err.no Git - linux-2.6/blob - fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/log2.h>
  38 #include <linux/crc16.h>
  39 #include <asm/uaccess.h>
  40
  41 #include "ext4.h"
  42 #include "ext4_jbd2.h"
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46 #include "group.h"
  47
  48 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  49                              unsigned long journal_devnum);
  50 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  51                                unsigned int);
  52 static void ext4_commit_super (struct super_block * sb,
  53                                struct ext4_super_block * es,
  54                                int sync);
  55 static void ext4_mark_recovery_complete(struct super_block * sb,
  56                                         struct ext4_super_block * es);
  57 static void ext4_clear_journal_err(struct super_block * sb,
  58                                    struct ext4_super_block * es);
  59 static int ext4_sync_fs(struct super_block *sb, int wait);
  60 static const char *ext4_decode_error(struct super_block * sb, int errno,
  61                                      char nbuf[16]);
  62 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  63 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  64 static void ext4_unlockfs(struct super_block *sb);
  65 static void ext4_write_super (struct super_block * sb);
  66 static void ext4_write_super_lockfs(struct super_block *sb);
  67
  68
  69 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  70                                struct ext4_group_desc *bg)
  71 {
  72         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  73                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  74                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  75 }
  76
  77 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  78                                struct ext4_group_desc *bg)
  79 {
  80         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  81                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  82                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  83 }
  84
  85 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  86                               struct ext4_group_desc *bg)
  87 {
  88         return le32_to_cpu(bg->bg_inode_table_lo) |
  89                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  90                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  91 }
  92
  93 void ext4_block_bitmap_set(struct super_block *sb,
  94                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  95 {
  96         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  97         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  98                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  99 }
 100
 101 void ext4_inode_bitmap_set(struct super_block *sb,
 102                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 103 {
 104         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 105         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 106                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 107 }
 108
 109 void ext4_inode_table_set(struct super_block *sb,
 110                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 111 {
 112         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 113         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 114                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 115 }
 116
 117 /*
 118  * Wrappers for jbd2_journal_start/end.
 119  *
 120  * The only special thing we need to do here is to make sure that all
 121  * journal_end calls result in the superblock being marked dirty, so
 122  * that sync() will call the filesystem's write_super callback if
 123  * appropriate.
 124  */
 125 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 126 {
 127         journal_t *journal;
 128
 129         if (sb->s_flags & MS_RDONLY)
 130                 return ERR_PTR(-EROFS);
 131
 132         /* Special case here: if the journal has aborted behind our
 133          * backs (eg. EIO in the commit thread), then we still need to
 134          * take the FS itself readonly cleanly. */
 135         journal = EXT4_SB(sb)->s_journal;
 136         if (is_journal_aborted(journal)) {
 137                 ext4_abort(sb, __func__,
 138                            "Detected aborted journal");
 139                 return ERR_PTR(-EROFS);
 140         }
 141
 142         return jbd2_journal_start(journal, nblocks);
 143 }
 144
 145 /*
 146  * The only special thing we need to do here is to make sure that all
 147  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 148  * that sync() will call the filesystem's write_super callback if
 149  * appropriate.
 150  */
 151 int __ext4_journal_stop(const char *where, handle_t *handle)
 152 {
 153         struct super_block *sb;
 154         int err;
 155         int rc;
 156
 157         sb = handle->h_transaction->t_journal->j_private;
 158         err = handle->h_err;
 159         rc = jbd2_journal_stop(handle);
 160
 161         if (!err)
 162                 err = rc;
 163         if (err)
 164                 __ext4_std_error(sb, where, err);
 165         return err;
 166 }
 167
 168 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 169                 struct buffer_head *bh, handle_t *handle, int err)
 170 {
 171         char nbuf[16];
 172         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 173
 174         if (bh)
 175                 BUFFER_TRACE(bh, "abort");
 176
 177         if (!handle->h_err)
 178                 handle->h_err = err;
 179
 180         if (is_handle_aborted(handle))
 181                 return;
 182
 183         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 184                caller, errstr, err_fn);
 185
 186         jbd2_journal_abort_handle(handle);
 187 }
 188
 189 /* Deal with the reporting of failure conditions on a filesystem such as
 190  * inconsistencies detected or read IO failures.
 191  *
 192  * On ext2, we can store the error state of the filesystem in the
 193  * superblock.  That is not possible on ext4, because we may have other
 194  * write ordering constraints on the superblock which prevent us from
 195  * writing it out straight away; and given that the journal is about to
 196  * be aborted, we can't rely on the current, or future, transactions to
 197  * write out the superblock safely.
 198  *
 199  * We'll just use the jbd2_journal_abort() error code to record an error in
 200  * the journal instead.  On recovery, the journal will compain about
 201  * that error until we've noted it down and cleared it.
 202  */
 203
 204 static void ext4_handle_error(struct super_block *sb)
 205 {
 206         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 207
 208         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 209         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 210
 211         if (sb->s_flags & MS_RDONLY)
 212                 return;
 213
 214         if (!test_opt (sb, ERRORS_CONT)) {
 215                 journal_t *journal = EXT4_SB(sb)->s_journal;
 216
 217                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 218                 if (journal)
 219                         jbd2_journal_abort(journal, -EIO);
 220         }
 221         if (test_opt (sb, ERRORS_RO)) {
 222                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 223                 sb->s_flags |= MS_RDONLY;
 224         }
 225         ext4_commit_super(sb, es, 1);
 226         if (test_opt(sb, ERRORS_PANIC))
 227                 panic("EXT4-fs (device %s): panic forced after error\n",
 228                         sb->s_id);
 229 }
 230
 231 void ext4_error (struct super_block * sb, const char * function,
 232                  const char * fmt, ...)
 233 {
 234         va_list args;
 235
 236         va_start(args, fmt);
 237         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 238         vprintk(fmt, args);
 239         printk("\n");
 240         va_end(args);
 241
 242         ext4_handle_error(sb);
 243 }
 244
 245 static const char *ext4_decode_error(struct super_block * sb, int errno,
 246                                      char nbuf[16])
 247 {
 248         char *errstr = NULL;
 249
 250         switch (errno) {
 251         case -EIO:
 252                 errstr = "IO failure";
 253                 break;
 254         case -ENOMEM:
 255                 errstr = "Out of memory";
 256                 break;
 257         case -EROFS:
 258                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 259                         errstr = "Journal has aborted";
 260                 else
 261                         errstr = "Readonly filesystem";
 262                 break;
 263         default:
 264                 /* If the caller passed in an extra buffer for unknown
 265                  * errors, textualise them now.  Else we just return
 266                  * NULL. */
 267                 if (nbuf) {
 268                         /* Check for truncated error codes... */
 269                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 270                                 errstr = nbuf;
 271                 }
 272                 break;
 273         }
 274
 275         return errstr;
 276 }
 277
 278 /* __ext4_std_error decodes expected errors from journaling functions
 279  * automatically and invokes the appropriate error response.  */
 280
 281 void __ext4_std_error (struct super_block * sb, const char * function,
 282                        int errno)
 283 {
 284         char nbuf[16];
 285         const char *errstr;
 286
 287         /* Special case: if the error is EROFS, and we're not already
 288          * inside a transaction, then there's really no point in logging
 289          * an error. */
 290         if (errno == -EROFS && journal_current_handle() == NULL &&
 291             (sb->s_flags & MS_RDONLY))
 292                 return;
 293
 294         errstr = ext4_decode_error(sb, errno, nbuf);
 295         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 296                 sb->s_id, function, errstr);
 297
 298         ext4_handle_error(sb);
 299 }
 300
 301 /*
 302  * ext4_abort is a much stronger failure handler than ext4_error.  The
 303  * abort function may be used to deal with unrecoverable failures such
 304  * as journal IO errors or ENOMEM at a critical moment in log management.
 305  *
 306  * We unconditionally force the filesystem into an ABORT|READONLY state,
 307  * unless the error response on the fs has been set to panic in which
 308  * case we take the easy way out and panic immediately.
 309  */
 310
 311 void ext4_abort (struct super_block * sb, const char * function,
 312                  const char * fmt, ...)
 313 {
 314         va_list args;
 315
 316         printk (KERN_CRIT "ext4_abort called.\n");
 317
 318         va_start(args, fmt);
 319         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 320         vprintk(fmt, args);
 321         printk("\n");
 322         va_end(args);
 323
 324         if (test_opt(sb, ERRORS_PANIC))
 325                 panic("EXT4-fs panic from previous error\n");
 326
 327         if (sb->s_flags & MS_RDONLY)
 328                 return;
 329
 330         printk(KERN_CRIT "Remounting filesystem read-only\n");
 331         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 332         sb->s_flags |= MS_RDONLY;
 333         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 334         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 335 }
 336
 337 void ext4_warning (struct super_block * sb, const char * function,
 338                    const char * fmt, ...)
 339 {
 340         va_list args;
 341
 342         va_start(args, fmt);
 343         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 344                sb->s_id, function);
 345         vprintk(fmt, args);
 346         printk("\n");
 347         va_end(args);
 348 }
 349
 350 void ext4_update_dynamic_rev(struct super_block *sb)
 351 {
 352         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 353
 354         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 355                 return;
 356
 357         ext4_warning(sb, __func__,
 358                      "updating to rev %d because of new feature flag, "
 359                      "running e2fsck is recommended",
 360                      EXT4_DYNAMIC_REV);
 361
 362         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 363         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 364         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 365         /* leave es->s_feature_*compat flags alone */
 366         /* es->s_uuid will be set by e2fsck if empty */
 367
 368         /*
 369          * The rest of the superblock fields should be zero, and if not it
 370          * means they are likely already in use, so leave them alone.  We
 371          * can leave it up to e2fsck to clean up any inconsistencies there.
 372          */
 373 }
 374
 375 int ext4_update_compat_feature(handle_t *handle,
 376                                         struct super_block *sb, __u32 compat)
 377 {
 378         int err = 0;
 379         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 380                 err = ext4_journal_get_write_access(handle,
 381                                 EXT4_SB(sb)->s_sbh);
 382                 if (err)
 383                         return err;
 384                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 385                 sb->s_dirt = 1;
 386                 handle->h_sync = 1;
 387                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 388                                         "call ext4_journal_dirty_met adata");
 389                 err = ext4_journal_dirty_metadata(handle,
 390                                 EXT4_SB(sb)->s_sbh);
 391         }
 392         return err;
 393 }
 394
 395 int ext4_update_rocompat_feature(handle_t *handle,
 396                                         struct super_block *sb, __u32 rocompat)
 397 {
 398         int err = 0;
 399         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 400                 err = ext4_journal_get_write_access(handle,
 401                                 EXT4_SB(sb)->s_sbh);
 402                 if (err)
 403                         return err;
 404                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 405                 sb->s_dirt = 1;
 406                 handle->h_sync = 1;
 407                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 408                                         "call ext4_journal_dirty_met adata");
 409                 err = ext4_journal_dirty_metadata(handle,
 410                                 EXT4_SB(sb)->s_sbh);
 411         }
 412         return err;
 413 }
 414
 415 int ext4_update_incompat_feature(handle_t *handle,
 416                                         struct super_block *sb, __u32 incompat)
 417 {
 418         int err = 0;
 419         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 420                 err = ext4_journal_get_write_access(handle,
 421                                 EXT4_SB(sb)->s_sbh);
 422                 if (err)
 423                         return err;
 424                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 425                 sb->s_dirt = 1;
 426                 handle->h_sync = 1;
 427                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 428                                         "call ext4_journal_dirty_met adata");
 429                 err = ext4_journal_dirty_metadata(handle,
 430                                 EXT4_SB(sb)->s_sbh);
 431         }
 432         return err;
 433 }
 434
 435 /*
 436  * Open the external journal device
 437  */
 438 static struct block_device *ext4_blkdev_get(dev_t dev)
 439 {
 440         struct block_device *bdev;
 441         char b[BDEVNAME_SIZE];
 442
 443         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 444         if (IS_ERR(bdev))
 445                 goto fail;
 446         return bdev;
 447
 448 fail:
 449         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 450                         __bdevname(dev, b), PTR_ERR(bdev));
 451         return NULL;
 452 }
 453
 454 /*
 455  * Release the journal device
 456  */
 457 static int ext4_blkdev_put(struct block_device *bdev)
 458 {
 459         bd_release(bdev);
 460         return blkdev_put(bdev);
 461 }
 462
 463 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 464 {
 465         struct block_device *bdev;
 466         int ret = -ENODEV;
 467
 468         bdev = sbi->journal_bdev;
 469         if (bdev) {
 470                 ret = ext4_blkdev_put(bdev);
 471                 sbi->journal_bdev = NULL;
 472         }
 473         return ret;
 474 }
 475
 476 static inline struct inode *orphan_list_entry(struct list_head *l)
 477 {
 478         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 479 }
 480
 481 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 482 {
 483         struct list_head *l;
 484
 485         printk(KERN_ERR "sb orphan head is %d\n",
 486                le32_to_cpu(sbi->s_es->s_last_orphan));
 487
 488         printk(KERN_ERR "sb_info orphan list:\n");
 489         list_for_each(l, &sbi->s_orphan) {
 490                 struct inode *inode = orphan_list_entry(l);
 491                 printk(KERN_ERR "  "
 492                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 493                        inode->i_sb->s_id, inode->i_ino, inode,
 494                        inode->i_mode, inode->i_nlink,
 495                        NEXT_ORPHAN(inode));
 496         }
 497 }
 498
 499 static void ext4_put_super (struct super_block * sb)
 500 {
 501         struct ext4_sb_info *sbi = EXT4_SB(sb);
 502         struct ext4_super_block *es = sbi->s_es;
 503         int i;
 504
 505         ext4_mb_release(sb);
 506         ext4_ext_release(sb);
 507         ext4_xattr_put_super(sb);
 508         jbd2_journal_destroy(sbi->s_journal);
 509         sbi->s_journal = NULL;
 510         if (!(sb->s_flags & MS_RDONLY)) {
 511                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 512                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 513                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 514                 mark_buffer_dirty(sbi->s_sbh);
 515                 ext4_commit_super(sb, es, 1);
 516         }
 517
 518         for (i = 0; i < sbi->s_gdb_count; i++)
 519                 brelse(sbi->s_group_desc[i]);
 520         kfree(sbi->s_group_desc);
 521         kfree(sbi->s_flex_groups);
 522         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 523         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 524         percpu_counter_destroy(&sbi->s_dirs_counter);
 525         brelse(sbi->s_sbh);
 526 #ifdef CONFIG_QUOTA
 527         for (i = 0; i < MAXQUOTAS; i++)
 528                 kfree(sbi->s_qf_names[i]);
 529 #endif
 530
 531         /* Debugging code just in case the in-memory inode orphan list
 532          * isn't empty.  The on-disk one can be non-empty if we've
 533          * detected an error and taken the fs readonly, but the
 534          * in-memory list had better be clean by this point. */
 535         if (!list_empty(&sbi->s_orphan))
 536                 dump_orphan_list(sb, sbi);
 537         J_ASSERT(list_empty(&sbi->s_orphan));
 538
 539         invalidate_bdev(sb->s_bdev);
 540         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 541                 /*
 542                  * Invalidate the journal device's buffers.  We don't want them
 543                  * floating about in memory - the physical journal device may
 544                  * hotswapped, and it breaks the `ro-after' testing code.
