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/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  50                                unsigned int);
  51 static void ext4_commit_super (struct super_block * sb,
  52                                struct ext4_super_block * es,
  53                                int sync);
  54 static void ext4_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext4_super_block * es);
  56 static void ext4_clear_journal_err(struct super_block * sb,
  57                                    struct ext4_super_block * es);
  58 static int ext4_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext4_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static void ext4_unlockfs(struct super_block *sb);
  64 static void ext4_write_super (struct super_block * sb);
  65 static void ext4_write_super_lockfs(struct super_block *sb);
  66
  67
  68 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  69                                struct ext4_group_desc *bg)
  70 {
  71         return le32_to_cpu(bg->bg_block_bitmap) |
  72                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  73                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  74 }
  75
  76 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  77                                struct ext4_group_desc *bg)
  78 {
  79         return le32_to_cpu(bg->bg_inode_bitmap) |
  80                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  81                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  82 }
  83
  84 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  85                               struct ext4_group_desc *bg)
  86 {
  87         return le32_to_cpu(bg->bg_inode_table) |
  88                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  89                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  90 }
  91
  92 void ext4_block_bitmap_set(struct super_block *sb,
  93                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  94 {
  95         bg->bg_block_bitmap = cpu_to_le32((u32)blk);
  96         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  97                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  98 }
  99
 100 void ext4_inode_bitmap_set(struct super_block *sb,
 101                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 102 {
 103         bg->bg_inode_bitmap  = cpu_to_le32((u32)blk);
 104         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 105                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 106 }
 107
 108 void ext4_inode_table_set(struct super_block *sb,
 109                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 110 {
 111         bg->bg_inode_table = cpu_to_le32((u32)blk);
 112         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 113                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 114 }
 115
 116 /*
 117  * Wrappers for jbd2_journal_start/end.
 118  *
 119  * The only special thing we need to do here is to make sure that all
 120  * journal_end calls result in the superblock being marked dirty, so
 121  * that sync() will call the filesystem's write_super callback if
 122  * appropriate.
 123  */
 124 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 125 {
 126         journal_t *journal;
 127
 128         if (sb->s_flags & MS_RDONLY)
 129                 return ERR_PTR(-EROFS);
 130
 131         /* Special case here: if the journal has aborted behind our
 132          * backs (eg. EIO in the commit thread), then we still need to
 133          * take the FS itself readonly cleanly. */
 134         journal = EXT4_SB(sb)->s_journal;
 135         if (is_journal_aborted(journal)) {
 136                 ext4_abort(sb, __FUNCTION__,
 137                            "Detected aborted journal");
 138                 return ERR_PTR(-EROFS);
 139         }
 140
 141         return jbd2_journal_start(journal, nblocks);
 142 }
 143
 144 /*
 145  * The only special thing we need to do here is to make sure that all
 146  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 147  * that sync() will call the filesystem's write_super callback if
 148  * appropriate.
 149  */
 150 int __ext4_journal_stop(const char *where, handle_t *handle)
 151 {
 152         struct super_block *sb;
 153         int err;
 154         int rc;
 155
 156         sb = handle->h_transaction->t_journal->j_private;
 157         err = handle->h_err;
 158         rc = jbd2_journal_stop(handle);
 159
 160         if (!err)
 161                 err = rc;
 162         if (err)
 163                 __ext4_std_error(sb, where, err);
 164         return err;
 165 }
 166
 167 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 168                 struct buffer_head *bh, handle_t *handle, int err)
 169 {
 170         char nbuf[16];
 171         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 172
 173         if (bh)
 174                 BUFFER_TRACE(bh, "abort");
 175
 176         if (!handle->h_err)
 177                 handle->h_err = err;
 178
 179         if (is_handle_aborted(handle))
 180                 return;
 181
 182         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 183                caller, errstr, err_fn);
 184
 185         jbd2_journal_abort_handle(handle);
 186 }
 187
 188 /* Deal with the reporting of failure conditions on a filesystem such as
 189  * inconsistencies detected or read IO failures.
 190  *
 191  * On ext2, we can store the error state of the filesystem in the
 192  * superblock.  That is not possible on ext4, because we may have other
 193  * write ordering constraints on the superblock which prevent us from
 194  * writing it out straight away; and given that the journal is about to
 195  * be aborted, we can't rely on the current, or future, transactions to
 196  * write out the superblock safely.
 197  *
 198  * We'll just use the jbd2_journal_abort() error code to record an error in
 199  * the journal instead.  On recovery, the journal will compain about
 200  * that error until we've noted it down and cleared it.
 201  */
 202
 203 static void ext4_handle_error(struct super_block *sb)
 204 {
 205         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 206
 207         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 208         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 209
 210         if (sb->s_flags & MS_RDONLY)
 211                 return;
 212
 213         if (!test_opt (sb, ERRORS_CONT)) {
 214                 journal_t *journal = EXT4_SB(sb)->s_journal;
 215
 216                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 217                 if (journal)
 218                         jbd2_journal_abort(journal, -EIO);
 219         }
 220         if (test_opt (sb, ERRORS_RO)) {
 221                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 222                 sb->s_flags |= MS_RDONLY;
 223         }
 224         ext4_commit_super(sb, es, 1);
 225         if (test_opt(sb, ERRORS_PANIC))
 226                 panic("EXT4-fs (device %s): panic forced after error\n",
 227                         sb->s_id);
 228 }
 229
 230 void ext4_error (struct super_block * sb, const char * function,
 231                  const char * fmt, ...)
 232 {
 233         va_list args;
 234
 235         va_start(args, fmt);
 236         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 237         vprintk(fmt, args);
 238         printk("\n");
 239         va_end(args);
 240
 241         ext4_handle_error(sb);
 242 }
 243
 244 static const char *ext4_decode_error(struct super_block * sb, int errno,
 245                                      char nbuf[16])
 246 {
 247         char *errstr = NULL;
 248
 249         switch (errno) {
 250         case -EIO:
 251                 errstr = "IO failure";
 252                 break;
 253         case -ENOMEM:
 254                 errstr = "Out of memory";
 255                 break;
 256         case -EROFS:
 257                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 258                         errstr = "Journal has aborted";
 259                 else
 260                         errstr = "Readonly filesystem";
 261                 break;
 262         default:
 263                 /* If the caller passed in an extra buffer for unknown
 264                  * errors, textualise them now.  Else we just return
 265                  * NULL. */
 266                 if (nbuf) {
 267                         /* Check for truncated error codes... */
 268                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 269                                 errstr = nbuf;
 270                 }
 271                 break;
 272         }
 273
 274         return errstr;
 275 }
 276
 277 /* __ext4_std_error decodes expected errors from journaling functions
 278  * automatically and invokes the appropriate error response.  */
 279
 280 void __ext4_std_error (struct super_block * sb, const char * function,
 281                        int errno)
 282 {
 283         char nbuf[16];
 284         const char *errstr;
 285
 286         /* Special case: if the error is EROFS, and we're not already
 287          * inside a transaction, then there's really no point in logging
 288          * an error. */
 289         if (errno == -EROFS && journal_current_handle() == NULL &&
 290             (sb->s_flags & MS_RDONLY))
 291                 return;
 292
 293         errstr = ext4_decode_error(sb, errno, nbuf);
 294         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 295                 sb->s_id, function, errstr);
 296
 297         ext4_handle_error(sb);
 298 }
 299
 300 /*
 301  * ext4_abort is a much stronger failure handler than ext4_error.  The
 302  * abort function may be used to deal with unrecoverable failures such
 303  * as journal IO errors or ENOMEM at a critical moment in log management.
 304  *
 305  * We unconditionally force the filesystem into an ABORT|READONLY state,
 306  * unless the error response on the fs has been set to panic in which
 307  * case we take the easy way out and panic immediately.
 308  */
 309
 310 void ext4_abort (struct super_block * sb, const char * function,
 311                  const char * fmt, ...)
 312 {
 313         va_list args;
 314
 315         printk (KERN_CRIT "ext4_abort called.\n");
 316
 317         va_start(args, fmt);
 318         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 319         vprintk(fmt, args);
 320         printk("\n");
 321         va_end(args);
 322
 323         if (test_opt(sb, ERRORS_PANIC))
 324                 panic("EXT4-fs panic from previous error\n");
 325
 326         if (sb->s_flags & MS_RDONLY)
 327                 return;
 328
 329         printk(KERN_CRIT "Remounting filesystem read-only\n");
 330         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 331         sb->s_flags |= MS_RDONLY;
 332         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 333         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 334 }
 335
 336 void ext4_warning (struct super_block * sb, const char * function,
 337                    const char * fmt, ...)
 338 {
 339         va_list args;
 340
 341         va_start(args, fmt);
 342         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 343                sb->s_id, function);
 344         vprintk(fmt, args);
 345         printk("\n");
 346         va_end(args);
 347 }
 348
 349 void ext4_update_dynamic_rev(struct super_block *sb)
 350 {
 351         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 352
 353         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 354                 return;
 355
 356         ext4_warning(sb, __FUNCTION__,
 357                      "updating to rev %d because of new feature flag, "
 358                      "running e2fsck is recommended",
 359                      EXT4_DYNAMIC_REV);
 360
 361         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 362         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 363         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 364         /* leave es->s_feature_*compat flags alone */
 365         /* es->s_uuid will be set by e2fsck if empty */
 366
 367         /*
 368          * The rest of the superblock fields should be zero, and if not it
 369          * means they are likely already in use, so leave them alone.  We
 370          * can leave it up to e2fsck to clean up any inconsistencies there.
 371          */
 372 }
 373
 374 /*
 375  * Open the external journal device
 376  */
 377 static struct block_device *ext4_blkdev_get(dev_t dev)
 378 {
 379         struct block_device *bdev;
 380         char b[BDEVNAME_SIZE];
 381
 382         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 383         if (IS_ERR(bdev))
 384                 goto fail;
 385         return bdev;
 386
 387 fail:
 388         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 389                         __bdevname(dev, b), PTR_ERR(bdev));
 390         return NULL;
 391 }
 392
 393 /*
 394  * Release the journal device
 395  */
 396 static int ext4_blkdev_put(struct block_device *bdev)
 397 {
 398         bd_release(bdev);
 399         return blkdev_put(bdev);
 400 }
 401
 402 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 403 {
 404         struct block_device *bdev;
 405         int ret = -ENODEV;
 406
 407         bdev = sbi->journal_bdev;
 408         if (bdev) {
 409                 ret = ext4_blkdev_put(bdev);
 410                 sbi->journal_bdev = NULL;
 411         }
 412         return ret;
 413 }
 414
 415 static inline struct inode *orphan_list_entry(struct list_head *l)
 416 {
 417         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 418 }
 419
 420 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 421 {
 422         struct list_head *l;
 423
 424         printk(KERN_ERR "sb orphan head is %d\n",
 425                le32_to_cpu(sbi->s_es->s_last_orphan));
 426
 427         printk(KERN_ERR "sb_info orphan list:\n");
 428         list_for_each(l, &sbi->s_orphan) {
 429                 struct inode *inode = orphan_list_entry(l);
 430                 printk(KERN_ERR "  "
 431                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 432                        inode->i_sb->s_id, inode->i_ino, inode,
 433                        inode->i_mode, inode->i_nlink,
 434                        NEXT_ORPHAN(inode));
 435         }
 436 }
 437
 438 static void ext4_put_super (struct super_block * sb)
 439 {
 440         struct ext4_sb_info *sbi = EXT4_SB(sb);
 441         struct ext4_super_block *es = sbi->s_es;
 442         int i;
 443
 444         ext4_ext_release(sb);
 445         ext4_xattr_put_super(sb);
 446         jbd2_journal_destroy(sbi->s_journal);
 447         if (!(sb->s_flags & MS_RDONLY)) {
 448                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 449                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 450                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 451                 mark_buffer_dirty(sbi->s_sbh);
 452                 ext4_commit_super(sb, es, 1);
 453         }
 454
 455         for (i = 0; i < sbi->s_gdb_count; i++)
 456                 brelse(sbi->s_group_desc[i]);
 457         kfree(sbi->s_group_desc);
 458         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 459         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 460         percpu_counter_destroy(&sbi->s_dirs_counter);
 461         brelse(sbi->s_sbh);
 462 #ifdef CONFIG_QUOTA
 463         for (i = 0; i < MAXQUOTAS; i++)
 464                 kfree(sbi->s_qf_names[i]);
 465 #endif
 466
 467         /* Debugging code just in case the in-memory inode orphan list
 468          * isn't empty.  The on-disk one can be non-empty if we've
 469          * detected an error and taken the fs readonly, but the
 470          * in-memory list had better be clean by this point. */
 471         if (!list_empty(&sbi->s_orphan))
 472                 dump_orphan_list(sb, sbi);
 473         J_ASSERT(list_empty(&sbi->s_orphan));
 474
 475         invalidate_bdev(sb->s_bdev);
 476         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 477                 /*
 478                  * Invalidate the journal device's buffers.  We don't want them
 479                  * floating about in memory - the physical journal device may
 480                  * hotswapped, and it breaks the `ro-after' testing code.
