remount\n");
1153                                         return 0;
1154                                 }
1155                         } else {
1156                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1157                                 sbi->s_mount_opt |= data_opt;
1158                         }
1159                         break;
1160 #ifdef CONFIG_QUOTA
1161                 case Opt_usrjquota:
1162                         qtype = USRQUOTA;
1163                         goto set_qf_name;
1164                 case Opt_grpjquota:
1165                         qtype = GRPQUOTA;
1166 set_qf_name:
1167                         if ((sb_any_quota_enabled(sb) ||
1168                              sb_any_quota_suspended(sb)) &&
1169                             !sbi->s_qf_names[qtype]) {
1170                                 printk(KERN_ERR
1171                                         "EXT4-fs: Cannot change journaled "
1172                                         "quota options when quota turned on.\n");
1173                                 return 0;
1174                         }
1175                         qname = match_strdup(&args[0]);
1176                         if (!qname) {
1177                                 printk(KERN_ERR
1178                                         "EXT4-fs: not enough memory for "
1179                                         "storing quotafile name.\n");
1180                                 return 0;
1181                         }
1182                         if (sbi->s_qf_names[qtype] &&
1183                             strcmp(sbi->s_qf_names[qtype], qname)) {
1184                                 printk(KERN_ERR
1185                                         "EXT4-fs: %s quota file already "
1186                                         "specified.\n", QTYPE2NAME(qtype));
1187                                 kfree(qname);
1188                                 return 0;
1189                         }
1190                         sbi->s_qf_names[qtype] = qname;
1191                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1192                                 printk(KERN_ERR
1193                                         "EXT4-fs: quotafile must be on "
1194                                         "filesystem root.\n");
1195                                 kfree(sbi->s_qf_names[qtype]);
1196                                 sbi->s_qf_names[qtype] = NULL;
1197                                 return 0;
1198                         }
1199                         set_opt(sbi->s_mount_opt, QUOTA);
1200                         break;
1201                 case Opt_offusrjquota:
1202                         qtype = USRQUOTA;
1203                         goto clear_qf_name;
1204                 case Opt_offgrpjquota:
1205                         qtype = GRPQUOTA;
1206 clear_qf_name:
1207                         if ((sb_any_quota_enabled(sb) ||
1208                              sb_any_quota_suspended(sb)) &&
1209                             sbi->s_qf_names[qtype]) {
1210                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1211                                         "journaled quota options when "
1212                                         "quota turned on.\n");
1213                                 return 0;
1214                         }
1215                         /*
1216                          * The space will be released later when all options
1217                          * are confirmed to be correct
1218                          */
1219                         sbi->s_qf_names[qtype] = NULL;
1220                         break;
1221                 case Opt_jqfmt_vfsold:
1222                         qfmt = QFMT_VFS_OLD;
1223                         goto set_qf_format;
1224                 case Opt_jqfmt_vfsv0:
1225                         qfmt = QFMT_VFS_V0;
1226 set_qf_format:
1227                         if ((sb_any_quota_enabled(sb) ||
1228                              sb_any_quota_suspended(sb)) &&
1229                             sbi->s_jquota_fmt != qfmt) {
1230                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1231                                         "journaled quota options when "
1232                                         "quota turned on.\n");
1233                                 return 0;
1234                         }
1235                         sbi->s_jquota_fmt = qfmt;
1236                         break;
1237                 case Opt_quota:
1238                 case Opt_usrquota:
1239                         set_opt(sbi->s_mount_opt, QUOTA);
1240                         set_opt(sbi->s_mount_opt, USRQUOTA);
1241                         break;
1242                 case Opt_grpquota:
1243                         set_opt(sbi->s_mount_opt, QUOTA);
1244                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1245                         break;
1246                 case Opt_noquota:
1247                         if (sb_any_quota_enabled(sb)) {
1248                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1249                                         "options when quota turned on.\n");
1250                                 return 0;
1251                         }
1252                         clear_opt(sbi->s_mount_opt, QUOTA);
1253                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1254                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1255                         break;
1256 #else
1257                 case Opt_quota:
1258                 case Opt_usrquota:
1259                 case Opt_grpquota:
1260                         printk(KERN_ERR
1261                                 "EXT4-fs: quota options not supported.\n");
1262                         break;
1263                 case Opt_usrjquota:
1264                 case Opt_grpjquota:
1265                 case Opt_offusrjquota:
1266                 case Opt_offgrpjquota:
1267                 case Opt_jqfmt_vfsold:
1268                 case Opt_jqfmt_vfsv0:
1269                         printk(KERN_ERR
1270                                 "EXT4-fs: journaled quota options not "
1271                                 "supported.\n");
1272                         break;
1273                 case Opt_noquota:
1274                         break;
1275 #endif
1276                 case Opt_abort:
1277                         set_opt(sbi->s_mount_opt, ABORT);
1278                         break;
1279                 case Opt_barrier:
1280                         if (match_int(&args[0], &option))
1281                                 return 0;
1282                         if (option)
1283                                 set_opt(sbi->s_mount_opt, BARRIER);
1284                         else
1285                                 clear_opt(sbi->s_mount_opt, BARRIER);
1286                         break;
1287                 case Opt_ignore:
1288                         break;
1289                 case Opt_resize:
1290                         if (!is_remount) {
1291                                 printk("EXT4-fs: resize option only available "
1292                                         "for remount\n");
1293                                 return 0;
1294                         }
1295                         if (match_int(&args[0], &option) != 0)
1296                                 return 0;
1297                         *n_blocks_count = option;
1298                         break;
1299                 case Opt_nobh:
1300                         set_opt(sbi->s_mount_opt, NOBH);
1301                         break;
1302                 case Opt_bh:
1303                         clear_opt(sbi->s_mount_opt, NOBH);
1304                         break;
1305                 case Opt_extents:
1306                         set_opt (sbi->s_mount_opt, EXTENTS);
1307                         break;
1308                 case Opt_noextents:
1309                         clear_opt (sbi->s_mount_opt, EXTENTS);
1310                         break;
1311                 case Opt_i_version:
1312                         set_opt(sbi->s_mount_opt, I_VERSION);
1313                         sb->s_flags |= MS_I_VERSION;
1314                         break;
1315                 case Opt_mballoc:
1316                         set_opt(sbi->s_mount_opt, MBALLOC);
1317                         break;
1318                 case Opt_nomballoc:
1319                         clear_opt(sbi->s_mount_opt, MBALLOC);
1320                         break;
1321                 case Opt_stripe:
1322                         if (match_int(&args[0], &option))
1323                                 return 0;
1324                         if (option < 0)
1325                                 return 0;
1326                         sbi->s_stripe = option;
1327                         break;
1328                 default:
1329                         printk (KERN_ERR
1330                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1331                                 "or missing value\n", p);
1332                         return 0;
1333                 }
1334         }
1335 #ifdef CONFIG_QUOTA
1336         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1337                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1338                      sbi->s_qf_names[USRQUOTA])
1339                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1340
1341                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1342                      sbi->s_qf_names[GRPQUOTA])
1343                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1344
1345                 if ((sbi->s_qf_names[USRQUOTA] &&
1346                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1347                     (sbi->s_qf_names[GRPQUOTA] &&
1348                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1349                         printk(KERN_ERR "EXT4-fs: old and new quota "
1350                                         "format mixing.\n");
1351                         return 0;
1352                 }
1353
1354                 if (!sbi->s_jquota_fmt) {
1355                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1356                                         "not specified.\n");
1357                         return 0;
1358                 }
1359         } else {
1360                 if (sbi->s_jquota_fmt) {
1361                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1362                                         "specified with no journaling "
1363                                         "enabled.\n");
1364                         return 0;
1365                 }
1366         }
1367 #endif
1368         return 1;
1369 }
1370
1371 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1372                             int read_only)
1373 {
1374         struct ext4_sb_info *sbi = EXT4_SB(sb);
1375         int res = 0;
1376
1377         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1378                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1379                         "forcing read-only mode\n");
1380                 res = MS_RDONLY;
1381         }
1382         if (read_only)
1383                 return res;
1384         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1385                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1386                         "running e2fsck is recommended\n");
1387         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1388                 printk (KERN_WARNING
1389                         "EXT4-fs warning: mounting fs with errors, "
1390                         "running e2fsck is recommended\n");
1391         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1392                  le16_to_cpu(es->s_mnt_count) >=
1393                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1394                 printk (KERN_WARNING
1395                         "EXT4-fs warning: maximal mount count reached, "
1396                         "running e2fsck is recommended\n");
1397         else if (le32_to_cpu(es->s_checkinterval) &&
1398                 (le32_to_cpu(es->s_lastcheck) +
1399                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1400                 printk (KERN_WARNING
1401                         "EXT4-fs warning: checktime reached, "
1402                         "running e2fsck is recommended\n");
1403 #if 0
1404                 /* @@@ We _will_ want to clear the valid bit if we find
1405                  * inconsistencies, to force a fsck at reboot.  But for
1406                  * a plain journaled filesystem we can keep it set as
1407                  * valid forever! :)
1408                  */
1409         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1410 #endif
