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_ext_release(sb);
 507         ext4_xattr_put_super(sb);
 508         jbd2_journal_destroy(sbi->s_journal);
 509         if (!(sb->s_flags & MS_RDONLY)) {
 510                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 511                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 512                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 513                 mark_buffer_dirty(sbi->s_sbh);
 514                 ext4_commit_super(sb, es, 1);
 515         }
 516
 517         for (i = 0; i < sbi->s_gdb_count; i++)
 518                 brelse(sbi->s_group_desc[i]);
 519         kfree(sbi->s_group_desc);
 520         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 521         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 522         percpu_counter_destroy(&sbi->s_dirs_counter);
 523         brelse(sbi->s_sbh);
 524 #ifdef CONFIG_QUOTA
 525         for (i = 0; i < MAXQUOTAS; i++)
 526                 kfree(sbi->s_qf_names[i]);
 527 #endif
 528
 529         /* Debugging code just in case the in-memory inode orphan list
 530          * isn't empty.  The on-disk one can be non-empty if we've
 531          * detected an error and taken the fs readonly, but the
 532          * in-memory list had better be clean by this point. */
 533         if (!list_empty(&sbi->s_orphan))
 534                 dump_orphan_list(sb, sbi);
 535         J_ASSERT(list_empty(&sbi->s_orphan));
 536
 537         invalidate_bdev(sb->s_bdev);
 538         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 539                 /*
 540                  * Invalidate the journal device's buffers.  We don't want them
 541                  * floating about in memory - the physical journal device may
 542                  * hotswapped, and it breaks the `ro-after' testing code.
 543                  */
 544                 sync_blockdev(sbi->journal_bdev);
 545                 invalidate_bdev(sbi->journal_bdev);
 546                 ext4_blkdev_remove(sbi);
 547         }
 548         sb->s_fs_info = NULL;
 549         kfree(sbi);
 550         return;
 551 }
 552
 553 static struct kmem_cache *ext4_inode_cachep;
 554
 555 /*
 556  * Called inside transaction, so use GFP_NOFS
 557  */
 558 static struct inode *ext4_alloc_inode(struct super_block *sb)
 559 {
 560         struct ext4_inode_info *ei;
 561
 562         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 563         if (!ei)
 564                 return NULL;
 565 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 566         ei->i_acl = EXT4_ACL_NOT_CACHED;
 567         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 568 #endif
 569         ei->i_block_alloc_info = NULL;
 570         ei->vfs_inode.i_version = 1;
 571         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 572         return &ei->vfs_inode;
 573 }
 574
 575 static void ext4_destroy_inode(struct inode *inode)
 576 {
 577         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 578                 printk("EXT4 Inode %p: orphan list check failed!\n",
 579                         EXT4_I(inode));
 580                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 581                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 582                                 true);
 583                 dump_stack();
 584         }
 585         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 586 }
 587
 588 static void init_once(struct kmem_cache *cachep, void *foo)
 589 {
 590         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 591
 592         INIT_LIST_HEAD(&ei->i_orphan);
 593 #ifdef CONFIG_EXT4DEV_FS_XATTR
 594         init_rwsem(&ei->xattr_sem);
 595 #endif
 596         mutex_init(&ei->truncate_mutex);
 597         inode_init_once(&ei->vfs_inode);
 598 }
 599
 600 static int init_inodecache(void)
 601 {
 602         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 603                                              sizeof(struct ext4_inode_info),
 604                                              0, (SLAB_RECLAIM_ACCOUNT|
 605                                                 SLAB_MEM_SPREAD),
 606                                              init_once);
 607         if (ext4_inode_cachep == NULL)
 608                 return -ENOMEM;
 609         return 0;
 610 }
 611
 612 static void destroy_inodecache(void)
 613 {
 614         kmem_cache_destroy(ext4_inode_cachep);
 615 }
 616
 617 static void ext4_clear_inode(struct inode *inode)
 618 {
 619         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 620 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 621         if (EXT4_I(inode)->i_acl &&
 622                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 623                 posix_acl_release(EXT4_I(inode)->i_acl);
 624                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 625         }
 626         if (EXT4_I(inode)->i_default_acl &&
 627                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 628                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 629                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 630         }
 631 #endif
 632         ext4_discard_reservation(inode);
 633         EXT4_I(inode)->i_block_alloc_info = NULL;
 634         if (unlikely(rsv))
 635                 kfree(rsv);
 636 }
 637
 638 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 639 {
 640 #if defined(CONFIG_QUOTA)
 641         struct ext4_sb_info *sbi = EXT4_SB(sb);
 642
 643         if (sbi->s_jquota_fmt)
 644                 seq_printf(seq, ",jqfmt=%s",
 645                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 646
 647         if (sbi->s_qf_names[USRQUOTA])
 648                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 649
 650         if (sbi->s_qf_names[GRPQUOTA])
 651                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 652
 653         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 654                 seq_puts(seq, ",usrquota");
 655
 656         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 657                 seq_puts(seq, ",grpquota");
 658 #endif
 659 }
 660
 661 /*
 662  * Show an option if
 663  *  - it's set to a non-default value OR
 664  *  - if the per-sb default is different from the global default
 665  */
 666 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 667 {
 668         struct super_block *sb = vfs->mnt_sb;
 669         struct ext4_sb_info *sbi = EXT4_SB(sb);
 670         struct ext4_super_block *es = sbi->s_es;
 671         unsigned long def_mount_opts;
 672
 673         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 674
 675         if (sbi->s_sb_block != 1)
 676                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 677         if (test_opt(sb, MINIX_DF))
 678                 seq_puts(seq, ",minixdf");
 679         if (test_opt(sb, GRPID))
 680                 seq_puts(seq, ",grpid");
 681         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 682                 seq_puts(seq, ",nogrpid");
 683         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 684             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 685                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 686         }
 687         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 688             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 689                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 690         }
 691         if (test_opt(sb, ERRORS_CONT)) {
 692                 int def_errors = le16_to_cpu(es->s_errors);
 693
 694                 if (def_errors == EXT4_ERRORS_PANIC ||
 695                     def_errors == EXT4_ERRORS_RO) {
 696                         seq_puts(seq, ",errors=continue");
 697                 }
 698         }
 699         if (test_opt(sb, ERRORS_RO))
 700                 seq_puts(seq, ",errors=remount-ro");
 701         if (test_opt(sb, ERRORS_PANIC))
 702                 seq_puts(seq, ",errors=panic");
 703         if (test_opt(sb, NO_UID32))
 704                 seq_puts(seq, ",nouid32");
 705         if (test_opt(sb, DEBUG))
 706                 seq_puts(seq, ",debug");
 707         if (test_opt(sb, OLDALLOC))
 708                 seq_puts(seq, ",oldalloc");
 709 #ifdef CONFIG_EXT4_FS_XATTR
 710         if (test_opt(sb, XATTR_USER))
 711                 seq_puts(seq, ",user_xattr");
 712         if (!test_opt(sb, XATTR_USER) &&
 713             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 714                 seq_puts(seq, ",nouser_xattr");
 715         }
 716 #endif
 717 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 718         if (test_opt(sb, POSIX_ACL))
 719                 seq_puts(seq, ",acl");
 720         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 721                 seq_puts(seq, ",noacl");
 722 #endif
 723         if (!test_opt(sb, RESERVATION))
 724                 seq_puts(seq, ",noreservation");
 725         if (sbi->s_commit_interval) {
 726                 seq_printf(seq, ",commit=%u",
 727                            (unsigned) (sbi->s_commit_interval / HZ));
 728         }
 729         if (test_opt(sb, BARRIER))
 730                 seq_puts(seq, ",barrier=1");
 731         if (test_opt(sb, NOBH))
 732                 seq_puts(seq, ",nobh");
 733         if (!test_opt(sb, EXTENTS))
 734                 seq_puts(seq, ",noextents");
 735
 736         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 737                 seq_puts(seq, ",data=journal");
 738         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 739                 seq_puts(seq, ",data=ordered");
 740         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 741                 seq_puts(seq, ",data=writeback");
 742
 743         ext4_show_quota_options(seq, sb);
 744
 745         return 0;
 746 }
 747
 748
 749 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 750                 u64 ino, u32 generation)
 751 {
 752         struct inode *inode;
 753
 754         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 755                 return ERR_PTR(-ESTALE);
 756         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 757                 return ERR_PTR(-ESTALE);
 758
 759         /* iget isn't really right if the inode is currently unallocated!!
 760          *
 761          * ext4_read_inode will return a bad_inode if the inode had been
 762          * deleted, so we should be safe.