 545                  */
 546                 sync_blockdev(sbi->journal_bdev);
 547                 invalidate_bdev(sbi->journal_bdev);
 548                 ext4_blkdev_remove(sbi);
 549         }
 550         sb->s_fs_info = NULL;
 551         kfree(sbi);
 552         return;
 553 }
 554
 555 static struct kmem_cache *ext4_inode_cachep;
 556
 557 /*
 558  * Called inside transaction, so use GFP_NOFS
 559  */
 560 static struct inode *ext4_alloc_inode(struct super_block *sb)
 561 {
 562         struct ext4_inode_info *ei;
 563
 564         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 565         if (!ei)
 566                 return NULL;
 567 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 568         ei->i_acl = EXT4_ACL_NOT_CACHED;
 569         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 570 #endif
 571         ei->i_block_alloc_info = NULL;
 572         ei->vfs_inode.i_version = 1;
 573         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 574         INIT_LIST_HEAD(&ei->i_prealloc_list);
 575         spin_lock_init(&ei->i_prealloc_lock);
 576         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 577         return &ei->vfs_inode;
 578 }
 579
 580 static void ext4_destroy_inode(struct inode *inode)
 581 {
 582         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 583                 printk("EXT4 Inode %p: orphan list check failed!\n",
 584                         EXT4_I(inode));
 585                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 586                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 587                                 true);
 588                 dump_stack();
 589         }
 590         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 591 }
 592
 593 static void init_once(struct kmem_cache *cachep, void *foo)
 594 {
 595         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 596
 597         INIT_LIST_HEAD(&ei->i_orphan);
 598 #ifdef CONFIG_EXT4DEV_FS_XATTR
 599         init_rwsem(&ei->xattr_sem);
 600 #endif
 601         init_rwsem(&ei->i_data_sem);
 602         inode_init_once(&ei->vfs_inode);
 603 }
 604
 605 static int init_inodecache(void)
 606 {
 607         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 608                                              sizeof(struct ext4_inode_info),
 609                                              0, (SLAB_RECLAIM_ACCOUNT|
 610                                                 SLAB_MEM_SPREAD),
 611                                              init_once);
 612         if (ext4_inode_cachep == NULL)
 613                 return -ENOMEM;
 614         return 0;
 615 }
 616
 617 static void destroy_inodecache(void)
 618 {
 619         kmem_cache_destroy(ext4_inode_cachep);
 620 }
 621
 622 static void ext4_clear_inode(struct inode *inode)
 623 {
 624         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 625 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 626         if (EXT4_I(inode)->i_acl &&
 627                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 628                 posix_acl_release(EXT4_I(inode)->i_acl);
 629                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 630         }
 631         if (EXT4_I(inode)->i_default_acl &&
 632                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 633                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 634                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 635         }
 636 #endif
 637         ext4_discard_reservation(inode);
 638         EXT4_I(inode)->i_block_alloc_info = NULL;
 639         if (unlikely(rsv))
 640                 kfree(rsv);
 641         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 642                                        &EXT4_I(inode)->jinode);
 643 }
 644
 645 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 646 {
 647 #if defined(CONFIG_QUOTA)
 648         struct ext4_sb_info *sbi = EXT4_SB(sb);
 649
 650         if (sbi->s_jquota_fmt)
 651                 seq_printf(seq, ",jqfmt=%s",
 652                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 653
 654         if (sbi->s_qf_names[USRQUOTA])
 655                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 656
 657         if (sbi->s_qf_names[GRPQUOTA])
 658                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 659
 660         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 661                 seq_puts(seq, ",usrquota");
 662
 663         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 664                 seq_puts(seq, ",grpquota");
 665 #endif
 666 }
 667
 668 /*
 669  * Show an option if
 670  *  - it's set to a non-default value OR
 671  *  - if the per-sb default is different from the global default
 672  */
 673 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 674 {
 675         int def_errors;
 676         unsigned long def_mount_opts;
 677         struct super_block *sb = vfs->mnt_sb;
 678         struct ext4_sb_info *sbi = EXT4_SB(sb);
 679         struct ext4_super_block *es = sbi->s_es;
 680
 681         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 682         def_errors     = le16_to_cpu(es->s_errors);
 683
 684         if (sbi->s_sb_block != 1)
 685                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 686         if (test_opt(sb, MINIX_DF))
 687                 seq_puts(seq, ",minixdf");
 688         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 689                 seq_puts(seq, ",grpid");
 690         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 691                 seq_puts(seq, ",nogrpid");
 692         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 693             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 694                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 695         }
 696         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 697             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 698                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 699         }
 700         if (test_opt(sb, ERRORS_RO)) {
 701                 if (def_errors == EXT4_ERRORS_PANIC ||
 702                     def_errors == EXT4_ERRORS_CONTINUE) {
 703                         seq_puts(seq, ",errors=remount-ro");
 704                 }
 705         }
 706         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 707                 seq_puts(seq, ",errors=continue");
 708         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 709                 seq_puts(seq, ",errors=panic");
 710         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 711                 seq_puts(seq, ",nouid32");
 712         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 713                 seq_puts(seq, ",debug");
 714         if (test_opt(sb, OLDALLOC))
 715                 seq_puts(seq, ",oldalloc");
 716 #ifdef CONFIG_EXT4DEV_FS_XATTR
 717         if (test_opt(sb, XATTR_USER) &&
 718                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 719                 seq_puts(seq, ",user_xattr");
 720         if (!test_opt(sb, XATTR_USER) &&
 721             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 722                 seq_puts(seq, ",nouser_xattr");
 723         }
 724 #endif
 725 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 726         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 727                 seq_puts(seq, ",acl");
 728         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 729                 seq_puts(seq, ",noacl");
 730 #endif
 731         if (!test_opt(sb, RESERVATION))
 732                 seq_puts(seq, ",noreservation");
 733         if (sbi->s_commit_interval) {
 734                 seq_printf(seq, ",commit=%u",
 735                            (unsigned) (sbi->s_commit_interval / HZ));
 736         }
 737         /*
 738          * We're changing the default of barrier mount option, so
 739          * let's always display its mount state so it's clear what its
 740          * status is.
 741          */
 742         seq_puts(seq, ",barrier=");
 743         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 744         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 745                 seq_puts(seq, ",journal_async_commit");
 746         if (test_opt(sb, NOBH))
 747                 seq_puts(seq, ",nobh");
 748         if (!test_opt(sb, EXTENTS))
 749                 seq_puts(seq, ",noextents");
 750         if (!test_opt(sb, MBALLOC))
 751                 seq_puts(seq, ",nomballoc");
 752         if (test_opt(sb, I_VERSION))
 753                 seq_puts(seq, ",i_version");
 754
 755         if (sbi->s_stripe)
 756                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 757         /*
 758          * journal mode get enabled in different ways
 759          * So just print the value even if we didn't specify it
 760          */
 761         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 762                 seq_puts(seq, ",data=journal");
 763         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 764                 seq_puts(seq, ",data=ordered");
 765         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 766                 seq_puts(seq, ",data=writeback");
 767
 768         ext4_show_quota_options(seq, sb);
 769         return 0;
 770 }
 771
 772
 773 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 774                 u64 ino, u32 generation)
 775 {
 776         struct inode *inode;
 777
 778         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 779                 return ERR_PTR(-ESTALE);
 780         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 781                 return ERR_PTR(-ESTALE);
 782
 783         /* iget isn't really right if the inode is currently unallocated!!
 784          *
 785          * ext4_read_inode will return a bad_inode if the inode had been
 786          * deleted, so we should be safe.
 787          *
 788          * Currently we don't know the generation for parent directory, so
 789          * a generation of 0 means "accept any"
 790          */
 791         inode = ext4_iget(sb, ino);
 792         if (IS_ERR(inode))
 793                 return ERR_CAST(inode);
 794         if (generation && inode->i_generation != generation) {
 795                 iput(inode);
 796                 return ERR_PTR(-ESTALE);
 797         }
 798
 799         return inode;
 800 }
 801
 802 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 803                 int fh_len, int fh_type)
 804 {
 805         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 806                                     ext4_nfs_get_inode);
 807 }
 808
 809 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 810                 int fh_len, int fh_type)
 811 {
 812         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 813                                     ext4_nfs_get_inode);
 814 }
 815
 816 #ifdef CONFIG_QUOTA
 817 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 818 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 819
 820 static int ext4_dquot_initialize(struct inode *inode, int type);
 821 static int ext4_dquot_drop(struct inode *inode);
 822 static int ext4_write_dquot(struct dquot *dquot);
 823 static int ext4_acquire_dquot(struct dquot *dquot);
 824 static int ext4_release_dquot(struct dquot *dquot);
 825 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 826 static int ext4_write_info(struct super_block *sb, int type);
 827 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 828                                 char *path, int remount);
 829 static int ext4_quota_on_mount(struct super_block *sb, int type);
 830 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 831                                size_t len, loff_t off);
 832 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 833                                 const char *data, size_t len, loff_t off);
 834
 835 static struct dquot_operations ext4_quota_operations = {
 836         .initialize     = ext4_dquot_initialize,
 837         .drop           = ext4_dquot_drop,
 838         .alloc_space    = dquot_alloc_space,
 839         .alloc_inode    = dquot_alloc_inode,
 840         .free_space     = dquot_free_space,
 841         .free_inode     = dquot_free_inode,
 842         .transfer       = dquot_transfer,
 843         .write_dquot    = ext4_write_dquot,
 844         .acquire_dquot  = ext4_acquire_dquot,
 845         .release_dquot  = ext4_release_dquot,
 846         .mark_dirty     = ext4_mark_dquot_dirty,
 847         .write_info     = ext4_write_info
 848 };
 849
 850 static struct quotactl_ops ext4_qctl_operations = {
 851         .quota_on       = ext4_quota_on,
 852         .quota_off      = vfs_quota_off,
 853         .quota_sync     = vfs_quota_sync,
 854         .get_info       = vfs_get_dqinfo,
 855         .set_info       = vfs_set_dqinfo,
 856         .get_dqblk      = vfs_get_dqblk,
 857         .set_dqblk      = vfs_set_dqblk
 858 };
 859 #endif
 860
 861 static const struct super_operations ext4_sops = {
 862         .alloc_inode    = ext4_alloc_inode,
 863         .destroy_inode  = ext4_destroy_inode,
 864         .write_inode    = ext4_write_inode,
 865         .dirty_inode    = ext4_dirty_inode,
 866         .delete_inode   = ext4_delete_inode,
 867         .put_super      = ext4_put_super,
 868         .write_super    = ext4_write_super,
 869         .sync_fs        = ext4_sync_fs,
 870         .write_super_lockfs = ext4_write_super_lockfs,
 871         .unlockfs       = ext4_unlockfs,
 872         .statfs         = ext4_statfs,
 873         .remount_fs     = ext4_remount,
 874         .clear_inode    = ext4_clear_inode,
 875         .show_options   = ext4_show_options,
 876 #ifdef CONFIG_QUOTA
 877         .quota_read     = ext4_quota_read,
 878         .quota_write    = ext4_quota_write,
 879 #endif
 880 };
 881
 882 static const struct export_operations ext4_export_ops = {
 883         .fh_to_dentry = ext4_fh_to_dentry,
 884         .fh_to_parent = ext4_fh_to_parent,
 885         .get_parent = ext4_get_parent,
 886 };
 887
 888 enum {
 889         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 890         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 891         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 892         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 893         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 894         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 895         Opt_journal_checksum, Opt_journal_async_commit,
 896         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 897         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 898         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 899         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 900         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 901         Opt_mballoc, Opt_nomballoc, Opt_stripe,
 902 };
 903
 904 static match_table_t tokens = {
 905         {Opt_bsd_df, "bsddf"},
 906         {Opt_minix_df, "minixdf"},
 907         {Opt_grpid, "grpid"},
 908         {Opt_grpid, "bsdgroups"},
 909         {Opt_nogrpid, "nogrpid"},
 910         {Opt_nogrpid, "sysvgroups"},
 911         {Opt_resgid, "resgid=%u"},
 912         {Opt_resuid, "resuid=%u"},
 913         {Opt_sb, "sb=%u"},
 914         {Opt_err_cont, "errors=continue"},
 915         {Opt_err_panic, "errors=panic"},
 916         {Opt_err_ro, "errors=remount-ro"},
 917         {Opt_nouid32, "nouid32"},
 918         {Opt_nocheck, "nocheck"},
 919         {Opt_nocheck, "check=none"},
 920         {Opt_debug, "debug"},
 921         {Opt_oldalloc, "oldalloc"},
 922         {Opt_orlov, "orlov"},
 923         {Opt_user_xattr, "user_xattr"},
 924         {Opt_nouser_xattr, "nouser_xattr"},
 925         {Opt_acl, "acl"},
 926         {Opt_noacl, "noacl"},
 927         {Opt_reservation, "reservation"},
 928         {Opt_noreservation, "noreservation"},
 929         {Opt_noload, "noload"},
 930         {Opt_nobh, "nobh"},
 931         {Opt_bh, "bh"},
 932         {Opt_commit, "commit=%u"},
 933         {Opt_journal_update, "journal=update"},
 934         {Opt_journal_inum, "journal=%u"},
 935         {Opt_journal_dev, "journal_dev=%u"},
 936         {Opt_journal_checksum, "journal_checksum"},
 937         {Opt_journal_async_commit, "journal_async_commit"},
 938         {Opt_abort, "abort"},
 939         {Opt_data_journal, "data=journal"},
 940         {Opt_data_ordered, "data=ordered"},
 941         {Opt_data_writeback, "data=writeback"},
 942         {Opt_offusrjquota, "usrjquota="},
 943         {Opt_usrjquota, "usrjquota=%s"},
 944         {Opt_offgrpjquota, "grpjquota="},
 945         {Opt_grpjquota, "grpjquota=%s"},
 946         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 947         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 948         {Opt_grpquota, "grpquota"},
 949         {Opt_noquota, "noquota"},
 950         {Opt_quota, "quota"},
 951         {Opt_usrquota, "usrquota"},
 952         {Opt_barrier, "barrier=%u"},
 953         {Opt_extents, "extents"},
 954         {Opt_noextents, "noextents"},
 955         {Opt_i_version, "i_version"},
 956         {Opt_mballoc, "mballoc"},
 957         {Opt_nomballoc, "nomballoc"},
 958         {Opt_stripe, "stripe=%u"},
 959         {Opt_resize, "resize"},
 960         {Opt_err, NULL},
 961 };
 962
 963 static ext4_fsblk_t get_sb_block(void **data)
 964 {
 965         ext4_fsblk_t    sb_block;
 966         char            *options = (char *) *data;
 967