1411         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1412                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1413         le16_add_cpu(&es->s_mnt_count, 1);
1414         es->s_mtime = cpu_to_le32(get_seconds());
1415         ext4_update_dynamic_rev(sb);
1416         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1417
1418         ext4_commit_super(sb, es, 1);
1419         if (test_opt(sb, DEBUG))
1420                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1421                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1422                         sb->s_blocksize,
1423                         sbi->s_groups_count,
1424                         EXT4_BLOCKS_PER_GROUP(sb),
1425                         EXT4_INODES_PER_GROUP(sb),
1426                         sbi->s_mount_opt);
1427
1428         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1429         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1430                 char b[BDEVNAME_SIZE];
1431
1432                 printk("external journal on %s\n",
1433                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1434         } else {
1435                 printk("internal journal\n");
1436         }
1437         return res;
1438 }
1439
1440 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1441                             struct ext4_group_desc *gdp)
1442 {
1443         __u16 crc = 0;
1444
1445         if (sbi->s_es->s_feature_ro_compat &
1446             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1447                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1448                 __le32 le_group = cpu_to_le32(block_group);
1449
1450                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1451                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1452                 crc = crc16(crc, (__u8 *)gdp, offset);
1453                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1454                 /* for checksum of struct ext4_group_desc do the rest...*/
1455                 if ((sbi->s_es->s_feature_incompat &
1456                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1457                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1458                         crc = crc16(crc, (__u8 *)gdp + offset,
1459                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1460                                         offset);
1461         }
1462
1463         return cpu_to_le16(crc);
1464 }
1465
1466 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1467                                 struct ext4_group_desc *gdp)
1468 {
1469         if ((sbi->s_es->s_feature_ro_compat &
1470              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1471             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1472                 return 0;
1473
1474         return 1;
1475 }
1476
1477 /* Called at mount-time, super-block is locked */
1478 static int ext4_check_descriptors(struct super_block *sb)
1479 {
1480         struct ext4_sb_info *sbi = EXT4_SB(sb);
1481         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1482         ext4_fsblk_t last_block;
1483         ext4_fsblk_t block_bitmap;
1484         ext4_fsblk_t inode_bitmap;
1485         ext4_fsblk_t inode_table;
1486         int flexbg_flag = 0;
1487         ext4_group_t i;
1488
1489         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1490                 flexbg_flag = 1;
1491
1492         ext4_debug ("Checking group descriptors");
1493
1494         for (i = 0; i < sbi->s_groups_count; i++) {
1495                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1496
1497                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1498                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1499                 else
1500                         last_block = first_block +
1501                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1502
1503                 block_bitmap = ext4_block_bitmap(sb, gdp);
1504                 if (block_bitmap < first_block || block_bitmap > last_block)
1505                 {
1506                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1507                                "Block bitmap for group %lu not in group "
1508                                "(block %llu)!", i, block_bitmap);
1509                         return 0;
1510                 }
1511                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1512                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1513                 {
1514                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1515                                "Inode bitmap for group %lu not in group "
1516                                "(block %llu)!", i, inode_bitmap);
1517                         return 0;
1518                 }
1519                 inode_table = ext4_inode_table(sb, gdp);
1520                 if (inode_table < first_block ||
1521                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1522                 {
1523                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1524                                "Inode table for group %lu not in group "
1525                                "(block %llu)!", i, inode_table);
1526                         return 0;
1527                 }
1528                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1529                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1530                                "Checksum for group %lu failed (%u!=%u)\n",
1531                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1532                                gdp)), le16_to_cpu(gdp->bg_checksum));
1533                         return 0;
1534                 }
1535                 if (!flexbg_flag)
1536                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1537         }
1538
1539         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1540         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1541         return 1;
1542 }
1543
1544 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1545  * the superblock) which were deleted from all directories, but held open by
1546  * a process at the time of a crash.  We walk the list and try to delete these
1547  * inodes at recovery time (only with a read-write filesystem).
1548  *
1549  * In order to keep the orphan inode chain consistent during traversal (in
1550  * case of crash during recovery), we link each inode into the superblock
1551  * orphan list_head and handle it the same way as an inode deletion during
1552  * normal operation (which journals the operations for us).
1553  *
1554  * We only do an iget() and an iput() on each inode, which is very safe if we
1555  * accidentally point at an in-use or already deleted inode.  The worst that
1556  * can happen in this case is that we get a "bit already cleared" message from
1557  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1558  * e2fsck was run on this filesystem, and it must have already done the orphan
1559  * inode cleanup for us, so we can safely abort without any further action.
1560  */
1561 static void ext4_orphan_cleanup (struct super_block * sb,
1562                                  struct ext4_super_block * es)
1563 {
1564         unsigned int s_flags = sb->s_flags;
1565         int nr_orphans = 0, nr_truncates = 0;
1566 #ifdef CONFIG_QUOTA
1567         int i;
1568 #endif
1569         if (!es->s_last_orphan) {
1570                 jbd_debug(4, "no orphan inodes to clean up\n");
1571                 return;
1572         }
1573
1574         if (bdev_read_only(sb->s_bdev)) {
1575                 printk(KERN_ERR "EXT4-fs: write access "
1576                         "unavailable, skipping orphan cleanup.\n");
1577                 return;
1578         }
1579
1580         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1581                 if (es->s_last_orphan)
1582                         jbd_debug(1, "Errors on filesystem, "
1583                                   "clearing orphan list.\n");
1584                 es->s_last_orphan = 0;
1585                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1586                 return;
1587         }
1588
1589         if (s_flags & MS_RDONLY) {
1590                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1591                        sb->s_id);
1592                 sb->s_flags &= ~MS_RDONLY;
1593         }
1594 #ifdef CONFIG_QUOTA
1595         /* Needed for iput() to work correctly and not trash data */
1596         sb->s_flags |= MS_ACTIVE;
1597         /* Turn on quotas so that they are updated correctly */
1598         for (i = 0; i < MAXQUOTAS; i++) {
1599                 if (EXT4_SB(sb)->s_qf_names[i]) {
1600                         int ret = ext4_quota_on_mount(sb, i);
1601                         if (ret < 0)
1602                                 printk(KERN_ERR
1603                                         "EXT4-fs: Cannot turn on journaled "
1604                                         "quota: error %d\n", ret);
1605                 }
1606         }
1607 #endif
1608
1609         while (es->s_last_orphan) {
1610                 struct inode *inode;
1611
1612                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1613                 if (IS_ERR(inode)) {
1614                         es->s_last_orphan = 0;
1615                         break;
1616                 }
1617
1618                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1619                 DQUOT_INIT(inode);
1620                 if (inode->i_nlink) {
1621                         printk(KERN_DEBUG
1622                                 "%s: truncating inode %lu to %Ld bytes\n",
1623                                 __func__, inode->i_ino, inode->i_size);
1624                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1625                                   inode->i_ino, inode->i_size);
1626                         ext4_truncate(inode);
1627                         nr_truncates++;
1628                 } else {
1629                         printk(KERN_DEBUG
1630                                 "%s: deleting unreferenced inode %lu\n",
1631                                 __func__, inode->i_ino);
1632                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1633                                   inode->i_ino);
1634                         nr_orphans++;
1635                 }
1636                 iput(inode);  /* The delete magic happens here! */
1637         }
1638
1639 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1640
1641         if (nr_orphans)
1642                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1643                        sb->s_id, PLURAL(nr_orphans));
1644         if (nr_truncates)
1645                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1646                        sb->s_id, PLURAL(nr_truncates));
1647 #ifdef CONFIG_QUOTA
1648         /* Turn quotas off */
1649         for (i = 0; i < MAXQUOTAS; i++) {
1650                 if (sb_dqopt(sb)->files[i])
1651                         vfs_quota_off(sb, i, 0);
1652         }
1653 #endif
1654         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1655 }
1656 /*
1657  * Maximal extent format file size.
1658  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1659  * extent format containers, within a sector_t, and within i_blocks
1660  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1661  * so that won't be a limiting factor.
1662  *
1663  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1664  */
1665 static loff_t ext4_max_size(int blkbits)
1666 {
1667         loff_t res;
1668         loff_t upper_limit = MAX_LFS_FILESIZE;
1669
1670         /* small i_blocks in vfs inode? */
1671         if (sizeof(blkcnt_t) < sizeof(u64)) {
1672                 /*
1673                  * CONFIG_LSF is not enabled implies the inode
1674                  * i_block represent total blocks in 512 bytes
1675                  * 32 == size of vfs inode i_blocks * 8
1676                  */
1677                 upper_limit = (1LL << 32) - 1;
1678
1679                 /* total blocks in file system block size */
1680                 upper_limit >>= (blkbits - 9);
1681                 upper_limit <<= blkbits;
1682         }
1683
1684         /* 32-bit extent-start container, ee_block */
1685         res = 1LL << 32;
1686         res <<= blkbits;
1687         res -= 1;
1688
1689         /* Sanity check against vm- & vfs- imposed limits */
1690         if (res > upper_limit)
1691                 res = upper_limit;
1692
1693         return res;
1694 }
1695
1696 /*
1697  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1698  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1699  * We need to be 1 filesystem block less than the 2^48 sector limit.
