fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  50                              unsigned long journal_devnum);
  51 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  52                                unsigned int);
  53 static void ext4_commit_super (struct super_block * sb,
  54                                struct ext4_super_block * es,
  55                                int sync);
  56 static void ext4_mark_recovery_complete(struct super_block * sb,
  57                                         struct ext4_super_block * es);
  58 static void ext4_clear_journal_err(struct super_block * sb,
  59                                    struct ext4_super_block * es);
  60 static int ext4_sync_fs(struct super_block *sb, int wait);
  61 static const char *ext4_decode_error(struct super_block * sb, int errno,
  62                                      char nbuf[16]);
  63 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  64 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  65 static void ext4_unlockfs(struct super_block *sb);
  66 static void ext4_write_super (struct super_block * sb);
  67 static void ext4_write_super_lockfs(struct super_block *sb);
  68
  69
  70 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  71                                struct ext4_group_desc *bg)
  72 {
  73         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  74                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  75                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  76 }
  77
  78 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  79                                struct ext4_group_desc *bg)
  80 {
  81         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  82                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  83                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  84 }
  85
  86 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  87                               struct ext4_group_desc *bg)
  88 {
  89         return le32_to_cpu(bg->bg_inode_table_lo) |
  90                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  91                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  92 }
  93
  94 void ext4_block_bitmap_set(struct super_block *sb,
  95                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  96 {
  97         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  98         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  99                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 100 }
 101
 102 void ext4_inode_bitmap_set(struct super_block *sb,
 103                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 104 {
 105         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 106         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 107                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 108 }
 109
 110 void ext4_inode_table_set(struct super_block *sb,
 111                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 112 {
 113         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 114         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 115                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 116 }
 117
 118 /*
 119  * Wrappers for jbd2_journal_start/end.
 120  *
 121  * The only special thing we need to do here is to make sure that all
 122  * journal_end calls result in the superblock being marked dirty, so
 123  * that sync() will call the filesystem's write_super callback if
 124  * appropriate.
 125  */
 126 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 127 {
 128         journal_t *journal;
 129
 130         if (sb->s_flags & MS_RDONLY)
 131                 return ERR_PTR(-EROFS);
 132
 133         /* Special case here: if the journal has aborted behind our
 134          * backs (eg. EIO in the commit thread), then we still need to
 135          * take the FS itself readonly cleanly. */
 136         journal = EXT4_SB(sb)->s_journal;
 137         if (is_journal_aborted(journal)) {
 138                 ext4_abort(sb, __FUNCTION__,
 139                            "Detected aborted journal");
 140                 return ERR_PTR(-EROFS);
 141         }
 142
 143         return jbd2_journal_start(journal, nblocks);
 144 }
 145
 146 /*
 147  * The only special thing we need to do here is to make sure that all
 148  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 149  * that sync() will call the filesystem's write_super callback if
 150  * appropriate.
 151  */
 152 int __ext4_journal_stop(const char *where, handle_t *handle)
 153 {
 154         struct super_block *sb;
 155         int err;
 156         int rc;
 157
 158         sb = handle->h_transaction->t_journal->j_private;
 159         err = handle->h_err;
 160         rc = jbd2_journal_stop(handle);
 161
 162         if (!err)
 163                 err = rc;
 164         if (err)
 165                 __ext4_std_error(sb, where, err);
 166         return err;
 167 }
 168
 169 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 170                 struct buffer_head *bh, handle_t *handle, int err)
 171 {
 172         char nbuf[16];
 173         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 174
 175         if (bh)
 176                 BUFFER_TRACE(bh, "abort");
 177
 178         if (!handle->h_err)
 179                 handle->h_err = err;
 180
 181         if (is_handle_aborted(handle))
 182                 return;
 183
 184         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 185                caller, errstr, err_fn);
 186
 187         jbd2_journal_abort_handle(handle);
 188 }
 189
 190 /* Deal with the reporting of failure conditions on a filesystem such as
 191  * inconsistencies detected or read IO failures.
 192  *
 193  * On ext2, we can store the error state of the filesystem in the
 194  * superblock.  That is not possible on ext4, because we may have other
 195  * write ordering constraints on the superblock which prevent us from
 196  * writing it out straight away; and given that the journal is about to
 197  * be aborted, we can't rely on the current, or future, transactions to
 198  * write out the superblock safely.
 199  *
 200  * We'll just use the jbd2_journal_abort() error code to record an error in
 201  * the journal instead.  On recovery, the journal will compain about
 202  * that error until we've noted it down and cleared it.
 203  */
 204
 205 static void ext4_handle_error(struct super_block *sb)
 206 {
 207         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 208
 209         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 210         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 211
 212         if (sb->s_flags & MS_RDONLY)
 213                 return;
 214
 215         if (!test_opt (sb, ERRORS_CONT)) {
 216                 journal_t *journal = EXT4_SB(sb)->s_journal;
 217
 218                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 219                 if (journal)
 220                         jbd2_journal_abort(journal, -EIO);
 221         }
 222         if (test_opt (sb, ERRORS_RO)) {
 223                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 224                 sb->s_flags |= MS_RDONLY;
 225         }
 226         ext4_commit_super(sb, es, 1);
 227         if (test_opt(sb, ERRORS_PANIC))
 228                 panic("EXT4-fs (device %s): panic forced after error\n",
 229                         sb->s_id);
 230 }
 231
 232 void ext4_error (struct super_block * sb, const char * function,
 233                  const char * fmt, ...)
 234 {
 235         va_list args;
 236
 237         va_start(args, fmt);
 238         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 239         vprintk(fmt, args);
 240         printk("\n");
 241         va_end(args);
 242
 243         ext4_handle_error(sb);
 244 }
 245
 246 static const char *ext4_decode_error(struct super_block * sb, int errno,
 247                                      char nbuf[16])
 248 {
 249         char *errstr = NULL;
 250
 251         switch (errno) {
 252         case -EIO:
 253                 errstr = "IO failure";
 254                 break;
 255         case -ENOMEM:
 256                 errstr = "Out of memory";
 257                 break;
 258         case -EROFS:
 259                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 260                         errstr = "Journal has aborted";
 261                 else
 262                         errstr = "Readonly filesystem";
 263                 break;
 264         default:
 265                 /* If the caller passed in an extra buffer for unknown
 266                  * errors, textualise them now.  Else we just return
 267                  * NULL. */
 268                 if (nbuf) {
 269                         /* Check for truncated error codes... */
 270                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 271                                 errstr = nbuf;
 272                 }
 273                 break;
 274         }
 275
 276         return errstr;
 277 }
 278
 279 /* __ext4_std_error decodes expected errors from journaling functions
 280  * automatically and invokes the appropriate error response.  */
 281
 282 void __ext4_std_error (struct super_block * sb, const char * function,
 283                        int errno)
 284 {
 285         char nbuf[16];
 286         const char *errstr;
 287
 288         /* Special case: if the error is EROFS, and we're not already
 289          * inside a transaction, then there's really no point in logging
 290          * an error. */
 291         if (errno == -EROFS && journal_current_handle() == NULL &&
 292             (sb->s_flags & MS_RDONLY))
 293                 return;
 294
 295         errstr = ext4_decode_error(sb, errno, nbuf);
 296         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 297                 sb->s_id, function, errstr);
 298
 299         ext4_handle_error(sb);
 300 }
 301
 302 /*
 303  * ext4_abort is a much stronger failure handler than ext4_error.  The
 304  * abort function may be used to deal with unrecoverable failures such
 305  * as journal IO errors or ENOMEM at a critical moment in log management.
 306  *
 307  * We unconditionally force the filesystem into an ABORT|READONLY state,
 308  * unless the error response on the fs has been set to panic in which
 309  * case we take the easy way out and panic immediately.
 310  */
 311
 312 void ext4_abort (struct super_block * sb, const char * function,
 313                  const char * fmt, ...)
 314 {
 315         va_list args;
 316
 317         printk (KERN_CRIT "ext4_abort called.\n");
 318
 319         va_start(args, fmt);
 320         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 321         vprintk(fmt, args);
 322         printk("\n");
 323         va_end(args);
 324
 325         if (test_opt(sb, ERRORS_PANIC))
 326                 panic("EXT4-fs panic from previous error\n");
 327
 328         if (sb->s_flags & MS_RDONLY)
 329                 return;
 330
 331         printk(KERN_CRIT "Remounting filesystem read-only\n");
 332         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 333         sb->s_flags |= MS_RDONLY;
 334         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 335         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 336 }
 337
 338 void ext4_warning (struct super_block * sb, const char * function,
 339                    const char * fmt, ...)
 340 {
 341         va_list args;
 342
 343         va_start(args, fmt);
 344         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 345                sb->s_id, function);
 346         vprintk(fmt, args);
 347         printk("\n");
 348         va_end(args);
 349 }
 350
 351 void ext4_update_dynamic_rev(struct super_block *sb)
 352 {
 353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 354
 355         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 356                 return;
 357
 358         ext4_warning(sb, __FUNCTION__,
 359                      "updating to rev %d because of new feature flag, "
 360                      "running e2fsck is recommended",
 361                      EXT4_DYNAMIC_REV);
 362
 363         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 364         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 365         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 366         /* leave es->s_feature_*compat flags alone */
 367         /* es->s_uuid will be set by e2fsck if empty */
 368
 369         /*
 370          * The rest of the superblock fields should be zero, and if not it
 371          * means they are likely already in use, so leave them alone.  We
 372          * can leave it up to e2fsck to clean up any inconsistencies there.
 373          */
 374 }
 375
 376 int ext4_update_compat_feature(handle_t *handle,
 377                                         struct super_block *sb, __u32 compat)
 378 {
 379         int err = 0;
 380         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 381                 err = ext4_journal_get_write_access(handle,
 382                                 EXT4_SB(sb)->s_sbh);
 383                 if (err)
 384                         return err;
 385                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 386                 sb->s_dirt = 1;
 387                 handle->h_sync = 1;
 388                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 389                                         "call ext4_journal_dirty_met adata");
 390                 err = ext4_journal_dirty_metadata(handle,
 391                                 EXT4_SB(sb)->s_sbh);
 392         }
 393         return err;
 394 }
 395
 396 int ext4_update_rocompat_feature(handle_t *handle,
 397                                         struct super_block *sb, __u32 rocompat)
 398 {
 399         int err = 0;
 400         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 401                 err = ext4_journal_get_write_access(handle,
 402                                 EXT4_SB(sb)->s_sbh);
 403                 if (err)
 404                         return err;
 405                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 406                 sb->s_dirt = 1;
 407                 handle->h_sync = 1;
 408                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 409                                         "call ext4_journal_dirty_met adata");
 410                 err = ext4_journal_dirty_metadata(handle,
 411                                 EXT4_SB(sb)->s_sbh);
 412         }
 413         return err;
 414 }
 415
 416 int ext4_update_incompat_feature(handle_t *handle,
 417                                         struct super_block *sb, __u32 incompat)
 418 {
 419         int err = 0;
 420         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 421                 err = ext4_journal_get_write_access(handle,
 422                                 EXT4_SB(sb)->s_sbh);
 423                 if (err)
 424                         return err;
 425                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 426                 sb->s_dirt = 1;
 427                 handle->h_sync = 1;
 428                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 429                                         "call ext4_journal_dirty_met adata");
 430                 err = ext4_journal_dirty_metadata(handle,
 431                                 EXT4_SB(sb)->s_sbh);
 432         }
 433         return err;
 434 }
 435
 436 /*
 437  * Open the external journal device
 438  */
 439 static struct block_device *ext4_blkdev_get(dev_t dev)
 440 {
 441         struct block_device *bdev;
 442         char b[BDEVNAME_SIZE];
 443
 444         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 445         if (IS_ERR(bdev))
 446                 goto fail;
 447         return bdev;
 448
 449 fail:
 450         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 451                         __bdevname(dev, b), PTR_ERR(bdev));
 452         return NULL;
 453 }
 454
 455 /*
 456  * Release the journal device
 457  */
 458 static int ext4_blkdev_put(struct block_device *bdev)
 459 {
 460         bd_release(bdev);
 461         return blkdev_put(bdev);
 462 }
 463
 464 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 465 {
 466         struct block_device *bdev;
 467         int ret = -ENODEV;
 468
 469         bdev = sbi->journal_bdev;
 470         if (bdev) {
 471                 ret = ext4_blkdev_put(bdev);
 472                 sbi->journal_bdev = NULL;
 473         }
 474         return ret;
 475 }
 476
 477 static inline struct inode *orphan_list_entry(struct list_head *l)
 478 {
 479         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 480 }
 481
 482 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 483 {
 484         struct list_head *l;
 485
 486         printk(KERN_ERR "sb orphan head is %d\n",
 487                le32_to_cpu(sbi->s_es->s_last_orphan));
 488
 489         printk(KERN_ERR "sb_info orphan list:\n");
 490         list_for_each(l, &sbi->s_orphan) {
 491                 struct inode *inode = orphan_list_entry(l);
 492                 printk(KERN_ERR "  "
 493                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 494                        inode->i_sb->s_id, inode->i_ino, inode,
 495                        inode->i_mode, inode->i_nlink,
 496                        NEXT_ORPHAN(inode));
 497         }
 498 }
 499
 500 static void ext4_put_super (struct super_block * sb)
 501 {
 502         struct ext4_sb_info *sbi = EXT4_SB(sb);
 503         struct ext4_super_block *es = sbi->s_es;
 504         int i;
 505
 506         ext4_mb_release(sb);
 507         ext4_ext_release(sb);
 508         ext4_xattr_put_super(sb);
 509         jbd2_journal_destroy(sbi->s_journal);
 510         if (!(sb->s_flags & MS_RDONLY)) {
 511                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 512                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 513                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 514                 mark_buffer_dirty(sbi->s_sbh);
 515                 ext4_commit_super(sb, es, 1);
 516         }
 517
 518         for (i = 0; i < sbi->s_gdb_count; i++)
 519                 brelse(sbi->s_group_desc[i]);
 520         kfree(sbi->s_group_desc);
 521         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 522         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 523         percpu_counter_destroy(&sbi->s_dirs_counter);
 524         brelse(sbi->s_sbh);
 525 #ifdef CONFIG_QUOTA
 526         for (i = 0; i < MAXQUOTAS; i++)
 527                 kfree(sbi->s_qf_names[i]);
 528 #endif
 529
 530         /* Debugging code just in case the in-memory inode orphan list
 531          * isn't empty.  The on-disk one can be non-empty if we've
 532          * detected an error and taken the fs readonly, but the
 533          * in-memory list had better be clean by this point. */
 534         if (!list_empty(&sbi->s_orphan))
 535                 dump_orphan_list(sb, sbi);
 536         J_ASSERT(list_empty(&sbi->s_orphan));
 537
 538         invalidate_bdev(sb->s_bdev);
 539         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 540                 /*
 541                  * Invalidate the journal device's buffers.  We don't want them
 542                  * floating about in memory - the physical journal device may
 543                  * hotswapped, and it breaks the `ro-after' testing code.