 481                  */
 482                 sync_blockdev(sbi->journal_bdev);
 483                 invalidate_bdev(sbi->journal_bdev);
 484                 ext4_blkdev_remove(sbi);
 485         }
 486         sb->s_fs_info = NULL;
 487         kfree(sbi);
 488         return;
 489 }
 490
 491 static struct kmem_cache *ext4_inode_cachep;
 492
 493 /*
 494  * Called inside transaction, so use GFP_NOFS
 495  */
 496 static struct inode *ext4_alloc_inode(struct super_block *sb)
 497 {
 498         struct ext4_inode_info *ei;
 499
 500         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 501         if (!ei)
 502                 return NULL;
 503 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 504         ei->i_acl = EXT4_ACL_NOT_CACHED;
 505         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 506 #endif
 507         ei->i_block_alloc_info = NULL;
 508         ei->vfs_inode.i_version = 1;
 509         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 510         return &ei->vfs_inode;
 511 }
 512
 513 static void ext4_destroy_inode(struct inode *inode)
 514 {
 515         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 516                 printk("EXT4 Inode %p: orphan list check failed!\n",
 517                         EXT4_I(inode));
 518                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 519                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 520                                 true);
 521                 dump_stack();
 522         }
 523         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 524 }
 525
 526 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
 527 {
 528         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 529
 530         INIT_LIST_HEAD(&ei->i_orphan);
 531 #ifdef CONFIG_EXT4DEV_FS_XATTR
 532         init_rwsem(&ei->xattr_sem);
 533 #endif
 534         mutex_init(&ei->truncate_mutex);
 535         inode_init_once(&ei->vfs_inode);
 536 }
 537
 538 static int init_inodecache(void)
 539 {
 540         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 541                                              sizeof(struct ext4_inode_info),
 542                                              0, (SLAB_RECLAIM_ACCOUNT|
 543                                                 SLAB_MEM_SPREAD),
 544                                              init_once);
 545         if (ext4_inode_cachep == NULL)
 546                 return -ENOMEM;
 547         return 0;
 548 }
 549
 550 static void destroy_inodecache(void)
 551 {
 552         kmem_cache_destroy(ext4_inode_cachep);
 553 }
 554
 555 static void ext4_clear_inode(struct inode *inode)
 556 {
 557         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 558 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 559         if (EXT4_I(inode)->i_acl &&
 560                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 561                 posix_acl_release(EXT4_I(inode)->i_acl);
 562                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 563         }
 564         if (EXT4_I(inode)->i_default_acl &&
 565                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 566                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 567                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 568         }
 569 #endif
 570         ext4_discard_reservation(inode);
 571         EXT4_I(inode)->i_block_alloc_info = NULL;
 572         if (unlikely(rsv))
 573                 kfree(rsv);
 574 }
 575
 576 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 577 {
 578 #if defined(CONFIG_QUOTA)
 579         struct ext4_sb_info *sbi = EXT4_SB(sb);
 580
 581         if (sbi->s_jquota_fmt)
 582                 seq_printf(seq, ",jqfmt=%s",
 583                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 584
 585         if (sbi->s_qf_names[USRQUOTA])
 586                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 587
 588         if (sbi->s_qf_names[GRPQUOTA])
 589                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 590
 591         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 592                 seq_puts(seq, ",usrquota");
 593
 594         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 595                 seq_puts(seq, ",grpquota");
 596 #endif
 597 }
 598
 599 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 600 {
 601         struct super_block *sb = vfs->mnt_sb;
 602
 603         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 604                 seq_puts(seq, ",data=journal");
 605         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 606                 seq_puts(seq, ",data=ordered");
 607         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 608                 seq_puts(seq, ",data=writeback");
 609
 610         ext4_show_quota_options(seq, sb);
 611
 612         return 0;
 613 }
 614
 615
 616 static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
 617 {
 618         __u32 *objp = vobjp;
 619         unsigned long ino = objp[0];
 620         __u32 generation = objp[1];
 621         struct inode *inode;
 622         struct dentry *result;
 623
 624         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 625                 return ERR_PTR(-ESTALE);
 626         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 627                 return ERR_PTR(-ESTALE);
 628
 629         /* iget isn't really right if the inode is currently unallocated!!
 630          *
 631          * ext4_read_inode will return a bad_inode if the inode had been
 632          * deleted, so we should be safe.
 633          *
 634          * Currently we don't know the generation for parent directory, so
 635          * a generation of 0 means "accept any"
 636          */
 637         inode = iget(sb, ino);
 638         if (inode == NULL)
 639                 return ERR_PTR(-ENOMEM);
 640         if (is_bad_inode(inode) ||
 641             (generation && inode->i_generation != generation)) {
 642                 iput(inode);
 643                 return ERR_PTR(-ESTALE);
 644         }
 645         /* now to find a dentry.
 646          * If possible, get a well-connected one
 647          */
 648         result = d_alloc_anon(inode);
 649         if (!result) {
 650                 iput(inode);
 651                 return ERR_PTR(-ENOMEM);
 652         }
 653         return result;
 654 }
 655
 656 #ifdef CONFIG_QUOTA
 657 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 658 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 659
 660 static int ext4_dquot_initialize(struct inode *inode, int type);
 661 static int ext4_dquot_drop(struct inode *inode);
 662 static int ext4_write_dquot(struct dquot *dquot);
 663 static int ext4_acquire_dquot(struct dquot *dquot);
 664 static int ext4_release_dquot(struct dquot *dquot);
 665 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 666 static int ext4_write_info(struct super_block *sb, int type);
 667 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 668 static int ext4_quota_on_mount(struct super_block *sb, int type);
 669 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 670                                size_t len, loff_t off);
 671 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 672                                 const char *data, size_t len, loff_t off);
 673
 674 static struct dquot_operations ext4_quota_operations = {
 675         .initialize     = ext4_dquot_initialize,
 676         .drop           = ext4_dquot_drop,
 677         .alloc_space    = dquot_alloc_space,
 678         .alloc_inode    = dquot_alloc_inode,
 679         .free_space     = dquot_free_space,
 680         .free_inode     = dquot_free_inode,
 681         .transfer       = dquot_transfer,
 682         .write_dquot    = ext4_write_dquot,
 683         .acquire_dquot  = ext4_acquire_dquot,
 684         .release_dquot  = ext4_release_dquot,
 685         .mark_dirty     = ext4_mark_dquot_dirty,
 686         .write_info     = ext4_write_info
 687 };
 688
 689 static struct quotactl_ops ext4_qctl_operations = {
 690         .quota_on       = ext4_quota_on,
 691         .quota_off      = vfs_quota_off,
 692         .quota_sync     = vfs_quota_sync,
 693         .get_info       = vfs_get_dqinfo,
 694         .set_info       = vfs_set_dqinfo,
 695         .get_dqblk      = vfs_get_dqblk,
 696         .set_dqblk      = vfs_set_dqblk
 697 };
 698 #endif
 699
 700 static const struct super_operations ext4_sops = {
 701         .alloc_inode    = ext4_alloc_inode,
 702         .destroy_inode  = ext4_destroy_inode,
 703         .read_inode     = ext4_read_inode,
 704         .write_inode    = ext4_write_inode,
 705         .dirty_inode    = ext4_dirty_inode,
 706         .delete_inode   = ext4_delete_inode,
 707         .put_super      = ext4_put_super,
 708         .write_super    = ext4_write_super,
 709         .sync_fs        = ext4_sync_fs,
 710         .write_super_lockfs = ext4_write_super_lockfs,
 711         .unlockfs       = ext4_unlockfs,
 712         .statfs         = ext4_statfs,
 713         .remount_fs     = ext4_remount,
 714         .clear_inode    = ext4_clear_inode,
 715         .show_options   = ext4_show_options,
 716 #ifdef CONFIG_QUOTA
 717         .quota_read     = ext4_quota_read,
 718         .quota_write    = ext4_quota_write,
 719 #endif
 720 };
 721
 722 static struct export_operations ext4_export_ops = {
 723         .get_parent = ext4_get_parent,
 724         .get_dentry = ext4_get_dentry,
 725 };
 726
 727 enum {
 728         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 729         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 730         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 731         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 732         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 733         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 734         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 735         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 736         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 737         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 738         Opt_grpquota, Opt_extents, Opt_noextents,
 739 };
 740
 741 static match_table_t tokens = {
 742         {Opt_bsd_df, "bsddf"},
 743         {Opt_minix_df, "minixdf"},
 744         {Opt_grpid, "grpid"},
 745         {Opt_grpid, "bsdgroups"},
 746         {Opt_nogrpid, "nogrpid"},
 747         {Opt_nogrpid, "sysvgroups"},
 748         {Opt_resgid, "resgid=%u"},
 749         {Opt_resuid, "resuid=%u"},
 750         {Opt_sb, "sb=%u"},
 751         {Opt_err_cont, "errors=continue"},
 752         {Opt_err_panic, "errors=panic"},
 753         {Opt_err_ro, "errors=remount-ro"},
 754         {Opt_nouid32, "nouid32"},
 755         {Opt_nocheck, "nocheck"},
 756         {Opt_nocheck, "check=none"},
 757         {Opt_debug, "debug"},
 758         {Opt_oldalloc, "oldalloc"},
 759         {Opt_orlov, "orlov"},
 760         {Opt_user_xattr, "user_xattr"},
 761         {Opt_nouser_xattr, "nouser_xattr"},
 762         {Opt_acl, "acl"},
 763         {Opt_noacl, "noacl"},
 764         {Opt_reservation, "reservation"},
 765         {Opt_noreservation, "noreservation"},
 766         {Opt_noload, "noload"},
 767         {Opt_nobh, "nobh"},
 768         {Opt_bh, "bh"},
 769         {Opt_commit, "commit=%u"},
 770         {Opt_journal_update, "journal=update"},
 771         {Opt_journal_inum, "journal=%u"},
 772         {Opt_journal_dev, "journal_dev=%u"},
 773         {Opt_abort, "abort"},
 774         {Opt_data_journal, "data=journal"},
 775         {Opt_data_ordered, "data=ordered"},
 776         {Opt_data_writeback, "data=writeback"},
 777         {Opt_offusrjquota, "usrjquota="},
 778         {Opt_usrjquota, "usrjquota=%s"},
 779         {Opt_offgrpjquota, "grpjquota="},
 780         {Opt_grpjquota, "grpjquota=%s"},
 781         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 782         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 783         {Opt_grpquota, "grpquota"},
 784         {Opt_noquota, "noquota"},
 785         {Opt_quota, "quota"},
 786         {Opt_usrquota, "usrquota"},
 787         {Opt_barrier, "barrier=%u"},
 788         {Opt_extents, "extents"},
 789         {Opt_noextents, "noextents"},
 790         {Opt_err, NULL},
 791         {Opt_resize, "resize"},
 792 };
 793
 794 static ext4_fsblk_t get_sb_block(void **data)
 795 {
 796         ext4_fsblk_t    sb_block;
 797         char            *options = (char *) *data;
 798
 799         if (!options || strncmp(options, "sb=", 3) != 0)
 800                 return 1;       /* Default location */
 801         options += 3;
 802         /*todo: use simple_strtoll with >32bit ext4 */
 803         sb_block = simple_strtoul(options, &options, 0);
 804         if (*options && *options != ',') {
 805                 printk("EXT4-fs: Invalid sb specification: %s\n",
 806                        (char *) *data);
 807                 return 1;
 808         }
 809         if (*options == ',')
 810                 options++;
 811         *data = (void *) options;
 812         return sb_block;
 813 }
 814
 815 static int parse_options (char *options, struct super_block *sb,
 816                           unsigned int *inum, unsigned long *journal_devnum,
 817                           ext4_fsblk_t *n_blocks_count, int is_remount)
 818 {
 819         struct ext4_sb_info *sbi = EXT4_SB(sb);
 820         char * p;
 821         substring_t args[MAX_OPT_ARGS];
 822         int data_opt = 0;
 823         int option;
 824 #ifdef CONFIG_QUOTA
 825         int qtype;
 826         char *qname;
 827 #endif
 828
 829         if (!options)
 830                 return 1;
 831
 832         while ((p = strsep (&options, ",")) != NULL) {
 833                 int token;
 834                 if (!*p)
 835                         continue;
 836
 837                 token = match_token(p, tokens, args);
 838                 switch (token) {
 839                 case Opt_bsd_df:
 840                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 841                         break;
 842                 case Opt_minix_df:
 843                         set_opt (sbi->s_mount_opt, MINIX_DF);
 844                         break;
 845                 case Opt_grpid:
 846                         set_opt (sbi->s_mount_opt, GRPID);
 847                         break;
 848                 case Opt_nogrpid:
 849                         clear_opt (sbi->s_mount_opt, GRPID);
 850                         break;
 851                 case Opt_resuid:
 852                         if (match_int(&args[0], &option))
 853                                 return 0;
 854                         sbi->s_resuid = option;
 855                         break;
 856                 case Opt_resgid:
 857                         if (match_int(&args[0], &option))
 858                                 return 0;
 859                         sbi->s_resgid = option;
 860                         break;
 861                 case Opt_sb:
 862                         /* handled by get_sb_block() instead of here */
 863                         /* *sb_block = match_int(&args[0]); */
 864                         break;
 865                 case Opt_err_panic:
 866                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 867                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 868                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 869                         break;
 870                 case Opt_err_ro:
 871                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 872                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 873                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 874                         break;
 875                 case Opt_err_cont:
 876                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 877                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 878                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 879                         break;
 880                 case Opt_nouid32:
 881                         set_opt (sbi->s_mount_opt, NO_UID32);
 882                         break;
 883                 case Opt_nocheck:
 884                         clear_opt (sbi->s_mount_opt, CHECK);
 885                         break;
 886                 case Opt_debug:
 887                         set_opt (sbi->s_mount_opt, DEBUG);
 888                         break;
 889                 case Opt_oldalloc:
 890                         set_opt (sbi->s_mount_opt, OLDALLOC);
 891                         break;
 892                 case Opt_orlov:
 893                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 894                         break;
 895 #ifdef CONFIG_EXT4DEV_FS_XATTR
 896                 case Opt_user_xattr:
 897                         set_opt (sbi->s_mount_opt, XATTR_USER);
 898                         break;
 899                 case Opt_nouser_xattr:
 900                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 901                         break;
 902 #else
 903                 case Opt_user_xattr:
 904                 case Opt_nouser_xattr:
 905                         printk("EXT4 (no)user_xattr options not supported\n");
 906                         break;
 907 #endif
 908 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 909                 case Opt_acl:
 910                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 911                         break;
 912                 case Opt_noacl:
 913                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 914                         break;
 915 #else
 916                 case Opt_acl:
 917                 case Opt_noacl:
 918                         printk("EXT4 (no)acl options not supported\n");
 919                         break;
 920 #endif
 921                 case Opt_reservation:
 922                         set_opt(sbi->s_mount_opt, RESERVATION);
 923                         break;
 924                 case Opt_noreservation:
 925                         clear_opt(sbi->s_mount_opt, RESERVATION);
 926                         break;
 927                 case Opt_journal_update:
 928                         /* @@@ FIXME */
 929                         /* Eventually we will want to be able to create
 930                            a journal file here.  For now, only allow the
 931                            user to specify an existing inode to be the
 932                            journal file. */
 933                         if (is_remount) {
 934                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 935                                        "journal on remount\n");
 936                                 return 0;
 937                         }
 938                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 939                         break;
 940                 case Opt_journal_inum:
 941                         if (is_remount) {
 942                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 943                                        "journal on remount\n");
 944                                 return 0;
 945                         }
 946                         if (match_int(&args[0], &option))
 947                                 return 0;
 948                         *inum = option;
 949                         break;
 950                 case Opt_journal_dev:
 951                         if (is_remount) {
 952                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 953                                        "journal on remount\n");
 954                                 return 0;
 955                         }
 956                         if (match_int(&args[0], &option))
 957                                 return 0;
 958                         *journal_devnum = option;
 959                         break;
 960                 case Opt_noload:
 961                         set_opt (sbi->s_mount_opt, NOLOAD);
 962                         break;
 963                 case Opt_commit:
 964                         if (match_int(&args[0], &option))
 965                                 return 0;
 966                         if (option < 0)
 967                                 return 0;
 968                         if (option == 0)
 969                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 970                         sbi->s_commit_interval = HZ * option;
 971                         break;
 972                 case Opt_data_journal:
 973                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
 974                         goto datacheck;
 975                 case Opt_data_ordered:
 976                         data_opt = EXT4_MOUNT_ORDERED_DATA;
 977                         goto datacheck;
 978                 case Opt_data_writeback:
 979                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
 980                 datacheck:
 981                         if (is_remount) {
 982                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
 983                                                 != data_opt) {
 984                                         printk(KERN_ERR
 985                                                 "EXT4-fs: cannot change data "
 986                                                 "mode on remount\n");
 987                                         return 0;
 988                                 }
 989                         } else {
 990                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
 991                                 sbi->s_mount_opt |= data_opt;
 992                         }
 993                         break;
 994 #ifdef CONFIG_QUOTA
 995                 case Opt_usrjquota:
 996                         qtype = USRQUOTA;
 997                         goto set_qf_name;
 998                 case Opt_grpjquota:
 999                         qtype = GRPQUOTA;
1000 set_qf_name:
1001                         if (sb_any_quota_enabled(sb)) {
1002                                 printk(KERN_ERR
1003                                         "EXT4-fs: Cannot change journalled "
1004                                         "quota options when quota turned on.\n");
1005                                 return 0;
1006                         }
1007                         qname = match_strdup(&args[0]);
1008                         if (!qname) {
1009                                 printk(KERN_ERR
1010                                         "EXT4-fs: not enough memory for "
1011                                         "storing quotafile name.\n");
1012                                 return 0;
1013                         }
1014                         if (sbi->s_qf_names[qtype] &&
1015                             strcmp(sbi->s_qf_names[qtype], qname)) {
1016                                 printk(KERN_ERR
1017                                         "EXT4-fs: %s quota file already "
1018                                         "specified.\n", QTYPE2NAME(qtype));
1019                                 kfree(qname);
1020                                 return 0;
1021                         }
1022                         sbi->s_qf_names[qtype] = qname;
1023                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1024                                 printk(KERN_ERR
1025                                         "EXT4-fs: quotafile must be on "
1026                                         "filesystem root.\n");
1027                                 kfree(sbi->s_qf_names[qtype]);
1028                                 sbi->s_qf_names[qtype] = NULL;
1029                                 return 0;
1030                         }
1031                         set_opt(sbi->s_mount_opt, QUOTA);
1032                         break;
1033                 case Opt_offusrjquota:
1034                         qtype = USRQUOTA;
1035                         goto clear_qf_name;
1036                 case Opt_offgrpjquota:
1037                         qtype = GRPQUOTA;
1038 clear_qf_name:
1039                         if (sb_any_quota_enabled(sb)) {
1040                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1041                                         "journalled quota options when "
1042                                         "quota turned on.\n");
1043                                 return 0;
1044                         }
1045                         /*
1046                          * The space will be released later when all options
1047                          * are confirmed to be correct
1048                          */
1049                         sbi->s_qf_names[qtype] = NULL;
1050                         break;
1051                 case Opt_jqfmt_vfsold:
1052                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1053                         break;
1054                 case Opt_jqfmt_vfsv0:
1055                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1056                         break;
1057                 case Opt_quota:
1058                 case Opt_usrquota:
1059                         set_opt(sbi->s_mount_opt, QUOTA);
1060                         set_opt(sbi->s_mount_opt, USRQUOTA);
1061                         break;
1062                 case Opt_grpquota:
1063                         set_opt(sbi->s_mount_opt, QUOTA);
1064                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1065                         break;
1066                 case Opt_noquota:
1067                         if (sb_any_quota_enabled(sb)) {
1068                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1069                                         "options when quota turned on.\n");
1070                                 return 0;
1071                         }
1072                         clear_opt(sbi->s_mount_opt, QUOTA);
1073                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1074                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1075                         break;
1076 #else
1077                 case Opt_quota:
1078                 case Opt_usrquota:
1079                 case Opt_grpquota:
1080                 case Opt_usrjquota:
1081                 case Opt_grpjquota:
1082                 case Opt_offusrjquota:
1083                 case Opt_offgrpjquota:
1084                 case Opt_jqfmt_vfsold:
1085                 case Opt_jqfmt_vfsv0:
1086                         printk(KERN_ERR
1087                                 "EXT4-fs: journalled quota options not "
1088                                 "supported.\n");
1089                         break;
1090                 case Opt_noquota:
1091                         break;
1092 #endif
1093                 case Opt_abort:
1094                         set_opt(sbi->s_mount_opt, ABORT);
1095                         break;
1096                 case Opt_barrier:
1097                         if (match_int(&args[0], &option))
1098                                 return 0;
1099                         if (option)
1100                                 set_opt(sbi->s_mount_opt, BARRIER);
1101                         else
1102                                 clear_opt(sbi->s_mount_opt, BARRIER);
1103                         break;
1104                 case Opt_ignore:
1105                         break;
1106                 case Opt_resize:
1107                         if (!is_remount) {
1108                                 printk("EXT4-fs: resize option only available "
1109                                         "for remount\n");
1110                                 return 0;
1111                         }
1112                         if (match_int(&args[0], &option) != 0)
1113                                 return 0;
1114                         *n_blocks_count = option;
1115                         break;
1116                 case Opt_nobh:
1117                         set_opt(sbi->s_mount_opt, NOBH);
1118                         break;
1119                 case Opt_bh:
1120                         clear_opt(sbi->s_mount_opt, NOBH);
1121                         break;
1122                 case Opt_extents:
1123                         set_opt (sbi->s_mount_opt, EXTENTS);
1124                         break;
1125                 case Opt_noextents:
1126                         clear_opt (sbi->s_mount_opt, EXTENTS);
1127                         break;
1128                 default:
1129                         printk (KERN_ERR
1130                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1131                                 "or missing value\n", p);
1132                         return 0;
1133                 }
1134         }
1135 #ifdef CONFIG_QUOTA
1136         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1137                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1138                      sbi->s_qf_names[USRQUOTA])
1139                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1140
1141                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1142                      sbi->s_qf_names[GRPQUOTA])
1143                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1144
1145                 if ((sbi->s_qf_names[USRQUOTA] &&
1146                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1147                     (sbi->s_qf_names[GRPQUOTA] &&
1148                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1149                         printk(KERN_ERR "EXT4-fs: old and new quota "
1150                                         "format mixing.\n");
1151                         return 0;
1152                 }
1153
1154                 if (!sbi->s_jquota_fmt) {
1155                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1156                                         "not specified.\n");
1157                         return 0;
1158                 }
1159         } else {
1160                 if (sbi->s_jquota_fmt) {
1161                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1162                                         "specified with no journalling "
1163                                         "enabled.\n");
1164                         return 0;
1165                 }
1166         }
1167 #endif
1168         return 1;
1169 }
1170
1171 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1172                             int read_only)
1173 {
1174         struct ext4_sb_info *sbi = EXT4_SB(sb);
1175         int res = 0;
1176
1177         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1178                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1179                         "forcing read-only mode\n");
1180                 res = MS_RDONLY;
1181         }
1182         if (read_only)
1183                 return res;
1184         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1185                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1186                         "running e2fsck is recommended\n");
1187         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1188                 printk (KERN_WARNING
1189                         "EXT4-fs warning: mounting fs with errors, "
1190                         "running e2fsck is recommended\n");
1191         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1192                  le16_to_cpu(es->s_mnt_count) >=
1193                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1194                 printk (KERN_WARNING
1195                         "EXT4-fs warning: maximal mount count reached, "
1196                         "running e2fsck is recommended\n");
1197         else if (le32_to_cpu(es->s_checkinterval) &&
1198                 (le32_to_cpu(es->s_lastcheck) +
1199                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1200                 printk (KERN_WARNING
1201                         "EXT4-fs warning: checktime reached, "
1202                         "running e2fsck is recommended\n");
1203 #if 0
1204                 /* @@@ We _will_ want to clear the valid bit if we find
1205                  * inconsistencies, to force a fsck at reboot.  But for
1206                  * a plain journaled filesystem we can keep it set as
1207                  * valid forever! :)
1208                  */
1209         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1210 #endif
1211         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1212                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1213         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1214         es->s_mtime = cpu_to_le32(get_seconds());
1215         ext4_update_dynamic_rev(sb);
1216         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1217
1218         ext4_commit_super(sb, es, 1);
1219         if (test_opt(sb, DEBUG))
1220                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1221                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1222                         sb->s_blocksize,
1223                         sbi->s_groups_count,
1224                         EXT4_BLOCKS_PER_GROUP(sb),
1225                         EXT4_INODES_PER_GROUP(sb),
1226                         sbi->s_mount_opt);
1227
1228         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1229         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1230                 char b[BDEVNAME_SIZE];
1231
1232                 printk("external journal on %s\n",
1233                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1234         } else {
1235                 printk("internal journal\n");
1236         }
1237         return res;
1238 }
1239
1240 /* Called at mount-time, super-block is locked */
1241 static int ext4_check_descriptors (struct super_block * sb)
1242 {
1243         struct ext4_sb_info *sbi = EXT4_SB(sb);
1244         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1245         ext4_fsblk_t last_block;
1246         ext4_fsblk_t block_bitmap;
1247         ext4_fsblk_t inode_bitmap;
1248         ext4_fsblk_t inode_table;
1249         struct ext4_group_desc * gdp = NULL;
1250         int desc_block = 0;
1251         int i;
1252
1253         ext4_debug ("Checking group descriptors");
1254
1255         for (i = 0; i < sbi->s_groups_count; i++)
1256         {
1257                 if (i == sbi->s_groups_count - 1)
1258                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1259                 else
1260                         last_block = first_block +
1261                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1262
1263                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1264                         gdp = (struct ext4_group_desc *)
1265                                         sbi->s_group_desc[desc_block++]->b_data;
1266                 block_bitmap = ext4_block_bitmap(sb, gdp);
1267                 if (block_bitmap < first_block || block_bitmap > last_block)
1268                 {
1269                         ext4_error (sb, "ext4_check_descriptors",
1270                                     "Block bitmap for group %d"
1271                                     " not in group (block %llu)!",
1272                                     i, block_bitmap);
1273                         return 0;
1274                 }
1275                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1276                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1277                 {
1278                         ext4_error (sb, "ext4_check_descriptors",
1279                                     "Inode bitmap for group %d"
1280                                     " not in group (block %llu)!",
1281                                     i, inode_bitmap);
1282                         return 0;
1283                 }
1284                 inode_table = ext4_inode_table(sb, gdp);
1285                 if (inode_table < first_block ||
1286                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1287                 {
1288                         ext4_error (sb, "ext4_check_descriptors",
1289                                     "Inode table for group %d"
1290                                     " not in group (block %llu)!",
1291                                     i, inode_table);
1292                         return 0;
1293                 }
1294                 first_block += EXT4_BLOCKS_PER_GROUP(sb);
1295                 gdp = (struct ext4_group_desc *)
1296                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1297         }
1298
1299         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1300         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1301         return 1;
1302 }
1303
1304
1305 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1306  * the superblock) which were deleted from all directories, but held open by
1307  * a process at the time of a crash.  We walk the list and try to delete these
1308  * inodes at recovery time (only with a read-write filesystem).
1309  *
1310  * In order to keep the orphan inode chain consistent during traversal (in
1311  * case of crash during recovery), we link each inode into the superblock
1312  * orphan list_head and handle it the same way as an inode deletion during
1313  * normal operation (which journals the operations for us).
1314  *
1315  * We only do an iget() and an iput() on each inode, which is very safe if we
1316  * accidentally point at an in-use or already deleted inode.  The worst that
1317  * can happen in this case is that we get a "bit already cleared" message from
1318  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1319  * e2fsck was run on this filesystem, and it must have already done the orphan
1320  * inode cleanup for us, so we can safely abort without any further action.
1321  */
1322 static void ext4_orphan_cleanup (struct super_block * sb,
1323                                  struct ext4_super_block * es)
1324 {
1325         unsigned int s_flags = sb->s_flags;
1326         int nr_orphans = 0, nr_truncates = 0;
1327 #ifdef CONFIG_QUOTA
1328         int i;
1329 #endif
1330         if (!es->s_last_orphan) {
1331                 jbd_debug(4, "no orphan inodes to clean up\n");
1332                 return;
1333         }
1334
1335         if (bdev_read_only(sb->s_bdev)) {
1336                 printk(KERN_ERR "EXT4-fs: write access "
1337                         "unavailable, skipping orphan cleanup.\n");
1338                 return;
1339         }
1340
1341         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1342                 if (es->s_last_orphan)
1343                         jbd_debug(1, "Errors on filesystem, "
1344                                   "clearing orphan list.\n");
1345                 es->s_last_orphan = 0;
1346                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1347                 return;
1348         }
1349
1350         if (s_flags & MS_RDONLY) {
1351                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1352                        sb->s_id);
1353                 sb->s_flags &= ~MS_RDONLY;
1354         }
1355 #ifdef CONFIG_QUOTA
1356         /* Needed for iput() to work correctly and not trash data */
1357         sb->s_flags |= MS_ACTIVE;
1358         /* Turn on quotas so that they are updated correctly */
1359         for (i = 0; i < MAXQUOTAS; i++) {
1360                 if (EXT4_SB(sb)->s_qf_names[i]) {
1361                         int ret = ext4_quota_on_mount(sb, i);
1362                         if (ret < 0)
1363                                 printk(KERN_ERR
1364                                         "EXT4-fs: Cannot turn on journalled "
1365                                         "quota: error %d\n", ret);
1366                 }
1367         }
1368 #endif
1369
1370         while (es->s_last_orphan) {
1371                 struct inode *inode;
1372
1373                 if (!(inode =
1374                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1375                         es->s_last_orphan = 0;
1376                         break;
1377                 }
1378
1379                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1380                 DQUOT_INIT(inode);
1381                 if (inode->i_nlink) {
1382                         printk(KERN_DEBUG
1383                                 "%s: truncating inode %lu to %Ld bytes\n",
1384                                 __FUNCTION__, inode->i_ino, inode->i_size);
1385                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1386                                   inode->i_ino, inode->i_size);
1387                         ext4_truncate(inode);
1388                         nr_truncates++;
1389                 } else {
1390                         printk(KERN_DEBUG
1391                                 "%s: deleting unreferenced inode %lu\n",
1392                                 __FUNCTION__, inode->i_ino);
1393                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1394                                   inode->i_ino);
1395                         nr_orphans++;
1396                 }
1397                 iput(inode);  /* The delete magic happens here! */
1398         }
1399
1400 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1401
1402         if (nr_orphans)
1403                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1404                        sb->s_id, PLURAL(nr_orphans));
1405         if (nr_truncates)
1406                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1407                        sb->s_id, PLURAL(nr_truncates));
1408 #ifdef CONFIG_QUOTA
1409         /* Turn quotas off */
1410         for (i = 0; i < MAXQUOTAS; i++) {
1411                 if (sb_dqopt(sb)->files[i])
1412                         vfs_quota_off(sb, i);
1413         }
1414 #endif
1415         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1416 }
1417
1418 #define log2(n) ffz(~(n))
1419
1420 /*
1421  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1422  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1423  * We need to be 1 filesystem block less than the 2^32 sector limit.