1700  */
1701 static loff_t ext4_max_bitmap_size(int bits)
1702 {
1703         loff_t res = EXT4_NDIR_BLOCKS;
1704         int meta_blocks;
1705         loff_t upper_limit;
1706         /* This is calculated to be the largest file size for a
1707          * dense, bitmapped file such that the total number of
1708          * sectors in the file, including data and all indirect blocks,
1709          * does not exceed 2^48 -1
1710          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1711          * total number of  512 bytes blocks of the file
1712          */
1713
1714         if (sizeof(blkcnt_t) < sizeof(u64)) {
1715                 /*
1716                  * CONFIG_LSF is not enabled implies the inode
1717                  * i_block represent total blocks in 512 bytes
1718                  * 32 == size of vfs inode i_blocks * 8
1719                  */
1720                 upper_limit = (1LL << 32) - 1;
1721
1722                 /* total blocks in file system block size */
1723                 upper_limit >>= (bits - 9);
1724
1725         } else {
1726                 /*
1727                  * We use 48 bit ext4_inode i_blocks
1728                  * With EXT4_HUGE_FILE_FL set the i_blocks
1729                  * represent total number of blocks in
1730                  * file system block size
1731                  */
1732                 upper_limit = (1LL << 48) - 1;
1733
1734         }
1735
1736         /* indirect blocks */
1737         meta_blocks = 1;
1738         /* double indirect blocks */
1739         meta_blocks += 1 + (1LL << (bits-2));
1740         /* tripple indirect blocks */
1741         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1742
1743         upper_limit -= meta_blocks;
1744         upper_limit <<= bits;
1745
1746         res += 1LL << (bits-2);
1747         res += 1LL << (2*(bits-2));
1748         res += 1LL << (3*(bits-2));
1749         res <<= bits;
1750         if (res > upper_limit)
1751                 res = upper_limit;
1752
1753         if (res > MAX_LFS_FILESIZE)
1754                 res = MAX_LFS_FILESIZE;
1755
1756         return res;
1757 }
1758
1759 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1760                                 ext4_fsblk_t logical_sb_block, int nr)
1761 {
1762         struct ext4_sb_info *sbi = EXT4_SB(sb);
1763         ext4_group_t bg, first_meta_bg;
1764         int has_super = 0;
1765
1766         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1767
1768         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1769             nr < first_meta_bg)
1770                 return logical_sb_block + nr + 1;
1771         bg = sbi->s_desc_per_block * nr;
1772         if (ext4_bg_has_super(sb, bg))
1773                 has_super = 1;
1774         return (has_super + ext4_group_first_block_no(sb, bg));
1775 }
1776
1777 /**
1778  * ext4_get_stripe_size: Get the stripe size.
1779  * @sbi: In memory super block info
1780  *
1781  * If we have specified it via mount option, then
1782  * use the mount option value. If the value specified at mount time is
1783  * greater than the blocks per group use the super block value.
1784  * If the super block value is greater than blocks per group return 0.
1785  * Allocator needs it be less than blocks per group.
1786  *
1787  */
1788 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1789 {
1790         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1791         unsigned long stripe_width =
1792                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1793
1794         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1795                 return sbi->s_stripe;
1796
1797         if (stripe_width <= sbi->s_blocks_per_group)
1798                 return stripe_width;
1799
1800         if (stride <= sbi->s_blocks_per_group)
1801                 return stride;
1802
1803         return 0;
1804 }
1805
1806 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1807                                 __releases(kernel_sem)
1808                                 __acquires(kernel_sem)
1809
1810 {
1811         struct buffer_head * bh;
1812         struct ext4_super_block *es = NULL;
1813         struct ext4_sb_info *sbi;
1814         ext4_fsblk_t block;
1815         ext4_fsblk_t sb_block = get_sb_block(&data);
1816         ext4_fsblk_t logical_sb_block;
1817         unsigned long offset = 0;
1818         unsigned int journal_inum = 0;
1819         unsigned long journal_devnum = 0;
1820         unsigned long def_mount_opts;
1821         struct inode *root;
1822         int ret = -EINVAL;
1823         int blocksize;
1824         int db_count;
1825         int i;
1826         int needs_recovery;
1827         __le32 features;
1828         __u64 blocks_count;
1829         int err;
1830
1831         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1832         if (!sbi)
1833                 return -ENOMEM;
1834         sb->s_fs_info = sbi;
1835         sbi->s_mount_opt = 0;
1836         sbi->s_resuid = EXT4_DEF_RESUID;
1837         sbi->s_resgid = EXT4_DEF_RESGID;
1838         sbi->s_sb_block = sb_block;
1839
1840         unlock_kernel();
1841
1842         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1843         if (!blocksize) {
1844                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1845                 goto out_fail;
1846         }
1847
1848         if (!sb_set_blocksize(sb, blocksize)) {
1849                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1850                 goto out_fail;
1851         }
1852
1853         /*
1854          * The ext4 superblock will not be buffer aligned for other than 1kB
1855          * block sizes.  We need to calculate the offset from buffer start.
1856          */
1857         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1858                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1859                 offset = do_div(logical_sb_block, blocksize);
1860         } else {
1861                 logical_sb_block = sb_block;
1862         }
1863
1864         if (!(bh = sb_bread(sb, logical_sb_block))) {
1865                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1866                 goto out_fail;
1867         }
1868         /*
1869          * Note: s_es must be initialized as soon as possible because
1870          *       some ext4 macro-instructions depend on its value
1871          */
1872         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1873         sbi->s_es = es;
1874         sb->s_magic = le16_to_cpu(es->s_magic);
1875         if (sb->s_magic != EXT4_SUPER_MAGIC)
1876                 goto cantfind_ext4;
1877
1878         /* Set defaults before we parse the mount options */
1879         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1880         if (def_mount_opts & EXT4_DEFM_DEBUG)
1881                 set_opt(sbi->s_mount_opt, DEBUG);
1882         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1883                 set_opt(sbi->s_mount_opt, GRPID);
1884         if (def_mount_opts & EXT4_DEFM_UID16)
1885                 set_opt(sbi->s_mount_opt, NO_UID32);
1886 #ifdef CONFIG_EXT4DEV_FS_XATTR
1887         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1888                 set_opt(sbi->s_mount_opt, XATTR_USER);
1889 #endif
1890 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1891         if (def_mount_opts & EXT4_DEFM_ACL)
1892                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1893 #endif
1894         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1895                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1896         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1897                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1898         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1899                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1900
1901         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1902                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1903         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1904                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1905         else
1906                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1907
1908         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1909         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1910
1911         set_opt(sbi->s_mount_opt, RESERVATION);
1912
1913         /*
1914          * turn on extents feature by default in ext4 filesystem
1915          * User -o noextents to turn it off
1916          */
1917         set_opt(sbi->s_mount_opt, EXTENTS);
1918         /*
1919          * turn on mballoc feature by default in ext4 filesystem
1920          * User -o nomballoc to turn it off
1921          */
1922         set_opt(sbi->s_mount_opt, MBALLOC);
1923
1924         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1925                             NULL, 0))
1926                 goto failed_mount;
1927
1928         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1929                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1930
1931         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1932             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1933              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1934              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1935                 printk(KERN_WARNING
1936                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1937                        "running e2fsck is recommended\n");
1938
1939         /*
1940          * Since ext4 is still considered development code, we require
1941          * that the TEST_FILESYS flag in s->flags be set.
1942          */
1943         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
1944                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
1945                        "OK to use with test code.\n", sb->s_id);
1946                 goto failed_mount;
1947         }
1948
1949         /*
1950          * Check feature flags regardless of the revision level, since we
1951          * previously didn't change the revision level when setting the flags,
1952          * so there is a chance incompat flags are set on a rev 0 filesystem.
