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_mb_release(sb);
 507         ext4_ext_release(sb);
 508         ext4_xattr_put_super(sb);
 509         jbd2_journal_destroy(sbi->s_journal);
 510         if (!(sb->s_flags & MS_RDONLY)) {
 511                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 512                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 513                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 514                 mark_buffer_dirty(sbi->s_sbh);
 515                 ext4_commit_super(sb, es, 1);
 516         }
 517
 518         for (i = 0; i < sbi->s_gdb_count; i++)
 519                 brelse(sbi->s_group_desc[i]);
 520         kfree(sbi->s_group_desc);
 521         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 522         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 523         percpu_counter_destroy(&sbi->s_dirs_counter);
 524         brelse(sbi->s_sbh);
 525 #ifdef CONFIG_QUOTA
 526         for (i = 0; i < MAXQUOTAS; i++)
 527                 kfree(sbi->s_qf_names[i]);
 528 #endif
 529
 530         /* Debugging code just in case the in-memory inode orphan list
 531          * isn't empty.  The on-disk one can be non-empty if we've
 532          * detected an error and taken the fs readonly, but the
 533          * in-memory list had better be clean by this point. */
 534         if (!list_empty(&sbi->s_orphan))
 535                 dump_orphan_list(sb, sbi);
 536         J_ASSERT(list_empty(&sbi->s_orphan));
 537
 538         invalidate_bdev(sb->s_bdev);
 539         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 540                 /*
 541                  * Invalidate the journal device's buffers.  We don't want them
 542                  * floating about in memory - the physical journal device may
 543                  * hotswapped, and it breaks the `ro-after' testing code.
 544                  */
 545                 sync_blockdev(sbi->journal_bdev);
 546                 invalidate_bdev(sbi->journal_bdev);
 547                 ext4_blkdev_remove(sbi);
 548         }
 549         sb->s_fs_info = NULL;
 550         kfree(sbi);
 551         return;
 552 }
 553
 554 static struct kmem_cache *ext4_inode_cachep;
 555
 556 /*
 557  * Called inside transaction, so use GFP_NOFS
 558  */
 559 static struct inode *ext4_alloc_inode(struct super_block *sb)
 560 {
 561         struct ext4_inode_info *ei;
 562
 563         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 564         if (!ei)
 565                 return NULL;
 566 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 567         ei->i_acl = EXT4_ACL_NOT_CACHED;
 568         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 569 #endif
 570         ei->i_block_alloc_info = NULL;
 571         ei->vfs_inode.i_version = 1;
 572         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 573         INIT_LIST_HEAD(&ei->i_prealloc_list);
 574         spin_lock_init(&ei->i_prealloc_lock);
 575         return &ei->vfs_inode;
 576 }
 577
 578 static void ext4_destroy_inode(struct inode *inode)
 579 {
 580         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 581                 printk("EXT4 Inode %p: orphan list check failed!\n",
 582                         EXT4_I(inode));
 583                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 584                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 585                                 true);
 586                 dump_stack();
 587         }
 588         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 589 }
 590
 591 static void init_once(struct kmem_cache *cachep, void *foo)
 592 {
 593         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 594
 595         INIT_LIST_HEAD(&ei->i_orphan);
 596 #ifdef CONFIG_EXT4DEV_FS_XATTR
 597         init_rwsem(&ei->xattr_sem);
 598 #endif
 599         init_rwsem(&ei->i_data_sem);
 600         inode_init_once(&ei->vfs_inode);
 601 }
 602
 603 static int init_inodecache(void)
 604 {
 605         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 606                                              sizeof(struct ext4_inode_info),
 607                                              0, (SLAB_RECLAIM_ACCOUNT|
 608                                                 SLAB_MEM_SPREAD),
 609                                              init_once);
 610         if (ext4_inode_cachep == NULL)
 611                 return -ENOMEM;
 612         return 0;
 613 }
 614
 615 static void destroy_inodecache(void)
 616 {
 617         kmem_cache_destroy(ext4_inode_cachep);
 618 }
 619
 620 static void ext4_clear_inode(struct inode *inode)
 621 {
 622         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 623 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 624         if (EXT4_I(inode)->i_acl &&
 625                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 626                 posix_acl_release(EXT4_I(inode)->i_acl);
 627                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 628         }
 629         if (EXT4_I(inode)->i_default_acl &&
 630                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 631                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 632                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 633         }
 634 #endif
 635         ext4_discard_reservation(inode);
 636         EXT4_I(inode)->i_block_alloc_info = NULL;
 637         if (unlikely(rsv))
 638                 kfree(rsv);
 639 }
 640
 641 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 642 {
 643 #if defined(CONFIG_QUOTA)
 644         struct ext4_sb_info *sbi = EXT4_SB(sb);
 645
 646         if (sbi->s_jquota_fmt)
 647                 seq_printf(seq, ",jqfmt=%s",
 648                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 649
 650         if (sbi->s_qf_names[USRQUOTA])
 651                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 652
 653         if (sbi->s_qf_names[GRPQUOTA])
 654                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 655
 656         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 657                 seq_puts(seq, ",usrquota");
 658
 659         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 660                 seq_puts(seq, ",grpquota");
 661 #endif
 662 }
 663
 664 /*
 665  * Show an option if
 666  *  - it's set to a non-default value OR
 667  *  - if the per-sb default is different from the global default
 668  */
 669 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 670 {
 671         int def_errors;
 672         unsigned long def_mount_opts;
 673         struct super_block *sb = vfs->mnt_sb;
 674         struct ext4_sb_info *sbi = EXT4_SB(sb);
 675         struct ext4_super_block *es = sbi->s_es;
 676
 677         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 678         def_errors     = le16_to_cpu(es->s_errors);
 679
 680         if (sbi->s_sb_block != 1)
 681                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 682         if (test_opt(sb, MINIX_DF))
 683                 seq_puts(seq, ",minixdf");
 684         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 685                 seq_puts(seq, ",grpid");
 686         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 687                 seq_puts(seq, ",nogrpid");
 688         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 689             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 690                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 691         }
 692         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 693             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 694                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 695         }
 696         if (test_opt(sb, ERRORS_RO)) {
 697                 if (def_errors == EXT4_ERRORS_PANIC ||
 698                     def_errors == EXT4_ERRORS_CONTINUE) {
 699                         seq_puts(seq, ",errors=remount-ro");
 700                 }
 701         }
 702         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 703                 seq_puts(seq, ",errors=continue");
 704         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 705                 seq_puts(seq, ",errors=panic");
 706         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 707                 seq_puts(seq, ",nouid32");
 708         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 709                 seq_puts(seq, ",debug");
 710         if (test_opt(sb, OLDALLOC))
 711                 seq_puts(seq, ",oldalloc");
 712 #ifdef CONFIG_EXT4DEV_FS_XATTR
 713         if (test_opt(sb, XATTR_USER) &&
 714                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 715                 seq_puts(seq, ",user_xattr");
 716         if (!test_opt(sb, XATTR_USER) &&
 717             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 718                 seq_puts(seq, ",nouser_xattr");
 719         }
 720 #endif
 721 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 722         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 723                 seq_puts(seq, ",acl");
 724         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 725                 seq_puts(seq, ",noacl");
 726 #endif
 727         if (!test_opt(sb, RESERVATION))
 728                 seq_puts(seq, ",noreservation");
 729         if (sbi->s_commit_interval) {
 730                 seq_printf(seq, ",commit=%u",
 731                            (unsigned) (sbi->s_commit_interval / HZ));
 732         }
 733         if (test_opt(sb, BARRIER))
 734                 seq_puts(seq, ",barrier=1");
 735         if (test_opt(sb, NOBH))
 736                 seq_puts(seq, ",nobh");
 737         if (!test_opt(sb, EXTENTS))
 738                 seq_puts(seq, ",noextents");
 739         if (!test_opt(sb, MBALLOC))
 740                 seq_puts(seq, ",nomballoc");
 741         if (test_opt(sb, I_VERSION))
 742                 seq_puts(seq, ",i_version");
 743
 744         if (sbi->s_stripe)
 745                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 746         /*
 747          * journal mode get enabled in different ways
 748          * So just print the value even if we didn't specify it
 749          */
 750         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 751                 seq_puts(seq, ",data=journal");
 752         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 753                 seq_puts(seq, ",data=ordered");
 754         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 755                 seq_puts(seq, ",data=writeback");
 756
 757         ext4_show_quota_options(seq, sb);
 758         return 0;
 759 }
 760
 761
 762 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 763                 u64 ino, u32 generation)
 764 {
 765         struct inode *inode;
 766
 767         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 768                 return ERR_PTR(-ESTALE);
 769         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 770                 return ERR_PTR(-ESTALE);
 771
 772         /* iget isn't really right if the inode is currently unallocated!!
 773          *
 774          * ext4_read_inode will return a bad_inode if the inode had been
 775          * deleted, so we should be safe.