 968         if (!options || strncmp(options, "sb=", 3) != 0)
 969                 return 1;       /* Default location */
 970         options += 3;
 971         /*todo: use simple_strtoll with >32bit ext4 */
 972         sb_block = simple_strtoul(options, &options, 0);
 973         if (*options && *options != ',') {
 974                 printk("EXT4-fs: Invalid sb specification: %s\n",
 975                        (char *) *data);
 976                 return 1;
 977         }
 978         if (*options == ',')
 979                 options++;
 980         *data = (void *) options;
 981         return sb_block;
 982 }
 983
 984 static int parse_options (char *options, struct super_block *sb,
 985                           unsigned int *inum, unsigned long *journal_devnum,
 986                           ext4_fsblk_t *n_blocks_count, int is_remount)
 987 {
 988         struct ext4_sb_info *sbi = EXT4_SB(sb);
 989         char * p;
 990         substring_t args[MAX_OPT_ARGS];
 991         int data_opt = 0;
 992         int option;
 993 #ifdef CONFIG_QUOTA
 994         int qtype, qfmt;
 995         char *qname;
 996 #endif
 997
 998         if (!options)
 999                 return 1;
1000
1001         while ((p = strsep (&options, ",")) != NULL) {
1002                 int token;
1003                 if (!*p)
1004                         continue;
1005
1006                 token = match_token(p, tokens, args);
1007                 switch (token) {
1008                 case Opt_bsd_df:
1009                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1010                         break;
1011                 case Opt_minix_df:
1012                         set_opt (sbi->s_mount_opt, MINIX_DF);
1013                         break;
1014                 case Opt_grpid:
1015                         set_opt (sbi->s_mount_opt, GRPID);
1016                         break;
1017                 case Opt_nogrpid:
1018                         clear_opt (sbi->s_mount_opt, GRPID);
1019                         break;
1020                 case Opt_resuid:
1021                         if (match_int(&args[0], &option))
1022                                 return 0;
1023                         sbi->s_resuid = option;
1024                         break;
1025                 case Opt_resgid:
1026                         if (match_int(&args[0], &option))
1027                                 return 0;
1028                         sbi->s_resgid = option;
1029                         break;
1030                 case Opt_sb:
1031                         /* handled by get_sb_block() instead of here */
1032                         /* *sb_block = match_int(&args[0]); */
1033                         break;
1034                 case Opt_err_panic:
1035                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1036                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1037                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1038                         break;
1039                 case Opt_err_ro:
1040                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1041                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1042                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1043                         break;
1044                 case Opt_err_cont:
1045                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1046                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1047                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1048                         break;
1049                 case Opt_nouid32:
1050                         set_opt (sbi->s_mount_opt, NO_UID32);
1051                         break;
1052                 case Opt_nocheck:
1053                         clear_opt (sbi->s_mount_opt, CHECK);
1054                         break;
1055                 case Opt_debug:
1056                         set_opt (sbi->s_mount_opt, DEBUG);
1057                         break;
1058                 case Opt_oldalloc:
1059                         set_opt (sbi->s_mount_opt, OLDALLOC);
1060                         break;
1061                 case Opt_orlov:
1062                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1063                         break;
1064 #ifdef CONFIG_EXT4DEV_FS_XATTR
1065                 case Opt_user_xattr:
1066                         set_opt (sbi->s_mount_opt, XATTR_USER);
1067                         break;
1068                 case Opt_nouser_xattr:
1069                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1070                         break;
1071 #else
1072                 case Opt_user_xattr:
1073                 case Opt_nouser_xattr:
1074                         printk("EXT4 (no)user_xattr options not supported\n");
1075                         break;
1076 #endif
1077 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1078                 case Opt_acl:
1079                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1080                         break;
1081                 case Opt_noacl:
1082                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1083                         break;
1084 #else
1085                 case Opt_acl:
1086                 case Opt_noacl:
1087                         printk("EXT4 (no)acl options not supported\n");
1088                         break;
1089 #endif
1090                 case Opt_reservation:
1091                         set_opt(sbi->s_mount_opt, RESERVATION);
1092                         break;
1093                 case Opt_noreservation:
1094                         clear_opt(sbi->s_mount_opt, RESERVATION);
1095                         break;
1096                 case Opt_journal_update:
1097                         /* @@@ FIXME */
1098                         /* Eventually we will want to be able to create
1099                            a journal file here.  For now, only allow the
1100                            user to specify an existing inode to be the
1101                            journal file. */
1102                         if (is_remount) {
1103                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1104                                        "journal on remount\n");
1105                                 return 0;
1106                         }
1107                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1108                         break;
1109                 case Opt_journal_inum:
1110                         if (is_remount) {
1111                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1112                                        "journal on remount\n");
1113                                 return 0;
1114                         }
1115                         if (match_int(&args[0], &option))
1116                                 return 0;
1117                         *inum = option;
1118                         break;
1119                 case Opt_journal_dev:
1120                         if (is_remount) {
1121                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1122                                        "journal on remount\n");
1123                                 return 0;
1124                         }
1125                         if (match_int(&args[0], &option))
1126                                 return 0;
1127                         *journal_devnum = option;
1128                         break;
1129                 case Opt_journal_checksum:
1130                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1131                         break;
1132                 case Opt_journal_async_commit:
1133                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1134                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1135                         break;
1136                 case Opt_noload:
1137                         set_opt (sbi->s_mount_opt, NOLOAD);
1138                         break;
1139                 case Opt_commit:
1140                         if (match_int(&args[0], &option))
1141                                 return 0;
1142                         if (option < 0)
1143                                 return 0;
1144                         if (option == 0)
1145                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1146                         sbi->s_commit_interval = HZ * option;
1147                         break;
1148                 case Opt_data_journal:
1149                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1150                         goto datacheck;
1151                 case Opt_data_ordered:
1152                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1153                         goto datacheck;
1154                 case Opt_data_writeback:
1155                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1156                 datacheck:
1157                         if (is_remount) {
1158                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1159                                                 != data_opt) {
1160                                         printk(KERN_ERR
1161                                                 "EXT4-fs: cannot change data "
1162                                                 "mode on remount\n");
1163                                         return 0;
1164                                 }
1165                         } else {
1166                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1167                                 sbi->s_mount_opt |= data_opt;
1168                         }
1169                         break;
1170 #ifdef CONFIG_QUOTA
1171                 case Opt_usrjquota:
1172                         qtype = USRQUOTA;
1173                         goto set_qf_name;
1174                 case Opt_grpjquota:
1175                         qtype = GRPQUOTA;
1176 set_qf_name:
1177                         if ((sb_any_quota_enabled(sb) ||
1178                              sb_any_quota_suspended(sb)) &&
1179                             !sbi->s_qf_names[qtype]) {
1180                                 printk(KERN_ERR
1181                                         "EXT4-fs: Cannot change journaled "
1182                                         "quota options when quota turned on.\n");
1183                                 return 0;
1184                         }
1185                         qname = match_strdup(&args[0]);
1186                         if (!qname) {
1187                                 printk(KERN_ERR
1188                                         "EXT4-fs: not enough memory for "
1189                                         "storing quotafile name.\n");
1190                                 return 0;
1191                         }
1192                         if (sbi->s_qf_names[qtype] &&
1193                             strcmp(sbi->s_qf_names[qtype], qname)) {
1194                                 printk(KERN_ERR
1195                                         "EXT4-fs: %s quota file already "
1196                                         "specified.\n", QTYPE2NAME(qtype));
1197                                 kfree(qname);
1198                                 return 0;
1199                         }
1200                         sbi->s_qf_names[qtype] = qname;
1201                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1202                                 printk(KERN_ERR
1203                                         "EXT4-fs: quotafile must be on "
1204                                         "filesystem root.\n");
1205                                 kfree(sbi->s_qf_names[qtype]);
1206                                 sbi->s_qf_names[qtype] = NULL;
1207                                 return 0;
1208                         }
1209                         set_opt(sbi->s_mount_opt, QUOTA);
1210                         break;
1211                 case Opt_offusrjquota:
1212                         qtype = USRQUOTA;
1213                         goto clear_qf_name;
1214                 case Opt_offgrpjquota:
1215                         qtype = GRPQUOTA;
1216 clear_qf_name:
1217                         if ((sb_any_quota_enabled(sb) ||
1218                              sb_any_quota_suspended(sb)) &&
1219                             sbi->s_qf_names[qtype]) {
1220                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1221                                         "journaled quota options when "
1222                                         "quota turned on.\n");
1223                                 return 0;
1224                         }
1225                         /*
1226                          * The space will be released later when all options
1227                          * are confirmed to be correct
1228                          */
1229                         sbi->s_qf_names[qtype] = NULL;
1230                         break;
1231                 case Opt_jqfmt_vfsold:
1232                         qfmt = QFMT_VFS_OLD;
1233                         goto set_qf_format;
1234                 case Opt_jqfmt_vfsv0:
1235                         qfmt = QFMT_VFS_V0;
1236 set_qf_format:
1237                         if ((sb_any_quota_enabled(sb) ||
1238                              sb_any_quota_suspended(sb)) &&
1239                             sbi->s_jquota_fmt != qfmt) {
1240                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1241                                         "journaled quota options when "
1242                                         "quota turned on.\n");
1243                                 return 0;
1244                         }
1245                         sbi->s_jquota_fmt = qfmt;
1246                         break;
1247                 case Opt_quota:
1248                 case Opt_usrquota:
1249                         set_opt(sbi->s_mount_opt, QUOTA);
1250                         set_opt(sbi->s_mount_opt, USRQUOTA);
1251                         break;
1252                 case Opt_grpquota:
1253                         set_opt(sbi->s_mount_opt, QUOTA);
1254                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1255                         break;
1256                 case Opt_noquota:
1257                         if (sb_any_quota_enabled(sb)) {
1258                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1259                                         "options when quota turned on.\n");
1260                                 return 0;
1261                         }
1262                         clear_opt(sbi->s_mount_opt, QUOTA);
1263                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1264                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1265                         break;
1266 #else
1267                 case Opt_quota:
1268                 case Opt_usrquota:
1269                 case Opt_grpquota:
1270                         printk(KERN_ERR
1271                                 "EXT4-fs: quota options not supported.\n");
1272                         break;
1273                 case Opt_usrjquota:
1274                 case Opt_grpjquota:
1275                 case Opt_offusrjquota:
1276                 case Opt_offgrpjquota:
1277                 case Opt_jqfmt_vfsold:
1278                 case Opt_jqfmt_vfsv0:
1279                         printk(KERN_ERR
1280                                 "EXT4-fs: journaled quota options not "
1281                                 "supported.\n");
1282                         break;
1283                 case Opt_noquota:
1284                         break;
1285 #endif
1286                 case Opt_abort:
1287                         set_opt(sbi->s_mount_opt, ABORT);
1288                         break;
1289                 case Opt_barrier:
1290                         if (match_int(&args[0], &option))
1291                                 return 0;
1292                         if (option)
1293                                 set_opt(sbi->s_mount_opt, BARRIER);
1294                         else
1295                                 clear_opt(sbi->s_mount_opt, BARRIER);
1296                         break;
1297                 case Opt_ignore:
1298                         break;
1299                 case Opt_resize:
1300                         if (!is_remount) {
1301                                 printk("EXT4-fs: resize option only available "
1302                                         "for remount\n");
1303                                 return 0;
1304                         }
1305                         if (match_int(&args[0], &option) != 0)
1306                                 return 0;
1307                         *n_blocks_count = option;
1308                         break;
1309                 case Opt_nobh:
1310                         set_opt(sbi->s_mount_opt, NOBH);
1311                         break;
1312                 case Opt_bh:
1313                         clear_opt(sbi->s_mount_opt, NOBH);
1314                         break;
1315                 case Opt_extents:
1316                         set_opt (sbi->s_mount_opt, EXTENTS);
1317                         break;
1318                 case Opt_noextents:
1319                         clear_opt (sbi->s_mount_opt, EXTENTS);
1320                         break;
1321                 case Opt_i_version:
1322                         set_opt(sbi->s_mount_opt, I_VERSION);
1323                         sb->s_flags |= MS_I_VERSION;
1324                         break;
1325                 case Opt_mballoc:
1326                         set_opt(sbi->s_mount_opt, MBALLOC);
1327                         break;