1953          */
1954         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1955         if (features) {
1956                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1957                        "unsupported optional features (%x).\n",
1958                        sb->s_id, le32_to_cpu(features));
1959                 goto failed_mount;
1960         }
1961         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1962         if (!(sb->s_flags & MS_RDONLY) && features) {
1963                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1964                        "unsupported optional features (%x).\n",
1965                        sb->s_id, le32_to_cpu(features));
1966                 goto failed_mount;
1967         }
1968         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
1969                 /*
1970                  * Large file size enabled file system can only be
1971                  * mount if kernel is build with CONFIG_LSF
1972                  */
1973                 if (sizeof(root->i_blocks) < sizeof(u64) &&
1974                                 !(sb->s_flags & MS_RDONLY)) {
1975                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
1976                                         "files cannot be mounted read-write "
1977                                         "without CONFIG_LSF.\n", sb->s_id);
1978                         goto failed_mount;
1979                 }
1980         }
1981         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1982
1983         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1984             blocksize > EXT4_MAX_BLOCK_SIZE) {
1985                 printk(KERN_ERR
1986                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1987                        blocksize, sb->s_id);
1988                 goto failed_mount;
1989         }
1990
1991         if (sb->s_blocksize != blocksize) {
1992
1993                 /* Validate the filesystem blocksize */
1994                 if (!sb_set_blocksize(sb, blocksize)) {
1995                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
1996                                         blocksize);
1997                         goto failed_mount;
1998                 }
1999
2000                 brelse (bh);
2001                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2002                 offset = do_div(logical_sb_block, blocksize);
2003                 bh = sb_bread(sb, logical_sb_block);
2004                 if (!bh) {
2005                         printk(KERN_ERR
2006                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2007                         goto failed_mount;
2008                 }
2009                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2010                 sbi->s_es = es;
2011                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2012                         printk (KERN_ERR
2013                                 "EXT4-fs: Magic mismatch, very weird !\n");
2014                         goto failed_mount;
2015                 }
2016         }
2017
2018         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2019         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2020
2021         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2022                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2023                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2024         } else {
2025                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2026                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2027                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2028                     (!is_power_of_2(sbi->s_inode_size)) ||
2029                     (sbi->s_inode_size > blocksize)) {
2030                         printk (KERN_ERR
2031                                 "EXT4-fs: unsupported inode size: %d\n",
2032                                 sbi->s_inode_size);
2033                         goto failed_mount;
2034                 }
2035                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2036                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2037         }
2038         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2039         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2040                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2041                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2042                     !is_power_of_2(sbi->s_desc_size)) {
2043                         printk(KERN_ERR
2044                                "EXT4-fs: unsupported descriptor size %lu\n",
2045                                sbi->s_desc_size);
2046                         goto failed_mount;
2047                 }
2048         } else
2049                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2050         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2051         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2052         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2053                 goto cantfind_ext4;
2054         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2055         if (sbi->s_inodes_per_block == 0)
2056                 goto cantfind_ext4;
2057         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2058                                         sbi->s_inodes_per_block;
2059         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2060         sbi->s_sbh = bh;
2061         sbi->s_mount_state = le16_to_cpu(es->s_state);
2062         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2063         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2064         for (i=0; i < 4; i++)
2065                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2066         sbi->s_def_hash_version = es->s_def_hash_version;
2067
2068         if (sbi->s_blocks_per_group > blocksize * 8) {
2069                 printk (KERN_ERR
2070                         "EXT4-fs: #blocks per group too big: %lu\n",
2071                         sbi->s_blocks_per_group);
2072                 goto failed_mount;
2073         }
2074         if (sbi->s_inodes_per_group > blocksize * 8) {
2075                 printk (KERN_ERR
2076                         "EXT4-fs: #inodes per group too big: %lu\n",
2077                         sbi->s_inodes_per_group);
2078                 goto failed_mount;
2079         }
2080
2081         if (ext4_blocks_count(es) >
2082                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2083                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2084                         " too large to mount safely\n", sb->s_id);
2085                 if (sizeof(sector_t) < 8)
2086                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2087                                         "enabled\n");
2088                 goto failed_mount;
2089         }
2090
2091         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2092                 goto cantfind_ext4;
2093
2094         /* ensure blocks_count calculation below doesn't sign-extend */
2095         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2096             le32_to_cpu(es->s_first_data_block) + 1) {
2097                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2098                        "first data block %u, blocks per group %lu\n",
2099                         ext4_blocks_count(es),
2100                         le32_to_cpu(es->s_first_data_block),
2101                         EXT4_BLOCKS_PER_GROUP(sb));
2102                 goto failed_mount;
2103         }
2104         blocks_count = (ext4_blocks_count(es) -
2105                         le32_to_cpu(es->s_first_data_block) +
2106                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2107         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2108         sbi->s_groups_count = blocks_count;
2109         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2110                    EXT4_DESC_PER_BLOCK(sb);
2111         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2112                                     GFP_KERNEL);
2113         if (sbi->s_group_desc == NULL) {
2114                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2115                 goto failed_mount;
2116         }
2117
2118         bgl_lock_init(&sbi->s_blockgroup_lock);
2119
2120         for (i = 0; i < db_count; i++) {
2121                 block = descriptor_loc(sb, logical_sb_block, i);
2122                 sbi->s_group_desc[i] = sb_bread(sb, block);
2123                 if (!sbi->s_group_desc[i]) {
2124                         printk (KERN_ERR "EXT4-fs: "
2125                                 "can't read group descriptor %d\n", i);
2126                         db_count = i;
2127                         goto failed_mount2;
2128                 }
2129         }
2130         if (!ext4_check_descriptors (sb)) {
2131                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2132                 goto failed_mount2;
2133         }
2134         sbi->s_gdb_count = db_count;
2135         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2136         spin_lock_init(&sbi->s_next_gen_lock);
2137
2138         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2139                         ext4_count_free_blocks(sb));
2140         if (!err) {
2141                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2142                                 ext4_count_free_inodes(sb));
2143         }
2144         if (!err) {
2145                 err = percpu_counter_init(&sbi->s_dirs_counter,
2146                                 ext4_count_dirs(sb));
2147         }
2148         if (err) {
2149                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2150                 goto failed_mount3;
2151         }
2152
2153         /* per fileystem reservation list head & lock */
2154         spin_lock_init(&sbi->s_rsv_window_lock);
2155         sbi->s_rsv_window_root = RB_ROOT;
2156         /* Add a single, static dummy reservation to the start of the
2157          * reservation window list --- it gives us a placeholder for
2158          * append-at-start-of-list which makes the allocation logic
2159          * _much_ simpler. */
2160         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2161         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2162         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2163         sbi->s_rsv_window_head.rsv_goal_size = 0;
2164         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2165
2166         sbi->s_stripe = ext4_get_stripe_size(sbi);
2167
2168         /*
2169          * set up enough so that it can read an inode
2170          */
2171         sb->s_op = &ext4_sops;
2172         sb->s_export_op = &ext4_export_ops;
2173         sb->s_xattr = ext4_xattr_handlers;
2174 #ifdef CONFIG_QUOTA
2175         sb->s_qcop = &ext4_qctl_operations;
2176         sb->dq_op = &ext4_quota_operations;
2177 #endif
2178         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2179
2180         sb->s_root = NULL;
2181
2182         needs_recovery = (es->s_last_orphan != 0 ||
2183                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2184                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2185
2186         /*
2187          * The first inode we look at is the journal inode.  Don't try
2188          * root first: it may be modified in the journal!
2189          */
2190         if (!test_opt(sb, NOLOAD) &&
2191             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2192                 if (ext4_load_journal(sb, es, journal_devnum))
2193                         goto failed_mount3;
2194                 if (!(sb->s_flags & MS_RDONLY) &&
2195                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2196                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2197                                "ext4_fill_super: Journal transaction "
2198                                "%u is corrupt\n", sb->s_id,
2199                                EXT4_SB(sb)->s_journal->j_failed_commit);
2200                         if (test_opt (sb, ERRORS_RO)) {
2201                                 printk (KERN_CRIT
2202                                         "Mounting filesystem read-only\n");
2203                                 sb->s_flags |= MS_RDONLY;
2204                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2205                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2206                         }
2207                         if (test_opt(sb, ERRORS_PANIC)) {
2208                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2209                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2210                                 ext4_commit_super(sb, es, 1);
2211                                 printk(KERN_CRIT
2212                                        "EXT4-fs (device %s): mount failed\n",
2213                                       sb->s_id);
2214                                 goto failed_mount4;
2215                         }
2216                 }
2217         } else if (journal_inum) {
2218                 if (ext4_create_journal(sb, es, journal_inum))
2219                         goto failed_mount3;
2220         } else {
2221                 if (!silent)
2222                         printk (KERN_ERR
2223                                 "ext4: No journal on filesystem on %s\n",
2224                                 sb->s_id);
2225                 goto failed_mount3;
2226         }
2227
2228         if (ext4_blocks_count(es) > 0xffffffffULL &&
2229             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2230                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2231                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2232                 goto failed_mount4;
2233         }
2234
2235         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2236                 jbd2_journal_set_features(sbi->s_journal,
2237                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2238                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2239         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2240                 jbd2_journal_set_features(sbi->s_journal,
2241                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2242                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2243                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2244         } else {
2245                 jbd2_journal_clear_features(sbi->s_journal,
2246                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2247                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2248         }
2249
2250         /* We have now updated the journal if required, so we can
2251          * validate the data journaling mode. */
2252         switch (test_opt(sb, DATA_FLAGS)) {
2253         case 0:
2254                 /* No mode set, assume a default based on the journal
2255                  * capabilities: ORDERED_DATA if the journal can
2256                  * cope, else JOURNAL_DATA
2257                  */
2258                 if (jbd2_journal_check_available_features
2259                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2260                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2261                 else
2262                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2263                 break;
2264
2265         case EXT4_MOUNT_ORDERED_DATA:
2266         case EXT4_MOUNT_WRITEBACK_DATA:
2267                 if (!jbd2_journal_check_available_features
2268                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2269                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2270                                "requested data journaling mode\n");
2271                         goto failed_mount4;
2272                 }
2273         default:
2274                 break;
2275         }
2276
2277         if (test_opt(sb, NOBH)) {
2278                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2279                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2280                                 "its supported only with writeback mode\n");
2281                         clear_opt(sbi->s_mount_opt, NOBH);
2282                 }
2283         }
2284         /*
2285          * The jbd2_journal_load will have done any necessary log recovery,
2286          * so we can safely mount the rest of the filesystem now.
