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         ei->i_reserved_data_blocks = 0;
 578         ei->i_reserved_meta_blocks = 0;
 579         ei->i_allocated_meta_blocks = 0;
 580         ei->i_delalloc_reserved_flag = 0;
 581         spin_lock_init(&(ei->i_block_reservation_lock));
 582         return &ei->vfs_inode;
 583 }
 584
 585 static void ext4_destroy_inode(struct inode *inode)
 586 {
 587         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 588                 printk("EXT4 Inode %p: orphan list check failed!\n",
 589                         EXT4_I(inode));
 590                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 591                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 592                                 true);
 593                 dump_stack();
 594         }
 595         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 596 }
 597
 598 static void init_once(void *foo)
 599 {
 600         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 601
 602         INIT_LIST_HEAD(&ei->i_orphan);
 603 #ifdef CONFIG_EXT4DEV_FS_XATTR
 604         init_rwsem(&ei->xattr_sem);
 605 #endif
 606         init_rwsem(&ei->i_data_sem);
 607         inode_init_once(&ei->vfs_inode);
 608 }
 609
 610 static int init_inodecache(void)
 611 {
 612         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 613                                              sizeof(struct ext4_inode_info),
 614                                              0, (SLAB_RECLAIM_ACCOUNT|
 615                                                 SLAB_MEM_SPREAD),
 616                                              init_once);
 617         if (ext4_inode_cachep == NULL)
 618                 return -ENOMEM;
 619         return 0;
 620 }
 621
 622 static void destroy_inodecache(void)
 623 {
 624         kmem_cache_destroy(ext4_inode_cachep);
 625 }
 626
 627 static void ext4_clear_inode(struct inode *inode)
 628 {
 629         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 630 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 631         if (EXT4_I(inode)->i_acl &&
 632                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 633                 posix_acl_release(EXT4_I(inode)->i_acl);
 634                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 635         }
 636         if (EXT4_I(inode)->i_default_acl &&
 637                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 638                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 639                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 640         }
 641 #endif
 642         ext4_discard_reservation(inode);
 643         EXT4_I(inode)->i_block_alloc_info = NULL;
 644         if (unlikely(rsv))
 645                 kfree(rsv);
 646         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 647                                        &EXT4_I(inode)->jinode);
 648 }
 649
 650 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 651 {
 652 #if defined(CONFIG_QUOTA)
 653         struct ext4_sb_info *sbi = EXT4_SB(sb);
 654
 655         if (sbi->s_jquota_fmt)
 656                 seq_printf(seq, ",jqfmt=%s",
 657                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 658
 659         if (sbi->s_qf_names[USRQUOTA])
 660                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 661
 662         if (sbi->s_qf_names[GRPQUOTA])
 663                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 664
 665         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 666                 seq_puts(seq, ",usrquota");
 667
 668         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 669                 seq_puts(seq, ",grpquota");
 670 #endif
 671 }
 672
 673 /*
 674  * Show an option if
 675  *  - it's set to a non-default value OR
 676  *  - if the per-sb default is different from the global default
 677  */
 678 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 679 {
 680         int def_errors;
 681         unsigned long def_mount_opts;
 682         struct super_block *sb = vfs->mnt_sb;
 683         struct ext4_sb_info *sbi = EXT4_SB(sb);
 684         struct ext4_super_block *es = sbi->s_es;
 685
 686         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 687         def_errors     = le16_to_cpu(es->s_errors);
 688
 689         if (sbi->s_sb_block != 1)
 690                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 691         if (test_opt(sb, MINIX_DF))
 692                 seq_puts(seq, ",minixdf");
 693         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 694                 seq_puts(seq, ",grpid");
 695         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 696                 seq_puts(seq, ",nogrpid");
 697         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 698             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 699                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 700         }
 701         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 702             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 703                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 704         }
 705         if (test_opt(sb, ERRORS_RO)) {
 706                 if (def_errors == EXT4_ERRORS_PANIC ||
 707                     def_errors == EXT4_ERRORS_CONTINUE) {
 708                         seq_puts(seq, ",errors=remount-ro");
 709                 }
 710         }
 711         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 712                 seq_puts(seq, ",errors=continue");
 713         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 714                 seq_puts(seq, ",errors=panic");
 715         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 716                 seq_puts(seq, ",nouid32");
 717         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 718                 seq_puts(seq, ",debug");
 719         if (test_opt(sb, OLDALLOC))
 720                 seq_puts(seq, ",oldalloc");
 721 #ifdef CONFIG_EXT4DEV_FS_XATTR
 722         if (test_opt(sb, XATTR_USER) &&
 723                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 724                 seq_puts(seq, ",user_xattr");
 725         if (!test_opt(sb, XATTR_USER) &&
 726             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 727                 seq_puts(seq, ",nouser_xattr");
 728         }
 729 #endif
 730 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 731         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 732                 seq_puts(seq, ",acl");
 733         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 734                 seq_puts(seq, ",noacl");
 735 #endif
 736         if (!test_opt(sb, RESERVATION))
 737                 seq_puts(seq, ",noreservation");
 738         if (sbi->s_commit_interval) {
 739                 seq_printf(seq, ",commit=%u",
 740                            (unsigned) (sbi->s_commit_interval / HZ));
 741         }
 742         /*
 743          * We're changing the default of barrier mount option, so
 744          * let's always display its mount state so it's clear what its
 745          * status is.
 746          */
 747         seq_puts(seq, ",barrier=");
 748         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 749         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 750                 seq_puts(seq, ",journal_async_commit");
 751         if (test_opt(sb, NOBH))
 752                 seq_puts(seq, ",nobh");
 753         if (!test_opt(sb, EXTENTS))
 754                 seq_puts(seq, ",noextents");
 755         if (!test_opt(sb, MBALLOC))
 756                 seq_puts(seq, ",nomballoc");
 757         if (test_opt(sb, I_VERSION))
 758                 seq_puts(seq, ",i_version");
 759         if (!test_opt(sb, DELALLOC))
 760                 seq_puts(seq, ",nodelalloc");
 761
 762
 763         if (sbi->s_stripe)
 764                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 765         /*
 766          * journal mode get enabled in different ways
 767          * So just print the value even if we didn't specify it
 768          */
 769         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 770                 seq_puts(seq, ",data=journal");
 771         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 772                 seq_puts(seq, ",data=ordered");
 773         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 774                 seq_puts(seq, ",data=writeback");
 775
 776         ext4_show_quota_options(seq, sb);
 777         return 0;
 778 }
 779
 780
 781 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 782                 u64 ino, u32 generation)
 783 {
 784         struct inode *inode;
 785
 786         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 787                 return ERR_PTR(-ESTALE);
 788         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 789                 return ERR_PTR(-ESTALE);
 790
 791         /* iget isn't really right if the inode is currently unallocated!!
 792          *
 793          * ext4_read_inode will return a bad_inode if the inode had been
 794          * deleted, so we should be safe.
 795          *
 796          * Currently we don't know the generation for parent directory, so
 797          * a generation of 0 means "accept any"
 798          */
 799         inode = ext4_iget(sb, ino);
 800         if (IS_ERR(inode))
 801                 return ERR_CAST(inode);
 802         if (generation && inode->i_generation != generation) {
 803                 iput(inode);
 804                 return ERR_PTR(-ESTALE);
 805         }
 806
 807         return inode;
 808 }
 809
 810 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 811                 int fh_len, int fh_type)
 812 {
 813         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 814                                     ext4_nfs_get_inode);
 815 }
 816
 817 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 818                 int fh_len, int fh_type)
 819 {
 820         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 821                                     ext4_nfs_get_inode);
 822 }
 823
 824 #ifdef CONFIG_QUOTA
 825 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 826 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 827
 828 static int ext4_dquot_initialize(struct inode *inode, int type);
 829 static int ext4_dquot_drop(struct inode *inode);
 830 static int ext4_write_dquot(struct dquot *dquot);
 831 static int ext4_acquire_dquot(struct dquot *dquot);
 832 static int ext4_release_dquot(struct dquot *dquot);
 833 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 834 static int ext4_write_info(struct super_block *sb, int type);
 835 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 836                                 char *path, int remount);
 837 static int ext4_quota_on_mount(struct super_block *sb, int type);
 838 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 839                                size_t len, loff_t off);
 840 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 841                                 const char *data, size_t len, loff_t off);
 842
 843 static struct dquot_operations ext4_quota_operations = {
 844         .initialize     = ext4_dquot_initialize,
 845         .drop           = ext4_dquot_drop,
 846         .alloc_space    = dquot_alloc_space,
 847         .alloc_inode    = dquot_alloc_inode,
 848         .free_space     = dquot_free_space,
 849         .free_inode     = dquot_free_inode,
 850         .transfer       = dquot_transfer,
 851         .write_dquot    = ext4_write_dquot,
 852         .acquire_dquot  = ext4_acquire_dquot,
 853         .release_dquot  = ext4_release_dquot,
 854         .mark_dirty     = ext4_mark_dquot_dirty,
 855         .write_info     = ext4_write_info
 856 };
 857
 858 static struct quotactl_ops ext4_qctl_operations = {
 859         .quota_on       = ext4_quota_on,
 860         .quota_off      = vfs_quota_off,
 861         .quota_sync     = vfs_quota_sync,
 862         .get_info       = vfs_get_dqinfo,
 863         .set_info       = vfs_set_dqinfo,
 864         .get_dqblk      = vfs_get_dqblk,
 865         .set_dqblk      = vfs_set_dqblk
 866 };
 867 #endif
 868
 869 static const struct super_operations ext4_sops = {
 870         .alloc_inode    = ext4_alloc_inode,
 871         .destroy_inode  = ext4_destroy_inode,
 872         .write_inode    = ext4_write_inode,
 873         .dirty_inode    = ext4_dirty_inode,
 874         .delete_inode   = ext4_delete_inode,
 875         .put_super      = ext4_put_super,
 876         .write_super    = ext4_write_super,
 877         .sync_fs        = ext4_sync_fs,
 878         .write_super_lockfs = ext4_write_super_lockfs,
 879         .unlockfs       = ext4_unlockfs,
 880         .statfs         = ext4_statfs,
 881         .remount_fs     = ext4_remount,
 882         .clear_inode    = ext4_clear_inode,
 883         .show_options   = ext4_show_options,
 884 #ifdef CONFIG_QUOTA
 885         .quota_read     = ext4_quota_read,
 886         .quota_write    = ext4_quota_write,
 887 #endif
 888 };
 889
 890 static const struct export_operations ext4_export_ops = {
 891         .fh_to_dentry = ext4_fh_to_dentry,
 892         .fh_to_parent = ext4_fh_to_parent,
 893         .get_parent = ext4_get_parent,
 894 };
 895
 896 enum {
 897         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 898         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 899         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 900         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 901         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 902         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 903         Opt_journal_checksum, Opt_journal_async_commit,
 904         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 905         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 906         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 907         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 908         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 909         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 910 };
 911
 912 static match_table_t tokens = {
 913         {Opt_bsd_df, "bsddf"},
 914         {Opt_minix_df, "minixdf"},
 915         {Opt_grpid, "grpid"},
 916         {Opt_grpid, "bsdgroups"},
 917         {Opt_nogrpid, "nogrpid"},
 918         {Opt_nogrpid, "sysvgroups"},
 919         {Opt_resgid, "resgid=%u"},
 920         {Opt_resuid, "resuid=%u"},
 921         {Opt_sb, "sb=%u"},
 922         {Opt_err_cont, "errors=continue"},
 923         {Opt_err_panic, "errors=panic"},
 924         {Opt_err_ro, "errors=remount-ro"},
 925         {Opt_nouid32, "nouid32"},
 926         {Opt_nocheck, "nocheck"},
 927         {Opt_nocheck, "check=none"},
 928         {Opt_debug, "debug"},
 929         {Opt_oldalloc, "oldalloc"},
 930         {Opt_orlov, "orlov"},
 931         {Opt_user_xattr, "user_xattr"},
 932         {Opt_nouser_xattr, "nouser_xattr"},
 933         {Opt_acl, "acl"},
 934         {Opt_noacl, "noacl"},
 935         {Opt_reservation, "reservation"},
 936         {Opt_noreservation, "noreservation"},
 937         {Opt_noload, "noload"},
 938         {Opt_nobh, "nobh"},
 939         {Opt_bh, "bh"},
 940         {Opt_commit, "commit=%u"},
 941         {Opt_journal_update, "journal=update"},
 942         {Opt_journal_inum, "journal=%u"},
 943         {Opt_journal_dev, "journal_dev=%u"},
 944         {Opt_journal_checksum, "journal_checksum"},
 945         {Opt_journal_async_commit, "journal_async_commit"},
 946         {Opt_abort, "abort"},
 947         {Opt_data_journal, "data=journal"},
 948         {Opt_data_ordered, "data=ordered"},
 949         {Opt_data_writeback, "data=writeback"},
 950         {Opt_offusrjquota, "usrjquota="},
 951         {Opt_usrjquota, "usrjquota=%s"},
 952         {Opt_offgrpjquota, "grpjquota="},
 953         {Opt_grpjquota, "grpjquota=%s"},
 954         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 955         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 956         {Opt_grpquota, "grpquota"},
 957         {Opt_noquota, "noquota"},
 958         {Opt_quota, "quota"},