 776          *
 777          * Currently we don't know the generation for parent directory, so
 778          * a generation of 0 means "accept any"
 779          */
 780         inode = ext4_iget(sb, ino);
 781         if (IS_ERR(inode))
 782                 return ERR_CAST(inode);
 783         if (generation && inode->i_generation != generation) {
 784                 iput(inode);
 785                 return ERR_PTR(-ESTALE);
 786         }
 787
 788         return inode;
 789 }
 790
 791 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 792                 int fh_len, int fh_type)
 793 {
 794         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 795                                     ext4_nfs_get_inode);
 796 }
 797
 798 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 799                 int fh_len, int fh_type)
 800 {
 801         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 802                                     ext4_nfs_get_inode);
 803 }
 804
 805 #ifdef CONFIG_QUOTA
 806 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 807 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 808
 809 static int ext4_dquot_initialize(struct inode *inode, int type);
 810 static int ext4_dquot_drop(struct inode *inode);
 811 static int ext4_write_dquot(struct dquot *dquot);
 812 static int ext4_acquire_dquot(struct dquot *dquot);
 813 static int ext4_release_dquot(struct dquot *dquot);
 814 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 815 static int ext4_write_info(struct super_block *sb, int type);
 816 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 817                                 char *path, int remount);
 818 static int ext4_quota_on_mount(struct super_block *sb, int type);
 819 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 820                                size_t len, loff_t off);
 821 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 822                                 const char *data, size_t len, loff_t off);
 823
 824 static struct dquot_operations ext4_quota_operations = {
 825         .initialize     = ext4_dquot_initialize,
 826         .drop           = ext4_dquot_drop,
 827         .alloc_space    = dquot_alloc_space,
 828         .alloc_inode    = dquot_alloc_inode,
 829         .free_space     = dquot_free_space,
 830         .free_inode     = dquot_free_inode,
 831         .transfer       = dquot_transfer,
 832         .write_dquot    = ext4_write_dquot,
 833         .acquire_dquot  = ext4_acquire_dquot,
 834         .release_dquot  = ext4_release_dquot,
 835         .mark_dirty     = ext4_mark_dquot_dirty,
 836         .write_info     = ext4_write_info
 837 };
 838
 839 static struct quotactl_ops ext4_qctl_operations = {
 840         .quota_on       = ext4_quota_on,
 841         .quota_off      = vfs_quota_off,
 842         .quota_sync     = vfs_quota_sync,
 843         .get_info       = vfs_get_dqinfo,
 844         .set_info       = vfs_set_dqinfo,
 845         .get_dqblk      = vfs_get_dqblk,
 846         .set_dqblk      = vfs_set_dqblk
 847 };
 848 #endif
 849
 850 static const struct super_operations ext4_sops = {
 851         .alloc_inode    = ext4_alloc_inode,
 852         .destroy_inode  = ext4_destroy_inode,
 853         .write_inode    = ext4_write_inode,
 854         .dirty_inode    = ext4_dirty_inode,
 855         .delete_inode   = ext4_delete_inode,
 856         .put_super      = ext4_put_super,
 857         .write_super    = ext4_write_super,
 858         .sync_fs        = ext4_sync_fs,
 859         .write_super_lockfs = ext4_write_super_lockfs,
 860         .unlockfs       = ext4_unlockfs,
 861         .statfs         = ext4_statfs,
 862         .remount_fs     = ext4_remount,
 863         .clear_inode    = ext4_clear_inode,
 864         .show_options   = ext4_show_options,
 865 #ifdef CONFIG_QUOTA
 866         .quota_read     = ext4_quota_read,
 867         .quota_write    = ext4_quota_write,
 868 #endif
 869 };
 870
 871 static const struct export_operations ext4_export_ops = {
 872         .fh_to_dentry = ext4_fh_to_dentry,
 873         .fh_to_parent = ext4_fh_to_parent,
 874         .get_parent = ext4_get_parent,
 875 };
 876
 877 enum {
 878         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 879         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 880         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 881         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 882         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 883         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 884         Opt_journal_checksum, Opt_journal_async_commit,
 885         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 886         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 887         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 888         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 889         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 890         Opt_mballoc, Opt_nomballoc, Opt_stripe,
 891 };
 892
 893 static match_table_t tokens = {
 894         {Opt_bsd_df, "bsddf"},
 895         {Opt_minix_df, "minixdf"},
 896         {Opt_grpid, "grpid"},
 897         {Opt_grpid, "bsdgroups"},
 898         {Opt_nogrpid, "nogrpid"},
 899         {Opt_nogrpid, "sysvgroups"},
 900         {Opt_resgid, "resgid=%u"},
 901         {Opt_resuid, "resuid=%u"},
 902         {Opt_sb, "sb=%u"},
 903         {Opt_err_cont, "errors=continue"},
 904         {Opt_err_panic, "errors=panic"},
 905         {Opt_err_ro, "errors=remount-ro"},
 906         {Opt_nouid32, "nouid32"},
 907         {Opt_nocheck, "nocheck"},
 908         {Opt_nocheck, "check=none"},
 909         {Opt_debug, "debug"},
 910         {Opt_oldalloc, "oldalloc"},
 911         {Opt_orlov, "orlov"},
 912         {Opt_user_xattr, "user_xattr"},
 913         {Opt_nouser_xattr, "nouser_xattr"},
 914         {Opt_acl, "acl"},
 915         {Opt_noacl, "noacl"},
 916         {Opt_reservation, "reservation"},
 917         {Opt_noreservation, "noreservation"},
 918         {Opt_noload, "noload"},
 919         {Opt_nobh, "nobh"},
 920         {Opt_bh, "bh"},
 921         {Opt_commit, "commit=%u"},
 922         {Opt_journal_update, "journal=update"},
 923         {Opt_journal_inum, "journal=%u"},
 924         {Opt_journal_dev, "journal_dev=%u"},
 925         {Opt_journal_checksum, "journal_checksum"},
 926         {Opt_journal_async_commit, "journal_async_commit"},
 927         {Opt_abort, "abort"},
 928         {Opt_data_journal, "data=journal"},
 929         {Opt_data_ordered, "data=ordered"},
 930         {Opt_data_writeback, "data=writeback"},
 931         {Opt_offusrjquota, "usrjquota="},
 932         {Opt_usrjquota, "usrjquota=%s"},
 933         {Opt_offgrpjquota, "grpjquota="},
 934         {Opt_grpjquota, "grpjquota=%s"},
 935         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 936         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 937         {Opt_grpquota, "grpquota"},
 938         {Opt_noquota, "noquota"},
 939         {Opt_quota, "quota"},
 940         {Opt_usrquota, "usrquota"},
 941         {Opt_barrier, "barrier=%u"},
 942         {Opt_extents, "extents"},
 943         {Opt_noextents, "noextents"},
 944         {Opt_i_version, "i_version"},
 945         {Opt_mballoc, "mballoc"},
 946         {Opt_nomballoc, "nomballoc"},
 947         {Opt_stripe, "stripe=%u"},
 948         {Opt_resize, "resize"},
 949         {Opt_err, NULL},
 950 };
 951
 952 static ext4_fsblk_t get_sb_block(void **data)
 953 {
 954         ext4_fsblk_t    sb_block;
 955         char            *options = (char *) *data;
 956
 957         if (!options || strncmp(options, "sb=", 3) != 0)
 958                 return 1;       /* Default location */
 959         options += 3;
 960         /*todo: use simple_strtoll with >32bit ext4 */
 961         sb_block = simple_strtoul(options, &options, 0);
 962         if (*options && *options != ',') {
 963                 printk("EXT4-fs: Invalid sb specification: %s\n",
 964                        (char *) *data);
 965                 return 1;
 966         }
 967         if (*options == ',')
 968                 options++;
 969         *data = (void *) options;
 970         return sb_block;
 971 }
 972
 973 static int parse_options (char *options, struct super_block *sb,
 974                           unsigned int *inum, unsigned long *journal_devnum,
 975                           ext4_fsblk_t *n_blocks_count, int is_remount)
 976 {
 977         struct ext4_sb_info *sbi = EXT4_SB(sb);
 978         char * p;
 979         substring_t args[MAX_OPT_ARGS];
 980         int data_opt = 0;
 981         int option;
 982 #ifdef CONFIG_QUOTA
 983         int qtype;
 984         char *qname;
 985 #endif
 986
 987         if (!options)
 988                 return 1;
 989
 990         while ((p = strsep (&options, ",")) != NULL) {
 991                 int token;
 992                 if (!*p)
 993                         continue;
 994
 995                 token = match_token(p, tokens, args);
 996                 switch (token) {
 997                 case Opt_bsd_df:
 998                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 999                         break;
1000                 case Opt_minix_df:
1001                         set_opt (sbi->s_mount_opt, MINIX_DF);
1002                         break;
1003                 case Opt_grpid:
1004                         set_opt (sbi->s_mount_opt, GRPID);
1005                         break;
1006                 case Opt_nogrpid:
1007                         clear_opt (sbi->s_mount_opt, GRPID);
1008                         break;
1009                 case Opt_resuid:
1010                         if (match_int(&args[0], &option))
1011                                 return 0;
1012                         sbi->s_resuid = option;
1013                         break;
1014                 case Opt_resgid:
1015                         if (match_int(&args[0], &option))
1016                                 return 0;
1017                         sbi->s_resgid = option;
1018                         break;
1019                 case Opt_sb:
1020                         /* handled by get_sb_block() instead of here */
1021                         /* *sb_block = match_int(&args[0]); */
1022                         break;
1023                 case Opt_err_panic:
1024                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1025                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1026                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1027                         break;
1028                 case Opt_err_ro:
1029                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1030                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1031                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1032                         break;
1033                 case Opt_err_cont:
1034                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1035                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1036                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1037                         break;
1038                 case Opt_nouid32:
1039                         set_opt (sbi->s_mount_opt, NO_UID32);
1040                         break;
1041                 case Opt_nocheck:
1042                         clear_opt (sbi->s_mount_opt, CHECK);
1043                         break;
1044                 case Opt_debug:
1045                         set_opt (sbi->s_mount_opt, DEBUG);
1046                         break;
1047                 case Opt_oldalloc:
1048                         set_opt (sbi->s_mount_opt, OLDALLOC);
1049                         break;
1050                 case Opt_orlov:
1051                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1052                         break;
1053 #ifdef CONFIG_EXT4DEV_FS_XATTR
1054                 case Opt_user_xattr:
1055                         set_opt (sbi->s_mount_opt, XATTR_USER);
1056                         break;
1057                 case Opt_nouser_xattr:
1058                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1059                         break;
1060 #else
1061                 case Opt_user_xattr:
1062                 case Opt_nouser_xattr:
1063                         printk("EXT4 (no)user_xattr options not supported\n");
1064                         break;
1065 #endif
1066 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1067                 case Opt_acl:
1068                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1069                         break;
1070                 case Opt_noacl:
1071                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1072                         break;
1073 #else
1074                 case Opt_acl:
1075                 case Opt_noacl:
1076                         printk("EXT4 (no)acl options not supported\n");
1077                         break;
1078 #endif
1079                 case Opt_reservation:
1080                         set_opt(sbi->s_mount_opt, RESERVATION);
1081                         break;
1082                 case Opt_noreservation:
1083                         clear_opt(sbi->s_mount_opt, RESERVATION);
1084                         break;
1085                 case Opt_journal_update:
1086                         /* @@@ FIXME */
1087                         /* Eventually we will want to be able to create
1088                            a journal file here.  For now, only allow the
1089                            user to specify an existing inode to be the
1090                            journal file. */
1091                         if (is_remount) {
1092                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1093                                        "journal on remount\n");
1094                                 return 0;
1095                         }
1096                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1097                         break;
1098                 case Opt_journal_inum:
1099                         if (is_remount) {
1100                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1101                                        "journal on remount\n");
1102                                 return 0;
1103                         }
1104                         if (match_int(&args[0], &option))
1105                                 return 0;
1106                         *inum = option;
1107                         break;
1108                 case Opt_journal_dev:
1109                         if (is_remount) {
1110                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1111                                        "journal on remount\n");
1112                                 return 0;
1113                         }
1114                         if (match_int(&args[0], &option))
1115                                 return 0;
1116                         *journal_devnum = option;
1117                         break;
1118                 case Opt_journal_checksum:
1119                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1120                         break;
1121                 case Opt_journal_async_commit:
1122                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1123                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1124                         break;
1125                 case Opt_noload:
1126                         set_opt (sbi->s_mount_opt, NOLOAD);
1127                         break;
1128                 case Opt_commit:
1129                         if (match_int(&args[0], &option))
1130                                 return 0;
1131                         if (option < 0)
1132                                 return 0;
1133                         if (option == 0)
1134                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1135                         sbi->s_commit_interval = HZ * option;
1136                         break;
1137                 case Opt_data_journal:
1138                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1139                         goto datacheck;
1140                 case Opt_data_ordered:
1141                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1142                         goto datacheck;
1143                 case Opt_data_writeback:
1144                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1145                 datacheck:
1146                         if (is_remount) {
1147                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1148                                                 != data_opt) {
1149                                         printk(KERN_ERR
1150                                                 "EXT4-fs: cannot change data "
1151                                                 "mode on remount\n");
1152                                         return 0;
1153                                 }
1154                         } else {
1155                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1156                                 sbi->s_mount_opt |= data_opt;
1157                         }
1158                         break;
1159 #ifdef CONFIG_QUOTA
1160                 case Opt_usrjquota:
1161                         qtype = USRQUOTA;
1162                         goto set_qf_name;
1163                 case Opt_grpjquota:
1164                         qtype = GRPQUOTA;
1165 set_qf_name:
1166                         if (sb_any_quota_enabled(sb)) {
1167                                 printk(KERN_ERR