 763          *
 764          * Currently we don't know the generation for parent directory, so
 765          * a generation of 0 means "accept any"
 766          */
 767         inode = iget(sb, ino);
 768         if (inode == NULL)
 769                 return ERR_PTR(-ENOMEM);
 770         if (is_bad_inode(inode) ||
 771             (generation && inode->i_generation != generation)) {
 772                 iput(inode);
 773                 return ERR_PTR(-ESTALE);
 774         }
 775
 776         return inode;
 777 }
 778
 779 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 780                 int fh_len, int fh_type)
 781 {
 782         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 783                                     ext4_nfs_get_inode);
 784 }
 785
 786 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 787                 int fh_len, int fh_type)
 788 {
 789         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 790                                     ext4_nfs_get_inode);
 791 }
 792
 793 #ifdef CONFIG_QUOTA
 794 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 795 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 796
 797 static int ext4_dquot_initialize(struct inode *inode, int type);
 798 static int ext4_dquot_drop(struct inode *inode);
 799 static int ext4_write_dquot(struct dquot *dquot);
 800 static int ext4_acquire_dquot(struct dquot *dquot);
 801 static int ext4_release_dquot(struct dquot *dquot);
 802 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 803 static int ext4_write_info(struct super_block *sb, int type);
 804 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 805 static int ext4_quota_on_mount(struct super_block *sb, int type);
 806 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 807                                size_t len, loff_t off);
 808 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 809                                 const char *data, size_t len, loff_t off);
 810
 811 static struct dquot_operations ext4_quota_operations = {
 812         .initialize     = ext4_dquot_initialize,
 813         .drop           = ext4_dquot_drop,
 814         .alloc_space    = dquot_alloc_space,
 815         .alloc_inode    = dquot_alloc_inode,
 816         .free_space     = dquot_free_space,
 817         .free_inode     = dquot_free_inode,
 818         .transfer       = dquot_transfer,
 819         .write_dquot    = ext4_write_dquot,
 820         .acquire_dquot  = ext4_acquire_dquot,
 821         .release_dquot  = ext4_release_dquot,
 822         .mark_dirty     = ext4_mark_dquot_dirty,
 823         .write_info     = ext4_write_info
 824 };
 825
 826 static struct quotactl_ops ext4_qctl_operations = {
 827         .quota_on       = ext4_quota_on,
 828         .quota_off      = vfs_quota_off,
 829         .quota_sync     = vfs_quota_sync,
 830         .get_info       = vfs_get_dqinfo,
 831         .set_info       = vfs_set_dqinfo,
 832         .get_dqblk      = vfs_get_dqblk,
 833         .set_dqblk      = vfs_set_dqblk
 834 };
 835 #endif
 836
 837 static const struct super_operations ext4_sops = {
 838         .alloc_inode    = ext4_alloc_inode,
 839         .destroy_inode  = ext4_destroy_inode,
 840         .read_inode     = ext4_read_inode,
 841         .write_inode    = ext4_write_inode,
 842         .dirty_inode    = ext4_dirty_inode,
 843         .delete_inode   = ext4_delete_inode,
 844         .put_super      = ext4_put_super,
 845         .write_super    = ext4_write_super,
 846         .sync_fs        = ext4_sync_fs,
 847         .write_super_lockfs = ext4_write_super_lockfs,
 848         .unlockfs       = ext4_unlockfs,
 849         .statfs         = ext4_statfs,
 850         .remount_fs     = ext4_remount,
 851         .clear_inode    = ext4_clear_inode,
 852         .show_options   = ext4_show_options,
 853 #ifdef CONFIG_QUOTA
 854         .quota_read     = ext4_quota_read,
 855         .quota_write    = ext4_quota_write,
 856 #endif
 857 };
 858
 859 static const struct export_operations ext4_export_ops = {
 860         .fh_to_dentry = ext4_fh_to_dentry,
 861         .fh_to_parent = ext4_fh_to_parent,
 862         .get_parent = ext4_get_parent,
 863 };
 864
 865 enum {
 866         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 867         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 868         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 869         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 870         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 871         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 872         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 873         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 874         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 875         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 876         Opt_grpquota, Opt_extents, Opt_noextents,
 877 };
 878
 879 static match_table_t tokens = {
 880         {Opt_bsd_df, "bsddf"},
 881         {Opt_minix_df, "minixdf"},
 882         {Opt_grpid, "grpid"},
 883         {Opt_grpid, "bsdgroups"},
 884         {Opt_nogrpid, "nogrpid"},
 885         {Opt_nogrpid, "sysvgroups"},
 886         {Opt_resgid, "resgid=%u"},
 887         {Opt_resuid, "resuid=%u"},
 888         {Opt_sb, "sb=%u"},
 889         {Opt_err_cont, "errors=continue"},
 890         {Opt_err_panic, "errors=panic"},
 891         {Opt_err_ro, "errors=remount-ro"},
 892         {Opt_nouid32, "nouid32"},
 893         {Opt_nocheck, "nocheck"},
 894         {Opt_nocheck, "check=none"},
 895         {Opt_debug, "debug"},
 896         {Opt_oldalloc, "oldalloc"},
 897         {Opt_orlov, "orlov"},
 898         {Opt_user_xattr, "user_xattr"},
 899         {Opt_nouser_xattr, "nouser_xattr"},
 900         {Opt_acl, "acl"},
 901         {Opt_noacl, "noacl"},
 902         {Opt_reservation, "reservation"},
 903         {Opt_noreservation, "noreservation"},
 904         {Opt_noload, "noload"},
 905         {Opt_nobh, "nobh"},
 906         {Opt_bh, "bh"},
 907         {Opt_commit, "commit=%u"},
 908         {Opt_journal_update, "journal=update"},
 909         {Opt_journal_inum, "journal=%u"},
 910         {Opt_journal_dev, "journal_dev=%u"},
 911         {Opt_abort, "abort"},
 912         {Opt_data_journal, "data=journal"},
 913         {Opt_data_ordered, "data=ordered"},
 914         {Opt_data_writeback, "data=writeback"},
 915         {Opt_offusrjquota, "usrjquota="},
 916         {Opt_usrjquota, "usrjquota=%s"},
 917         {Opt_offgrpjquota, "grpjquota="},
 918         {Opt_grpjquota, "grpjquota=%s"},
 919         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 920         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 921         {Opt_grpquota, "grpquota"},
 922         {Opt_noquota, "noquota"},
 923         {Opt_quota, "quota"},
 924         {Opt_usrquota, "usrquota"},
 925         {Opt_barrier, "barrier=%u"},
 926         {Opt_extents, "extents"},
 927         {Opt_noextents, "noextents"},
 928         {Opt_err, NULL},
 929         {Opt_resize, "resize"},
 930 };
 931
 932 static ext4_fsblk_t get_sb_block(void **data)
 933 {
 934         ext4_fsblk_t    sb_block;
 935         char            *options = (char *) *data;
 936
 937         if (!options || strncmp(options, "sb=", 3) != 0)
 938                 return 1;       /* Default location */
 939         options += 3;
 940         /*todo: use simple_strtoll with >32bit ext4 */
 941         sb_block = simple_strtoul(options, &options, 0);
 942         if (*options && *options != ',') {
 943                 printk("EXT4-fs: Invalid sb specification: %s\n",
 944                        (char *) *data);
 945                 return 1;
 946         }
 947         if (*options == ',')
 948                 options++;
 949         *data = (void *) options;
 950         return sb_block;
 951 }
 952
 953 static int parse_options (char *options, struct super_block *sb,
 954                           unsigned int *inum, unsigned long *journal_devnum,
 955                           ext4_fsblk_t *n_blocks_count, int is_remount)
 956 {
 957         struct ext4_sb_info *sbi = EXT4_SB(sb);
 958         char * p;
 959         substring_t args[MAX_OPT_ARGS];
 960         int data_opt = 0;
 961         int option;
 962 #ifdef CONFIG_QUOTA
 963         int qtype;
 964         char *qname;
 965 #endif
 966
 967         if (!options)
 968                 return 1;
 969
 970         while ((p = strsep (&options, ",")) != NULL) {
 971                 int token;
 972                 if (!*p)
 973                         continue;
 974
 975                 token = match_token(p, tokens, args);
 976                 switch (token) {
 977                 case Opt_bsd_df:
 978                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 979                         break;
 980                 case Opt_minix_df:
 981                         set_opt (sbi->s_mount_opt, MINIX_DF);
 982                         break;
 983                 case Opt_grpid:
 984                         set_opt (sbi->s_mount_opt, GRPID);
 985                         break;
 986                 case Opt_nogrpid:
 987                         clear_opt (sbi->s_mount_opt, GRPID);
 988                         break;
 989                 case Opt_resuid:
 990                         if (match_int(&args[0], &option))
 991                                 return 0;
 992                         sbi->s_resuid = option;
 993                         break;
 994                 case Opt_resgid:
 995                         if (match_int(&args[0], &option))
 996                                 return 0;
 997                         sbi->s_resgid = option;
 998                         break;
 999                 case Opt_sb:
1000                         /* handled by get_sb_block() instead of here */
1001                         /* *sb_block = match_int(&args[0]); */
1002                         break;
1003                 case Opt_err_panic:
1004                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1005                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1006                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1007                         break;
1008                 case Opt_err_ro:
1009                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1010                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1011                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1012                         break;
1013                 case Opt_err_cont:
1014                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1015                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1016                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1017                         break;
1018                 case Opt_nouid32:
1019                         set_opt (sbi->s_mount_opt, NO_UID32);
1020                         break;
1021                 case Opt_nocheck:
1022                         clear_opt (sbi->s_mount_opt, CHECK);
1023                         break;
1024                 case Opt_debug:
1025                         set_opt (sbi->s_mount_opt, DEBUG);
1026                         break;
1027                 case Opt_oldalloc:
1028                         set_opt (sbi->s_mount_opt, OLDALLOC);
1029                         break;
1030                 case Opt_orlov:
1031                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1032                         break;
1033 #ifdef CONFIG_EXT4DEV_FS_XATTR
1034                 case Opt_user_xattr:
1035                         set_opt (sbi->s_mount_opt, XATTR_USER);
1036                         break;
1037                 case Opt_nouser_xattr:
1038                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1039                         break;
1040 #else
1041                 case Opt_user_xattr:
1042                 case Opt_nouser_xattr:
1043                         printk("EXT4 (no)user_xattr options not supported\n");
1044                         break;
1045 #endif
1046 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1047                 case Opt_acl:
1048                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1049                         break;
1050                 case Opt_noacl:
1051                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1052                         break;
1053 #else
1054                 case Opt_acl:
1055                 case Opt_noacl:
1056                         printk("EXT4 (no)acl options not supported\n");
1057                         break;
1058 #endif
1059                 case Opt_reservation:
1060                         set_opt(sbi->s_mount_opt, RESERVATION);
1061                         break;
1062                 case Opt_noreservation:
1063                         clear_opt(sbi->s_mount_opt, RESERVATION);
1064                         break;
1065                 case Opt_journal_update:
1066                         /* @@@ FIXME */
1067                         /* Eventually we will want to be able to create
1068                            a journal file here.  For now, only allow the
1069                            user to specify an existing inode to be the
1070                            journal file. */
1071                         if (is_remount) {
1072                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1073                                        "journal on remount\n");
1074                                 return 0;
1075                         }
1076                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1077                         break;
1078                 case Opt_journal_inum:
1079                         if (is_remount) {
1080                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1081                                        "journal on remount\n");
1082                                 return 0;
1083                         }
1084                         if (match_int(&args[0], &option))
1085                                 return 0;
1086                         *inum = option;
1087                         break;
1088                 case Opt_journal_dev:
1089                         if (is_remount) {
1090                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1091                                        "journal on remount\n");
1092                                 return 0;
1093                         }
1094                         if (match_int(&args[0], &option))
1095                                 return 0;
1096                         *journal_devnum = option;
1097                         break;
1098                 case Opt_noload:
1099                         set_opt (sbi->s_mount_opt, NOLOAD);
1100                         break;
1101                 case Opt_commit:
1102                         if (match_int(&args[0], &option))
1103                                 return 0;
1104                         if (option < 0)
1105                                 return 0;
1106                         if (option == 0)
1107                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1108                         sbi->s_commit_interval = HZ * option;
1109                         break;
1110                 case Opt_data_journal:
1111                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1112                         goto datacheck;
1113                 case Opt_data_ordered:
1114                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1115                         goto datacheck;
1116                 case Opt_data_writeback:
1117                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1118                 datacheck:
1119                         if (is_remount) {
1120                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1121                                                 != data_opt) {
1122                                         printk(KERN_ERR
1123                                                 "EXT4-fs: cannot change data "
1124                                                 "mode on remount\n");
1125                                         return 0;
1126                                 }
1127                         } else {
1128                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1129                                 sbi->s_mount_opt |= data_opt;
1130                         }
1131                         break;
1132 #ifdef CONFIG_QUOTA
1133                 case Opt_usrjquota:
1134                         qtype = USRQUOTA;
1135                         goto set_qf_name;
1136                 case Opt_grpjquota:
1137                         qtype = GRPQUOTA;
1138 set_qf_name:
1139                         if (sb_any_quota_enabled(sb)) {
1140                                 printk(KERN_ERR
1141                                         "EXT4-fs: Cannot change journalled "
1142                                         "quota options when quota turned on.\n");