 959         {Opt_usrquota, "usrquota"},
 960         {Opt_barrier, "barrier=%u"},
 961         {Opt_extents, "extents"},
 962         {Opt_noextents, "noextents"},
 963         {Opt_i_version, "i_version"},
 964         {Opt_mballoc, "mballoc"},
 965         {Opt_nomballoc, "nomballoc"},
 966         {Opt_stripe, "stripe=%u"},
 967         {Opt_resize, "resize"},
 968         {Opt_delalloc, "delalloc"},
 969         {Opt_nodelalloc, "nodelalloc"},
 970         {Opt_err, NULL},
 971 };
 972
 973 static ext4_fsblk_t get_sb_block(void **data)
 974 {
 975         ext4_fsblk_t    sb_block;
 976         char            *options = (char *) *data;
 977
 978         if (!options || strncmp(options, "sb=", 3) != 0)
 979                 return 1;       /* Default location */
 980         options += 3;
 981         /*todo: use simple_strtoll with >32bit ext4 */
 982         sb_block = simple_strtoul(options, &options, 0);
 983         if (*options && *options != ',') {
 984                 printk("EXT4-fs: Invalid sb specification: %s\n",
 985                        (char *) *data);
 986                 return 1;
 987         }
 988         if (*options == ',')
 989                 options++;
 990         *data = (void *) options;
 991         return sb_block;
 992 }
 993
 994 static int parse_options (char *options, struct super_block *sb,
 995                           unsigned int *inum, unsigned long *journal_devnum,
 996                           ext4_fsblk_t *n_blocks_count, int is_remount)
 997 {
 998         struct ext4_sb_info *sbi = EXT4_SB(sb);
 999         char * p;
1000         substring_t args[MAX_OPT_ARGS];
1001         int data_opt = 0;
1002         int option;
1003 #ifdef CONFIG_QUOTA
1004         int qtype, qfmt;
1005         char *qname;
1006 #endif
1007         ext4_fsblk_t last_block;
1008
1009         if (!options)
1010                 return 1;
1011
1012         while ((p = strsep (&options, ",")) != NULL) {
1013                 int token;
1014                 if (!*p)
1015                         continue;
1016
1017                 token = match_token(p, tokens, args);
1018                 switch (token) {
1019                 case Opt_bsd_df:
1020                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1021                         break;
1022                 case Opt_minix_df:
1023                         set_opt (sbi->s_mount_opt, MINIX_DF);
1024                         break;
1025                 case Opt_grpid:
1026                         set_opt (sbi->s_mount_opt, GRPID);
1027                         break;
1028                 case Opt_nogrpid:
1029                         clear_opt (sbi->s_mount_opt, GRPID);
1030                         break;
1031                 case Opt_resuid:
1032                         if (match_int(&args[0], &option))
1033                                 return 0;
1034                         sbi->s_resuid = option;
1035                         break;
1036                 case Opt_resgid:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         sbi->s_resgid = option;
1040                         break;
1041                 case Opt_sb:
1042                         /* handled by get_sb_block() instead of here */
1043                         /* *sb_block = match_int(&args[0]); */
1044                         break;
1045                 case Opt_err_panic:
1046                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1047                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1048                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1049                         break;
1050                 case Opt_err_ro:
1051                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1052                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1053                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1054                         break;
1055                 case Opt_err_cont:
1056                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1057                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1058                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1059                         break;
1060                 case Opt_nouid32:
1061                         set_opt (sbi->s_mount_opt, NO_UID32);
1062                         break;
1063                 case Opt_nocheck:
1064                         clear_opt (sbi->s_mount_opt, CHECK);
1065                         break;
1066                 case Opt_debug:
1067                         set_opt (sbi->s_mount_opt, DEBUG);
1068                         break;
1069                 case Opt_oldalloc:
1070                         set_opt (sbi->s_mount_opt, OLDALLOC);
1071                         break;
1072                 case Opt_orlov:
1073                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1074                         break;
1075 #ifdef CONFIG_EXT4DEV_FS_XATTR
1076                 case Opt_user_xattr:
1077                         set_opt (sbi->s_mount_opt, XATTR_USER);
1078                         break;
1079                 case Opt_nouser_xattr:
1080                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1081                         break;
1082 #else
1083                 case Opt_user_xattr:
1084                 case Opt_nouser_xattr:
1085                         printk("EXT4 (no)user_xattr options not supported\n");
1086                         break;
1087 #endif
1088 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1089                 case Opt_acl:
1090                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1091                         break;
1092                 case Opt_noacl:
1093                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1094                         break;
1095 #else
1096                 case Opt_acl:
1097                 case Opt_noacl:
1098                         printk("EXT4 (no)acl options not supported\n");
1099                         break;
1100 #endif
1101                 case Opt_reservation:
1102                         set_opt(sbi->s_mount_opt, RESERVATION);
1103                         break;
1104                 case Opt_noreservation:
1105                         clear_opt(sbi->s_mount_opt, RESERVATION);
1106                         break;
1107                 case Opt_journal_update:
1108                         /* @@@ FIXME */
1109                         /* Eventually we will want to be able to create
1110                            a journal file here.  For now, only allow the
1111                            user to specify an existing inode to be the
1112                            journal file. */
1113                         if (is_remount) {
1114                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1115                                        "journal on remount\n");
1116                                 return 0;
1117                         }
1118                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1119                         break;
1120                 case Opt_journal_inum:
1121                         if (is_remount) {
1122                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1123                                        "journal on remount\n");
1124                                 return 0;
1125                         }
1126                         if (match_int(&args[0], &option))
1127                                 return 0;
1128                         *inum = option;
1129                         break;
1130                 case Opt_journal_dev:
1131                         if (is_remount) {
1132                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1133                                        "journal on remount\n");
1134                                 return 0;
1135                         }
1136                         if (match_int(&args[0], &option))
1137                                 return 0;
1138                         *journal_devnum = option;
1139                         break;
1140                 case Opt_journal_checksum:
1141                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1142                         break;
1143                 case Opt_journal_async_commit:
1144                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1145                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1146                         break;
1147                 case Opt_noload:
1148                         set_opt (sbi->s_mount_opt, NOLOAD);
1149                         break;
1150                 case Opt_commit:
1151                         if (match_int(&args[0], &option))
1152                                 return 0;
1153                         if (option < 0)
1154                                 return 0;
1155                         if (option == 0)
1156                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1157                         sbi->s_commit_interval = HZ * option;
1158                         break;
1159                 case Opt_data_journal:
1160                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1161                         goto datacheck;
1162                 case Opt_data_ordered:
1163                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1164                         goto datacheck;
1165                 case Opt_data_writeback:
1166                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1167                 datacheck:
1168                         if (is_remount) {
1169                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1170                                                 != data_opt) {
1171                                         printk(KERN_ERR
1172                                                 "EXT4-fs: cannot change data "
1173                                                 "mode on remount\n");
1174                                         return 0;
1175                                 }
1176                         } else {
1177                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1178                                 sbi->s_mount_opt |= data_opt;
1179                         }
1180                         break;
1181 #ifdef CONFIG_QUOTA
1182                 case Opt_usrjquota:
1183                         qtype = USRQUOTA;
1184                         goto set_qf_name;
1185                 case Opt_grpjquota:
1186                         qtype = GRPQUOTA;
1187 set_qf_name:
1188                         if ((sb_any_quota_enabled(sb) ||
1189                              sb_any_quota_suspended(sb)) &&
1190                             !sbi->s_qf_names[qtype]) {
1191                                 printk(KERN_ERR
1192                                         "EXT4-fs: Cannot change journaled "
1193                                         "quota options when quota turned on.\n");
1194                                 return 0;
1195                         }
1196                         qname = match_strdup(&args[0]);
1197                         if (!qname) {
1198                                 printk(KERN_ERR
1199                                         "EXT4-fs: not enough memory for "
1200                                         "storing quotafile name.\n");
1201                                 return 0;
1202                         }
1203                         if (sbi->s_qf_names[qtype] &&
1204                             strcmp(sbi->s_qf_names[qtype], qname)) {
1205                                 printk(KERN_ERR
1206                                         "EXT4-fs: %s quota file already "
1207                                         "specified.\n", QTYPE2NAME(qtype));
1208                                 kfree(qname);
1209                                 return 0;
1210                         }
1211                         sbi->s_qf_names[qtype] = qname;
1212                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1213                                 printk(KERN_ERR
1214                                         "EXT4-fs: quotafile must be on "
1215                                         "filesystem root.\n");
1216                                 kfree(sbi->s_qf_names[qtype]);
1217                                 sbi->s_qf_names[qtype] = NULL;
1218                                 return 0;
1219                         }
1220                         set_opt(sbi->s_mount_opt, QUOTA);
1221                         break;
1222                 case Opt_offusrjquota:
1223                         qtype = USRQUOTA;
1224                         goto clear_qf_name;
1225                 case Opt_offgrpjquota:
1226                         qtype = GRPQUOTA;
1227 clear_qf_name:
1228                         if ((sb_any_quota_enabled(sb) ||
1229                              sb_any_quota_suspended(sb)) &&
1230                             sbi->s_qf_names[qtype]) {
1231                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1232                                         "journaled quota options when "
1233                                         "quota turned on.\n");
1234                                 return 0;
1235                         }
1236                         /*
1237                          * The space will be released later when all options
1238                          * are confirmed to be correct
1239                          */
1240                         sbi->s_qf_names[qtype] = NULL;
1241                         break;
1242                 case Opt_jqfmt_vfsold:
1243                         qfmt = QFMT_VFS_OLD;
1244                         goto set_qf_format;
1245                 case Opt_jqfmt_vfsv0:
1246                         qfmt = QFMT_VFS_V0;
1247 set_qf_format:
1248                         if ((sb_any_quota_enabled(sb) ||
1249                              sb_any_quota_suspended(sb)) &&
1250                             sbi->s_jquota_fmt != qfmt) {
1251                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1252                                         "journaled quota options when "
1253                                         "quota turned on.\n");
1254                                 return 0;
1255                         }
1256                         sbi->s_jquota_fmt = qfmt;
1257                         break;
1258                 case Opt_quota:
1259                 case Opt_usrquota:
1260                         set_opt(sbi->s_mount_opt, QUOTA);
1261                         set_opt(sbi->s_mount_opt, USRQUOTA);
1262                         break;
1263                 case Opt_grpquota:
1264                         set_opt(sbi->s_mount_opt, QUOTA);
1265                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1266                         break;
1267                 case Opt_noquota:
1268                         if (sb_any_quota_enabled(sb)) {
1269                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1270                                         "options when quota turned on.\n");
1271                                 return 0;
1272                         }
1273                         clear_opt(sbi->s_mount_opt, QUOTA);
1274                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1275                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1276                         break;
1277 #else
1278                 case Opt_quota:
1279                 case Opt_usrquota:
1280                 case Opt_grpquota:
1281                         printk(KERN_ERR
1282                                 "EXT4-fs: quota options not supported.\n");
1283                         break;
1284                 case Opt_usrjquota:
1285                 case Opt_grpjquota:
1286                 case Opt_offusrjquota:
1287                 case Opt_offgrpjquota:
1288                 case Opt_jqfmt_vfsold:
1289                 case Opt_jqfmt_vfsv0:
1290                         printk(KERN_ERR
1291                                 "EXT4-fs: journaled quota options not "
1292                                 "supported.\n");
1293                         break;
1294                 case Opt_noquota:
1295                         break;
1296 #endif
1297                 case Opt_abort:
1298                         set_opt(sbi->s_mount_opt, ABORT);