1168                                         "EXT4-fs: Cannot change journalled "
1169                                         "quota options when quota turned on.\n");
1170                                 return 0;
1171                         }
1172                         qname = match_strdup(&args[0]);
1173                         if (!qname) {
1174                                 printk(KERN_ERR
1175                                         "EXT4-fs: not enough memory for "
1176                                         "storing quotafile name.\n");
1177                                 return 0;
1178                         }
1179                         if (sbi->s_qf_names[qtype] &&
1180                             strcmp(sbi->s_qf_names[qtype], qname)) {
1181                                 printk(KERN_ERR
1182                                         "EXT4-fs: %s quota file already "
1183                                         "specified.\n", QTYPE2NAME(qtype));
1184                                 kfree(qname);
1185                                 return 0;
1186                         }
1187                         sbi->s_qf_names[qtype] = qname;
1188                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1189                                 printk(KERN_ERR
1190                                         "EXT4-fs: quotafile must be on "
1191                                         "filesystem root.\n");
1192                                 kfree(sbi->s_qf_names[qtype]);
1193                                 sbi->s_qf_names[qtype] = NULL;
1194                                 return 0;
1195                         }
1196                         set_opt(sbi->s_mount_opt, QUOTA);
1197                         break;
1198                 case Opt_offusrjquota:
1199                         qtype = USRQUOTA;
1200                         goto clear_qf_name;
1201                 case Opt_offgrpjquota:
1202                         qtype = GRPQUOTA;
1203 clear_qf_name:
1204                         if (sb_any_quota_enabled(sb)) {
1205                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1206                                         "journalled quota options when "
1207                                         "quota turned on.\n");
1208                                 return 0;
1209                         }
1210                         /*
1211                          * The space will be released later when all options
1212                          * are confirmed to be correct
1213                          */
1214                         sbi->s_qf_names[qtype] = NULL;
1215                         break;
1216                 case Opt_jqfmt_vfsold:
1217                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1218                         break;
1219                 case Opt_jqfmt_vfsv0:
1220                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1221                         break;
1222                 case Opt_quota:
1223                 case Opt_usrquota:
1224                         set_opt(sbi->s_mount_opt, QUOTA);
1225                         set_opt(sbi->s_mount_opt, USRQUOTA);
1226                         break;
1227                 case Opt_grpquota:
1228                         set_opt(sbi->s_mount_opt, QUOTA);
1229                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1230                         break;
1231                 case Opt_noquota:
1232                         if (sb_any_quota_enabled(sb)) {
1233                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1234                                         "options when quota turned on.\n");
1235                                 return 0;
1236                         }
1237                         clear_opt(sbi->s_mount_opt, QUOTA);
1238                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1239                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1240                         break;
1241 #else
1242                 case Opt_quota:
1243                 case Opt_usrquota:
1244                 case Opt_grpquota:
1245                 case Opt_usrjquota:
1246                 case Opt_grpjquota:
1247                 case Opt_offusrjquota:
1248                 case Opt_offgrpjquota:
1249                 case Opt_jqfmt_vfsold:
1250                 case Opt_jqfmt_vfsv0:
1251                         printk(KERN_ERR
1252                                 "EXT4-fs: journalled quota options not "
1253                                 "supported.\n");
1254                         break;
1255                 case Opt_noquota:
1256                         break;
1257 #endif
1258                 case Opt_abort:
1259                         set_opt(sbi->s_mount_opt, ABORT);
1260                         break;
1261                 case Opt_barrier:
1262                         if (match_int(&args[0], &option))
1263                                 return 0;
1264                         if (option)
1265                                 set_opt(sbi->s_mount_opt, BARRIER);
1266                         else
1267                                 clear_opt(sbi->s_mount_opt, BARRIER);
1268                         break;
1269                 case Opt_ignore:
1270                         break;
1271                 case Opt_resize:
1272                         if (!is_remount) {
1273                                 printk("EXT4-fs: resize option only available "
1274                                         "for remount\n");
1275                                 return 0;
1276                         }
1277                         if (match_int(&args[0], &option) != 0)
1278                                 return 0;
1279                         *n_blocks_count = option;
1280                         break;
1281                 case Opt_nobh:
1282                         set_opt(sbi->s_mount_opt, NOBH);
1283                         break;
1284                 case Opt_bh:
1285                         clear_opt(sbi->s_mount_opt, NOBH);
1286                         break;
1287                 case Opt_extents:
1288                         set_opt (sbi->s_mount_opt, EXTENTS);
1289                         break;
1290                 case Opt_noextents:
1291                         clear_opt (sbi->s_mount_opt, EXTENTS);
1292                         break;
1293                 case Opt_i_version:
1294                         set_opt(sbi->s_mount_opt, I_VERSION);
1295                         sb->s_flags |= MS_I_VERSION;
1296                         break;
1297                 case Opt_mballoc:
1298                         set_opt(sbi->s_mount_opt, MBALLOC);
1299                         break;
1300                 case Opt_nomballoc:
1301                         clear_opt(sbi->s_mount_opt, MBALLOC);
1302                         break;
1303                 case Opt_stripe:
1304                         if (match_int(&args[0], &option))
1305                                 return 0;
1306                         if (option < 0)
1307                                 return 0;
1308                         sbi->s_stripe = option;
1309                         break;
1310                 default:
1311                         printk (KERN_ERR
1312                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1313                                 "or missing value\n", p);
1314                         return 0;
1315                 }
1316         }
1317 #ifdef CONFIG_QUOTA
1318         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1319                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1320                      sbi->s_qf_names[USRQUOTA])
1321                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1322
1323                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1324                      sbi->s_qf_names[GRPQUOTA])
1325                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1326
1327                 if ((sbi->s_qf_names[USRQUOTA] &&
1328                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1329                     (sbi->s_qf_names[GRPQUOTA] &&
1330                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1331                         printk(KERN_ERR "EXT4-fs: old and new quota "
1332                                         "format mixing.\n");
1333                         return 0;
1334                 }
1335
1336                 if (!sbi->s_jquota_fmt) {
1337                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1338                                         "not specified.\n");
1339                         return 0;
1340                 }
1341         } else {
1342                 if (sbi->s_jquota_fmt) {
1343                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1344                                         "specified with no journalling "
1345                                         "enabled.\n");
1346                         return 0;
1347                 }
1348         }
1349 #endif
1350         return 1;
1351 }
1352
1353 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1354                             int read_only)
1355 {
1356         struct ext4_sb_info *sbi = EXT4_SB(sb);
1357         int res = 0;
1358
1359         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1360                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1361                         "forcing read-only mode\n");
1362                 res = MS_RDONLY;
1363         }
1364         if (read_only)
1365                 return res;
1366         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1367                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1368                         "running e2fsck is recommended\n");
1369         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1370                 printk (KERN_WARNING
1371                         "EXT4-fs warning: mounting fs with errors, "
1372                         "running e2fsck is recommended\n");
1373         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1374                  le16_to_cpu(es->s_mnt_count) >=
1375                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1376                 printk (KERN_WARNING
1377                         "EXT4-fs warning: maximal mount count reached, "
1378                         "running e2fsck is recommended\n");
1379         else if (le32_to_cpu(es->s_checkinterval) &&
1380                 (le32_to_cpu(es->s_lastcheck) +
1381                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1382                 printk (KERN_WARNING
1383                         "EXT4-fs warning: checktime reached, "
1384                         "running e2fsck is recommended\n");
1385 #if 0
1386                 /* @@@ We _will_ want to clear the valid bit if we find
1387                  * inconsistencies, to force a fsck at reboot.  But for
1388                  * a plain journaled filesystem we can keep it set as
1389                  * valid forever! :)
1390                  */
1391         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1392 #endif
1393         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1394                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1395         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1396         es->s_mtime = cpu_to_le32(get_seconds());
1397         ext4_update_dynamic_rev(sb);
1398         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1399
1400         ext4_commit_super(sb, es, 1);
1401         if (test_opt(sb, DEBUG))
1402                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1403                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1404                         sb->s_blocksize,
1405                         sbi->s_groups_count,
1406                         EXT4_BLOCKS_PER_GROUP(sb),
1407                         EXT4_INODES_PER_GROUP(sb),
1408                         sbi->s_mount_opt);
1409
1410         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1411         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1412                 char b[BDEVNAME_SIZE];
1413
1414                 printk("external journal on %s\n",
1415                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1416         } else {
1417                 printk("internal journal\n");
1418         }
1419         return res;
1420 }
1421
1422 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1423                             struct ext4_group_desc *gdp)
1424 {
1425         __u16 crc = 0;
1426
1427         if (sbi->s_es->s_feature_ro_compat &
1428             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1429                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1430                 __le32 le_group = cpu_to_le32(block_group);
1431
1432                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1433                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1434                 crc = crc16(crc, (__u8 *)gdp, offset);
1435                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1436                 /* for checksum of struct ext4_group_desc do the rest...*/
1437                 if ((sbi->s_es->s_feature_incompat &
1438                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1439                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1440                         crc = crc16(crc, (__u8 *)gdp + offset,
1441                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1442                                         offset);
1443         }
1444
1445         return cpu_to_le16(crc);
1446 }
1447
1448 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1449                                 struct ext4_group_desc *gdp)
1450 {
1451         if ((sbi->s_es->s_feature_ro_compat &
1452              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1453             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1454                 return 0;
1455
1456         return 1;
1457 }
1458
1459 /* Called at mount-time, super-block is locked */
1460 static int ext4_check_descriptors(struct super_block *sb)
1461 {
1462         struct ext4_sb_info *sbi = EXT4_SB(sb);
1463         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1464         ext4_fsblk_t last_block;
1465         ext4_fsblk_t block_bitmap;
1466         ext4_fsblk_t inode_bitmap;
1467         ext4_fsblk_t inode_table;
1468         int flexbg_flag = 0;
1469         ext4_group_t i;
1470
1471         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1472                 flexbg_flag = 1;
1473
1474         ext4_debug ("Checking group descriptors");
1475
1476         for (i = 0; i < sbi->s_groups_count; i++) {
1477                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1478
1479                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1480                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1481                 else
1482                         last_block = first_block +
1483                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1484
1485                 block_bitmap = ext4_block_bitmap(sb, gdp);
1486                 if (block_bitmap < first_block || block_bitmap > last_block)
1487                 {
1488                         ext4_error (sb, "ext4_check_descriptors",
1489                                     "Block bitmap for group %lu"
1490                                     " not in group (block %llu)!",
1491                                     i, block_bitmap);
1492                         return 0;
1493                 }
1494                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1495                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1496                 {
1497                         ext4_error (sb, "ext4_check_descriptors",
1498                                     "Inode bitmap for group %lu"
1499                                     " not in group (block %llu)!",
1500                                     i, inode_bitmap);
1501                         return 0;
1502                 }
1503                 inode_table = ext4_inode_table(sb, gdp);
1504                 if (inode_table < first_block ||
1505                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1506                 {
1507                         ext4_error (sb, "ext4_check_descriptors",
1508                                     "Inode table for group %lu"
1509                                     " not in group (block %llu)!",
1510                                     i, inode_table);
1511                         return 0;
1512                 }
1513                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1514                         ext4_error(sb, __FUNCTION__,
1515                                    "Checksum for group %lu failed (%u!=%u)\n",
1516                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1517                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1518                         return 0;
1519                 }
1520                 if (!flexbg_flag)
1521                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1522         }
1523
1524         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1525         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1526         return 1;
1527 }
1528
1529 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1530  * the superblock) which were deleted from all directories, but held open by
1531  * a process at the time of a crash.  We walk the list and try to delete these
1532  * inodes at recovery time (only with a read-write filesystem).