1143                                 return 0;
1144                         }
1145                         qname = match_strdup(&args[0]);
1146                         if (!qname) {
1147                                 printk(KERN_ERR
1148                                         "EXT4-fs: not enough memory for "
1149                                         "storing quotafile name.\n");
1150                                 return 0;
1151                         }
1152                         if (sbi->s_qf_names[qtype] &&
1153                             strcmp(sbi->s_qf_names[qtype], qname)) {
1154                                 printk(KERN_ERR
1155                                         "EXT4-fs: %s quota file already "
1156                                         "specified.\n", QTYPE2NAME(qtype));
1157                                 kfree(qname);
1158                                 return 0;
1159                         }
1160                         sbi->s_qf_names[qtype] = qname;
1161                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1162                                 printk(KERN_ERR
1163                                         "EXT4-fs: quotafile must be on "
1164                                         "filesystem root.\n");
1165                                 kfree(sbi->s_qf_names[qtype]);
1166                                 sbi->s_qf_names[qtype] = NULL;
1167                                 return 0;
1168                         }
1169                         set_opt(sbi->s_mount_opt, QUOTA);
1170                         break;
1171                 case Opt_offusrjquota:
1172                         qtype = USRQUOTA;
1173                         goto clear_qf_name;
1174                 case Opt_offgrpjquota:
1175                         qtype = GRPQUOTA;
1176 clear_qf_name:
1177                         if (sb_any_quota_enabled(sb)) {
1178                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1179                                         "journalled quota options when "
1180                                         "quota turned on.\n");
1181                                 return 0;
1182                         }
1183                         /*
1184                          * The space will be released later when all options
1185                          * are confirmed to be correct
1186                          */
1187                         sbi->s_qf_names[qtype] = NULL;
1188                         break;
1189                 case Opt_jqfmt_vfsold:
1190                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1191                         break;
1192                 case Opt_jqfmt_vfsv0:
1193                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1194                         break;
1195                 case Opt_quota:
1196                 case Opt_usrquota:
1197                         set_opt(sbi->s_mount_opt, QUOTA);
1198                         set_opt(sbi->s_mount_opt, USRQUOTA);
1199                         break;
1200                 case Opt_grpquota:
1201                         set_opt(sbi->s_mount_opt, QUOTA);
1202                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1203                         break;
1204                 case Opt_noquota:
1205                         if (sb_any_quota_enabled(sb)) {
1206                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1207                                         "options when quota turned on.\n");
1208                                 return 0;
1209                         }
1210                         clear_opt(sbi->s_mount_opt, QUOTA);
1211                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1212                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1213                         break;
1214 #else
1215                 case Opt_quota:
1216                 case Opt_usrquota:
1217                 case Opt_grpquota:
1218                 case Opt_usrjquota:
1219                 case Opt_grpjquota:
1220                 case Opt_offusrjquota:
1221                 case Opt_offgrpjquota:
1222                 case Opt_jqfmt_vfsold:
1223                 case Opt_jqfmt_vfsv0:
1224                         printk(KERN_ERR
1225                                 "EXT4-fs: journalled quota options not "
1226                                 "supported.\n");
1227                         break;
1228                 case Opt_noquota:
1229                         break;
1230 #endif
1231                 case Opt_abort:
1232                         set_opt(sbi->s_mount_opt, ABORT);
1233                         break;
1234                 case Opt_barrier:
1235                         if (match_int(&args[0], &option))
1236                                 return 0;
1237                         if (option)
1238                                 set_opt(sbi->s_mount_opt, BARRIER);
1239                         else
1240                                 clear_opt(sbi->s_mount_opt, BARRIER);
1241                         break;
1242                 case Opt_ignore:
1243                         break;
1244                 case Opt_resize:
1245                         if (!is_remount) {
1246                                 printk("EXT4-fs: resize option only available "
1247                                         "for remount\n");
1248                                 return 0;
1249                         }
1250                         if (match_int(&args[0], &option) != 0)
1251                                 return 0;
1252                         *n_blocks_count = option;
1253                         break;
1254                 case Opt_nobh:
1255                         set_opt(sbi->s_mount_opt, NOBH);
1256                         break;
1257                 case Opt_bh:
1258                         clear_opt(sbi->s_mount_opt, NOBH);
1259                         break;
1260                 case Opt_extents:
1261                         set_opt (sbi->s_mount_opt, EXTENTS);
1262                         break;
1263                 case Opt_noextents:
1264                         clear_opt (sbi->s_mount_opt, EXTENTS);
1265                         break;
1266                 default:
1267                         printk (KERN_ERR
1268                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1269                                 "or missing value\n", p);
1270                         return 0;
1271                 }
1272         }
1273 #ifdef CONFIG_QUOTA
1274         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1275                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1276                      sbi->s_qf_names[USRQUOTA])
1277                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1278
1279                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1280                      sbi->s_qf_names[GRPQUOTA])
1281                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1282
1283                 if ((sbi->s_qf_names[USRQUOTA] &&
1284                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1285                     (sbi->s_qf_names[GRPQUOTA] &&
1286                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1287                         printk(KERN_ERR "EXT4-fs: old and new quota "
1288                                         "format mixing.\n");
1289                         return 0;
1290                 }
1291
1292                 if (!sbi->s_jquota_fmt) {
1293                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1294                                         "not specified.\n");
1295                         return 0;
1296                 }
1297         } else {
1298                 if (sbi->s_jquota_fmt) {
1299                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1300                                         "specified with no journalling "
1301                                         "enabled.\n");
1302                         return 0;
1303                 }
1304         }
1305 #endif
1306         return 1;
1307 }
1308
1309 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1310                             int read_only)
1311 {
1312         struct ext4_sb_info *sbi = EXT4_SB(sb);
1313         int res = 0;
1314
1315         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1316                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1317                         "forcing read-only mode\n");
1318                 res = MS_RDONLY;
1319         }
1320         if (read_only)
1321                 return res;
1322         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1323                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1324                         "running e2fsck is recommended\n");
1325         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1326                 printk (KERN_WARNING
1327                         "EXT4-fs warning: mounting fs with errors, "
1328                         "running e2fsck is recommended\n");
1329         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1330                  le16_to_cpu(es->s_mnt_count) >=
1331                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1332                 printk (KERN_WARNING
1333                         "EXT4-fs warning: maximal mount count reached, "
1334                         "running e2fsck is recommended\n");
1335         else if (le32_to_cpu(es->s_checkinterval) &&
1336                 (le32_to_cpu(es->s_lastcheck) +
1337                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1338                 printk (KERN_WARNING
1339                         "EXT4-fs warning: checktime reached, "
1340                         "running e2fsck is recommended\n");
1341 #if 0
1342                 /* @@@ We _will_ want to clear the valid bit if we find
1343                  * inconsistencies, to force a fsck at reboot.  But for
1344                  * a plain journaled filesystem we can keep it set as
1345                  * valid forever! :)
1346                  */
1347         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1348 #endif
1349         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1350                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1351         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1352         es->s_mtime = cpu_to_le32(get_seconds());
1353         ext4_update_dynamic_rev(sb);
1354         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1355
1356         ext4_commit_super(sb, es, 1);
1357         if (test_opt(sb, DEBUG))
1358                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1359                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1360                         sb->s_blocksize,
1361                         sbi->s_groups_count,
1362                         EXT4_BLOCKS_PER_GROUP(sb),
1363                         EXT4_INODES_PER_GROUP(sb),
1364                         sbi->s_mount_opt);
1365
1366         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1367         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1368                 char b[BDEVNAME_SIZE];
1369
1370                 printk("external journal on %s\n",
1371                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1372         } else {
1373                 printk("internal journal\n");
1374         }
1375         return res;
1376 }
1377
1378 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1379                             struct ext4_group_desc *gdp)
1380 {
1381         __u16 crc = 0;
1382
1383         if (sbi->s_es->s_feature_ro_compat &
1384             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1385                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1386                 __le32 le_group = cpu_to_le32(block_group);
1387
1388                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1389                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1390                 crc = crc16(crc, (__u8 *)gdp, offset);
1391                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1392                 /* for checksum of struct ext4_group_desc do the rest...*/
1393                 if ((sbi->s_es->s_feature_incompat &
1394                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1395                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1396                         crc = crc16(crc, (__u8 *)gdp + offset,
1397                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1398                                         offset);
1399         }
1400
1401         return cpu_to_le16(crc);
1402 }
1403
1404 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1405                                 struct ext4_group_desc *gdp)
1406 {
1407         if ((sbi->s_es->s_feature_ro_compat &
1408              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1409             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1410                 return 0;
1411
1412         return 1;
1413 }
1414
1415 /* Called at mount-time, super-block is locked */
1416 static int ext4_check_descriptors (struct super_block * sb)
1417 {
1418         struct ext4_sb_info *sbi = EXT4_SB(sb);
1419         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1420         ext4_fsblk_t last_block;
1421         ext4_fsblk_t block_bitmap;
1422         ext4_fsblk_t inode_bitmap;
1423         ext4_fsblk_t inode_table;
1424         struct ext4_group_desc * gdp = NULL;
1425         int desc_block = 0;
1426         int flexbg_flag = 0;
1427         ext4_group_t i;
1428
1429         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1430                 flexbg_flag = 1;
1431
1432         ext4_debug ("Checking group descriptors");
1433
1434         for (i = 0; i < sbi->s_groups_count; i++)
1435         {
1436                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1437                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1438                 else
1439                         last_block = first_block +
1440                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1441
1442                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1443                         gdp = (struct ext4_group_desc *)
1444                                         sbi->s_group_desc[desc_block++]->b_data;
1445                 block_bitmap = ext4_block_bitmap(sb, gdp);
1446                 if (block_bitmap < first_block || block_bitmap > last_block)
1447                 {
1448                         ext4_error (sb, "ext4_check_descriptors",
1449                                     "Block bitmap for group %lu"
1450                                     " not in group (block %llu)!",
1451                                     i, block_bitmap);
1452                         return 0;
1453                 }
1454                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1455                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1456                 {
1457                         ext4_error (sb, "ext4_check_descriptors",
1458                                     "Inode bitmap for group %lu"
1459                                     " not in group (block %llu)!",
1460                                     i, inode_bitmap);
1461                         return 0;
1462                 }
1463                 inode_table = ext4_inode_table(sb, gdp);
1464                 if (inode_table < first_block ||
1465                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1466                 {
1467                         ext4_error (sb, "ext4_check_descriptors",
1468                                     "Inode table for group %lu"
1469                                     " not in group (block %llu)!",
1470                                     i, inode_table);
1471                         return 0;
1472                 }
1473                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1474                         ext4_error(sb, __FUNCTION__,
1475                                    "Checksum for group %lu failed (%u!=%u)\n",
1476                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1477                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1478                         return 0;
1479                 }
1480                 if (!flexbg_flag)
1481                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1482                 gdp = (struct ext4_group_desc *)
1483                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1484         }
1485
1486         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1487         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1488         return 1;
1489 }
1490
1491 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1492  * the superblock) which were deleted from all directories, but held open by
1493  * a process at the time of a crash.  We walk the list and try to delete these
1494  * inodes at recovery time (only with a read-write filesystem).