1424  */
1425 static loff_t ext4_max_size(int bits)
1426 {
1427         loff_t res = EXT4_NDIR_BLOCKS;
1428         /* This constant is calculated to be the largest file size for a
1429          * dense, 4k-blocksize file such that the total number of
1430          * sectors in the file, including data and all indirect blocks,
1431          * does not exceed 2^32. */
1432         const loff_t upper_limit = 0x1ff7fffd000LL;
1433
1434         res += 1LL << (bits-2);
1435         res += 1LL << (2*(bits-2));
1436         res += 1LL << (3*(bits-2));
1437         res <<= bits;
1438         if (res > upper_limit)
1439                 res = upper_limit;
1440         return res;
1441 }
1442
1443 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1444                                 ext4_fsblk_t logical_sb_block, int nr)
1445 {
1446         struct ext4_sb_info *sbi = EXT4_SB(sb);
1447         unsigned long bg, first_meta_bg;
1448         int has_super = 0;
1449
1450         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1451
1452         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1453             nr < first_meta_bg)
1454                 return logical_sb_block + nr + 1;
1455         bg = sbi->s_desc_per_block * nr;
1456         if (ext4_bg_has_super(sb, bg))
1457                 has_super = 1;
1458         return (has_super + ext4_group_first_block_no(sb, bg));
1459 }
1460
1461
1462 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1463 {
1464         struct buffer_head * bh;
1465         struct ext4_super_block *es = NULL;
1466         struct ext4_sb_info *sbi;
1467         ext4_fsblk_t block;
1468         ext4_fsblk_t sb_block = get_sb_block(&data);
1469         ext4_fsblk_t logical_sb_block;
1470         unsigned long offset = 0;
1471         unsigned int journal_inum = 0;
1472         unsigned long journal_devnum = 0;
1473         unsigned long def_mount_opts;
1474         struct inode *root;
1475         int blocksize;
1476         int hblock;
1477         int db_count;
1478         int i;
1479         int needs_recovery;
1480         __le32 features;
1481         __u64 blocks_count;
1482         int err;
1483
1484         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1485         if (!sbi)
1486                 return -ENOMEM;
1487         sb->s_fs_info = sbi;
1488         sbi->s_mount_opt = 0;
1489         sbi->s_resuid = EXT4_DEF_RESUID;
1490         sbi->s_resgid = EXT4_DEF_RESGID;
1491
1492         unlock_kernel();
1493
1494         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1495         if (!blocksize) {
1496                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1497                 goto out_fail;
1498         }
1499
1500         /*
1501          * The ext4 superblock will not be buffer aligned for other than 1kB
1502          * block sizes.  We need to calculate the offset from buffer start.
1503          */
1504         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1505                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1506                 offset = do_div(logical_sb_block, blocksize);
1507         } else {
1508                 logical_sb_block = sb_block;
1509         }
1510
1511         if (!(bh = sb_bread(sb, logical_sb_block))) {
1512                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1513                 goto out_fail;
1514         }
1515         /*
1516          * Note: s_es must be initialized as soon as possible because
1517          *       some ext4 macro-instructions depend on its value
1518          */
1519         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1520         sbi->s_es = es;
1521         sb->s_magic = le16_to_cpu(es->s_magic);
1522         if (sb->s_magic != EXT4_SUPER_MAGIC)
1523                 goto cantfind_ext4;
1524
1525         /* Set defaults before we parse the mount options */
1526         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1527         if (def_mount_opts & EXT4_DEFM_DEBUG)
1528                 set_opt(sbi->s_mount_opt, DEBUG);
1529         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1530                 set_opt(sbi->s_mount_opt, GRPID);
1531         if (def_mount_opts & EXT4_DEFM_UID16)
1532                 set_opt(sbi->s_mount_opt, NO_UID32);
1533 #ifdef CONFIG_EXT4DEV_FS_XATTR
1534         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1535                 set_opt(sbi->s_mount_opt, XATTR_USER);
1536 #endif
1537 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1538         if (def_mount_opts & EXT4_DEFM_ACL)
1539                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1540 #endif
1541         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1542                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1543         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1544                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1545         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1546                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1547
1548         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1549                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1550         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1551                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1552         else
1553                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1554
1555         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1556         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1557
1558         set_opt(sbi->s_mount_opt, RESERVATION);
1559
1560         /*
1561          * turn on extents feature by default in ext4 filesystem
1562          * User -o noextents to turn it off
1563          */
1564         set_opt(sbi->s_mount_opt, EXTENTS);
1565
1566         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1567                             NULL, 0))
1568                 goto failed_mount;
1569
1570         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1571                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1572
1573         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1574             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1575              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1576              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1577                 printk(KERN_WARNING
1578                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1579                        "running e2fsck is recommended\n");
1580         /*
1581          * Check feature flags regardless of the revision level, since we
1582          * previously didn't change the revision level when setting the flags,
1583          * so there is a chance incompat flags are set on a rev 0 filesystem.
1584          */
1585         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1586         if (features) {
1587                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1588                        "unsupported optional features (%x).\n",
1589                        sb->s_id, le32_to_cpu(features));
1590                 goto failed_mount;
1591         }
1592         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1593         if (!(sb->s_flags & MS_RDONLY) && features) {
1594                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1595                        "unsupported optional features (%x).\n",
1596                        sb->s_id, le32_to_cpu(features));
1597                 goto failed_mount;
1598         }
1599         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1600
1601         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1602             blocksize > EXT4_MAX_BLOCK_SIZE) {
1603                 printk(KERN_ERR
1604                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1605                        blocksize, sb->s_id);
1606                 goto failed_mount;
1607         }
1608
1609         hblock = bdev_hardsect_size(sb->s_bdev);
1610         if (sb->s_blocksize != blocksize) {
1611                 /*
1612                  * Make sure the blocksize for the filesystem is larger
1613                  * than the hardware sectorsize for the machine.
1614                  */
1615                 if (blocksize < hblock) {
1616                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1617                                "device blocksize %d.\n", blocksize, hblock);
1618                         goto failed_mount;
1619                 }
1620
1621                 brelse (bh);
1622                 sb_set_blocksize(sb, blocksize);
1623                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1624                 offset = do_div(logical_sb_block, blocksize);
1625                 bh = sb_bread(sb, logical_sb_block);
1626                 if (!bh) {
1627                         printk(KERN_ERR
1628                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1629                         goto failed_mount;
1630                 }
1631                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1632                 sbi->s_es = es;
1633                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1634                         printk (KERN_ERR
1635                                 "EXT4-fs: Magic mismatch, very weird !\n");
1636                         goto failed_mount;
1637                 }
1638         }
1639
1640         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1641
1642         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1643                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1644                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1645         } else {
1646                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1647                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1648                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1649                     (!is_power_of_2(sbi->s_inode_size)) ||
1650                     (sbi->s_inode_size > blocksize)) {
1651                         printk (KERN_ERR
1652                                 "EXT4-fs: unsupported inode size: %d\n",
1653                                 sbi->s_inode_size);
1654                         goto failed_mount;
1655                 }
1656                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
1657                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
1658         }
1659         sbi->s_frag_size = EXT4_MIN_FRAG_SIZE <<
1660                                    le32_to_cpu(es->s_log_frag_size);
1661         if (blocksize != sbi->s_frag_size) {
1662                 printk(KERN_ERR
1663                        "EXT4-fs: fragsize %lu != blocksize %u (unsupported)\n",
1664                        sbi->s_frag_size, blocksize);
1665                 goto failed_mount;
1666         }
1667         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1668         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1669                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1670                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1671                     sbi->s_desc_size & (sbi->s_desc_size - 1)) {
1672                         printk(KERN_ERR
1673                                "EXT4-fs: unsupported descriptor size %lu\n",
1674                                sbi->s_desc_size);
1675                         goto failed_mount;
1676                 }
1677         } else
1678                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1679         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1680         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1681         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1682         if (EXT4_INODE_SIZE(sb) == 0)
1683                 goto cantfind_ext4;
1684         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1685         if (sbi->s_inodes_per_block == 0)
1686                 goto cantfind_ext4;
1687         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1688                                         sbi->s_inodes_per_block;
1689         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1690         sbi->s_sbh = bh;
1691         sbi->s_mount_state = le16_to_cpu(es->s_state);
1692         sbi->s_addr_per_block_bits = log2(EXT4_ADDR_PER_BLOCK(sb));
1693         sbi->s_desc_per_block_bits = log2(EXT4_DESC_PER_BLOCK(sb));
1694         for (i=0; i < 4; i++)
1695                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1696         sbi->s_def_hash_version = es->s_def_hash_version;
1697
1698         if (sbi->s_blocks_per_group > blocksize * 8) {
1699                 printk (KERN_ERR
1700                         "EXT4-fs: #blocks per group too big: %lu\n",
1701                         sbi->s_blocks_per_group);
1702                 goto failed_mount;
1703         }
1704         if (sbi->s_frags_per_group > blocksize * 8) {
1705                 printk (KERN_ERR
1706                         "EXT4-fs: #fragments per group too big: %lu\n",
1707                         sbi->s_frags_per_group);
1708                 goto failed_mount;
1709         }
1710         if (sbi->s_inodes_per_group > blocksize * 8) {
1711                 printk (KERN_ERR
1712                         "EXT4-fs: #inodes per group too big: %lu\n",
1713                         sbi->s_inodes_per_group);
1714                 goto failed_mount;
1715         }
1716
1717         if (ext4_blocks_count(es) >
1718                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1719                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1720                         " too large to mount safely\n", sb->s_id);
1721                 if (sizeof(sector_t) < 8)
1722                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1723                                         "enabled\n");
1724                 goto failed_mount;
1725         }
1726
1727         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1728                 goto cantfind_ext4;
1729         blocks_count = (ext4_blocks_count(es) -
1730                         le32_to_cpu(es->s_first_data_block) +
1731                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1732         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1733         sbi->s_groups_count = blocks_count;
1734         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1735                    EXT4_DESC_PER_BLOCK(sb);
1736         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1737                                     GFP_KERNEL);
1738         if (sbi->s_group_desc == NULL) {
1739                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1740                 goto failed_mount;
1741         }
1742
1743         bgl_lock_init(&sbi->s_blockgroup_lock);
1744
1745         for (i = 0; i < db_count; i++) {
1746                 block = descriptor_loc(sb, logical_sb_block, i);
1747                 sbi->s_group_desc[i] = sb_bread(sb, block);
1748                 if (!sbi->s_group_desc[i]) {
1749                         printk (KERN_ERR "EXT4-fs: "
1750                                 "can't read group descriptor %d\n", i);
1751                         db_count = i;
1752                         goto failed_mount2;
1753                 }
1754         }
1755         if (!ext4_check_descriptors (sb)) {
1756                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1757                 goto failed_mount2;
1758         }
1759         sbi->s_gdb_count = db_count;
1760         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1761         spin_lock_init(&sbi->s_next_gen_lock);
1762
1763         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1764                         ext4_count_free_blocks(sb));
1765         if (!err) {
1766                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1767                                 ext4_count_free_inodes(sb));
1768         }
1769         if (!err) {
1770                 err = percpu_counter_init(&sbi->s_dirs_counter,
1771                                 ext4_count_dirs(sb));
1772         }
1773         if (err) {
1774                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
1775                 goto failed_mount3;
1776         }
1777
1778         /* per fileystem reservation list head & lock */
1779         spin_lock_init(&sbi->s_rsv_window_lock);
1780         sbi->s_rsv_window_root = RB_ROOT;
1781         /* Add a single, static dummy reservation to the start of the
1782          * reservation window list --- it gives us a placeholder for
1783          * append-at-start-of-list which makes the allocation logic
1784          * _much_ simpler. */
1785         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1786         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1787         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1788         sbi->s_rsv_window_head.rsv_goal_size = 0;
1789         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1790
1791         /*
1792          * set up enough so that it can read an inode
1793          */
1794         sb->s_op = &ext4_sops;
1795         sb->s_export_op = &ext4_export_ops;
1796         sb->s_xattr = ext4_xattr_handlers;
1797 #ifdef CONFIG_QUOTA
1798         sb->s_qcop = &ext4_qctl_operations;
1799         sb->dq_op = &ext4_quota_operations;
1800 #endif
1801         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1802
1803         sb->s_root = NULL;
1804
1805         needs_recovery = (es->s_last_orphan != 0 ||
1806                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1807                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1808
1809         /*
1810          * The first inode we look at is the journal inode.  Don't try
1811          * root first: it may be modified in the journal!