1328                 case Opt_nomballoc:
1329                         clear_opt(sbi->s_mount_opt, MBALLOC);
1330                         break;
1331                 case Opt_stripe:
1332                         if (match_int(&args[0], &option))
1333                                 return 0;
1334                         if (option < 0)
1335                                 return 0;
1336                         sbi->s_stripe = option;
1337                         break;
1338                 default:
1339                         printk (KERN_ERR
1340                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1341                                 "or missing value\n", p);
1342                         return 0;
1343                 }
1344         }
1345 #ifdef CONFIG_QUOTA
1346         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1347                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1348                      sbi->s_qf_names[USRQUOTA])
1349                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1350
1351                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1352                      sbi->s_qf_names[GRPQUOTA])
1353                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1354
1355                 if ((sbi->s_qf_names[USRQUOTA] &&
1356                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1357                     (sbi->s_qf_names[GRPQUOTA] &&
1358                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1359                         printk(KERN_ERR "EXT4-fs: old and new quota "
1360                                         "format mixing.\n");
1361                         return 0;
1362                 }
1363
1364                 if (!sbi->s_jquota_fmt) {
1365                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1366                                         "not specified.\n");
1367                         return 0;
1368                 }
1369         } else {
1370                 if (sbi->s_jquota_fmt) {
1371                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1372                                         "specified with no journaling "
1373                                         "enabled.\n");
1374                         return 0;
1375                 }
1376         }
1377 #endif
1378         return 1;
1379 }
1380
1381 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1382                             int read_only)
1383 {
1384         struct ext4_sb_info *sbi = EXT4_SB(sb);
1385         int res = 0;
1386
1387         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1388                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1389                         "forcing read-only mode\n");
1390                 res = MS_RDONLY;
1391         }
1392         if (read_only)
1393                 return res;
1394         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1395                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1396                         "running e2fsck is recommended\n");
1397         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1398                 printk (KERN_WARNING
1399                         "EXT4-fs warning: mounting fs with errors, "
1400                         "running e2fsck is recommended\n");
1401         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1402                  le16_to_cpu(es->s_mnt_count) >=
1403                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1404                 printk (KERN_WARNING
1405                         "EXT4-fs warning: maximal mount count reached, "
1406                         "running e2fsck is recommended\n");
1407         else if (le32_to_cpu(es->s_checkinterval) &&
1408                 (le32_to_cpu(es->s_lastcheck) +
1409                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1410                 printk (KERN_WARNING
1411                         "EXT4-fs warning: checktime reached, "
1412                         "running e2fsck is recommended\n");
1413 #if 0
1414                 /* @@@ We _will_ want to clear the valid bit if we find
1415                  * inconsistencies, to force a fsck at reboot.  But for
1416                  * a plain journaled filesystem we can keep it set as
1417                  * valid forever! :)
1418                  */
1419         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1420 #endif
1421         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1422                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1423         le16_add_cpu(&es->s_mnt_count, 1);
1424         es->s_mtime = cpu_to_le32(get_seconds());
1425         ext4_update_dynamic_rev(sb);
1426         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1427
1428         ext4_commit_super(sb, es, 1);
1429         if (test_opt(sb, DEBUG))
1430                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1431                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1432                         sb->s_blocksize,
1433                         sbi->s_groups_count,
1434                         EXT4_BLOCKS_PER_GROUP(sb),
1435                         EXT4_INODES_PER_GROUP(sb),
1436                         sbi->s_mount_opt);
1437
1438         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1439         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1440                 char b[BDEVNAME_SIZE];
1441
1442                 printk("external journal on %s\n",
1443                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1444         } else {
1445                 printk("internal journal\n");
1446         }
1447         return res;
1448 }
1449
1450 static int ext4_fill_flex_info(struct super_block *sb)
1451 {
1452         struct ext4_sb_info *sbi = EXT4_SB(sb);
1453         struct ext4_group_desc *gdp = NULL;
1454         struct buffer_head *bh;
1455         ext4_group_t flex_group_count;
1456         ext4_group_t flex_group;
1457         int groups_per_flex = 0;
1458         __u64 block_bitmap = 0;
1459         int i;
1460
1461         if (!sbi->s_es->s_log_groups_per_flex) {
1462                 sbi->s_log_groups_per_flex = 0;
1463                 return 1;
1464         }
1465
1466         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1467         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1468
1469         flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
1470                 groups_per_flex;
1471         sbi->s_flex_groups = kmalloc(flex_group_count *
1472                                      sizeof(struct flex_groups), GFP_KERNEL);
1473         if (sbi->s_flex_groups == NULL) {
1474                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
1475                 goto failed;
1476         }
1477         memset(sbi->s_flex_groups, 0, flex_group_count *
1478                sizeof(struct flex_groups));
1479
1480         gdp = ext4_get_group_desc(sb, 1, &bh);
1481         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1482
1483         for (i = 0; i < sbi->s_groups_count; i++) {
1484                 gdp = ext4_get_group_desc(sb, i, &bh);
1485
1486                 flex_group = ext4_flex_group(sbi, i);
1487                 sbi->s_flex_groups[flex_group].free_inodes +=
1488                         le16_to_cpu(gdp->bg_free_inodes_count);
1489                 sbi->s_flex_groups[flex_group].free_blocks +=
1490                         le16_to_cpu(gdp->bg_free_blocks_count);
1491         }
1492
1493         return 1;
1494 failed:
1495         return 0;
1496 }
1497
1498 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1499                             struct ext4_group_desc *gdp)
1500 {
1501         __u16 crc = 0;
1502
1503         if (sbi->s_es->s_feature_ro_compat &
1504             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1505                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1506                 __le32 le_group = cpu_to_le32(block_group);
1507
1508                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1509                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1510                 crc = crc16(crc, (__u8 *)gdp, offset);
1511                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1512                 /* for checksum of struct ext4_group_desc do the rest...*/
1513                 if ((sbi->s_es->s_feature_incompat &
1514                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1515                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1516                         crc = crc16(crc, (__u8 *)gdp + offset,
1517                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1518                                         offset);
1519         }
1520
1521         return cpu_to_le16(crc);
1522 }
1523
1524 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1525                                 struct ext4_group_desc *gdp)
1526 {
1527         if ((sbi->s_es->s_feature_ro_compat &
1528              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1529             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1530                 return 0;
1531
1532         return 1;
1533 }
1534
1535 /* Called at mount-time, super-block is locked */
1536 static int ext4_check_descriptors(struct super_block *sb)
1537 {
1538         struct ext4_sb_info *sbi = EXT4_SB(sb);
1539         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1540         ext4_fsblk_t last_block;
1541         ext4_fsblk_t block_bitmap;
1542         ext4_fsblk_t inode_bitmap;
1543         ext4_fsblk_t inode_table;
1544         int flexbg_flag = 0;
1545         ext4_group_t i;
1546
1547         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1548                 flexbg_flag = 1;
1549
1550         ext4_debug ("Checking group descriptors");
1551
1552         for (i = 0; i < sbi->s_groups_count; i++) {
1553                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1554
1555                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1556                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1557                 else
1558                         last_block = first_block +
1559                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1560
1561                 block_bitmap = ext4_block_bitmap(sb, gdp);
1562                 if (block_bitmap < first_block || block_bitmap > last_block)
1563                 {
1564                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1565                                "Block bitmap for group %lu not in group "
1566                                "(block %llu)!", i, block_bitmap);
1567                         return 0;
1568                 }
1569                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1570                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1571                 {
1572                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1573                                "Inode bitmap for group %lu not in group "
1574                                "(block %llu)!", i, inode_bitmap);
1575                         return 0;
1576                 }
1577                 inode_table = ext4_inode_table(sb, gdp);
1578                 if (inode_table < first_block ||
1579                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1580                 {
1581                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1582                                "Inode table for group %lu not in group "
1583                                "(block %llu)!", i, inode_table);
1584                         return 0;
1585                 }
1586                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1587                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1588                                "Checksum for group %lu failed (%u!=%u)\n",
1589                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1590                                gdp)), le16_to_cpu(gdp->bg_checksum));
1591                         return 0;
1592                 }
1593                 if (!flexbg_flag)
1594                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1595         }
1596
1597         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1598         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1599         return 1;
1600 }
1601
1602 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1603  * the superblock) which were deleted from all directories, but held open by
1604  * a process at the time of a crash.  We walk the list and try to delete these
1605  * inodes at recovery time (only with a read-write filesystem).
1606  *
1607  * In order to keep the orphan inode chain consistent during traversal (in
1608  * case of crash during recovery), we link each inode into the superblock
1609  * orphan list_head and handle it the same way as an inode deletion during
1610  * normal operation (which journals the operations for us).
1611  *
1612  * We only do an iget() and an iput() on each inode, which is very safe if we
1613  * accidentally point at an in-use or already deleted inode.  The worst that
1614  * can happen in this case is that we get a "bit already cleared" message from
1615  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1616  * e2fsck was run on this filesystem, and it must have already done the orphan
1617  * inode cleanup for us, so we can safely abort without any further action.
1618  */
1619 static void ext4_orphan_cleanup (struct super_block * sb,
1620                                  struct ext4_super_block * es)
1621 {
1622         unsigned int s_flags = sb->s_flags;
1623         int nr_orphans = 0, nr_truncates = 0;
1624 #ifdef CONFIG_QUOTA
1625         int i;
1626 #endif
1627         if (!es->s_last_orphan) {
1628                 jbd_debug(4, "no orphan inodes to clean up\n");
1629                 return;
1630         }
1631
1632         if (bdev_read_only(sb->s_bdev)) {
1633                 printk(KERN_ERR "EXT4-fs: write access "
1634                         "unavailable, skipping orphan cleanup.\n");
1635                 return;
1636         }
1637
1638         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1639                 if (es->s_last_orphan)
1640                         jbd_debug(1, "Errors on filesystem, "
1641                                   "clearing orphan list.\n");
1642                 es->s_last_orphan = 0;
1643                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1644                 return;
1645         }
1646
1647         if (s_flags & MS_RDONLY) {
1648                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1649                        sb->s_id);
1650                 sb->s_flags &= ~MS_RDONLY;
1651         }
1652 #ifdef CONFIG_QUOTA
1653         /* Needed for iput() to work correctly and not trash data */
1654         sb->s_flags |= MS_ACTIVE;
1655         /* Turn on quotas so that they are updated correctly */
1656         for (i = 0; i < MAXQUOTAS; i++) {
1657                 if (EXT4_SB(sb)->s_qf_names[i]) {
1658                         int ret = ext4_quota_on_mount(sb, i);
1659                         if (ret < 0)
1660                                 printk(KERN_ERR
1661                                         "EXT4-fs: Cannot turn on journaled "
1662                                         "quota: error %d\n", ret);
1663                 }
1664         }
1665 #endif
1666
1667         while (es->s_last_orphan) {
1668                 struct inode *inode;
1669
1670                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1671                 if (IS_ERR(inode)) {
1672                         es->s_last_orphan = 0;
1673                         break;
1674                 }
1675
1676                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1677                 DQUOT_INIT(inode);
1678                 if (inode->i_nlink) {
1679                         printk(KERN_DEBUG
1680                                 "%s: truncating inode %lu to %Ld bytes\n",
1681                                 __func__, inode->i_ino, inode->i_size);
1682                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1683                                   inode->i_ino, inode->i_size);
1684                         ext4_truncate(inode);
1685                         nr_truncates++;
1686                 } else {
1687                         printk(KERN_DEBUG
1688                                 "%s: deleting unreferenced inode %lu\n",
1689                                 __func__, inode->i_ino);
1690                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1691                                   inode->i_ino);
1692                         nr_orphans++;
1693                 }
1694                 iput(inode);  /* The delete magic happens here! */
1695         }
1696
1697 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1698
1699         if (nr_orphans)
1700                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1701                        sb->s_id, PLURAL(nr_orphans));
1702         if (nr_truncates)
1703                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1704                        sb->s_id, PLURAL(nr_truncates));
1705 #ifdef CONFIG_QUOTA
1706         /* Turn quotas off */
1707         for (i = 0; i < MAXQUOTAS; i++) {
1708                 if (sb_dqopt(sb)->files[i])
1709                         vfs_quota_off(sb, i, 0);
1710         }
1711 #endif
1712         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1713 }
1714 /*
1715  * Maximal extent format file size.