1299                         break;
1300                 case Opt_barrier:
1301                         if (match_int(&args[0], &option))
1302                                 return 0;
1303                         if (option)
1304                                 set_opt(sbi->s_mount_opt, BARRIER);
1305                         else
1306                                 clear_opt(sbi->s_mount_opt, BARRIER);
1307                         break;
1308                 case Opt_ignore:
1309                         break;
1310                 case Opt_resize:
1311                         if (!is_remount) {
1312                                 printk("EXT4-fs: resize option only available "
1313                                         "for remount\n");
1314                                 return 0;
1315                         }
1316                         if (match_int(&args[0], &option) != 0)
1317                                 return 0;
1318                         *n_blocks_count = option;
1319                         break;
1320                 case Opt_nobh:
1321                         set_opt(sbi->s_mount_opt, NOBH);
1322                         break;
1323                 case Opt_bh:
1324                         clear_opt(sbi->s_mount_opt, NOBH);
1325                         break;
1326                 case Opt_extents:
1327                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1328                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1329                                 ext4_warning(sb, __func__,
1330                                         "extents feature not enabled "
1331                                         "on this filesystem, use tune2fs\n");
1332                                 return 0;
1333                         }
1334                         set_opt (sbi->s_mount_opt, EXTENTS);
1335                         break;
1336                 case Opt_noextents:
1337                         /*
1338                          * When e2fsprogs support resizing an already existing
1339                          * ext3 file system to greater than 2**32 we need to
1340                          * add support to block allocator to handle growing
1341                          * already existing block  mapped inode so that blocks
1342                          * allocated for them fall within 2**32
1343                          */
1344                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1345                         if (last_block  > 0xffffffffULL) {
1346                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1347                                                 "large to mount with "
1348                                                 "-o noextents options\n");
1349                                 return 0;
1350                         }
1351                         clear_opt (sbi->s_mount_opt, EXTENTS);
1352                         break;
1353                 case Opt_i_version:
1354                         set_opt(sbi->s_mount_opt, I_VERSION);
1355                         sb->s_flags |= MS_I_VERSION;
1356                         break;
1357                 case Opt_nodelalloc:
1358                         clear_opt(sbi->s_mount_opt, DELALLOC);
1359                         break;
1360                 case Opt_mballoc:
1361                         set_opt(sbi->s_mount_opt, MBALLOC);
1362                         break;
1363                 case Opt_nomballoc:
1364                         clear_opt(sbi->s_mount_opt, MBALLOC);
1365                         break;
1366                 case Opt_stripe:
1367                         if (match_int(&args[0], &option))
1368                                 return 0;
1369                         if (option < 0)
1370                                 return 0;
1371                         sbi->s_stripe = option;
1372                         break;
1373                 case Opt_delalloc:
1374                         set_opt(sbi->s_mount_opt, DELALLOC);
1375                         break;
1376                 default:
1377                         printk (KERN_ERR
1378                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1379                                 "or missing value\n", p);
1380                         return 0;
1381                 }
1382         }
1383 #ifdef CONFIG_QUOTA
1384         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1385                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1386                      sbi->s_qf_names[USRQUOTA])
1387                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1388
1389                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1390                      sbi->s_qf_names[GRPQUOTA])
1391                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1392
1393                 if ((sbi->s_qf_names[USRQUOTA] &&
1394                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1395                     (sbi->s_qf_names[GRPQUOTA] &&
1396                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1397                         printk(KERN_ERR "EXT4-fs: old and new quota "
1398                                         "format mixing.\n");
1399                         return 0;
1400                 }
1401
1402                 if (!sbi->s_jquota_fmt) {
1403                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1404                                         "not specified.\n");
1405                         return 0;
1406                 }
1407         } else {
1408                 if (sbi->s_jquota_fmt) {
1409                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1410                                         "specified with no journaling "
1411                                         "enabled.\n");
1412                         return 0;
1413                 }
1414         }
1415 #endif
1416         return 1;
1417 }
1418
1419 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1420                             int read_only)
1421 {
1422         struct ext4_sb_info *sbi = EXT4_SB(sb);
1423         int res = 0;
1424
1425         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1426                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1427                         "forcing read-only mode\n");
1428                 res = MS_RDONLY;
1429         }
1430         if (read_only)
1431                 return res;
1432         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1433                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1434                         "running e2fsck is recommended\n");
1435         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1436                 printk (KERN_WARNING
1437                         "EXT4-fs warning: mounting fs with errors, "
1438                         "running e2fsck is recommended\n");
1439         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1440                  le16_to_cpu(es->s_mnt_count) >=
1441                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1442                 printk (KERN_WARNING
1443                         "EXT4-fs warning: maximal mount count reached, "
1444                         "running e2fsck is recommended\n");
1445         else if (le32_to_cpu(es->s_checkinterval) &&
1446                 (le32_to_cpu(es->s_lastcheck) +
1447                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1448                 printk (KERN_WARNING
1449                         "EXT4-fs warning: checktime reached, "
1450                         "running e2fsck is recommended\n");
1451 #if 0
1452                 /* @@@ We _will_ want to clear the valid bit if we find
1453                  * inconsistencies, to force a fsck at reboot.  But for
1454                  * a plain journaled filesystem we can keep it set as
1455                  * valid forever! :)
1456                  */
1457         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1458 #endif
1459         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1460                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1461         le16_add_cpu(&es->s_mnt_count, 1);
1462         es->s_mtime = cpu_to_le32(get_seconds());
1463         ext4_update_dynamic_rev(sb);
1464         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1465
1466         ext4_commit_super(sb, es, 1);
1467         if (test_opt(sb, DEBUG))
1468                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1469                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1470                         sb->s_blocksize,
1471                         sbi->s_groups_count,
1472                         EXT4_BLOCKS_PER_GROUP(sb),
1473                         EXT4_INODES_PER_GROUP(sb),
1474                         sbi->s_mount_opt);
1475
1476         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1477         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1478                 char b[BDEVNAME_SIZE];
1479
1480                 printk("external journal on %s\n",
1481                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1482         } else {
1483                 printk("internal journal\n");
1484         }
1485         return res;
1486 }
1487
1488 static int ext4_fill_flex_info(struct super_block *sb)
1489 {
1490         struct ext4_sb_info *sbi = EXT4_SB(sb);
1491         struct ext4_group_desc *gdp = NULL;
1492         struct buffer_head *bh;
1493         ext4_group_t flex_group_count;
1494         ext4_group_t flex_group;
1495         int groups_per_flex = 0;
1496         __u64 block_bitmap = 0;
1497         int i;
1498
1499         if (!sbi->s_es->s_log_groups_per_flex) {
1500                 sbi->s_log_groups_per_flex = 0;
1501                 return 1;
1502         }
1503
1504         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1505         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1506
1507         flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
1508                 groups_per_flex;
1509         sbi->s_flex_groups = kmalloc(flex_group_count *
1510                                      sizeof(struct flex_groups), GFP_KERNEL);
1511         if (sbi->s_flex_groups == NULL) {
1512                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
1513                 goto failed;
1514         }
1515         memset(sbi->s_flex_groups, 0, flex_group_count *
1516                sizeof(struct flex_groups));
1517
1518         gdp = ext4_get_group_desc(sb, 1, &bh);
1519         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1520
1521         for (i = 0; i < sbi->s_groups_count; i++) {
1522                 gdp = ext4_get_group_desc(sb, i, &bh);
1523
1524                 flex_group = ext4_flex_group(sbi, i);
1525                 sbi->s_flex_groups[flex_group].free_inodes +=
1526                         le16_to_cpu(gdp->bg_free_inodes_count);
1527                 sbi->s_flex_groups[flex_group].free_blocks +=
1528                         le16_to_cpu(gdp->bg_free_blocks_count);
1529         }
1530
1531         return 1;
1532 failed:
1533         return 0;
1534 }
1535
1536 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1537                             struct ext4_group_desc *gdp)
1538 {
1539         __u16 crc = 0;
1540
1541         if (sbi->s_es->s_feature_ro_compat &
1542             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1543                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1544                 __le32 le_group = cpu_to_le32(block_group);
1545
1546                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1547                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1548                 crc = crc16(crc, (__u8 *)gdp, offset);
1549                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1550                 /* for checksum of struct ext4_group_desc do the rest...*/
1551                 if ((sbi->s_es->s_feature_incompat &
1552                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1553                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1554                         crc = crc16(crc, (__u8 *)gdp + offset,
1555                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1556                                         offset);
1557         }
1558
1559         return cpu_to_le16(crc);
1560 }
1561
1562 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1563                                 struct ext4_group_desc *gdp)
1564 {
1565         if ((sbi->s_es->s_feature_ro_compat &
1566              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1567             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1568                 return 0;
1569
1570         return 1;
1571 }
1572
1573 /* Called at mount-time, super-block is locked */
1574 static int ext4_check_descriptors(struct super_block *sb)
1575 {
1576         struct ext4_sb_info *sbi = EXT4_SB(sb);
1577         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1578         ext4_fsblk_t last_block;
1579         ext4_fsblk_t block_bitmap;
1580         ext4_fsblk_t inode_bitmap;
1581         ext4_fsblk_t inode_table;
1582         int flexbg_flag = 0;
1583         ext4_group_t i;
1584
1585         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1586                 flexbg_flag = 1;
1587
1588         ext4_debug ("Checking group descriptors");
1589
1590         for (i = 0; i < sbi->s_groups_count; i++) {
1591                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1592
1593                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1594                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1595                 else
1596                         last_block = first_block +
1597                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1598
1599                 block_bitmap = ext4_block_bitmap(sb, gdp);
1600                 if (block_bitmap < first_block || block_bitmap > last_block)
1601                 {
1602                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1603                                "Block bitmap for group %lu not in group "
1604                                "(block %llu)!", i, block_bitmap);
1605                         return 0;
1606                 }
1607                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1608                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1609                 {
1610                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1611                                "Inode bitmap for group %lu not in group "
1612                                "(block %llu)!", i, inode_bitmap);
1613                         return 0;
1614                 }
1615                 inode_table = ext4_inode_table(sb, gdp);
1616                 if (inode_table < first_block ||
1617                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1618                 {
1619                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1620                                "Inode table for group %lu not in group "
1621                                "(block %llu)!", i, inode_table);
1622                         return 0;
1623                 }
1624                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1625                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1626                                "Checksum for group %lu failed (%u!=%u)\n",
1627                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1628                                gdp)), le16_to_cpu(gdp->bg_checksum));
1629                         return 0;
1630                 }
1631                 if (!flexbg_flag)
1632                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1633         }
1634
1635         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1636         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1637         return 1;
1638 }
1639
1640 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1641  * the superblock) which were deleted from all directories, but held open by
1642  * a process at the time of a crash.  We walk the list and try to delete these
1643  * inodes at recovery time (only with a read-write filesystem).