1495  *
1496  * In order to keep the orphan inode chain consistent during traversal (in
1497  * case of crash during recovery), we link each inode into the superblock
1498  * orphan list_head and handle it the same way as an inode deletion during
1499  * normal operation (which journals the operations for us).
1500  *
1501  * We only do an iget() and an iput() on each inode, which is very safe if we
1502  * accidentally point at an in-use or already deleted inode.  The worst that
1503  * can happen in this case is that we get a "bit already cleared" message from
1504  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1505  * e2fsck was run on this filesystem, and it must have already done the orphan
1506  * inode cleanup for us, so we can safely abort without any further action.
1507  */
1508 static void ext4_orphan_cleanup (struct super_block * sb,
1509                                  struct ext4_super_block * es)
1510 {
1511         unsigned int s_flags = sb->s_flags;
1512         int nr_orphans = 0, nr_truncates = 0;
1513 #ifdef CONFIG_QUOTA
1514         int i;
1515 #endif
1516         if (!es->s_last_orphan) {
1517                 jbd_debug(4, "no orphan inodes to clean up\n");
1518                 return;
1519         }
1520
1521         if (bdev_read_only(sb->s_bdev)) {
1522                 printk(KERN_ERR "EXT4-fs: write access "
1523                         "unavailable, skipping orphan cleanup.\n");
1524                 return;
1525         }
1526
1527         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1528                 if (es->s_last_orphan)
1529                         jbd_debug(1, "Errors on filesystem, "
1530                                   "clearing orphan list.\n");
1531                 es->s_last_orphan = 0;
1532                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1533                 return;
1534         }
1535
1536         if (s_flags & MS_RDONLY) {
1537                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1538                        sb->s_id);
1539                 sb->s_flags &= ~MS_RDONLY;
1540         }
1541 #ifdef CONFIG_QUOTA
1542         /* Needed for iput() to work correctly and not trash data */
1543         sb->s_flags |= MS_ACTIVE;
1544         /* Turn on quotas so that they are updated correctly */
1545         for (i = 0; i < MAXQUOTAS; i++) {
1546                 if (EXT4_SB(sb)->s_qf_names[i]) {
1547                         int ret = ext4_quota_on_mount(sb, i);
1548                         if (ret < 0)
1549                                 printk(KERN_ERR
1550                                         "EXT4-fs: Cannot turn on journalled "
1551                                         "quota: error %d\n", ret);
1552                 }
1553         }
1554 #endif
1555
1556         while (es->s_last_orphan) {
1557                 struct inode *inode;
1558
1559                 if (!(inode =
1560                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1561                         es->s_last_orphan = 0;
1562                         break;
1563                 }
1564
1565                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1566                 DQUOT_INIT(inode);
1567                 if (inode->i_nlink) {
1568                         printk(KERN_DEBUG
1569                                 "%s: truncating inode %lu to %Ld bytes\n",
1570                                 __FUNCTION__, inode->i_ino, inode->i_size);
1571                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1572                                   inode->i_ino, inode->i_size);
1573                         ext4_truncate(inode);
1574                         nr_truncates++;
1575                 } else {
1576                         printk(KERN_DEBUG
1577                                 "%s: deleting unreferenced inode %lu\n",
1578                                 __FUNCTION__, inode->i_ino);
1579                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1580                                   inode->i_ino);
1581                         nr_orphans++;
1582                 }
1583                 iput(inode);  /* The delete magic happens here! */
1584         }
1585
1586 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1587
1588         if (nr_orphans)
1589                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1590                        sb->s_id, PLURAL(nr_orphans));
1591         if (nr_truncates)
1592                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1593                        sb->s_id, PLURAL(nr_truncates));
1594 #ifdef CONFIG_QUOTA
1595         /* Turn quotas off */
1596         for (i = 0; i < MAXQUOTAS; i++) {
1597                 if (sb_dqopt(sb)->files[i])
1598                         vfs_quota_off(sb, i);
1599         }
1600 #endif
1601         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1602 }
1603
1604 /*
1605  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1606  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1607  * We need to be 1 filesystem block less than the 2^32 sector limit.
1608  */
1609 static loff_t ext4_max_size(int bits)
1610 {
1611         loff_t res = EXT4_NDIR_BLOCKS;
1612         /* This constant is calculated to be the largest file size for a
1613          * dense, 4k-blocksize file such that the total number of
1614          * sectors in the file, including data and all indirect blocks,
1615          * does not exceed 2^32. */
1616         const loff_t upper_limit = 0x1ff7fffd000LL;
1617
1618         res += 1LL << (bits-2);
1619         res += 1LL << (2*(bits-2));
1620         res += 1LL << (3*(bits-2));
1621         res <<= bits;
1622         if (res > upper_limit)
1623                 res = upper_limit;
1624         return res;
1625 }
1626
1627 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1628                                 ext4_fsblk_t logical_sb_block, int nr)
1629 {
1630         struct ext4_sb_info *sbi = EXT4_SB(sb);
1631         ext4_group_t bg, first_meta_bg;
1632         int has_super = 0;
1633
1634         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1635
1636         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1637             nr < first_meta_bg)
1638                 return logical_sb_block + nr + 1;
1639         bg = sbi->s_desc_per_block * nr;
1640         if (ext4_bg_has_super(sb, bg))
1641                 has_super = 1;
1642         return (has_super + ext4_group_first_block_no(sb, bg));
1643 }
1644
1645
1646 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1647                                 __releases(kernel_sem)
1648                                 __acquires(kernel_sem)
1649
1650 {
1651         struct buffer_head * bh;
1652         struct ext4_super_block *es = NULL;
1653         struct ext4_sb_info *sbi;
1654         ext4_fsblk_t block;
1655         ext4_fsblk_t sb_block = get_sb_block(&data);
1656         ext4_fsblk_t logical_sb_block;
1657         unsigned long offset = 0;
1658         unsigned int journal_inum = 0;
1659         unsigned long journal_devnum = 0;
1660         unsigned long def_mount_opts;
1661         struct inode *root;
1662         int blocksize;
1663         int hblock;
1664         int db_count;
1665         int i;
1666         int needs_recovery;
1667         __le32 features;
1668         __u64 blocks_count;
1669         int err;
1670
1671         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1672         if (!sbi)
1673                 return -ENOMEM;
1674         sb->s_fs_info = sbi;
1675         sbi->s_mount_opt = 0;
1676         sbi->s_resuid = EXT4_DEF_RESUID;
1677         sbi->s_resgid = EXT4_DEF_RESGID;
1678         sbi->s_sb_block = sb_block;
1679
1680         unlock_kernel();
1681
1682         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1683         if (!blocksize) {
1684                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1685                 goto out_fail;
1686         }
1687
1688         if (!sb_set_blocksize(sb, blocksize)) {
1689                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1690                 goto out_fail;
1691         }
1692
1693         /*
1694          * The ext4 superblock will not be buffer aligned for other than 1kB
1695          * block sizes.  We need to calculate the offset from buffer start.
1696          */
1697         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1698                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1699                 offset = do_div(logical_sb_block, blocksize);
1700         } else {
1701                 logical_sb_block = sb_block;
1702         }
1703
1704         if (!(bh = sb_bread(sb, logical_sb_block))) {
1705                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1706                 goto out_fail;
1707         }
1708         /*
1709          * Note: s_es must be initialized as soon as possible because
1710          *       some ext4 macro-instructions depend on its value
1711          */
1712         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1713         sbi->s_es = es;
1714         sb->s_magic = le16_to_cpu(es->s_magic);
1715         if (sb->s_magic != EXT4_SUPER_MAGIC)
1716                 goto cantfind_ext4;
1717
1718         /* Set defaults before we parse the mount options */
1719         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1720         if (def_mount_opts & EXT4_DEFM_DEBUG)
1721                 set_opt(sbi->s_mount_opt, DEBUG);
1722         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1723                 set_opt(sbi->s_mount_opt, GRPID);
1724         if (def_mount_opts & EXT4_DEFM_UID16)
1725                 set_opt(sbi->s_mount_opt, NO_UID32);
1726 #ifdef CONFIG_EXT4DEV_FS_XATTR
1727         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1728                 set_opt(sbi->s_mount_opt, XATTR_USER);
1729 #endif
1730 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1731         if (def_mount_opts & EXT4_DEFM_ACL)
1732                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1733 #endif
1734         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1735                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1736         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1737                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1738         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1739                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1740
1741         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1742                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1743         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1744                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1745         else
1746                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1747
1748         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1749         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1750
1751         set_opt(sbi->s_mount_opt, RESERVATION);
1752
1753         /*
1754          * turn on extents feature by default in ext4 filesystem
1755          * User -o noextents to turn it off
1756          */
1757         set_opt(sbi->s_mount_opt, EXTENTS);
1758
1759         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1760                             NULL, 0))
1761                 goto failed_mount;
1762
1763         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1764                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1765
1766         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1767             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1768              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1769              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1770                 printk(KERN_WARNING
1771                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1772                        "running e2fsck is recommended\n");
1773         /*
1774          * Check feature flags regardless of the revision level, since we
1775          * previously didn't change the revision level when setting the flags,
1776          * so there is a chance incompat flags are set on a rev 0 filesystem.