 544                  */
 545                 sync_blockdev(sbi->journal_bdev);
 546                 invalidate_bdev(sbi->journal_bdev);
 547                 ext4_blkdev_remove(sbi);
 548         }
 549         sb->s_fs_info = NULL;
 550         kfree(sbi);
 551         return;
 552 }
 553
 554 static struct kmem_cache *ext4_inode_cachep;
 555
 556 /*
 557  * Called inside transaction, so use GFP_NOFS
 558  */
 559 static struct inode *ext4_alloc_inode(struct super_block *sb)
 560 {
 561         struct ext4_inode_info *ei;
 562
 563         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 564         if (!ei)
 565                 return NULL;
 566 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 567         ei->i_acl = EXT4_ACL_NOT_CACHED;
 568         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 569 #endif
 570         ei->i_block_alloc_info = NULL;
 571         ei->vfs_inode.i_version = 1;
 572         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 573         INIT_LIST_HEAD(&ei->i_prealloc_list);
 574         spin_lock_init(&ei->i_prealloc_lock);
 575         return &ei->vfs_inode;
 576 }
 577
 578 static void ext4_destroy_inode(struct inode *inode)
 579 {
 580         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 581                 printk("EXT4 Inode %p: orphan list check failed!\n",
 582                         EXT4_I(inode));
 583                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 584                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 585                                 true);
 586                 dump_stack();
 587         }
 588         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 589 }
 590
 591 static void init_once(struct kmem_cache *cachep, void *foo)
 592 {
 593         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 594
 595         INIT_LIST_HEAD(&ei->i_orphan);
 596 #ifdef CONFIG_EXT4DEV_FS_XATTR
 597         init_rwsem(&ei->xattr_sem);
 598 #endif
 599         init_rwsem(&ei->i_data_sem);
 600         inode_init_once(&ei->vfs_inode);
 601 }
 602
 603 static int init_inodecache(void)
 604 {
 605         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 606                                              sizeof(struct ext4_inode_info),
 607                                              0, (SLAB_RECLAIM_ACCOUNT|
 608                                                 SLAB_MEM_SPREAD),
 609                                              init_once);
 610         if (ext4_inode_cachep == NULL)
 611                 return -ENOMEM;
 612         return 0;
 613 }
 614
 615 static void destroy_inodecache(void)
 616 {
 617         kmem_cache_destroy(ext4_inode_cachep);
 618 }
 619
 620 static void ext4_clear_inode(struct inode *inode)
 621 {
 622         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 623 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 624         if (EXT4_I(inode)->i_acl &&
 625                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 626                 posix_acl_release(EXT4_I(inode)->i_acl);
 627                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 628         }
 629         if (EXT4_I(inode)->i_default_acl &&
 630                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 631                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 632                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 633         }
 634 #endif
 635         ext4_discard_reservation(inode);
 636         EXT4_I(inode)->i_block_alloc_info = NULL;
 637         if (unlikely(rsv))
 638                 kfree(rsv);
 639 }
 640
 641 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 642 {
 643 #if defined(CONFIG_QUOTA)
 644         struct ext4_sb_info *sbi = EXT4_SB(sb);
 645
 646         if (sbi->s_jquota_fmt)
 647                 seq_printf(seq, ",jqfmt=%s",
 648                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 649
 650         if (sbi->s_qf_names[USRQUOTA])
 651                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 652
 653         if (sbi->s_qf_names[GRPQUOTA])
 654                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 655
 656         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 657                 seq_puts(seq, ",usrquota");
 658
 659         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 660                 seq_puts(seq, ",grpquota");
 661 #endif
 662 }
 663
 664 /*
 665  * Show an option if
 666  *  - it's set to a non-default value OR
 667  *  - if the per-sb default is different from the global default
 668  */
 669 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 670 {
 671         int def_errors;
 672         unsigned long def_mount_opts;
 673         struct super_block *sb = vfs->mnt_sb;
 674         struct ext4_sb_info *sbi = EXT4_SB(sb);
 675         struct ext4_super_block *es = sbi->s_es;
 676
 677         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 678         def_errors     = le16_to_cpu(es->s_errors);
 679
 680         if (sbi->s_sb_block != 1)
 681                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 682         if (test_opt(sb, MINIX_DF))
 683                 seq_puts(seq, ",minixdf");
 684         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 685                 seq_puts(seq, ",grpid");
 686         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 687                 seq_puts(seq, ",nogrpid");
 688         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 689             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 690                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 691         }
 692         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 693             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 694                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 695         }
 696         if (test_opt(sb, ERRORS_RO)) {
 697                 if (def_errors == EXT4_ERRORS_PANIC ||
 698                     def_errors == EXT4_ERRORS_CONTINUE) {
 699                         seq_puts(seq, ",errors=remount-ro");
 700                 }
 701         }
 702         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 703                 seq_puts(seq, ",errors=continue");
 704         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 705                 seq_puts(seq, ",errors=panic");
 706         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 707                 seq_puts(seq, ",nouid32");
 708         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 709                 seq_puts(seq, ",debug");
 710         if (test_opt(sb, OLDALLOC))
 711                 seq_puts(seq, ",oldalloc");
 712 #ifdef CONFIG_EXT4DEV_FS_XATTR
 713         if (test_opt(sb, XATTR_USER) &&
 714                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 715                 seq_puts(seq, ",user_xattr");
 716         if (!test_opt(sb, XATTR_USER) &&
 717             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 718                 seq_puts(seq, ",nouser_xattr");
 719         }
 720 #endif
 721 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 722         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 723                 seq_puts(seq, ",acl");
 724         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 725                 seq_puts(seq, ",noacl");
 726 #endif
 727         if (!test_opt(sb, RESERVATION))
 728                 seq_puts(seq, ",noreservation");
 729         if (sbi->s_commit_interval) {
 730                 seq_printf(seq, ",commit=%u",
 731                            (unsigned) (sbi->s_commit_interval / HZ));
 732         }
 733         if (test_opt(sb, BARRIER))
 734                 seq_puts(seq, ",barrier=1");
 735         if (test_opt(sb, NOBH))
 736                 seq_puts(seq, ",nobh");
 737         if (!test_opt(sb, EXTENTS))
 738                 seq_puts(seq, ",noextents");
 739         if (!test_opt(sb, MBALLOC))
 740                 seq_puts(seq, ",nomballoc");
 741         if (test_opt(sb, I_VERSION))
 742                 seq_puts(seq, ",i_version");
 743
 744         /*
 745          * journal mode get enabled in different ways
 746          * So just print the value even if we didn't specify it
 747          */
 748         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 749                 seq_puts(seq, ",data=journal");
 750         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 751                 seq_puts(seq, ",data=ordered");
 752         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 753                 seq_puts(seq, ",data=writeback");
 754
 755         ext4_show_quota_options(seq, sb);
 756         return 0;
 757 }
 758
 759
 760 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 761                 u64 ino, u32 generation)
 762 {
 763         struct inode *inode;
 764
 765         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 766                 return ERR_PTR(-ESTALE);
 767         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 768                 return ERR_PTR(-ESTALE);
 769
 770         /* iget isn't really right if the inode is currently unallocated!!
 771          *
 772          * ext4_read_inode will return a bad_inode if the inode had been
 773          * deleted, so we should be safe.