1533  *
1534  * In order to keep the orphan inode chain consistent during traversal (in
1535  * case of crash during recovery), we link each inode into the superblock
1536  * orphan list_head and handle it the same way as an inode deletion during
1537  * normal operation (which journals the operations for us).
1538  *
1539  * We only do an iget() and an iput() on each inode, which is very safe if we
1540  * accidentally point at an in-use or already deleted inode.  The worst that
1541  * can happen in this case is that we get a "bit already cleared" message from
1542  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1543  * e2fsck was run on this filesystem, and it must have already done the orphan
1544  * inode cleanup for us, so we can safely abort without any further action.
1545  */
1546 static void ext4_orphan_cleanup (struct super_block * sb,
1547                                  struct ext4_super_block * es)
1548 {
1549         unsigned int s_flags = sb->s_flags;
1550         int nr_orphans = 0, nr_truncates = 0;
1551 #ifdef CONFIG_QUOTA
1552         int i;
1553 #endif
1554         if (!es->s_last_orphan) {
1555                 jbd_debug(4, "no orphan inodes to clean up\n");
1556                 return;
1557         }
1558
1559         if (bdev_read_only(sb->s_bdev)) {
1560                 printk(KERN_ERR "EXT4-fs: write access "
1561                         "unavailable, skipping orphan cleanup.\n");
1562                 return;
1563         }
1564
1565         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1566                 if (es->s_last_orphan)
1567                         jbd_debug(1, "Errors on filesystem, "
1568                                   "clearing orphan list.\n");
1569                 es->s_last_orphan = 0;
1570                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1571                 return;
1572         }
1573
1574         if (s_flags & MS_RDONLY) {
1575                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1576                        sb->s_id);
1577                 sb->s_flags &= ~MS_RDONLY;
1578         }
1579 #ifdef CONFIG_QUOTA
1580         /* Needed for iput() to work correctly and not trash data */
1581         sb->s_flags |= MS_ACTIVE;
1582         /* Turn on quotas so that they are updated correctly */
1583         for (i = 0; i < MAXQUOTAS; i++) {
1584                 if (EXT4_SB(sb)->s_qf_names[i]) {
1585                         int ret = ext4_quota_on_mount(sb, i);
1586                         if (ret < 0)
1587                                 printk(KERN_ERR
1588                                         "EXT4-fs: Cannot turn on journalled "
1589                                         "quota: error %d\n", ret);
1590                 }
1591         }
1592 #endif
1593
1594         while (es->s_last_orphan) {
1595                 struct inode *inode;
1596
1597                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1598                 if (IS_ERR(inode)) {
1599                         es->s_last_orphan = 0;
1600                         break;
1601                 }
1602
1603                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1604                 DQUOT_INIT(inode);
1605                 if (inode->i_nlink) {
1606                         printk(KERN_DEBUG
1607                                 "%s: truncating inode %lu to %Ld bytes\n",
1608                                 __FUNCTION__, inode->i_ino, inode->i_size);
1609                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1610                                   inode->i_ino, inode->i_size);
1611                         ext4_truncate(inode);
1612                         nr_truncates++;
1613                 } else {
1614                         printk(KERN_DEBUG
1615                                 "%s: deleting unreferenced inode %lu\n",
1616                                 __FUNCTION__, inode->i_ino);
1617                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1618                                   inode->i_ino);
1619                         nr_orphans++;
1620                 }
1621                 iput(inode);  /* The delete magic happens here! */
1622         }
1623
1624 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1625
1626         if (nr_orphans)
1627                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1628                        sb->s_id, PLURAL(nr_orphans));
1629         if (nr_truncates)
1630                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1631                        sb->s_id, PLURAL(nr_truncates));
1632 #ifdef CONFIG_QUOTA
1633         /* Turn quotas off */
1634         for (i = 0; i < MAXQUOTAS; i++) {
1635                 if (sb_dqopt(sb)->files[i])
1636                         vfs_quota_off(sb, i, 0);
1637         }
1638 #endif
1639         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1640 }
1641 /*
1642  * Maximal extent format file size.
1643  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1644  * extent format containers, within a sector_t, and within i_blocks
1645  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1646  * so that won't be a limiting factor.
1647  *
1648  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1649  */
1650 static loff_t ext4_max_size(int blkbits)
1651 {
1652         loff_t res;
1653         loff_t upper_limit = MAX_LFS_FILESIZE;
1654
1655         /* small i_blocks in vfs inode? */
1656         if (sizeof(blkcnt_t) < sizeof(u64)) {
1657                 /*
1658                  * CONFIG_LSF is not enabled implies the inode
1659                  * i_block represent total blocks in 512 bytes
1660                  * 32 == size of vfs inode i_blocks * 8
1661                  */
1662                 upper_limit = (1LL << 32) - 1;
1663
1664                 /* total blocks in file system block size */
1665                 upper_limit >>= (blkbits - 9);
1666                 upper_limit <<= blkbits;
1667         }
1668
1669         /* 32-bit extent-start container, ee_block */
1670         res = 1LL << 32;
1671         res <<= blkbits;
1672         res -= 1;
1673
1674         /* Sanity check against vm- & vfs- imposed limits */
1675         if (res > upper_limit)
1676                 res = upper_limit;
1677
1678         return res;
1679 }
1680
1681 /*
1682  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1683  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1684  * We need to be 1 filesystem block less than the 2^48 sector limit.
1685  */
1686 static loff_t ext4_max_bitmap_size(int bits)
1687 {
1688         loff_t res = EXT4_NDIR_BLOCKS;
1689         int meta_blocks;
1690         loff_t upper_limit;
1691         /* This is calculated to be the largest file size for a
1692          * dense, bitmapped file such that the total number of
1693          * sectors in the file, including data and all indirect blocks,
1694          * does not exceed 2^48 -1
1695          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1696          * total number of  512 bytes blocks of the file
1697          */
1698
1699         if (sizeof(blkcnt_t) < sizeof(u64)) {
1700                 /*
1701                  * CONFIG_LSF is not enabled implies the inode
1702                  * i_block represent total blocks in 512 bytes
1703                  * 32 == size of vfs inode i_blocks * 8
1704                  */
1705                 upper_limit = (1LL << 32) - 1;
1706
1707                 /* total blocks in file system block size */
1708                 upper_limit >>= (bits - 9);
1709
1710         } else {
1711                 /*
1712                  * We use 48 bit ext4_inode i_blocks
1713                  * With EXT4_HUGE_FILE_FL set the i_blocks
1714                  * represent total number of blocks in
1715                  * file system block size
1716                  */
1717                 upper_limit = (1LL << 48) - 1;
1718
1719         }
1720
1721         /* indirect blocks */
1722         meta_blocks = 1;
1723         /* double indirect blocks */
1724         meta_blocks += 1 + (1LL << (bits-2));
1725         /* tripple indirect blocks */
1726         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1727
1728         upper_limit -= meta_blocks;
1729         upper_limit <<= bits;
1730
1731         res += 1LL << (bits-2);
1732         res += 1LL << (2*(bits-2));
1733         res += 1LL << (3*(bits-2));
1734         res <<= bits;
1735         if (res > upper_limit)
1736                 res = upper_limit;
1737
1738         if (res > MAX_LFS_FILESIZE)
1739                 res = MAX_LFS_FILESIZE;
1740
1741         return res;
1742 }
1743
1744 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1745                                 ext4_fsblk_t logical_sb_block, int nr)
1746 {
1747         struct ext4_sb_info *sbi = EXT4_SB(sb);
1748         ext4_group_t bg, first_meta_bg;
1749         int has_super = 0;
1750
1751         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1752
1753         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1754             nr < first_meta_bg)
1755                 return logical_sb_block + nr + 1;
1756         bg = sbi->s_desc_per_block * nr;
1757         if (ext4_bg_has_super(sb, bg))
1758                 has_super = 1;
1759         return (has_super + ext4_group_first_block_no(sb, bg));
1760 }
1761
1762 /**
1763  * ext4_get_stripe_size: Get the stripe size.
1764  * @sbi: In memory super block info
1765  *
1766  * If we have specified it via mount option, then
1767  * use the mount option value. If the value specified at mount time is
1768  * greater than the blocks per group use the super block value.
1769  * If the super block value is greater than blocks per group return 0.
1770  * Allocator needs it be less than blocks per group.