1777          */
1778         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1779         if (features) {
1780                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1781                        "unsupported optional features (%x).\n",
1782                        sb->s_id, le32_to_cpu(features));
1783                 goto failed_mount;
1784         }
1785         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1786         if (!(sb->s_flags & MS_RDONLY) && features) {
1787                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1788                        "unsupported optional features (%x).\n",
1789                        sb->s_id, le32_to_cpu(features));
1790                 goto failed_mount;
1791         }
1792         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1793
1794         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1795             blocksize > EXT4_MAX_BLOCK_SIZE) {
1796                 printk(KERN_ERR
1797                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1798                        blocksize, sb->s_id);
1799                 goto failed_mount;
1800         }
1801
1802         hblock = bdev_hardsect_size(sb->s_bdev);
1803         if (sb->s_blocksize != blocksize) {
1804                 /*
1805                  * Make sure the blocksize for the filesystem is larger
1806                  * than the hardware sectorsize for the machine.
1807                  */
1808                 if (blocksize < hblock) {
1809                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1810                                "device blocksize %d.\n", blocksize, hblock);
1811                         goto failed_mount;
1812                 }
1813
1814                 brelse (bh);
1815                 sb_set_blocksize(sb, blocksize);
1816                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1817                 offset = do_div(logical_sb_block, blocksize);
1818                 bh = sb_bread(sb, logical_sb_block);
1819                 if (!bh) {
1820                         printk(KERN_ERR
1821                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1822                         goto failed_mount;
1823                 }
1824                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1825                 sbi->s_es = es;
1826                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1827                         printk (KERN_ERR
1828                                 "EXT4-fs: Magic mismatch, very weird !\n");
1829                         goto failed_mount;
1830                 }
1831         }
1832
1833         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1834
1835         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1836                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1837                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1838         } else {
1839                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1840                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1841                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1842                     (!is_power_of_2(sbi->s_inode_size)) ||
1843                     (sbi->s_inode_size > blocksize)) {
1844                         printk (KERN_ERR
1845                                 "EXT4-fs: unsupported inode size: %d\n",
1846                                 sbi->s_inode_size);
1847                         goto failed_mount;
1848                 }
1849                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
1850                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
1851         }
1852         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1853         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1854                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1855                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1856                     !is_power_of_2(sbi->s_desc_size)) {
1857                         printk(KERN_ERR
1858                                "EXT4-fs: unsupported descriptor size %lu\n",
1859                                sbi->s_desc_size);
1860                         goto failed_mount;
1861                 }
1862         } else
1863                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1864         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1865         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1866         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
1867                 goto cantfind_ext4;
1868         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1869         if (sbi->s_inodes_per_block == 0)
1870                 goto cantfind_ext4;
1871         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1872                                         sbi->s_inodes_per_block;
1873         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1874         sbi->s_sbh = bh;
1875         sbi->s_mount_state = le16_to_cpu(es->s_state);
1876         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
1877         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
1878         for (i=0; i < 4; i++)
1879                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1880         sbi->s_def_hash_version = es->s_def_hash_version;
1881
1882         if (sbi->s_blocks_per_group > blocksize * 8) {
1883                 printk (KERN_ERR
1884                         "EXT4-fs: #blocks per group too big: %lu\n",
1885                         sbi->s_blocks_per_group);
1886                 goto failed_mount;
1887         }
1888         if (sbi->s_inodes_per_group > blocksize * 8) {
1889                 printk (KERN_ERR
1890                         "EXT4-fs: #inodes per group too big: %lu\n",
1891                         sbi->s_inodes_per_group);
1892                 goto failed_mount;
1893         }
1894
1895         if (ext4_blocks_count(es) >
1896                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1897                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1898                         " too large to mount safely\n", sb->s_id);
1899                 if (sizeof(sector_t) < 8)
1900                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1901                                         "enabled\n");
1902                 goto failed_mount;
1903         }
1904
1905         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1906                 goto cantfind_ext4;
1907         blocks_count = (ext4_blocks_count(es) -
1908                         le32_to_cpu(es->s_first_data_block) +
1909                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1910         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1911         sbi->s_groups_count = blocks_count;
1912         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1913                    EXT4_DESC_PER_BLOCK(sb);
1914         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1915                                     GFP_KERNEL);
1916         if (sbi->s_group_desc == NULL) {
1917                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1918                 goto failed_mount;
1919         }
1920
1921         bgl_lock_init(&sbi->s_blockgroup_lock);
1922
1923         for (i = 0; i < db_count; i++) {
1924                 block = descriptor_loc(sb, logical_sb_block, i);
1925                 sbi->s_group_desc[i] = sb_bread(sb, block);
1926                 if (!sbi->s_group_desc[i]) {
1927                         printk (KERN_ERR "EXT4-fs: "
1928                                 "can't read group descriptor %d\n", i);
1929                         db_count = i;
1930                         goto failed_mount2;
1931                 }
1932         }
1933         if (!ext4_check_descriptors (sb)) {
1934                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1935                 goto failed_mount2;
1936         }
1937         sbi->s_gdb_count = db_count;
1938         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1939         spin_lock_init(&sbi->s_next_gen_lock);
1940
1941         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1942                         ext4_count_free_blocks(sb));
1943         if (!err) {
1944                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1945                                 ext4_count_free_inodes(sb));
1946         }
1947         if (!err) {
1948                 err = percpu_counter_init(&sbi->s_dirs_counter,
1949                                 ext4_count_dirs(sb));
1950         }
1951         if (err) {
1952                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
1953                 goto failed_mount3;
1954         }
1955
1956         /* per fileystem reservation list head & lock */
1957         spin_lock_init(&sbi->s_rsv_window_lock);
1958         sbi->s_rsv_window_root = RB_ROOT;
1959         /* Add a single, static dummy reservation to the start of the
1960          * reservation window list --- it gives us a placeholder for
1961          * append-at-start-of-list which makes the allocation logic
1962          * _much_ simpler. */
1963         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1964         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1965         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1966         sbi->s_rsv_window_head.rsv_goal_size = 0;
1967         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1968
1969         /*
1970          * set up enough so that it can read an inode
1971          */
1972         sb->s_op = &ext4_sops;
1973         sb->s_export_op = &ext4_export_ops;
1974         sb->s_xattr = ext4_xattr_handlers;
1975 #ifdef CONFIG_QUOTA
1976         sb->s_qcop = &ext4_qctl_operations;
1977         sb->dq_op = &ext4_quota_operations;
1978 #endif
1979         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1980
1981         sb->s_root = NULL;
1982
1983         needs_recovery = (es->s_last_orphan != 0 ||
1984                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1985                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1986
1987         /*
1988          * The first inode we look at is the journal inode.  Don't try
1989          * root first: it may be modified in the journal!
1990          */
1991         if (!test_opt(sb, NOLOAD) &&
1992             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1993                 if (ext4_load_journal(sb, es, journal_devnum))
1994                         goto failed_mount3;
1995         } else if (journal_inum) {
1996                 if (ext4_create_journal(sb, es, journal_inum))
1997                         goto failed_mount3;
1998         } else {
1999                 if (!silent)
2000                         printk (KERN_ERR
2001                                 "ext4: No journal on filesystem on %s\n",
2002                                 sb->s_id);
2003                 goto failed_mount3;
2004         }
2005
2006         if (ext4_blocks_count(es) > 0xffffffffULL &&
2007             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2008                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2009                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2010                 goto failed_mount4;
2011         }
2012
2013         /* We have now updated the journal if required, so we can
2014          * validate the data journaling mode. */
2015         switch (test_opt(sb, DATA_FLAGS)) {
2016         case 0:
2017                 /* No mode set, assume a default based on the journal
2018                  * capabilities: ORDERED_DATA if the journal can
2019                  * cope, else JOURNAL_DATA
2020                  */
2021                 if (jbd2_journal_check_available_features
2022                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2023                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2024                 else
2025                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2026                 break;
2027
2028         case EXT4_MOUNT_ORDERED_DATA:
2029         case EXT4_MOUNT_WRITEBACK_DATA:
2030                 if (!jbd2_journal_check_available_features
2031                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2032                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2033                                "requested data journaling mode\n");
2034                         goto failed_mount4;
2035                 }
2036         default:
2037                 break;
2038         }
2039
2040         if (test_opt(sb, NOBH)) {
2041                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2042                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2043                                 "its supported only with writeback mode\n");
2044                         clear_opt(sbi->s_mount_opt, NOBH);
2045                 }
2046         }
2047         /*
2048          * The jbd2_journal_load will have done any necessary log recovery,
2049          * so we can safely mount the rest of the filesystem now.
2050          */
2051
2052         root = iget(sb, EXT4_ROOT_INO);
2053         sb->s_root = d_alloc_root(root);
2054         if (!sb->s_root) {
2055                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2056                 iput(root);
2057                 goto failed_mount4;
2058         }
2059         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2060                 dput(sb->s_root);
2061                 sb->s_root = NULL;
2062                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2063                 goto failed_mount4;
2064         }
2065
2066         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2067
2068         /* determine the minimum size of new large inodes, if present */
2069         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2070                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2071                                                      EXT4_GOOD_OLD_INODE_SIZE;
2072                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2073                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2074                         if (sbi->s_want_extra_isize <
2075                             le16_to_cpu(es->s_want_extra_isize))
2076                                 sbi->s_want_extra_isize =
2077                                         le16_to_cpu(es->s_want_extra_isize);
2078                         if (sbi->s_want_extra_isize <
2079                             le16_to_cpu(es->s_min_extra_isize))
2080                                 sbi->s_want_extra_isize =
2081                                         le16_to_cpu(es->s_min_extra_isize);
2082                 }
2083         }
2084         /* Check if enough inode space is available */
2085         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2086                                                         sbi->s_inode_size) {
2087                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2088                                                        EXT4_GOOD_OLD_INODE_SIZE;
2089                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2090                         "available.\n");
2091         }
2092
2093         /*
2094          * akpm: core read_super() calls in here with the superblock locked.
