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

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