 774          *
 775          * Currently we don't know the generation for parent directory, so
 776          * a generation of 0 means "accept any"
 777          */
 778         inode = iget(sb, ino);
 779         if (inode == NULL)
 780                 return ERR_PTR(-ENOMEM);
 781         if (is_bad_inode(inode) ||
 782             (generation && inode->i_generation != generation)) {
 783                 iput(inode);
 784                 return ERR_PTR(-ESTALE);
 785         }
 786
 787         return inode;
 788 }
 789
 790 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 791                 int fh_len, int fh_type)
 792 {
 793         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 794                                     ext4_nfs_get_inode);
 795 }
 796
 797 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 798                 int fh_len, int fh_type)
 799 {
 800         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 801                                     ext4_nfs_get_inode);
 802 }
 803
 804 #ifdef CONFIG_QUOTA
 805 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 806 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 807
 808 static int ext4_dquot_initialize(struct inode *inode, int type);
 809 static int ext4_dquot_drop(struct inode *inode);
 810 static int ext4_write_dquot(struct dquot *dquot);
 811 static int ext4_acquire_dquot(struct dquot *dquot);
 812 static int ext4_release_dquot(struct dquot *dquot);
 813 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 814 static int ext4_write_info(struct super_block *sb, int type);
 815 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 816 static int ext4_quota_on_mount(struct super_block *sb, int type);
 817 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 818                                size_t len, loff_t off);
 819 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 820                                 const char *data, size_t len, loff_t off);
 821
 822 static struct dquot_operations ext4_quota_operations = {
 823         .initialize     = ext4_dquot_initialize,
 824         .drop           = ext4_dquot_drop,
 825         .alloc_space    = dquot_alloc_space,
 826         .alloc_inode    = dquot_alloc_inode,
 827         .free_space     = dquot_free_space,
 828         .free_inode     = dquot_free_inode,
 829         .transfer       = dquot_transfer,
 830         .write_dquot    = ext4_write_dquot,
 831         .acquire_dquot  = ext4_acquire_dquot,
 832         .release_dquot  = ext4_release_dquot,
 833         .mark_dirty     = ext4_mark_dquot_dirty,
 834         .write_info     = ext4_write_info
 835 };
 836
 837 static struct quotactl_ops ext4_qctl_operations = {
 838         .quota_on       = ext4_quota_on,
 839         .quota_off      = vfs_quota_off,
 840         .quota_sync     = vfs_quota_sync,
 841         .get_info       = vfs_get_dqinfo,
 842         .set_info       = vfs_set_dqinfo,
 843         .get_dqblk      = vfs_get_dqblk,
 844         .set_dqblk      = vfs_set_dqblk
 845 };
 846 #endif
 847
 848 static const struct super_operations ext4_sops = {
 849         .alloc_inode    = ext4_alloc_inode,
 850         .destroy_inode  = ext4_destroy_inode,
 851         .read_inode     = ext4_read_inode,
 852         .write_inode    = ext4_write_inode,
 853         .dirty_inode    = ext4_dirty_inode,
 854         .delete_inode   = ext4_delete_inode,
 855         .put_super      = ext4_put_super,
 856         .write_super    = ext4_write_super,
 857         .sync_fs        = ext4_sync_fs,
 858         .write_super_lockfs = ext4_write_super_lockfs,
 859         .unlockfs       = ext4_unlockfs,
 860         .statfs         = ext4_statfs,
 861         .remount_fs     = ext4_remount,
 862         .clear_inode    = ext4_clear_inode,
 863         .show_options   = ext4_show_options,
 864 #ifdef CONFIG_QUOTA
 865         .quota_read     = ext4_quota_read,
 866         .quota_write    = ext4_quota_write,
 867 #endif
 868 };
 869
 870 static const struct export_operations ext4_export_ops = {
 871         .fh_to_dentry = ext4_fh_to_dentry,
 872         .fh_to_parent = ext4_fh_to_parent,
 873         .get_parent = ext4_get_parent,
 874 };
 875
 876 enum {
 877         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 878         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 879         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 880         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 881         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 882         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 883         Opt_journal_checksum, Opt_journal_async_commit,
 884         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 885         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 886         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 887         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 888         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 889         Opt_mballoc, Opt_nomballoc, Opt_stripe,
 890 };
 891
 892 static match_table_t tokens = {
 893         {Opt_bsd_df, "bsddf"},
 894         {Opt_minix_df, "minixdf"},
 895         {Opt_grpid, "grpid"},
 896         {Opt_grpid, "bsdgroups"},
 897         {Opt_nogrpid, "nogrpid"},
 898         {Opt_nogrpid, "sysvgroups"},
 899         {Opt_resgid, "resgid=%u"},
 900         {Opt_resuid, "resuid=%u"},
 901         {Opt_sb, "sb=%u"},
 902         {Opt_err_cont, "errors=continue"},
 903         {Opt_err_panic, "errors=panic"},
 904         {Opt_err_ro, "errors=remount-ro"},
 905         {Opt_nouid32, "nouid32"},
 906         {Opt_nocheck, "nocheck"},
 907         {Opt_nocheck, "check=none"},
 908         {Opt_debug, "debug"},
 909         {Opt_oldalloc, "oldalloc"},
 910         {Opt_orlov, "orlov"},
 911         {Opt_user_xattr, "user_xattr"},
 912         {Opt_nouser_xattr, "nouser_xattr"},
 913         {Opt_acl, "acl"},
 914         {Opt_noacl, "noacl"},
 915         {Opt_reservation, "reservation"},
 916         {Opt_noreservation, "noreservation"},
 917         {Opt_noload, "noload"},
 918         {Opt_nobh, "nobh"},
 919         {Opt_bh, "bh"},
 920         {Opt_commit, "commit=%u"},
 921         {Opt_journal_update, "journal=update"},
 922         {Opt_journal_inum, "journal=%u"},
 923         {Opt_journal_dev, "journal_dev=%u"},
 924         {Opt_journal_checksum, "journal_checksum"},
 925         {Opt_journal_async_commit, "journal_async_commit"},
 926         {Opt_abort, "abort"},
 927         {Opt_data_journal, "data=journal"},
 928         {Opt_data_ordered, "data=ordered"},
 929         {Opt_data_writeback, "data=writeback"},
 930         {Opt_offusrjquota, "usrjquota="},
 931         {Opt_usrjquota, "usrjquota=%s"},
 932         {Opt_offgrpjquota, "grpjquota="},
 933         {Opt_grpjquota, "grpjquota=%s"},
 934         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 935         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 936         {Opt_grpquota, "grpquota"},
 937         {Opt_noquota, "noquota"},
 938         {Opt_quota, "quota"},
 939         {Opt_usrquota, "usrquota"},
 940         {Opt_barrier, "barrier=%u"},
 941         {Opt_extents, "extents"},
 942         {Opt_noextents, "noextents"},
 943         {Opt_i_version, "i_version"},
 944         {Opt_mballoc, "mballoc"},
 945         {Opt_nomballoc, "nomballoc"},
 946         {Opt_stripe, "stripe=%u"},
 947         {Opt_err, NULL},
 948         {Opt_resize, "resize"},
 949 };
 950
 951 static ext4_fsblk_t get_sb_block(void **data)
 952 {
 953         ext4_fsblk_t    sb_block;
 954         char            *options = (char *) *data;
 955
 956         if (!options || strncmp(options, "sb=", 3) != 0)
 957                 return 1;       /* Default location */
 958         options += 3;
 959         /*todo: use simple_strtoll with >32bit ext4 */
 960         sb_block = simple_strtoul(options, &options, 0);
 961         if (*options && *options != ',') {
 962                 printk("EXT4-fs: Invalid sb specification: %s\n",
 963                        (char *) *data);
 964                 return 1;
 965         }
 966         if (*options == ',')
 967                 options++;
 968         *data = (void *) options;
 969         return sb_block;
 970 }
 971
 972 static int parse_options (char *options, struct super_block *sb,
 973                           unsigned int *inum, unsigned long *journal_devnum,
 974                           ext4_fsblk_t *n_blocks_count, int is_remount)
 975 {
 976         struct ext4_sb_info *sbi = EXT4_SB(sb);
 977         char * p;
 978         substring_t args[MAX_OPT_ARGS];
 979         int data_opt = 0;
 980         int option;
 981 #ifdef CONFIG_QUOTA
 982         int qtype;
 983         char *qname;
 984 #endif
 985
 986         if (!options)
 987                 return 1;
 988
 989         while ((p = strsep (&options, ",")) != NULL) {
 990                 int token;
 991                 if (!*p)
 992                         continue;
 993
 994                 token = match_token(p, tokens, args);
 995                 switch (token) {
 996                 case Opt_bsd_df:
 997                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 998                         break;
 999                 case Opt_minix_df:
1000                         set_opt (sbi->s_mount_opt, MINIX_DF);
1001                         break;
1002                 case Opt_grpid:
1003                         set_opt (sbi->s_mount_opt, GRPID);
1004                         break;
1005                 case Opt_nogrpid:
1006                         clear_opt (sbi->s_mount_opt, GRPID);
1007                         break;
1008                 case Opt_resuid:
1009                         if (match_int(&args[0], &option))
1010                                 return 0;
1011                         sbi->s_resuid = option;
1012                         break;
1013                 case Opt_resgid:
1014                         if (match_int(&args[0], &option))
1015                                 return 0;
1016                         sbi->s_resgid = option;
1017                         break;
1018                 case Opt_sb:
1019                         /* handled by get_sb_block() instead of here */
1020                         /* *sb_block = match_int(&args[0]); */
1021                         break;
1022                 case Opt_err_panic:
1023                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1024                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1025                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1026                         break;
1027                 case Opt_err_ro:
1028                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1029                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1030                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1031                         break;
1032                 case Opt_err_cont:
1033                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1034                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1035                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1036                         break;
1037                 case Opt_nouid32:
1038                         set_opt (sbi->s_mount_opt, NO_UID32);
1039                         break;
1040                 case Opt_nocheck:
1041                         clear_opt (sbi->s_mount_opt, CHECK);
1042                         break;
1043                 case Opt_debug:
1044                         set_opt (sbi->s_mount_opt, DEBUG);
1045                         break;
1046                 case Opt_oldalloc:
1047                         set_opt (sbi->s_mount_opt, OLDALLOC);
1048                         break;
1049                 case Opt_orlov:
1050                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1051                         break;
1052 #ifdef CONFIG_EXT4DEV_FS_XATTR
1053                 case Opt_user_xattr:
1054                         set_opt (sbi->s_mount_opt, XATTR_USER);
1055                         break;
1056                 case Opt_nouser_xattr:
1057                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1058                         break;
1059 #else
1060                 case Opt_user_xattr:
1061                 case Opt_nouser_xattr:
1062                         printk("EXT4 (no)user_xattr options not supported\n");
1063                         break;
1064 #endif
1065 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1066                 case Opt_acl:
1067                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1068                         break;
1069                 case Opt_noacl:
1070                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1071                         break;
1072 #else
1073                 case Opt_acl:
1074                 case Opt_noacl:
1075                         printk("EXT4 (no)acl options not supported\n");
1076                         break;
1077 #endif
1078                 case Opt_reservation:
1079                         set_opt(sbi->s_mount_opt, RESERVATION);
1080                         break;
1081                 case Opt_noreservation:
1082                         clear_opt(sbi->s_mount_opt, RESERVATION);
1083                         break;
1084                 case Opt_journal_update:
1085                         /* @@@ FIXME */
1086                         /* Eventually we will want to be able to create
1087                            a journal file here.  For now, only allow the
1088                            user to specify an existing inode to be the
1089                            journal file. */
1090                         if (is_remount) {
1091                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1092                                        "journal on remount\n");
1093                                 return 0;
1094                         }
1095                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1096                         break;
1097                 case Opt_journal_inum:
1098                         if (is_remount) {
1099                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1100                                        "journal on remount\n");
1101                                 return 0;
1102                         }
1103                         if (match_int(&args[0], &option))
1104                                 return 0;
1105                         *inum = option;
1106                         break;
1107                 case Opt_journal_dev:
1108                         if (is_remount) {
1109                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1110                                        "journal on remount\n");
1111                                 return 0;
1112                         }
1113                         if (match_int(&args[0], &option))
1114                                 return 0;
1115                         *journal_devnum = option;
1116                         break;
1117                 case Opt_journal_checksum:
1118                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1119                         break;
1120                 case Opt_journal_async_commit:
1121                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1122                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1123                         break;
1124                 case Opt_noload:
1125                         set_opt (sbi->s_mount_opt, NOLOAD);
1126                         break;
1127                 case Opt_commit:
1128                         if (match_int(&args[0], &option))
1129                                 return 0;
1130                         if (option < 0)
1131                                 return 0;
1132                         if (option == 0)
1133                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1134                         sbi->s_commit_interval = HZ * option;
1135                         break;
1136                 case Opt_data_journal:
1137                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1138                         goto datacheck;
1139                 case Opt_data_ordered:
1140                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1141                         goto datacheck;
1142                 case Opt_data_writeback:
1143                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1144                 datacheck:
1145                         if (is_remount) {
1146                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1147                                                 != data_opt) {
1148                                         printk(KERN_ERR
1149                                                 "EXT4-fs: cannot change data "
1150                                                 "mode on remount\n");
1151                                         return 0;
1152                                 }
1153                         } else {
1154                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1155                                 sbi->s_mount_opt |= data_opt;
1156                         }
1157                         break;
1158 #ifdef CONFIG_QUOTA
1159                 case Opt_usrjquota:
1160                         qtype = USRQUOTA;
1161                         goto set_qf_name;
1162                 case Opt_grpjquota:
1163                         qtype = GRPQUOTA;
1164 set_qf_name:
1165                         if (sb_any_quota_enabled(sb)) {
1166                                 printk(KERN_ERR
1167                                         "EXT4-fs: Cannot change journalled "
1168                                         "quota options when quota turned on.\n");
1169                                 return 0;
1170                         }
1171                         qname = match_strdup(&args[0]);
1172                         if (!qname) {
1173                                 printk(KERN_ERR
1174                                         "EXT4-fs: not enough memory for "
1175                                         "storing quotafile name.\n");
1176                                 return 0;
1177                         }
1178                         if (sbi->s_qf_names[qtype] &&
1179                             strcmp(sbi->s_qf_names[qtype], qname)) {
1180                                 printk(KERN_ERR
1181                                         "EXT4-fs: %s quota file already "
1182                                         "specified.\n", QTYPE2NAME(qtype));
1183                                 kfree(qname);
1184                                 return 0;
1185                         }
1186                         sbi->s_qf_names[qtype] = qname;
1187                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1188                                 printk(KERN_ERR
1189                                         "EXT4-fs: quotafile must be on "
1190                                         "filesystem root.\n");
1191                                 kfree(sbi->s_qf_names[qtype]);
1192                                 sbi->s_qf_names[qtype] = NULL;
1193                                 return 0;
1194                         }
1195                         set_opt(sbi->s_mount_opt, QUOTA);
1196                         break;
1197                 case Opt_offusrjquota:
1198                         qtype = USRQUOTA;
1199                         goto clear_qf_name;
1200                 case Opt_offgrpjquota:
1201                         qtype = GRPQUOTA;
1202 clear_qf_name:
1203                         if (sb_any_quota_enabled(sb)) {
1204                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1205                                         "journalled quota options when "
1206                                         "quota turned on.\n");
1207                                 return 0;
1208                         }
1209                         /*
1210                          * The space will be released later when all options
1211                          * are confirmed to be correct
1212                          */
1213                         sbi->s_qf_names[qtype] = NULL;
1214                         break;