2095          * That deadlocks, because orphan cleanup needs to lock the superblock
2096          * in numerous places.  Here we just pop the lock - it's relatively
2097          * harmless, because we are now ready to accept write_super() requests,
2098          * and aviro says that's the only reason for hanging onto the
2099          * superblock lock.
2100          */
2101         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2102         ext4_orphan_cleanup(sb, es);
2103         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2104         if (needs_recovery)
2105                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2106         ext4_mark_recovery_complete(sb, es);
2107         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2108                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2109                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2110                 "writeback");
2111
2112         ext4_ext_init(sb);
2113
2114         lock_kernel();
2115         return 0;
2116
2117 cantfind_ext4:
2118         if (!silent)
2119                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2120                        sb->s_id);
2121         goto failed_mount;
2122
2123 failed_mount4:
2124         jbd2_journal_destroy(sbi->s_journal);
2125 failed_mount3:
2126         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2127         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2128         percpu_counter_destroy(&sbi->s_dirs_counter);
2129 failed_mount2:
2130         for (i = 0; i < db_count; i++)
2131                 brelse(sbi->s_group_desc[i]);
2132         kfree(sbi->s_group_desc);
2133 failed_mount:
2134 #ifdef CONFIG_QUOTA
2135         for (i = 0; i < MAXQUOTAS; i++)
2136                 kfree(sbi->s_qf_names[i]);
2137 #endif
2138         ext4_blkdev_remove(sbi);
2139         brelse(bh);
2140 out_fail:
2141         sb->s_fs_info = NULL;
2142         kfree(sbi);
2143         lock_kernel();
2144         return -EINVAL;
2145 }
2146
2147 /*
2148  * Setup any per-fs journal parameters now.  We'll do this both on
2149  * initial mount, once the journal has been initialised but before we've
2150  * done any recovery; and again on any subsequent remount.
2151  */
2152 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2153 {
2154         struct ext4_sb_info *sbi = EXT4_SB(sb);
2155
2156         if (sbi->s_commit_interval)
2157                 journal->j_commit_interval = sbi->s_commit_interval;
2158         /* We could also set up an ext4-specific default for the commit
2159          * interval here, but for now we'll just fall back to the jbd
2160          * default. */
2161
2162         spin_lock(&journal->j_state_lock);
2163         if (test_opt(sb, BARRIER))
2164                 journal->j_flags |= JBD2_BARRIER;
2165         else
2166                 journal->j_flags &= ~JBD2_BARRIER;
2167         spin_unlock(&journal->j_state_lock);
2168 }
2169
2170 static journal_t *ext4_get_journal(struct super_block *sb,
2171                                    unsigned int journal_inum)
2172 {
2173         struct inode *journal_inode;
2174         journal_t *journal;
2175
2176         /* First, test for the existence of a valid inode on disk.  Bad
2177          * things happen if we iget() an unused inode, as the subsequent
2178          * iput() will try to delete it. */
2179
2180         journal_inode = iget(sb, journal_inum);
2181         if (!journal_inode) {
2182                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2183                 return NULL;
2184         }
2185         if (!journal_inode->i_nlink) {
2186                 make_bad_inode(journal_inode);
2187                 iput(journal_inode);
2188                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2189                 return NULL;
2190         }
2191
2192         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2193                   journal_inode, journal_inode->i_size);
2194         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2195                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2196                 iput(journal_inode);
2197                 return NULL;
2198         }
2199
2200         journal = jbd2_journal_init_inode(journal_inode);
2201         if (!journal) {
2202                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2203                 iput(journal_inode);
2204                 return NULL;
2205         }
2206         journal->j_private = sb;
2207         ext4_init_journal_params(sb, journal);
2208         return journal;
2209 }
2210
2211 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2212                                        dev_t j_dev)
2213 {
2214         struct buffer_head * bh;
2215         journal_t *journal;
2216         ext4_fsblk_t start;
2217         ext4_fsblk_t len;
2218         int hblock, blocksize;
2219         ext4_fsblk_t sb_block;
2220         unsigned long offset;
2221         struct ext4_super_block * es;
2222         struct block_device *bdev;
2223
2224         bdev = ext4_blkdev_get(j_dev);
2225         if (bdev == NULL)
2226                 return NULL;
2227
2228         if (bd_claim(bdev, sb)) {
2229                 printk(KERN_ERR
2230                         "EXT4: failed to claim external journal device.\n");
2231                 blkdev_put(bdev);
2232                 return NULL;
2233         }
2234
2235         blocksize = sb->s_blocksize;
2236         hblock = bdev_hardsect_size(bdev);
2237         if (blocksize < hblock) {
2238                 printk(KERN_ERR
2239                         "EXT4-fs: blocksize too small for journal device.\n");
2240                 goto out_bdev;
2241         }
2242
2243         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2244         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2245         set_blocksize(bdev, blocksize);
2246         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2247                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2248                        "external journal\n");
2249                 goto out_bdev;
2250         }
2251
2252         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2253         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2254             !(le32_to_cpu(es->s_feature_incompat) &
2255               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2256                 printk(KERN_ERR "EXT4-fs: external journal has "
2257                                         "bad superblock\n");
2258                 brelse(bh);
2259                 goto out_bdev;
2260         }
2261
2262         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2263                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2264                 brelse(bh);
2265                 goto out_bdev;
2266         }
2267
2268         len = ext4_blocks_count(es);
2269         start = sb_block + 1;
2270         brelse(bh);     /* we're done with the superblock */
2271
2272         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2273                                         start, len, blocksize);
2274         if (!journal) {
2275                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2276                 goto out_bdev;
2277         }
2278         journal->j_private = sb;
2279         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2280         wait_on_buffer(journal->j_sb_buffer);
2281         if (!buffer_uptodate(journal->j_sb_buffer)) {
2282                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2283                 goto out_journal;
2284         }
2285         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2286                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2287                                         "user (unsupported) - %d\n",
2288                         be32_to_cpu(journal->j_superblock->s_nr_users));
2289                 goto out_journal;
2290         }
2291         EXT4_SB(sb)->journal_bdev = bdev;
2292         ext4_init_journal_params(sb, journal);
2293         return journal;
2294 out_journal:
2295         jbd2_journal_destroy(journal);
2296 out_bdev:
2297         ext4_blkdev_put(bdev);
2298         return NULL;
2299 }
2300
2301 static int ext4_load_journal(struct super_block *sb,
2302                              struct ext4_super_block *es,
2303                              unsigned long journal_devnum)
2304 {
2305         journal_t *journal;
2306         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2307         dev_t journal_dev;
2308         int err = 0;
2309         int really_read_only;
2310
2311         if (journal_devnum &&
2312             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2313                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2314                         "numbers have changed\n");
2315                 journal_dev = new_decode_dev(journal_devnum);
2316         } else
2317                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2318
2319         really_read_only = bdev_read_only(sb->s_bdev);
2320
2321         /*
2322          * Are we loading a blank journal or performing recovery after a
2323          * crash?  For recovery, we need to check in advance whether we
2324          * can get read-write access to the device.
2325          */
2326
2327         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2328                 if (sb->s_flags & MS_RDONLY) {
2329                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2330                                         "required on readonly filesystem.\n");
2331                         if (really_read_only) {
2332                                 printk(KERN_ERR "EXT4-fs: write access "
2333                                         "unavailable, cannot proceed.\n");
2334                                 return -EROFS;
2335                         }
2336                         printk (KERN_INFO "EXT4-fs: write access will "
2337                                         "be enabled during recovery.\n");
2338                 }
2339         }
2340
2341         if (journal_inum && journal_dev) {
2342                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2343                        "and inode journals!\n");
2344                 return -EINVAL;
2345         }
2346
2347         if (journal_inum) {
2348                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2349                         return -EINVAL;
2350         } else {
2351                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2352                         return -EINVAL;
2353         }
2354
2355         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2356                 err = jbd2_journal_update_format(journal);
2357                 if (err)  {
2358                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2359                         jbd2_journal_destroy(journal);
2360                         return err;
2361                 }
2362         }
2363
2364         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2365                 err = jbd2_journal_wipe(journal, !really_read_only);
2366         if (!err)
2367                 err = jbd2_journal_load(journal);
2368
2369         if (err) {
2370                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2371                 jbd2_journal_destroy(journal);
2372                 return err;
2373         }
2374
2375         EXT4_SB(sb)->s_journal = journal;
2376         ext4_clear_journal_err(sb, es);
2377
2378         if (journal_devnum &&
2379             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2380                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2381                 sb->s_dirt = 1;
2382
2383                 /* Make sure we flush the recovery flag to disk. */
2384                 ext4_commit_super(sb, es, 1);
2385         }
2386
2387         return 0;
2388 }
2389
2390 static int ext4_create_journal(struct super_block * sb,
2391                                struct ext4_super_block * es,
2392                                unsigned int journal_inum)
2393 {
2394         journal_t *journal;
2395         int err;
2396
2397         if (sb->s_flags & MS_RDONLY) {
2398                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2399                                 "create journal.\n");
2400                 return -EROFS;
2401         }
2402
2403         journal = ext4_get_journal(sb, journal_inum);
2404         if (!journal)
2405                 return -EINVAL;
2406
2407         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2408                journal_inum);
2409
2410         err = jbd2_journal_create(journal);
2411         if (err) {
2412                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2413                 jbd2_journal_destroy(journal);
2414                 return -EIO;
2415         }
2416
2417         EXT4_SB(sb)->s_journal = journal;
2418
2419         ext4_update_dynamic_rev(sb);
2420         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2421         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2422
2423         es->s_journal_inum = cpu_to_le32(journal_inum);
2424         sb->s_dirt = 1;
2425
2426         /* Make sure we flush the recovery flag to disk. */
2427         ext4_commit_super(sb, es, 1);
2428
2429         return 0;
2430 }
2431
2432 static void ext4_commit_super (struct super_block * sb,
2433                                struct ext4_super_block * es,
2434                                int sync)
2435 {
2436         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2437
2438         if (!sbh)
2439                 return;
2440         es->s_wtime = cpu_to_le32(get_seconds());
2441         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2442         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2443         BUFFER_TRACE(sbh, "marking dirty");
2444         mark_buffer_dirty(sbh);
2445         if (sync)
2446                 sync_dirty_buffer(sbh);
2447 }
2448
2449
2450 /*
2451  * Have we just finished recovery?  If so, and if we are mounting (or
2452  * remounting) the filesystem readonly, then we will end up with a
2453  * consistent fs on disk.  Record that fact.