1812          */
1813         if (!test_opt(sb, NOLOAD) &&
1814             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1815                 if (ext4_load_journal(sb, es, journal_devnum))
1816                         goto failed_mount3;
1817         } else if (journal_inum) {
1818                 if (ext4_create_journal(sb, es, journal_inum))
1819                         goto failed_mount3;
1820         } else {
1821                 if (!silent)
1822                         printk (KERN_ERR
1823                                 "ext4: No journal on filesystem on %s\n",
1824                                 sb->s_id);
1825                 goto failed_mount3;
1826         }
1827
1828         if (ext4_blocks_count(es) > 0xffffffffULL &&
1829             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
1830                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
1831                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
1832                 goto failed_mount4;
1833         }
1834
1835         /* We have now updated the journal if required, so we can
1836          * validate the data journaling mode. */
1837         switch (test_opt(sb, DATA_FLAGS)) {
1838         case 0:
1839                 /* No mode set, assume a default based on the journal
1840                  * capabilities: ORDERED_DATA if the journal can
1841                  * cope, else JOURNAL_DATA
1842                  */
1843                 if (jbd2_journal_check_available_features
1844                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
1845                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1846                 else
1847                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1848                 break;
1849
1850         case EXT4_MOUNT_ORDERED_DATA:
1851         case EXT4_MOUNT_WRITEBACK_DATA:
1852                 if (!jbd2_journal_check_available_features
1853                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
1854                         printk(KERN_ERR "EXT4-fs: Journal does not support "
1855                                "requested data journaling mode\n");
1856                         goto failed_mount4;
1857                 }
1858         default:
1859                 break;
1860         }
1861
1862         if (test_opt(sb, NOBH)) {
1863                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
1864                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
1865                                 "its supported only with writeback mode\n");
1866                         clear_opt(sbi->s_mount_opt, NOBH);
1867                 }
1868         }
1869         /*
1870          * The jbd2_journal_load will have done any necessary log recovery,
1871          * so we can safely mount the rest of the filesystem now.
1872          */
1873
1874         root = iget(sb, EXT4_ROOT_INO);
1875         sb->s_root = d_alloc_root(root);
1876         if (!sb->s_root) {
1877                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
1878                 iput(root);
1879                 goto failed_mount4;
1880         }
1881         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1882                 dput(sb->s_root);
1883                 sb->s_root = NULL;
1884                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
1885                 goto failed_mount4;
1886         }
1887
1888         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1889
1890         /* determine the minimum size of new large inodes, if present */
1891         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
1892                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
1893                                                      EXT4_GOOD_OLD_INODE_SIZE;
1894                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1895                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
1896                         if (sbi->s_want_extra_isize <
1897                             le16_to_cpu(es->s_want_extra_isize))
1898                                 sbi->s_want_extra_isize =
1899                                         le16_to_cpu(es->s_want_extra_isize);
1900                         if (sbi->s_want_extra_isize <
1901                             le16_to_cpu(es->s_min_extra_isize))
1902                                 sbi->s_want_extra_isize =
1903                                         le16_to_cpu(es->s_min_extra_isize);
1904                 }
1905         }
1906         /* Check if enough inode space is available */
1907         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
1908                                                         sbi->s_inode_size) {
1909                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
1910                                                        EXT4_GOOD_OLD_INODE_SIZE;
1911                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
1912                         "available.\n");
1913         }
1914
1915         /*
1916          * akpm: core read_super() calls in here with the superblock locked.
1917          * That deadlocks, because orphan cleanup needs to lock the superblock
1918          * in numerous places.  Here we just pop the lock - it's relatively
1919          * harmless, because we are now ready to accept write_super() requests,
1920          * and aviro says that's the only reason for hanging onto the
1921          * superblock lock.
1922          */
1923         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
1924         ext4_orphan_cleanup(sb, es);
1925         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
1926         if (needs_recovery)
1927                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
1928         ext4_mark_recovery_complete(sb, es);
1929         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
1930                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
1931                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
1932                 "writeback");
1933
1934         ext4_ext_init(sb);
1935
1936         lock_kernel();
1937         return 0;
1938
1939 cantfind_ext4:
1940         if (!silent)
1941                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
1942                        sb->s_id);
1943         goto failed_mount;
1944
1945 failed_mount4:
1946         jbd2_journal_destroy(sbi->s_journal);
1947 failed_mount3:
1948         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1949         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1950         percpu_counter_destroy(&sbi->s_dirs_counter);
1951 failed_mount2:
1952         for (i = 0; i < db_count; i++)
1953                 brelse(sbi->s_group_desc[i]);
1954         kfree(sbi->s_group_desc);
1955 failed_mount:
1956 #ifdef CONFIG_QUOTA
1957         for (i = 0; i < MAXQUOTAS; i++)
1958                 kfree(sbi->s_qf_names[i]);
1959 #endif
1960         ext4_blkdev_remove(sbi);
1961         brelse(bh);
1962 out_fail:
1963         sb->s_fs_info = NULL;
1964         kfree(sbi);
1965         lock_kernel();
1966         return -EINVAL;
1967 }
1968
1969 /*
1970  * Setup any per-fs journal parameters now.  We'll do this both on
1971  * initial mount, once the journal has been initialised but before we've
1972  * done any recovery; and again on any subsequent remount.
1973  */
1974 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
1975 {
1976         struct ext4_sb_info *sbi = EXT4_SB(sb);
1977
1978         if (sbi->s_commit_interval)
1979                 journal->j_commit_interval = sbi->s_commit_interval;
1980         /* We could also set up an ext4-specific default for the commit
1981          * interval here, but for now we'll just fall back to the jbd
1982          * default. */
1983
1984         spin_lock(&journal->j_state_lock);
1985         if (test_opt(sb, BARRIER))
1986                 journal->j_flags |= JBD2_BARRIER;
1987         else
1988                 journal->j_flags &= ~JBD2_BARRIER;
1989         spin_unlock(&journal->j_state_lock);
1990 }
1991
1992 static journal_t *ext4_get_journal(struct super_block *sb,
1993                                    unsigned int journal_inum)
1994 {
1995         struct inode *journal_inode;
1996         journal_t *journal;
1997
1998         /* First, test for the existence of a valid inode on disk.  Bad
1999          * things happen if we iget() an unused inode, as the subsequent
2000          * iput() will try to delete it. */
2001
2002         journal_inode = iget(sb, journal_inum);
2003         if (!journal_inode) {
2004                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2005                 return NULL;
2006         }
2007         if (!journal_inode->i_nlink) {
2008                 make_bad_inode(journal_inode);
2009                 iput(journal_inode);
2010                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2011                 return NULL;
2012         }
2013
2014         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2015                   journal_inode, journal_inode->i_size);
2016         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2017                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2018                 iput(journal_inode);
2019                 return NULL;
2020         }
2021
2022         journal = jbd2_journal_init_inode(journal_inode);
2023         if (!journal) {
2024                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2025                 iput(journal_inode);
2026                 return NULL;
2027         }
2028         journal->j_private = sb;
2029         ext4_init_journal_params(sb, journal);
2030         return journal;
2031 }
2032
2033 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2034                                        dev_t j_dev)
2035 {
2036         struct buffer_head * bh;
2037         journal_t *journal;
2038         ext4_fsblk_t start;
2039         ext4_fsblk_t len;
2040         int hblock, blocksize;
2041         ext4_fsblk_t sb_block;
2042         unsigned long offset;
2043         struct ext4_super_block * es;
2044         struct block_device *bdev;
2045
2046         bdev = ext4_blkdev_get(j_dev);
2047         if (bdev == NULL)
2048                 return NULL;
2049
2050         if (bd_claim(bdev, sb)) {
2051                 printk(KERN_ERR
2052                         "EXT4: failed to claim external journal device.\n");
2053                 blkdev_put(bdev);
2054                 return NULL;
2055         }
2056
2057         blocksize = sb->s_blocksize;
2058         hblock = bdev_hardsect_size(bdev);
2059         if (blocksize < hblock) {
2060                 printk(KERN_ERR
2061                         "EXT4-fs: blocksize too small for journal device.\n");
2062                 goto out_bdev;
2063         }
2064
2065         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2066         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2067         set_blocksize(bdev, blocksize);
2068         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2069                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2070                        "external journal\n");
2071                 goto out_bdev;
2072         }
2073
2074         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2075         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2076             !(le32_to_cpu(es->s_feature_incompat) &
2077               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2078                 printk(KERN_ERR "EXT4-fs: external journal has "
2079                                         "bad superblock\n");
2080                 brelse(bh);
2081                 goto out_bdev;
2082         }
2083
2084         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2085                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2086                 brelse(bh);
2087                 goto out_bdev;
2088         }
2089
2090         len = ext4_blocks_count(es);
2091         start = sb_block + 1;
2092         brelse(bh);     /* we're done with the superblock */
2093
2094         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2095                                         start, len, blocksize);
2096         if (!journal) {
2097                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2098                 goto out_bdev;
2099         }
2100         journal->j_private = sb;
2101         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2102         wait_on_buffer(journal->j_sb_buffer);
2103         if (!buffer_uptodate(journal->j_sb_buffer)) {
2104                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2105                 goto out_journal;
2106         }
2107         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2108                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2109                                         "user (unsupported) - %d\n",
2110                         be32_to_cpu(journal->j_superblock->s_nr_users));
2111                 goto out_journal;
2112         }
2113         EXT4_SB(sb)->journal_bdev = bdev;
2114         ext4_init_journal_params(sb, journal);
2115         return journal;
2116 out_journal:
2117         jbd2_journal_destroy(journal);
2118 out_bdev:
2119         ext4_blkdev_put(bdev);
2120         return NULL;
2121 }
2122
2123 static int ext4_load_journal(struct super_block *sb,
2124                              struct ext4_super_block *es,
2125                              unsigned long journal_devnum)
2126 {
2127         journal_t *journal;
2128         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2129         dev_t journal_dev;
2130         int err = 0;
2131         int really_read_only;
2132
2133         if (journal_devnum &&
2134             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2135                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2136                         "numbers have changed\n");
2137                 journal_dev = new_decode_dev(journal_devnum);
2138         } else
2139                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2140
2141         really_read_only = bdev_read_only(sb->s_bdev);
2142
2143         /*
2144          * Are we loading a blank journal or performing recovery after a
2145          * crash?  For recovery, we need to check in advance whether we
2146          * can get read-write access to the device.
