fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  50                              unsigned long journal_devnum);
  51 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  52                                unsigned int);
  53 static void ext4_commit_super (struct super_block * sb,
  54                                struct ext4_super_block * es,
  55                                int sync);
  56 static void ext4_mark_recovery_complete(struct super_block * sb,
  57                                         struct ext4_super_block * es);
  58 static void ext4_clear_journal_err(struct super_block * sb,
  59                                    struct ext4_super_block * es);
  60 static int ext4_sync_fs(struct super_block *sb, int wait);
  61 static const char *ext4_decode_error(struct super_block * sb, int errno,
  62                                      char nbuf[16]);
  63 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  64 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  65 static void ext4_unlockfs(struct super_block *sb);
  66 static void ext4_write_super (struct super_block * sb);
  67 static void ext4_write_super_lockfs(struct super_block *sb);
  68
  69
  70 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  71                                struct ext4_group_desc *bg)
  72 {
  73         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  74                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  75                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  76 }
  77
  78 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  79                                struct ext4_group_desc *bg)
  80 {
  81         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  82                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  83                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  84 }
  85
  86 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  87                               struct ext4_group_desc *bg)
  88 {
  89         return le32_to_cpu(bg->bg_inode_table_lo) |
  90                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  91                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  92 }
  93
  94 void ext4_block_bitmap_set(struct super_block *sb,
  95                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  96 {
  97         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  98         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  99                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 100 }
 101
 102 void ext4_inode_bitmap_set(struct super_block *sb,
 103                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 104 {
 105         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 106         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 107                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 108 }
 109
 110 void ext4_inode_table_set(struct super_block *sb,
 111                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 112 {
 113         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 114         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 115                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 116 }
 117
 118 /*
 119  * Wrappers for jbd2_journal_start/end.
 120  *
 121  * The only special thing we need to do here is to make sure that all
 122  * journal_end calls result in the superblock being marked dirty, so
 123  * that sync() will call the filesystem's write_super callback if
 124  * appropriate.
 125  */
 126 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 127 {
 128         journal_t *journal;
 129
 130         if (sb->s_flags & MS_RDONLY)
 131                 return ERR_PTR(-EROFS);
 132
 133         /* Special case here: if the journal has aborted behind our
 134          * backs (eg. EIO in the commit thread), then we still need to
 135          * take the FS itself readonly cleanly. */
 136         journal = EXT4_SB(sb)->s_journal;
 137         if (is_journal_aborted(journal)) {
 138                 ext4_abort(sb, __FUNCTION__,
 139                            "Detected aborted journal");
 140                 return ERR_PTR(-EROFS);
 141         }
 142
 143         return jbd2_journal_start(journal, nblocks);
 144 }
 145
 146 /*
 147  * The only special thing we need to do here is to make sure that all
 148  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 149  * that sync() will call the filesystem's write_super callback if
 150  * appropriate.
 151  */
 152 int __ext4_journal_stop(const char *where, handle_t *handle)
 153 {
 154         struct super_block *sb;
 155         int err;
 156         int rc;
 157
 158         sb = handle->h_transaction->t_journal->j_private;
 159         err = handle->h_err;
 160         rc = jbd2_journal_stop(handle);
 161
 162         if (!err)
 163                 err = rc;
 164         if (err)
 165                 __ext4_std_error(sb, where, err);
 166         return err;
 167 }
 168
 169 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 170                 struct buffer_head *bh, handle_t *handle, int err)
 171 {
 172         char nbuf[16];
 173         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 174
 175         if (bh)
 176                 BUFFER_TRACE(bh, "abort");
 177
 178         if (!handle->h_err)
 179                 handle->h_err = err;
 180
 181         if (is_handle_aborted(handle))
 182                 return;
 183
 184         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 185                caller, errstr, err_fn);
 186
 187         jbd2_journal_abort_handle(handle);
 188 }
 189
 190 /* Deal with the reporting of failure conditions on a filesystem such as
 191  * inconsistencies detected or read IO failures.
 192  *
 193  * On ext2, we can store the error state of the filesystem in the
 194  * superblock.  That is not possible on ext4, because we may have other
 195  * write ordering constraints on the superblock which prevent us from
 196  * writing it out straight away; and given that the journal is about to
 197  * be aborted, we can't rely on the current, or future, transactions to
 198  * write out the superblock safely.
 199  *
 200  * We'll just use the jbd2_journal_abort() error code to record an error in
 201  * the journal instead.  On recovery, the journal will compain about
 202  * that error until we've noted it down and cleared it.
 203  */
 204
 205 static void ext4_handle_error(struct super_block *sb)
 206 {
 207         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 208
 209         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 210         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 211
 212         if (sb->s_flags & MS_RDONLY)
 213                 return;
 214
 215         if (!test_opt (sb, ERRORS_CONT)) {
 216                 journal_t *journal = EXT4_SB(sb)->s_journal;
 217
 218                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 219                 if (journal)
 220                         jbd2_journal_abort(journal, -EIO);
 221         }
 222         if (test_opt (sb, ERRORS_RO)) {
 223                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 224                 sb->s_flags |= MS_RDONLY;
 225         }
 226         ext4_commit_super(sb, es, 1);
 227         if (test_opt(sb, ERRORS_PANIC))
 228                 panic("EXT4-fs (device %s): panic forced after error\n",
 229                         sb->s_id);
 230 }
 231
 232 void ext4_error (struct super_block * sb, const char * function,
 233                  const char * fmt, ...)
 234 {
 235         va_list args;
 236
 237         va_start(args, fmt);
 238         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 239         vprintk(fmt, args);
 240         printk("\n");
 241         va_end(args);
 242
 243         ext4_handle_error(sb);
 244 }
 245
 246 static const char *ext4_decode_error(struct super_block * sb, int errno,
 247                                      char nbuf[16])
 248 {
 249         char *errstr = NULL;
 250
 251         switch (errno) {
 252         case -EIO:
 253                 errstr = "IO failure";
 254                 break;
 255         case -ENOMEM:
 256                 errstr = "Out of memory";
 257                 break;
 258         case -EROFS:
 259                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 260                         errstr = "Journal has aborted";
 261                 else
 262                         errstr = "Readonly filesystem";
 263                 break;
 264         default:
 265                 /* If the caller passed in an extra buffer for unknown
 266                  * errors, textualise them now.  Else we just return
 267                  * NULL. */
 268                 if (nbuf) {
 269                         /* Check for truncated error codes... */
 270                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 271                                 errstr = nbuf;
 272                 }
 273                 break;
 274         }
 275
 276         return errstr;
 277 }
 278
 279 /* __ext4_std_error decodes expected errors from journaling functions
 280  * automatically and invokes the appropriate error response.  */
 281
 282 void __ext4_std_error (struct super_block * sb, const char * function,
 283                        int errno)
 284 {
 285         char nbuf[16];
 286         const char *errstr;
 287
 288         /* Special case: if the error is EROFS, and we're not already
 289          * inside a transaction, then there's really no point in logging
 290          * an error. */
 291         if (errno == -EROFS && journal_current_handle() == NULL &&
 292             (sb->s_flags & MS_RDONLY))
 293                 return;
 294
 295         errstr = ext4_decode_error(sb, errno, nbuf);
 296         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 297                 sb->s_id, function, errstr);
 298
 299         ext4_handle_error(sb);
 300 }
 301
 302 /*
 303  * ext4_abort is a much stronger failure handler than ext4_error.  The
 304  * abort function may be used to deal with unrecoverable failures such
 305  * as journal IO errors or ENOMEM at a critical moment in log management.
 306  *
 307  * We unconditionally force the filesystem into an ABORT|READONLY state,
 308  * unless the error response on the fs has been set to panic in which
 309  * case we take the easy way out and panic immediately.
 310  */
 311
 312 void ext4_abort (struct super_block * sb, const char * function,
 313                  const char * fmt, ...)
 314 {
 315         va_list args;
 316
 317         printk (KERN_CRIT "ext4_abort called.\n");
 318
 319         va_start(args, fmt);
 320         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 321         vprintk(fmt, args);
 322         printk("\n");
 323         va_end(args);
 324
 325         if (test_opt(sb, ERRORS_PANIC))
 326                 panic("EXT4-fs panic from previous error\n");
 327
 328         if (sb->s_flags & MS_RDONLY)
 329                 return;
 330
 331         printk(KERN_CRIT "Remounting filesystem read-only\n");
 332         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 333         sb->s_flags |= MS_RDONLY;
 334         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 335         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 336 }
 337
 338 void ext4_warning (struct super_block * sb, const char * function,
 339                    const char * fmt, ...)
 340 {
 341         va_list args;
 342
 343         va_start(args, fmt);
 344         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 345                sb->s_id, function);
 346         vprintk(fmt, args);
 347         printk("\n");
 348         va_end(args);
 349 }
 350
 351 void ext4_update_dynamic_rev(struct super_block *sb)
 352 {
 353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 354
 355         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 356                 return;
 357
 358         ext4_warning(sb, __FUNCTION__,
 359                      "updating to rev %d because of new feature flag, "
 360                      "running e2fsck is recommended",
 361                      EXT4_DYNAMIC_REV);
 362
 363         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 364         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 365         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 366         /* leave es->s_feature_*compat flags alone */
 367         /* es->s_uuid will be set by e2fsck if empty */
 368
 369         /*
 370          * The rest of the superblock fields should be zero, and if not it
 371          * means they are likely already in use, so leave them alone.  We
 372          * can leave it up to e2fsck to clean up any inconsistencies there.
 373          */
 374 }
 375
 376 int ext4_update_compat_feature(handle_t *handle,
 377                                         struct super_block *sb, __u32 compat)
 378 {
 379         int err = 0;
 380         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 381                 err = ext4_journal_get_write_access(handle,
 382                                 EXT4_SB(sb)->s_sbh);
 383                 if (err)
 384                         return err;
 385                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 386                 sb->s_dirt = 1;
 387                 handle->h_sync = 1;
 388                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 389                                         "call ext4_journal_dirty_met adata");
 390                 err = ext4_journal_dirty_metadata(handle,
 391                                 EXT4_SB(sb)->s_sbh);
 392         }
 393         return err;
 394 }
 395
 396 int ext4_update_rocompat_feature(handle_t *handle,
 397                                         struct super_block *sb, __u32 rocompat)
 398 {
 399         int err = 0;
 400         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 401                 err = ext4_journal_get_write_access(handle,
 402                                 EXT4_SB(sb)->s_sbh);
 403                 if (err)
 404                         return err;
 405                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 406                 sb->s_dirt = 1;
 407                 handle->h_sync = 1;
 408                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 409                                         "call ext4_journal_dirty_met adata");
 410                 err = ext4_journal_dirty_metadata(handle,
 411                                 EXT4_SB(sb)->s_sbh);
 412         }
 413         return err;
 414 }
 415
 416 int ext4_update_incompat_feature(handle_t *handle,
 417                                         struct super_block *sb, __u32 incompat)
 418 {
 419         int err = 0;
 420         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 421                 err = ext4_journal_get_write_access(handle,
 422                                 EXT4_SB(sb)->s_sbh);
 423                 if (err)
 424                         return err;
 425                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 426                 sb->s_dirt = 1;
 427                 handle->h_sync = 1;
 428                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 429                                         "call ext4_journal_dirty_met adata");
 430                 err = ext4_journal_dirty_metadata(handle,
 431                                 EXT4_SB(sb)->s_sbh);
 432         }
 433         return err;
 434 }
 435
 436 /*
 437  * Open the external journal device
 438  */
 439 static struct block_device *ext4_blkdev_get(dev_t dev)
 440 {
 441         struct block_device *bdev;
 442         char b[BDEVNAME_SIZE];
 443
 444         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 445         if (IS_ERR(bdev))
 446                 goto fail;
 447         return bdev;
 448
 449 fail:
 450         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 451                         __bdevname(dev, b), PTR_ERR(bdev));
 452         return NULL;
 453 }
 454
 455 /*
 456  * Release the journal device
 457  */
 458 static int ext4_blkdev_put(struct block_device *bdev)
 459 {
 460         bd_release(bdev);
 461         return blkdev_put(bdev);
 462 }
 463
 464 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 465 {
 466         struct block_device *bdev;
 467         int ret = -ENODEV;
 468
 469         bdev = sbi->journal_bdev;
 470         if (bdev) {
 471                 ret = ext4_blkdev_put(bdev);
 472                 sbi->journal_bdev = NULL;
 473         }
 474         return ret;
 475 }
 476
 477 static inline struct inode *orphan_list_entry(struct list_head *l)
 478 {
 479         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 480 }
 481
 482 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 483 {
 484         struct list_head *l;
 485
 486         printk(KERN_ERR "sb orphan head is %d\n",
 487                le32_to_cpu(sbi->s_es->s_last_orphan));
 488
 489         printk(KERN_ERR "sb_info orphan list:\n");
 490         list_for_each(l, &sbi->s_orphan) {
 491                 struct inode *inode = orphan_list_entry(l);
 492                 printk(KERN_ERR "  "
 493                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 494                        inode->i_sb->s_id, inode->i_ino, inode,
 495                        inode->i_mode, inode->i_nlink,
 496                        NEXT_ORPHAN(inode));
 497         }
 498 }
 499
 500 static void ext4_put_super (struct super_block * sb)
 501 {
 502         struct ext4_sb_info *sbi = EXT4_SB(sb);
 503         struct ext4_super_block *es = sbi->s_es;
 504         int i;
 505
 506         ext4_mb_release(sb);
 507         ext4_ext_release(sb);
 508         ext4_xattr_put_super(sb);
 509         jbd2_journal_destroy(sbi->s_journal);
 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         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 522         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 523         percpu_counter_destroy(&sbi->s_dirs_counter);
 524         brelse(sbi->s_sbh);
 525 #ifdef CONFIG_QUOTA
 526         for (i = 0; i < MAXQUOTAS; i++)
 527                 kfree(sbi->s_qf_names[i]);
 528 #endif
 529
 530         /* Debugging code just in case the in-memory inode orphan list
 531          * isn't empty.  The on-disk one can be non-empty if we've
 532          * detected an error and taken the fs readonly, but the
 533          * in-memory list had better be clean by this point. */
 534         if (!list_empty(&sbi->s_orphan))
 535                 dump_orphan_list(sb, sbi);
 536         J_ASSERT(list_empty(&sbi->s_orphan));
 537
 538         invalidate_bdev(sb->s_bdev);
 539         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 540                 /*
 541                  * Invalidate the journal device's buffers.  We don't want them
 542                  * floating about in memory - the physical journal device may
 543                  * hotswapped, and it breaks the `ro-after' testing code.