1771  *
1772  */
1773 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1774 {
1775         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1776         unsigned long stripe_width =
1777                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1778
1779         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1780                 return sbi->s_stripe;
1781
1782         if (stripe_width <= sbi->s_blocks_per_group)
1783                 return stripe_width;
1784
1785         if (stride <= sbi->s_blocks_per_group)
1786                 return stride;
1787
1788         return 0;
1789 }
1790
1791 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1792                                 __releases(kernel_sem)
1793                                 __acquires(kernel_sem)
1794
1795 {
1796         struct buffer_head * bh;
1797         struct ext4_super_block *es = NULL;
1798         struct ext4_sb_info *sbi;
1799         ext4_fsblk_t block;
1800         ext4_fsblk_t sb_block = get_sb_block(&data);
1801         ext4_fsblk_t logical_sb_block;
1802         unsigned long offset = 0;
1803         unsigned int journal_inum = 0;
1804         unsigned long journal_devnum = 0;
1805         unsigned long def_mount_opts;
1806         struct inode *root;
1807         int ret = -EINVAL;
1808         int blocksize;
1809         int db_count;
1810         int i;
1811         int needs_recovery;
1812         __le32 features;
1813         __u64 blocks_count;
1814         int err;
1815
1816         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1817         if (!sbi)
1818                 return -ENOMEM;
1819         sb->s_fs_info = sbi;
1820         sbi->s_mount_opt = 0;
1821         sbi->s_resuid = EXT4_DEF_RESUID;
1822         sbi->s_resgid = EXT4_DEF_RESGID;
1823         sbi->s_sb_block = sb_block;
1824
1825         unlock_kernel();
1826
1827         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1828         if (!blocksize) {
1829                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1830                 goto out_fail;
1831         }
1832
1833         if (!sb_set_blocksize(sb, blocksize)) {
1834                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1835                 goto out_fail;
1836         }
1837
1838         /*
1839          * The ext4 superblock will not be buffer aligned for other than 1kB
1840          * block sizes.  We need to calculate the offset from buffer start.
1841          */
1842         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1843                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1844                 offset = do_div(logical_sb_block, blocksize);
1845         } else {
1846                 logical_sb_block = sb_block;
1847         }
1848
1849         if (!(bh = sb_bread(sb, logical_sb_block))) {
1850                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1851                 goto out_fail;
1852         }
1853         /*
1854          * Note: s_es must be initialized as soon as possible because
1855          *       some ext4 macro-instructions depend on its value
1856          */
1857         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1858         sbi->s_es = es;
1859         sb->s_magic = le16_to_cpu(es->s_magic);
1860         if (sb->s_magic != EXT4_SUPER_MAGIC)
1861                 goto cantfind_ext4;
1862
1863         /* Set defaults before we parse the mount options */
1864         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1865         if (def_mount_opts & EXT4_DEFM_DEBUG)
1866                 set_opt(sbi->s_mount_opt, DEBUG);
1867         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1868                 set_opt(sbi->s_mount_opt, GRPID);
1869         if (def_mount_opts & EXT4_DEFM_UID16)
1870                 set_opt(sbi->s_mount_opt, NO_UID32);
1871 #ifdef CONFIG_EXT4DEV_FS_XATTR
1872         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1873                 set_opt(sbi->s_mount_opt, XATTR_USER);
1874 #endif
1875 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1876         if (def_mount_opts & EXT4_DEFM_ACL)
1877                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1878 #endif
1879         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1880                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1881         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1882                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1883         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1884                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1885
1886         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1887                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1888         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1889                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1890         else
1891                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1892
1893         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1894         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1895
1896         set_opt(sbi->s_mount_opt, RESERVATION);
1897
1898         /*
1899          * turn on extents feature by default in ext4 filesystem
1900          * User -o noextents to turn it off
1901          */
1902         set_opt(sbi->s_mount_opt, EXTENTS);
1903         /*
1904          * turn on mballoc feature by default in ext4 filesystem
1905          * User -o nomballoc to turn it off
1906          */
1907         set_opt(sbi->s_mount_opt, MBALLOC);
1908
1909         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1910                             NULL, 0))
1911                 goto failed_mount;
1912
1913         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1914                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1915
1916         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1917             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1918              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1919              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1920                 printk(KERN_WARNING
1921                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1922                        "running e2fsck is recommended\n");
1923
1924         /*
1925          * Since ext4 is still considered development code, we require
1926          * that the TEST_FILESYS flag in s->flags be set.
1927          */
1928         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
1929                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
1930                        "OK to use with test code.\n", sb->s_id);
1931                 goto failed_mount;
1932         }
1933
1934         /*
1935          * Check feature flags regardless of the revision level, since we
1936          * previously didn't change the revision level when setting the flags,
1937          * so there is a chance incompat flags are set on a rev 0 filesystem.
1938          */
1939         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1940         if (features) {
1941                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1942                        "unsupported optional features (%x).\n",
1943                        sb->s_id, le32_to_cpu(features));
1944                 goto failed_mount;
1945         }
1946         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1947         if (!(sb->s_flags & MS_RDONLY) && features) {
1948                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1949                        "unsupported optional features (%x).\n",
1950                        sb->s_id, le32_to_cpu(features));
1951                 goto failed_mount;
1952         }
1953         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
1954                 /*
1955                  * Large file size enabled file system can only be
1956                  * mount if kernel is build with CONFIG_LSF
1957                  */
1958                 if (sizeof(root->i_blocks) < sizeof(u64) &&
1959                                 !(sb->s_flags & MS_RDONLY)) {
1960                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
1961                                         "files cannot be mounted read-write "
1962                                         "without CONFIG_LSF.\n", sb->s_id);
1963                         goto failed_mount;
1964                 }
1965         }
1966         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1967
1968         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1969             blocksize > EXT4_MAX_BLOCK_SIZE) {
1970                 printk(KERN_ERR
1971                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1972                        blocksize, sb->s_id);
1973                 goto failed_mount;
1974         }
1975
1976         if (sb->s_blocksize != blocksize) {
1977
1978                 /* Validate the filesystem blocksize */
1979                 if (!sb_set_blocksize(sb, blocksize)) {
1980                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
1981                                         blocksize);
1982                         goto failed_mount;
1983                 }
1984
1985                 brelse (bh);
1986                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1987                 offset = do_div(logical_sb_block, blocksize);
1988                 bh = sb_bread(sb, logical_sb_block);
1989                 if (!bh) {
1990                         printk(KERN_ERR
1991                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1992                         goto failed_mount;
1993                 }
1994                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1995                 sbi->s_es = es;
1996                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1997                         printk (KERN_ERR
1998                                 "EXT4-fs: Magic mismatch, very weird !\n");
1999                         goto failed_mount;
2000                 }
2001         }
2002
2003         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2004         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2005
2006         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2007                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2008                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2009         } else {
2010                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2011                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2012                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2013                     (!is_power_of_2(sbi->s_inode_size)) ||
2014                     (sbi->s_inode_size > blocksize)) {
2015                         printk (KERN_ERR
2016                                 "EXT4-fs: unsupported inode size: %d\n",
2017                                 sbi->s_inode_size);
2018                         goto failed_mount;
2019                 }
2020                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2021                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2022         }
2023         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2024         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2025                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2026                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2027                     !is_power_of_2(sbi->s_desc_size)) {
2028                         printk(KERN_ERR
2029                                "EXT4-fs: unsupported descriptor size %lu\n",
2030                                sbi->s_desc_size);
2031                         goto failed_mount;
2032                 }
2033         } else
2034                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2035         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2036         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2037         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2038                 goto cantfind_ext4;
2039         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2040         if (sbi->s_inodes_per_block == 0)
2041                 goto cantfind_ext4;
2042         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2043                                         sbi->s_inodes_per_block;
2044         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2045         sbi->s_sbh = bh;
2046         sbi->s_mount_state = le16_to_cpu(es->s_state);
2047         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2048         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2049         for (i=0; i < 4; i++)
2050                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2051         sbi->s_def_hash_version = es->s_def_hash_version;
2052
2053         if (sbi->s_blocks_per_group > blocksize * 8) {
2054                 printk (KERN_ERR
2055                         "EXT4-fs: #blocks per group too big: %lu\n",
2056                         sbi->s_blocks_per_group);
2057                 goto failed_mount;
2058         }
2059         if (sbi->s_inodes_per_group > blocksize * 8) {
2060                 printk (KERN_ERR
2061                         "EXT4-fs: #inodes per group too big: %lu\n",
2062                         sbi->s_inodes_per_group);
2063                 goto failed_mount;
2064         }
2065
2066         if (ext4_blocks_count(es) >
2067                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2068                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2069                         " too large to mount safely\n", sb->s_id);
2070                 if (sizeof(sector_t) < 8)
2071                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2072                                         "enabled\n");
2073                 goto failed_mount;
2074         }
2075
2076         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2077                 goto cantfind_ext4;
2078
2079         /* ensure blocks_count calculation below doesn't sign-extend */
2080         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2081             le32_to_cpu(es->s_first_data_block) + 1) {
2082                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2083                        "first data block %u, blocks per group %lu\n",
2084                         ext4_blocks_count(es),
2085                         le32_to_cpu(es->s_first_data_block),
2086                         EXT4_BLOCKS_PER_GROUP(sb));
2087                 goto failed_mount;
2088         }
2089         blocks_count = (ext4_blocks_count(es) -
2090                         le32_to_cpu(es->s_first_data_block) +
2091                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2092         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2093         sbi->s_groups_count = blocks_count;
2094         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2095                    EXT4_DESC_PER_BLOCK(sb);
2096         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2097                                     GFP_KERNEL);
2098         if (sbi->s_group_desc == NULL) {
2099                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2100                 goto failed_mount;
2101         }
2102
2103         bgl_lock_init(&sbi->s_blockgroup_lock);
2104
2105         for (i = 0; i < db_count; i++) {
2106                 block = descriptor_loc(sb, logical_sb_block, i);
2107                 sbi->s_group_desc[i] = sb_bread(sb, block);
2108                 if (!sbi->s_group_desc[i]) {
2109                         printk (KERN_ERR "EXT4-fs: "
2110                                 "can't read group descriptor %d\n", i);
2111                         db_count = i;
2112                         goto failed_mount2;
2113                 }
2114         }
2115         if (!ext4_check_descriptors (sb)) {
2116                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2117                 goto failed_mount2;
2118         }
2119         sbi->s_gdb_count = db_count;
2120         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2121         spin_lock_init(&sbi->s_next_gen_lock);
2122
2123         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2124                         ext4_count_free_blocks(sb));
2125         if (!err) {
2126                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2127                                 ext4_count_free_inodes(sb));
2128         }
2129         if (!err) {
2130                 err = percpu_counter_init(&sbi->s_dirs_counter,
2131                                 ext4_count_dirs(sb));
2132         }
2133         if (err) {
2134                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2135                 goto failed_mount3;
2136         }
2137
2138         /* per fileystem reservation list head & lock */
2139         spin_lock_init(&sbi->s_rsv_window_lock);
2140         sbi->s_rsv_window_root = RB_ROOT;
2141         /* Add a single, static dummy reservation to the start of the
2142          * reservation window list --- it gives us a placeholder for
2143          * append-at-start-of-list which makes the allocation logic
2144          * _much_ simpler. */
2145         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2146         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2147         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2148         sbi->s_rsv_window_head.rsv_goal_size = 0;
2149         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2150
2151         sbi->s_stripe = ext4_get_stripe_size(sbi);
2152
2153         /*
2154          * set up enough so that it can read an inode
2155          */
2156         sb->s_op = &ext4_sops;
2157         sb->s_export_op = &ext4_export_ops;
2158         sb->s_xattr = ext4_xattr_handlers;
2159 #ifdef CONFIG_QUOTA
2160         sb->s_qcop = &ext4_qctl_operations;
2161         sb->dq_op = &ext4_quota_operations;
2162 #endif
2163         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2164
2165         sb->s_root = NULL;
2166
2167         needs_recovery = (es->s_last_orphan != 0 ||
2168                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2169                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2170
2171         /*
2172          * The first inode we look at is the journal inode.  Don't try
2173          * root first: it may be modified in the journal!