1215                 case Opt_jqfmt_vfsold:
1216                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1217                         break;
1218                 case Opt_jqfmt_vfsv0:
1219                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1220                         break;
1221                 case Opt_quota:
1222                 case Opt_usrquota:
1223                         set_opt(sbi->s_mount_opt, QUOTA);
1224                         set_opt(sbi->s_mount_opt, USRQUOTA);
1225                         break;
1226                 case Opt_grpquota:
1227                         set_opt(sbi->s_mount_opt, QUOTA);
1228                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1229                         break;
1230                 case Opt_noquota:
1231                         if (sb_any_quota_enabled(sb)) {
1232                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1233                                         "options when quota turned on.\n");
1234                                 return 0;
1235                         }
1236                         clear_opt(sbi->s_mount_opt, QUOTA);
1237                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1238                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1239                         break;
1240 #else
1241                 case Opt_quota:
1242                 case Opt_usrquota:
1243                 case Opt_grpquota:
1244                 case Opt_usrjquota:
1245                 case Opt_grpjquota:
1246                 case Opt_offusrjquota:
1247                 case Opt_offgrpjquota:
1248                 case Opt_jqfmt_vfsold:
1249                 case Opt_jqfmt_vfsv0:
1250                         printk(KERN_ERR
1251                                 "EXT4-fs: journalled quota options not "
1252                                 "supported.\n");
1253                         break;
1254                 case Opt_noquota:
1255                         break;
1256 #endif
1257                 case Opt_abort:
1258                         set_opt(sbi->s_mount_opt, ABORT);
1259                         break;
1260                 case Opt_barrier:
1261                         if (match_int(&args[0], &option))
1262                                 return 0;
1263                         if (option)
1264                                 set_opt(sbi->s_mount_opt, BARRIER);
1265                         else
1266                                 clear_opt(sbi->s_mount_opt, BARRIER);
1267                         break;
1268                 case Opt_ignore:
1269                         break;
1270                 case Opt_resize:
1271                         if (!is_remount) {
1272                                 printk("EXT4-fs: resize option only available "
1273                                         "for remount\n");
1274                                 return 0;
1275                         }
1276                         if (match_int(&args[0], &option) != 0)
1277                                 return 0;
1278                         *n_blocks_count = option;
1279                         break;
1280                 case Opt_nobh:
1281                         set_opt(sbi->s_mount_opt, NOBH);
1282                         break;
1283                 case Opt_bh:
1284                         clear_opt(sbi->s_mount_opt, NOBH);
1285                         break;
1286                 case Opt_extents:
1287                         set_opt (sbi->s_mount_opt, EXTENTS);
1288                         break;
1289                 case Opt_noextents:
1290                         clear_opt (sbi->s_mount_opt, EXTENTS);
1291                         break;
1292                 case Opt_i_version:
1293                         set_opt(sbi->s_mount_opt, I_VERSION);
1294                         sb->s_flags |= MS_I_VERSION;
1295                         break;
1296                 case Opt_mballoc:
1297                         set_opt(sbi->s_mount_opt, MBALLOC);
1298                         break;
1299                 case Opt_nomballoc:
1300                         clear_opt(sbi->s_mount_opt, MBALLOC);
1301                         break;
1302                 case Opt_stripe:
1303                         if (match_int(&args[0], &option))
1304                                 return 0;
1305                         if (option < 0)
1306                                 return 0;
1307                         sbi->s_stripe = option;
1308                         break;
1309                 default:
1310                         printk (KERN_ERR
1311                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1312                                 "or missing value\n", p);
1313                         return 0;
1314                 }
1315         }
1316 #ifdef CONFIG_QUOTA
1317         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1318                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1319                      sbi->s_qf_names[USRQUOTA])
1320                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1321
1322                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1323                      sbi->s_qf_names[GRPQUOTA])
1324                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1325
1326                 if ((sbi->s_qf_names[USRQUOTA] &&
1327                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1328                     (sbi->s_qf_names[GRPQUOTA] &&
1329                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1330                         printk(KERN_ERR "EXT4-fs: old and new quota "
1331                                         "format mixing.\n");
1332                         return 0;
1333                 }
1334
1335                 if (!sbi->s_jquota_fmt) {
1336                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1337                                         "not specified.\n");
1338                         return 0;
1339                 }
1340         } else {
1341                 if (sbi->s_jquota_fmt) {
1342                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1343                                         "specified with no journalling "
1344                                         "enabled.\n");
1345                         return 0;
1346                 }
1347         }
1348 #endif
1349         return 1;
1350 }
1351
1352 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1353                             int read_only)
1354 {
1355         struct ext4_sb_info *sbi = EXT4_SB(sb);
1356         int res = 0;
1357
1358         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1359                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1360                         "forcing read-only mode\n");
1361                 res = MS_RDONLY;
1362         }
1363         if (read_only)
1364                 return res;
1365         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1366                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1367                         "running e2fsck is recommended\n");
1368         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1369                 printk (KERN_WARNING
1370                         "EXT4-fs warning: mounting fs with errors, "
1371                         "running e2fsck is recommended\n");
1372         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1373                  le16_to_cpu(es->s_mnt_count) >=
1374                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1375                 printk (KERN_WARNING
1376                         "EXT4-fs warning: maximal mount count reached, "
1377                         "running e2fsck is recommended\n");
1378         else if (le32_to_cpu(es->s_checkinterval) &&
1379                 (le32_to_cpu(es->s_lastcheck) +
1380                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1381                 printk (KERN_WARNING
1382                         "EXT4-fs warning: checktime reached, "
1383                         "running e2fsck is recommended\n");
1384 #if 0
1385                 /* @@@ We _will_ want to clear the valid bit if we find
1386                  * inconsistencies, to force a fsck at reboot.  But for
1387                  * a plain journaled filesystem we can keep it set as
1388                  * valid forever! :)
1389                  */
1390         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1391 #endif
1392         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1393                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1394         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1395         es->s_mtime = cpu_to_le32(get_seconds());
1396         ext4_update_dynamic_rev(sb);
1397         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1398
1399         ext4_commit_super(sb, es, 1);
1400         if (test_opt(sb, DEBUG))
1401                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1402                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1403                         sb->s_blocksize,
1404                         sbi->s_groups_count,
1405                         EXT4_BLOCKS_PER_GROUP(sb),
1406                         EXT4_INODES_PER_GROUP(sb),
1407                         sbi->s_mount_opt);
1408
1409         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1410         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1411                 char b[BDEVNAME_SIZE];
1412
1413                 printk("external journal on %s\n",
1414                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1415         } else {
1416                 printk("internal journal\n");
1417         }
1418         return res;
1419 }
1420
1421 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1422                             struct ext4_group_desc *gdp)
1423 {
1424         __u16 crc = 0;
1425
1426         if (sbi->s_es->s_feature_ro_compat &
1427             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1428                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1429                 __le32 le_group = cpu_to_le32(block_group);
1430
1431                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1432                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1433                 crc = crc16(crc, (__u8 *)gdp, offset);
1434                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1435                 /* for checksum of struct ext4_group_desc do the rest...*/
1436                 if ((sbi->s_es->s_feature_incompat &
1437                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1438                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1439                         crc = crc16(crc, (__u8 *)gdp + offset,
1440                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1441                                         offset);
1442         }
1443
1444         return cpu_to_le16(crc);
1445 }
1446
1447 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1448                                 struct ext4_group_desc *gdp)
1449 {
1450         if ((sbi->s_es->s_feature_ro_compat &
1451              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1452             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1453                 return 0;
1454
1455         return 1;
1456 }
1457
1458 /* Called at mount-time, super-block is locked */
1459 static int ext4_check_descriptors (struct super_block * sb)
1460 {
1461         struct ext4_sb_info *sbi = EXT4_SB(sb);
1462         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1463         ext4_fsblk_t last_block;
1464         ext4_fsblk_t block_bitmap;
1465         ext4_fsblk_t inode_bitmap;
1466         ext4_fsblk_t inode_table;
1467         struct ext4_group_desc * gdp = NULL;
1468         int desc_block = 0;
1469         int flexbg_flag = 0;
1470         ext4_group_t i;
1471
1472         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1473                 flexbg_flag = 1;
1474
1475         ext4_debug ("Checking group descriptors");
1476
1477         for (i = 0; i < sbi->s_groups_count; i++)
1478         {
1479                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1480                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1481                 else
1482                         last_block = first_block +
1483                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1484
1485                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1486                         gdp = (struct ext4_group_desc *)
1487                                         sbi->s_group_desc[desc_block++]->b_data;
1488                 block_bitmap = ext4_block_bitmap(sb, gdp);
1489                 if (block_bitmap < first_block || block_bitmap > last_block)
1490                 {
1491                         ext4_error (sb, "ext4_check_descriptors",
1492                                     "Block bitmap for group %lu"
1493                                     " not in group (block %llu)!",
1494                                     i, block_bitmap);
1495                         return 0;
1496                 }
1497                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1498                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1499                 {
1500                         ext4_error (sb, "ext4_check_descriptors",
1501                                     "Inode bitmap for group %lu"
1502                                     " not in group (block %llu)!",
1503                                     i, inode_bitmap);
1504                         return 0;
1505                 }
1506                 inode_table = ext4_inode_table(sb, gdp);
1507                 if (inode_table < first_block ||
1508                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1509                 {
1510                         ext4_error (sb, "ext4_check_descriptors",
1511                                     "Inode table for group %lu"
1512                                     " not in group (block %llu)!",
1513                                     i, inode_table);
1514                         return 0;
1515                 }
1516                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1517                         ext4_error(sb, __FUNCTION__,
1518                                    "Checksum for group %lu failed (%u!=%u)\n",
1519                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1520                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1521                         return 0;
1522                 }
1523                 if (!flexbg_flag)
1524                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1525                 gdp = (struct ext4_group_desc *)
1526                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1527         }
1528
1529         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1530         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1531         return 1;
1532 }
1533
1534 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1535  * the superblock) which were deleted from all directories, but held open by
1536  * a process at the time of a crash.  We walk the list and try to delete these
1537  * inodes at recovery time (only with a read-write filesystem).
1538  *
1539  * In order to keep the orphan inode chain consistent during traversal (in
1540  * case of crash during recovery), we link each inode into the superblock
1541  * orphan list_head and handle it the same way as an inode deletion during
1542  * normal operation (which journals the operations for us).
1543  *
1544  * We only do an iget() and an iput() on each inode, which is very safe if we
1545  * accidentally point at an in-use or already deleted inode.  The worst that
1546  * can happen in this case is that we get a "bit already cleared" message from
1547  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1548  * e2fsck was run on this filesystem, and it must have already done the orphan
1549  * inode cleanup for us, so we can safely abort without any further action.
1550  */
1551 static void ext4_orphan_cleanup (struct super_block * sb,
1552                                  struct ext4_super_block * es)
1553 {
1554         unsigned int s_flags = sb->s_flags;
1555         int nr_orphans = 0, nr_truncates = 0;
1556 #ifdef CONFIG_QUOTA
1557         int i;
1558 #endif
1559         if (!es->s_last_orphan) {
1560                 jbd_debug(4, "no orphan inodes to clean up\n");
1561                 return;
1562         }
1563
1564         if (bdev_read_only(sb->s_bdev)) {
1565                 printk(KERN_ERR "EXT4-fs: write access "
1566                         "unavailable, skipping orphan cleanup.\n");
1567                 return;
1568         }
1569
1570         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1571                 if (es->s_last_orphan)
1572                         jbd_debug(1, "Errors on filesystem, "
1573                                   "clearing orphan list.\n");
1574                 es->s_last_orphan = 0;
1575                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1576                 return;
1577         }
1578
1579         if (s_flags & MS_RDONLY) {
1580                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1581                        sb->s_id);
1582                 sb->s_flags &= ~MS_RDONLY;
1583         }
1584 #ifdef CONFIG_QUOTA
1585         /* Needed for iput() to work correctly and not trash data */
1586         sb->s_flags |= MS_ACTIVE;
1587         /* Turn on quotas so that they are updated correctly */
1588         for (i = 0; i < MAXQUOTAS; i++) {
1589                 if (EXT4_SB(sb)->s_qf_names[i]) {
1590                         int ret = ext4_quota_on_mount(sb, i);
1591                         if (ret < 0)
1592                                 printk(KERN_ERR
1593                                         "EXT4-fs: Cannot turn on journalled "
1594                                         "quota: error %d\n", ret);
1595                 }
1596         }
1597 #endif
1598
1599         while (es->s_last_orphan) {
1600                 struct inode *inode;
1601
1602                 if (!(inode =
1603                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1604                         es->s_last_orphan = 0;
1605                         break;
1606                 }
1607
1608                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1609                 DQUOT_INIT(inode);
1610                 if (inode->i_nlink) {
1611                         printk(KERN_DEBUG
1612                                 "%s: truncating inode %lu to %Ld bytes\n",
1613                                 __FUNCTION__, inode->i_ino, inode->i_size);
1614                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1615                                   inode->i_ino, inode->i_size);
1616                         ext4_truncate(inode);
1617                         nr_truncates++;
1618                 } else {
1619                         printk(KERN_DEBUG
1620                                 "%s: deleting unreferenced inode %lu\n",
1621                                 __FUNCTION__, inode->i_ino);
1622                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1623                                   inode->i_ino);
1624                         nr_orphans++;
1625                 }
1626                 iput(inode);  /* The delete magic happens here! */
1627         }
1628
1629 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1630
1631         if (nr_orphans)
1632                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1633                        sb->s_id, PLURAL(nr_orphans));
1634         if (nr_truncates)
1635                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1636                        sb->s_id, PLURAL(nr_truncates));
1637 #ifdef CONFIG_QUOTA
1638         /* Turn quotas off */
1639         for (i = 0; i < MAXQUOTAS; i++) {
1640                 if (sb_dqopt(sb)->files[i])
1641                         vfs_quota_off(sb, i);
1642         }
1643 #endif
1644         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1645 }
1646 /*
1647  * Maximal extent format file size.