 544                  */
 545                 sync_blockdev(sbi->journal_bdev);
 546                 invalidate_bdev(sbi->journal_bdev);
 547                 ext4_blkdev_remove(sbi);
 548         }
 549         sb->s_fs_info = NULL;
 550         kfree(sbi);
 551         return;
 552 }
 553
 554 static struct kmem_cache *ext4_inode_cachep;
 555
 556 /*
 557  * Called inside transaction, so use GFP_NOFS
 558  */
 559 static struct inode *ext4_alloc_inode(struct super_block *sb)
 560 {
 561         struct ext4_inode_info *ei;
 562
 563         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 564         if (!ei)
 565                 return NULL;
 566 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 567         ei->i_acl = EXT4_ACL_NOT_CACHED;
 568         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 569 #endif
 570         ei->i_block_alloc_info = NULL;
 571         ei->vfs_inode.i_version = 1;
 572         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 573         INIT_LIST_HEAD(&ei->i_prealloc_list);
 574         spin_lock_init(&ei->i_prealloc_lock);
 575         return &ei->vfs_inode;
 576 }
 577
 578 static void ext4_destroy_inode(struct inode *inode)
 579 {
 580         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 581                 printk("EXT4 Inode %p: orphan list check failed!\n",
 582                         EXT4_I(inode));
 583                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 584                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 585                                 true);
 586                 dump_stack();
 587         }
 588         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 589 }
 590
 591 static void init_once(struct kmem_cache *cachep, void *foo)
 592 {
 593         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 594
 595         INIT_LIST_HEAD(&ei->i_orphan);
 596 #ifdef CONFIG_EXT4DEV_FS_XATTR
 597         init_rwsem(&ei->xattr_sem);
 598 #endif
 599         init_rwsem(&ei->i_data_sem);
 600         inode_init_once(&ei->vfs_inode);
 601 }
 602
 603 static int init_inodecache(void)
 604 {
 605         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 606                                              sizeof(struct ext4_inode_info),
 607                                              0, (SLAB_RECLAIM_ACCOUNT|
 608                                                 SLAB_MEM_SPREAD),
 609                                              init_once);
 610         if (ext4_inode_cachep == NULL)
 611                 return -ENOMEM;
 612         return 0;
 613 }
 614
 615 static void destroy_inodecache(void)
 616 {
 617         kmem_cache_destroy(ext4_inode_cachep);
 618 }
 619
 620 static void ext4_clear_inode(struct inode *inode)
 621 {
 622         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 623 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 624         if (EXT4_I(inode)->i_acl &&
 625                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 626                 posix_acl_release(EXT4_I(inode)->i_acl);
 627                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 628         }
 629         if (EXT4_I(inode)->i_default_acl &&
 630                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 631                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 632                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 633         }
 634 #endif
 635         ext4_discard_reservation(inode);
 636         EXT4_I(inode)->i_block_alloc_info = NULL;
 637         if (unlikely(rsv))
 638                 kfree(rsv);
 639 }
 640
 641 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 642 {
 643 #if defined(CONFIG_QUOTA)
 644         struct ext4_sb_info *sbi = EXT4_SB(sb);
 645
 646         if (sbi->s_jquota_fmt)
 647                 seq_printf(seq, ",jqfmt=%s",
 648                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 649
 650         if (sbi->s_qf_names[USRQUOTA])
 651                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 652
 653         if (sbi->s_qf_names[GRPQUOTA])
 654                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 655
 656         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 657                 seq_puts(seq, ",usrquota");
 658
 659         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 660                 seq_puts(seq, ",grpquota");
 661 #endif
 662 }
 663
 664 /*
 665  * Show an option if
 666  *  - it's set to a non-default value OR
 667  *  - if the per-sb default is different from the global default
 668  */
 669 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 670 {
 671         int def_errors;
 672         unsigned long def_mount_opts;
 673         struct super_block *sb = vfs->mnt_sb;
 674         struct ext4_sb_info *sbi = EXT4_SB(sb);
 675         struct ext4_super_block *es = sbi->s_es;
 676
 677         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 678         def_errors     = le16_to_cpu(es->s_errors);
 679
 680         if (sbi->s_sb_block != 1)
 681                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 682         if (test_opt(sb, MINIX_DF))
 683                 seq_puts(seq, ",minixdf");
 684         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 685                 seq_puts(seq, ",grpid");
 686         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 687                 seq_puts(seq, ",nogrpid");
 688         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 689             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 690                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 691         }
 692         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 693             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 694                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 695         }
 696         if (test_opt(sb, ERRORS_RO)) {
 697                 if (def_errors == EXT4_ERRORS_PANIC ||
 698                     def_errors == EXT4_ERRORS_CONTINUE) {
 699                         seq_puts(seq, ",errors=remount-ro");
 700                 }
 701         }
 702         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 703                 seq_puts(seq, ",errors=continue");
 704         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 705                 seq_puts(seq, ",errors=panic");
 706         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 707                 seq_puts(seq, ",nouid32");
 708         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 709                 seq_puts(seq, ",debug");
 710         if (test_opt(sb, OLDALLOC))
 711                 seq_puts(seq, ",oldalloc");
 712 #ifdef CONFIG_EXT4DEV_FS_XATTR
 713         if (test_opt(sb, XATTR_USER) &&
 714                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 715                 seq_puts(seq, ",user_xattr");
 716         if (!test_opt(sb, XATTR_USER) &&
 717             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 718                 seq_puts(seq, ",nouser_xattr");
 719         }
 720 #endif
 721 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 722         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 723                 seq_puts(seq, ",acl");
 724         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 725                 seq_puts(seq, ",noacl");
 726 #endif
 727         if (!test_opt(sb, RESERVATION))
 728                 seq_puts(seq, ",noreservation");
 729         if (sbi->s_commit_interval) {
 730                 seq_printf(seq, ",commit=%u",
 731                            (unsigned) (sbi->s_commit_interval / HZ));
 732         }
 733         if (test_opt(sb, BARRIER))
 734                 seq_puts(seq, ",barrier=1");
 735         if (test_opt(sb, NOBH))
 736                 seq_puts(seq, ",nobh");
 737         if (!test_opt(sb, EXTENTS))
 738                 seq_puts(seq, ",noextents");
 739         if (test_opt(sb, I_VERSION))
 740                 seq_puts(seq, ",i_version");
 741
 742         /*
 743          * journal mode get enabled in different ways
 744          * So just print the value even if we didn't specify it
 745          */
 746         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 747                 seq_puts(seq, ",data=journal");
 748         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 749                 seq_puts(seq, ",data=ordered");
 750         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 751                 seq_puts(seq, ",data=writeback");
 752
 753         ext4_show_quota_options(seq, sb);
 754         return 0;
 755 }
 756
 757
 758 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 759                 u64 ino, u32 generation)
 760 {
 761         struct inode *inode;
 762
 763         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 764                 return ERR_PTR(-ESTALE);
 765         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 766                 return ERR_PTR(-ESTALE);
 767
 768         /* iget isn't really right if the inode is currently unallocated!!
 769          *
 770          * ext4_read_inode will return a bad_inode if the inode had been
 771          * deleted, so we should be safe.
 772          *
 773          * Currently we don't know the generation for parent directory, so
 774          * a generation of 0 means "accept any"
 775          */
 776         inode = iget(sb, ino);
 777         if (inode == NULL)
 778                 return ERR_PTR(-ENOMEM);
 779         if (is_bad_inode(inode) ||
 780             (generation && inode->i_generation != generation)) {
 781                 iput(inode);
 782                 return ERR_PTR(-ESTALE);
 783         }
 784
 785         return inode;
 786 }
 787
 788 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 789                 int fh_len, int fh_type)
 790 {
 791         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 792                                     ext4_nfs_get_inode);
 793 }
 794
 795 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 796                 int fh_len, int fh_type)
 797 {
 798         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 799                                     ext4_nfs_get_inode);
 800 }
 801
 802 #ifdef CONFIG_QUOTA
 803 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 804 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 805
 806 static int ext4_dquot_initialize(struct inode *inode, int type);
 807 static int ext4_dquot_drop(struct inode *inode);
 808 static int ext4_write_dquot(struct dquot *dquot);
 809 static int ext4_acquire_dquot(struct dquot *dquot);
 810 static int ext4_release_dquot(struct dquot *dquot);
 811 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 812 static int ext4_write_info(struct super_block *sb, int type);
 813 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 814 static int ext4_quota_on_mount(struct super_block *sb, int type);
 815 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 816                                size_t len, loff_t off);
 817 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 818                                 const char *data, size_t len, loff_t off);
 819
 820 static struct dquot_operations ext4_quota_operations = {
 821         .initialize     = ext4_dquot_initialize,
 822         .drop           = ext4_dquot_drop,
 823         .alloc_space    = dquot_alloc_space,
 824         .alloc_inode    = dquot_alloc_inode,
 825         .free_space     = dquot_free_space,
 826         .free_inode     = dquot_free_inode,
 827         .transfer       = dquot_transfer,
 828         .write_dquot    = ext4_write_dquot,
 829         .acquire_dquot  = ext4_acquire_dquot,
 830         .release_dquot  = ext4_release_dquot,
 831         .mark_dirty     = ext4_mark_dquot_dirty,
 832         .write_info     = ext4_write_info
 833 };
 834
 835 static struct quotactl_ops ext4_qctl_operations = {
 836         .quota_on       = ext4_quota_on,
 837         .quota_off      = vfs_quota_off,
 838         .quota_sync     = vfs_quota_sync,
 839         .get_info       = vfs_get_dqinfo,
 840         .set_info       = vfs_set_dqinfo,
 841         .get_dqblk      = vfs_get_dqblk,
 842         .set_dqblk      = vfs_set_dqblk
 843 };
 844 #endif
 845
 846 static const struct super_operations ext4_sops = {
 847         .alloc_inode    = ext4_alloc_inode,
 848         .destroy_inode  = ext4_destroy_inode,
 849         .read_inode     = ext4_read_inode,
 850         .write_inode    = ext4_write_inode,
 851         .dirty_inode    = ext4_dirty_inode,
 852         .delete_inode   = ext4_delete_inode,
 853         .put_super      = ext4_put_super,
 854         .write_super    = ext4_write_super,
 855         .sync_fs        = ext4_sync_fs,
 856         .write_super_lockfs = ext4_write_super_lockfs,
 857         .unlockfs       = ext4_unlockfs,
 858         .statfs         = ext4_statfs,
 859         .remount_fs     = ext4_remount,
 860         .clear_inode    = ext4_clear_inode,
 861         .show_options   = ext4_show_options,
 862 #ifdef CONFIG_QUOTA
 863         .quota_read     = ext4_quota_read,
 864         .quota_write    = ext4_quota_write,
 865 #endif
 866 };
 867
 868 static const struct export_operations ext4_export_ops = {
 869         .fh_to_dentry = ext4_fh_to_dentry,
 870         .fh_to_parent = ext4_fh_to_parent,
 871         .get_parent = ext4_get_parent,
 872 };
 873
 874 enum {
 875         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 876         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 877         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 878         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 879         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 880         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 881         Opt_journal_checksum, Opt_journal_async_commit,
 882         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 883         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 884         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 885         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 886         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 887         Opt_mballoc, Opt_nomballoc, Opt_stripe,
 888 };
 889
 890 static match_table_t tokens = {
 891         {Opt_bsd_df, "bsddf"},
 892         {Opt_minix_df, "minixdf"},
 893         {Opt_grpid, "grpid"},
 894         {Opt_grpid, "bsdgroups"},
 895         {Opt_nogrpid, "nogrpid"},
 896         {Opt_nogrpid, "sysvgroups"},
 897         {Opt_resgid, "resgid=%u"},
 898         {Opt_resuid, "resuid=%u"},
 899         {Opt_sb, "sb=%u"},
 900         {Opt_err_cont, "errors=continue"},
 901         {Opt_err_panic, "errors=panic"},
 902         {Opt_err_ro, "errors=remount-ro"},
 903         {Opt_nouid32, "nouid32"},
 904         {Opt_nocheck, "nocheck"},
 905         {Opt_nocheck, "check=none"},
 906         {Opt_debug, "debug"},
 907         {Opt_oldalloc, "oldalloc"},
 908         {Opt_orlov, "orlov"},
 909         {Opt_user_xattr, "user_xattr"},
 910         {Opt_nouser_xattr, "nouser_xattr"},
 911         {Opt_acl, "acl"},
 912         {Opt_noacl, "noacl"},
 913         {Opt_reservation, "reservation"},
 914         {Opt_noreservation, "noreservation"},
 915         {Opt_noload, "noload"},
 916         {Opt_nobh, "nobh"},
 917         {Opt_bh, "bh"},
 918         {Opt_commit, "commit=%u"},
 919         {Opt_journal_update, "journal=update"},
 920         {Opt_journal_inum, "journal=%u"},
 921         {Opt_journal_dev, "journal_dev=%u"},
 922         {Opt_journal_checksum, "journal_checksum"},
 923         {Opt_journal_async_commit, "journal_async_commit"},
 924         {Opt_abort, "abort"},
 925         {Opt_data_journal, "data=journal"},
 926         {Opt_data_ordered, "data=ordered"},
 927         {Opt_data_writeback, "data=writeback"},
 928         {Opt_offusrjquota, "usrjquota="},
 929         {Opt_usrjquota, "usrjquota=%s"},
 930         {Opt_offgrpjquota, "grpjquota="},
 931         {Opt_grpjquota, "grpjquota=%s"},
 932         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 933         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 934         {Opt_grpquota, "grpquota"},
 935         {Opt_noquota, "noquota"},
 936         {Opt_quota, "quota"},
 937         {Opt_usrquota, "usrquota"},
 938         {Opt_barrier, "barrier=%u"},
 939         {Opt_extents, "extents"},
 940         {Opt_noextents, "noextents"},
 941         {Opt_i_version, "i_version"},
 942         {Opt_mballoc, "mballoc"},
 943         {Opt_nomballoc, "nomballoc"},