1644  *
1645  * In order to keep the orphan inode chain consistent during traversal (in
1646  * case of crash during recovery), we link each inode into the superblock
1647  * orphan list_head and handle it the same way as an inode deletion during
1648  * normal operation (which journals the operations for us).
1649  *
1650  * We only do an iget() and an iput() on each inode, which is very safe if we
1651  * accidentally point at an in-use or already deleted inode.  The worst that
1652  * can happen in this case is that we get a "bit already cleared" message from
1653  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1654  * e2fsck was run on this filesystem, and it must have already done the orphan
1655  * inode cleanup for us, so we can safely abort without any further action.
1656  */
1657 static void ext4_orphan_cleanup (struct super_block * sb,
1658                                  struct ext4_super_block * es)
1659 {
1660         unsigned int s_flags = sb->s_flags;
1661         int nr_orphans = 0, nr_truncates = 0;
1662 #ifdef CONFIG_QUOTA
1663         int i;
1664 #endif
1665         if (!es->s_last_orphan) {
1666                 jbd_debug(4, "no orphan inodes to clean up\n");
1667                 return;
1668         }
1669
1670         if (bdev_read_only(sb->s_bdev)) {
1671                 printk(KERN_ERR "EXT4-fs: write access "
1672                         "unavailable, skipping orphan cleanup.\n");
1673                 return;
1674         }
1675
1676         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1677                 if (es->s_last_orphan)
1678                         jbd_debug(1, "Errors on filesystem, "
1679                                   "clearing orphan list.\n");
1680                 es->s_last_orphan = 0;
1681                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1682                 return;
1683         }
1684
1685         if (s_flags & MS_RDONLY) {
1686                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1687                        sb->s_id);
1688                 sb->s_flags &= ~MS_RDONLY;
1689         }
1690 #ifdef CONFIG_QUOTA
1691         /* Needed for iput() to work correctly and not trash data */
1692         sb->s_flags |= MS_ACTIVE;
1693         /* Turn on quotas so that they are updated correctly */
1694         for (i = 0; i < MAXQUOTAS; i++) {
1695                 if (EXT4_SB(sb)->s_qf_names[i]) {
1696                         int ret = ext4_quota_on_mount(sb, i);
1697                         if (ret < 0)
1698                                 printk(KERN_ERR
1699                                         "EXT4-fs: Cannot turn on journaled "
1700                                         "quota: error %d\n", ret);
1701                 }
1702         }
1703 #endif
1704
1705         while (es->s_last_orphan) {
1706                 struct inode *inode;
1707
1708                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1709                 if (IS_ERR(inode)) {
1710                         es->s_last_orphan = 0;
1711                         break;
1712                 }
1713
1714                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1715                 DQUOT_INIT(inode);
1716                 if (inode->i_nlink) {
1717                         printk(KERN_DEBUG
1718                                 "%s: truncating inode %lu to %Ld bytes\n",
1719                                 __func__, inode->i_ino, inode->i_size);
1720                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1721                                   inode->i_ino, inode->i_size);
1722                         ext4_truncate(inode);
1723                         nr_truncates++;
1724                 } else {
1725                         printk(KERN_DEBUG
1726                                 "%s: deleting unreferenced inode %lu\n",
1727                                 __func__, inode->i_ino);
1728                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1729                                   inode->i_ino);
1730                         nr_orphans++;
1731                 }
1732                 iput(inode);  /* The delete magic happens here! */
1733         }
1734
1735 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1736
1737         if (nr_orphans)
1738                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1739                        sb->s_id, PLURAL(nr_orphans));
1740         if (nr_truncates)
1741                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1742                        sb->s_id, PLURAL(nr_truncates));
1743 #ifdef CONFIG_QUOTA
1744         /* Turn quotas off */
1745         for (i = 0; i < MAXQUOTAS; i++) {
1746                 if (sb_dqopt(sb)->files[i])
1747                         vfs_quota_off(sb, i, 0);
1748         }
1749 #endif
1750         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1751 }
1752 /*
1753  * Maximal extent format file size.
1754  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1755  * extent format containers, within a sector_t, and within i_blocks
1756  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1757  * so that won't be a limiting factor.
1758  *
1759  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1760  */
1761 static loff_t ext4_max_size(int blkbits)
1762 {
1763         loff_t res;
1764         loff_t upper_limit = MAX_LFS_FILESIZE;
1765
1766         /* small i_blocks in vfs inode? */
1767         if (sizeof(blkcnt_t) < sizeof(u64)) {
1768                 /*
1769                  * CONFIG_LSF is not enabled implies the inode
1770                  * i_block represent total blocks in 512 bytes
1771                  * 32 == size of vfs inode i_blocks * 8
1772                  */
1773                 upper_limit = (1LL << 32) - 1;
1774
1775                 /* total blocks in file system block size */
1776                 upper_limit >>= (blkbits - 9);
1777                 upper_limit <<= blkbits;
1778         }
1779
1780         /* 32-bit extent-start container, ee_block */
1781         res = 1LL << 32;
1782         res <<= blkbits;
1783         res -= 1;
1784
1785         /* Sanity check against vm- & vfs- imposed limits */
1786         if (res > upper_limit)
1787                 res = upper_limit;
1788
1789         return res;
1790 }
1791
1792 /*
1793  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1794  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1795  * We need to be 1 filesystem block less than the 2^48 sector limit.
1796  */
1797 static loff_t ext4_max_bitmap_size(int bits)
1798 {
1799         loff_t res = EXT4_NDIR_BLOCKS;
1800         int meta_blocks;
1801         loff_t upper_limit;
1802         /* This is calculated to be the largest file size for a
1803          * dense, bitmapped file such that the total number of
1804          * sectors in the file, including data and all indirect blocks,
1805          * does not exceed 2^48 -1
1806          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1807          * total number of  512 bytes blocks of the file
1808          */
1809
1810         if (sizeof(blkcnt_t) < sizeof(u64)) {
1811                 /*
1812                  * CONFIG_LSF is not enabled implies the inode
1813                  * i_block represent total blocks in 512 bytes
1814                  * 32 == size of vfs inode i_blocks * 8
1815                  */
1816                 upper_limit = (1LL << 32) - 1;
1817
1818                 /* total blocks in file system block size */
1819                 upper_limit >>= (bits - 9);
1820
1821         } else {
1822                 /*
1823                  * We use 48 bit ext4_inode i_blocks
1824                  * With EXT4_HUGE_FILE_FL set the i_blocks
1825                  * represent total number of blocks in
1826                  * file system block size
1827                  */
1828                 upper_limit = (1LL << 48) - 1;
1829
1830         }
1831
1832         /* indirect blocks */
1833         meta_blocks = 1;
1834         /* double indirect blocks */
1835         meta_blocks += 1 + (1LL << (bits-2));
1836         /* tripple indirect blocks */
1837         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1838
1839         upper_limit -= meta_blocks;
1840         upper_limit <<= bits;
1841
1842         res += 1LL << (bits-2);
1843         res += 1LL << (2*(bits-2));
1844         res += 1LL << (3*(bits-2));
1845         res <<= bits;
1846         if (res > upper_limit)
1847                 res = upper_limit;
1848
1849         if (res > MAX_LFS_FILESIZE)
1850                 res = MAX_LFS_FILESIZE;
1851
1852         return res;
1853 }
1854
1855 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1856                                 ext4_fsblk_t logical_sb_block, int nr)
1857 {
1858         struct ext4_sb_info *sbi = EXT4_SB(sb);
1859         ext4_group_t bg, first_meta_bg;
1860         int has_super = 0;
1861
1862         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1863
1864         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1865             nr < first_meta_bg)
1866                 return logical_sb_block + nr + 1;
1867         bg = sbi->s_desc_per_block * nr;
1868         if (ext4_bg_has_super(sb, bg))
1869                 has_super = 1;
1870         return (has_super + ext4_group_first_block_no(sb, bg));
1871 }
1872
1873 /**
1874  * ext4_get_stripe_size: Get the stripe size.
1875  * @sbi: In memory super block info
1876  *
1877  * If we have specified it via mount option, then
1878  * use the mount option value. If the value specified at mount time is
1879  * greater than the blocks per group use the super block value.
1880  * If the super block value is greater than blocks per group return 0.
1881  * Allocator needs it be less than blocks per group.
1882  *
1883  */
1884 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1885 {
1886         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1887         unsigned long stripe_width =
1888                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1889
1890         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1891                 return sbi->s_stripe;
1892
1893         if (stripe_width <= sbi->s_blocks_per_group)
1894                 return stripe_width;
1895
1896         if (stride <= sbi->s_blocks_per_group)
1897                 return stride;
1898
1899         return 0;
1900 }
1901
1902 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1903                                 __releases(kernel_lock)
1904                                 __acquires(kernel_lock)
1905
1906 {
1907         struct buffer_head * bh;
1908         struct ext4_super_block *es = NULL;
1909         struct ext4_sb_info *sbi;
1910         ext4_fsblk_t block;
1911         ext4_fsblk_t sb_block = get_sb_block(&data);
1912         ext4_fsblk_t logical_sb_block;
1913         unsigned long offset = 0;
1914         unsigned int journal_inum = 0;
1915         unsigned long journal_devnum = 0;
1916         unsigned long def_mount_opts;
1917         struct inode *root;
1918         int ret = -EINVAL;
1919         int blocksize;
1920         int db_count;
1921         int i;
1922         int needs_recovery;
1923         __le32 features;
1924         __u64 blocks_count;
1925         int err;
1926
1927         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1928         if (!sbi)
1929                 return -ENOMEM;
1930         sb->s_fs_info = sbi;
1931         sbi->s_mount_opt = 0;
1932         sbi->s_resuid = EXT4_DEF_RESUID;
1933         sbi->s_resgid = EXT4_DEF_RESGID;
1934         sbi->s_sb_block = sb_block;
1935
1936         unlock_kernel();
1937
1938         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1939         if (!blocksize) {
1940                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1941                 goto out_fail;
1942         }
1943
1944         /*
1945          * The ext4 superblock will not be buffer aligned for other than 1kB
1946          * block sizes.  We need to calculate the offset from buffer start.
1947          */
1948         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1949                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1950                 offset = do_div(logical_sb_block, blocksize);
1951         } else {
1952                 logical_sb_block = sb_block;
1953         }
1954
1955         if (!(bh = sb_bread(sb, logical_sb_block))) {
1956                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1957                 goto out_fail;
1958         }
1959         /*
1960          * Note: s_es must be initialized as soon as possible because
1961          *       some ext4 macro-instructions depend on its value
1962          */
1963         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1964         sbi->s_es = es;
1965         sb->s_magic = le16_to_cpu(es->s_magic);
1966         if (sb->s_magic != EXT4_SUPER_MAGIC)
1967                 goto cantfind_ext4;
1968
1969         /* Set defaults before we parse the mount options */
1970         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1971         if (def_mount_opts & EXT4_DEFM_DEBUG)
1972                 set_opt(sbi->s_mount_opt, DEBUG);
1973         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1974                 set_opt(sbi->s_mount_opt, GRPID);
1975         if (def_mount_opts & EXT4_DEFM_UID16)
1976                 set_opt(sbi->s_mount_opt, NO_UID32);
1977 #ifdef CONFIG_EXT4DEV_FS_XATTR
1978         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1979                 set_opt(sbi->s_mount_opt, XATTR_USER);
1980 #endif
1981 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1982         if (def_mount_opts & EXT4_DEFM_ACL)
1983                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1984 #endif
1985         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1986                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1987         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1988                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1989         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1990                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1991
1992         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1993                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1994         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1995                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1996         else
1997                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1998
1999         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2000         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2001
2002         set_opt(sbi->s_mount_opt, RESERVATION);
2003         set_opt(sbi->s_mount_opt, BARRIER);
2004
2005         /*
2006          * turn on extents feature by default in ext4 filesystem
2007          * only if feature flag already set by mkfs or tune2fs.