2174          */
2175         if (!test_opt(sb, NOLOAD) &&
2176             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2177                 if (ext4_load_journal(sb, es, journal_devnum))
2178                         goto failed_mount3;
2179         } else if (journal_inum) {
2180                 if (ext4_create_journal(sb, es, journal_inum))
2181                         goto failed_mount3;
2182         } else {
2183                 if (!silent)
2184                         printk (KERN_ERR
2185                                 "ext4: No journal on filesystem on %s\n",
2186                                 sb->s_id);
2187                 goto failed_mount3;
2188         }
2189
2190         if (ext4_blocks_count(es) > 0xffffffffULL &&
2191             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2192                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2193                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2194                 goto failed_mount4;
2195         }
2196
2197         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2198                 jbd2_journal_set_features(sbi->s_journal,
2199                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2200                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2201         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2202                 jbd2_journal_set_features(sbi->s_journal,
2203                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2204                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2205                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2206         } else {
2207                 jbd2_journal_clear_features(sbi->s_journal,
2208                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2209                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2210         }
2211
2212         /* We have now updated the journal if required, so we can
2213          * validate the data journaling mode. */
2214         switch (test_opt(sb, DATA_FLAGS)) {
2215         case 0:
2216                 /* No mode set, assume a default based on the journal
2217                  * capabilities: ORDERED_DATA if the journal can
2218                  * cope, else JOURNAL_DATA
2219                  */
2220                 if (jbd2_journal_check_available_features
2221                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2222                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2223                 else
2224                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2225                 break;
2226
2227         case EXT4_MOUNT_ORDERED_DATA:
2228         case EXT4_MOUNT_WRITEBACK_DATA:
2229                 if (!jbd2_journal_check_available_features
2230                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2231                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2232                                "requested data journaling mode\n");
2233                         goto failed_mount4;
2234                 }
2235         default:
2236                 break;
2237         }
2238
2239         if (test_opt(sb, NOBH)) {
2240                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2241                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2242                                 "its supported only with writeback mode\n");
2243                         clear_opt(sbi->s_mount_opt, NOBH);
2244                 }
2245         }
2246         /*
2247          * The jbd2_journal_load will have done any necessary log recovery,
2248          * so we can safely mount the rest of the filesystem now.
2249          */
2250
2251         root = ext4_iget(sb, EXT4_ROOT_INO);
2252         if (IS_ERR(root)) {
2253                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2254                 ret = PTR_ERR(root);
2255                 goto failed_mount4;
2256         }
2257         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2258                 iput(root);
2259                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2260                 goto failed_mount4;
2261         }
2262         sb->s_root = d_alloc_root(root);
2263         if (!sb->s_root) {
2264                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2265                 iput(root);
2266                 ret = -ENOMEM;
2267                 goto failed_mount4;
2268         }
2269
2270         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2271
2272         /* determine the minimum size of new large inodes, if present */
2273         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2274                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2275                                                      EXT4_GOOD_OLD_INODE_SIZE;
2276                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2277                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2278                         if (sbi->s_want_extra_isize <
2279                             le16_to_cpu(es->s_want_extra_isize))
2280                                 sbi->s_want_extra_isize =
2281                                         le16_to_cpu(es->s_want_extra_isize);
2282                         if (sbi->s_want_extra_isize <
2283                             le16_to_cpu(es->s_min_extra_isize))
2284                                 sbi->s_want_extra_isize =
2285                                         le16_to_cpu(es->s_min_extra_isize);
2286                 }
2287         }
2288         /* Check if enough inode space is available */
2289         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2290                                                         sbi->s_inode_size) {
2291                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2292                                                        EXT4_GOOD_OLD_INODE_SIZE;
2293                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2294                         "available.\n");
2295         }
2296
2297         /*
2298          * akpm: core read_super() calls in here with the superblock locked.
2299          * That deadlocks, because orphan cleanup needs to lock the superblock
2300          * in numerous places.  Here we just pop the lock - it's relatively
2301          * harmless, because we are now ready to accept write_super() requests,
2302          * and aviro says that's the only reason for hanging onto the
2303          * superblock lock.
2304          */
2305         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2306         ext4_orphan_cleanup(sb, es);
2307         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2308         if (needs_recovery)
2309                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2310         ext4_mark_recovery_complete(sb, es);
2311         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2312                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2313                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2314                 "writeback");
2315
2316         ext4_ext_init(sb);
2317         ext4_mb_init(sb, needs_recovery);
2318
2319         lock_kernel();
2320         return 0;
2321
2322 cantfind_ext4:
2323         if (!silent)
2324                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2325                        sb->s_id);
2326         goto failed_mount;
2327
2328 failed_mount4:
2329         jbd2_journal_destroy(sbi->s_journal);
2330 failed_mount3:
2331         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2332         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2333         percpu_counter_destroy(&sbi->s_dirs_counter);
2334 failed_mount2:
2335         for (i = 0; i < db_count; i++)
2336                 brelse(sbi->s_group_desc[i]);
2337         kfree(sbi->s_group_desc);
2338 failed_mount:
2339 #ifdef CONFIG_QUOTA
2340         for (i = 0; i < MAXQUOTAS; i++)
2341                 kfree(sbi->s_qf_names[i]);
2342 #endif
2343         ext4_blkdev_remove(sbi);
2344         brelse(bh);
2345 out_fail:
2346         sb->s_fs_info = NULL;
2347         kfree(sbi);
2348         lock_kernel();
2349         return ret;
2350 }
2351
2352 /*
2353  * Setup any per-fs journal parameters now.  We'll do this both on
2354  * initial mount, once the journal has been initialised but before we've
2355  * done any recovery; and again on any subsequent remount.
2356  */
2357 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2358 {
2359         struct ext4_sb_info *sbi = EXT4_SB(sb);
2360
2361         if (sbi->s_commit_interval)
2362                 journal->j_commit_interval = sbi->s_commit_interval;
2363         /* We could also set up an ext4-specific default for the commit
2364          * interval here, but for now we'll just fall back to the jbd
2365          * default. */
2366
2367         spin_lock(&journal->j_state_lock);
2368         if (test_opt(sb, BARRIER))
2369                 journal->j_flags |= JBD2_BARRIER;
2370         else
2371                 journal->j_flags &= ~JBD2_BARRIER;
2372         spin_unlock(&journal->j_state_lock);
2373 }
2374
2375 static journal_t *ext4_get_journal(struct super_block *sb,
2376                                    unsigned int journal_inum)
2377 {
2378         struct inode *journal_inode;
2379         journal_t *journal;
2380
2381         /* First, test for the existence of a valid inode on disk.  Bad
2382          * things happen if we iget() an unused inode, as the subsequent
2383          * iput() will try to delete it. */
2384
2385         journal_inode = ext4_iget(sb, journal_inum);
2386         if (IS_ERR(journal_inode)) {
2387                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2388                 return NULL;
2389         }
2390         if (!journal_inode->i_nlink) {
2391                 make_bad_inode(journal_inode);
2392                 iput(journal_inode);
2393                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2394                 return NULL;
2395         }
2396
2397         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2398                   journal_inode, journal_inode->i_size);
2399         if (!S_ISREG(journal_inode->i_mode)) {
2400                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2401                 iput(journal_inode);
2402                 return NULL;
2403         }
2404
2405         journal = jbd2_journal_init_inode(journal_inode);
2406         if (!journal) {
2407                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2408                 iput(journal_inode);
2409                 return NULL;
2410         }
2411         journal->j_private = sb;
2412         ext4_init_journal_params(sb, journal);
2413         return journal;
2414 }
2415
2416 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2417                                        dev_t j_dev)
2418 {
2419         struct buffer_head * bh;
2420         journal_t *journal;
2421         ext4_fsblk_t start;
2422         ext4_fsblk_t len;
2423         int hblock, blocksize;
2424         ext4_fsblk_t sb_block;
2425         unsigned long offset;
2426         struct ext4_super_block * es;
2427         struct block_device *bdev;
2428
2429         bdev = ext4_blkdev_get(j_dev);
2430         if (bdev == NULL)
2431                 return NULL;
2432
2433         if (bd_claim(bdev, sb)) {
2434                 printk(KERN_ERR
2435                         "EXT4: failed to claim external journal device.\n");
2436                 blkdev_put(bdev);
2437                 return NULL;
2438         }
2439
2440         blocksize = sb->s_blocksize;
2441         hblock = bdev_hardsect_size(bdev);
2442         if (blocksize < hblock) {
2443                 printk(KERN_ERR
2444                         "EXT4-fs: blocksize too small for journal device.\n");
2445                 goto out_bdev;
2446         }
2447
2448         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2449         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2450         set_blocksize(bdev, blocksize);
2451         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2452                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2453                        "external journal\n");
2454                 goto out_bdev;
2455         }
2456
2457         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2458         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2459             !(le32_to_cpu(es->s_feature_incompat) &
2460               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2461                 printk(KERN_ERR "EXT4-fs: external journal has "
2462                                         "bad superblock\n");
2463                 brelse(bh);
2464                 goto out_bdev;
2465         }
2466
2467         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2468                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2469                 brelse(bh);
2470                 goto out_bdev;
2471         }
2472
2473         len = ext4_blocks_count(es);
2474         start = sb_block + 1;
2475         brelse(bh);     /* we're done with the superblock */
2476
2477         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2478                                         start, len, blocksize);
2479         if (!journal) {
2480                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2481                 goto out_bdev;
2482         }
2483         journal->j_private = sb;
2484         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2485         wait_on_buffer(journal->j_sb_buffer);
2486         if (!buffer_uptodate(journal->j_sb_buffer)) {
2487                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2488                 goto out_journal;
2489         }
2490         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2491                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2492                                         "user (unsupported) - %d\n",
2493                         be32_to_cpu(journal->j_superblock->s_nr_users));
2494                 goto out_journal;
2495         }
2496         EXT4_SB(sb)->journal_bdev = bdev;
2497         ext4_init_journal_params(sb, journal);
2498         return journal;
2499 out_journal:
2500         jbd2_journal_destroy(journal);
2501 out_bdev:
2502         ext4_blkdev_put(bdev);
2503         return NULL;
2504 }
2505
2506 static int ext4_load_journal(struct super_block *sb,
2507                              struct ext4_super_block *es,
2508                              unsigned long journal_devnum)
2509 {
2510         journal_t *journal;
2511         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2512         dev_t journal_dev;
2513         int err = 0;
2514         int really_read_only;
2515
2516         if (journal_devnum &&
2517             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2518                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2519                         "numbers have changed\n");
2520                 journal_dev = new_decode_dev(journal_devnum);
2521         } else
2522                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2523
2524         really_read_only = bdev_read_only(sb->s_bdev);
2525
2526         /*
2527          * Are we loading a blank journal or performing recovery after a
2528          * crash?  For recovery, we need to check in advance whether we
2529          * can get read-write access to the device.