 944         {Opt_stripe, "stripe=%u"},
 945         {Opt_err, NULL},
 946         {Opt_resize, "resize"},
 947 };
 948
 949 static ext4_fsblk_t get_sb_block(void **data)
 950 {
 951         ext4_fsblk_t    sb_block;
 952         char            *options = (char *) *data;
 953
 954         if (!options || strncmp(options, "sb=", 3) != 0)
 955                 return 1;       /* Default location */
 956         options += 3;
 957         /*todo: use simple_strtoll with >32bit ext4 */
 958         sb_block = simple_strtoul(options, &options, 0);
 959         if (*options && *options != ',') {
 960                 printk("EXT4-fs: Invalid sb specification: %s\n",
 961                        (char *) *data);
 962                 return 1;
 963         }
 964         if (*options == ',')
 965                 options++;
 966         *data = (void *) options;
 967         return sb_block;
 968 }
 969
 970 static int parse_options (char *options, struct super_block *sb,
 971                           unsigned int *inum, unsigned long *journal_devnum,
 972                           ext4_fsblk_t *n_blocks_count, int is_remount)
 973 {
 974         struct ext4_sb_info *sbi = EXT4_SB(sb);
 975         char * p;
 976         substring_t args[MAX_OPT_ARGS];
 977         int data_opt = 0;
 978         int option;
 979 #ifdef CONFIG_QUOTA
 980         int qtype;
 981         char *qname;
 982 #endif
 983
 984         if (!options)
 985                 return 1;
 986
 987         while ((p = strsep (&options, ",")) != NULL) {
 988                 int token;
 989                 if (!*p)
 990                         continue;
 991
 992                 token = match_token(p, tokens, args);
 993                 switch (token) {
 994                 case Opt_bsd_df:
 995                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 996                         break;
 997                 case Opt_minix_df:
 998                         set_opt (sbi->s_mount_opt, MINIX_DF);
 999                         break;
1000                 case Opt_grpid:
1001                         set_opt (sbi->s_mount_opt, GRPID);
1002                         break;
1003                 case Opt_nogrpid:
1004                         clear_opt (sbi->s_mount_opt, GRPID);
1005                         break;
1006                 case Opt_resuid:
1007                         if (match_int(&args[0], &option))
1008                                 return 0;
1009                         sbi->s_resuid = option;
1010                         break;
1011                 case Opt_resgid:
1012                         if (match_int(&args[0], &option))
1013                                 return 0;
1014                         sbi->s_resgid = option;
1015                         break;
1016                 case Opt_sb:
1017                         /* handled by get_sb_block() instead of here */
1018                         /* *sb_block = match_int(&args[0]); */
1019                         break;
1020                 case Opt_err_panic:
1021                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1022                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1023                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1024                         break;
1025                 case Opt_err_ro:
1026                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1027                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1028                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1029                         break;
1030                 case Opt_err_cont:
1031                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1032                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1033                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1034                         break;
1035                 case Opt_nouid32:
1036                         set_opt (sbi->s_mount_opt, NO_UID32);
1037                         break;
1038                 case Opt_nocheck:
1039                         clear_opt (sbi->s_mount_opt, CHECK);
1040                         break;
1041                 case Opt_debug:
1042                         set_opt (sbi->s_mount_opt, DEBUG);
1043                         break;
1044                 case Opt_oldalloc:
1045                         set_opt (sbi->s_mount_opt, OLDALLOC);
1046                         break;
1047                 case Opt_orlov:
1048                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1049                         break;
1050 #ifdef CONFIG_EXT4DEV_FS_XATTR
1051                 case Opt_user_xattr:
1052                         set_opt (sbi->s_mount_opt, XATTR_USER);
1053                         break;
1054                 case Opt_nouser_xattr:
1055                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1056                         break;
1057 #else
1058                 case Opt_user_xattr:
1059                 case Opt_nouser_xattr:
1060                         printk("EXT4 (no)user_xattr options not supported\n");
1061                         break;
1062 #endif
1063 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1064                 case Opt_acl:
1065                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1066                         break;
1067                 case Opt_noacl:
1068                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1069                         break;
1070 #else
1071                 case Opt_acl:
1072                 case Opt_noacl:
1073                         printk("EXT4 (no)acl options not supported\n");
1074                         break;
1075 #endif
1076                 case Opt_reservation:
1077                         set_opt(sbi->s_mount_opt, RESERVATION);
1078                         break;
1079                 case Opt_noreservation:
1080                         clear_opt(sbi->s_mount_opt, RESERVATION);
1081                         break;
1082                 case Opt_journal_update:
1083                         /* @@@ FIXME */
1084                         /* Eventually we will want to be able to create
1085                            a journal file here.  For now, only allow the
1086                            user to specify an existing inode to be the
1087                            journal file. */
1088                         if (is_remount) {
1089                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1090                                        "journal on remount\n");
1091                                 return 0;
1092                         }
1093                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1094                         break;
1095                 case Opt_journal_inum:
1096                         if (is_remount) {
1097                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1098                                        "journal on remount\n");
1099                                 return 0;
1100                         }
1101                         if (match_int(&args[0], &option))
1102                                 return 0;
1103                         *inum = option;
1104                         break;
1105                 case Opt_journal_dev:
1106                         if (is_remount) {
1107                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1108                                        "journal on remount\n");
1109                                 return 0;
1110                         }
1111                         if (match_int(&args[0], &option))
1112                                 return 0;
1113                         *journal_devnum = option;
1114                         break;
1115                 case Opt_journal_checksum:
1116                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1117                         break;
1118                 case Opt_journal_async_commit:
1119                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1120                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1121                         break;
1122                 case Opt_noload:
1123                         set_opt (sbi->s_mount_opt, NOLOAD);
1124                         break;
1125                 case Opt_commit:
1126                         if (match_int(&args[0], &option))
1127                                 return 0;
1128                         if (option < 0)
1129                                 return 0;
1130                         if (option == 0)
1131                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1132                         sbi->s_commit_interval = HZ * option;
1133                         break;
1134                 case Opt_data_journal:
1135                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1136                         goto datacheck;
1137                 case Opt_data_ordered:
1138                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1139                         goto datacheck;
1140                 case Opt_data_writeback:
1141                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1142                 datacheck:
1143                         if (is_remount) {
1144                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1145                                                 != data_opt) {
1146                                         printk(KERN_ERR
1147                                                 "EXT4-fs: cannot change data "
1148                                                 "mode on remount\n");
1149                                         return 0;
1150                                 }
1151                         } else {
1152                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1153                                 sbi->s_mount_opt |= data_opt;
1154                         }
1155                         break;
1156 #ifdef CONFIG_QUOTA
1157                 case Opt_usrjquota:
1158                         qtype = USRQUOTA;
1159                         goto set_qf_name;
1160                 case Opt_grpjquota:
1161                         qtype = GRPQUOTA;
1162 set_qf_name:
1163                         if (sb_any_quota_enabled(sb)) {
1164                                 printk(KERN_ERR
1165                                         "EXT4-fs: Cannot change journalled "
1166                                         "quota options when quota turned on.\n");
1167                                 return 0;
1168                         }
1169                         qname = match_strdup(&args[0]);
1170                         if (!qname) {
1171                                 printk(KERN_ERR
1172                                         "EXT4-fs: not enough memory for "
1173                                         "storing quotafile name.\n");
1174                                 return 0;
1175                         }
1176                         if (sbi->s_qf_names[qtype] &&
1177                             strcmp(sbi->s_qf_names[qtype], qname)) {
1178                                 printk(KERN_ERR
1179                                         "EXT4-fs: %s quota file already "
1180                                         "specified.\n", QTYPE2NAME(qtype));
1181                                 kfree(qname);
1182                                 return 0;
1183                         }
1184                         sbi->s_qf_names[qtype] = qname;
1185                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1186                                 printk(KERN_ERR
1187                                         "EXT4-fs: quotafile must be on "
1188                                         "filesystem root.\n");
1189                                 kfree(sbi->s_qf_names[qtype]);
1190                                 sbi->s_qf_names[qtype] = NULL;
1191                                 return 0;
1192                         }
1193                         set_opt(sbi->s_mount_opt, QUOTA);
1194                         break;
1195                 case Opt_offusrjquota:
1196                         qtype = USRQUOTA;
1197                         goto clear_qf_name;
1198                 case Opt_offgrpjquota:
1199                         qtype = GRPQUOTA;
1200 clear_qf_name:
1201                         if (sb_any_quota_enabled(sb)) {
1202                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1203                                         "journalled quota options when "
1204                                         "quota turned on.\n");
1205                                 return 0;
1206                         }
1207                         /*
1208                          * The space will be released later when all options
1209                          * are confirmed to be correct
1210                          */
1211                         sbi->s_qf_names[qtype] = NULL;
1212                         break;
1213                 case Opt_jqfmt_vfsold:
1214                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1215                         break;
1216                 case Opt_jqfmt_vfsv0:
1217                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1218                         break;
1219                 case Opt_quota:
1220                 case Opt_usrquota:
1221                         set_opt(sbi->s_mount_opt, QUOTA);
1222                         set_opt(sbi->s_mount_opt, USRQUOTA);
1223                         break;
1224                 case Opt_grpquota:
1225                         set_opt(sbi->s_mount_opt, QUOTA);
1226                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1227                         break;
1228                 case Opt_noquota:
1229                         if (sb_any_quota_enabled(sb)) {
1230                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1231                                         "options when quota turned on.\n");
1232                                 return 0;
1233                         }
1234                         clear_opt(sbi->s_mount_opt, QUOTA);
1235                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1236                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1237                         break;
1238 #else
1239                 case Opt_quota:
1240                 case Opt_usrquota:
1241                 case Opt_grpquota:
1242                 case Opt_usrjquota:
1243                 case Opt_grpjquota:
1244                 case Opt_offusrjquota:
1245                 case Opt_offgrpjquota:
1246                 case Opt_jqfmt_vfsold:
1247                 case Opt_jqfmt_vfsv0:
1248                         printk(KERN_ERR
1249                                 "EXT4-fs: journalled quota options not "
1250                                 "supported.\n");
1251                         break;
1252                 case Opt_noquota:
1253                         break;
1254 #endif
1255                 case Opt_abort:
1256                         set_opt(sbi->s_mount_opt, ABORT);
1257                         break;
1258                 case Opt_barrier:
1259                         if (match_int(&args[0], &option))
1260                                 return 0;
1261                         if (option)
1262                                 set_opt(sbi->s_mount_opt, BARRIER);
1263                         else
1264                                 clear_opt(sbi->s_mount_opt, BARRIER);
1265                         break;
1266                 case Opt_ignore:
1267                         break;
1268                 case Opt_resize:
1269                         if (!is_remount) {
1270                                 printk("EXT4-fs: resize option only available "
1271                                         "for remount\n");
1272                                 return 0;
1273                         }
1274                         if (match_int(&args[0], &option) != 0)
1275                                 return 0;
1276                         *n_blocks_count = option;
1277                         break;
1278                 case Opt_nobh:
1279                         set_opt(sbi->s_mount_opt, NOBH);
1280                         break;
1281                 case Opt_bh:
1282                         clear_opt(sbi->s_mount_opt, NOBH);
1283                         break;
1284                 case Opt_extents:
1285                         set_opt (sbi->s_mount_opt, EXTENTS);
1286                         break;
1287                 case Opt_noextents:
1288                         clear_opt (sbi->s_mount_opt, EXTENTS);
1289                         break;