1716  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1717  * extent format containers, within a sector_t, and within i_blocks
1718  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1719  * so that won't be a limiting factor.
1720  *
1721  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1722  */
1723 static loff_t ext4_max_size(int blkbits)
1724 {
1725         loff_t res;
1726         loff_t upper_limit = MAX_LFS_FILESIZE;
1727
1728         /* small i_blocks in vfs inode? */
1729         if (sizeof(blkcnt_t) < sizeof(u64)) {
1730                 /*
1731                  * CONFIG_LSF is not enabled implies the inode
1732                  * i_block represent total blocks in 512 bytes
1733                  * 32 == size of vfs inode i_blocks * 8
1734                  */
1735                 upper_limit = (1LL << 32) - 1;
1736
1737                 /* total blocks in file system block size */
1738                 upper_limit >>= (blkbits - 9);
1739                 upper_limit <<= blkbits;
1740         }
1741
1742         /* 32-bit extent-start container, ee_block */
1743         res = 1LL << 32;
1744         res <<= blkbits;
1745         res -= 1;
1746
1747         /* Sanity check against vm- & vfs- imposed limits */
1748         if (res > upper_limit)
1749                 res = upper_limit;
1750
1751         return res;
1752 }
1753
1754 /*
1755  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1756  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1757  * We need to be 1 filesystem block less than the 2^48 sector limit.
1758  */
1759 static loff_t ext4_max_bitmap_size(int bits)
1760 {
1761         loff_t res = EXT4_NDIR_BLOCKS;
1762         int meta_blocks;
1763         loff_t upper_limit;
1764         /* This is calculated to be the largest file size for a
1765          * dense, bitmapped file such that the total number of
1766          * sectors in the file, including data and all indirect blocks,
1767          * does not exceed 2^48 -1
1768          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1769          * total number of  512 bytes blocks of the file
1770          */
1771
1772         if (sizeof(blkcnt_t) < sizeof(u64)) {
1773                 /*
1774                  * CONFIG_LSF is not enabled implies the inode
1775                  * i_block represent total blocks in 512 bytes
1776                  * 32 == size of vfs inode i_blocks * 8
1777                  */
1778                 upper_limit = (1LL << 32) - 1;
1779
1780                 /* total blocks in file system block size */
1781                 upper_limit >>= (bits - 9);
1782
1783         } else {
1784                 /*
1785                  * We use 48 bit ext4_inode i_blocks
1786                  * With EXT4_HUGE_FILE_FL set the i_blocks
1787                  * represent total number of blocks in
1788                  * file system block size
1789                  */
1790                 upper_limit = (1LL << 48) - 1;
1791
1792         }
1793
1794         /* indirect blocks */
1795         meta_blocks = 1;
1796         /* double indirect blocks */
1797         meta_blocks += 1 + (1LL << (bits-2));
1798         /* tripple indirect blocks */
1799         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1800
1801         upper_limit -= meta_blocks;
1802         upper_limit <<= bits;
1803
1804         res += 1LL << (bits-2);
1805         res += 1LL << (2*(bits-2));
1806         res += 1LL << (3*(bits-2));
1807         res <<= bits;
1808         if (res > upper_limit)
1809                 res = upper_limit;
1810
1811         if (res > MAX_LFS_FILESIZE)
1812                 res = MAX_LFS_FILESIZE;
1813
1814         return res;
1815 }
1816
1817 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1818                                 ext4_fsblk_t logical_sb_block, int nr)
1819 {
1820         struct ext4_sb_info *sbi = EXT4_SB(sb);
1821         ext4_group_t bg, first_meta_bg;
1822         int has_super = 0;
1823
1824         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1825
1826         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1827             nr < first_meta_bg)
1828                 return logical_sb_block + nr + 1;
1829         bg = sbi->s_desc_per_block * nr;
1830         if (ext4_bg_has_super(sb, bg))
1831                 has_super = 1;
1832         return (has_super + ext4_group_first_block_no(sb, bg));
1833 }
1834
1835 /**
1836  * ext4_get_stripe_size: Get the stripe size.
1837  * @sbi: In memory super block info
1838  *
1839  * If we have specified it via mount option, then
1840  * use the mount option value. If the value specified at mount time is
1841  * greater than the blocks per group use the super block value.
1842  * If the super block value is greater than blocks per group return 0.
1843  * Allocator needs it be less than blocks per group.
1844  *
1845  */
1846 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1847 {
1848         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1849         unsigned long stripe_width =
1850                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1851
1852         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1853                 return sbi->s_stripe;
1854
1855         if (stripe_width <= sbi->s_blocks_per_group)
1856                 return stripe_width;
1857
1858         if (stride <= sbi->s_blocks_per_group)
1859                 return stride;
1860
1861         return 0;
1862 }
1863
1864 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1865                                 __releases(kernel_lock)
1866                                 __acquires(kernel_lock)
1867
1868 {
1869         struct buffer_head * bh;
1870         struct ext4_super_block *es = NULL;
1871         struct ext4_sb_info *sbi;
1872         ext4_fsblk_t block;
1873         ext4_fsblk_t sb_block = get_sb_block(&data);
1874         ext4_fsblk_t logical_sb_block;
1875         unsigned long offset = 0;
1876         unsigned int journal_inum = 0;
1877         unsigned long journal_devnum = 0;
1878         unsigned long def_mount_opts;
1879         struct inode *root;
1880         int ret = -EINVAL;
1881         int blocksize;
1882         int db_count;
1883         int i;
1884         int needs_recovery;
1885         __le32 features;
1886         __u64 blocks_count;
1887         int err;
1888
1889         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1890         if (!sbi)
1891                 return -ENOMEM;
1892         sb->s_fs_info = sbi;
1893         sbi->s_mount_opt = 0;
1894         sbi->s_resuid = EXT4_DEF_RESUID;
1895         sbi->s_resgid = EXT4_DEF_RESGID;
1896         sbi->s_sb_block = sb_block;
1897
1898         unlock_kernel();
1899
1900         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1901         if (!blocksize) {
1902                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1903                 goto out_fail;
1904         }
1905
1906         /*
1907          * The ext4 superblock will not be buffer aligned for other than 1kB
1908          * block sizes.  We need to calculate the offset from buffer start.
1909          */
1910         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1911                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1912                 offset = do_div(logical_sb_block, blocksize);
1913         } else {
1914                 logical_sb_block = sb_block;
1915         }
1916
1917         if (!(bh = sb_bread(sb, logical_sb_block))) {
1918                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1919                 goto out_fail;
1920         }
1921         /*
1922          * Note: s_es must be initialized as soon as possible because
1923          *       some ext4 macro-instructions depend on its value
1924          */
1925         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1926         sbi->s_es = es;
1927         sb->s_magic = le16_to_cpu(es->s_magic);
1928         if (sb->s_magic != EXT4_SUPER_MAGIC)
1929                 goto cantfind_ext4;
1930
1931         /* Set defaults before we parse the mount options */
1932         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1933         if (def_mount_opts & EXT4_DEFM_DEBUG)
1934                 set_opt(sbi->s_mount_opt, DEBUG);
1935         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1936                 set_opt(sbi->s_mount_opt, GRPID);
1937         if (def_mount_opts & EXT4_DEFM_UID16)
1938                 set_opt(sbi->s_mount_opt, NO_UID32);
1939 #ifdef CONFIG_EXT4DEV_FS_XATTR
1940         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1941                 set_opt(sbi->s_mount_opt, XATTR_USER);
1942 #endif
1943 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1944         if (def_mount_opts & EXT4_DEFM_ACL)
1945                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1946 #endif
1947         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1948                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1949         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1950                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1951         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1952                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1953
1954         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1955                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1956         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1957                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1958         else
1959                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1960
1961         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1962         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1963
1964         set_opt(sbi->s_mount_opt, RESERVATION);
1965         set_opt(sbi->s_mount_opt, BARRIER);
1966
1967         /*
1968          * turn on extents feature by default in ext4 filesystem
1969          * User -o noextents to turn it off
1970          */
1971         set_opt(sbi->s_mount_opt, EXTENTS);
1972         /*
1973          * turn on mballoc feature by default in ext4 filesystem
1974          * User -o nomballoc to turn it off
1975          */
1976         set_opt(sbi->s_mount_opt, MBALLOC);
1977
1978         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1979                             NULL, 0))
1980                 goto failed_mount;
1981
1982         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1983                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1984
1985         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1986             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1987              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1988              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1989                 printk(KERN_WARNING
1990                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1991                        "running e2fsck is recommended\n");
1992
1993         /*
1994          * Since ext4 is still considered development code, we require
1995          * that the TEST_FILESYS flag in s->flags be set.
1996          */
1997         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
1998                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
1999                        "OK to use with test code.\n", sb->s_id);
2000                 goto failed_mount;
2001         }
2002
2003         /*
2004          * Check feature flags regardless of the revision level, since we
2005          * previously didn't change the revision level when setting the flags,
2006          * so there is a chance incompat flags are set on a rev 0 filesystem.
2007          */
2008         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2009         if (features) {
2010                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2011                        "unsupported optional features (%x).\n",
2012                        sb->s_id, le32_to_cpu(features));
2013                 goto failed_mount;
2014         }
2015         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2016         if (!(sb->s_flags & MS_RDONLY) && features) {
2017                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2018                        "unsupported optional features (%x).\n",
2019                        sb->s_id, le32_to_cpu(features));
2020                 goto failed_mount;
2021         }
2022         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2023                 /*
2024                  * Large file size enabled file system can only be
2025                  * mount if kernel is build with CONFIG_LSF
2026                  */
2027                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2028                                 !(sb->s_flags & MS_RDONLY)) {
2029                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2030                                         "files cannot be mounted read-write "
2031                                         "without CONFIG_LSF.\n", sb->s_id);
2032                         goto failed_mount;
2033                 }
2034         }
2035         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2036
2037         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2038             blocksize > EXT4_MAX_BLOCK_SIZE) {
2039                 printk(KERN_ERR
2040                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2041                        blocksize, sb->s_id);
2042                 goto failed_mount;
2043         }
2044
2045         if (sb->s_blocksize != blocksize) {
2046
2047                 /* Validate the filesystem blocksize */
2048                 if (!sb_set_blocksize(sb, blocksize)) {
2049                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2050                                         blocksize);
2051                         goto failed_mount;
2052                 }
2053
2054                 brelse (bh);
2055                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2056                 offset = do_div(logical_sb_block, blocksize);
2057                 bh = sb_bread(sb, logical_sb_block);
2058                 if (!bh) {
2059                         printk(KERN_ERR
2060                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2061                         goto failed_mount;
2062                 }
2063                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2064                 sbi->s_es = es;
2065                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2066                         printk (KERN_ERR
2067                                 "EXT4-fs: Magic mismatch, very weird !\n");
2068                         goto failed_mount;
2069                 }
2070         }
2071
2072         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2073         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2074
2075         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2076                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2077                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2078         } else {
2079                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2080                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2081                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2082                     (!is_power_of_2(sbi->s_inode_size)) ||
2083                     (sbi->s_inode_size > blocksize)) {
2084                         printk (KERN_ERR
2085                                 "EXT4-fs: unsupported inode size: %d\n",
2086                                 sbi->s_inode_size);
2087                         goto failed_mount;
2088                 }
2089                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2090                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2091         }
2092         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2093         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2094                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2095                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2096                     !is_power_of_2(sbi->s_desc_size)) {
2097                         printk(KERN_ERR
2098                                "EXT4-fs: unsupported descriptor size %lu\n",
2099                                sbi->s_desc_size);
2100                         goto failed_mount;
2101                 }
2102         } else
2103                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2104         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2105         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2106         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2107                 goto cantfind_ext4;
2108         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2109         if (sbi->s_inodes_per_block == 0)
2110                 goto cantfind_ext4;
2111         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2112                                         sbi->s_inodes_per_block;
2113         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2114         sbi->s_sbh = bh;
2115         sbi->s_mount_state = le16_to_cpu(es->s_state);
2116         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2117         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2118         for (i=0; i < 4; i++)
2119                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2120         sbi->s_def_hash_version = es->s_def_hash_version;
2121
2122         if (sbi->s_blocks_per_group > blocksize * 8) {
2123                 printk (KERN_ERR
2124                         "EXT4-fs: #blocks per group too big: %lu\n",
2125                         sbi->s_blocks_per_group);
2126                 goto failed_mount;
2127         }
2128         if (sbi->s_inodes_per_group > blocksize * 8) {
2129                 printk (KERN_ERR
2130                         "EXT4-fs: #inodes per group too big: %lu\n",
2131                         sbi->s_inodes_per_group);
2132                 goto failed_mount;
2133         }
2134
2135         if (ext4_blocks_count(es) >
2136                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2137                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2138                         " too large to mount safely\n", sb->s_id);
2139                 if (sizeof(sector_t) < 8)
2140                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2141                                         "enabled\n");
2142                 goto failed_mount;
2143         }
2144
2145         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2146                 goto cantfind_ext4;
2147
2148         /* ensure blocks_count calculation below doesn't sign-extend */
2149         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2150             le32_to_cpu(es->s_first_data_block) + 1) {
2151                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2152                        "first data block %u, blocks per group %lu\n",
2153                         ext4_blocks_count(es),
2154                         le32_to_cpu(es->s_first_data_block),
2155                         EXT4_BLOCKS_PER_GROUP(sb));
2156                 goto failed_mount;
2157         }
2158         blocks_count = (ext4_blocks_count(es) -
2159                         le32_to_cpu(es->s_first_data_block) +
2160                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2161         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2162         sbi->s_groups_count = blocks_count;
2163         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2164                    EXT4_DESC_PER_BLOCK(sb);
2165         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2166                                     GFP_KERNEL);
2167         if (sbi->s_group_desc == NULL) {
2168                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2169                 goto failed_mount;
2170         }
2171
2172         bgl_lock_init(&sbi->s_blockgroup_lock);
2173
2174         for (i = 0; i < db_count; i++) {
2175                 block = descriptor_loc(sb, logical_sb_block, i);
2176                 sbi->s_group_desc[i] = sb_bread(sb, block);
2177                 if (!sbi->s_group_desc[i]) {
2178                         printk (KERN_ERR "EXT4-fs: "
2179                                 "can't read group descriptor %d\n", i);
2180                         db_count = i;
2181                         goto failed_mount2;
2182                 }
2183         }
2184         if (!ext4_check_descriptors (sb)) {
2185                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2186                 goto failed_mount2;
2187         }
2188         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2189                 if (!ext4_fill_flex_info(sb)) {
2190                         printk(KERN_ERR
2191                                "EXT4-fs: unable to initialize "
2192                                "flex_bg meta info!\n");
2193                         goto failed_mount2;
2194                 }
2195
2196         sbi->s_gdb_count = db_count;
2197         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2198         spin_lock_init(&sbi->s_next_gen_lock);
2199
2200         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2201                         ext4_count_free_blocks(sb));
2202         if (!err) {
2203                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2204                                 ext4_count_free_inodes(sb));
2205         }
2206         if (!err) {
2207                 err = percpu_counter_init(&sbi->s_dirs_counter,
2208                                 ext4_count_dirs(sb));
2209         }
2210         if (err) {
2211                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2212                 goto failed_mount3;
2213         }
2214
2215         /* per fileystem reservation list head & lock */
2216         spin_lock_init(&sbi->s_rsv_window_lock);
2217         sbi->s_rsv_window_root = RB_ROOT;
2218         /* Add a single, static dummy reservation to the start of the
2219          * reservation window list --- it gives us a placeholder for
2220          * append-at-start-of-list which makes the allocation logic
2221          * _much_ simpler. */
2222         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2223         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2224         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2225         sbi->s_rsv_window_head.rsv_goal_size = 0;
2226         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2227
2228         sbi->s_stripe = ext4_get_stripe_size(sbi);
2229
2230         /*
2231          * set up enough so that it can read an inode
2232          */
2233         sb->s_op = &ext4_sops;
2234         sb->s_export_op = &ext4_export_ops;
2235         sb->s_xattr = ext4_xattr_handlers;
2236 #ifdef CONFIG_QUOTA
2237         sb->s_qcop = &ext4_qctl_operations;
2238         sb->dq_op = &ext4_quota_operations;
2239 #endif
2240         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2241
2242         sb->s_root = NULL;
2243
2244         needs_recovery = (es->s_last_orphan != 0 ||
2245                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2246                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2247
2248         /*
2249          * The first inode we look at is the journal inode.  Don't try
2250          * root first: it may be modified in the journal!