2287          */
2288
2289         root = ext4_iget(sb, EXT4_ROOT_INO);
2290         if (IS_ERR(root)) {
2291                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2292                 ret = PTR_ERR(root);
2293                 goto failed_mount4;
2294         }
2295         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2296                 iput(root);
2297                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2298                 goto failed_mount4;
2299         }
2300         sb->s_root = d_alloc_root(root);
2301         if (!sb->s_root) {
2302                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2303                 iput(root);
2304                 ret = -ENOMEM;
2305                 goto failed_mount4;
2306         }
2307
2308         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2309
2310         /* determine the minimum size of new large inodes, if present */
2311         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2312                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2313                                                      EXT4_GOOD_OLD_INODE_SIZE;
2314                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2315                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2316                         if (sbi->s_want_extra_isize <
2317                             le16_to_cpu(es->s_want_extra_isize))
2318                                 sbi->s_want_extra_isize =
2319                                         le16_to_cpu(es->s_want_extra_isize);
2320                         if (sbi->s_want_extra_isize <
2321                             le16_to_cpu(es->s_min_extra_isize))
2322                                 sbi->s_want_extra_isize =
2323                                         le16_to_cpu(es->s_min_extra_isize);
2324                 }
2325         }
2326         /* Check if enough inode space is available */
2327         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2328                                                         sbi->s_inode_size) {
2329                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2330                                                        EXT4_GOOD_OLD_INODE_SIZE;
2331                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2332                         "available.\n");
2333         }
2334
2335         /*
2336          * akpm: core read_super() calls in here with the superblock locked.
2337          * That deadlocks, because orphan cleanup needs to lock the superblock
2338          * in numerous places.  Here we just pop the lock - it's relatively
2339          * harmless, because we are now ready to accept write_super() requests,
2340          * and aviro says that's the only reason for hanging onto the
2341          * superblock lock.
2342          */
2343         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2344         ext4_orphan_cleanup(sb, es);
2345         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2346         if (needs_recovery)
2347                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2348         ext4_mark_recovery_complete(sb, es);
2349         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2350                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2351                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2352                 "writeback");
2353
2354         ext4_ext_init(sb);
2355         ext4_mb_init(sb, needs_recovery);
2356
2357         lock_kernel();
2358         return 0;
2359
2360 cantfind_ext4:
2361         if (!silent)
2362                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2363                        sb->s_id);
2364         goto failed_mount;
2365
2366 failed_mount4:
2367         jbd2_journal_destroy(sbi->s_journal);
2368 failed_mount3:
2369         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2370         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2371         percpu_counter_destroy(&sbi->s_dirs_counter);
2372 failed_mount2:
2373         for (i = 0; i < db_count; i++)
2374                 brelse(sbi->s_group_desc[i]);
2375         kfree(sbi->s_group_desc);
2376 failed_mount:
2377 #ifdef CONFIG_QUOTA
2378         for (i = 0; i < MAXQUOTAS; i++)
2379                 kfree(sbi->s_qf_names[i]);
2380 #endif
2381         ext4_blkdev_remove(sbi);
2382         brelse(bh);
2383 out_fail:
2384         sb->s_fs_info = NULL;
2385         kfree(sbi);
2386         lock_kernel();
2387         return ret;
2388 }
2389
2390 /*
2391  * Setup any per-fs journal parameters now.  We'll do this both on
2392  * initial mount, once the journal has been initialised but before we've
2393  * done any recovery; and again on any subsequent remount.
2394  */
2395 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2396 {
2397         struct ext4_sb_info *sbi = EXT4_SB(sb);
2398
2399         if (sbi->s_commit_interval)
2400                 journal->j_commit_interval = sbi->s_commit_interval;
2401         /* We could also set up an ext4-specific default for the commit
2402          * interval here, but for now we'll just fall back to the jbd
2403          * default. */
2404
2405         spin_lock(&journal->j_state_lock);
2406         if (test_opt(sb, BARRIER))
2407                 journal->j_flags |= JBD2_BARRIER;
2408         else
2409                 journal->j_flags &= ~JBD2_BARRIER;
2410         spin_unlock(&journal->j_state_lock);
2411 }
2412
2413 static journal_t *ext4_get_journal(struct super_block *sb,
2414                                    unsigned int journal_inum)
2415 {
2416         struct inode *journal_inode;
2417         journal_t *journal;
2418
2419         /* First, test for the existence of a valid inode on disk.  Bad
2420          * things happen if we iget() an unused inode, as the subsequent
2421          * iput() will try to delete it. */
2422
2423         journal_inode = ext4_iget(sb, journal_inum);
2424         if (IS_ERR(journal_inode)) {
2425                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2426                 return NULL;
2427         }
2428         if (!journal_inode->i_nlink) {
2429                 make_bad_inode(journal_inode);
2430                 iput(journal_inode);
2431                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2432                 return NULL;
2433         }
2434
2435         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2436                   journal_inode, journal_inode->i_size);
2437         if (!S_ISREG(journal_inode->i_mode)) {
2438                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2439                 iput(journal_inode);
2440                 return NULL;
2441         }
2442
2443         journal = jbd2_journal_init_inode(journal_inode);
2444         if (!journal) {
2445                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2446                 iput(journal_inode);
2447                 return NULL;
2448         }
2449         journal->j_private = sb;
2450         ext4_init_journal_params(sb, journal);
2451         return journal;
2452 }
2453
2454 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2455                                        dev_t j_dev)
2456 {
2457         struct buffer_head * bh;
2458         journal_t *journal;
2459         ext4_fsblk_t start;
2460         ext4_fsblk_t len;
2461         int hblock, blocksize;
2462         ext4_fsblk_t sb_block;
2463         unsigned long offset;
2464         struct ext4_super_block * es;
2465         struct block_device *bdev;
2466
2467         bdev = ext4_blkdev_get(j_dev);
2468         if (bdev == NULL)
2469                 return NULL;
2470
2471         if (bd_claim(bdev, sb)) {
2472                 printk(KERN_ERR
2473                         "EXT4: failed to claim external journal device.\n");
2474                 blkdev_put(bdev);
2475                 return NULL;
2476         }
2477
2478         blocksize = sb->s_blocksize;
2479         hblock = bdev_hardsect_size(bdev);
2480         if (blocksize < hblock) {
2481                 printk(KERN_ERR
2482                         "EXT4-fs: blocksize too small for journal device.\n");
2483                 goto out_bdev;
2484         }
2485
2486         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2487         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2488         set_blocksize(bdev, blocksize);
2489         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2490                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2491                        "external journal\n");
2492                 goto out_bdev;
2493         }
2494
2495         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2496         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2497             !(le32_to_cpu(es->s_feature_incompat) &
2498               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2499                 printk(KERN_ERR "EXT4-fs: external journal has "
2500                                         "bad superblock\n");
2501                 brelse(bh);
2502                 goto out_bdev;
2503         }
2504
2505         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2506                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2507                 brelse(bh);
2508                 goto out_bdev;
2509         }
2510
2511         len = ext4_blocks_count(es);
2512         start = sb_block + 1;
2513         brelse(bh);     /* we're done with the superblock */
2514
2515         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2516                                         start, len, blocksize);
2517         if (!journal) {
2518                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2519                 goto out_bdev;
2520         }
2521         journal->j_private = sb;
2522         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2523         wait_on_buffer(journal->j_sb_buffer);
2524         if (!buffer_uptodate(journal->j_sb_buffer)) {
2525                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2526                 goto out_journal;
2527         }
2528         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2529                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2530                                         "user (unsupported) - %d\n",
2531                         be32_to_cpu(journal->j_superblock->s_nr_users));
2532                 goto out_journal;
2533         }
2534         EXT4_SB(sb)->journal_bdev = bdev;
2535         ext4_init_journal_params(sb, journal);
2536         return journal;
2537 out_journal:
2538         jbd2_journal_destroy(journal);
2539 out_bdev:
2540         ext4_blkdev_put(bdev);
2541         return NULL;
2542 }
2543
2544 static int ext4_load_journal(struct super_block *sb,
2545                              struct ext4_super_block *es,
2546                              unsigned long journal_devnum)
2547 {
2548         journal_t *journal;
2549         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2550         dev_t journal_dev;
2551         int err = 0;
2552         int really_read_only;
2553
2554         if (journal_devnum &&
2555             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2556                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2557                         "numbers have changed\n");
2558                 journal_dev = new_decode_dev(journal_devnum);
2559         } else
2560                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2561
2562         really_read_only = bdev_read_only(sb->s_bdev);
2563
2564         /*
2565          * Are we loading a blank journal or performing recovery after a
2566          * crash?  For recovery, we need to check in advance whether we
2567          * can get read-write access to the device.