1290                 case Opt_i_version:
1291                         set_opt(sbi->s_mount_opt, I_VERSION);
1292                         sb->s_flags |= MS_I_VERSION;
1293                         break;
1294                 case Opt_mballoc:
1295                         set_opt(sbi->s_mount_opt, MBALLOC);
1296                         break;
1297                 case Opt_nomballoc:
1298                         clear_opt(sbi->s_mount_opt, MBALLOC);
1299                         break;
1300                 case Opt_stripe:
1301                         if (match_int(&args[0], &option))
1302                                 return 0;
1303                         if (option < 0)
1304                                 return 0;
1305                         sbi->s_stripe = option;
1306                         break;
1307                 default:
1308                         printk (KERN_ERR
1309                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1310                                 "or missing value\n", p);
1311                         return 0;
1312                 }
1313         }
1314 #ifdef CONFIG_QUOTA
1315         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1316                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1317                      sbi->s_qf_names[USRQUOTA])
1318                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1319
1320                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1321                      sbi->s_qf_names[GRPQUOTA])
1322                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1323
1324                 if ((sbi->s_qf_names[USRQUOTA] &&
1325                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1326                     (sbi->s_qf_names[GRPQUOTA] &&
1327                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1328                         printk(KERN_ERR "EXT4-fs: old and new quota "
1329                                         "format mixing.\n");
1330                         return 0;
1331                 }
1332
1333                 if (!sbi->s_jquota_fmt) {
1334                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1335                                         "not specified.\n");
1336                         return 0;
1337                 }
1338         } else {
1339                 if (sbi->s_jquota_fmt) {
1340                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1341                                         "specified with no journalling "
1342                                         "enabled.\n");
1343                         return 0;
1344                 }
1345         }
1346 #endif
1347         return 1;
1348 }
1349
1350 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1351                             int read_only)
1352 {
1353         struct ext4_sb_info *sbi = EXT4_SB(sb);
1354         int res = 0;
1355
1356         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1357                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1358                         "forcing read-only mode\n");
1359                 res = MS_RDONLY;
1360         }
1361         if (read_only)
1362                 return res;
1363         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1364                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1365                         "running e2fsck is recommended\n");
1366         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1367                 printk (KERN_WARNING
1368                         "EXT4-fs warning: mounting fs with errors, "
1369                         "running e2fsck is recommended\n");
1370         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1371                  le16_to_cpu(es->s_mnt_count) >=
1372                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1373                 printk (KERN_WARNING
1374                         "EXT4-fs warning: maximal mount count reached, "
1375                         "running e2fsck is recommended\n");
1376         else if (le32_to_cpu(es->s_checkinterval) &&
1377                 (le32_to_cpu(es->s_lastcheck) +
1378                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1379                 printk (KERN_WARNING
1380                         "EXT4-fs warning: checktime reached, "
1381                         "running e2fsck is recommended\n");
1382 #if 0
1383                 /* @@@ We _will_ want to clear the valid bit if we find
1384                  * inconsistencies, to force a fsck at reboot.  But for
1385                  * a plain journaled filesystem we can keep it set as
1386                  * valid forever! :)
1387                  */
1388         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1389 #endif
1390         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1391                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1392         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1393         es->s_mtime = cpu_to_le32(get_seconds());
1394         ext4_update_dynamic_rev(sb);
1395         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1396
1397         ext4_commit_super(sb, es, 1);
1398         if (test_opt(sb, DEBUG))
1399                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1400                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1401                         sb->s_blocksize,
1402                         sbi->s_groups_count,
1403                         EXT4_BLOCKS_PER_GROUP(sb),
1404                         EXT4_INODES_PER_GROUP(sb),
1405                         sbi->s_mount_opt);
1406
1407         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1408         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1409                 char b[BDEVNAME_SIZE];
1410
1411                 printk("external journal on %s\n",
1412                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1413         } else {
1414                 printk("internal journal\n");
1415         }
1416         return res;
1417 }
1418
1419 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1420                             struct ext4_group_desc *gdp)
1421 {
1422         __u16 crc = 0;
1423
1424         if (sbi->s_es->s_feature_ro_compat &
1425             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1426                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1427                 __le32 le_group = cpu_to_le32(block_group);
1428
1429                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1430                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1431                 crc = crc16(crc, (__u8 *)gdp, offset);
1432                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1433                 /* for checksum of struct ext4_group_desc do the rest...*/
1434                 if ((sbi->s_es->s_feature_incompat &
1435                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1436                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1437                         crc = crc16(crc, (__u8 *)gdp + offset,
1438                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1439                                         offset);
1440         }
1441
1442         return cpu_to_le16(crc);
1443 }
1444
1445 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1446                                 struct ext4_group_desc *gdp)
1447 {
1448         if ((sbi->s_es->s_feature_ro_compat &
1449              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1450             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1451                 return 0;
1452
1453         return 1;
1454 }
1455
1456 /* Called at mount-time, super-block is locked */
1457 static int ext4_check_descriptors (struct super_block * sb)
1458 {
1459         struct ext4_sb_info *sbi = EXT4_SB(sb);
1460         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1461         ext4_fsblk_t last_block;
1462         ext4_fsblk_t block_bitmap;
1463         ext4_fsblk_t inode_bitmap;
1464         ext4_fsblk_t inode_table;
1465         struct ext4_group_desc * gdp = NULL;
1466         int desc_block = 0;
1467         int flexbg_flag = 0;
1468         ext4_group_t i;
1469
1470         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1471                 flexbg_flag = 1;
1472
1473         ext4_debug ("Checking group descriptors");
1474
1475         for (i = 0; i < sbi->s_groups_count; i++)
1476         {
1477                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1478                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1479                 else
1480                         last_block = first_block +
1481                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1482
1483                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1484                         gdp = (struct ext4_group_desc *)
1485                                         sbi->s_group_desc[desc_block++]->b_data;
1486                 block_bitmap = ext4_block_bitmap(sb, gdp);
1487                 if (block_bitmap < first_block || block_bitmap > last_block)
1488                 {
1489                         ext4_error (sb, "ext4_check_descriptors",
1490                                     "Block bitmap for group %lu"
1491                                     " not in group (block %llu)!",
1492                                     i, block_bitmap);
1493                         return 0;
1494                 }
1495                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1496                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1497                 {
1498                         ext4_error (sb, "ext4_check_descriptors",
1499                                     "Inode bitmap for group %lu"
1500                                     " not in group (block %llu)!",
1501                                     i, inode_bitmap);
1502                         return 0;
1503                 }
1504                 inode_table = ext4_inode_table(sb, gdp);
1505                 if (inode_table < first_block ||
1506                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1507                 {
1508                         ext4_error (sb, "ext4_check_descriptors",
1509                                     "Inode table for group %lu"
1510                                     " not in group (block %llu)!",
1511                                     i, inode_table);
1512                         return 0;
1513                 }
1514                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1515                         ext4_error(sb, __FUNCTION__,
1516                                    "Checksum for group %lu failed (%u!=%u)\n",
1517                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1518                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1519                         return 0;
1520                 }
1521                 if (!flexbg_flag)
1522                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1523                 gdp = (struct ext4_group_desc *)
1524                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1525         }
1526
1527         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1528         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1529         return 1;
1530 }
1531
1532 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1533  * the superblock) which were deleted from all directories, but held open by
1534  * a process at the time of a crash.  We walk the list and try to delete these
1535  * inodes at recovery time (only with a read-write filesystem).
1536  *
1537  * In order to keep the orphan inode chain consistent during traversal (in
1538  * case of crash during recovery), we link each inode into the superblock
1539  * orphan list_head and handle it the same way as an inode deletion during
1540  * normal operation (which journals the operations for us).
1541  *
1542  * We only do an iget() and an iput() on each inode, which is very safe if we
1543  * accidentally point at an in-use or already deleted inode.  The worst that
1544  * can happen in this case is that we get a "bit already cleared" message from
1545  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1546  * e2fsck was run on this filesystem, and it must have already done the orphan
1547  * inode cleanup for us, so we can safely abort without any further action.
1548  */
1549 static void ext4_orphan_cleanup (struct super_block * sb,
1550                                  struct ext4_super_block * es)
1551 {
1552         unsigned int s_flags = sb->s_flags;
1553         int nr_orphans = 0, nr_truncates = 0;
1554 #ifdef CONFIG_QUOTA
1555         int i;
1556 #endif
1557         if (!es->s_last_orphan) {
1558                 jbd_debug(4, "no orphan inodes to clean up\n");
1559                 return;
1560         }
1561
1562         if (bdev_read_only(sb->s_bdev)) {
1563                 printk(KERN_ERR "EXT4-fs: write access "
1564                         "unavailable, skipping orphan cleanup.\n");
1565                 return;
1566         }
1567
1568         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1569                 if (es->s_last_orphan)
1570                         jbd_debug(1, "Errors on filesystem, "
1571                                   "clearing orphan list.\n");
1572                 es->s_last_orphan = 0;
1573                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1574                 return;
1575         }
1576
1577         if (s_flags & MS_RDONLY) {
1578                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1579                        sb->s_id);
1580                 sb->s_flags &= ~MS_RDONLY;
1581         }
1582 #ifdef CONFIG_QUOTA
1583         /* Needed for iput() to work correctly and not trash data */
1584         sb->s_flags |= MS_ACTIVE;
1585         /* Turn on quotas so that they are updated correctly */
1586         for (i = 0; i < MAXQUOTAS; i++) {
1587                 if (EXT4_SB(sb)->s_qf_names[i]) {
1588                         int ret = ext4_quota_on_mount(sb, i);
1589                         if (ret < 0)
1590                                 printk(KERN_ERR
1591                                         "EXT4-fs: Cannot turn on journalled "
1592                                         "quota: error %d\n", ret);
1593                 }
1594         }
1595 #endif
1596
1597         while (es->s_last_orphan) {
1598                 struct inode *inode;
1599
1600                 if (!(inode =
1601                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1602                         es->s_last_orphan = 0;
1603                         break;
1604                 }
1605
1606                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1607                 DQUOT_INIT(inode);
1608                 if (inode->i_nlink) {
1609                         printk(KERN_DEBUG
1610                                 "%s: truncating inode %lu to %Ld bytes\n",
1611                                 __FUNCTION__, inode->i_ino, inode->i_size);
1612                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1613                                   inode->i_ino, inode->i_size);
1614                         ext4_truncate(inode);
1615                         nr_truncates++;
1616                 } else {
1617                         printk(KERN_DEBUG
1618                                 "%s: deleting unreferenced inode %lu\n",
1619                                 __FUNCTION__, inode->i_ino);
1620                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1621                                   inode->i_ino);
1622                         nr_orphans++;
1623                 }
1624                 iput(inode);  /* The delete magic happens here! */
1625         }
1626
1627 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1628
1629         if (nr_orphans)
1630                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1631                        sb->s_id, PLURAL(nr_orphans));
1632         if (nr_truncates)
1633                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1634                        sb->s_id, PLURAL(nr_truncates));
1635 #ifdef CONFIG_QUOTA
1636         /* Turn quotas off */
1637         for (i = 0; i < MAXQUOTAS; i++) {
1638                 if (sb_dqopt(sb)->files[i])
1639                         vfs_quota_off(sb, i);
1640         }
1641 #endif
1642         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1643 }
1644 /*
1645  * Maximal extent format file size.