2008          * Use -o noextents to turn it off
2009          */
2010         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2011                 set_opt(sbi->s_mount_opt, EXTENTS);
2012         else
2013                 ext4_warning(sb, __func__,
2014                         "extents feature not enabled on this filesystem, "
2015                         "use tune2fs.\n");
2016         /*
2017          * turn on mballoc code by default in ext4 filesystem
2018          * Use -o nomballoc to turn it off
2019          */
2020         set_opt(sbi->s_mount_opt, MBALLOC);
2021
2022         /*
2023          * enable delayed allocation by default
2024          * Use -o nodelalloc to turn it off
2025          */
2026         set_opt(sbi->s_mount_opt, DELALLOC);
2027
2028
2029         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
2030                             NULL, 0))
2031                 goto failed_mount;
2032
2033         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2034                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2035
2036         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2037             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2038              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2039              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2040                 printk(KERN_WARNING
2041                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2042                        "running e2fsck is recommended\n");
2043
2044         /*
2045          * Since ext4 is still considered development code, we require
2046          * that the TEST_FILESYS flag in s->flags be set.
2047          */
2048         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
2049                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
2050                        "OK to use with test code.\n", sb->s_id);
2051                 goto failed_mount;
2052         }
2053
2054         /*
2055          * Check feature flags regardless of the revision level, since we
2056          * previously didn't change the revision level when setting the flags,
2057          * so there is a chance incompat flags are set on a rev 0 filesystem.
2058          */
2059         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2060         if (features) {
2061                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2062                        "unsupported optional features (%x).\n",
2063                        sb->s_id, le32_to_cpu(features));
2064                 goto failed_mount;
2065         }
2066         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2067         if (!(sb->s_flags & MS_RDONLY) && features) {
2068                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2069                        "unsupported optional features (%x).\n",
2070                        sb->s_id, le32_to_cpu(features));
2071                 goto failed_mount;
2072         }
2073         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2074                 /*
2075                  * Large file size enabled file system can only be
2076                  * mount if kernel is build with CONFIG_LSF
2077                  */
2078                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2079                                 !(sb->s_flags & MS_RDONLY)) {
2080                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2081                                         "files cannot be mounted read-write "
2082                                         "without CONFIG_LSF.\n", sb->s_id);
2083                         goto failed_mount;
2084                 }
2085         }
2086         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2087
2088         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2089             blocksize > EXT4_MAX_BLOCK_SIZE) {
2090                 printk(KERN_ERR
2091                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2092                        blocksize, sb->s_id);
2093                 goto failed_mount;
2094         }
2095
2096         if (sb->s_blocksize != blocksize) {
2097
2098                 /* Validate the filesystem blocksize */
2099                 if (!sb_set_blocksize(sb, blocksize)) {
2100                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2101                                         blocksize);
2102                         goto failed_mount;
2103                 }
2104
2105                 brelse (bh);
2106                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2107                 offset = do_div(logical_sb_block, blocksize);
2108                 bh = sb_bread(sb, logical_sb_block);
2109                 if (!bh) {
2110                         printk(KERN_ERR
2111                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2112                         goto failed_mount;
2113                 }
2114                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2115                 sbi->s_es = es;
2116                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2117                         printk (KERN_ERR
2118                                 "EXT4-fs: Magic mismatch, very weird !\n");
2119                         goto failed_mount;
2120                 }
2121         }
2122
2123         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2124         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2125
2126         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2127                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2128                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2129         } else {
2130                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2131                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2132                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2133                     (!is_power_of_2(sbi->s_inode_size)) ||
2134                     (sbi->s_inode_size > blocksize)) {
2135                         printk (KERN_ERR
2136                                 "EXT4-fs: unsupported inode size: %d\n",
2137                                 sbi->s_inode_size);
2138                         goto failed_mount;
2139                 }
2140                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2141                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2142         }
2143         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2144         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2145                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2146                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2147                     !is_power_of_2(sbi->s_desc_size)) {
2148                         printk(KERN_ERR
2149                                "EXT4-fs: unsupported descriptor size %lu\n",
2150                                sbi->s_desc_size);
2151                         goto failed_mount;
2152                 }
2153         } else
2154                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2155         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2156         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2157         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2158                 goto cantfind_ext4;
2159         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2160         if (sbi->s_inodes_per_block == 0)
2161                 goto cantfind_ext4;
2162         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2163                                         sbi->s_inodes_per_block;
2164         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2165         sbi->s_sbh = bh;
2166         sbi->s_mount_state = le16_to_cpu(es->s_state);
2167         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2168         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2169         for (i=0; i < 4; i++)
2170                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2171         sbi->s_def_hash_version = es->s_def_hash_version;
2172
2173         if (sbi->s_blocks_per_group > blocksize * 8) {
2174                 printk (KERN_ERR
2175                         "EXT4-fs: #blocks per group too big: %lu\n",
2176                         sbi->s_blocks_per_group);
2177                 goto failed_mount;
2178         }
2179         if (sbi->s_inodes_per_group > blocksize * 8) {
2180                 printk (KERN_ERR
2181                         "EXT4-fs: #inodes per group too big: %lu\n",
2182                         sbi->s_inodes_per_group);
2183                 goto failed_mount;
2184         }
2185
2186         if (ext4_blocks_count(es) >
2187                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2188                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2189                         " too large to mount safely\n", sb->s_id);
2190                 if (sizeof(sector_t) < 8)
2191                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2192                                         "enabled\n");
2193                 goto failed_mount;
2194         }
2195
2196         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2197                 goto cantfind_ext4;
2198
2199         /* ensure blocks_count calculation below doesn't sign-extend */
2200         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2201             le32_to_cpu(es->s_first_data_block) + 1) {
2202                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2203                        "first data block %u, blocks per group %lu\n",
2204                         ext4_blocks_count(es),
2205                         le32_to_cpu(es->s_first_data_block),
2206                         EXT4_BLOCKS_PER_GROUP(sb));
2207                 goto failed_mount;
2208         }
2209         blocks_count = (ext4_blocks_count(es) -
2210                         le32_to_cpu(es->s_first_data_block) +
2211                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2212         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2213         sbi->s_groups_count = blocks_count;
2214         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2215                    EXT4_DESC_PER_BLOCK(sb);
2216         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2217                                     GFP_KERNEL);
2218         if (sbi->s_group_desc == NULL) {
2219                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2220                 goto failed_mount;
2221         }
2222
2223         bgl_lock_init(&sbi->s_blockgroup_lock);
2224
2225         for (i = 0; i < db_count; i++) {
2226                 block = descriptor_loc(sb, logical_sb_block, i);
2227                 sbi->s_group_desc[i] = sb_bread(sb, block);
2228                 if (!sbi->s_group_desc[i]) {
2229                         printk (KERN_ERR "EXT4-fs: "
2230                                 "can't read group descriptor %d\n", i);
2231                         db_count = i;
2232                         goto failed_mount2;
2233                 }
2234         }
2235         if (!ext4_check_descriptors (sb)) {
2236                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2237                 goto failed_mount2;
2238         }
2239         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2240                 if (!ext4_fill_flex_info(sb)) {
2241                         printk(KERN_ERR
2242                                "EXT4-fs: unable to initialize "
2243                                "flex_bg meta info!\n");
2244                         goto failed_mount2;
2245                 }
2246
2247         sbi->s_gdb_count = db_count;
2248         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2249         spin_lock_init(&sbi->s_next_gen_lock);
2250
2251         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2252                         ext4_count_free_blocks(sb));
2253         if (!err) {
2254                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2255                                 ext4_count_free_inodes(sb));
2256         }
2257         if (!err) {
2258                 err = percpu_counter_init(&sbi->s_dirs_counter,
2259                                 ext4_count_dirs(sb));
2260         }
2261         if (err) {
2262                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2263                 goto failed_mount3;
2264         }
2265
2266         /* per fileystem reservation list head & lock */
2267         spin_lock_init(&sbi->s_rsv_window_lock);
2268         sbi->s_rsv_window_root = RB_ROOT;
2269         /* Add a single, static dummy reservation to the start of the
2270          * reservation window list --- it gives us a placeholder for
2271          * append-at-start-of-list which makes the allocation logic
2272          * _much_ simpler. */
2273         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2274         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2275         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2276         sbi->s_rsv_window_head.rsv_goal_size = 0;
2277         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2278
2279         sbi->s_stripe = ext4_get_stripe_size(sbi);
2280
2281         /*
2282          * set up enough so that it can read an inode
2283          */
2284         sb->s_op = &ext4_sops;
2285         sb->s_export_op = &ext4_export_ops;
2286         sb->s_xattr = ext4_xattr_handlers;
2287 #ifdef CONFIG_QUOTA
2288         sb->s_qcop = &ext4_qctl_operations;
2289         sb->dq_op = &ext4_quota_operations;
2290 #endif
2291         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2292
2293         sb->s_root = NULL;
2294
2295         needs_recovery = (es->s_last_orphan != 0 ||
2296                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2297                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2298
2299         /*
2300          * The first inode we look at is the journal inode.  Don't try
2301          * root first: it may be modified in the journal!
2302          */
2303         if (!test_opt(sb, NOLOAD) &&
2304             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2305                 if (ext4_load_journal(sb, es, journal_devnum))
2306                         goto failed_mount3;
2307                 if (!(sb->s_flags & MS_RDONLY) &&
2308                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2309                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2310                                "ext4_fill_super: Journal transaction "
2311                                "%u is corrupt\n", sb->s_id,
2312                                EXT4_SB(sb)->s_journal->j_failed_commit);
2313                         if (test_opt (sb, ERRORS_RO)) {
2314                                 printk (KERN_CRIT
2315                                         "Mounting filesystem read-only\n");
2316                                 sb->s_flags |= MS_RDONLY;
2317                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2318                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2319                         }
2320                         if (test_opt(sb, ERRORS_PANIC)) {
2321                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2322                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2323                                 ext4_commit_super(sb, es, 1);
2324                                 printk(KERN_CRIT
2325                                        "EXT4-fs (device %s): mount failed\n",
2326                                       sb->s_id);
2327                                 goto failed_mount4;
2328                         }
2329                 }
2330         } else if (journal_inum) {
2331                 if (ext4_create_journal(sb, es, journal_inum))
2332                         goto failed_mount3;
2333         } else {
2334                 if (!silent)
2335                         printk (KERN_ERR
2336                                 "ext4: No journal on filesystem on %s\n",
2337                                 sb->s_id);
2338                 goto failed_mount3;
2339         }
2340
2341         if (ext4_blocks_count(es) > 0xffffffffULL &&
2342             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2343                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2344                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2345                 goto failed_mount4;
2346         }
2347
2348         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2349                 jbd2_journal_set_features(sbi->s_journal,
2350                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2351                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2352         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2353                 jbd2_journal_set_features(sbi->s_journal,
2354                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2355                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2356                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2357         } else {
2358                 jbd2_journal_clear_features(sbi->s_journal,
2359                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2360                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2361         }
2362
2363         /* We have now updated the journal if required, so we can
2364          * validate the data journaling mode. */
2365         switch (test_opt(sb, DATA_FLAGS)) {
2366         case 0:
2367                 /* No mode set, assume a default based on the journal
2368                  * capabilities: ORDERED_DATA if the journal can
2369                  * cope, else JOURNAL_DATA
2370                  */
2371                 if (jbd2_journal_check_available_features
2372                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2373                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2374                 else
2375                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2376                 break;
2377
2378         case EXT4_MOUNT_ORDERED_DATA:
2379         case EXT4_MOUNT_WRITEBACK_DATA:
2380                 if (!jbd2_journal_check_available_features
2381                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2382                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2383                                "requested data journaling mode\n");
2384                         goto failed_mount4;
2385                 }
2386         default:
2387                 break;
2388         }
2389
2390         if (test_opt(sb, NOBH)) {
2391                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2392                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2393                                 "its supported only with writeback mode\n");
2394                         clear_opt(sbi->s_mount_opt, NOBH);
2395                 }
2396         }
2397         /*
2398          * The jbd2_journal_load will have done any necessary log recovery,
2399          * so we can safely mount the rest of the filesystem now.