2568          */
2569
2570         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2571                 if (sb->s_flags & MS_RDONLY) {
2572                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2573                                         "required on readonly filesystem.\n");
2574                         if (really_read_only) {
2575                                 printk(KERN_ERR "EXT4-fs: write access "
2576                                         "unavailable, cannot proceed.\n");
2577                                 return -EROFS;
2578                         }
2579                         printk (KERN_INFO "EXT4-fs: write access will "
2580                                         "be enabled during recovery.\n");
2581                 }
2582         }
2583
2584         if (journal_inum && journal_dev) {
2585                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2586                        "and inode journals!\n");
2587                 return -EINVAL;
2588         }
2589
2590         if (journal_inum) {
2591                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2592                         return -EINVAL;
2593         } else {
2594                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2595                         return -EINVAL;
2596         }
2597
2598         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2599                 err = jbd2_journal_update_format(journal);
2600                 if (err)  {
2601                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2602                         jbd2_journal_destroy(journal);
2603                         return err;
2604                 }
2605         }
2606
2607         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2608                 err = jbd2_journal_wipe(journal, !really_read_only);
2609         if (!err)
2610                 err = jbd2_journal_load(journal);
2611
2612         if (err) {
2613                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2614                 jbd2_journal_destroy(journal);
2615                 return err;
2616         }
2617
2618         EXT4_SB(sb)->s_journal = journal;
2619         ext4_clear_journal_err(sb, es);
2620
2621         if (journal_devnum &&
2622             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2623                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2624                 sb->s_dirt = 1;
2625
2626                 /* Make sure we flush the recovery flag to disk. */
2627                 ext4_commit_super(sb, es, 1);
2628         }
2629
2630         return 0;
2631 }
2632
2633 static int ext4_create_journal(struct super_block * sb,
2634                                struct ext4_super_block * es,
2635                                unsigned int journal_inum)
2636 {
2637         journal_t *journal;
2638         int err;
2639
2640         if (sb->s_flags & MS_RDONLY) {
2641                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2642                                 "create journal.\n");
2643                 return -EROFS;
2644         }
2645
2646         journal = ext4_get_journal(sb, journal_inum);
2647         if (!journal)
2648                 return -EINVAL;
2649
2650         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2651                journal_inum);
2652
2653         err = jbd2_journal_create(journal);
2654         if (err) {
2655                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2656                 jbd2_journal_destroy(journal);
2657                 return -EIO;
2658         }
2659
2660         EXT4_SB(sb)->s_journal = journal;
2661
2662         ext4_update_dynamic_rev(sb);
2663         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2664         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2665
2666         es->s_journal_inum = cpu_to_le32(journal_inum);
2667         sb->s_dirt = 1;
2668
2669         /* Make sure we flush the recovery flag to disk. */
2670         ext4_commit_super(sb, es, 1);
2671
2672         return 0;
2673 }
2674
2675 static void ext4_commit_super (struct super_block * sb,
2676                                struct ext4_super_block * es,
2677                                int sync)
2678 {
2679         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2680
2681         if (!sbh)
2682                 return;
2683         es->s_wtime = cpu_to_le32(get_seconds());
2684         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2685         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2686         BUFFER_TRACE(sbh, "marking dirty");
2687         mark_buffer_dirty(sbh);
2688         if (sync)
2689                 sync_dirty_buffer(sbh);
2690 }
2691
2692
2693 /*
2694  * Have we just finished recovery?  If so, and if we are mounting (or
2695  * remounting) the filesystem readonly, then we will end up with a
2696  * consistent fs on disk.  Record that fact.
2697  */
2698 static void ext4_mark_recovery_complete(struct super_block * sb,
2699                                         struct ext4_super_block * es)
2700 {
2701         journal_t *journal = EXT4_SB(sb)->s_journal;
2702
2703         jbd2_journal_lock_updates(journal);
2704         jbd2_journal_flush(journal);
2705         lock_super(sb);
2706         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2707             sb->s_flags & MS_RDONLY) {
2708                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2709                 sb->s_dirt = 0;
2710                 ext4_commit_super(sb, es, 1);
2711         }
2712         unlock_super(sb);
2713         jbd2_journal_unlock_updates(journal);
2714 }
2715
2716 /*
2717  * If we are mounting (or read-write remounting) a filesystem whose journal
2718  * has recorded an error from a previous lifetime, move that error to the
2719  * main filesystem now.
2720  */
2721 static void ext4_clear_journal_err(struct super_block * sb,
2722                                    struct ext4_super_block * es)
2723 {
2724         journal_t *journal;
2725         int j_errno;
2726         const char *errstr;
2727
2728         journal = EXT4_SB(sb)->s_journal;
2729
2730         /*
2731          * Now check for any error status which may have been recorded in the
2732          * journal by a prior ext4_error() or ext4_abort()
2733          */
2734
2735         j_errno = jbd2_journal_errno(journal);
2736         if (j_errno) {
2737                 char nbuf[16];
2738
2739                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2740                 ext4_warning(sb, __func__, "Filesystem error recorded "
2741                              "from previous mount: %s", errstr);
2742                 ext4_warning(sb, __func__, "Marking fs in need of "
2743                              "filesystem check.");
2744
2745                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2746                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2747                 ext4_commit_super (sb, es, 1);
2748
2749                 jbd2_journal_clear_err(journal);
2750         }
2751 }
2752
2753 /*
2754  * Force the running and committing transactions to commit,
2755  * and wait on the commit.
2756  */
2757 int ext4_force_commit(struct super_block *sb)
2758 {
2759         journal_t *journal;
2760         int ret;
2761
2762         if (sb->s_flags & MS_RDONLY)
2763                 return 0;
2764
2765         journal = EXT4_SB(sb)->s_journal;
2766         sb->s_dirt = 0;
2767         ret = ext4_journal_force_commit(journal);
2768         return ret;
2769 }
2770
2771 /*
2772  * Ext4 always journals updates to the superblock itself, so we don't
2773  * have to propagate any other updates to the superblock on disk at this
2774  * point.  Just start an async writeback to get the buffers on their way
2775  * to the disk.
2776  *
2777  * This implicitly triggers the writebehind on sync().
2778  */
2779
2780 static void ext4_write_super (struct super_block * sb)
2781 {
2782         if (mutex_trylock(&sb->s_lock) != 0)
2783                 BUG();
2784         sb->s_dirt = 0;
2785 }
2786
2787 static int ext4_sync_fs(struct super_block *sb, int wait)
2788 {
2789         tid_t target;
2790
2791         sb->s_dirt = 0;
2792         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2793                 if (wait)
2794                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2795         }
2796         return 0;
2797 }
2798
2799 /*
2800  * LVM calls this function before a (read-only) snapshot is created.  This
2801  * gives us a chance to flush the journal completely and mark the fs clean.
2802  */
2803 static void ext4_write_super_lockfs(struct super_block *sb)
2804 {
2805         sb->s_dirt = 0;
2806
2807         if (!(sb->s_flags & MS_RDONLY)) {
2808                 journal_t *journal = EXT4_SB(sb)->s_journal;
2809
2810                 /* Now we set up the journal barrier. */
2811                 jbd2_journal_lock_updates(journal);
2812                 jbd2_journal_flush(journal);
2813
2814                 /* Journal blocked and flushed, clear needs_recovery flag. */
2815                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2816                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2817         }
2818 }
2819
2820 /*
2821  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2822  * flag here, even though the filesystem is not technically dirty yet.