1646  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1647  * extent format containers, within a sector_t, and within i_blocks
1648  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1649  * so that won't be a limiting factor.
1650  *
1651  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1652  */
1653 static loff_t ext4_max_size(int blkbits)
1654 {
1655         loff_t res;
1656         loff_t upper_limit = MAX_LFS_FILESIZE;
1657
1658         /* small i_blocks in vfs inode? */
1659         if (sizeof(blkcnt_t) < sizeof(u64)) {
1660                 /*
1661                  * CONFIG_LSF is not enabled implies the inode
1662                  * i_block represent total blocks in 512 bytes
1663                  * 32 == size of vfs inode i_blocks * 8
1664                  */
1665                 upper_limit = (1LL << 32) - 1;
1666
1667                 /* total blocks in file system block size */
1668                 upper_limit >>= (blkbits - 9);
1669                 upper_limit <<= blkbits;
1670         }
1671
1672         /* 32-bit extent-start container, ee_block */
1673         res = 1LL << 32;
1674         res <<= blkbits;
1675         res -= 1;
1676
1677         /* Sanity check against vm- & vfs- imposed limits */
1678         if (res > upper_limit)
1679                 res = upper_limit;
1680
1681         return res;
1682 }
1683
1684 /*
1685  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1686  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1687  * We need to be 1 filesystem block less than the 2^48 sector limit.
1688  */
1689 static loff_t ext4_max_bitmap_size(int bits)
1690 {
1691         loff_t res = EXT4_NDIR_BLOCKS;
1692         int meta_blocks;
1693         loff_t upper_limit;
1694         /* This is calculated to be the largest file size for a
1695          * dense, bitmapped file such that the total number of
1696          * sectors in the file, including data and all indirect blocks,
1697          * does not exceed 2^48 -1
1698          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1699          * total number of  512 bytes blocks of the file
1700          */
1701
1702         if (sizeof(blkcnt_t) < sizeof(u64)) {
1703                 /*
1704                  * CONFIG_LSF is not enabled implies the inode
1705                  * i_block represent total blocks in 512 bytes
1706                  * 32 == size of vfs inode i_blocks * 8
1707                  */
1708                 upper_limit = (1LL << 32) - 1;
1709
1710                 /* total blocks in file system block size */
1711                 upper_limit >>= (bits - 9);
1712
1713         } else {
1714                 /*
1715                  * We use 48 bit ext4_inode i_blocks
1716                  * With EXT4_HUGE_FILE_FL set the i_blocks
1717                  * represent total number of blocks in
1718                  * file system block size
1719                  */
1720                 upper_limit = (1LL << 48) - 1;
1721
1722         }
1723
1724         /* indirect blocks */
1725         meta_blocks = 1;
1726         /* double indirect blocks */
1727         meta_blocks += 1 + (1LL << (bits-2));
1728         /* tripple indirect blocks */
1729         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1730
1731         upper_limit -= meta_blocks;
1732         upper_limit <<= bits;
1733
1734         res += 1LL << (bits-2);
1735         res += 1LL << (2*(bits-2));
1736         res += 1LL << (3*(bits-2));
1737         res <<= bits;
1738         if (res > upper_limit)
1739                 res = upper_limit;
1740
1741         if (res > MAX_LFS_FILESIZE)
1742                 res = MAX_LFS_FILESIZE;
1743
1744         return res;
1745 }
1746
1747 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1748                                 ext4_fsblk_t logical_sb_block, int nr)
1749 {
1750         struct ext4_sb_info *sbi = EXT4_SB(sb);
1751         ext4_group_t bg, first_meta_bg;
1752         int has_super = 0;
1753
1754         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1755
1756         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1757             nr < first_meta_bg)
1758                 return logical_sb_block + nr + 1;
1759         bg = sbi->s_desc_per_block * nr;
1760         if (ext4_bg_has_super(sb, bg))
1761                 has_super = 1;
1762         return (has_super + ext4_group_first_block_no(sb, bg));
1763 }
1764
1765 /**
1766  * ext4_get_stripe_size: Get the stripe size.
1767  * @sbi: In memory super block info
1768  *
1769  * If we have specified it via mount option, then
1770  * use the mount option value. If the value specified at mount time is
1771  * greater than the blocks per group use the super block value.
1772  * If the super block value is greater than blocks per group return 0.
1773  * Allocator needs it be less than blocks per group.
1774  *
1775  */
1776 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1777 {
1778         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1779         unsigned long stripe_width =
1780                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1781
1782         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1783                 return sbi->s_stripe;
1784
1785         if (stripe_width <= sbi->s_blocks_per_group)
1786                 return stripe_width;
1787
1788         if (stride <= sbi->s_blocks_per_group)
1789                 return stride;
1790
1791         return 0;
1792 }
1793
1794 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1795                                 __releases(kernel_sem)
1796                                 __acquires(kernel_sem)
1797
1798 {
1799         struct buffer_head * bh;
1800         struct ext4_super_block *es = NULL;
1801         struct ext4_sb_info *sbi;
1802         ext4_fsblk_t block;
1803         ext4_fsblk_t sb_block = get_sb_block(&data);
1804         ext4_fsblk_t logical_sb_block;
1805         unsigned long offset = 0;
1806         unsigned int journal_inum = 0;
1807         unsigned long journal_devnum = 0;
1808         unsigned long def_mount_opts;
1809         struct inode *root;
1810         int blocksize;
1811         int hblock;
1812         int db_count;
1813         int i;
1814         int needs_recovery;
1815         __le32 features;
1816         __u64 blocks_count;
1817         int err;
1818
1819         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1820         if (!sbi)
1821                 return -ENOMEM;
1822         sb->s_fs_info = sbi;
1823         sbi->s_mount_opt = 0;
1824         sbi->s_resuid = EXT4_DEF_RESUID;
1825         sbi->s_resgid = EXT4_DEF_RESGID;
1826         sbi->s_sb_block = sb_block;
1827
1828         unlock_kernel();
1829
1830         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1831         if (!blocksize) {
1832                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1833                 goto out_fail;
1834         }
1835
1836         if (!sb_set_blocksize(sb, blocksize)) {
1837                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1838                 goto out_fail;
1839         }
1840
1841         /*
1842          * The ext4 superblock will not be buffer aligned for other than 1kB
1843          * block sizes.  We need to calculate the offset from buffer start.
1844          */
1845         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1846                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1847                 offset = do_div(logical_sb_block, blocksize);
1848         } else {
1849                 logical_sb_block = sb_block;
1850         }
1851
1852         if (!(bh = sb_bread(sb, logical_sb_block))) {
1853                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1854                 goto out_fail;
1855         }
1856         /*
1857          * Note: s_es must be initialized as soon as possible because
1858          *       some ext4 macro-instructions depend on its value
1859          */
1860         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1861         sbi->s_es = es;
1862         sb->s_magic = le16_to_cpu(es->s_magic);
1863         if (sb->s_magic != EXT4_SUPER_MAGIC)
1864                 goto cantfind_ext4;
1865
1866         /* Set defaults before we parse the mount options */
1867         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1868         if (def_mount_opts & EXT4_DEFM_DEBUG)
1869                 set_opt(sbi->s_mount_opt, DEBUG);
1870         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1871                 set_opt(sbi->s_mount_opt, GRPID);
1872         if (def_mount_opts & EXT4_DEFM_UID16)
1873                 set_opt(sbi->s_mount_opt, NO_UID32);
1874 #ifdef CONFIG_EXT4DEV_FS_XATTR
1875         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1876                 set_opt(sbi->s_mount_opt, XATTR_USER);
1877 #endif
1878 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1879         if (def_mount_opts & EXT4_DEFM_ACL)
1880                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1881 #endif
1882         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1883                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1884         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1885                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1886         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1887                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1888
1889         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1890                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1891         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1892                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1893         else
1894                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1895
1896         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1897         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1898
1899         set_opt(sbi->s_mount_opt, RESERVATION);
1900
1901         /*
1902          * turn on extents feature by default in ext4 filesystem
1903          * User -o noextents to turn it off
1904          */
1905         set_opt(sbi->s_mount_opt, EXTENTS);
1906
1907         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1908                             NULL, 0))
1909                 goto failed_mount;
1910
1911         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1912                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1913
1914         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1915             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1916              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1917              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1918                 printk(KERN_WARNING
1919                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1920                        "running e2fsck is recommended\n");
1921         /*
1922          * Check feature flags regardless of the revision level, since we
1923          * previously didn't change the revision level when setting the flags,
1924          * so there is a chance incompat flags are set on a rev 0 filesystem.
1925          */
1926         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1927         if (features) {
1928                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1929                        "unsupported optional features (%x).\n",
1930                        sb->s_id, le32_to_cpu(features));
1931                 goto failed_mount;
1932         }
1933         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1934         if (!(sb->s_flags & MS_RDONLY) && features) {
1935                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1936                        "unsupported optional features (%x).\n",
1937                        sb->s_id, le32_to_cpu(features));
1938                 goto failed_mount;
1939         }
1940         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
1941                 /*
1942                  * Large file size enabled file system can only be
1943                  * mount if kernel is build with CONFIG_LSF
1944                  */
1945                 if (sizeof(root->i_blocks) < sizeof(u64) &&
1946                                 !(sb->s_flags & MS_RDONLY)) {
1947                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
1948                                         "files cannot be mounted read-write "
1949                                         "without CONFIG_LSF.\n", sb->s_id);
1950                         goto failed_mount;
1951                 }
1952         }
1953         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1954
1955         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1956             blocksize > EXT4_MAX_BLOCK_SIZE) {
1957                 printk(KERN_ERR
1958                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1959                        blocksize, sb->s_id);
1960                 goto failed_mount;
1961         }
1962
1963         hblock = bdev_hardsect_size(sb->s_bdev);
1964         if (sb->s_blocksize != blocksize) {
1965                 /*
1966                  * Make sure the blocksize for the filesystem is larger
1967                  * than the hardware sectorsize for the machine.