2251          */
2252         if (!test_opt(sb, NOLOAD) &&
2253             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2254                 if (ext4_load_journal(sb, es, journal_devnum))
2255                         goto failed_mount3;
2256                 if (!(sb->s_flags & MS_RDONLY) &&
2257                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2258                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2259                                "ext4_fill_super: Journal transaction "
2260                                "%u is corrupt\n", sb->s_id,
2261                                EXT4_SB(sb)->s_journal->j_failed_commit);
2262                         if (test_opt (sb, ERRORS_RO)) {
2263                                 printk (KERN_CRIT
2264                                         "Mounting filesystem read-only\n");
2265                                 sb->s_flags |= MS_RDONLY;
2266                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2267                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2268                         }
2269                         if (test_opt(sb, ERRORS_PANIC)) {
2270                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2271                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2272                                 ext4_commit_super(sb, es, 1);
2273                                 printk(KERN_CRIT
2274                                        "EXT4-fs (device %s): mount failed\n",
2275                                       sb->s_id);
2276                                 goto failed_mount4;
2277                         }
2278                 }
2279         } else if (journal_inum) {
2280                 if (ext4_create_journal(sb, es, journal_inum))
2281                         goto failed_mount3;
2282         } else {
2283                 if (!silent)
2284                         printk (KERN_ERR
2285                                 "ext4: No journal on filesystem on %s\n",
2286                                 sb->s_id);
2287                 goto failed_mount3;
2288         }
2289
2290         if (ext4_blocks_count(es) > 0xffffffffULL &&
2291             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2292                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2293                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2294                 goto failed_mount4;
2295         }
2296
2297         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2298                 jbd2_journal_set_features(sbi->s_journal,
2299                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2300                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2301         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2302                 jbd2_journal_set_features(sbi->s_journal,
2303                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2304                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2305                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2306         } else {
2307                 jbd2_journal_clear_features(sbi->s_journal,
2308                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2309                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2310         }
2311
2312         /* We have now updated the journal if required, so we can
2313          * validate the data journaling mode. */
2314         switch (test_opt(sb, DATA_FLAGS)) {
2315         case 0:
2316                 /* No mode set, assume a default based on the journal
2317                  * capabilities: ORDERED_DATA if the journal can
2318                  * cope, else JOURNAL_DATA
2319                  */
2320                 if (jbd2_journal_check_available_features
2321                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2322                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2323                 else
2324                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2325                 break;
2326
2327         case EXT4_MOUNT_ORDERED_DATA:
2328         case EXT4_MOUNT_WRITEBACK_DATA:
2329                 if (!jbd2_journal_check_available_features
2330                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2331                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2332                                "requested data journaling mode\n");
2333                         goto failed_mount4;
2334                 }
2335         default:
2336                 break;
2337         }
2338
2339         if (test_opt(sb, NOBH)) {
2340                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2341                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2342                                 "its supported only with writeback mode\n");
2343                         clear_opt(sbi->s_mount_opt, NOBH);
2344                 }
2345         }
2346         /*
2347          * The jbd2_journal_load will have done any necessary log recovery,
2348          * so we can safely mount the rest of the filesystem now.
2349          */
2350
2351         root = ext4_iget(sb, EXT4_ROOT_INO);
2352         if (IS_ERR(root)) {
2353                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2354                 ret = PTR_ERR(root);
2355                 goto failed_mount4;
2356         }
2357         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2358                 iput(root);
2359                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2360                 goto failed_mount4;
2361         }
2362         sb->s_root = d_alloc_root(root);
2363         if (!sb->s_root) {
2364                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2365                 iput(root);
2366                 ret = -ENOMEM;
2367                 goto failed_mount4;
2368         }
2369
2370         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2371
2372         /* determine the minimum size of new large inodes, if present */
2373         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2374                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2375                                                      EXT4_GOOD_OLD_INODE_SIZE;
2376                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2377                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2378                         if (sbi->s_want_extra_isize <
2379                             le16_to_cpu(es->s_want_extra_isize))
2380                                 sbi->s_want_extra_isize =
2381                                         le16_to_cpu(es->s_want_extra_isize);
2382                         if (sbi->s_want_extra_isize <
2383                             le16_to_cpu(es->s_min_extra_isize))
2384                                 sbi->s_want_extra_isize =
2385                                         le16_to_cpu(es->s_min_extra_isize);
2386                 }
2387         }
2388         /* Check if enough inode space is available */
2389         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2390                                                         sbi->s_inode_size) {
2391                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2392                                                        EXT4_GOOD_OLD_INODE_SIZE;
2393                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2394                         "available.\n");
2395         }
2396
2397         /*
2398          * akpm: core read_super() calls in here with the superblock locked.
2399          * That deadlocks, because orphan cleanup needs to lock the superblock
2400          * in numerous places.  Here we just pop the lock - it's relatively
2401          * harmless, because we are now ready to accept write_super() requests,
2402          * and aviro says that's the only reason for hanging onto the
2403          * superblock lock.
2404          */
2405         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2406         ext4_orphan_cleanup(sb, es);
2407         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2408         if (needs_recovery)
2409                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2410         ext4_mark_recovery_complete(sb, es);
2411         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2412                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2413                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2414                 "writeback");
2415
2416         ext4_ext_init(sb);
2417         ext4_mb_init(sb, needs_recovery);
2418
2419         lock_kernel();
2420         return 0;
2421
2422 cantfind_ext4:
2423         if (!silent)
2424                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2425                        sb->s_id);
2426         goto failed_mount;
2427
2428 failed_mount4:
2429         jbd2_journal_destroy(sbi->s_journal);
2430         sbi->s_journal = NULL;
2431 failed_mount3:
2432         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2433         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2434         percpu_counter_destroy(&sbi->s_dirs_counter);
2435 failed_mount2:
2436         for (i = 0; i < db_count; i++)
2437                 brelse(sbi->s_group_desc[i]);
2438         kfree(sbi->s_group_desc);
2439 failed_mount:
2440 #ifdef CONFIG_QUOTA
2441         for (i = 0; i < MAXQUOTAS; i++)
2442                 kfree(sbi->s_qf_names[i]);
2443 #endif
2444         ext4_blkdev_remove(sbi);
2445         brelse(bh);
2446 out_fail:
2447         sb->s_fs_info = NULL;
2448         kfree(sbi);
2449         lock_kernel();
2450         return ret;
2451 }
2452
2453 /*
2454  * Setup any per-fs journal parameters now.  We'll do this both on
2455  * initial mount, once the journal has been initialised but before we've
2456  * done any recovery; and again on any subsequent remount.
2457  */
2458 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2459 {
2460         struct ext4_sb_info *sbi = EXT4_SB(sb);
2461
2462         if (sbi->s_commit_interval)
2463                 journal->j_commit_interval = sbi->s_commit_interval;
2464         /* We could also set up an ext4-specific default for the commit
2465          * interval here, but for now we'll just fall back to the jbd
2466          * default. */
2467
2468         spin_lock(&journal->j_state_lock);
2469         if (test_opt(sb, BARRIER))
2470                 journal->j_flags |= JBD2_BARRIER;
2471         else
2472                 journal->j_flags &= ~JBD2_BARRIER;
2473         spin_unlock(&journal->j_state_lock);
2474 }
2475
2476 static journal_t *ext4_get_journal(struct super_block *sb,
2477                                    unsigned int journal_inum)
2478 {
2479         struct inode *journal_inode;
2480         journal_t *journal;
2481
2482         /* First, test for the existence of a valid inode on disk.  Bad
2483          * things happen if we iget() an unused inode, as the subsequent
2484          * iput() will try to delete it. */
2485
2486         journal_inode = ext4_iget(sb, journal_inum);
2487         if (IS_ERR(journal_inode)) {
2488                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2489                 return NULL;
2490         }
2491         if (!journal_inode->i_nlink) {
2492                 make_bad_inode(journal_inode);
2493                 iput(journal_inode);
2494                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2495                 return NULL;
2496         }
2497
2498         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2499                   journal_inode, journal_inode->i_size);
2500         if (!S_ISREG(journal_inode->i_mode)) {
2501                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2502                 iput(journal_inode);
2503                 return NULL;
2504         }
2505
2506         journal = jbd2_journal_init_inode(journal_inode);
2507         if (!journal) {
2508                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2509                 iput(journal_inode);
2510                 return NULL;
2511         }
2512         journal->j_private = sb;
2513         ext4_init_journal_params(sb, journal);
2514         return journal;
2515 }
2516
2517 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2518                                        dev_t j_dev)
2519 {
2520         struct buffer_head * bh;
2521         journal_t *journal;
2522         ext4_fsblk_t start;
2523         ext4_fsblk_t len;
2524         int hblock, blocksize;
2525         ext4_fsblk_t sb_block;
2526         unsigned long offset;
2527         struct ext4_super_block * es;
2528         struct block_device *bdev;
2529
2530         bdev = ext4_blkdev_get(j_dev);
2531         if (bdev == NULL)
2532                 return NULL;
2533
2534         if (bd_claim(bdev, sb)) {
2535                 printk(KERN_ERR
2536                         "EXT4: failed to claim external journal device.\n");
2537                 blkdev_put(bdev);
2538                 return NULL;
2539         }
2540
2541         blocksize = sb->s_blocksize;
2542         hblock = bdev_hardsect_size(bdev);
2543         if (blocksize < hblock) {
2544                 printk(KERN_ERR
2545                         "EXT4-fs: blocksize too small for journal device.\n");
2546                 goto out_bdev;
2547         }
2548
2549         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2550         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2551         set_blocksize(bdev, blocksize);
2552         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2553                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2554                        "external journal\n");
2555                 goto out_bdev;
2556         }
2557
2558         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2559         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2560             !(le32_to_cpu(es->s_feature_incompat) &
2561               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2562                 printk(KERN_ERR "EXT4-fs: external journal has "
2563                                         "bad superblock\n");
2564                 brelse(bh);
2565                 goto out_bdev;
2566         }
2567
2568         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2569                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2570                 brelse(bh);
2571                 goto out_bdev;
2572         }
2573
2574         len = ext4_blocks_count(es);
2575         start = sb_block + 1;
2576         brelse(bh);     /* we're done with the superblock */
2577
2578         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2579                                         start, len, blocksize);
2580         if (!journal) {
2581                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2582                 goto out_bdev;
2583         }
2584         journal->j_private = sb;
2585         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2586         wait_on_buffer(journal->j_sb_buffer);
2587         if (!buffer_uptodate(journal->j_sb_buffer)) {
2588                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2589                 goto out_journal;
2590         }
2591         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2592                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2593                                         "user (unsupported) - %d\n",
2594                         be32_to_cpu(journal->j_superblock->s_nr_users));
2595                 goto out_journal;
2596         }
2597         EXT4_SB(sb)->journal_bdev = bdev;
2598         ext4_init_journal_params(sb, journal);
2599         return journal;
2600 out_journal:
2601         jbd2_journal_destroy(journal);
2602 out_bdev:
2603         ext4_blkdev_put(bdev);
2604         return NULL;
2605 }
2606
2607 static int ext4_load_journal(struct super_block *sb,
2608                              struct ext4_super_block *es,
2609                              unsigned long journal_devnum)
2610 {
2611         journal_t *journal;
2612         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2613         dev_t journal_dev;
2614         int err = 0;
2615         int really_read_only;
2616
2617         if (journal_devnum &&
2618             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2619                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2620                         "numbers have changed\n");
2621                 journal_dev = new_decode_dev(journal_devnum);
2622         } else
2623                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2624
2625         really_read_only = bdev_read_only(sb->s_bdev);
2626
2627         /*
2628          * Are we loading a blank journal or performing recovery after a
2629          * crash?  For recovery, we need to check in advance whether we
2630          * can get read-write access to the device.