2400          */
2401
2402         root = ext4_iget(sb, EXT4_ROOT_INO);
2403         if (IS_ERR(root)) {
2404                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2405                 ret = PTR_ERR(root);
2406                 goto failed_mount4;
2407         }
2408         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2409                 iput(root);
2410                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2411                 goto failed_mount4;
2412         }
2413         sb->s_root = d_alloc_root(root);
2414         if (!sb->s_root) {
2415                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2416                 iput(root);
2417                 ret = -ENOMEM;
2418                 goto failed_mount4;
2419         }
2420
2421         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2422
2423         /* determine the minimum size of new large inodes, if present */
2424         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2425                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2426                                                      EXT4_GOOD_OLD_INODE_SIZE;
2427                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2428                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2429                         if (sbi->s_want_extra_isize <
2430                             le16_to_cpu(es->s_want_extra_isize))
2431                                 sbi->s_want_extra_isize =
2432                                         le16_to_cpu(es->s_want_extra_isize);
2433                         if (sbi->s_want_extra_isize <
2434                             le16_to_cpu(es->s_min_extra_isize))
2435                                 sbi->s_want_extra_isize =
2436                                         le16_to_cpu(es->s_min_extra_isize);
2437                 }
2438         }
2439         /* Check if enough inode space is available */
2440         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2441                                                         sbi->s_inode_size) {
2442                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2443                                                        EXT4_GOOD_OLD_INODE_SIZE;
2444                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2445                         "available.\n");
2446         }
2447
2448         /*
2449          * akpm: core read_super() calls in here with the superblock locked.
2450          * That deadlocks, because orphan cleanup needs to lock the superblock
2451          * in numerous places.  Here we just pop the lock - it's relatively
2452          * harmless, because we are now ready to accept write_super() requests,
2453          * and aviro says that's the only reason for hanging onto the
2454          * superblock lock.
2455          */
2456         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2457         ext4_orphan_cleanup(sb, es);
2458         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2459         if (needs_recovery)
2460                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2461         ext4_mark_recovery_complete(sb, es);
2462         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2463                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2464                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2465                 "writeback");
2466
2467         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2468                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2469                                 "requested data journaling mode\n");
2470                 clear_opt(sbi->s_mount_opt, DELALLOC);
2471         } else if (test_opt(sb, DELALLOC))
2472                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2473
2474         ext4_ext_init(sb);
2475         ext4_mb_init(sb, needs_recovery);
2476
2477         lock_kernel();
2478         return 0;
2479
2480 cantfind_ext4:
2481         if (!silent)
2482                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2483                        sb->s_id);
2484         goto failed_mount;
2485
2486 failed_mount4:
2487         jbd2_journal_destroy(sbi->s_journal);
2488         sbi->s_journal = NULL;
2489 failed_mount3:
2490         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2491         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2492         percpu_counter_destroy(&sbi->s_dirs_counter);
2493 failed_mount2:
2494         for (i = 0; i < db_count; i++)
2495                 brelse(sbi->s_group_desc[i]);
2496         kfree(sbi->s_group_desc);
2497 failed_mount:
2498 #ifdef CONFIG_QUOTA
2499         for (i = 0; i < MAXQUOTAS; i++)
2500                 kfree(sbi->s_qf_names[i]);
2501 #endif
2502         ext4_blkdev_remove(sbi);
2503         brelse(bh);
2504 out_fail:
2505         sb->s_fs_info = NULL;
2506         kfree(sbi);
2507         lock_kernel();
2508         return ret;
2509 }
2510
2511 /*
2512  * Setup any per-fs journal parameters now.  We'll do this both on
2513  * initial mount, once the journal has been initialised but before we've
2514  * done any recovery; and again on any subsequent remount.
2515  */
2516 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2517 {
2518         struct ext4_sb_info *sbi = EXT4_SB(sb);
2519
2520         if (sbi->s_commit_interval)
2521                 journal->j_commit_interval = sbi->s_commit_interval;
2522         /* We could also set up an ext4-specific default for the commit
2523          * interval here, but for now we'll just fall back to the jbd
2524          * default. */
2525
2526         spin_lock(&journal->j_state_lock);
2527         if (test_opt(sb, BARRIER))
2528                 journal->j_flags |= JBD2_BARRIER;
2529         else
2530                 journal->j_flags &= ~JBD2_BARRIER;
2531         spin_unlock(&journal->j_state_lock);
2532 }
2533
2534 static journal_t *ext4_get_journal(struct super_block *sb,
2535                                    unsigned int journal_inum)
2536 {
2537         struct inode *journal_inode;
2538         journal_t *journal;
2539
2540         /* First, test for the existence of a valid inode on disk.  Bad
2541          * things happen if we iget() an unused inode, as the subsequent
2542          * iput() will try to delete it. */
2543
2544         journal_inode = ext4_iget(sb, journal_inum);
2545         if (IS_ERR(journal_inode)) {
2546                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2547                 return NULL;
2548         }
2549         if (!journal_inode->i_nlink) {
2550                 make_bad_inode(journal_inode);
2551                 iput(journal_inode);
2552                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2553                 return NULL;
2554         }
2555
2556         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2557                   journal_inode, journal_inode->i_size);
2558         if (!S_ISREG(journal_inode->i_mode)) {
2559                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2560                 iput(journal_inode);
2561                 return NULL;
2562         }
2563
2564         journal = jbd2_journal_init_inode(journal_inode);
2565         if (!journal) {
2566                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2567                 iput(journal_inode);
2568                 return NULL;
2569         }
2570         journal->j_private = sb;
2571         ext4_init_journal_params(sb, journal);
2572         return journal;
2573 }
2574
2575 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2576                                        dev_t j_dev)
2577 {
2578         struct buffer_head * bh;
2579         journal_t *journal;
2580         ext4_fsblk_t start;
2581         ext4_fsblk_t len;
2582         int hblock, blocksize;
2583         ext4_fsblk_t sb_block;
2584         unsigned long offset;
2585         struct ext4_super_block * es;
2586         struct block_device *bdev;
2587
2588         bdev = ext4_blkdev_get(j_dev);
2589         if (bdev == NULL)
2590                 return NULL;
2591
2592         if (bd_claim(bdev, sb)) {
2593                 printk(KERN_ERR
2594                         "EXT4: failed to claim external journal device.\n");
2595                 blkdev_put(bdev);
2596                 return NULL;
2597         }
2598
2599         blocksize = sb->s_blocksize;
2600         hblock = bdev_hardsect_size(bdev);
2601         if (blocksize < hblock) {
2602                 printk(KERN_ERR
2603                         "EXT4-fs: blocksize too small for journal device.\n");
2604                 goto out_bdev;
2605         }
2606
2607         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2608         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2609         set_blocksize(bdev, blocksize);
2610         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2611                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2612                        "external journal\n");
2613                 goto out_bdev;
2614         }
2615
2616         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2617         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2618             !(le32_to_cpu(es->s_feature_incompat) &
2619               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2620                 printk(KERN_ERR "EXT4-fs: external journal has "
2621                                         "bad superblock\n");
2622                 brelse(bh);
2623                 goto out_bdev;
2624         }
2625
2626         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2627                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2628                 brelse(bh);
2629                 goto out_bdev;
2630         }
2631
2632         len = ext4_blocks_count(es);
2633         start = sb_block + 1;
2634         brelse(bh);     /* we're done with the superblock */
2635
2636         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2637                                         start, len, blocksize);
2638         if (!journal) {
2639                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2640                 goto out_bdev;
2641         }
2642         journal->j_private = sb;
2643         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2644         wait_on_buffer(journal->j_sb_buffer);
2645         if (!buffer_uptodate(journal->j_sb_buffer)) {
2646                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2647                 goto out_journal;
2648         }
2649         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2650                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2651                                         "user (unsupported) - %d\n",
2652                         be32_to_cpu(journal->j_superblock->s_nr_users));
2653                 goto out_journal;
2654         }
2655         EXT4_SB(sb)->journal_bdev = bdev;
2656         ext4_init_journal_params(sb, journal);
2657         return journal;
2658 out_journal:
2659         jbd2_journal_destroy(journal);
2660 out_bdev:
2661         ext4_blkdev_put(bdev);
2662         return NULL;
2663 }
2664
2665 static int ext4_load_journal(struct super_block *sb,
2666                              struct ext4_super_block *es,
2667                              unsigned long journal_devnum)
2668 {
2669         journal_t *journal;
2670         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2671         dev_t journal_dev;
2672         int err = 0;
2673         int really_read_only;
2674
2675         if (journal_devnum &&
2676             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2677                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2678                         "numbers have changed\n");
2679                 journal_dev = new_decode_dev(journal_devnum);
2680         } else
2681                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2682
2683         really_read_only = bdev_read_only(sb->s_bdev);
2684
2685         /*
2686          * Are we loading a blank journal or performing recovery after a
2687          * crash?  For recovery, we need to check in advance whether we
2688          * can get read-write access to the device.
2689          */
2690
2691         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2692                 if (sb->s_flags & MS_RDONLY) {
2693                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2694                                         "required on readonly filesystem.\n");
2695                         if (really_read_only) {
2696                                 printk(KERN_ERR "EXT4-fs: write access "
2697                                         "unavailable, cannot proceed.\n");
2698                                 return -EROFS;
2699                         }
2700                         printk (KERN_INFO "EXT4-fs: write access will "
2701                                         "be enabled during recovery.\n");
2702                 }
2703         }
2704
2705         if (journal_inum && journal_dev) {
2706                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2707                        "and inode journals!\n");
2708                 return -EINVAL;
2709         }
2710
2711         if (journal_inum) {
2712                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2713                         return -EINVAL;
2714         } else {
2715                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2716                         return -EINVAL;
2717         }
2718
2719         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2720                 err = jbd2_journal_update_format(journal);
2721                 if (err)  {
2722                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2723                         jbd2_journal_destroy(journal);
2724                         return err;
2725                 }
2726         }
2727
2728         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2729                 err = jbd2_journal_wipe(journal, !really_read_only);
2730         if (!err)
2731                 err = jbd2_journal_load(journal);
2732
2733         if (err) {
2734                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2735                 jbd2_journal_destroy(journal);
2736                 return err;
2737         }
2738
2739         EXT4_SB(sb)->s_journal = journal;
2740         ext4_clear_journal_err(sb, es);
2741
2742         if (journal_devnum &&
2743             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2744                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2745                 sb->s_dirt = 1;
2746
2747                 /* Make sure we flush the recovery flag to disk. */
2748                 ext4_commit_super(sb, es, 1);
2749         }
2750
2751         return 0;
2752 }
2753
2754 static int ext4_create_journal(struct super_block * sb,
2755                                struct ext4_super_block * es,
2756                                unsigned int journal_inum)
2757 {
2758         journal_t *journal;
2759         int err;
2760
2761         if (sb->s_flags & MS_RDONLY) {
2762                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2763                                 "create journal.\n");
2764                 return -EROFS;
2765         }
2766
2767         journal = ext4_get_journal(sb, journal_inum);
2768         if (!journal)
2769                 return -EINVAL;
2770
2771         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2772                journal_inum);
2773
2774         err = jbd2_journal_create(journal);
2775         if (err) {
2776                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2777                 jbd2_journal_destroy(journal);
2778                 return -EIO;
2779         }
2780
2781         EXT4_SB(sb)->s_journal = journal;
2782
2783         ext4_update_dynamic_rev(sb);
2784         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2785         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2786
2787         es->s_journal_inum = cpu_to_le32(journal_inum);
2788         sb->s_dirt = 1;
2789
2790         /* Make sure we flush the recovery flag to disk. */
2791         ext4_commit_super(sb, es, 1);
2792
2793         return 0;
2794 }
2795
2796 static void ext4_commit_super (struct super_block * sb,
2797                                struct ext4_super_block * es,
2798                                int sync)
2799 {
2800         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2801
2802         if (!sbh)
2803                 return;
2804         es->s_wtime = cpu_to_le32(get_seconds());
2805         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2806         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2807         BUFFER_TRACE(sbh, "marking dirty");
2808         mark_buffer_dirty(sbh);
2809         if (sync)
2810                 sync_dirty_buffer(sbh);
2811 }
2812
2813
2814 /*
2815  * Have we just finished recovery?  If so, and if we are mounting (or
2816  * remounting) the filesystem readonly, then we will end up with a
2817  * consistent fs on disk.  Record that fact.