1968                  */
1969                 if (blocksize < hblock) {
1970                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1971                                "device blocksize %d.\n", blocksize, hblock);
1972                         goto failed_mount;
1973                 }
1974
1975                 brelse (bh);
1976                 sb_set_blocksize(sb, blocksize);
1977                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1978                 offset = do_div(logical_sb_block, blocksize);
1979                 bh = sb_bread(sb, logical_sb_block);
1980                 if (!bh) {
1981                         printk(KERN_ERR
1982                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1983                         goto failed_mount;
1984                 }
1985                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1986                 sbi->s_es = es;
1987                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1988                         printk (KERN_ERR
1989                                 "EXT4-fs: Magic mismatch, very weird !\n");
1990                         goto failed_mount;
1991                 }
1992         }
1993
1994         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
1995         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1996
1997         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1998                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1999                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2000         } else {
2001                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2002                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2003                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2004                     (!is_power_of_2(sbi->s_inode_size)) ||
2005                     (sbi->s_inode_size > blocksize)) {
2006                         printk (KERN_ERR
2007                                 "EXT4-fs: unsupported inode size: %d\n",
2008                                 sbi->s_inode_size);
2009                         goto failed_mount;
2010                 }
2011                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2012                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2013         }
2014         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2015         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2016                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2017                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2018                     !is_power_of_2(sbi->s_desc_size)) {
2019                         printk(KERN_ERR
2020                                "EXT4-fs: unsupported descriptor size %lu\n",
2021                                sbi->s_desc_size);
2022                         goto failed_mount;
2023                 }
2024         } else
2025                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2026         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2027         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2028         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2029                 goto cantfind_ext4;
2030         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2031         if (sbi->s_inodes_per_block == 0)
2032                 goto cantfind_ext4;
2033         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2034                                         sbi->s_inodes_per_block;
2035         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2036         sbi->s_sbh = bh;
2037         sbi->s_mount_state = le16_to_cpu(es->s_state);
2038         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2039         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2040         for (i=0; i < 4; i++)
2041                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2042         sbi->s_def_hash_version = es->s_def_hash_version;
2043
2044         if (sbi->s_blocks_per_group > blocksize * 8) {
2045                 printk (KERN_ERR
2046                         "EXT4-fs: #blocks per group too big: %lu\n",
2047                         sbi->s_blocks_per_group);
2048                 goto failed_mount;
2049         }
2050         if (sbi->s_inodes_per_group > blocksize * 8) {
2051                 printk (KERN_ERR
2052                         "EXT4-fs: #inodes per group too big: %lu\n",
2053                         sbi->s_inodes_per_group);
2054                 goto failed_mount;
2055         }
2056
2057         if (ext4_blocks_count(es) >
2058                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2059                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2060                         " too large to mount safely\n", sb->s_id);
2061                 if (sizeof(sector_t) < 8)
2062                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2063                                         "enabled\n");
2064                 goto failed_mount;
2065         }
2066
2067         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2068                 goto cantfind_ext4;
2069
2070         /* ensure blocks_count calculation below doesn't sign-extend */
2071         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2072             le32_to_cpu(es->s_first_data_block) + 1) {
2073                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2074                        "first data block %u, blocks per group %lu\n",
2075                         ext4_blocks_count(es),
2076                         le32_to_cpu(es->s_first_data_block),
2077                         EXT4_BLOCKS_PER_GROUP(sb));
2078                 goto failed_mount;
2079         }
2080         blocks_count = (ext4_blocks_count(es) -
2081                         le32_to_cpu(es->s_first_data_block) +
2082                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2083         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2084         sbi->s_groups_count = blocks_count;
2085         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2086                    EXT4_DESC_PER_BLOCK(sb);
2087         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2088                                     GFP_KERNEL);
2089         if (sbi->s_group_desc == NULL) {
2090                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2091                 goto failed_mount;
2092         }
2093
2094         bgl_lock_init(&sbi->s_blockgroup_lock);
2095
2096         for (i = 0; i < db_count; i++) {
2097                 block = descriptor_loc(sb, logical_sb_block, i);
2098                 sbi->s_group_desc[i] = sb_bread(sb, block);
2099                 if (!sbi->s_group_desc[i]) {
2100                         printk (KERN_ERR "EXT4-fs: "
2101                                 "can't read group descriptor %d\n", i);
2102                         db_count = i;
2103                         goto failed_mount2;
2104                 }
2105         }
2106         if (!ext4_check_descriptors (sb)) {
2107                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2108                 goto failed_mount2;
2109         }
2110         sbi->s_gdb_count = db_count;
2111         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2112         spin_lock_init(&sbi->s_next_gen_lock);
2113
2114         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2115                         ext4_count_free_blocks(sb));
2116         if (!err) {
2117                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2118                                 ext4_count_free_inodes(sb));
2119         }
2120         if (!err) {
2121                 err = percpu_counter_init(&sbi->s_dirs_counter,
2122                                 ext4_count_dirs(sb));
2123         }
2124         if (err) {
2125                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2126                 goto failed_mount3;
2127         }
2128
2129         /* per fileystem reservation list head & lock */
2130         spin_lock_init(&sbi->s_rsv_window_lock);
2131         sbi->s_rsv_window_root = RB_ROOT;
2132         /* Add a single, static dummy reservation to the start of the
2133          * reservation window list --- it gives us a placeholder for
2134          * append-at-start-of-list which makes the allocation logic
2135          * _much_ simpler. */
2136         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2137         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2138         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2139         sbi->s_rsv_window_head.rsv_goal_size = 0;
2140         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2141
2142         sbi->s_stripe = ext4_get_stripe_size(sbi);
2143
2144         /*
2145          * set up enough so that it can read an inode
2146          */
2147         sb->s_op = &ext4_sops;
2148         sb->s_export_op = &ext4_export_ops;
2149         sb->s_xattr = ext4_xattr_handlers;
2150 #ifdef CONFIG_QUOTA
2151         sb->s_qcop = &ext4_qctl_operations;
2152         sb->dq_op = &ext4_quota_operations;
2153 #endif
2154         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2155
2156         sb->s_root = NULL;
2157
2158         needs_recovery = (es->s_last_orphan != 0 ||
2159                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2160                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2161
2162         /*
2163          * The first inode we look at is the journal inode.  Don't try
2164          * root first: it may be modified in the journal!
2165          */
2166         if (!test_opt(sb, NOLOAD) &&
2167             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2168                 if (ext4_load_journal(sb, es, journal_devnum))
2169                         goto failed_mount3;
2170         } else if (journal_inum) {
2171                 if (ext4_create_journal(sb, es, journal_inum))
2172                         goto failed_mount3;
2173         } else {
2174                 if (!silent)
2175                         printk (KERN_ERR
2176                                 "ext4: No journal on filesystem on %s\n",
2177                                 sb->s_id);
2178                 goto failed_mount3;
2179         }
2180
2181         if (ext4_blocks_count(es) > 0xffffffffULL &&
2182             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2183                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2184                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2185                 goto failed_mount4;
2186         }
2187
2188         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2189                 jbd2_journal_set_features(sbi->s_journal,
2190                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2191                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2192         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2193                 jbd2_journal_set_features(sbi->s_journal,
2194                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2195                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2196                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2197         } else {
2198                 jbd2_journal_clear_features(sbi->s_journal,
2199                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2200                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2201         }
2202
2203         /* We have now updated the journal if required, so we can
2204          * validate the data journaling mode. */
2205         switch (test_opt(sb, DATA_FLAGS)) {
2206         case 0:
2207                 /* No mode set, assume a default based on the journal
2208                  * capabilities: ORDERED_DATA if the journal can
2209                  * cope, else JOURNAL_DATA
2210                  */
2211                 if (jbd2_journal_check_available_features
2212                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2213                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2214                 else
2215                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2216                 break;
2217
2218         case EXT4_MOUNT_ORDERED_DATA:
2219         case EXT4_MOUNT_WRITEBACK_DATA:
2220                 if (!jbd2_journal_check_available_features
2221                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2222                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2223                                "requested data journaling mode\n");
2224                         goto failed_mount4;
2225                 }
2226         default:
2227                 break;
2228         }
2229
2230         if (test_opt(sb, NOBH)) {
2231                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2232                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2233                                 "its supported only with writeback mode\n");
2234                         clear_opt(sbi->s_mount_opt, NOBH);
2235                 }
2236         }
2237         /*
2238          * The jbd2_journal_load will have done any necessary log recovery,
2239          * so we can safely mount the rest of the filesystem now.
2240          */
2241
2242         root = iget(sb, EXT4_ROOT_INO);
2243         sb->s_root = d_alloc_root(root);
2244         if (!sb->s_root) {
2245                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2246                 iput(root);
2247                 goto failed_mount4;
2248         }
2249         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2250                 dput(sb->s_root);
2251                 sb->s_root = NULL;
2252                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2253                 goto failed_mount4;
2254         }
2255
2256         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2257
2258         /* determine the minimum size of new large inodes, if present */
2259         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2260                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2261                                                      EXT4_GOOD_OLD_INODE_SIZE;
2262                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2263                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2264                         if (sbi->s_want_extra_isize <
2265                             le16_to_cpu(es->s_want_extra_isize))
2266                                 sbi->s_want_extra_isize =
2267                                         le16_to_cpu(es->s_want_extra_isize);
2268                         if (sbi->s_want_extra_isize <
2269                             le16_to_cpu(es->s_min_extra_isize))
2270                                 sbi->s_want_extra_isize =
2271                                         le16_to_cpu(es->s_min_extra_isize);
2272                 }
2273         }
2274         /* Check if enough inode space is available */
2275         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2276                                                         sbi->s_inode_size) {
2277                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2278                                                        EXT4_GOOD_OLD_INODE_SIZE;
2279                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2280                         "available.\n");
2281         }
2282
2283         /*
2284          * akpm: core read_super() calls in here with the superblock locked.
2285          * That deadlocks, because orphan cleanup needs to lock the superblock
2286          * in numerous places.  Here we just pop the lock - it's relatively
2287          * harmless, because we are now ready to accept write_super() requests,
2288          * and aviro says that's the only reason for hanging onto the
2289          * superblock lock.
2290          */
2291         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2292         ext4_orphan_cleanup(sb, es);
2293         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2294         if (needs_recovery)
2295                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2296         ext4_mark_recovery_complete(sb, es);
2297         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2298                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2299                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2300                 "writeback");
2301
2302         ext4_ext_init(sb);
2303         ext4_mb_init(sb, needs_recovery);
2304
2305         lock_kernel();
2306         return 0;
2307
2308 cantfind_ext4:
2309         if (!silent)
2310                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2311                        sb->s_id);
2312         goto failed_mount;
2313
2314 failed_mount4:
2315         jbd2_journal_destroy(sbi->s_journal);
2316 failed_mount3:
2317         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2318         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2319         percpu_counter_destroy(&sbi->s_dirs_counter);
2320 failed_mount2:
2321         for (i = 0; i < db_count; i++)
2322                 brelse(sbi->s_group_desc[i]);
2323         kfree(sbi->s_group_desc);
2324 failed_mount:
2325 #ifdef CONFIG_QUOTA
2326         for (i = 0; i < MAXQUOTAS; i++)
2327                 kfree(sbi->s_qf_names[i]);
2328 #endif
2329         ext4_blkdev_remove(sbi);
2330         brelse(bh);
2331 out_fail:
2332         sb->s_fs_info = NULL;
2333         kfree(sbi);
2334         lock_kernel();
2335         return -EINVAL;
2336 }
2337
2338 /*
2339  * Setup any per-fs journal parameters now.  We'll do this both on
2340  * initial mount, once the journal has been initialised but before we've
2341  * done any recovery; and again on any subsequent remount.
2342  */
2343 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2344 {
2345         struct ext4_sb_info *sbi = EXT4_SB(sb);
2346
2347         if (sbi->s_commit_interval)
2348                 journal->j_commit_interval = sbi->s_commit_interval;
2349         /* We could also set up an ext4-specific default for the commit
2350          * interval here, but for now we'll just fall back to the jbd
2351          * default. */
2352
2353         spin_lock(&journal->j_state_lock);
2354         if (test_opt(sb, BARRIER))
2355                 journal->j_flags |= JBD2_BARRIER;
2356         else
2357                 journal->j_flags &= ~JBD2_BARRIER;
2358         spin_unlock(&journal->j_state_lock);
2359 }
2360
2361 static journal_t *ext4_get_journal(struct super_block *sb,
2362                                    unsigned int journal_inum)
2363 {
2364         struct inode *journal_inode;
2365         journal_t *journal;
2366
2367         /* First, test for the existence of a valid inode on disk.  Bad
2368          * things happen if we iget() an unused inode, as the subsequent
2369          * iput() will try to delete it. */
2370
2371         journal_inode = iget(sb, journal_inum);
2372         if (!journal_inode) {
2373                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2374                 return NULL;
2375         }
2376         if (!journal_inode->i_nlink) {
2377                 make_bad_inode(journal_inode);
2378                 iput(journal_inode);
2379                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2380                 return NULL;
2381         }
2382
2383         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2384                   journal_inode, journal_inode->i_size);
2385         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2386                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2387                 iput(journal_inode);
2388                 return NULL;
2389         }
2390
2391         journal = jbd2_journal_init_inode(journal_inode);
2392         if (!journal) {
2393                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2394                 iput(journal_inode);
2395                 return NULL;
2396         }
2397         journal->j_private = sb;
2398         ext4_init_journal_params(sb, journal);
2399         return journal;
2400 }
2401
2402 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2403                                        dev_t j_dev)
2404 {
2405         struct buffer_head * bh;
2406         journal_t *journal;
2407         ext4_fsblk_t start;
2408         ext4_fsblk_t len;
2409         int hblock, blocksize;
2410         ext4_fsblk_t sb_block;
2411         unsigned long offset;
2412         struct ext4_super_block * es;
2413         struct block_device *bdev;
2414
2415         bdev = ext4_blkdev_get(j_dev);
2416         if (bdev == NULL)
2417                 return NULL;
2418
2419         if (bd_claim(bdev, sb)) {
2420                 printk(KERN_ERR
2421                         "EXT4: failed to claim external journal device.\n");
2422                 blkdev_put(bdev);
2423                 return NULL;
2424         }
2425
2426         blocksize = sb->s_blocksize;
2427         hblock = bdev_hardsect_size(bdev);
2428         if (blocksize < hblock) {
2429                 printk(KERN_ERR
2430                         "EXT4-fs: blocksize too small for journal device.\n");
2431                 goto out_bdev;
2432         }
2433
2434         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2435         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2436         set_blocksize(bdev, blocksize);
2437         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2438                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2439                        "external journal\n");
2440                 goto out_bdev;
2441         }
2442
2443         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2444         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2445             !(le32_to_cpu(es->s_feature_incompat) &
2446               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2447                 printk(KERN_ERR "EXT4-fs: external journal has "
2448                                         "bad superblock\n");
2449                 brelse(bh);
2450                 goto out_bdev;
2451         }
2452
2453         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2454                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2455                 brelse(bh);
2456                 goto out_bdev;
2457         }
2458
2459         len = ext4_blocks_count(es);
2460         start = sb_block + 1;
2461         brelse(bh);     /* we're done with the superblock */
2462
2463         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2464                                         start, len, blocksize);
2465         if (!journal) {
2466                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2467                 goto out_bdev;
2468         }
2469         journal->j_private = sb;
2470         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2471         wait_on_buffer(journal->j_sb_buffer);
2472         if (!buffer_uptodate(journal->j_sb_buffer)) {
2473                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2474                 goto out_journal;
2475         }
2476         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2477                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2478                                         "user (unsupported) - %d\n",
2479                         be32_to_cpu(journal->j_superblock->s_nr_users));
2480                 goto out_journal;
2481         }
2482         EXT4_SB(sb)->journal_bdev = bdev;
2483         ext4_init_journal_params(sb, journal);
2484         return journal;
2485 out_journal:
2486         jbd2_journal_destroy(journal);
2487 out_bdev:
2488         ext4_blkdev_put(bdev);
2489         return NULL;
2490 }
2491
2492 static int ext4_load_journal(struct super_block *sb,
2493                              struct ext4_super_block *es,
2494                              unsigned long journal_devnum)
2495 {
2496         journal_t *journal;
2497         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2498         dev_t journal_dev;
2499         int err = 0;
2500         int really_read_only;
2501
2502         if (journal_devnum &&
2503             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2504                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2505                         "numbers have changed\n");
2506                 journal_dev = new_decode_dev(journal_devnum);
2507         } else
2508                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2509
2510         really_read_only = bdev_read_only(sb->s_bdev);
2511
2512         /*
2513          * Are we loading a blank journal or performing recovery after a
2514          * crash?  For recovery, we need to check in advance whether we
2515          * can get read-write access to the device.