1648  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1649  * extent format containers, within a sector_t, and within i_blocks
1650  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1651  * so that won't be a limiting factor.
1652  *
1653  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1654  */
1655 static loff_t ext4_max_size(int blkbits)
1656 {
1657         loff_t res;
1658         loff_t upper_limit = MAX_LFS_FILESIZE;
1659
1660         /* small i_blocks in vfs inode? */
1661         if (sizeof(blkcnt_t) < sizeof(u64)) {
1662                 /*
1663                  * CONFIG_LSF is not enabled implies the inode
1664                  * i_block represent total blocks in 512 bytes
1665                  * 32 == size of vfs inode i_blocks * 8
1666                  */
1667                 upper_limit = (1LL << 32) - 1;
1668
1669                 /* total blocks in file system block size */
1670                 upper_limit >>= (blkbits - 9);
1671                 upper_limit <<= blkbits;
1672         }
1673
1674         /* 32-bit extent-start container, ee_block */
1675         res = 1LL << 32;
1676         res <<= blkbits;
1677         res -= 1;
1678
1679         /* Sanity check against vm- & vfs- imposed limits */
1680         if (res > upper_limit)
1681                 res = upper_limit;
1682
1683         return res;
1684 }
1685
1686 /*
1687  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1688  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1689  * We need to be 1 filesystem block less than the 2^48 sector limit.
1690  */
1691 static loff_t ext4_max_bitmap_size(int bits)
1692 {
1693         loff_t res = EXT4_NDIR_BLOCKS;
1694         int meta_blocks;
1695         loff_t upper_limit;
1696         /* This is calculated to be the largest file size for a
1697          * dense, bitmapped file such that the total number of
1698          * sectors in the file, including data and all indirect blocks,
1699          * does not exceed 2^48 -1
1700          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1701          * total number of  512 bytes blocks of the file
1702          */
1703
1704         if (sizeof(blkcnt_t) < sizeof(u64)) {
1705                 /*
1706                  * CONFIG_LSF is not enabled implies the inode
1707                  * i_block represent total blocks in 512 bytes
1708                  * 32 == size of vfs inode i_blocks * 8
1709                  */
1710                 upper_limit = (1LL << 32) - 1;
1711
1712                 /* total blocks in file system block size */
1713                 upper_limit >>= (bits - 9);
1714
1715         } else {
1716                 /*
1717                  * We use 48 bit ext4_inode i_blocks
1718                  * With EXT4_HUGE_FILE_FL set the i_blocks
1719                  * represent total number of blocks in
1720                  * file system block size
1721                  */
1722                 upper_limit = (1LL << 48) - 1;
1723
1724         }
1725
1726         /* indirect blocks */
1727         meta_blocks = 1;
1728         /* double indirect blocks */
1729         meta_blocks += 1 + (1LL << (bits-2));
1730         /* tripple indirect blocks */
1731         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1732
1733         upper_limit -= meta_blocks;
1734         upper_limit <<= bits;
1735
1736         res += 1LL << (bits-2);
1737         res += 1LL << (2*(bits-2));
1738         res += 1LL << (3*(bits-2));
1739         res <<= bits;
1740         if (res > upper_limit)
1741                 res = upper_limit;
1742
1743         if (res > MAX_LFS_FILESIZE)
1744                 res = MAX_LFS_FILESIZE;
1745
1746         return res;
1747 }
1748
1749 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1750                                 ext4_fsblk_t logical_sb_block, int nr)
1751 {
1752         struct ext4_sb_info *sbi = EXT4_SB(sb);
1753         ext4_group_t bg, first_meta_bg;
1754         int has_super = 0;
1755
1756         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1757
1758         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1759             nr < first_meta_bg)
1760                 return logical_sb_block + nr + 1;
1761         bg = sbi->s_desc_per_block * nr;
1762         if (ext4_bg_has_super(sb, bg))
1763                 has_super = 1;
1764         return (has_super + ext4_group_first_block_no(sb, bg));
1765 }
1766
1767 /**
1768  * ext4_get_stripe_size: Get the stripe size.
1769  * @sbi: In memory super block info
1770  *
1771  * If we have specified it via mount option, then
1772  * use the mount option value. If the value specified at mount time is
1773  * greater than the blocks per group use the super block value.
1774  * If the super block value is greater than blocks per group return 0.
1775  * Allocator needs it be less than blocks per group.
1776  *
1777  */
1778 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1779 {
1780         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1781         unsigned long stripe_width =
1782                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1783
1784         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1785                 return sbi->s_stripe;
1786
1787         if (stripe_width <= sbi->s_blocks_per_group)
1788                 return stripe_width;
1789
1790         if (stride <= sbi->s_blocks_per_group)
1791                 return stride;
1792
1793         return 0;
1794 }
1795
1796 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1797                                 __releases(kernel_sem)
1798                                 __acquires(kernel_sem)
1799
1800 {
1801         struct buffer_head * bh;
1802         struct ext4_super_block *es = NULL;
1803         struct ext4_sb_info *sbi;
1804         ext4_fsblk_t block;
1805         ext4_fsblk_t sb_block = get_sb_block(&data);
1806         ext4_fsblk_t logical_sb_block;
1807         unsigned long offset = 0;
1808         unsigned int journal_inum = 0;
1809         unsigned long journal_devnum = 0;
1810         unsigned long def_mount_opts;
1811         struct inode *root;
1812         int blocksize;
1813         int hblock;
1814         int db_count;
1815         int i;
1816         int needs_recovery;
1817         __le32 features;
1818         __u64 blocks_count;
1819         int err;
1820
1821         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1822         if (!sbi)
1823                 return -ENOMEM;
1824         sb->s_fs_info = sbi;
1825         sbi->s_mount_opt = 0;
1826         sbi->s_resuid = EXT4_DEF_RESUID;
1827         sbi->s_resgid = EXT4_DEF_RESGID;
1828         sbi->s_sb_block = sb_block;
1829
1830         unlock_kernel();
1831
1832         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1833         if (!blocksize) {
1834                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1835                 goto out_fail;
1836         }
1837
1838         if (!sb_set_blocksize(sb, blocksize)) {
1839                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1840                 goto out_fail;
1841         }
1842
1843         /*
1844          * The ext4 superblock will not be buffer aligned for other than 1kB
1845          * block sizes.  We need to calculate the offset from buffer start.
1846          */
1847         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1848                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1849                 offset = do_div(logical_sb_block, blocksize);
1850         } else {
1851                 logical_sb_block = sb_block;
1852         }
1853
1854         if (!(bh = sb_bread(sb, logical_sb_block))) {
1855                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1856                 goto out_fail;
1857         }
1858         /*
1859          * Note: s_es must be initialized as soon as possible because
1860          *       some ext4 macro-instructions depend on its value
1861          */
1862         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1863         sbi->s_es = es;
1864         sb->s_magic = le16_to_cpu(es->s_magic);
1865         if (sb->s_magic != EXT4_SUPER_MAGIC)
1866                 goto cantfind_ext4;
1867
1868         /* Set defaults before we parse the mount options */
1869         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1870         if (def_mount_opts & EXT4_DEFM_DEBUG)
1871                 set_opt(sbi->s_mount_opt, DEBUG);
1872         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1873                 set_opt(sbi->s_mount_opt, GRPID);
1874         if (def_mount_opts & EXT4_DEFM_UID16)
1875                 set_opt(sbi->s_mount_opt, NO_UID32);
1876 #ifdef CONFIG_EXT4DEV_FS_XATTR
1877         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1878                 set_opt(sbi->s_mount_opt, XATTR_USER);
1879 #endif
1880 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1881         if (def_mount_opts & EXT4_DEFM_ACL)
1882                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1883 #endif
1884         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1885                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1886         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1887                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1888         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1889                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1890
1891         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1892                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1893         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1894                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1895         else
1896                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1897
1898         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1899         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1900
1901         set_opt(sbi->s_mount_opt, RESERVATION);
1902
1903         /*
1904          * turn on extents feature by default in ext4 filesystem
1905          * User -o noextents to turn it off
1906          */
1907         set_opt(sbi->s_mount_opt, EXTENTS);
1908         /*
1909          * turn on mballoc feature by default in ext4 filesystem
1910          * User -o nomballoc to turn it off
1911          */
1912         set_opt(sbi->s_mount_opt, MBALLOC);
1913
1914         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1915                             NULL, 0))
1916                 goto failed_mount;
1917
1918         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1919                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1920
1921         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1922             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1923              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1924              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1925                 printk(KERN_WARNING
1926                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1927                        "running e2fsck is recommended\n");
1928         /*
1929          * Check feature flags regardless of the revision level, since we
1930          * previously didn't change the revision level when setting the flags,
1931          * so there is a chance incompat flags are set on a rev 0 filesystem.
1932          */
1933         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1934         if (features) {
1935                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1936                        "unsupported optional features (%x).\n",
1937                        sb->s_id, le32_to_cpu(features));
1938                 goto failed_mount;
1939         }
1940         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1941         if (!(sb->s_flags & MS_RDONLY) && features) {
1942                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1943                        "unsupported optional features (%x).\n",
1944                        sb->s_id, le32_to_cpu(features));
1945                 goto failed_mount;
1946         }
1947         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
1948                 /*
1949                  * Large file size enabled file system can only be
1950                  * mount if kernel is build with CONFIG_LSF
1951                  */
1952                 if (sizeof(root->i_blocks) < sizeof(u64) &&
1953                                 !(sb->s_flags & MS_RDONLY)) {
1954                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
1955                                         "files cannot be mounted read-write "
1956                                         "without CONFIG_LSF.\n", sb->s_id);
1957                         goto failed_mount;
1958                 }
1959         }
1960         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1961
1962         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1963             blocksize > EXT4_MAX_BLOCK_SIZE) {
1964                 printk(KERN_ERR
1965                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1966                        blocksize, sb->s_id);
1967                 goto failed_mount;
1968         }
1969
1970         hblock = bdev_hardsect_size(sb->s_bdev);
1971         if (sb->s_blocksize != blocksize) {
1972                 /*
1973                  * Make sure the blocksize for the filesystem is larger
1974                  * than the hardware sectorsize for the machine.