2454  */
2455 static void ext4_mark_recovery_complete(struct super_block * sb,
2456                                         struct ext4_super_block * es)
2457 {
2458         journal_t *journal = EXT4_SB(sb)->s_journal;
2459
2460         jbd2_journal_lock_updates(journal);
2461         jbd2_journal_flush(journal);
2462         lock_super(sb);
2463         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2464             sb->s_flags & MS_RDONLY) {
2465                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2466                 sb->s_dirt = 0;
2467                 ext4_commit_super(sb, es, 1);
2468         }
2469         unlock_super(sb);
2470         jbd2_journal_unlock_updates(journal);
2471 }
2472
2473 /*
2474  * If we are mounting (or read-write remounting) a filesystem whose journal
2475  * has recorded an error from a previous lifetime, move that error to the
2476  * main filesystem now.
2477  */
2478 static void ext4_clear_journal_err(struct super_block * sb,
2479                                    struct ext4_super_block * es)
2480 {
2481         journal_t *journal;
2482         int j_errno;
2483         const char *errstr;
2484
2485         journal = EXT4_SB(sb)->s_journal;
2486
2487         /*
2488          * Now check for any error status which may have been recorded in the
2489          * journal by a prior ext4_error() or ext4_abort()
2490          */
2491
2492         j_errno = jbd2_journal_errno(journal);
2493         if (j_errno) {
2494                 char nbuf[16];
2495
2496                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2497                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2498                              "from previous mount: %s", errstr);
2499                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2500                              "filesystem check.");
2501
2502                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2503                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2504                 ext4_commit_super (sb, es, 1);
2505
2506                 jbd2_journal_clear_err(journal);
2507         }
2508 }
2509
2510 /*
2511  * Force the running and committing transactions to commit,
2512  * and wait on the commit.
2513  */
2514 int ext4_force_commit(struct super_block *sb)
2515 {
2516         journal_t *journal;
2517         int ret;
2518
2519         if (sb->s_flags & MS_RDONLY)
2520                 return 0;
2521
2522         journal = EXT4_SB(sb)->s_journal;
2523         sb->s_dirt = 0;
2524         ret = ext4_journal_force_commit(journal);
2525         return ret;
2526 }
2527
2528 /*
2529  * Ext4 always journals updates to the superblock itself, so we don't
2530  * have to propagate any other updates to the superblock on disk at this
2531  * point.  Just start an async writeback to get the buffers on their way
2532  * to the disk.
2533  *
2534  * This implicitly triggers the writebehind on sync().
2535  */
2536
2537 static void ext4_write_super (struct super_block * sb)
2538 {
2539         if (mutex_trylock(&sb->s_lock) != 0)
2540                 BUG();
2541         sb->s_dirt = 0;
2542 }
2543
2544 static int ext4_sync_fs(struct super_block *sb, int wait)
2545 {
2546         tid_t target;
2547
2548         sb->s_dirt = 0;
2549         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2550                 if (wait)
2551                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2552         }
2553         return 0;
2554 }
2555
2556 /*
2557  * LVM calls this function before a (read-only) snapshot is created.  This
2558  * gives us a chance to flush the journal completely and mark the fs clean.
2559  */
2560 static void ext4_write_super_lockfs(struct super_block *sb)
2561 {
2562         sb->s_dirt = 0;
2563
2564         if (!(sb->s_flags & MS_RDONLY)) {
2565                 journal_t *journal = EXT4_SB(sb)->s_journal;
2566
2567                 /* Now we set up the journal barrier. */
2568                 jbd2_journal_lock_updates(journal);
2569                 jbd2_journal_flush(journal);
2570
2571                 /* Journal blocked and flushed, clear needs_recovery flag. */
2572                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2573                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2574         }
2575 }
2576
2577 /*
2578  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2579  * flag here, even though the filesystem is not technically dirty yet.
2580  */
2581 static void ext4_unlockfs(struct super_block *sb)
2582 {
2583         if (!(sb->s_flags & MS_RDONLY)) {
2584                 lock_super(sb);
2585                 /* Reser the needs_recovery flag before the fs is unlocked. */
2586                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2587                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2588                 unlock_super(sb);
2589                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2590         }
2591 }
2592
2593 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2594 {
2595         struct ext4_super_block * es;
2596         struct ext4_sb_info *sbi = EXT4_SB(sb);
2597         ext4_fsblk_t n_blocks_count = 0;
2598         unsigned long old_sb_flags;
2599         struct ext4_mount_options old_opts;
2600         int err;
2601 #ifdef CONFIG_QUOTA
2602         int i;
2603 #endif
2604
2605         /* Store the original options */
2606         old_sb_flags = sb->s_flags;
2607         old_opts.s_mount_opt = sbi->s_mount_opt;
2608         old_opts.s_resuid = sbi->s_resuid;
2609         old_opts.s_resgid = sbi->s_resgid;
2610         old_opts.s_commit_interval = sbi->s_commit_interval;
2611 #ifdef CONFIG_QUOTA
2612         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2613         for (i = 0; i < MAXQUOTAS; i++)
2614                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2615 #endif
2616
2617         /*
2618          * Allow the "check" option to be passed as a remount option.
2619          */
2620         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2621                 err = -EINVAL;
2622                 goto restore_opts;
2623         }
2624
2625         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2626                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2627
2628         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2629                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2630
2631         es = sbi->s_es;
2632
2633         ext4_init_journal_params(sb, sbi->s_journal);
2634
2635         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2636                 n_blocks_count > ext4_blocks_count(es)) {
2637                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2638                         err = -EROFS;
2639                         goto restore_opts;
2640                 }
2641
2642                 if (*flags & MS_RDONLY) {
2643                         /*
2644                          * First of all, the unconditional stuff we have to do
2645                          * to disable replay of the journal when we next remount
2646                          */
2647                         sb->s_flags |= MS_RDONLY;
2648
2649                         /*
2650                          * OK, test if we are remounting a valid rw partition
2651                          * readonly, and if so set the rdonly flag and then
2652                          * mark the partition as valid again.
2653                          */
2654                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2655                             (sbi->s_mount_state & EXT4_VALID_FS))
2656                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2657
2658                         /*
2659                          * We have to unlock super so that we can wait for
2660                          * transactions.
2661                          */
2662                         unlock_super(sb);
2663                         ext4_mark_recovery_complete(sb, es);
2664                         lock_super(sb);
2665                 } else {
2666                         __le32 ret;
2667                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2668                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2669                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2670                                        "remount RDWR because of unsupported "
2671                                        "optional features (%x).\n",
2672                                        sb->s_id, le32_to_cpu(ret));
2673                                 err = -EROFS;
2674                                 goto restore_opts;
2675                         }
2676
2677                         /*
2678                          * If we have an unprocessed orphan list hanging
2679                          * around from a previously readonly bdev mount,
2680                          * require a full umount/remount for now.
2681                          */
2682                         if (es->s_last_orphan) {
2683                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2684                                        "remount RDWR because of unprocessed "
2685                                        "orphan inode list.  Please "
2686                                        "umount/remount instead.\n",
2687                                        sb->s_id);
2688                                 err = -EINVAL;
2689                                 goto restore_opts;
2690                         }
2691
2692                         /*
2693                          * Mounting a RDONLY partition read-write, so reread
2694                          * and store the current valid flag.  (It may have
2695                          * been changed by e2fsck since we originally mounted
2696                          * the partition.)
2697                          */
2698                         ext4_clear_journal_err(sb, es);
2699                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2700                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2701                                 goto restore_opts;
2702                         if (!ext4_setup_super (sb, es, 0))
2703                                 sb->s_flags &= ~MS_RDONLY;
2704                 }
2705         }
2706 #ifdef CONFIG_QUOTA
2707         /* Release old quota file names */
2708         for (i = 0; i < MAXQUOTAS; i++)
2709                 if (old_opts.s_qf_names[i] &&
2710                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2711                         kfree(old_opts.s_qf_names[i]);
2712 #endif
2713         return 0;
2714 restore_opts:
2715         sb->s_flags = old_sb_flags;
2716         sbi->s_mount_opt = old_opts.s_mount_opt;
2717         sbi->s_resuid = old_opts.s_resuid;
2718         sbi->s_resgid = old_opts.s_resgid;
2719         sbi->s_commit_interval = old_opts.s_commit_interval;
2720 #ifdef CONFIG_QUOTA
2721         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2722         for (i = 0; i < MAXQUOTAS; i++) {
2723                 if (sbi->s_qf_names[i] &&
2724                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2725                         kfree(sbi->s_qf_names[i]);
2726                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2727         }
2728 #endif
2729         return err;
2730 }
2731
2732 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2733 {
2734         struct super_block *sb = dentry->d_sb;
2735         struct ext4_sb_info *sbi = EXT4_SB(sb);
2736         struct ext4_super_block *es = sbi->s_es;
2737         u64 fsid;
2738
2739         if (test_opt(sb, MINIX_DF)) {
2740                 sbi->s_overhead_last = 0;
2741         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2742                 ext4_group_t ngroups = sbi->s_groups_count, i;
2743                 ext4_fsblk_t overhead = 0;
2744                 smp_rmb();
2745
2746                 /*
2747                  * Compute the overhead (FS structures).  This is constant
2748                  * for a given filesystem unless the number of block groups
2749                  * changes so we cache the previous value until it does.
2750                  */
2751
2752                 /*
2753                  * All of the blocks before first_data_block are
2754                  * overhead
2755                  */
2756                 overhead = le32_to_cpu(es->s_first_data_block);
2757
2758                 /*
2759                  * Add the overhead attributed to the superblock and
2760                  * block group descriptors.  If the sparse superblocks
2761                  * feature is turned on, then not all groups have this.
2762                  */
2763                 for (i = 0; i < ngroups; i++) {
2764                         overhead += ext4_bg_has_super(sb, i) +
2765                                 ext4_bg_num_gdb(sb, i);
2766                         cond_resched();
2767                 }
2768
2769                 /*
2770                  * Every block group has an inode bitmap, a block
2771                  * bitmap, and an inode table.