2530          */
2531
2532         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2533                 if (sb->s_flags & MS_RDONLY) {
2534                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2535                                         "required on readonly filesystem.\n");
2536                         if (really_read_only) {
2537                                 printk(KERN_ERR "EXT4-fs: write access "
2538                                         "unavailable, cannot proceed.\n");
2539                                 return -EROFS;
2540                         }
2541                         printk (KERN_INFO "EXT4-fs: write access will "
2542                                         "be enabled during recovery.\n");
2543                 }
2544         }
2545
2546         if (journal_inum && journal_dev) {
2547                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2548                        "and inode journals!\n");
2549                 return -EINVAL;
2550         }
2551
2552         if (journal_inum) {
2553                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2554                         return -EINVAL;
2555         } else {
2556                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2557                         return -EINVAL;
2558         }
2559
2560         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2561                 err = jbd2_journal_update_format(journal);
2562                 if (err)  {
2563                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2564                         jbd2_journal_destroy(journal);
2565                         return err;
2566                 }
2567         }
2568
2569         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2570                 err = jbd2_journal_wipe(journal, !really_read_only);
2571         if (!err)
2572                 err = jbd2_journal_load(journal);
2573
2574         if (err) {
2575                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2576                 jbd2_journal_destroy(journal);
2577                 return err;
2578         }
2579
2580         EXT4_SB(sb)->s_journal = journal;
2581         ext4_clear_journal_err(sb, es);
2582
2583         if (journal_devnum &&
2584             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2585                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2586                 sb->s_dirt = 1;
2587
2588                 /* Make sure we flush the recovery flag to disk. */
2589                 ext4_commit_super(sb, es, 1);
2590         }
2591
2592         return 0;
2593 }
2594
2595 static int ext4_create_journal(struct super_block * sb,
2596                                struct ext4_super_block * es,
2597                                unsigned int journal_inum)
2598 {
2599         journal_t *journal;
2600         int err;
2601
2602         if (sb->s_flags & MS_RDONLY) {
2603                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2604                                 "create journal.\n");
2605                 return -EROFS;
2606         }
2607
2608         journal = ext4_get_journal(sb, journal_inum);
2609         if (!journal)
2610                 return -EINVAL;
2611
2612         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2613                journal_inum);
2614
2615         err = jbd2_journal_create(journal);
2616         if (err) {
2617                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2618                 jbd2_journal_destroy(journal);
2619                 return -EIO;
2620         }
2621
2622         EXT4_SB(sb)->s_journal = journal;
2623
2624         ext4_update_dynamic_rev(sb);
2625         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2626         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2627
2628         es->s_journal_inum = cpu_to_le32(journal_inum);
2629         sb->s_dirt = 1;
2630
2631         /* Make sure we flush the recovery flag to disk. */
2632         ext4_commit_super(sb, es, 1);
2633
2634         return 0;
2635 }
2636
2637 static void ext4_commit_super (struct super_block * sb,
2638                                struct ext4_super_block * es,
2639                                int sync)
2640 {
2641         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2642
2643         if (!sbh)
2644                 return;
2645         es->s_wtime = cpu_to_le32(get_seconds());
2646         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2647         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2648         BUFFER_TRACE(sbh, "marking dirty");
2649         mark_buffer_dirty(sbh);
2650         if (sync)
2651                 sync_dirty_buffer(sbh);
2652 }
2653
2654
2655 /*
2656  * Have we just finished recovery?  If so, and if we are mounting (or
2657  * remounting) the filesystem readonly, then we will end up with a
2658  * consistent fs on disk.  Record that fact.
2659  */
2660 static void ext4_mark_recovery_complete(struct super_block * sb,
2661                                         struct ext4_super_block * es)
2662 {
2663         journal_t *journal = EXT4_SB(sb)->s_journal;
2664
2665         jbd2_journal_lock_updates(journal);
2666         jbd2_journal_flush(journal);
2667         lock_super(sb);
2668         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2669             sb->s_flags & MS_RDONLY) {
2670                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2671                 sb->s_dirt = 0;
2672                 ext4_commit_super(sb, es, 1);
2673         }
2674         unlock_super(sb);
2675         jbd2_journal_unlock_updates(journal);
2676 }
2677
2678 /*
2679  * If we are mounting (or read-write remounting) a filesystem whose journal
2680  * has recorded an error from a previous lifetime, move that error to the
2681  * main filesystem now.
2682  */
2683 static void ext4_clear_journal_err(struct super_block * sb,
2684                                    struct ext4_super_block * es)
2685 {
2686         journal_t *journal;
2687         int j_errno;
2688         const char *errstr;
2689
2690         journal = EXT4_SB(sb)->s_journal;
2691
2692         /*
2693          * Now check for any error status which may have been recorded in the
2694          * journal by a prior ext4_error() or ext4_abort()
2695          */
2696
2697         j_errno = jbd2_journal_errno(journal);
2698         if (j_errno) {
2699                 char nbuf[16];
2700
2701                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2702                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2703                              "from previous mount: %s", errstr);
2704                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2705                              "filesystem check.");
2706
2707                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2708                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2709                 ext4_commit_super (sb, es, 1);
2710
2711                 jbd2_journal_clear_err(journal);
2712         }
2713 }
2714
2715 /*
2716  * Force the running and committing transactions to commit,
2717  * and wait on the commit.
2718  */
2719 int ext4_force_commit(struct super_block *sb)
2720 {
2721         journal_t *journal;
2722         int ret;
2723
2724         if (sb->s_flags & MS_RDONLY)
2725                 return 0;
2726
2727         journal = EXT4_SB(sb)->s_journal;
2728         sb->s_dirt = 0;
2729         ret = ext4_journal_force_commit(journal);
2730         return ret;
2731 }
2732
2733 /*
2734  * Ext4 always journals updates to the superblock itself, so we don't
2735  * have to propagate any other updates to the superblock on disk at this
2736  * point.  Just start an async writeback to get the buffers on their way
2737  * to the disk.
2738  *
2739  * This implicitly triggers the writebehind on sync().
2740  */
2741
2742 static void ext4_write_super (struct super_block * sb)
2743 {
2744         if (mutex_trylock(&sb->s_lock) != 0)
2745                 BUG();
2746         sb->s_dirt = 0;
2747 }
2748
2749 static int ext4_sync_fs(struct super_block *sb, int wait)
2750 {
2751         tid_t target;
2752
2753         sb->s_dirt = 0;
2754         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2755                 if (wait)
2756                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2757         }
2758         return 0;
2759 }
2760
2761 /*
2762  * LVM calls this function before a (read-only) snapshot is created.  This
2763  * gives us a chance to flush the journal completely and mark the fs clean.
2764  */
2765 static void ext4_write_super_lockfs(struct super_block *sb)
2766 {
2767         sb->s_dirt = 0;
2768
2769         if (!(sb->s_flags & MS_RDONLY)) {
2770                 journal_t *journal = EXT4_SB(sb)->s_journal;
2771
2772                 /* Now we set up the journal barrier. */
2773                 jbd2_journal_lock_updates(journal);
2774                 jbd2_journal_flush(journal);
2775
2776                 /* Journal blocked and flushed, clear needs_recovery flag. */
2777                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2778                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2779         }
2780 }
2781
2782 /*
2783  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2784  * flag here, even though the filesystem is not technically dirty yet.
2785  */
2786 static void ext4_unlockfs(struct super_block *sb)
2787 {
2788         if (!(sb->s_flags & MS_RDONLY)) {
2789                 lock_super(sb);
2790                 /* Reser the needs_recovery flag before the fs is unlocked. */
2791                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2792                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2793                 unlock_super(sb);
2794                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2795         }
2796 }
2797
2798 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2799 {
2800         struct ext4_super_block * es;
2801         struct ext4_sb_info *sbi = EXT4_SB(sb);
2802         ext4_fsblk_t n_blocks_count = 0;
2803         unsigned long old_sb_flags;
2804         struct ext4_mount_options old_opts;
2805         int err;
2806 #ifdef CONFIG_QUOTA
2807         int i;
2808 #endif
2809
2810         /* Store the original options */
2811         old_sb_flags = sb->s_flags;
2812         old_opts.s_mount_opt = sbi->s_mount_opt;
2813         old_opts.s_resuid = sbi->s_resuid;
2814         old_opts.s_resgid = sbi->s_resgid;
2815         old_opts.s_commit_interval = sbi->s_commit_interval;
2816 #ifdef CONFIG_QUOTA
2817         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2818         for (i = 0; i < MAXQUOTAS; i++)
2819                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2820 #endif
2821
2822         /*
2823          * Allow the "check" option to be passed as a remount option.
2824          */
2825         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2826                 err = -EINVAL;
2827                 goto restore_opts;
2828         }
2829
2830         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2831                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2832
2833         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2834                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2835
2836         es = sbi->s_es;
2837
2838         ext4_init_journal_params(sb, sbi->s_journal);
2839
2840         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2841                 n_blocks_count > ext4_blocks_count(es)) {
2842                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2843                         err = -EROFS;
2844                         goto restore_opts;
2845                 }
2846
2847                 if (*flags & MS_RDONLY) {
2848                         /*
2849                          * First of all, the unconditional stuff we have to do
2850                          * to disable replay of the journal when we next remount
2851                          */
2852                         sb->s_flags |= MS_RDONLY;
2853
2854                         /*
2855                          * OK, test if we are remounting a valid rw partition
2856                          * readonly, and if so set the rdonly flag and then
2857                          * mark the partition as valid again.