2823  */
2824 static void ext4_unlockfs(struct super_block *sb)
2825 {
2826         if (!(sb->s_flags & MS_RDONLY)) {
2827                 lock_super(sb);
2828                 /* Reser the needs_recovery flag before the fs is unlocked. */
2829                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2830                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2831                 unlock_super(sb);
2832                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2833         }
2834 }
2835
2836 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2837 {
2838         struct ext4_super_block * es;
2839         struct ext4_sb_info *sbi = EXT4_SB(sb);
2840         ext4_fsblk_t n_blocks_count = 0;
2841         unsigned long old_sb_flags;
2842         struct ext4_mount_options old_opts;
2843         int err;
2844 #ifdef CONFIG_QUOTA
2845         int i;
2846 #endif
2847
2848         /* Store the original options */
2849         old_sb_flags = sb->s_flags;
2850         old_opts.s_mount_opt = sbi->s_mount_opt;
2851         old_opts.s_resuid = sbi->s_resuid;
2852         old_opts.s_resgid = sbi->s_resgid;
2853         old_opts.s_commit_interval = sbi->s_commit_interval;
2854 #ifdef CONFIG_QUOTA
2855         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2856         for (i = 0; i < MAXQUOTAS; i++)
2857                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2858 #endif
2859
2860         /*
2861          * Allow the "check" option to be passed as a remount option.
2862          */
2863         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2864                 err = -EINVAL;
2865                 goto restore_opts;
2866         }
2867
2868         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2869                 ext4_abort(sb, __func__, "Abort forced by user");
2870
2871         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2872                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2873
2874         es = sbi->s_es;
2875
2876         ext4_init_journal_params(sb, sbi->s_journal);
2877
2878         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2879                 n_blocks_count > ext4_blocks_count(es)) {
2880                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2881                         err = -EROFS;
2882                         goto restore_opts;
2883                 }
2884
2885                 if (*flags & MS_RDONLY) {
2886                         /*
2887                          * First of all, the unconditional stuff we have to do
2888                          * to disable replay of the journal when we next remount
2889                          */
2890                         sb->s_flags |= MS_RDONLY;
2891
2892                         /*
2893                          * OK, test if we are remounting a valid rw partition
2894                          * readonly, and if so set the rdonly flag and then
2895                          * mark the partition as valid again.
2896                          */
2897                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2898                             (sbi->s_mount_state & EXT4_VALID_FS))
2899                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2900
2901                         /*
2902                          * We have to unlock super so that we can wait for
2903                          * transactions.
2904                          */
2905                         unlock_super(sb);
2906                         ext4_mark_recovery_complete(sb, es);
2907                         lock_super(sb);
2908                 } else {
2909                         __le32 ret;
2910                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2911                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2912                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2913                                        "remount RDWR because of unsupported "
2914                                        "optional features (%x).\n",
2915                                        sb->s_id, le32_to_cpu(ret));
2916                                 err = -EROFS;
2917                                 goto restore_opts;
2918                         }
2919
2920                         /*
2921                          * If we have an unprocessed orphan list hanging
2922                          * around from a previously readonly bdev mount,
2923                          * require a full umount/remount for now.
2924                          */
2925                         if (es->s_last_orphan) {
2926                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2927                                        "remount RDWR because of unprocessed "
2928                                        "orphan inode list.  Please "
2929                                        "umount/remount instead.\n",
2930                                        sb->s_id);
2931                                 err = -EINVAL;
2932                                 goto restore_opts;
2933                         }
2934
2935                         /*
2936                          * Mounting a RDONLY partition read-write, so reread
2937                          * and store the current valid flag.  (It may have
2938                          * been changed by e2fsck since we originally mounted
2939                          * the partition.)
2940                          */
2941                         ext4_clear_journal_err(sb, es);
2942                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2943                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2944                                 goto restore_opts;
2945                         if (!ext4_setup_super (sb, es, 0))
2946                                 sb->s_flags &= ~MS_RDONLY;
2947                 }
2948         }
2949 #ifdef CONFIG_QUOTA
2950         /* Release old quota file names */
2951         for (i = 0; i < MAXQUOTAS; i++)
2952                 if (old_opts.s_qf_names[i] &&
2953                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2954                         kfree(old_opts.s_qf_names[i]);
2955 #endif
2956         return 0;
2957 restore_opts:
2958         sb->s_flags = old_sb_flags;
2959         sbi->s_mount_opt = old_opts.s_mount_opt;
2960         sbi->s_resuid = old_opts.s_resuid;
2961         sbi->s_resgid = old_opts.s_resgid;
2962         sbi->s_commit_interval = old_opts.s_commit_interval;
2963 #ifdef CONFIG_QUOTA
2964         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2965         for (i = 0; i < MAXQUOTAS; i++) {
2966                 if (sbi->s_qf_names[i] &&
2967                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2968                         kfree(sbi->s_qf_names[i]);
2969                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2970         }
2971 #endif
2972         return err;
2973 }
2974
2975 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2976 {
2977         struct super_block *sb = dentry->d_sb;
2978         struct ext4_sb_info *sbi = EXT4_SB(sb);
2979         struct ext4_super_block *es = sbi->s_es;
2980         u64 fsid;
2981
2982         if (test_opt(sb, MINIX_DF)) {
2983                 sbi->s_overhead_last = 0;
2984         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2985                 ext4_group_t ngroups = sbi->s_groups_count, i;
2986                 ext4_fsblk_t overhead = 0;
2987                 smp_rmb();
2988
2989                 /*
2990                  * Compute the overhead (FS structures).  This is constant
2991                  * for a given filesystem unless the number of block groups
2992                  * changes so we cache the previous value until it does.
2993                  */
2994
2995                 /*
2996                  * All of the blocks before first_data_block are
2997                  * overhead
2998                  */
2999                 overhead = le32_to_cpu(es->s_first_data_block);
3000
3001                 /*
3002                  * Add the overhead attributed to the superblock and
3003                  * block group descriptors.  If the sparse superblocks
3004                  * feature is turned on, then not all groups have this.
3005                  */
3006                 for (i = 0; i < ngroups; i++) {
3007                         overhead += ext4_bg_has_super(sb, i) +
3008                                 ext4_bg_num_gdb(sb, i);
3009                         cond_resched();
3010                 }
3011
3012                 /*
3013                  * Every block group has an inode bitmap, a block
3014                  * bitmap, and an inode table.
3015                  */
3016                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3017                 sbi->s_overhead_last = overhead;
3018                 smp_wmb();
3019                 sbi->s_blocks_last = ext4_blocks_count(es);
3020         }
3021
3022         buf->f_type = EXT4_SUPER_MAGIC;
3023         buf->f_bsize = sb->s_blocksize;
3024         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3025         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
3026         ext4_free_blocks_count_set(es, buf->f_bfree);
3027         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3028         if (buf->f_bfree < ext4_r_blocks_count(es))
3029                 buf->f_bavail = 0;
3030         buf->f_files = le32_to_cpu(es->s_inodes_count);
3031         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3032         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3033         buf->f_namelen = EXT4_NAME_LEN;
3034         fsid = le64_to_cpup((void *)es->s_uuid) ^
3035                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3036         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3037         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3038         return 0;
3039 }
3040
3041 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3042  * is locked for write. Otherwise the are possible deadlocks:
3043  * Process 1                         Process 2
3044  * ext4_create()                     quota_sync()
3045  *   jbd2_journal_start()                   write_dquot()
3046  *   DQUOT_INIT()                        down(dqio_mutex)
3047  *     down(dqio_mutex)                    jbd2_journal_start()
3048  *
3049  */
3050
3051 #ifdef CONFIG_QUOTA
3052
3053 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3054 {
3055         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3056 }
3057
3058 static int ext4_dquot_initialize(struct inode *inode, int type)
3059 {
3060         handle_t *handle;
3061         int ret, err;
3062
3063         /* We may create quota structure so we need to reserve enough blocks */
3064         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3065         if (IS_ERR(handle))
3066                 return PTR_ERR(handle);
3067         ret = dquot_initialize(inode, type);
3068         err = ext4_journal_stop(handle);
3069         if (!ret)
3070                 ret = err;
3071         return ret;
3072 }
3073
3074 static int ext4_dquot_drop(struct inode *inode)
3075 {
3076         handle_t *handle;
3077         int ret, err;
3078
3079         /* We may delete quota structure so we need to reserve enough blocks */
3080         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3081         if (IS_ERR(handle)) {
3082                 /*
3083                  * We call dquot_drop() anyway to at least release references
3084                  * to quota structures so that umount does not hang.