1975                  */
1976                 if (blocksize < hblock) {
1977                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1978                                "device blocksize %d.\n", blocksize, hblock);
1979                         goto failed_mount;
1980                 }
1981
1982                 brelse (bh);
1983                 sb_set_blocksize(sb, blocksize);
1984                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1985                 offset = do_div(logical_sb_block, blocksize);
1986                 bh = sb_bread(sb, logical_sb_block);
1987                 if (!bh) {
1988                         printk(KERN_ERR
1989                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1990                         goto failed_mount;
1991                 }
1992                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1993                 sbi->s_es = es;
1994                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1995                         printk (KERN_ERR
1996                                 "EXT4-fs: Magic mismatch, very weird !\n");
1997                         goto failed_mount;
1998                 }
1999         }
2000
2001         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2002         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2003
2004         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2005                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2006                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2007         } else {
2008                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2009                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2010                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2011                     (!is_power_of_2(sbi->s_inode_size)) ||
2012                     (sbi->s_inode_size > blocksize)) {
2013                         printk (KERN_ERR
2014                                 "EXT4-fs: unsupported inode size: %d\n",
2015                                 sbi->s_inode_size);
2016                         goto failed_mount;
2017                 }
2018                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2019                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2020         }
2021         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2022         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2023                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2024                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2025                     !is_power_of_2(sbi->s_desc_size)) {
2026                         printk(KERN_ERR
2027                                "EXT4-fs: unsupported descriptor size %lu\n",
2028                                sbi->s_desc_size);
2029                         goto failed_mount;
2030                 }
2031         } else
2032                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2033         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2034         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2035         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2036                 goto cantfind_ext4;
2037         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2038         if (sbi->s_inodes_per_block == 0)
2039                 goto cantfind_ext4;
2040         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2041                                         sbi->s_inodes_per_block;
2042         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2043         sbi->s_sbh = bh;
2044         sbi->s_mount_state = le16_to_cpu(es->s_state);
2045         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2046         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2047         for (i=0; i < 4; i++)
2048                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2049         sbi->s_def_hash_version = es->s_def_hash_version;
2050
2051         if (sbi->s_blocks_per_group > blocksize * 8) {
2052                 printk (KERN_ERR
2053                         "EXT4-fs: #blocks per group too big: %lu\n",
2054                         sbi->s_blocks_per_group);
2055                 goto failed_mount;
2056         }
2057         if (sbi->s_inodes_per_group > blocksize * 8) {
2058                 printk (KERN_ERR
2059                         "EXT4-fs: #inodes per group too big: %lu\n",
2060                         sbi->s_inodes_per_group);
2061                 goto failed_mount;
2062         }
2063
2064         if (ext4_blocks_count(es) >
2065                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2066                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2067                         " too large to mount safely\n", sb->s_id);
2068                 if (sizeof(sector_t) < 8)
2069                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2070                                         "enabled\n");
2071                 goto failed_mount;
2072         }
2073
2074         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2075                 goto cantfind_ext4;
2076
2077         /* ensure blocks_count calculation below doesn't sign-extend */
2078         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2079             le32_to_cpu(es->s_first_data_block) + 1) {
2080                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2081                        "first data block %u, blocks per group %lu\n",
2082                         ext4_blocks_count(es),
2083                         le32_to_cpu(es->s_first_data_block),
2084                         EXT4_BLOCKS_PER_GROUP(sb));
2085                 goto failed_mount;
2086         }
2087         blocks_count = (ext4_blocks_count(es) -
2088                         le32_to_cpu(es->s_first_data_block) +
2089                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2090         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2091         sbi->s_groups_count = blocks_count;
2092         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2093                    EXT4_DESC_PER_BLOCK(sb);
2094         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2095                                     GFP_KERNEL);
2096         if (sbi->s_group_desc == NULL) {
2097                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2098                 goto failed_mount;
2099         }
2100
2101         bgl_lock_init(&sbi->s_blockgroup_lock);
2102
2103         for (i = 0; i < db_count; i++) {
2104                 block = descriptor_loc(sb, logical_sb_block, i);
2105                 sbi->s_group_desc[i] = sb_bread(sb, block);
2106                 if (!sbi->s_group_desc[i]) {
2107                         printk (KERN_ERR "EXT4-fs: "
2108                                 "can't read group descriptor %d\n", i);
2109                         db_count = i;
2110                         goto failed_mount2;
2111                 }
2112         }
2113         if (!ext4_check_descriptors (sb)) {
2114                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2115                 goto failed_mount2;
2116         }
2117         sbi->s_gdb_count = db_count;
2118         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2119         spin_lock_init(&sbi->s_next_gen_lock);
2120
2121         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2122                         ext4_count_free_blocks(sb));
2123         if (!err) {
2124                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2125                                 ext4_count_free_inodes(sb));
2126         }
2127         if (!err) {
2128                 err = percpu_counter_init(&sbi->s_dirs_counter,
2129                                 ext4_count_dirs(sb));
2130         }
2131         if (err) {
2132                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2133                 goto failed_mount3;
2134         }
2135
2136         /* per fileystem reservation list head & lock */
2137         spin_lock_init(&sbi->s_rsv_window_lock);
2138         sbi->s_rsv_window_root = RB_ROOT;
2139         /* Add a single, static dummy reservation to the start of the
2140          * reservation window list --- it gives us a placeholder for
2141          * append-at-start-of-list which makes the allocation logic
2142          * _much_ simpler. */
2143         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2144         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2145         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2146         sbi->s_rsv_window_head.rsv_goal_size = 0;
2147         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2148
2149         sbi->s_stripe = ext4_get_stripe_size(sbi);
2150
2151         /*
2152          * set up enough so that it can read an inode
2153          */
2154         sb->s_op = &ext4_sops;
2155         sb->s_export_op = &ext4_export_ops;
2156         sb->s_xattr = ext4_xattr_handlers;
2157 #ifdef CONFIG_QUOTA
2158         sb->s_qcop = &ext4_qctl_operations;
2159         sb->dq_op = &ext4_quota_operations;
2160 #endif
2161         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2162
2163         sb->s_root = NULL;
2164
2165         needs_recovery = (es->s_last_orphan != 0 ||
2166                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2167                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2168
2169         /*
2170          * The first inode we look at is the journal inode.  Don't try
2171          * root first: it may be modified in the journal!
2172          */
2173         if (!test_opt(sb, NOLOAD) &&
2174             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2175                 if (ext4_load_journal(sb, es, journal_devnum))
2176                         goto failed_mount3;
2177         } else if (journal_inum) {
2178                 if (ext4_create_journal(sb, es, journal_inum))
2179                         goto failed_mount3;
2180         } else {
2181                 if (!silent)
2182                         printk (KERN_ERR
2183                                 "ext4: No journal on filesystem on %s\n",
2184                                 sb->s_id);
2185                 goto failed_mount3;
2186         }
2187
2188         if (ext4_blocks_count(es) > 0xffffffffULL &&
2189             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2190                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2191                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2192                 goto failed_mount4;
2193         }
2194
2195         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2196                 jbd2_journal_set_features(sbi->s_journal,
2197                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2198                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2199         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2200                 jbd2_journal_set_features(sbi->s_journal,
2201                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2202                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2203                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2204         } else {
2205                 jbd2_journal_clear_features(sbi->s_journal,
2206                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2207                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2208         }
2209
2210         /* We have now updated the journal if required, so we can
2211          * validate the data journaling mode. */
2212         switch (test_opt(sb, DATA_FLAGS)) {
2213         case 0:
2214                 /* No mode set, assume a default based on the journal
2215                  * capabilities: ORDERED_DATA if the journal can
2216                  * cope, else JOURNAL_DATA
2217                  */
2218                 if (jbd2_journal_check_available_features
2219                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2220                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2221                 else
2222                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2223                 break;
2224
2225         case EXT4_MOUNT_ORDERED_DATA:
2226         case EXT4_MOUNT_WRITEBACK_DATA:
2227                 if (!jbd2_journal_check_available_features
2228                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2229                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2230                                "requested data journaling mode\n");
2231                         goto failed_mount4;
2232                 }
2233         default:
2234                 break;
2235         }
2236
2237         if (test_opt(sb, NOBH)) {
2238                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2239                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2240                                 "its supported only with writeback mode\n");
2241                         clear_opt(sbi->s_mount_opt, NOBH);
2242                 }
2243         }
2244         /*
2245          * The jbd2_journal_load will have done any necessary log recovery,
2246          * so we can safely mount the rest of the filesystem now.
2247          */
2248
2249         root = iget(sb, EXT4_ROOT_INO);
2250         sb->s_root = d_alloc_root(root);
2251         if (!sb->s_root) {
2252                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2253                 iput(root);
2254                 goto failed_mount4;
2255         }
2256         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2257                 dput(sb->s_root);
2258                 sb->s_root = NULL;
2259                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2260                 goto failed_mount4;
2261         }
2262
2263         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2264
2265         /* determine the minimum size of new large inodes, if present */
2266         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2267                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2268                                                      EXT4_GOOD_OLD_INODE_SIZE;
2269                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2270                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2271                         if (sbi->s_want_extra_isize <
2272                             le16_to_cpu(es->s_want_extra_isize))
2273                                 sbi->s_want_extra_isize =
2274                                         le16_to_cpu(es->s_want_extra_isize);
2275                         if (sbi->s_want_extra_isize <
2276                             le16_to_cpu(es->s_min_extra_isize))
2277                                 sbi->s_want_extra_isize =
2278                                         le16_to_cpu(es->s_min_extra_isize);
2279                 }
2280         }
2281         /* Check if enough inode space is available */
2282         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2283                                                         sbi->s_inode_size) {
2284                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2285                                                        EXT4_GOOD_OLD_INODE_SIZE;
2286                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2287                         "available.\n");
2288         }
2289
2290         /*
2291          * akpm: core read_super() calls in here with the superblock locked.
2292          * That deadlocks, because orphan cleanup needs to lock the superblock
2293          * in numerous places.  Here we just pop the lock - it's relatively
2294          * harmless, because we are now ready to accept write_super() requests,
2295          * and aviro says that's the only reason for hanging onto the
2296          * superblock lock.
2297          */
2298         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2299         ext4_orphan_cleanup(sb, es);
2300         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2301         if (needs_recovery)
2302                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2303         ext4_mark_recovery_complete(sb, es);
2304         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2305                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2306                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2307                 "writeback");
2308
2309         ext4_ext_init(sb);
2310         ext4_mb_init(sb, needs_recovery);
2311
2312         lock_kernel();
2313         return 0;
2314
2315 cantfind_ext4:
2316         if (!silent)
2317                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2318                        sb->s_id);
2319         goto failed_mount;
2320
2321 failed_mount4:
2322         jbd2_journal_destroy(sbi->s_journal);
2323 failed_mount3:
2324         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2325         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2326         percpu_counter_destroy(&sbi->s_dirs_counter);
2327 failed_mount2:
2328         for (i = 0; i < db_count; i++)
2329                 brelse(sbi->s_group_desc[i]);
2330         kfree(sbi->s_group_desc);
2331 failed_mount:
2332 #ifdef CONFIG_QUOTA
2333         for (i = 0; i < MAXQUOTAS; i++)
2334                 kfree(sbi->s_qf_names[i]);
2335 #endif
2336         ext4_blkdev_remove(sbi);
2337         brelse(bh);
2338 out_fail:
2339         sb->s_fs_info = NULL;
2340         kfree(sbi);
2341         lock_kernel();
2342         return -EINVAL;
2343 }
2344
2345 /*
2346  * Setup any per-fs journal parameters now.  We'll do this both on
2347  * initial mount, once the journal has been initialised but before we've
2348  * done any recovery; and again on any subsequent remount.
2349  */
2350 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2351 {
2352         struct ext4_sb_info *sbi = EXT4_SB(sb);
2353
2354         if (sbi->s_commit_interval)
2355                 journal->j_commit_interval = sbi->s_commit_interval;
2356         /* We could also set up an ext4-specific default for the commit
2357          * interval here, but for now we'll just fall back to the jbd
2358          * default. */
2359
2360         spin_lock(&journal->j_state_lock);
2361         if (test_opt(sb, BARRIER))
2362                 journal->j_flags |= JBD2_BARRIER;
2363         else
2364                 journal->j_flags &= ~JBD2_BARRIER;
2365         spin_unlock(&journal->j_state_lock);
2366 }
2367
2368 static journal_t *ext4_get_journal(struct super_block *sb,
2369                                    unsigned int journal_inum)
2370 {
2371         struct inode *journal_inode;
2372         journal_t *journal;
2373
2374         /* First, test for the existence of a valid inode on disk.  Bad
2375          * things happen if we iget() an unused inode, as the subsequent
2376          * iput() will try to delete it. */
2377
2378         journal_inode = iget(sb, journal_inum);
2379         if (!journal_inode) {
2380                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2381                 return NULL;
2382         }
2383         if (!journal_inode->i_nlink) {
2384                 make_bad_inode(journal_inode);
2385                 iput(journal_inode);
2386                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2387                 return NULL;
2388         }
2389
2390         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2391                   journal_inode, journal_inode->i_size);
2392         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2393                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2394                 iput(journal_inode);
2395                 return NULL;
2396         }
2397
2398         journal = jbd2_journal_init_inode(journal_inode);
2399         if (!journal) {
2400                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2401                 iput(journal_inode);
2402                 return NULL;
2403         }
2404         journal->j_private = sb;
2405         ext4_init_journal_params(sb, journal);
2406         return journal;
2407 }
2408
2409 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2410                                        dev_t j_dev)
2411 {
2412         struct buffer_head * bh;
2413         journal_t *journal;
2414         ext4_fsblk_t start;
2415         ext4_fsblk_t len;
2416         int hblock, blocksize;
2417         ext4_fsblk_t sb_block;
2418         unsigned long offset;
2419         struct ext4_super_block * es;
2420         struct block_device *bdev;
2421
2422         bdev = ext4_blkdev_get(j_dev);
2423         if (bdev == NULL)
2424                 return NULL;
2425
2426         if (bd_claim(bdev, sb)) {
2427                 printk(KERN_ERR
2428                         "EXT4: failed to claim external journal device.\n");
2429                 blkdev_put(bdev);
2430                 return NULL;
2431         }
2432
2433         blocksize = sb->s_blocksize;
2434         hblock = bdev_hardsect_size(bdev);
2435         if (blocksize < hblock) {
2436                 printk(KERN_ERR
2437                         "EXT4-fs: blocksize too small for journal device.\n");
2438                 goto out_bdev;
2439         }
2440
2441         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2442         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2443         set_blocksize(bdev, blocksize);
2444         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2445                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2446                        "external journal\n");
2447                 goto out_bdev;
2448         }
2449
2450         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2451         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2452             !(le32_to_cpu(es->s_feature_incompat) &
2453               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2454                 printk(KERN_ERR "EXT4-fs: external journal has "
2455                                         "bad superblock\n");
2456                 brelse(bh);
2457                 goto out_bdev;
2458         }
2459
2460         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2461                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2462                 brelse(bh);
2463                 goto out_bdev;
2464         }
2465
2466         len = ext4_blocks_count(es);
2467         start = sb_block + 1;
2468         brelse(bh);     /* we're done with the superblock */
2469
2470         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2471                                         start, len, blocksize);
2472         if (!journal) {
2473                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2474                 goto out_bdev;
2475         }
2476         journal->j_private = sb;
2477         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2478         wait_on_buffer(journal->j_sb_buffer);
2479         if (!buffer_uptodate(journal->j_sb_buffer)) {
2480                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2481                 goto out_journal;
2482         }
2483         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2484                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2485                                         "user (unsupported) - %d\n",
2486                         be32_to_cpu(journal->j_superblock->s_nr_users));
2487                 goto out_journal;
2488         }
2489         EXT4_SB(sb)->journal_bdev = bdev;
2490         ext4_init_journal_params(sb, journal);
2491         return journal;
2492 out_journal:
2493         jbd2_journal_destroy(journal);
2494 out_bdev:
2495         ext4_blkdev_put(bdev);
2496         return NULL;
2497 }
2498
2499 static int ext4_load_journal(struct super_block *sb,
2500                              struct ext4_super_block *es,
2501                              unsigned long journal_devnum)
2502 {
2503         journal_t *journal;
2504         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2505         dev_t journal_dev;
2506         int err = 0;
2507         int really_read_only;
2508
2509         if (journal_devnum &&
2510             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2511                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2512                         "numbers have changed\n");
2513                 journal_dev = new_decode_dev(journal_devnum);
2514         } else
2515                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2516
2517         really_read_only = bdev_read_only(sb->s_bdev);
2518
2519         /*
2520          * Are we loading a blank journal or performing recovery after a
2521          * crash?  For recovery, we need to check in advance whether we
2522          * can get read-write access to the device.