2516          */
2517
2518         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2519                 if (sb->s_flags & MS_RDONLY) {
2520                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2521                                         "required on readonly filesystem.\n");
2522                         if (really_read_only) {
2523                                 printk(KERN_ERR "EXT4-fs: write access "
2524                                         "unavailable, cannot proceed.\n");
2525                                 return -EROFS;
2526                         }
2527                         printk (KERN_INFO "EXT4-fs: write access will "
2528                                         "be enabled during recovery.\n");
2529                 }
2530         }
2531
2532         if (journal_inum && journal_dev) {
2533                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2534                        "and inode journals!\n");
2535                 return -EINVAL;
2536         }
2537
2538         if (journal_inum) {
2539                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2540                         return -EINVAL;
2541         } else {
2542                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2543                         return -EINVAL;
2544         }
2545
2546         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2547                 err = jbd2_journal_update_format(journal);
2548                 if (err)  {
2549                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2550                         jbd2_journal_destroy(journal);
2551                         return err;
2552                 }
2553         }
2554
2555         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2556                 err = jbd2_journal_wipe(journal, !really_read_only);
2557         if (!err)
2558                 err = jbd2_journal_load(journal);
2559
2560         if (err) {
2561                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2562                 jbd2_journal_destroy(journal);
2563                 return err;
2564         }
2565
2566         EXT4_SB(sb)->s_journal = journal;
2567         ext4_clear_journal_err(sb, es);
2568
2569         if (journal_devnum &&
2570             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2571                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2572                 sb->s_dirt = 1;
2573
2574                 /* Make sure we flush the recovery flag to disk. */
2575                 ext4_commit_super(sb, es, 1);
2576         }
2577
2578         return 0;
2579 }
2580
2581 static int ext4_create_journal(struct super_block * sb,
2582                                struct ext4_super_block * es,
2583                                unsigned int journal_inum)
2584 {
2585         journal_t *journal;
2586         int err;
2587
2588         if (sb->s_flags & MS_RDONLY) {
2589                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2590                                 "create journal.\n");
2591                 return -EROFS;
2592         }
2593
2594         journal = ext4_get_journal(sb, journal_inum);
2595         if (!journal)
2596                 return -EINVAL;
2597
2598         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2599                journal_inum);
2600
2601         err = jbd2_journal_create(journal);
2602         if (err) {
2603                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2604                 jbd2_journal_destroy(journal);
2605                 return -EIO;
2606         }
2607
2608         EXT4_SB(sb)->s_journal = journal;
2609
2610         ext4_update_dynamic_rev(sb);
2611         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2612         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2613
2614         es->s_journal_inum = cpu_to_le32(journal_inum);
2615         sb->s_dirt = 1;
2616
2617         /* Make sure we flush the recovery flag to disk. */
2618         ext4_commit_super(sb, es, 1);
2619
2620         return 0;
2621 }
2622
2623 static void ext4_commit_super (struct super_block * sb,
2624                                struct ext4_super_block * es,
2625                                int sync)
2626 {
2627         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2628
2629         if (!sbh)
2630                 return;
2631         es->s_wtime = cpu_to_le32(get_seconds());
2632         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2633         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2634         BUFFER_TRACE(sbh, "marking dirty");
2635         mark_buffer_dirty(sbh);
2636         if (sync)
2637                 sync_dirty_buffer(sbh);
2638 }
2639
2640
2641 /*
2642  * Have we just finished recovery?  If so, and if we are mounting (or
2643  * remounting) the filesystem readonly, then we will end up with a
2644  * consistent fs on disk.  Record that fact.
2645  */
2646 static void ext4_mark_recovery_complete(struct super_block * sb,
2647                                         struct ext4_super_block * es)
2648 {
2649         journal_t *journal = EXT4_SB(sb)->s_journal;
2650
2651         jbd2_journal_lock_updates(journal);
2652         jbd2_journal_flush(journal);
2653         lock_super(sb);
2654         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2655             sb->s_flags & MS_RDONLY) {
2656                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2657                 sb->s_dirt = 0;
2658                 ext4_commit_super(sb, es, 1);
2659         }
2660         unlock_super(sb);
2661         jbd2_journal_unlock_updates(journal);
2662 }
2663
2664 /*
2665  * If we are mounting (or read-write remounting) a filesystem whose journal
2666  * has recorded an error from a previous lifetime, move that error to the
2667  * main filesystem now.
2668  */
2669 static void ext4_clear_journal_err(struct super_block * sb,
2670                                    struct ext4_super_block * es)
2671 {
2672         journal_t *journal;
2673         int j_errno;
2674         const char *errstr;
2675
2676         journal = EXT4_SB(sb)->s_journal;
2677
2678         /*
2679          * Now check for any error status which may have been recorded in the
2680          * journal by a prior ext4_error() or ext4_abort()
2681          */
2682
2683         j_errno = jbd2_journal_errno(journal);
2684         if (j_errno) {
2685                 char nbuf[16];
2686
2687                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2688                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2689                              "from previous mount: %s", errstr);
2690                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2691                              "filesystem check.");
2692
2693                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2694                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2695                 ext4_commit_super (sb, es, 1);
2696
2697                 jbd2_journal_clear_err(journal);
2698         }
2699 }
2700
2701 /*
2702  * Force the running and committing transactions to commit,
2703  * and wait on the commit.
2704  */
2705 int ext4_force_commit(struct super_block *sb)
2706 {
2707         journal_t *journal;
2708         int ret;
2709
2710         if (sb->s_flags & MS_RDONLY)
2711                 return 0;
2712
2713         journal = EXT4_SB(sb)->s_journal;
2714         sb->s_dirt = 0;
2715         ret = ext4_journal_force_commit(journal);
2716         return ret;
2717 }
2718
2719 /*
2720  * Ext4 always journals updates to the superblock itself, so we don't
2721  * have to propagate any other updates to the superblock on disk at this
2722  * point.  Just start an async writeback to get the buffers on their way
2723  * to the disk.
2724  *
2725  * This implicitly triggers the writebehind on sync().
2726  */
2727
2728 static void ext4_write_super (struct super_block * sb)
2729 {
2730         if (mutex_trylock(&sb->s_lock) != 0)
2731                 BUG();
2732         sb->s_dirt = 0;
2733 }
2734
2735 static int ext4_sync_fs(struct super_block *sb, int wait)
2736 {
2737         tid_t target;
2738
2739         sb->s_dirt = 0;
2740         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2741                 if (wait)
2742                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2743         }
2744         return 0;
2745 }
2746
2747 /*
2748  * LVM calls this function before a (read-only) snapshot is created.  This
2749  * gives us a chance to flush the journal completely and mark the fs clean.
2750  */
2751 static void ext4_write_super_lockfs(struct super_block *sb)
2752 {
2753         sb->s_dirt = 0;
2754
2755         if (!(sb->s_flags & MS_RDONLY)) {
2756                 journal_t *journal = EXT4_SB(sb)->s_journal;
2757
2758                 /* Now we set up the journal barrier. */
2759                 jbd2_journal_lock_updates(journal);
2760                 jbd2_journal_flush(journal);
2761
2762                 /* Journal blocked and flushed, clear needs_recovery flag. */
2763                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2764                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2765         }
2766 }
2767
2768 /*
2769  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2770  * flag here, even though the filesystem is not technically dirty yet.
2771  */
2772 static void ext4_unlockfs(struct super_block *sb)
2773 {
2774         if (!(sb->s_flags & MS_RDONLY)) {
2775                 lock_super(sb);
2776                 /* Reser the needs_recovery flag before the fs is unlocked. */
2777                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2778                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2779                 unlock_super(sb);
2780                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2781         }
2782 }
2783
2784 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2785 {
2786         struct ext4_super_block * es;
2787         struct ext4_sb_info *sbi = EXT4_SB(sb);
2788         ext4_fsblk_t n_blocks_count = 0;
2789         unsigned long old_sb_flags;
2790         struct ext4_mount_options old_opts;
2791         int err;
2792 #ifdef CONFIG_QUOTA
2793         int i;
2794 #endif
2795
2796         /* Store the original options */
2797         old_sb_flags = sb->s_flags;
2798         old_opts.s_mount_opt = sbi->s_mount_opt;
2799         old_opts.s_resuid = sbi->s_resuid;
2800         old_opts.s_resgid = sbi->s_resgid;
2801         old_opts.s_commit_interval = sbi->s_commit_interval;
2802 #ifdef CONFIG_QUOTA
2803         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2804         for (i = 0; i < MAXQUOTAS; i++)
2805                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2806 #endif
2807
2808         /*
2809          * Allow the "check" option to be passed as a remount option.
2810          */
2811         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2812                 err = -EINVAL;
2813                 goto restore_opts;
2814         }
2815
2816         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2817                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2818
2819         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2820                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2821
2822         es = sbi->s_es;
2823
2824         ext4_init_journal_params(sb, sbi->s_journal);
2825
2826         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2827                 n_blocks_count > ext4_blocks_count(es)) {
2828                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2829                         err = -EROFS;
2830                         goto restore_opts;
2831                 }
2832
2833                 if (*flags & MS_RDONLY) {
2834                         /*
2835                          * First of all, the unconditional stuff we have to do
2836                          * to disable replay of the journal when we next remount
2837                          */
2838                         sb->s_flags |= MS_RDONLY;
2839
2840                         /*
2841                          * OK, test if we are remounting a valid rw partition
2842                          * readonly, and if so set the rdonly flag and then
2843                          * mark the partition as valid again.
2844                          */
2845                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2846                             (sbi->s_mount_state & EXT4_VALID_FS))
2847                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2848
2849                         /*
2850                          * We have to unlock super so that we can wait for
2851                          * transactions.
2852                          */
2853                         unlock_super(sb);
2854                         ext4_mark_recovery_complete(sb, es);
2855                         lock_super(sb);
2856                 } else {
2857                         __le32 ret;
2858                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2859                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2860                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2861                                        "remount RDWR because of unsupported "
2862                                        "optional features (%x).\n",
2863                                        sb->s_id, le32_to_cpu(ret));
2864                                 err = -EROFS;
2865                                 goto restore_opts;
2866                         }
2867
2868                         /*
2869                          * If we have an unprocessed orphan list hanging
2870                          * around from a previously readonly bdev mount,
2871                          * require a full umount/remount for now.
2872                          */
2873                         if (es->s_last_orphan) {
2874                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2875                                        "remount RDWR because of unprocessed "
2876                                        "orphan inode list.  Please "
2877                                        "umount/remount instead.\n",
2878                                        sb->s_id);
2879                                 err = -EINVAL;
2880                                 goto restore_opts;
2881                         }
2882
2883                         /*
2884                          * Mounting a RDONLY partition read-write, so reread
2885                          * and store the current valid flag.  (It may have
2886                          * been changed by e2fsck since we originally mounted
2887                          * the partition.)