2818  */
2819 static void ext4_mark_recovery_complete(struct super_block * sb,
2820                                         struct ext4_super_block * es)
2821 {
2822         journal_t *journal = EXT4_SB(sb)->s_journal;
2823
2824         jbd2_journal_lock_updates(journal);
2825         jbd2_journal_flush(journal);
2826         lock_super(sb);
2827         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2828             sb->s_flags & MS_RDONLY) {
2829                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2830                 sb->s_dirt = 0;
2831                 ext4_commit_super(sb, es, 1);
2832         }
2833         unlock_super(sb);
2834         jbd2_journal_unlock_updates(journal);
2835 }
2836
2837 /*
2838  * If we are mounting (or read-write remounting) a filesystem whose journal
2839  * has recorded an error from a previous lifetime, move that error to the
2840  * main filesystem now.
2841  */
2842 static void ext4_clear_journal_err(struct super_block * sb,
2843                                    struct ext4_super_block * es)
2844 {
2845         journal_t *journal;
2846         int j_errno;
2847         const char *errstr;
2848
2849         journal = EXT4_SB(sb)->s_journal;
2850
2851         /*
2852          * Now check for any error status which may have been recorded in the
2853          * journal by a prior ext4_error() or ext4_abort()
2854          */
2855
2856         j_errno = jbd2_journal_errno(journal);
2857         if (j_errno) {
2858                 char nbuf[16];
2859
2860                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2861                 ext4_warning(sb, __func__, "Filesystem error recorded "
2862                              "from previous mount: %s", errstr);
2863                 ext4_warning(sb, __func__, "Marking fs in need of "
2864                              "filesystem check.");
2865
2866                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2867                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2868                 ext4_commit_super (sb, es, 1);
2869
2870                 jbd2_journal_clear_err(journal);
2871         }
2872 }
2873
2874 /*
2875  * Force the running and committing transactions to commit,
2876  * and wait on the commit.
2877  */
2878 int ext4_force_commit(struct super_block *sb)
2879 {
2880         journal_t *journal;
2881         int ret;
2882
2883         if (sb->s_flags & MS_RDONLY)
2884                 return 0;
2885
2886         journal = EXT4_SB(sb)->s_journal;
2887         sb->s_dirt = 0;
2888         ret = ext4_journal_force_commit(journal);
2889         return ret;
2890 }
2891
2892 /*
2893  * Ext4 always journals updates to the superblock itself, so we don't
2894  * have to propagate any other updates to the superblock on disk at this
2895  * point.  Just start an async writeback to get the buffers on their way
2896  * to the disk.
2897  *
2898  * This implicitly triggers the writebehind on sync().
2899  */
2900
2901 static void ext4_write_super (struct super_block * sb)
2902 {
2903         if (mutex_trylock(&sb->s_lock) != 0)
2904                 BUG();
2905         sb->s_dirt = 0;
2906 }
2907
2908 static int ext4_sync_fs(struct super_block *sb, int wait)
2909 {
2910         tid_t target;
2911
2912         sb->s_dirt = 0;
2913         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2914                 if (wait)
2915                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2916         }
2917         return 0;
2918 }
2919
2920 /*
2921  * LVM calls this function before a (read-only) snapshot is created.  This
2922  * gives us a chance to flush the journal completely and mark the fs clean.
2923  */
2924 static void ext4_write_super_lockfs(struct super_block *sb)
2925 {
2926         sb->s_dirt = 0;
2927
2928         if (!(sb->s_flags & MS_RDONLY)) {
2929                 journal_t *journal = EXT4_SB(sb)->s_journal;
2930
2931                 /* Now we set up the journal barrier. */
2932                 jbd2_journal_lock_updates(journal);
2933                 jbd2_journal_flush(journal);
2934
2935                 /* Journal blocked and flushed, clear needs_recovery flag. */
2936                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2937                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2938         }
2939 }
2940
2941 /*
2942  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2943  * flag here, even though the filesystem is not technically dirty yet.
2944  */
2945 static void ext4_unlockfs(struct super_block *sb)
2946 {
2947         if (!(sb->s_flags & MS_RDONLY)) {
2948                 lock_super(sb);
2949                 /* Reser the needs_recovery flag before the fs is unlocked. */
2950                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2951                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2952                 unlock_super(sb);
2953                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2954         }
2955 }
2956
2957 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2958 {
2959         struct ext4_super_block * es;
2960         struct ext4_sb_info *sbi = EXT4_SB(sb);
2961         ext4_fsblk_t n_blocks_count = 0;
2962         unsigned long old_sb_flags;
2963         struct ext4_mount_options old_opts;
2964         int err;
2965 #ifdef CONFIG_QUOTA
2966         int i;
2967 #endif
2968
2969         /* Store the original options */
2970         old_sb_flags = sb->s_flags;
2971         old_opts.s_mount_opt = sbi->s_mount_opt;
2972         old_opts.s_resuid = sbi->s_resuid;
2973         old_opts.s_resgid = sbi->s_resgid;
2974         old_opts.s_commit_interval = sbi->s_commit_interval;
2975 #ifdef CONFIG_QUOTA
2976         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2977         for (i = 0; i < MAXQUOTAS; i++)
2978                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2979 #endif
2980
2981         /*
2982          * Allow the "check" option to be passed as a remount option.
2983          */
2984         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2985                 err = -EINVAL;
2986                 goto restore_opts;
2987         }
2988
2989         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2990                 ext4_abort(sb, __func__, "Abort forced by user");
2991
2992         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2993                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2994
2995         es = sbi->s_es;
2996
2997         ext4_init_journal_params(sb, sbi->s_journal);
2998
2999         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3000                 n_blocks_count > ext4_blocks_count(es)) {
3001                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3002                         err = -EROFS;
3003                         goto restore_opts;
3004                 }
3005
3006                 if (*flags & MS_RDONLY) {
3007                         /*
3008                          * First of all, the unconditional stuff we have to do
3009                          * to disable replay of the journal when we next remount
3010                          */
3011                         sb->s_flags |= MS_RDONLY;
3012
3013                         /*
3014                          * OK, test if we are remounting a valid rw partition
3015                          * readonly, and if so set the rdonly flag and then
3016                          * mark the partition as valid again.
3017                          */
3018                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3019                             (sbi->s_mount_state & EXT4_VALID_FS))
3020                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3021
3022                         /*
3023                          * We have to unlock super so that we can wait for
3024                          * transactions.
3025                          */
3026                         unlock_super(sb);
3027                         ext4_mark_recovery_complete(sb, es);
3028                         lock_super(sb);
3029                 } else {
3030                         __le32 ret;
3031                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3032                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3033                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3034                                        "remount RDWR because of unsupported "
3035                                        "optional features (%x).\n",
3036                                        sb->s_id, le32_to_cpu(ret));
3037                                 err = -EROFS;
3038                                 goto restore_opts;
3039                         }
3040
3041                         /*
3042                          * If we have an unprocessed orphan list hanging
3043                          * around from a previously readonly bdev mount,
3044                          * require a full umount/remount for now.
3045                          */
3046                         if (es->s_last_orphan) {
3047                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3048                                        "remount RDWR because of unprocessed "
3049                                        "orphan inode list.  Please "
3050                                        "umount/remount instead.\n",
3051                                        sb->s_id);
3052                                 err = -EINVAL;
3053                                 goto restore_opts;
3054                         }
3055
3056                         /*
3057                          * Mounting a RDONLY partition read-write, so reread
3058                          * and store the current valid flag.  (It may have
3059                          * been changed by e2fsck since we originally mounted
3060                          * the partition.)
3061                          */
3062                         ext4_clear_journal_err(sb, es);
3063                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3064                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3065                                 goto restore_opts;
3066                         if (!ext4_setup_super (sb, es, 0))
3067                                 sb->s_flags &= ~MS_RDONLY;
3068                 }
3069         }
3070 #ifdef CONFIG_QUOTA
3071         /* Release old quota file names */
3072         for (i = 0; i < MAXQUOTAS; i++)
3073                 if (old_opts.s_qf_names[i] &&
3074                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3075                         kfree(old_opts.s_qf_names[i]);
3076 #endif
3077         return 0;
3078 restore_opts:
3079         sb->s_flags = old_sb_flags;
3080         sbi->s_mount_opt = old_opts.s_mount_opt;
3081         sbi->s_resuid = old_opts.s_resuid;
3082         sbi->s_resgid = old_opts.s_resgid;
3083         sbi->s_commit_interval = old_opts.s_commit_interval;
3084 #ifdef CONFIG_QUOTA
3085         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3086         for (i = 0; i < MAXQUOTAS; i++) {
3087                 if (sbi->s_qf_names[i] &&
3088                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3089                         kfree(sbi->s_qf_names[i]);
3090                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3091         }
3092 #endif
3093         return err;
3094 }
3095
3096 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
3097 {
3098         struct super_block *sb = dentry->d_sb;
3099         struct ext4_sb_info *sbi = EXT4_SB(sb);
3100         struct ext4_super_block *es = sbi->s_es;
3101         u64 fsid;
3102
3103         if (test_opt(sb, MINIX_DF)) {
3104                 sbi->s_overhead_last = 0;
3105         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3106                 ext4_group_t ngroups = sbi->s_groups_count, i;
3107                 ext4_fsblk_t overhead = 0;
3108                 smp_rmb();
3109
3110                 /*
3111                  * Compute the overhead (FS structures).  This is constant
3112                  * for a given filesystem unless the number of block groups
3113                  * changes so we cache the previous value until it does.
3114                  */
3115
3116                 /*
3117                  * All of the blocks before first_data_block are
3118                  * overhead
3119                  */
3120                 overhead = le32_to_cpu(es->s_first_data_block);
3121
3122                 /*
3123                  * Add the overhead attributed to the superblock and
3124                  * block group descriptors.  If the sparse superblocks
3125                  * feature is turned on, then not all groups have this.
3126                  */
3127                 for (i = 0; i < ngroups; i++) {
3128                         overhead += ext4_bg_has_super(sb, i) +
3129                                 ext4_bg_num_gdb(sb, i);
3130                         cond_resched();
3131                 }
3132
3133                 /*
3134                  * Every block group has an inode bitmap, a block
3135                  * bitmap, and an inode table.