2523          */
2524
2525         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2526                 if (sb->s_flags & MS_RDONLY) {
2527                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2528                                         "required on readonly filesystem.\n");
2529                         if (really_read_only) {
2530                                 printk(KERN_ERR "EXT4-fs: write access "
2531                                         "unavailable, cannot proceed.\n");
2532                                 return -EROFS;
2533                         }
2534                         printk (KERN_INFO "EXT4-fs: write access will "
2535                                         "be enabled during recovery.\n");
2536                 }
2537         }
2538
2539         if (journal_inum && journal_dev) {
2540                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2541                        "and inode journals!\n");
2542                 return -EINVAL;
2543         }
2544
2545         if (journal_inum) {
2546                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2547                         return -EINVAL;
2548         } else {
2549                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2550                         return -EINVAL;
2551         }
2552
2553         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2554                 err = jbd2_journal_update_format(journal);
2555                 if (err)  {
2556                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2557                         jbd2_journal_destroy(journal);
2558                         return err;
2559                 }
2560         }
2561
2562         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2563                 err = jbd2_journal_wipe(journal, !really_read_only);
2564         if (!err)
2565                 err = jbd2_journal_load(journal);
2566
2567         if (err) {
2568                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2569                 jbd2_journal_destroy(journal);
2570                 return err;
2571         }
2572
2573         EXT4_SB(sb)->s_journal = journal;
2574         ext4_clear_journal_err(sb, es);
2575
2576         if (journal_devnum &&
2577             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2578                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2579                 sb->s_dirt = 1;
2580
2581                 /* Make sure we flush the recovery flag to disk. */
2582                 ext4_commit_super(sb, es, 1);
2583         }
2584
2585         return 0;
2586 }
2587
2588 static int ext4_create_journal(struct super_block * sb,
2589                                struct ext4_super_block * es,
2590                                unsigned int journal_inum)
2591 {
2592         journal_t *journal;
2593         int err;
2594
2595         if (sb->s_flags & MS_RDONLY) {
2596                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2597                                 "create journal.\n");
2598                 return -EROFS;
2599         }
2600
2601         journal = ext4_get_journal(sb, journal_inum);
2602         if (!journal)
2603                 return -EINVAL;
2604
2605         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2606                journal_inum);
2607
2608         err = jbd2_journal_create(journal);
2609         if (err) {
2610                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2611                 jbd2_journal_destroy(journal);
2612                 return -EIO;
2613         }
2614
2615         EXT4_SB(sb)->s_journal = journal;
2616
2617         ext4_update_dynamic_rev(sb);
2618         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2619         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2620
2621         es->s_journal_inum = cpu_to_le32(journal_inum);
2622         sb->s_dirt = 1;
2623
2624         /* Make sure we flush the recovery flag to disk. */
2625         ext4_commit_super(sb, es, 1);
2626
2627         return 0;
2628 }
2629
2630 static void ext4_commit_super (struct super_block * sb,
2631                                struct ext4_super_block * es,
2632                                int sync)
2633 {
2634         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2635
2636         if (!sbh)
2637                 return;
2638         es->s_wtime = cpu_to_le32(get_seconds());
2639         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2640         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2641         BUFFER_TRACE(sbh, "marking dirty");
2642         mark_buffer_dirty(sbh);
2643         if (sync)
2644                 sync_dirty_buffer(sbh);
2645 }
2646
2647
2648 /*
2649  * Have we just finished recovery?  If so, and if we are mounting (or
2650  * remounting) the filesystem readonly, then we will end up with a
2651  * consistent fs on disk.  Record that fact.
2652  */
2653 static void ext4_mark_recovery_complete(struct super_block * sb,
2654                                         struct ext4_super_block * es)
2655 {
2656         journal_t *journal = EXT4_SB(sb)->s_journal;
2657
2658         jbd2_journal_lock_updates(journal);
2659         jbd2_journal_flush(journal);
2660         lock_super(sb);
2661         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2662             sb->s_flags & MS_RDONLY) {
2663                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2664                 sb->s_dirt = 0;
2665                 ext4_commit_super(sb, es, 1);
2666         }
2667         unlock_super(sb);
2668         jbd2_journal_unlock_updates(journal);
2669 }
2670
2671 /*
2672  * If we are mounting (or read-write remounting) a filesystem whose journal
2673  * has recorded an error from a previous lifetime, move that error to the
2674  * main filesystem now.
2675  */
2676 static void ext4_clear_journal_err(struct super_block * sb,
2677                                    struct ext4_super_block * es)
2678 {
2679         journal_t *journal;
2680         int j_errno;
2681         const char *errstr;
2682
2683         journal = EXT4_SB(sb)->s_journal;
2684
2685         /*
2686          * Now check for any error status which may have been recorded in the
2687          * journal by a prior ext4_error() or ext4_abort()
2688          */
2689
2690         j_errno = jbd2_journal_errno(journal);
2691         if (j_errno) {
2692                 char nbuf[16];
2693
2694                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2695                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2696                              "from previous mount: %s", errstr);
2697                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2698                              "filesystem check.");
2699
2700                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2701                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2702                 ext4_commit_super (sb, es, 1);
2703
2704                 jbd2_journal_clear_err(journal);
2705         }
2706 }
2707
2708 /*
2709  * Force the running and committing transactions to commit,
2710  * and wait on the commit.
2711  */
2712 int ext4_force_commit(struct super_block *sb)
2713 {
2714         journal_t *journal;
2715         int ret;
2716
2717         if (sb->s_flags & MS_RDONLY)
2718                 return 0;
2719
2720         journal = EXT4_SB(sb)->s_journal;
2721         sb->s_dirt = 0;
2722         ret = ext4_journal_force_commit(journal);
2723         return ret;
2724 }
2725
2726 /*
2727  * Ext4 always journals updates to the superblock itself, so we don't
2728  * have to propagate any other updates to the superblock on disk at this
2729  * point.  Just start an async writeback to get the buffers on their way
2730  * to the disk.
2731  *
2732  * This implicitly triggers the writebehind on sync().
2733  */
2734
2735 static void ext4_write_super (struct super_block * sb)
2736 {
2737         if (mutex_trylock(&sb->s_lock) != 0)
2738                 BUG();
2739         sb->s_dirt = 0;
2740 }
2741
2742 static int ext4_sync_fs(struct super_block *sb, int wait)
2743 {
2744         tid_t target;
2745
2746         sb->s_dirt = 0;
2747         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2748                 if (wait)
2749                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2750         }
2751         return 0;
2752 }
2753
2754 /*
2755  * LVM calls this function before a (read-only) snapshot is created.  This
2756  * gives us a chance to flush the journal completely and mark the fs clean.
2757  */
2758 static void ext4_write_super_lockfs(struct super_block *sb)
2759 {
2760         sb->s_dirt = 0;
2761
2762         if (!(sb->s_flags & MS_RDONLY)) {
2763                 journal_t *journal = EXT4_SB(sb)->s_journal;
2764
2765                 /* Now we set up the journal barrier. */
2766                 jbd2_journal_lock_updates(journal);
2767                 jbd2_journal_flush(journal);
2768
2769                 /* Journal blocked and flushed, clear needs_recovery flag. */
2770                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2771                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2772         }
2773 }
2774
2775 /*
2776  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2777  * flag here, even though the filesystem is not technically dirty yet.
2778  */
2779 static void ext4_unlockfs(struct super_block *sb)
2780 {
2781         if (!(sb->s_flags & MS_RDONLY)) {
2782                 lock_super(sb);
2783                 /* Reser the needs_recovery flag before the fs is unlocked. */
2784                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2785                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2786                 unlock_super(sb);
2787                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2788         }
2789 }
2790
2791 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2792 {
2793         struct ext4_super_block * es;
2794         struct ext4_sb_info *sbi = EXT4_SB(sb);
2795         ext4_fsblk_t n_blocks_count = 0;
2796         unsigned long old_sb_flags;
2797         struct ext4_mount_options old_opts;
2798         int err;
2799 #ifdef CONFIG_QUOTA
2800         int i;
2801 #endif
2802
2803         /* Store the original options */
2804         old_sb_flags = sb->s_flags;
2805         old_opts.s_mount_opt = sbi->s_mount_opt;
2806         old_opts.s_resuid = sbi->s_resuid;
2807         old_opts.s_resgid = sbi->s_resgid;
2808         old_opts.s_commit_interval = sbi->s_commit_interval;
2809 #ifdef CONFIG_QUOTA
2810         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2811         for (i = 0; i < MAXQUOTAS; i++)
2812                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2813 #endif
2814
2815         /*
2816          * Allow the "check" option to be passed as a remount option.
2817          */
2818         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2819                 err = -EINVAL;
2820                 goto restore_opts;
2821         }
2822
2823         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2824                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2825
2826         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2827                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2828
2829         es = sbi->s_es;
2830
2831         ext4_init_journal_params(sb, sbi->s_journal);
2832
2833         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2834                 n_blocks_count > ext4_blocks_count(es)) {
2835                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2836                         err = -EROFS;
2837                         goto restore_opts;
2838                 }
2839
2840                 if (*flags & MS_RDONLY) {
2841                         /*
2842                          * First of all, the unconditional stuff we have to do
2843                          * to disable replay of the journal when we next remount
2844                          */
2845                         sb->s_flags |= MS_RDONLY;
2846
2847                         /*
2848                          * OK, test if we are remounting a valid rw partition
2849                          * readonly, and if so set the rdonly flag and then
2850                          * mark the partition as valid again.
2851                          */
2852                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2853                             (sbi->s_mount_state & EXT4_VALID_FS))
2854                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2855
2856                         /*
2857                          * We have to unlock super so that we can wait for
2858                          * transactions.
2859                          */
2860                         unlock_super(sb);
2861                         ext4_mark_recovery_complete(sb, es);
2862                         lock_super(sb);
2863                 } else {
2864                         __le32 ret;
2865                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2866                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2867                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2868                                        "remount RDWR because of unsupported "
2869                                        "optional features (%x).\n",
2870                                        sb->s_id, le32_to_cpu(ret));
2871                                 err = -EROFS;
2872                                 goto restore_opts;
2873                         }
2874
2875                         /*
2876                          * If we have an unprocessed orphan list hanging
2877                          * around from a previously readonly bdev mount,
2878                          * require a full umount/remount for now.
2879                          */
2880                         if (es->s_last_orphan) {
2881                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2882                                        "remount RDWR because of unprocessed "
2883                                        "orphan inode list.  Please "
2884                                        "umount/remount instead.\n",
2885                                        sb->s_id);
2886                                 err = -EINVAL;
2887                                 goto restore_opts;
2888                         }
2889
2890                         /*
2891                          * Mounting a RDONLY partition read-write, so reread
2892                          * and store the current valid flag.  (It may have
2893                          * been changed by e2fsck since we originally mounted
2894                          * the partition.)