2772                  */
2773                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2774                 sbi->s_overhead_last = overhead;
2775                 smp_wmb();
2776                 sbi->s_blocks_last = ext4_blocks_count(es);
2777         }
2778
2779         buf->f_type = EXT4_SUPER_MAGIC;
2780         buf->f_bsize = sb->s_blocksize;
2781         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2782         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2783         ext4_free_blocks_count_set(es, buf->f_bfree);
2784         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2785         if (buf->f_bfree < ext4_r_blocks_count(es))
2786                 buf->f_bavail = 0;
2787         buf->f_files = le32_to_cpu(es->s_inodes_count);
2788         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2789         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2790         buf->f_namelen = EXT4_NAME_LEN;
2791         fsid = le64_to_cpup((void *)es->s_uuid) ^
2792                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2793         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2794         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2795         return 0;
2796 }
2797
2798 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2799  * is locked for write. Otherwise the are possible deadlocks:
2800  * Process 1                         Process 2
2801  * ext4_create()                     quota_sync()
2802  *   jbd2_journal_start()                   write_dquot()
2803  *   DQUOT_INIT()                        down(dqio_mutex)
2804  *     down(dqio_mutex)                    jbd2_journal_start()
2805  *
2806  */
2807
2808 #ifdef CONFIG_QUOTA
2809
2810 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2811 {
2812         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2813 }
2814
2815 static int ext4_dquot_initialize(struct inode *inode, int type)
2816 {
2817         handle_t *handle;
2818         int ret, err;
2819
2820         /* We may create quota structure so we need to reserve enough blocks */
2821         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2822         if (IS_ERR(handle))
2823                 return PTR_ERR(handle);
2824         ret = dquot_initialize(inode, type);
2825         err = ext4_journal_stop(handle);
2826         if (!ret)
2827                 ret = err;
2828         return ret;
2829 }
2830
2831 static int ext4_dquot_drop(struct inode *inode)
2832 {
2833         handle_t *handle;
2834         int ret, err;
2835
2836         /* We may delete quota structure so we need to reserve enough blocks */
2837         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2838         if (IS_ERR(handle))
2839                 return PTR_ERR(handle);
2840         ret = dquot_drop(inode);
2841         err = ext4_journal_stop(handle);
2842         if (!ret)
2843                 ret = err;
2844         return ret;
2845 }
2846
2847 static int ext4_write_dquot(struct dquot *dquot)
2848 {
2849         int ret, err;
2850         handle_t *handle;
2851         struct inode *inode;
2852
2853         inode = dquot_to_inode(dquot);
2854         handle = ext4_journal_start(inode,
2855                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2856         if (IS_ERR(handle))
2857                 return PTR_ERR(handle);
2858         ret = dquot_commit(dquot);
2859         err = ext4_journal_stop(handle);
2860         if (!ret)
2861                 ret = err;
2862         return ret;
2863 }
2864
2865 static int ext4_acquire_dquot(struct dquot *dquot)
2866 {
2867         int ret, err;
2868         handle_t *handle;
2869
2870         handle = ext4_journal_start(dquot_to_inode(dquot),
2871                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2872         if (IS_ERR(handle))
2873                 return PTR_ERR(handle);
2874         ret = dquot_acquire(dquot);
2875         err = ext4_journal_stop(handle);
2876         if (!ret)
2877                 ret = err;
2878         return ret;
2879 }
2880
2881 static int ext4_release_dquot(struct dquot *dquot)
2882 {
2883         int ret, err;
2884         handle_t *handle;
2885
2886         handle = ext4_journal_start(dquot_to_inode(dquot),
2887                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2888         if (IS_ERR(handle)) {
2889                 /* Release dquot anyway to avoid endless cycle in dqput() */
2890                 dquot_release(dquot);
2891                 return PTR_ERR(handle);
2892         }
2893         ret = dquot_release(dquot);
2894         err = ext4_journal_stop(handle);
2895         if (!ret)
2896                 ret = err;
2897         return ret;
2898 }
2899
2900 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2901 {
2902         /* Are we journalling quotas? */
2903         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2904             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2905                 dquot_mark_dquot_dirty(dquot);
2906                 return ext4_write_dquot(dquot);
2907         } else {
2908                 return dquot_mark_dquot_dirty(dquot);
2909         }
2910 }
2911
2912 static int ext4_write_info(struct super_block *sb, int type)
2913 {
2914         int ret, err;
2915         handle_t *handle;
2916
2917         /* Data block + inode block */
2918         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2919         if (IS_ERR(handle))
2920                 return PTR_ERR(handle);
2921         ret = dquot_commit_info(sb, type);
2922         err = ext4_journal_stop(handle);
2923         if (!ret)
2924                 ret = err;
2925         return ret;
2926 }
2927
2928 /*
2929  * Turn on quotas during mount time - we need to find
2930  * the quota file and such...
2931  */
2932 static int ext4_quota_on_mount(struct super_block *sb, int type)
2933 {
2934         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2935                         EXT4_SB(sb)->s_jquota_fmt, type);
2936 }
2937
2938 /*
2939  * Standard function to be called on quota_on
2940  */
2941 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2942                          char *path)
2943 {
2944         int err;
2945         struct nameidata nd;
2946
2947         if (!test_opt(sb, QUOTA))
2948                 return -EINVAL;
2949         /* Not journalling quota? */
2950         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2951             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2952                 return vfs_quota_on(sb, type, format_id, path);
2953         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2954         if (err)
2955                 return err;
2956         /* Quotafile not on the same filesystem? */
2957         if (nd.mnt->mnt_sb != sb) {
2958                 path_release(&nd);
2959                 return -EXDEV;
2960         }
2961         /* Quotafile not of fs root? */
2962         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2963                 printk(KERN_WARNING
2964                         "EXT4-fs: Quota file not on filesystem root. "
2965                         "Journalled quota will not work.\n");
2966         path_release(&nd);
2967         return vfs_quota_on(sb, type, format_id, path);
2968 }
2969
2970 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2971  * acquiring the locks... As quota files are never truncated and quota code
2972  * itself serializes the operations (and noone else should touch the files)
2973  * we don't have to be afraid of races */
2974 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2975                                size_t len, loff_t off)
2976 {
2977         struct inode *inode = sb_dqopt(sb)->files[type];
2978         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2979         int err = 0;
2980         int offset = off & (sb->s_blocksize - 1);
2981         int tocopy;
2982         size_t toread;
2983         struct buffer_head *bh;
2984         loff_t i_size = i_size_read(inode);
2985
2986         if (off > i_size)
2987                 return 0;
2988         if (off+len > i_size)
2989                 len = i_size-off;
2990         toread = len;
2991         while (toread > 0) {
2992                 tocopy = sb->s_blocksize - offset < toread ?
2993                                 sb->s_blocksize - offset : toread;
2994                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2995                 if (err)
2996                         return err;
2997                 if (!bh)        /* A hole? */
2998                         memset(data, 0, tocopy);
2999                 else
3000                         memcpy(data, bh->b_data+offset, tocopy);
3001                 brelse(bh);
3002                 offset = 0;
3003                 toread -= tocopy;
3004                 data += tocopy;
3005                 blk++;
3006         }
3007         return len;
3008 }
3009
3010 /* Write to quotafile (we know the transaction is already started and has
3011  * enough credits) */
3012 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3013                                 const char *data, size_t len, loff_t off)
3014 {
3015         struct inode *inode = sb_dqopt(sb)->files[type];
3016         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3017         int err = 0;
3018         int offset = off & (sb->s_blocksize - 1);
3019         int tocopy;
3020         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3021         size_t towrite = len;
3022         struct buffer_head *bh;
3023         handle_t *handle = journal_current_handle();
3024
3025         if (!handle) {
3026                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3027                         " cancelled because transaction is not started.\n",
3028                         (unsigned long long)off, (unsigned long long)len);
3029                 return -EIO;
3030         }
3031         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3032         while (towrite > 0) {
3033                 tocopy = sb->s_blocksize - offset < towrite ?
3034                                 sb->s_blocksize - offset : towrite;
3035                 bh = ext4_bread(handle, inode, blk, 1, &err);
3036                 if (!bh)
3037                         goto out;
3038                 if (journal_quota) {
3039                         err = ext4_journal_get_write_access(handle, bh);
3040                         if (err) {
3041                                 brelse(bh);
3042                                 goto out;
3043                         }
3044                 }
3045                 lock_buffer(bh);
3046                 memcpy(bh->b_data+offset, data, tocopy);
3047                 flush_dcache_page(bh->b_page);
3048                 unlock_buffer(bh);
3049                 if (journal_quota)
3050                         err = ext4_journal_dirty_metadata(handle, bh);
3051                 else {
3052                         /* Always do at least ordered writes for quotas */
3053                         err = ext4_journal_dirty_data(handle, bh);
3054                         mark_buffer_dirty(bh);
3055                 }
3056                 brelse(bh);
3057                 if (err)
3058                         goto out;
3059                 offset = 0;
3060                 towrite -= tocopy;
3061                 data += tocopy;
3062                 blk++;
3063         }
3064 out:
3065         if (len == towrite)
3066                 return err;
3067         if (inode->i_size < off+len-towrite) {
3068                 i_size_write(inode, off+len-towrite);
3069                 EXT4_I(inode)->i_disksize = inode->i_size;
3070         }
3071         inode->i_version++;
3072         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3073         ext4_mark_inode_dirty(handle, inode);
3074         mutex_unlock(&inode->i_mutex);
3075         return len - towrite;
3076 }
3077
3078 #endif
3079
3080 static int ext4_get_sb(struct file_system_type *fs_type,
3081         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3082 {
3083         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3084 }
3085
3086 static struct file_system_type ext4dev_fs_type = {
3087         .owner          = THIS_MODULE,
3088         .name           = "ext4dev",
3089         .get_sb         = ext4_get_sb,
3090         .kill_sb        = kill_block_super,
3091         .fs_flags       = FS_REQUIRES_DEV,
3092 };
3093
3094 static int __init init_ext4_fs(void)
3095 {
3096         int err = init_ext4_xattr();
3097         if (err)
3098                 return err;
3099         err = init_inodecache();
3100         if (err)
3101                 goto out1;
3102         err = register_filesystem(&ext4dev_fs_type);
3103         if (err)
3104                 goto out;
3105         return 0;
3106 out:
3107         destroy_inodecache();
3108 out1:
3109         exit_ext4_xattr();
3110         return err;
3111 }
3112
3113 static void __exit exit_ext4_fs(void)
3114 {
3115         unregister_filesystem(&ext4dev_fs_type);
3116         destroy_inodecache();
3117         exit_ext4_xattr();
3118 }
3119
3120 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3121 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3122 MODULE_LICENSE("GPL");
3123 module_init(init_ext4_fs)
3124 module_exit(exit_ext4_fs)