2147          */
2148
2149         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2150                 if (sb->s_flags & MS_RDONLY) {
2151                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2152                                         "required on readonly filesystem.\n");
2153                         if (really_read_only) {
2154                                 printk(KERN_ERR "EXT4-fs: write access "
2155                                         "unavailable, cannot proceed.\n");
2156                                 return -EROFS;
2157                         }
2158                         printk (KERN_INFO "EXT4-fs: write access will "
2159                                         "be enabled during recovery.\n");
2160                 }
2161         }
2162
2163         if (journal_inum && journal_dev) {
2164                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2165                        "and inode journals!\n");
2166                 return -EINVAL;
2167         }
2168
2169         if (journal_inum) {
2170                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2171                         return -EINVAL;
2172         } else {
2173                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2174                         return -EINVAL;
2175         }
2176
2177         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2178                 err = jbd2_journal_update_format(journal);
2179                 if (err)  {
2180                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2181                         jbd2_journal_destroy(journal);
2182                         return err;
2183                 }
2184         }
2185
2186         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2187                 err = jbd2_journal_wipe(journal, !really_read_only);
2188         if (!err)
2189                 err = jbd2_journal_load(journal);
2190
2191         if (err) {
2192                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2193                 jbd2_journal_destroy(journal);
2194                 return err;
2195         }
2196
2197         EXT4_SB(sb)->s_journal = journal;
2198         ext4_clear_journal_err(sb, es);
2199
2200         if (journal_devnum &&
2201             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2202                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2203                 sb->s_dirt = 1;
2204
2205                 /* Make sure we flush the recovery flag to disk. */
2206                 ext4_commit_super(sb, es, 1);
2207         }
2208
2209         return 0;
2210 }
2211
2212 static int ext4_create_journal(struct super_block * sb,
2213                                struct ext4_super_block * es,
2214                                unsigned int journal_inum)
2215 {
2216         journal_t *journal;
2217         int err;
2218
2219         if (sb->s_flags & MS_RDONLY) {
2220                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2221                                 "create journal.\n");
2222                 return -EROFS;
2223         }
2224
2225         journal = ext4_get_journal(sb, journal_inum);
2226         if (!journal)
2227                 return -EINVAL;
2228
2229         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2230                journal_inum);
2231
2232         err = jbd2_journal_create(journal);
2233         if (err) {
2234                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2235                 jbd2_journal_destroy(journal);
2236                 return -EIO;
2237         }
2238
2239         EXT4_SB(sb)->s_journal = journal;
2240
2241         ext4_update_dynamic_rev(sb);
2242         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2243         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2244
2245         es->s_journal_inum = cpu_to_le32(journal_inum);
2246         sb->s_dirt = 1;
2247
2248         /* Make sure we flush the recovery flag to disk. */
2249         ext4_commit_super(sb, es, 1);
2250
2251         return 0;
2252 }
2253
2254 static void ext4_commit_super (struct super_block * sb,
2255                                struct ext4_super_block * es,
2256                                int sync)
2257 {
2258         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2259
2260         if (!sbh)
2261                 return;
2262         es->s_wtime = cpu_to_le32(get_seconds());
2263         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2264         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2265         BUFFER_TRACE(sbh, "marking dirty");
2266         mark_buffer_dirty(sbh);
2267         if (sync)
2268                 sync_dirty_buffer(sbh);
2269 }
2270
2271
2272 /*
2273  * Have we just finished recovery?  If so, and if we are mounting (or
2274  * remounting) the filesystem readonly, then we will end up with a
2275  * consistent fs on disk.  Record that fact.
2276  */
2277 static void ext4_mark_recovery_complete(struct super_block * sb,
2278                                         struct ext4_super_block * es)
2279 {
2280         journal_t *journal = EXT4_SB(sb)->s_journal;
2281
2282         jbd2_journal_lock_updates(journal);
2283         jbd2_journal_flush(journal);
2284         lock_super(sb);
2285         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2286             sb->s_flags & MS_RDONLY) {
2287                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2288                 sb->s_dirt = 0;
2289                 ext4_commit_super(sb, es, 1);
2290         }
2291         unlock_super(sb);
2292         jbd2_journal_unlock_updates(journal);
2293 }
2294
2295 /*
2296  * If we are mounting (or read-write remounting) a filesystem whose journal
2297  * has recorded an error from a previous lifetime, move that error to the
2298  * main filesystem now.
2299  */
2300 static void ext4_clear_journal_err(struct super_block * sb,
2301                                    struct ext4_super_block * es)
2302 {
2303         journal_t *journal;
2304         int j_errno;
2305         const char *errstr;
2306
2307         journal = EXT4_SB(sb)->s_journal;
2308
2309         /*
2310          * Now check for any error status which may have been recorded in the
2311          * journal by a prior ext4_error() or ext4_abort()
2312          */
2313
2314         j_errno = jbd2_journal_errno(journal);
2315         if (j_errno) {
2316                 char nbuf[16];
2317
2318                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2319                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2320                              "from previous mount: %s", errstr);
2321                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2322                              "filesystem check.");
2323
2324                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2325                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2326                 ext4_commit_super (sb, es, 1);
2327
2328                 jbd2_journal_clear_err(journal);
2329         }
2330 }
2331
2332 /*
2333  * Force the running and committing transactions to commit,
2334  * and wait on the commit.
2335  */
2336 int ext4_force_commit(struct super_block *sb)
2337 {
2338         journal_t *journal;
2339         int ret;
2340
2341         if (sb->s_flags & MS_RDONLY)
2342                 return 0;
2343
2344         journal = EXT4_SB(sb)->s_journal;
2345         sb->s_dirt = 0;
2346         ret = ext4_journal_force_commit(journal);
2347         return ret;
2348 }
2349
2350 /*
2351  * Ext4 always journals updates to the superblock itself, so we don't
2352  * have to propagate any other updates to the superblock on disk at this
2353  * point.  Just start an async writeback to get the buffers on their way
2354  * to the disk.
2355  *
2356  * This implicitly triggers the writebehind on sync().
2357  */
2358
2359 static void ext4_write_super (struct super_block * sb)
2360 {
2361         if (mutex_trylock(&sb->s_lock) != 0)
2362                 BUG();
2363         sb->s_dirt = 0;
2364 }
2365
2366 static int ext4_sync_fs(struct super_block *sb, int wait)
2367 {
2368         tid_t target;
2369
2370         sb->s_dirt = 0;
2371         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2372                 if (wait)
2373                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2374         }
2375         return 0;
2376 }
2377
2378 /*
2379  * LVM calls this function before a (read-only) snapshot is created.  This
2380  * gives us a chance to flush the journal completely and mark the fs clean.
2381  */
2382 static void ext4_write_super_lockfs(struct super_block *sb)
2383 {
2384         sb->s_dirt = 0;
2385
2386         if (!(sb->s_flags & MS_RDONLY)) {
2387                 journal_t *journal = EXT4_SB(sb)->s_journal;
2388
2389                 /* Now we set up the journal barrier. */
2390                 jbd2_journal_lock_updates(journal);
2391                 jbd2_journal_flush(journal);
2392
2393                 /* Journal blocked and flushed, clear needs_recovery flag. */
2394                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2395                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2396         }
2397 }
2398
2399 /*
2400  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2401  * flag here, even though the filesystem is not technically dirty yet.
2402  */
2403 static void ext4_unlockfs(struct super_block *sb)
2404 {
2405         if (!(sb->s_flags & MS_RDONLY)) {
2406                 lock_super(sb);
2407                 /* Reser the needs_recovery flag before the fs is unlocked. */
2408                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2409                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2410                 unlock_super(sb);
2411                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2412         }
2413 }
2414
2415 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2416 {
2417         struct ext4_super_block * es;
2418         struct ext4_sb_info *sbi = EXT4_SB(sb);
2419         ext4_fsblk_t n_blocks_count = 0;
2420         unsigned long old_sb_flags;
2421         struct ext4_mount_options old_opts;
2422         int err;
2423 #ifdef CONFIG_QUOTA
2424         int i;
2425 #endif
2426
2427         /* Store the original options */
2428         old_sb_flags = sb->s_flags;
2429         old_opts.s_mount_opt = sbi->s_mount_opt;
2430         old_opts.s_resuid = sbi->s_resuid;
2431         old_opts.s_resgid = sbi->s_resgid;
2432         old_opts.s_commit_interval = sbi->s_commit_interval;
2433 #ifdef CONFIG_QUOTA
2434         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2435         for (i = 0; i < MAXQUOTAS; i++)
2436                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2437 #endif
2438
2439         /*
2440          * Allow the "check" option to be passed as a remount option.
2441          */
2442         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2443                 err = -EINVAL;
2444                 goto restore_opts;
2445         }
2446
2447         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2448                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2449
2450         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2451                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2452
2453         es = sbi->s_es;
2454
2455         ext4_init_journal_params(sb, sbi->s_journal);
2456
2457         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2458                 n_blocks_count > ext4_blocks_count(es)) {
2459                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2460                         err = -EROFS;
2461                         goto restore_opts;
2462                 }
2463
2464                 if (*flags & MS_RDONLY) {
2465                         /*
2466                          * First of all, the unconditional stuff we have to do
2467                          * to disable replay of the journal when we next remount
2468                          */
2469                         sb->s_flags |= MS_RDONLY;
2470
2471                         /*
2472                          * OK, test if we are remounting a valid rw partition
2473                          * readonly, and if so set the rdonly flag and then
2474                          * mark the partition as valid again.
2475                          */
2476                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2477                             (sbi->s_mount_state & EXT4_VALID_FS))
2478                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2479
2480                         /*
2481                          * We have to unlock super so that we can wait for
2482                          * transactions.
2483                          */
2484                         unlock_super(sb);
2485                         ext4_mark_recovery_complete(sb, es);
2486                         lock_super(sb);
2487                 } else {
2488                         __le32 ret;
2489                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2490                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2491                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2492                                        "remount RDWR because of unsupported "
2493                                        "optional features (%x).\n",
2494                                        sb->s_id, le32_to_cpu(ret));
2495                                 err = -EROFS;
2496                                 goto restore_opts;
2497                         }
2498
2499                         /*
2500                          * If we have an unprocessed orphan list hanging
2501                          * around from a previously readonly bdev mount,
2502                          * require a full umount/remount for now.
2503                          */
2504                         if (es->s_last_orphan) {
2505                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2506                                        "remount RDWR because of unprocessed "
2507                                        "orphan inode list.  Please "
2508                                        "umount/remount instead.\n",
2509                                        sb->s_id);
2510                                 err = -EINVAL;
2511                                 goto restore_opts;
2512                         }
2513
2514                         /*
2515                          * Mounting a RDONLY partition read-write, so reread
2516                          * and store the current valid flag.  (It may have
2517                          * been changed by e2fsck since we originally mounted
2518                          * the partition.)
2519                          */
2520                         ext4_clear_journal_err(sb, es);
2521                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2522                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2523                                 goto restore_opts;
2524                         if (!ext4_setup_super (sb, es, 0))
2525                                 sb->s_flags &= ~MS_RDONLY;
2526                 }
2527         }
2528 #ifdef CONFIG_QUOTA
2529         /* Release old quota file names */
2530         for (i = 0; i < MAXQUOTAS; i++)
2531                 if (old_opts.s_qf_names[i] &&
2532                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2533                         kfree(old_opts.s_qf_names[i]);
2534 #endif
2535         return 0;
2536 restore_opts:
2537         sb->s_flags = old_sb_flags;
2538         sbi->s_mount_opt = old_opts.s_mount_opt;
2539         sbi->s_resuid = old_opts.s_resuid;
2540         sbi->s_resgid = old_opts.s_resgid;
2541         sbi->s_commit_interval = old_opts.s_commit_interval;
2542 #ifdef CONFIG_QUOTA
2543         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2544         for (i = 0; i < MAXQUOTAS; i++) {
2545                 if (sbi->s_qf_names[i] &&
2546                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2547                         kfree(sbi->s_qf_names[i]);
2548                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2549         }
2550 #endif
2551         return err;
2552 }
2553
2554 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2555 {
2556         struct super_block *sb = dentry->d_sb;
2557         struct ext4_sb_info *sbi = EXT4_SB(sb);
2558         struct ext4_super_block *es = sbi->s_es;
2559         u64 fsid;
2560
2561         if (test_opt(sb, MINIX_DF)) {
2562                 sbi->s_overhead_last = 0;
2563         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2564                 unsigned long ngroups = sbi->s_groups_count, i;
2565                 ext4_fsblk_t overhead = 0;
2566                 smp_rmb();
2567
2568                 /*
2569                  * Compute the overhead (FS structures).  This is constant
2570                  * for a given filesystem unless the number of block groups
2571                  * changes so we cache the previous value until it does.