2895                          */
2896                         ext4_clear_journal_err(sb, es);
2897                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2898                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2899                                 goto restore_opts;
2900                         if (!ext4_setup_super (sb, es, 0))
2901                                 sb->s_flags &= ~MS_RDONLY;
2902                 }
2903         }
2904 #ifdef CONFIG_QUOTA
2905         /* Release old quota file names */
2906         for (i = 0; i < MAXQUOTAS; i++)
2907                 if (old_opts.s_qf_names[i] &&
2908                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2909                         kfree(old_opts.s_qf_names[i]);
2910 #endif
2911         return 0;
2912 restore_opts:
2913         sb->s_flags = old_sb_flags;
2914         sbi->s_mount_opt = old_opts.s_mount_opt;
2915         sbi->s_resuid = old_opts.s_resuid;
2916         sbi->s_resgid = old_opts.s_resgid;
2917         sbi->s_commit_interval = old_opts.s_commit_interval;
2918 #ifdef CONFIG_QUOTA
2919         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2920         for (i = 0; i < MAXQUOTAS; i++) {
2921                 if (sbi->s_qf_names[i] &&
2922                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2923                         kfree(sbi->s_qf_names[i]);
2924                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2925         }
2926 #endif
2927         return err;
2928 }
2929
2930 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2931 {
2932         struct super_block *sb = dentry->d_sb;
2933         struct ext4_sb_info *sbi = EXT4_SB(sb);
2934         struct ext4_super_block *es = sbi->s_es;
2935         u64 fsid;
2936
2937         if (test_opt(sb, MINIX_DF)) {
2938                 sbi->s_overhead_last = 0;
2939         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2940                 ext4_group_t ngroups = sbi->s_groups_count, i;
2941                 ext4_fsblk_t overhead = 0;
2942                 smp_rmb();
2943
2944                 /*
2945                  * Compute the overhead (FS structures).  This is constant
2946                  * for a given filesystem unless the number of block groups
2947                  * changes so we cache the previous value until it does.
2948                  */
2949
2950                 /*
2951                  * All of the blocks before first_data_block are
2952                  * overhead
2953                  */
2954                 overhead = le32_to_cpu(es->s_first_data_block);
2955
2956                 /*
2957                  * Add the overhead attributed to the superblock and
2958                  * block group descriptors.  If the sparse superblocks
2959                  * feature is turned on, then not all groups have this.
2960                  */
2961                 for (i = 0; i < ngroups; i++) {
2962                         overhead += ext4_bg_has_super(sb, i) +
2963                                 ext4_bg_num_gdb(sb, i);
2964                         cond_resched();
2965                 }
2966
2967                 /*
2968                  * Every block group has an inode bitmap, a block
2969                  * bitmap, and an inode table.
2970                  */
2971                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2972                 sbi->s_overhead_last = overhead;
2973                 smp_wmb();
2974                 sbi->s_blocks_last = ext4_blocks_count(es);
2975         }
2976
2977         buf->f_type = EXT4_SUPER_MAGIC;
2978         buf->f_bsize = sb->s_blocksize;
2979         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2980         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2981         ext4_free_blocks_count_set(es, buf->f_bfree);
2982         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2983         if (buf->f_bfree < ext4_r_blocks_count(es))
2984                 buf->f_bavail = 0;
2985         buf->f_files = le32_to_cpu(es->s_inodes_count);
2986         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2987         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2988         buf->f_namelen = EXT4_NAME_LEN;
2989         fsid = le64_to_cpup((void *)es->s_uuid) ^
2990                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2991         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2992         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2993         return 0;
2994 }
2995
2996 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2997  * is locked for write. Otherwise the are possible deadlocks:
2998  * Process 1                         Process 2
2999  * ext4_create()                     quota_sync()
3000  *   jbd2_journal_start()                   write_dquot()
3001  *   DQUOT_INIT()                        down(dqio_mutex)
3002  *     down(dqio_mutex)                    jbd2_journal_start()
3003  *
3004  */
3005
3006 #ifdef CONFIG_QUOTA
3007
3008 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3009 {
3010         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3011 }
3012
3013 static int ext4_dquot_initialize(struct inode *inode, int type)
3014 {
3015         handle_t *handle;
3016         int ret, err;
3017
3018         /* We may create quota structure so we need to reserve enough blocks */
3019         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3020         if (IS_ERR(handle))
3021                 return PTR_ERR(handle);
3022         ret = dquot_initialize(inode, type);
3023         err = ext4_journal_stop(handle);
3024         if (!ret)
3025                 ret = err;
3026         return ret;
3027 }
3028
3029 static int ext4_dquot_drop(struct inode *inode)
3030 {
3031         handle_t *handle;
3032         int ret, err;
3033
3034         /* We may delete quota structure so we need to reserve enough blocks */
3035         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3036         if (IS_ERR(handle))
3037                 return PTR_ERR(handle);
3038         ret = dquot_drop(inode);
3039         err = ext4_journal_stop(handle);
3040         if (!ret)
3041                 ret = err;
3042         return ret;
3043 }
3044
3045 static int ext4_write_dquot(struct dquot *dquot)
3046 {
3047         int ret, err;
3048         handle_t *handle;
3049         struct inode *inode;
3050
3051         inode = dquot_to_inode(dquot);
3052         handle = ext4_journal_start(inode,
3053                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3054         if (IS_ERR(handle))
3055                 return PTR_ERR(handle);
3056         ret = dquot_commit(dquot);
3057         err = ext4_journal_stop(handle);
3058         if (!ret)
3059                 ret = err;
3060         return ret;
3061 }
3062
3063 static int ext4_acquire_dquot(struct dquot *dquot)
3064 {
3065         int ret, err;
3066         handle_t *handle;
3067
3068         handle = ext4_journal_start(dquot_to_inode(dquot),
3069                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3070         if (IS_ERR(handle))
3071                 return PTR_ERR(handle);
3072         ret = dquot_acquire(dquot);
3073         err = ext4_journal_stop(handle);
3074         if (!ret)
3075                 ret = err;
3076         return ret;
3077 }
3078
3079 static int ext4_release_dquot(struct dquot *dquot)
3080 {
3081         int ret, err;
3082         handle_t *handle;
3083
3084         handle = ext4_journal_start(dquot_to_inode(dquot),
3085                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3086         if (IS_ERR(handle)) {
3087                 /* Release dquot anyway to avoid endless cycle in dqput() */
3088                 dquot_release(dquot);
3089                 return PTR_ERR(handle);
3090         }
3091         ret = dquot_release(dquot);
3092         err = ext4_journal_stop(handle);
3093         if (!ret)
3094                 ret = err;
3095         return ret;
3096 }
3097
3098 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3099 {
3100         /* Are we journalling quotas? */
3101         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3102             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3103                 dquot_mark_dquot_dirty(dquot);
3104                 return ext4_write_dquot(dquot);
3105         } else {
3106                 return dquot_mark_dquot_dirty(dquot);
3107         }
3108 }
3109
3110 static int ext4_write_info(struct super_block *sb, int type)
3111 {
3112         int ret, err;
3113         handle_t *handle;
3114
3115         /* Data block + inode block */
3116         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3117         if (IS_ERR(handle))
3118                 return PTR_ERR(handle);
3119         ret = dquot_commit_info(sb, type);
3120         err = ext4_journal_stop(handle);
3121         if (!ret)
3122                 ret = err;
3123         return ret;
3124 }
3125
3126 /*
3127  * Turn on quotas during mount time - we need to find
3128  * the quota file and such...
3129  */
3130 static int ext4_quota_on_mount(struct super_block *sb, int type)
3131 {
3132         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3133                         EXT4_SB(sb)->s_jquota_fmt, type);
3134 }
3135
3136 /*
3137  * Standard function to be called on quota_on
3138  */
3139 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3140                          char *path)
3141 {
3142         int err;
3143         struct nameidata nd;
3144
3145         if (!test_opt(sb, QUOTA))
3146                 return -EINVAL;
3147         /* Not journalling quota? */
3148         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
3149             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
3150                 return vfs_quota_on(sb, type, format_id, path);
3151         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3152         if (err)
3153                 return err;
3154         /* Quotafile not on the same filesystem? */
3155         if (nd.mnt->mnt_sb != sb) {
3156                 path_release(&nd);
3157                 return -EXDEV;
3158         }
3159         /* Quotafile not of fs root? */
3160         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
3161                 printk(KERN_WARNING
3162                         "EXT4-fs: Quota file not on filesystem root. "
3163                         "Journalled quota will not work.\n");
3164         path_release(&nd);
3165         return vfs_quota_on(sb, type, format_id, path);
3166 }
3167
3168 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3169  * acquiring the locks... As quota files are never truncated and quota code
3170  * itself serializes the operations (and noone else should touch the files)
3171  * we don't have to be afraid of races */
3172 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3173                                size_t len, loff_t off)
3174 {
3175         struct inode *inode = sb_dqopt(sb)->files[type];
3176         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3177         int err = 0;
3178         int offset = off & (sb->s_blocksize - 1);
3179         int tocopy;
3180         size_t toread;
3181         struct buffer_head *bh;
3182         loff_t i_size = i_size_read(inode);
3183
3184         if (off > i_size)
3185                 return 0;
3186         if (off+len > i_size)
3187                 len = i_size-off;
3188         toread = len;
3189         while (toread > 0) {
3190                 tocopy = sb->s_blocksize - offset < toread ?
3191                                 sb->s_blocksize - offset : toread;
3192                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3193                 if (err)
3194                         return err;
3195                 if (!bh)        /* A hole? */
3196                         memset(data, 0, tocopy);
3197                 else
3198                         memcpy(data, bh->b_data+offset, tocopy);
3199                 brelse(bh);
3200                 offset = 0;
3201                 toread -= tocopy;
3202                 data += tocopy;
3203                 blk++;
3204         }
3205         return len;
3206 }
3207
3208 /* Write to quotafile (we know the transaction is already started and has
3209  * enough credits) */
3210 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3211                                 const char *data, size_t len, loff_t off)
3212 {
3213         struct inode *inode = sb_dqopt(sb)->files[type];
3214         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3215         int err = 0;
3216         int offset = off & (sb->s_blocksize - 1);
3217         int tocopy;
3218         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3219         size_t towrite = len;
3220         struct buffer_head *bh;
3221         handle_t *handle = journal_current_handle();
3222
3223         if (!handle) {
3224                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3225                         " cancelled because transaction is not started.\n",
3226                         (unsigned long long)off, (unsigned long long)len);
3227                 return -EIO;
3228         }
3229         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3230         while (towrite > 0) {
3231                 tocopy = sb->s_blocksize - offset < towrite ?
3232                                 sb->s_blocksize - offset : towrite;
3233                 bh = ext4_bread(handle, inode, blk, 1, &err);
3234                 if (!bh)
3235                         goto out;
3236                 if (journal_quota) {
3237                         err = ext4_journal_get_write_access(handle, bh);
3238                         if (err) {
3239                                 brelse(bh);
3240                                 goto out;
3241                         }
3242                 }
3243                 lock_buffer(bh);
3244                 memcpy(bh->b_data+offset, data, tocopy);
3245                 flush_dcache_page(bh->b_page);
3246                 unlock_buffer(bh);
3247                 if (journal_quota)
3248                         err = ext4_journal_dirty_metadata(handle, bh);
3249                 else {
3250                         /* Always do at least ordered writes for quotas */
3251                         err = ext4_journal_dirty_data(handle, bh);
3252                         mark_buffer_dirty(bh);
3253                 }
3254                 brelse(bh);
3255                 if (err)
3256                         goto out;
3257                 offset = 0;
3258                 towrite -= tocopy;
3259                 data += tocopy;
3260                 blk++;
3261         }
3262 out:
3263         if (len == towrite)
3264                 return err;
3265         if (inode->i_size < off+len-towrite) {
3266                 i_size_write(inode, off+len-towrite);
3267                 EXT4_I(inode)->i_disksize = inode->i_size;
3268         }
3269         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3270         ext4_mark_inode_dirty(handle, inode);
3271         mutex_unlock(&inode->i_mutex);
3272         return len - towrite;
3273 }
3274
3275 #endif
3276
3277 static int ext4_get_sb(struct file_system_type *fs_type,
3278         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3279 {
3280         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3281 }
3282
3283 static struct file_system_type ext4dev_fs_type = {
3284         .owner          = THIS_MODULE,
3285         .name           = "ext4dev",
3286         .get_sb         = ext4_get_sb,
3287         .kill_sb        = kill_block_super,
3288         .fs_flags       = FS_REQUIRES_DEV,
3289 };
3290
3291 static int __init init_ext4_fs(void)
3292 {
3293         int err;
3294
3295         err = init_ext4_mballoc();
3296         if (err)
3297                 return err;
3298
3299         err = init_ext4_xattr();
3300         if (err)
3301                 goto out2;
3302         err = init_inodecache();
3303         if (err)
3304                 goto out1;
3305         err = register_filesystem(&ext4dev_fs_type);
3306         if (err)
3307                 goto out;
3308         return 0;
3309 out:
3310         destroy_inodecache();
3311 out1:
3312         exit_ext4_xattr();
3313 out2:
3314         exit_ext4_mballoc();
3315         return err;
3316 }
3317
3318 static void __exit exit_ext4_fs(void)
3319 {
3320         unregister_filesystem(&ext4dev_fs_type);
3321         destroy_inodecache();
3322         exit_ext4_xattr();
3323         exit_ext4_mballoc();
3324 }
3325
3326 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3327 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3328 MODULE_LICENSE("GPL");
3329 module_init(init_ext4_fs)
3330 module_exit(exit_ext4_fs)