2631          */
2632
2633         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2634                 if (sb->s_flags & MS_RDONLY) {
2635                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2636                                         "required on readonly filesystem.\n");
2637                         if (really_read_only) {
2638                                 printk(KERN_ERR "EXT4-fs: write access "
2639                                         "unavailable, cannot proceed.\n");
2640                                 return -EROFS;
2641                         }
2642                         printk (KERN_INFO "EXT4-fs: write access will "
2643                                         "be enabled during recovery.\n");
2644                 }
2645         }
2646
2647         if (journal_inum && journal_dev) {
2648                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2649                        "and inode journals!\n");
2650                 return -EINVAL;
2651         }
2652
2653         if (journal_inum) {
2654                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2655                         return -EINVAL;
2656         } else {
2657                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2658                         return -EINVAL;
2659         }
2660
2661         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2662                 err = jbd2_journal_update_format(journal);
2663                 if (err)  {
2664                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2665                         jbd2_journal_destroy(journal);
2666                         return err;
2667                 }
2668         }
2669
2670         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2671                 err = jbd2_journal_wipe(journal, !really_read_only);
2672         if (!err)
2673                 err = jbd2_journal_load(journal);
2674
2675         if (err) {
2676                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2677                 jbd2_journal_destroy(journal);
2678                 return err;
2679         }
2680
2681         EXT4_SB(sb)->s_journal = journal;
2682         ext4_clear_journal_err(sb, es);
2683
2684         if (journal_devnum &&
2685             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2686                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2687                 sb->s_dirt = 1;
2688
2689                 /* Make sure we flush the recovery flag to disk. */
2690                 ext4_commit_super(sb, es, 1);
2691         }
2692
2693         return 0;
2694 }
2695
2696 static int ext4_create_journal(struct super_block * sb,
2697                                struct ext4_super_block * es,
2698                                unsigned int journal_inum)
2699 {
2700         journal_t *journal;
2701         int err;
2702
2703         if (sb->s_flags & MS_RDONLY) {
2704                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2705                                 "create journal.\n");
2706                 return -EROFS;
2707         }
2708
2709         journal = ext4_get_journal(sb, journal_inum);
2710         if (!journal)
2711                 return -EINVAL;
2712
2713         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2714                journal_inum);
2715
2716         err = jbd2_journal_create(journal);
2717         if (err) {
2718                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2719                 jbd2_journal_destroy(journal);
2720                 return -EIO;
2721         }
2722
2723         EXT4_SB(sb)->s_journal = journal;
2724
2725         ext4_update_dynamic_rev(sb);
2726         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2727         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2728
2729         es->s_journal_inum = cpu_to_le32(journal_inum);
2730         sb->s_dirt = 1;
2731
2732         /* Make sure we flush the recovery flag to disk. */
2733         ext4_commit_super(sb, es, 1);
2734
2735         return 0;
2736 }
2737
2738 static void ext4_commit_super (struct super_block * sb,
2739                                struct ext4_super_block * es,
2740                                int sync)
2741 {
2742         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2743
2744         if (!sbh)
2745                 return;
2746         es->s_wtime = cpu_to_le32(get_seconds());
2747         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2748         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2749         BUFFER_TRACE(sbh, "marking dirty");
2750         mark_buffer_dirty(sbh);
2751         if (sync)
2752                 sync_dirty_buffer(sbh);
2753 }
2754
2755
2756 /*
2757  * Have we just finished recovery?  If so, and if we are mounting (or
2758  * remounting) the filesystem readonly, then we will end up with a
2759  * consistent fs on disk.  Record that fact.
2760  */
2761 static void ext4_mark_recovery_complete(struct super_block * sb,
2762                                         struct ext4_super_block * es)
2763 {
2764         journal_t *journal = EXT4_SB(sb)->s_journal;
2765
2766         jbd2_journal_lock_updates(journal);
2767         jbd2_journal_flush(journal);
2768         lock_super(sb);
2769         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2770             sb->s_flags & MS_RDONLY) {
2771                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2772                 sb->s_dirt = 0;
2773                 ext4_commit_super(sb, es, 1);
2774         }
2775         unlock_super(sb);
2776         jbd2_journal_unlock_updates(journal);
2777 }
2778
2779 /*
2780  * If we are mounting (or read-write remounting) a filesystem whose journal
2781  * has recorded an error from a previous lifetime, move that error to the
2782  * main filesystem now.
2783  */
2784 static void ext4_clear_journal_err(struct super_block * sb,
2785                                    struct ext4_super_block * es)
2786 {
2787         journal_t *journal;
2788         int j_errno;
2789         const char *errstr;
2790
2791         journal = EXT4_SB(sb)->s_journal;
2792
2793         /*
2794          * Now check for any error status which may have been recorded in the
2795          * journal by a prior ext4_error() or ext4_abort()
2796          */
2797
2798         j_errno = jbd2_journal_errno(journal);
2799         if (j_errno) {
2800                 char nbuf[16];
2801
2802                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2803                 ext4_warning(sb, __func__, "Filesystem error recorded "
2804                              "from previous mount: %s", errstr);
2805                 ext4_warning(sb, __func__, "Marking fs in need of "
2806                              "filesystem check.");
2807
2808                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2809                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2810                 ext4_commit_super (sb, es, 1);
2811
2812                 jbd2_journal_clear_err(journal);
2813         }
2814 }
2815
2816 /*
2817  * Force the running and committing transactions to commit,
2818  * and wait on the commit.
2819  */
2820 int ext4_force_commit(struct super_block *sb)
2821 {
2822         journal_t *journal;
2823         int ret;
2824
2825         if (sb->s_flags & MS_RDONLY)
2826                 return 0;
2827
2828         journal = EXT4_SB(sb)->s_journal;
2829         sb->s_dirt = 0;
2830         ret = ext4_journal_force_commit(journal);
2831         return ret;
2832 }
2833
2834 /*
2835  * Ext4 always journals updates to the superblock itself, so we don't
2836  * have to propagate any other updates to the superblock on disk at this
2837  * point.  Just start an async writeback to get the buffers on their way
2838  * to the disk.
2839  *
2840  * This implicitly triggers the writebehind on sync().
2841  */
2842
2843 static void ext4_write_super (struct super_block * sb)
2844 {
2845         if (mutex_trylock(&sb->s_lock) != 0)
2846                 BUG();
2847         sb->s_dirt = 0;
2848 }
2849
2850 static int ext4_sync_fs(struct super_block *sb, int wait)
2851 {
2852         tid_t target;
2853
2854         sb->s_dirt = 0;
2855         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2856                 if (wait)
2857                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2858         }
2859         return 0;
2860 }
2861
2862 /*
2863  * LVM calls this function before a (read-only) snapshot is created.  This
2864  * gives us a chance to flush the journal completely and mark the fs clean.
2865  */
2866 static void ext4_write_super_lockfs(struct super_block *sb)
2867 {
2868         sb->s_dirt = 0;
2869
2870         if (!(sb->s_flags & MS_RDONLY)) {
2871                 journal_t *journal = EXT4_SB(sb)->s_journal;
2872
2873                 /* Now we set up the journal barrier. */
2874                 jbd2_journal_lock_updates(journal);
2875                 jbd2_journal_flush(journal);
2876
2877                 /* Journal blocked and flushed, clear needs_recovery flag. */
2878                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2879                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2880         }
2881 }
2882
2883 /*
2884  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2885  * flag here, even though the filesystem is not technically dirty yet.
2886  */
2887 static void ext4_unlockfs(struct super_block *sb)
2888 {
2889         if (!(sb->s_flags & MS_RDONLY)) {
2890                 lock_super(sb);
2891                 /* Reser the needs_recovery flag before the fs is unlocked. */
2892                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2893                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2894                 unlock_super(sb);
2895                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2896         }
2897 }
2898
2899 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2900 {
2901         struct ext4_super_block * es;
2902         struct ext4_sb_info *sbi = EXT4_SB(sb);
2903         ext4_fsblk_t n_blocks_count = 0;
2904         unsigned long old_sb_flags;
2905         struct ext4_mount_options old_opts;
2906         int err;
2907 #ifdef CONFIG_QUOTA
2908         int i;
2909 #endif
2910
2911         /* Store the original options */
2912         old_sb_flags = sb->s_flags;
2913         old_opts.s_mount_opt = sbi->s_mount_opt;
2914         old_opts.s_resuid = sbi->s_resuid;
2915         old_opts.s_resgid = sbi->s_resgid;
2916         old_opts.s_commit_interval = sbi->s_commit_interval;
2917 #ifdef CONFIG_QUOTA
2918         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2919         for (i = 0; i < MAXQUOTAS; i++)
2920                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2921 #endif
2922
2923         /*
2924          * Allow the "check" option to be passed as a remount option.
2925          */
2926         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2927                 err = -EINVAL;
2928                 goto restore_opts;
2929         }
2930
2931         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2932                 ext4_abort(sb, __func__, "Abort forced by user");
2933
2934         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2935                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2936
2937         es = sbi->s_es;
2938
2939         ext4_init_journal_params(sb, sbi->s_journal);
2940
2941         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2942                 n_blocks_count > ext4_blocks_count(es)) {
2943                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2944                         err = -EROFS;
2945                         goto restore_opts;
2946                 }
2947
2948                 if (*flags & MS_RDONLY) {
2949                         /*
2950                          * First of all, the unconditional stuff we have to do
2951                          * to disable replay of the journal when we next remount
2952                          */
2953                         sb->s_flags |= MS_RDONLY;
2954
2955                         /*
2956                          * OK, test if we are remounting a valid rw partition
2957                          * readonly, and if so set the rdonly flag and then
2958                          * mark the partition as valid again.
2959                          */
2960                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2961                             (sbi->s_mount_state & EXT4_VALID_FS))
2962                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2963
2964                         /*
2965                          * We have to unlock super so that we can wait for
2966                          * transactions.
2967                          */
2968                         unlock_super(sb);
2969                         ext4_mark_recovery_complete(sb, es);
2970                         lock_super(sb);
2971                 } else {
2972                         __le32 ret;
2973                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2974                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2975                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2976                                        "remount RDWR because of unsupported "
2977                                        "optional features (%x).\n",
2978                                        sb->s_id, le32_to_cpu(ret));
2979                                 err = -EROFS;
2980                                 goto restore_opts;
2981                         }
2982
2983                         /*
2984                          * If we have an unprocessed orphan list hanging
2985                          * around from a previously readonly bdev mount,
2986                          * require a full umount/remount for now.
2987                          */
2988                         if (es->s_last_orphan) {
2989                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2990                                        "remount RDWR because of unprocessed "
2991                                        "orphan inode list.  Please "
2992                                        "umount/remount instead.\n",
2993                                        sb->s_id);
2994                                 err = -EINVAL;
2995                                 goto restore_opts;
2996                         }
2997
2998                         /*
2999                          * Mounting a RDONLY partition read-write, so reread
3000                          * and store the current valid flag.  (It may have
3001                          * been changed by e2fsck since we originally mounted
3002                          * the partition.)