3085                  */
3086                 dquot_drop(inode);
3087                 return PTR_ERR(handle);
3088         }
3089         ret = dquot_drop(inode);
3090         err = ext4_journal_stop(handle);
3091         if (!ret)
3092                 ret = err;
3093         return ret;
3094 }
3095
3096 static int ext4_write_dquot(struct dquot *dquot)
3097 {
3098         int ret, err;
3099         handle_t *handle;
3100         struct inode *inode;
3101
3102         inode = dquot_to_inode(dquot);
3103         handle = ext4_journal_start(inode,
3104                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3105         if (IS_ERR(handle))
3106                 return PTR_ERR(handle);
3107         ret = dquot_commit(dquot);
3108         err = ext4_journal_stop(handle);
3109         if (!ret)
3110                 ret = err;
3111         return ret;
3112 }
3113
3114 static int ext4_acquire_dquot(struct dquot *dquot)
3115 {
3116         int ret, err;
3117         handle_t *handle;
3118
3119         handle = ext4_journal_start(dquot_to_inode(dquot),
3120                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3121         if (IS_ERR(handle))
3122                 return PTR_ERR(handle);
3123         ret = dquot_acquire(dquot);
3124         err = ext4_journal_stop(handle);
3125         if (!ret)
3126                 ret = err;
3127         return ret;
3128 }
3129
3130 static int ext4_release_dquot(struct dquot *dquot)
3131 {
3132         int ret, err;
3133         handle_t *handle;
3134
3135         handle = ext4_journal_start(dquot_to_inode(dquot),
3136                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3137         if (IS_ERR(handle)) {
3138                 /* Release dquot anyway to avoid endless cycle in dqput() */
3139                 dquot_release(dquot);
3140                 return PTR_ERR(handle);
3141         }
3142         ret = dquot_release(dquot);
3143         err = ext4_journal_stop(handle);
3144         if (!ret)
3145                 ret = err;
3146         return ret;
3147 }
3148
3149 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3150 {
3151         /* Are we journaling quotas? */
3152         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3153             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3154                 dquot_mark_dquot_dirty(dquot);
3155                 return ext4_write_dquot(dquot);
3156         } else {
3157                 return dquot_mark_dquot_dirty(dquot);
3158         }
3159 }
3160
3161 static int ext4_write_info(struct super_block *sb, int type)
3162 {
3163         int ret, err;
3164         handle_t *handle;
3165
3166         /* Data block + inode block */
3167         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3168         if (IS_ERR(handle))
3169                 return PTR_ERR(handle);
3170         ret = dquot_commit_info(sb, type);
3171         err = ext4_journal_stop(handle);
3172         if (!ret)
3173                 ret = err;
3174         return ret;
3175 }
3176
3177 /*
3178  * Turn on quotas during mount time - we need to find
3179  * the quota file and such...
3180  */
3181 static int ext4_quota_on_mount(struct super_block *sb, int type)
3182 {
3183         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3184                         EXT4_SB(sb)->s_jquota_fmt, type);
3185 }
3186
3187 /*
3188  * Standard function to be called on quota_on
3189  */
3190 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3191                          char *path, int remount)
3192 {
3193         int err;
3194         struct nameidata nd;
3195
3196         if (!test_opt(sb, QUOTA))
3197                 return -EINVAL;
3198         /* When remounting, no checks are needed and in fact, path is NULL */
3199         if (remount)
3200                 return vfs_quota_on(sb, type, format_id, path, remount);
3201
3202         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3203         if (err)
3204                 return err;
3205
3206         /* Quotafile not on the same filesystem? */
3207         if (nd.path.mnt->mnt_sb != sb) {
3208                 path_put(&nd.path);
3209                 return -EXDEV;
3210         }
3211         /* Journaling quota? */
3212         if (EXT4_SB(sb)->s_qf_names[type]) {
3213                 /* Quotafile not of fs root? */
3214                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3215                         printk(KERN_WARNING
3216                                 "EXT4-fs: Quota file not on filesystem root. "
3217                                 "Journaled quota will not work.\n");
3218         }
3219
3220         /*
3221          * When we journal data on quota file, we have to flush journal to see
3222          * all updates to the file when we bypass pagecache...
3223          */
3224         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3225                 /*
3226                  * We don't need to lock updates but journal_flush() could
3227                  * otherwise be livelocked...
3228                  */
3229                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3230                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3231                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3232         }
3233
3234         path_put(&nd.path);
3235         return vfs_quota_on(sb, type, format_id, path, remount);
3236 }
3237
3238 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3239  * acquiring the locks... As quota files are never truncated and quota code
3240  * itself serializes the operations (and noone else should touch the files)
3241  * we don't have to be afraid of races */
3242 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3243                                size_t len, loff_t off)
3244 {
3245         struct inode *inode = sb_dqopt(sb)->files[type];
3246         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3247         int err = 0;
3248         int offset = off & (sb->s_blocksize - 1);
3249         int tocopy;
3250         size_t toread;
3251         struct buffer_head *bh;
3252         loff_t i_size = i_size_read(inode);
3253
3254         if (off > i_size)
3255                 return 0;
3256         if (off+len > i_size)
3257                 len = i_size-off;
3258         toread = len;
3259         while (toread > 0) {
3260                 tocopy = sb->s_blocksize - offset < toread ?
3261                                 sb->s_blocksize - offset : toread;
3262                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3263                 if (err)
3264                         return err;
3265                 if (!bh)        /* A hole? */
3266                         memset(data, 0, tocopy);
3267                 else
3268                         memcpy(data, bh->b_data+offset, tocopy);
3269                 brelse(bh);
3270                 offset = 0;
3271                 toread -= tocopy;
3272                 data += tocopy;
3273                 blk++;
3274         }
3275         return len;
3276 }
3277
3278 /* Write to quotafile (we know the transaction is already started and has
3279  * enough credits) */
3280 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3281                                 const char *data, size_t len, loff_t off)
3282 {
3283         struct inode *inode = sb_dqopt(sb)->files[type];
3284         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3285         int err = 0;
3286         int offset = off & (sb->s_blocksize - 1);
3287         int tocopy;
3288         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3289         size_t towrite = len;
3290         struct buffer_head *bh;
3291         handle_t *handle = journal_current_handle();
3292
3293         if (!handle) {
3294                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3295                         " cancelled because transaction is not started.\n",
3296                         (unsigned long long)off, (unsigned long long)len);
3297                 return -EIO;
3298         }
3299         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3300         while (towrite > 0) {
3301                 tocopy = sb->s_blocksize - offset < towrite ?
3302                                 sb->s_blocksize - offset : towrite;
3303                 bh = ext4_bread(handle, inode, blk, 1, &err);
3304                 if (!bh)
3305                         goto out;
3306                 if (journal_quota) {
3307                         err = ext4_journal_get_write_access(handle, bh);
3308                         if (err) {
3309                                 brelse(bh);
3310                                 goto out;
3311                         }
3312                 }
3313                 lock_buffer(bh);
3314                 memcpy(bh->b_data+offset, data, tocopy);
3315                 flush_dcache_page(bh->b_page);
3316                 unlock_buffer(bh);
3317                 if (journal_quota)
3318                         err = ext4_journal_dirty_metadata(handle, bh);
3319                 else {
3320                         /* Always do at least ordered writes for quotas */
3321                         err = ext4_journal_dirty_data(handle, bh);
3322                         mark_buffer_dirty(bh);
3323                 }
3324                 brelse(bh);
3325                 if (err)
3326                         goto out;
3327                 offset = 0;
3328                 towrite -= tocopy;
3329                 data += tocopy;
3330                 blk++;
3331         }
3332 out:
3333         if (len == towrite)
3334                 return err;
3335         if (inode->i_size < off+len-towrite) {
3336                 i_size_write(inode, off+len-towrite);
3337                 EXT4_I(inode)->i_disksize = inode->i_size;
3338         }
3339         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3340         ext4_mark_inode_dirty(handle, inode);
3341         mutex_unlock(&inode->i_mutex);
3342         return len - towrite;
3343 }
3344
3345 #endif
3346
3347 static int ext4_get_sb(struct file_system_type *fs_type,
3348         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3349 {
3350         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3351 }
3352
3353 static struct file_system_type ext4dev_fs_type = {
3354         .owner          = THIS_MODULE,
3355         .name           = "ext4dev",
3356         .get_sb         = ext4_get_sb,
3357         .kill_sb        = kill_block_super,
3358         .fs_flags       = FS_REQUIRES_DEV,
3359 };
3360
3361 static int __init init_ext4_fs(void)
3362 {
3363         int err;
3364
3365         err = init_ext4_mballoc();
3366         if (err)
3367                 return err;
3368
3369         err = init_ext4_xattr();
3370         if (err)
3371                 goto out2;
3372         err = init_inodecache();
3373         if (err)
3374                 goto out1;
3375         err = register_filesystem(&ext4dev_fs_type);
3376         if (err)
3377                 goto out;
3378         return 0;
3379 out:
3380         destroy_inodecache();
3381 out1:
3382         exit_ext4_xattr();
3383 out2:
3384         exit_ext4_mballoc();
3385         return err;
3386 }
3387
3388 static void __exit exit_ext4_fs(void)
3389 {
3390         unregister_filesystem(&ext4dev_fs_type);
3391         destroy_inodecache();
3392         exit_ext4_xattr();
3393         exit_ext4_mballoc();
3394 }
3395
3396 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3397 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3398 MODULE_LICENSE("GPL");
3399 module_init(init_ext4_fs)
3400 module_exit(exit_ext4_fs)