3136                  */
3137                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3138                 sbi->s_overhead_last = overhead;
3139                 smp_wmb();
3140                 sbi->s_blocks_last = ext4_blocks_count(es);
3141         }
3142
3143         buf->f_type = EXT4_SUPER_MAGIC;
3144         buf->f_bsize = sb->s_blocksize;
3145         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3146         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
3147         ext4_free_blocks_count_set(es, buf->f_bfree);
3148         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3149         if (buf->f_bfree < ext4_r_blocks_count(es))
3150                 buf->f_bavail = 0;
3151         buf->f_files = le32_to_cpu(es->s_inodes_count);
3152         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3153         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3154         buf->f_namelen = EXT4_NAME_LEN;
3155         fsid = le64_to_cpup((void *)es->s_uuid) ^
3156                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3157         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3158         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3159         return 0;
3160 }
3161
3162 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3163  * is locked for write. Otherwise the are possible deadlocks:
3164  * Process 1                         Process 2
3165  * ext4_create()                     quota_sync()
3166  *   jbd2_journal_start()                   write_dquot()
3167  *   DQUOT_INIT()                        down(dqio_mutex)
3168  *     down(dqio_mutex)                    jbd2_journal_start()
3169  *
3170  */
3171
3172 #ifdef CONFIG_QUOTA
3173
3174 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3175 {
3176         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3177 }
3178
3179 static int ext4_dquot_initialize(struct inode *inode, int type)
3180 {
3181         handle_t *handle;
3182         int ret, err;
3183
3184         /* We may create quota structure so we need to reserve enough blocks */
3185         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3186         if (IS_ERR(handle))
3187                 return PTR_ERR(handle);
3188         ret = dquot_initialize(inode, type);
3189         err = ext4_journal_stop(handle);
3190         if (!ret)
3191                 ret = err;
3192         return ret;
3193 }
3194
3195 static int ext4_dquot_drop(struct inode *inode)
3196 {
3197         handle_t *handle;
3198         int ret, err;
3199
3200         /* We may delete quota structure so we need to reserve enough blocks */
3201         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3202         if (IS_ERR(handle)) {
3203                 /*
3204                  * We call dquot_drop() anyway to at least release references
3205                  * to quota structures so that umount does not hang.
3206                  */
3207                 dquot_drop(inode);
3208                 return PTR_ERR(handle);
3209         }
3210         ret = dquot_drop(inode);
3211         err = ext4_journal_stop(handle);
3212         if (!ret)
3213                 ret = err;
3214         return ret;
3215 }
3216
3217 static int ext4_write_dquot(struct dquot *dquot)
3218 {
3219         int ret, err;
3220         handle_t *handle;
3221         struct inode *inode;
3222
3223         inode = dquot_to_inode(dquot);
3224         handle = ext4_journal_start(inode,
3225                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3226         if (IS_ERR(handle))
3227                 return PTR_ERR(handle);
3228         ret = dquot_commit(dquot);
3229         err = ext4_journal_stop(handle);
3230         if (!ret)
3231                 ret = err;
3232         return ret;
3233 }
3234
3235 static int ext4_acquire_dquot(struct dquot *dquot)
3236 {
3237         int ret, err;
3238         handle_t *handle;
3239
3240         handle = ext4_journal_start(dquot_to_inode(dquot),
3241                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3242         if (IS_ERR(handle))
3243                 return PTR_ERR(handle);
3244         ret = dquot_acquire(dquot);
3245         err = ext4_journal_stop(handle);
3246         if (!ret)
3247                 ret = err;
3248         return ret;
3249 }
3250
3251 static int ext4_release_dquot(struct dquot *dquot)
3252 {
3253         int ret, err;
3254         handle_t *handle;
3255
3256         handle = ext4_journal_start(dquot_to_inode(dquot),
3257                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3258         if (IS_ERR(handle)) {
3259                 /* Release dquot anyway to avoid endless cycle in dqput() */
3260                 dquot_release(dquot);
3261                 return PTR_ERR(handle);
3262         }
3263         ret = dquot_release(dquot);
3264         err = ext4_journal_stop(handle);
3265         if (!ret)
3266                 ret = err;
3267         return ret;
3268 }
3269
3270 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3271 {
3272         /* Are we journaling quotas? */
3273         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3274             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3275                 dquot_mark_dquot_dirty(dquot);
3276                 return ext4_write_dquot(dquot);
3277         } else {
3278                 return dquot_mark_dquot_dirty(dquot);
3279         }
3280 }
3281
3282 static int ext4_write_info(struct super_block *sb, int type)
3283 {
3284         int ret, err;
3285         handle_t *handle;
3286
3287         /* Data block + inode block */
3288         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3289         if (IS_ERR(handle))
3290                 return PTR_ERR(handle);
3291         ret = dquot_commit_info(sb, type);
3292         err = ext4_journal_stop(handle);
3293         if (!ret)
3294                 ret = err;
3295         return ret;
3296 }
3297
3298 /*
3299  * Turn on quotas during mount time - we need to find
3300  * the quota file and such...
3301  */
3302 static int ext4_quota_on_mount(struct super_block *sb, int type)
3303 {
3304         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3305                         EXT4_SB(sb)->s_jquota_fmt, type);
3306 }
3307
3308 /*
3309  * Standard function to be called on quota_on
3310  */
3311 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3312                          char *path, int remount)
3313 {
3314         int err;
3315         struct nameidata nd;
3316
3317         if (!test_opt(sb, QUOTA))
3318                 return -EINVAL;
3319         /* When remounting, no checks are needed and in fact, path is NULL */
3320         if (remount)
3321                 return vfs_quota_on(sb, type, format_id, path, remount);
3322
3323         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3324         if (err)
3325                 return err;
3326
3327         /* Quotafile not on the same filesystem? */
3328         if (nd.path.mnt->mnt_sb != sb) {
3329                 path_put(&nd.path);
3330                 return -EXDEV;
3331         }
3332         /* Journaling quota? */
3333         if (EXT4_SB(sb)->s_qf_names[type]) {
3334                 /* Quotafile not of fs root? */
3335                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3336                         printk(KERN_WARNING
3337                                 "EXT4-fs: Quota file not on filesystem root. "
3338                                 "Journaled quota will not work.\n");
3339         }
3340
3341         /*
3342          * When we journal data on quota file, we have to flush journal to see
3343          * all updates to the file when we bypass pagecache...
3344          */
3345         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3346                 /*
3347                  * We don't need to lock updates but journal_flush() could
3348                  * otherwise be livelocked...
3349                  */
3350                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3351                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3352                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3353         }
3354
3355         path_put(&nd.path);
3356         return vfs_quota_on(sb, type, format_id, path, remount);
3357 }
3358
3359 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3360  * acquiring the locks... As quota files are never truncated and quota code
3361  * itself serializes the operations (and noone else should touch the files)
3362  * we don't have to be afraid of races */
3363 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3364                                size_t len, loff_t off)
3365 {
3366         struct inode *inode = sb_dqopt(sb)->files[type];
3367         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3368         int err = 0;
3369         int offset = off & (sb->s_blocksize - 1);
3370         int tocopy;
3371         size_t toread;
3372         struct buffer_head *bh;
3373         loff_t i_size = i_size_read(inode);
3374
3375         if (off > i_size)
3376                 return 0;
3377         if (off+len > i_size)
3378                 len = i_size-off;
3379         toread = len;
3380         while (toread > 0) {
3381                 tocopy = sb->s_blocksize - offset < toread ?
3382                                 sb->s_blocksize - offset : toread;
3383                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3384                 if (err)
3385                         return err;
3386                 if (!bh)        /* A hole? */
3387                         memset(data, 0, tocopy);
3388                 else
3389                         memcpy(data, bh->b_data+offset, tocopy);
3390                 brelse(bh);
3391                 offset = 0;
3392                 toread -= tocopy;
3393                 data += tocopy;
3394                 blk++;
3395         }
3396         return len;
3397 }
3398
3399 /* Write to quotafile (we know the transaction is already started and has
3400  * enough credits) */
3401 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3402                                 const char *data, size_t len, loff_t off)
3403 {
3404         struct inode *inode = sb_dqopt(sb)->files[type];
3405         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3406         int err = 0;
3407         int offset = off & (sb->s_blocksize - 1);
3408         int tocopy;
3409         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3410         size_t towrite = len;
3411         struct buffer_head *bh;
3412         handle_t *handle = journal_current_handle();
3413
3414         if (!handle) {
3415                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3416                         " cancelled because transaction is not started.\n",
3417                         (unsigned long long)off, (unsigned long long)len);
3418                 return -EIO;
3419         }
3420         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3421         while (towrite > 0) {
3422                 tocopy = sb->s_blocksize - offset < towrite ?
3423                                 sb->s_blocksize - offset : towrite;
3424                 bh = ext4_bread(handle, inode, blk, 1, &err);
3425                 if (!bh)
3426                         goto out;
3427                 if (journal_quota) {
3428                         err = ext4_journal_get_write_access(handle, bh);
3429                         if (err) {
3430                                 brelse(bh);
3431                                 goto out;
3432                         }
3433                 }
3434                 lock_buffer(bh);
3435                 memcpy(bh->b_data+offset, data, tocopy);
3436                 flush_dcache_page(bh->b_page);
3437                 unlock_buffer(bh);
3438                 if (journal_quota)
3439                         err = ext4_journal_dirty_metadata(handle, bh);
3440                 else {
3441                         /* Always do at least ordered writes for quotas */
3442                         err = ext4_jbd2_file_inode(handle, inode);
3443                         mark_buffer_dirty(bh);
3444                 }
3445                 brelse(bh);
3446                 if (err)
3447                         goto out;
3448                 offset = 0;
3449                 towrite -= tocopy;
3450                 data += tocopy;
3451                 blk++;
3452         }
3453 out:
3454         if (len == towrite) {
3455                 mutex_unlock(&inode->i_mutex);
3456                 return err;
3457         }
3458         if (inode->i_size < off+len-towrite) {
3459                 i_size_write(inode, off+len-towrite);
3460                 EXT4_I(inode)->i_disksize = inode->i_size;
3461         }
3462         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3463         ext4_mark_inode_dirty(handle, inode);
3464         mutex_unlock(&inode->i_mutex);
3465         return len - towrite;
3466 }
3467
3468 #endif
3469
3470 static int ext4_get_sb(struct file_system_type *fs_type,
3471         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3472 {
3473         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3474 }
3475
3476 static struct file_system_type ext4dev_fs_type = {
3477         .owner          = THIS_MODULE,
3478         .name           = "ext4dev",
3479         .get_sb         = ext4_get_sb,
3480         .kill_sb        = kill_block_super,
3481         .fs_flags       = FS_REQUIRES_DEV,
3482 };
3483
3484 static int __init init_ext4_fs(void)
3485 {
3486         int err;
3487
3488         err = init_ext4_mballoc();
3489         if (err)
3490                 return err;
3491
3492         err = init_ext4_xattr();
3493         if (err)
3494                 goto out2;
3495         err = init_inodecache();
3496         if (err)
3497                 goto out1;
3498         err = register_filesystem(&ext4dev_fs_type);
3499         if (err)
3500                 goto out;
3501         return 0;
3502 out:
3503         destroy_inodecache();
3504 out1:
3505         exit_ext4_xattr();
3506 out2:
3507         exit_ext4_mballoc();
3508         return err;
3509 }
3510
3511 static void __exit exit_ext4_fs(void)
3512 {
3513         unregister_filesystem(&ext4dev_fs_type);
3514         destroy_inodecache();
3515         exit_ext4_xattr();
3516         exit_ext4_mballoc();
3517 }
3518
3519 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3520 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3521 MODULE_LICENSE("GPL");
3522 module_init(init_ext4_fs)
3523 module_exit(exit_ext4_fs)