3003                          */
3004                         ext4_clear_journal_err(sb, es);
3005                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3006                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3007                                 goto restore_opts;
3008                         if (!ext4_setup_super (sb, es, 0))
3009                                 sb->s_flags &= ~MS_RDONLY;
3010                 }
3011         }
3012 #ifdef CONFIG_QUOTA
3013         /* Release old quota file names */
3014         for (i = 0; i < MAXQUOTAS; i++)
3015                 if (old_opts.s_qf_names[i] &&
3016                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3017                         kfree(old_opts.s_qf_names[i]);
3018 #endif
3019         return 0;
3020 restore_opts:
3021         sb->s_flags = old_sb_flags;
3022         sbi->s_mount_opt = old_opts.s_mount_opt;
3023         sbi->s_resuid = old_opts.s_resuid;
3024         sbi->s_resgid = old_opts.s_resgid;
3025         sbi->s_commit_interval = old_opts.s_commit_interval;
3026 #ifdef CONFIG_QUOTA
3027         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3028         for (i = 0; i < MAXQUOTAS; i++) {
3029                 if (sbi->s_qf_names[i] &&
3030                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3031                         kfree(sbi->s_qf_names[i]);
3032                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3033         }
3034 #endif
3035         return err;
3036 }
3037
3038 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
3039 {
3040         struct super_block *sb = dentry->d_sb;
3041         struct ext4_sb_info *sbi = EXT4_SB(sb);
3042         struct ext4_super_block *es = sbi->s_es;
3043         u64 fsid;
3044
3045         if (test_opt(sb, MINIX_DF)) {
3046                 sbi->s_overhead_last = 0;
3047         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3048                 ext4_group_t ngroups = sbi->s_groups_count, i;
3049                 ext4_fsblk_t overhead = 0;
3050                 smp_rmb();
3051
3052                 /*
3053                  * Compute the overhead (FS structures).  This is constant
3054                  * for a given filesystem unless the number of block groups
3055                  * changes so we cache the previous value until it does.
3056                  */
3057
3058                 /*
3059                  * All of the blocks before first_data_block are
3060                  * overhead
3061                  */
3062                 overhead = le32_to_cpu(es->s_first_data_block);
3063
3064                 /*
3065                  * Add the overhead attributed to the superblock and
3066                  * block group descriptors.  If the sparse superblocks
3067                  * feature is turned on, then not all groups have this.
3068                  */
3069                 for (i = 0; i < ngroups; i++) {
3070                         overhead += ext4_bg_has_super(sb, i) +
3071                                 ext4_bg_num_gdb(sb, i);
3072                         cond_resched();
3073                 }
3074
3075                 /*
3076                  * Every block group has an inode bitmap, a block
3077                  * bitmap, and an inode table.
3078                  */
3079                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3080                 sbi->s_overhead_last = overhead;
3081                 smp_wmb();
3082                 sbi->s_blocks_last = ext4_blocks_count(es);
3083         }
3084
3085         buf->f_type = EXT4_SUPER_MAGIC;
3086         buf->f_bsize = sb->s_blocksize;
3087         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3088         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
3089         ext4_free_blocks_count_set(es, buf->f_bfree);
3090         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3091         if (buf->f_bfree < ext4_r_blocks_count(es))
3092                 buf->f_bavail = 0;
3093         buf->f_files = le32_to_cpu(es->s_inodes_count);
3094         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3095         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3096         buf->f_namelen = EXT4_NAME_LEN;
3097         fsid = le64_to_cpup((void *)es->s_uuid) ^
3098                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3099         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3100         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3101         return 0;
3102 }
3103
3104 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3105  * is locked for write. Otherwise the are possible deadlocks:
3106  * Process 1                         Process 2
3107  * ext4_create()                     quota_sync()
3108  *   jbd2_journal_start()                   write_dquot()
3109  *   DQUOT_INIT()                        down(dqio_mutex)
3110  *     down(dqio_mutex)                    jbd2_journal_start()
3111  *
3112  */
3113
3114 #ifdef CONFIG_QUOTA
3115
3116 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3117 {
3118         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3119 }
3120
3121 static int ext4_dquot_initialize(struct inode *inode, int type)
3122 {
3123         handle_t *handle;
3124         int ret, err;
3125
3126         /* We may create quota structure so we need to reserve enough blocks */
3127         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3128         if (IS_ERR(handle))
3129                 return PTR_ERR(handle);
3130         ret = dquot_initialize(inode, type);
3131         err = ext4_journal_stop(handle);
3132         if (!ret)
3133                 ret = err;
3134         return ret;
3135 }
3136
3137 static int ext4_dquot_drop(struct inode *inode)
3138 {
3139         handle_t *handle;
3140         int ret, err;
3141
3142         /* We may delete quota structure so we need to reserve enough blocks */
3143         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3144         if (IS_ERR(handle)) {
3145                 /*
3146                  * We call dquot_drop() anyway to at least release references
3147                  * to quota structures so that umount does not hang.
3148                  */
3149                 dquot_drop(inode);
3150                 return PTR_ERR(handle);
3151         }
3152         ret = dquot_drop(inode);
3153         err = ext4_journal_stop(handle);
3154         if (!ret)
3155                 ret = err;
3156         return ret;
3157 }
3158
3159 static int ext4_write_dquot(struct dquot *dquot)
3160 {
3161         int ret, err;
3162         handle_t *handle;
3163         struct inode *inode;
3164
3165         inode = dquot_to_inode(dquot);
3166         handle = ext4_journal_start(inode,
3167                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3168         if (IS_ERR(handle))
3169                 return PTR_ERR(handle);
3170         ret = dquot_commit(dquot);
3171         err = ext4_journal_stop(handle);
3172         if (!ret)
3173                 ret = err;
3174         return ret;
3175 }
3176
3177 static int ext4_acquire_dquot(struct dquot *dquot)
3178 {
3179         int ret, err;
3180         handle_t *handle;
3181
3182         handle = ext4_journal_start(dquot_to_inode(dquot),
3183                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3184         if (IS_ERR(handle))
3185                 return PTR_ERR(handle);
3186         ret = dquot_acquire(dquot);
3187         err = ext4_journal_stop(handle);
3188         if (!ret)
3189                 ret = err;
3190         return ret;
3191 }
3192
3193 static int ext4_release_dquot(struct dquot *dquot)
3194 {
3195         int ret, err;
3196         handle_t *handle;
3197
3198         handle = ext4_journal_start(dquot_to_inode(dquot),
3199                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3200         if (IS_ERR(handle)) {
3201                 /* Release dquot anyway to avoid endless cycle in dqput() */
3202                 dquot_release(dquot);
3203                 return PTR_ERR(handle);
3204         }
3205         ret = dquot_release(dquot);
3206         err = ext4_journal_stop(handle);
3207         if (!ret)
3208                 ret = err;
3209         return ret;
3210 }
3211
3212 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3213 {
3214         /* Are we journaling quotas? */
3215         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3216             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3217                 dquot_mark_dquot_dirty(dquot);
3218                 return ext4_write_dquot(dquot);
3219         } else {
3220                 return dquot_mark_dquot_dirty(dquot);
3221         }
3222 }
3223
3224 static int ext4_write_info(struct super_block *sb, int type)
3225 {
3226         int ret, err;
3227         handle_t *handle;
3228
3229         /* Data block + inode block */
3230         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3231         if (IS_ERR(handle))
3232                 return PTR_ERR(handle);
3233         ret = dquot_commit_info(sb, type);
3234         err = ext4_journal_stop(handle);
3235         if (!ret)
3236                 ret = err;
3237         return ret;
3238 }
3239
3240 /*
3241  * Turn on quotas during mount time - we need to find
3242  * the quota file and such...
3243  */
3244 static int ext4_quota_on_mount(struct super_block *sb, int type)
3245 {
3246         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3247                         EXT4_SB(sb)->s_jquota_fmt, type);
3248 }
3249
3250 /*
3251  * Standard function to be called on quota_on
3252  */
3253 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3254                          char *path, int remount)
3255 {
3256         int err;
3257         struct nameidata nd;
3258
3259         if (!test_opt(sb, QUOTA))
3260                 return -EINVAL;
3261         /* When remounting, no checks are needed and in fact, path is NULL */
3262         if (remount)
3263                 return vfs_quota_on(sb, type, format_id, path, remount);
3264
3265         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3266         if (err)
3267                 return err;
3268
3269         /* Quotafile not on the same filesystem? */
3270         if (nd.path.mnt->mnt_sb != sb) {
3271                 path_put(&nd.path);
3272                 return -EXDEV;
3273         }
3274         /* Journaling quota? */
3275         if (EXT4_SB(sb)->s_qf_names[type]) {
3276                 /* Quotafile not of fs root? */
3277                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3278                         printk(KERN_WARNING
3279                                 "EXT4-fs: Quota file not on filesystem root. "
3280                                 "Journaled quota will not work.\n");
3281         }
3282
3283         /*
3284          * When we journal data on quota file, we have to flush journal to see
3285          * all updates to the file when we bypass pagecache...
3286          */
3287         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3288                 /*
3289                  * We don't need to lock updates but journal_flush() could
3290                  * otherwise be livelocked...
3291                  */
3292                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3293                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3294                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3295         }
3296
3297         path_put(&nd.path);
3298         return vfs_quota_on(sb, type, format_id, path, remount);
3299 }
3300
3301 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3302  * acquiring the locks... As quota files are never truncated and quota code
3303  * itself serializes the operations (and noone else should touch the files)
3304  * we don't have to be afraid of races */
3305 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3306                                size_t len, loff_t off)
3307 {
3308         struct inode *inode = sb_dqopt(sb)->files[type];
3309         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3310         int err = 0;
3311         int offset = off & (sb->s_blocksize - 1);
3312         int tocopy;
3313         size_t toread;
3314         struct buffer_head *bh;
3315         loff_t i_size = i_size_read(inode);
3316
3317         if (off > i_size)
3318                 return 0;
3319         if (off+len > i_size)
3320                 len = i_size-off;
3321         toread = len;
3322         while (toread > 0) {
3323                 tocopy = sb->s_blocksize - offset < toread ?
3324                                 sb->s_blocksize - offset : toread;
3325                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3326                 if (err)
3327                         return err;
3328                 if (!bh)        /* A hole? */
3329                         memset(data, 0, tocopy);
3330                 else
3331                         memcpy(data, bh->b_data+offset, tocopy);
3332                 brelse(bh);
3333                 offset = 0;
3334                 toread -= tocopy;
3335                 data += tocopy;
3336                 blk++;
3337         }
3338         return len;
3339 }
3340
3341 /* Write to quotafile (we know the transaction is already started and has
3342  * enough credits) */
3343 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3344                                 const char *data, size_t len, loff_t off)
3345 {
3346         struct inode *inode = sb_dqopt(sb)->files[type];
3347         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3348         int err = 0;
3349         int offset = off & (sb->s_blocksize - 1);
3350         int tocopy;
3351         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3352         size_t towrite = len;
3353         struct buffer_head *bh;
3354         handle_t *handle = journal_current_handle();
3355
3356         if (!handle) {
3357                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3358                         " cancelled because transaction is not started.\n",
3359                         (unsigned long long)off, (unsigned long long)len);
3360                 return -EIO;
3361         }
3362         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3363         while (towrite > 0) {
3364                 tocopy = sb->s_blocksize - offset < towrite ?
3365                                 sb->s_blocksize - offset : towrite;
3366                 bh = ext4_bread(handle, inode, blk, 1, &err);
3367                 if (!bh)
3368                         goto out;
3369                 if (journal_quota) {
3370                         err = ext4_journal_get_write_access(handle, bh);
3371                         if (err) {
3372                                 brelse(bh);
3373                                 goto out;
3374                         }
3375                 }
3376                 lock_buffer(bh);
3377                 memcpy(bh->b_data+offset, data, tocopy);
3378                 flush_dcache_page(bh->b_page);
3379                 unlock_buffer(bh);
3380                 if (journal_quota)
3381                         err = ext4_journal_dirty_metadata(handle, bh);
3382                 else {
3383                         /* Always do at least ordered writes for quotas */
3384                         err = ext4_jbd2_file_inode(handle, inode);
3385                         mark_buffer_dirty(bh);
3386                 }
3387                 brelse(bh);
3388                 if (err)
3389                         goto out;
3390                 offset = 0;
3391                 towrite -= tocopy;
3392                 data += tocopy;
3393                 blk++;
3394         }
3395 out:
3396         if (len == towrite) {
3397                 mutex_unlock(&inode->i_mutex);
3398                 return err;
3399         }
3400         if (inode->i_size < off+len-towrite) {
3401                 i_size_write(inode, off+len-towrite);
3402                 EXT4_I(inode)->i_disksize = inode->i_size;
3403         }
3404         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3405         ext4_mark_inode_dirty(handle, inode);
3406         mutex_unlock(&inode->i_mutex);
3407         return len - towrite;
3408 }
3409
3410 #endif
3411
3412 static int ext4_get_sb(struct file_system_type *fs_type,
3413         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3414 {
3415         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3416 }
3417
3418 static struct file_system_type ext4dev_fs_type = {
3419         .owner          = THIS_MODULE,
3420         .name           = "ext4dev",
3421         .get_sb         = ext4_get_sb,
3422         .kill_sb        = kill_block_super,
3423         .fs_flags       = FS_REQUIRES_DEV,
3424 };
3425
3426 static int __init init_ext4_fs(void)
3427 {
3428         int err;
3429
3430         err = init_ext4_mballoc();
3431         if (err)
3432                 return err;
3433
3434         err = init_ext4_xattr();
3435         if (err)
3436                 goto out2;
3437         err = init_inodecache();
3438         if (err)
3439                 goto out1;
3440         err = register_filesystem(&ext4dev_fs_type);
3441         if (err)
3442                 goto out;
3443         return 0;
3444 out:
3445         destroy_inodecache();
3446 out1:
3447         exit_ext4_xattr();
3448 out2:
3449         exit_ext4_mballoc();
3450         return err;
3451 }
3452
3453 static void __exit exit_ext4_fs(void)
3454 {
3455         unregister_filesystem(&ext4dev_fs_type);
3456         destroy_inodecache();
3457         exit_ext4_xattr();
3458         exit_ext4_mballoc();
3459 }
3460
3461 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3462 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3463 MODULE_LICENSE("GPL");
3464 module_init(init_ext4_fs)
3465 module_exit(exit_ext4_fs)