2572                  */
2573
2574                 /*
2575                  * All of the blocks before first_data_block are
2576                  * overhead
2577                  */
2578                 overhead = le32_to_cpu(es->s_first_data_block);
2579
2580                 /*
2581                  * Add the overhead attributed to the superblock and
2582                  * block group descriptors.  If the sparse superblocks
2583                  * feature is turned on, then not all groups have this.
2584                  */
2585                 for (i = 0; i < ngroups; i++) {
2586                         overhead += ext4_bg_has_super(sb, i) +
2587                                 ext4_bg_num_gdb(sb, i);
2588                         cond_resched();
2589                 }
2590
2591                 /*
2592                  * Every block group has an inode bitmap, a block
2593                  * bitmap, and an inode table.
2594                  */
2595                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2596                 sbi->s_overhead_last = overhead;
2597                 smp_wmb();
2598                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2599         }
2600
2601         buf->f_type = EXT4_SUPER_MAGIC;
2602         buf->f_bsize = sb->s_blocksize;
2603         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2604         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2605         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2606         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2607         if (buf->f_bfree < ext4_r_blocks_count(es))
2608                 buf->f_bavail = 0;
2609         buf->f_files = le32_to_cpu(es->s_inodes_count);
2610         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2611         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2612         buf->f_namelen = EXT4_NAME_LEN;
2613         fsid = le64_to_cpup((void *)es->s_uuid) ^
2614                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2615         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2616         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2617         return 0;
2618 }
2619
2620 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2621  * is locked for write. Otherwise the are possible deadlocks:
2622  * Process 1                         Process 2
2623  * ext4_create()                     quota_sync()
2624  *   jbd2_journal_start()                   write_dquot()
2625  *   DQUOT_INIT()                        down(dqio_mutex)
2626  *     down(dqio_mutex)                    jbd2_journal_start()
2627  *
2628  */
2629
2630 #ifdef CONFIG_QUOTA
2631
2632 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2633 {
2634         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2635 }
2636
2637 static int ext4_dquot_initialize(struct inode *inode, int type)
2638 {
2639         handle_t *handle;
2640         int ret, err;
2641
2642         /* We may create quota structure so we need to reserve enough blocks */
2643         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2644         if (IS_ERR(handle))
2645                 return PTR_ERR(handle);
2646         ret = dquot_initialize(inode, type);
2647         err = ext4_journal_stop(handle);
2648         if (!ret)
2649                 ret = err;
2650         return ret;
2651 }
2652
2653 static int ext4_dquot_drop(struct inode *inode)
2654 {
2655         handle_t *handle;
2656         int ret, err;
2657
2658         /* We may delete quota structure so we need to reserve enough blocks */
2659         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2660         if (IS_ERR(handle))
2661                 return PTR_ERR(handle);
2662         ret = dquot_drop(inode);
2663         err = ext4_journal_stop(handle);
2664         if (!ret)
2665                 ret = err;
2666         return ret;
2667 }
2668
2669 static int ext4_write_dquot(struct dquot *dquot)
2670 {
2671         int ret, err;
2672         handle_t *handle;
2673         struct inode *inode;
2674
2675         inode = dquot_to_inode(dquot);
2676         handle = ext4_journal_start(inode,
2677                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2678         if (IS_ERR(handle))
2679                 return PTR_ERR(handle);
2680         ret = dquot_commit(dquot);
2681         err = ext4_journal_stop(handle);
2682         if (!ret)
2683                 ret = err;
2684         return ret;
2685 }
2686
2687 static int ext4_acquire_dquot(struct dquot *dquot)
2688 {
2689         int ret, err;
2690         handle_t *handle;
2691
2692         handle = ext4_journal_start(dquot_to_inode(dquot),
2693                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2694         if (IS_ERR(handle))
2695                 return PTR_ERR(handle);
2696         ret = dquot_acquire(dquot);
2697         err = ext4_journal_stop(handle);
2698         if (!ret)
2699                 ret = err;
2700         return ret;
2701 }
2702
2703 static int ext4_release_dquot(struct dquot *dquot)
2704 {
2705         int ret, err;
2706         handle_t *handle;
2707
2708         handle = ext4_journal_start(dquot_to_inode(dquot),
2709                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2710         if (IS_ERR(handle)) {
2711                 /* Release dquot anyway to avoid endless cycle in dqput() */
2712                 dquot_release(dquot);
2713                 return PTR_ERR(handle);
2714         }
2715         ret = dquot_release(dquot);
2716         err = ext4_journal_stop(handle);
2717         if (!ret)
2718                 ret = err;
2719         return ret;
2720 }
2721
2722 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2723 {
2724         /* Are we journalling quotas? */
2725         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2726             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2727                 dquot_mark_dquot_dirty(dquot);
2728                 return ext4_write_dquot(dquot);
2729         } else {
2730                 return dquot_mark_dquot_dirty(dquot);
2731         }
2732 }
2733
2734 static int ext4_write_info(struct super_block *sb, int type)
2735 {
2736         int ret, err;
2737         handle_t *handle;
2738
2739         /* Data block + inode block */
2740         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2741         if (IS_ERR(handle))
2742                 return PTR_ERR(handle);
2743         ret = dquot_commit_info(sb, type);
2744         err = ext4_journal_stop(handle);
2745         if (!ret)
2746                 ret = err;
2747         return ret;
2748 }
2749
2750 /*
2751  * Turn on quotas during mount time - we need to find
2752  * the quota file and such...
2753  */
2754 static int ext4_quota_on_mount(struct super_block *sb, int type)
2755 {
2756         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2757                         EXT4_SB(sb)->s_jquota_fmt, type);
2758 }
2759
2760 /*
2761  * Standard function to be called on quota_on
2762  */
2763 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2764                          char *path)
2765 {
2766         int err;
2767         struct nameidata nd;
2768
2769         if (!test_opt(sb, QUOTA))
2770                 return -EINVAL;
2771         /* Not journalling quota? */
2772         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2773             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2774                 return vfs_quota_on(sb, type, format_id, path);
2775         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2776         if (err)
2777                 return err;
2778         /* Quotafile not on the same filesystem? */
2779         if (nd.mnt->mnt_sb != sb) {
2780                 path_release(&nd);
2781                 return -EXDEV;
2782         }
2783         /* Quotafile not of fs root? */
2784         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2785                 printk(KERN_WARNING
2786                         "EXT4-fs: Quota file not on filesystem root. "
2787                         "Journalled quota will not work.\n");
2788         path_release(&nd);
2789         return vfs_quota_on(sb, type, format_id, path);
2790 }
2791
2792 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2793  * acquiring the locks... As quota files are never truncated and quota code
2794  * itself serializes the operations (and noone else should touch the files)
2795  * we don't have to be afraid of races */
2796 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2797                                size_t len, loff_t off)
2798 {
2799         struct inode *inode = sb_dqopt(sb)->files[type];
2800         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2801         int err = 0;
2802         int offset = off & (sb->s_blocksize - 1);
2803         int tocopy;
2804         size_t toread;
2805         struct buffer_head *bh;
2806         loff_t i_size = i_size_read(inode);
2807
2808         if (off > i_size)
2809                 return 0;
2810         if (off+len > i_size)
2811                 len = i_size-off;
2812         toread = len;
2813         while (toread > 0) {
2814                 tocopy = sb->s_blocksize - offset < toread ?
2815                                 sb->s_blocksize - offset : toread;
2816                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2817                 if (err)
2818                         return err;
2819                 if (!bh)        /* A hole? */
2820                         memset(data, 0, tocopy);
2821                 else
2822                         memcpy(data, bh->b_data+offset, tocopy);
2823                 brelse(bh);
2824                 offset = 0;
2825                 toread -= tocopy;
2826                 data += tocopy;
2827                 blk++;
2828         }
2829         return len;
2830 }
2831
2832 /* Write to quotafile (we know the transaction is already started and has
2833  * enough credits) */
2834 static ssize_t ext4_quota_write(struct super_block *sb, int type,
2835                                 const char *data, size_t len, loff_t off)
2836 {
2837         struct inode *inode = sb_dqopt(sb)->files[type];
2838         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2839         int err = 0;
2840         int offset = off & (sb->s_blocksize - 1);
2841         int tocopy;
2842         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
2843         size_t towrite = len;
2844         struct buffer_head *bh;
2845         handle_t *handle = journal_current_handle();
2846
2847         if (!handle) {
2848                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
2849                         " cancelled because transaction is not started.\n",
2850                         (unsigned long long)off, (unsigned long long)len);
2851                 return -EIO;
2852         }
2853         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2854         while (towrite > 0) {
2855                 tocopy = sb->s_blocksize - offset < towrite ?
2856                                 sb->s_blocksize - offset : towrite;
2857                 bh = ext4_bread(handle, inode, blk, 1, &err);
2858                 if (!bh)
2859                         goto out;
2860                 if (journal_quota) {
2861                         err = ext4_journal_get_write_access(handle, bh);
2862                         if (err) {
2863                                 brelse(bh);
2864                                 goto out;
2865                         }
2866                 }
2867                 lock_buffer(bh);
2868                 memcpy(bh->b_data+offset, data, tocopy);
2869                 flush_dcache_page(bh->b_page);
2870                 unlock_buffer(bh);
2871                 if (journal_quota)
2872                         err = ext4_journal_dirty_metadata(handle, bh);
2873                 else {
2874                         /* Always do at least ordered writes for quotas */
2875                         err = ext4_journal_dirty_data(handle, bh);
2876                         mark_buffer_dirty(bh);
2877                 }
2878                 brelse(bh);
2879                 if (err)
2880                         goto out;
2881                 offset = 0;
2882                 towrite -= tocopy;
2883                 data += tocopy;
2884                 blk++;
2885         }
2886 out:
2887         if (len == towrite)
2888                 return err;
2889         if (inode->i_size < off+len-towrite) {
2890                 i_size_write(inode, off+len-towrite);
2891                 EXT4_I(inode)->i_disksize = inode->i_size;
2892         }
2893         inode->i_version++;
2894         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2895         ext4_mark_inode_dirty(handle, inode);
2896         mutex_unlock(&inode->i_mutex);
2897         return len - towrite;
2898 }
2899
2900 #endif
2901
2902 static int ext4_get_sb(struct file_system_type *fs_type,
2903         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2904 {
2905         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
2906 }
2907
2908 static struct file_system_type ext4dev_fs_type = {
2909         .owner          = THIS_MODULE,
2910         .name           = "ext4dev",
2911         .get_sb         = ext4_get_sb,
2912         .kill_sb        = kill_block_super,
2913         .fs_flags       = FS_REQUIRES_DEV,
2914 };
2915
2916 static int __init init_ext4_fs(void)
2917 {
2918         int err = init_ext4_xattr();
2919         if (err)
2920                 return err;
2921         err = init_inodecache();
2922         if (err)
2923                 goto out1;
2924         err = register_filesystem(&ext4dev_fs_type);
2925         if (err)
2926                 goto out;
2927         return 0;
2928 out:
2929         destroy_inodecache();
2930 out1:
2931         exit_ext4_xattr();
2932         return err;
2933 }
2934
2935 static void __exit exit_ext4_fs(void)
2936 {
2937         unregister_filesystem(&ext4dev_fs_type);
2938         destroy_inodecache();
2939         exit_ext4_xattr();
2940 }
2941
2942 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2943 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
2944 MODULE_LICENSE("GPL");
2945 module_init(init_ext4_fs)
2946 module_exit(exit_ext4_fs)