2888                          */
2889                         ext4_clear_journal_err(sb, es);
2890                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2891                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2892                                 goto restore_opts;
2893                         if (!ext4_setup_super (sb, es, 0))
2894                                 sb->s_flags &= ~MS_RDONLY;
2895                 }
2896         }
2897 #ifdef CONFIG_QUOTA
2898         /* Release old quota file names */
2899         for (i = 0; i < MAXQUOTAS; i++)
2900                 if (old_opts.s_qf_names[i] &&
2901                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2902                         kfree(old_opts.s_qf_names[i]);
2903 #endif
2904         return 0;
2905 restore_opts:
2906         sb->s_flags = old_sb_flags;
2907         sbi->s_mount_opt = old_opts.s_mount_opt;
2908         sbi->s_resuid = old_opts.s_resuid;
2909         sbi->s_resgid = old_opts.s_resgid;
2910         sbi->s_commit_interval = old_opts.s_commit_interval;
2911 #ifdef CONFIG_QUOTA
2912         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2913         for (i = 0; i < MAXQUOTAS; i++) {
2914                 if (sbi->s_qf_names[i] &&
2915                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2916                         kfree(sbi->s_qf_names[i]);
2917                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2918         }
2919 #endif
2920         return err;
2921 }
2922
2923 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2924 {
2925         struct super_block *sb = dentry->d_sb;
2926         struct ext4_sb_info *sbi = EXT4_SB(sb);
2927         struct ext4_super_block *es = sbi->s_es;
2928         u64 fsid;
2929
2930         if (test_opt(sb, MINIX_DF)) {
2931                 sbi->s_overhead_last = 0;
2932         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2933                 ext4_group_t ngroups = sbi->s_groups_count, i;
2934                 ext4_fsblk_t overhead = 0;
2935                 smp_rmb();
2936
2937                 /*
2938                  * Compute the overhead (FS structures).  This is constant
2939                  * for a given filesystem unless the number of block groups
2940                  * changes so we cache the previous value until it does.
2941                  */
2942
2943                 /*
2944                  * All of the blocks before first_data_block are
2945                  * overhead
2946                  */
2947                 overhead = le32_to_cpu(es->s_first_data_block);
2948
2949                 /*
2950                  * Add the overhead attributed to the superblock and
2951                  * block group descriptors.  If the sparse superblocks
2952                  * feature is turned on, then not all groups have this.
2953                  */
2954                 for (i = 0; i < ngroups; i++) {
2955                         overhead += ext4_bg_has_super(sb, i) +
2956                                 ext4_bg_num_gdb(sb, i);
2957                         cond_resched();
2958                 }
2959
2960                 /*
2961                  * Every block group has an inode bitmap, a block
2962                  * bitmap, and an inode table.
2963                  */
2964                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2965                 sbi->s_overhead_last = overhead;
2966                 smp_wmb();
2967                 sbi->s_blocks_last = ext4_blocks_count(es);
2968         }
2969
2970         buf->f_type = EXT4_SUPER_MAGIC;
2971         buf->f_bsize = sb->s_blocksize;
2972         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2973         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2974         ext4_free_blocks_count_set(es, buf->f_bfree);
2975         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2976         if (buf->f_bfree < ext4_r_blocks_count(es))
2977                 buf->f_bavail = 0;
2978         buf->f_files = le32_to_cpu(es->s_inodes_count);
2979         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2980         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2981         buf->f_namelen = EXT4_NAME_LEN;
2982         fsid = le64_to_cpup((void *)es->s_uuid) ^
2983                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2984         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2985         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2986         return 0;
2987 }
2988
2989 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2990  * is locked for write. Otherwise the are possible deadlocks:
2991  * Process 1                         Process 2
2992  * ext4_create()                     quota_sync()
2993  *   jbd2_journal_start()                   write_dquot()
2994  *   DQUOT_INIT()                        down(dqio_mutex)
2995  *     down(dqio_mutex)                    jbd2_journal_start()
2996  *
2997  */
2998
2999 #ifdef CONFIG_QUOTA
3000
3001 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3002 {
3003         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3004 }
3005
3006 static int ext4_dquot_initialize(struct inode *inode, int type)
3007 {
3008         handle_t *handle;
3009         int ret, err;
3010
3011         /* We may create quota structure so we need to reserve enough blocks */
3012         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3013         if (IS_ERR(handle))
3014                 return PTR_ERR(handle);
3015         ret = dquot_initialize(inode, type);
3016         err = ext4_journal_stop(handle);
3017         if (!ret)
3018                 ret = err;
3019         return ret;
3020 }
3021
3022 static int ext4_dquot_drop(struct inode *inode)
3023 {
3024         handle_t *handle;
3025         int ret, err;
3026
3027         /* We may delete quota structure so we need to reserve enough blocks */
3028         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3029         if (IS_ERR(handle))
3030                 return PTR_ERR(handle);
3031         ret = dquot_drop(inode);
3032         err = ext4_journal_stop(handle);
3033         if (!ret)
3034                 ret = err;
3035         return ret;
3036 }
3037
3038 static int ext4_write_dquot(struct dquot *dquot)
3039 {
3040         int ret, err;
3041         handle_t *handle;
3042         struct inode *inode;
3043
3044         inode = dquot_to_inode(dquot);
3045         handle = ext4_journal_start(inode,
3046                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3047         if (IS_ERR(handle))
3048                 return PTR_ERR(handle);
3049         ret = dquot_commit(dquot);
3050         err = ext4_journal_stop(handle);
3051         if (!ret)
3052                 ret = err;
3053         return ret;
3054 }
3055
3056 static int ext4_acquire_dquot(struct dquot *dquot)
3057 {
3058         int ret, err;
3059         handle_t *handle;
3060
3061         handle = ext4_journal_start(dquot_to_inode(dquot),
3062                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3063         if (IS_ERR(handle))
3064                 return PTR_ERR(handle);
3065         ret = dquot_acquire(dquot);
3066         err = ext4_journal_stop(handle);
3067         if (!ret)
3068                 ret = err;
3069         return ret;
3070 }
3071
3072 static int ext4_release_dquot(struct dquot *dquot)
3073 {
3074         int ret, err;
3075         handle_t *handle;
3076
3077         handle = ext4_journal_start(dquot_to_inode(dquot),
3078                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3079         if (IS_ERR(handle)) {
3080                 /* Release dquot anyway to avoid endless cycle in dqput() */
3081                 dquot_release(dquot);
3082                 return PTR_ERR(handle);
3083         }
3084         ret = dquot_release(dquot);
3085         err = ext4_journal_stop(handle);
3086         if (!ret)
3087                 ret = err;
3088         return ret;
3089 }
3090
3091 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3092 {
3093         /* Are we journalling quotas? */
3094         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3095             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3096                 dquot_mark_dquot_dirty(dquot);
3097                 return ext4_write_dquot(dquot);
3098         } else {
3099                 return dquot_mark_dquot_dirty(dquot);
3100         }
3101 }
3102
3103 static int ext4_write_info(struct super_block *sb, int type)
3104 {
3105         int ret, err;
3106         handle_t *handle;
3107
3108         /* Data block + inode block */
3109         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3110         if (IS_ERR(handle))
3111                 return PTR_ERR(handle);
3112         ret = dquot_commit_info(sb, type);
3113         err = ext4_journal_stop(handle);
3114         if (!ret)
3115                 ret = err;
3116         return ret;
3117 }
3118
3119 /*
3120  * Turn on quotas during mount time - we need to find
3121  * the quota file and such...
3122  */
3123 static int ext4_quota_on_mount(struct super_block *sb, int type)
3124 {
3125         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3126                         EXT4_SB(sb)->s_jquota_fmt, type);
3127 }
3128
3129 /*
3130  * Standard function to be called on quota_on
3131  */
3132 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3133                          char *path)
3134 {
3135         int err;
3136         struct nameidata nd;
3137
3138         if (!test_opt(sb, QUOTA))
3139                 return -EINVAL;
3140         /* Not journalling quota? */
3141         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
3142             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
3143                 return vfs_quota_on(sb, type, format_id, path);
3144         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3145         if (err)
3146                 return err;
3147         /* Quotafile not on the same filesystem? */
3148         if (nd.mnt->mnt_sb != sb) {
3149                 path_release(&nd);
3150                 return -EXDEV;
3151         }
3152         /* Quotafile not of fs root? */
3153         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
3154                 printk(KERN_WARNING
3155                         "EXT4-fs: Quota file not on filesystem root. "
3156                         "Journalled quota will not work.\n");
3157         path_release(&nd);
3158         return vfs_quota_on(sb, type, format_id, path);
3159 }
3160
3161 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3162  * acquiring the locks... As quota files are never truncated and quota code
3163  * itself serializes the operations (and noone else should touch the files)
3164  * we don't have to be afraid of races */
3165 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3166                                size_t len, loff_t off)
3167 {
3168         struct inode *inode = sb_dqopt(sb)->files[type];
3169         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3170         int err = 0;
3171         int offset = off & (sb->s_blocksize - 1);
3172         int tocopy;
3173         size_t toread;
3174         struct buffer_head *bh;
3175         loff_t i_size = i_size_read(inode);
3176
3177         if (off > i_size)
3178                 return 0;
3179         if (off+len > i_size)
3180                 len = i_size-off;
3181         toread = len;
3182         while (toread > 0) {
3183                 tocopy = sb->s_blocksize - offset < toread ?
3184                                 sb->s_blocksize - offset : toread;
3185                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3186                 if (err)
3187                         return err;
3188                 if (!bh)        /* A hole? */
3189                         memset(data, 0, tocopy);
3190                 else
3191                         memcpy(data, bh->b_data+offset, tocopy);
3192                 brelse(bh);
3193                 offset = 0;
3194                 toread -= tocopy;
3195                 data += tocopy;
3196                 blk++;
3197         }
3198         return len;
3199 }
3200
3201 /* Write to quotafile (we know the transaction is already started and has
3202  * enough credits) */
3203 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3204                                 const char *data, size_t len, loff_t off)
3205 {
3206         struct inode *inode = sb_dqopt(sb)->files[type];
3207         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3208         int err = 0;
3209         int offset = off & (sb->s_blocksize - 1);
3210         int tocopy;
3211         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3212         size_t towrite = len;
3213         struct buffer_head *bh;
3214         handle_t *handle = journal_current_handle();
3215
3216         if (!handle) {
3217                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3218                         " cancelled because transaction is not started.\n",
3219                         (unsigned long long)off, (unsigned long long)len);
3220                 return -EIO;
3221         }
3222         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3223         while (towrite > 0) {
3224                 tocopy = sb->s_blocksize - offset < towrite ?
3225                                 sb->s_blocksize - offset : towrite;
3226                 bh = ext4_bread(handle, inode, blk, 1, &err);
3227                 if (!bh)
3228                         goto out;
3229                 if (journal_quota) {
3230                         err = ext4_journal_get_write_access(handle, bh);
3231                         if (err) {
3232                                 brelse(bh);
3233                                 goto out;
3234                         }
3235                 }
3236                 lock_buffer(bh);
3237                 memcpy(bh->b_data+offset, data, tocopy);
3238                 flush_dcache_page(bh->b_page);
3239                 unlock_buffer(bh);
3240                 if (journal_quota)
3241                         err = ext4_journal_dirty_metadata(handle, bh);
3242                 else {
3243                         /* Always do at least ordered writes for quotas */
3244                         err = ext4_journal_dirty_data(handle, bh);
3245                         mark_buffer_dirty(bh);
3246                 }
3247                 brelse(bh);
3248                 if (err)
3249                         goto out;
3250                 offset = 0;
3251                 towrite -= tocopy;
3252                 data += tocopy;
3253                 blk++;
3254         }
3255 out:
3256         if (len == towrite)
3257                 return err;
3258         if (inode->i_size < off+len-towrite) {
3259                 i_size_write(inode, off+len-towrite);
3260                 EXT4_I(inode)->i_disksize = inode->i_size;
3261         }
3262         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3263         ext4_mark_inode_dirty(handle, inode);
3264         mutex_unlock(&inode->i_mutex);
3265         return len - towrite;
3266 }
3267
3268 #endif
3269
3270 static int ext4_get_sb(struct file_system_type *fs_type,
3271         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3272 {
3273         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3274 }
3275
3276 static struct file_system_type ext4dev_fs_type = {
3277         .owner          = THIS_MODULE,
3278         .name           = "ext4dev",
3279         .get_sb         = ext4_get_sb,
3280         .kill_sb        = kill_block_super,
3281         .fs_flags       = FS_REQUIRES_DEV,
3282 };
3283
3284 static int __init init_ext4_fs(void)
3285 {
3286         int err;
3287
3288         err = init_ext4_mballoc();
3289         if (err)
3290                 return err;
3291
3292         err = init_ext4_xattr();
3293         if (err)
3294                 goto out2;
3295         err = init_inodecache();
3296         if (err)
3297                 goto out1;
3298         err = register_filesystem(&ext4dev_fs_type);
3299         if (err)
3300                 goto out;
3301         return 0;
3302 out:
3303         destroy_inodecache();
3304 out1:
3305         exit_ext4_xattr();
3306 out2:
3307         exit_ext4_mballoc();
3308         return err;
3309 }
3310
3311 static void __exit exit_ext4_fs(void)
3312 {
3313         unregister_filesystem(&ext4dev_fs_type);
3314         destroy_inodecache();
3315         exit_ext4_xattr();
3316         exit_ext4_mballoc();
3317 }
3318
3319 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3320 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3321 MODULE_LICENSE("GPL");
3322 module_init(init_ext4_fs)
3323 module_exit(exit_ext4_fs)