2858                          */
2859                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2860                             (sbi->s_mount_state & EXT4_VALID_FS))
2861                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2862
2863                         /*
2864                          * We have to unlock super so that we can wait for
2865                          * transactions.
2866                          */
2867                         unlock_super(sb);
2868                         ext4_mark_recovery_complete(sb, es);
2869                         lock_super(sb);
2870                 } else {
2871                         __le32 ret;
2872                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2873                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2874                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2875                                        "remount RDWR because of unsupported "
2876                                        "optional features (%x).\n",
2877                                        sb->s_id, le32_to_cpu(ret));
2878                                 err = -EROFS;
2879                                 goto restore_opts;
2880                         }
2881
2882                         /*
2883                          * If we have an unprocessed orphan list hanging
2884                          * around from a previously readonly bdev mount,
2885                          * require a full umount/remount for now.
2886                          */
2887                         if (es->s_last_orphan) {
2888                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2889                                        "remount RDWR because of unprocessed "
2890                                        "orphan inode list.  Please "
2891                                        "umount/remount instead.\n",
2892                                        sb->s_id);
2893                                 err = -EINVAL;
2894                                 goto restore_opts;
2895                         }
2896
2897                         /*
2898                          * Mounting a RDONLY partition read-write, so reread
2899                          * and store the current valid flag.  (It may have
2900                          * been changed by e2fsck since we originally mounted
2901                          * the partition.)
2902                          */
2903                         ext4_clear_journal_err(sb, es);
2904                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2905                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2906                                 goto restore_opts;
2907                         if (!ext4_setup_super (sb, es, 0))
2908                                 sb->s_flags &= ~MS_RDONLY;
2909                 }
2910         }
2911 #ifdef CONFIG_QUOTA
2912         /* Release old quota file names */
2913         for (i = 0; i < MAXQUOTAS; i++)
2914                 if (old_opts.s_qf_names[i] &&
2915                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2916                         kfree(old_opts.s_qf_names[i]);
2917 #endif
2918         return 0;
2919 restore_opts:
2920         sb->s_flags = old_sb_flags;
2921         sbi->s_mount_opt = old_opts.s_mount_opt;
2922         sbi->s_resuid = old_opts.s_resuid;
2923         sbi->s_resgid = old_opts.s_resgid;
2924         sbi->s_commit_interval = old_opts.s_commit_interval;
2925 #ifdef CONFIG_QUOTA
2926         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2927         for (i = 0; i < MAXQUOTAS; i++) {
2928                 if (sbi->s_qf_names[i] &&
2929                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2930                         kfree(sbi->s_qf_names[i]);
2931                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2932         }
2933 #endif
2934         return err;
2935 }
2936
2937 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2938 {
2939         struct super_block *sb = dentry->d_sb;
2940         struct ext4_sb_info *sbi = EXT4_SB(sb);
2941         struct ext4_super_block *es = sbi->s_es;
2942         u64 fsid;
2943
2944         if (test_opt(sb, MINIX_DF)) {
2945                 sbi->s_overhead_last = 0;
2946         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2947                 ext4_group_t ngroups = sbi->s_groups_count, i;
2948                 ext4_fsblk_t overhead = 0;
2949                 smp_rmb();
2950
2951                 /*
2952                  * Compute the overhead (FS structures).  This is constant
2953                  * for a given filesystem unless the number of block groups
2954                  * changes so we cache the previous value until it does.
2955                  */
2956
2957                 /*
2958                  * All of the blocks before first_data_block are
2959                  * overhead
2960                  */
2961                 overhead = le32_to_cpu(es->s_first_data_block);
2962
2963                 /*
2964                  * Add the overhead attributed to the superblock and
2965                  * block group descriptors.  If the sparse superblocks
2966                  * feature is turned on, then not all groups have this.
2967                  */
2968                 for (i = 0; i < ngroups; i++) {
2969                         overhead += ext4_bg_has_super(sb, i) +
2970                                 ext4_bg_num_gdb(sb, i);
2971                         cond_resched();
2972                 }
2973
2974                 /*
2975                  * Every block group has an inode bitmap, a block
2976                  * bitmap, and an inode table.
2977                  */
2978                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2979                 sbi->s_overhead_last = overhead;
2980                 smp_wmb();
2981                 sbi->s_blocks_last = ext4_blocks_count(es);
2982         }
2983
2984         buf->f_type = EXT4_SUPER_MAGIC;
2985         buf->f_bsize = sb->s_blocksize;
2986         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2987         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2988         ext4_free_blocks_count_set(es, buf->f_bfree);
2989         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2990         if (buf->f_bfree < ext4_r_blocks_count(es))
2991                 buf->f_bavail = 0;
2992         buf->f_files = le32_to_cpu(es->s_inodes_count);
2993         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2994         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2995         buf->f_namelen = EXT4_NAME_LEN;
2996         fsid = le64_to_cpup((void *)es->s_uuid) ^
2997                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2998         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2999         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3000         return 0;
3001 }
3002
3003 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3004  * is locked for write. Otherwise the are possible deadlocks:
3005  * Process 1                         Process 2
3006  * ext4_create()                     quota_sync()
3007  *   jbd2_journal_start()                   write_dquot()
3008  *   DQUOT_INIT()                        down(dqio_mutex)
3009  *     down(dqio_mutex)                    jbd2_journal_start()
3010  *
3011  */
3012
3013 #ifdef CONFIG_QUOTA
3014
3015 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3016 {
3017         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3018 }
3019
3020 static int ext4_dquot_initialize(struct inode *inode, int type)
3021 {
3022         handle_t *handle;
3023         int ret, err;
3024
3025         /* We may create quota structure so we need to reserve enough blocks */
3026         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3027         if (IS_ERR(handle))
3028                 return PTR_ERR(handle);
3029         ret = dquot_initialize(inode, type);
3030         err = ext4_journal_stop(handle);
3031         if (!ret)
3032                 ret = err;
3033         return ret;
3034 }
3035
3036 static int ext4_dquot_drop(struct inode *inode)
3037 {
3038         handle_t *handle;
3039         int ret, err;
3040
3041         /* We may delete quota structure so we need to reserve enough blocks */
3042         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3043         if (IS_ERR(handle))
3044                 return PTR_ERR(handle);
3045         ret = dquot_drop(inode);
3046         err = ext4_journal_stop(handle);
3047         if (!ret)
3048                 ret = err;
3049         return ret;
3050 }
3051
3052 static int ext4_write_dquot(struct dquot *dquot)
3053 {
3054         int ret, err;
3055         handle_t *handle;
3056         struct inode *inode;
3057
3058         inode = dquot_to_inode(dquot);
3059         handle = ext4_journal_start(inode,
3060                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3061         if (IS_ERR(handle))
3062                 return PTR_ERR(handle);
3063         ret = dquot_commit(dquot);
3064         err = ext4_journal_stop(handle);
3065         if (!ret)
3066                 ret = err;
3067         return ret;
3068 }
3069
3070 static int ext4_acquire_dquot(struct dquot *dquot)
3071 {
3072         int ret, err;
3073         handle_t *handle;
3074
3075         handle = ext4_journal_start(dquot_to_inode(dquot),
3076                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3077         if (IS_ERR(handle))
3078                 return PTR_ERR(handle);
3079         ret = dquot_acquire(dquot);
3080         err = ext4_journal_stop(handle);
3081         if (!ret)
3082                 ret = err;
3083         return ret;
3084 }
3085
3086 static int ext4_release_dquot(struct dquot *dquot)
3087 {
3088         int ret, err;
3089         handle_t *handle;
3090
3091         handle = ext4_journal_start(dquot_to_inode(dquot),
3092                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3093         if (IS_ERR(handle)) {
3094                 /* Release dquot anyway to avoid endless cycle in dqput() */
3095                 dquot_release(dquot);
3096                 return PTR_ERR(handle);
3097         }
3098         ret = dquot_release(dquot);
3099         err = ext4_journal_stop(handle);
3100         if (!ret)
3101                 ret = err;
3102         return ret;
3103 }
3104
3105 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3106 {
3107         /* Are we journalling quotas? */
3108         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3109             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3110                 dquot_mark_dquot_dirty(dquot);
3111                 return ext4_write_dquot(dquot);
3112         } else {
3113                 return dquot_mark_dquot_dirty(dquot);
3114         }
3115 }
3116
3117 static int ext4_write_info(struct super_block *sb, int type)
3118 {
3119         int ret, err;
3120         handle_t *handle;
3121
3122         /* Data block + inode block */
3123         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3124         if (IS_ERR(handle))
3125                 return PTR_ERR(handle);
3126         ret = dquot_commit_info(sb, type);
3127         err = ext4_journal_stop(handle);
3128         if (!ret)
3129                 ret = err;
3130         return ret;
3131 }
3132
3133 /*
3134  * Turn on quotas during mount time - we need to find
3135  * the quota file and such...
3136  */
3137 static int ext4_quota_on_mount(struct super_block *sb, int type)
3138 {
3139         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3140                         EXT4_SB(sb)->s_jquota_fmt, type);
3141 }
3142
3143 /*
3144  * Standard function to be called on quota_on
3145  */
3146 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3147                          char *path, int remount)
3148 {
3149         int err;
3150         struct nameidata nd;
3151
3152         if (!test_opt(sb, QUOTA))
3153                 return -EINVAL;
3154         /* Not journalling quota? */
3155         if ((!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
3156             !EXT4_SB(sb)->s_qf_names[GRPQUOTA]) || remount)
3157                 return vfs_quota_on(sb, type, format_id, path, remount);
3158         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3159         if (err)
3160                 return err;
3161         /* Quotafile not on the same filesystem? */
3162         if (nd.path.mnt->mnt_sb != sb) {
3163                 path_put(&nd.path);
3164                 return -EXDEV;
3165         }
3166         /* Quotafile not of fs root? */
3167         if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3168                 printk(KERN_WARNING
3169                         "EXT4-fs: Quota file not on filesystem root. "
3170                         "Journalled quota will not work.\n");
3171         path_put(&nd.path);
3172         return vfs_quota_on(sb, type, format_id, path, remount);
3173 }
3174
3175 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3176  * acquiring the locks... As quota files are never truncated and quota code
3177  * itself serializes the operations (and noone else should touch the files)
3178  * we don't have to be afraid of races */
3179 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3180                                size_t len, loff_t off)
3181 {
3182         struct inode *inode = sb_dqopt(sb)->files[type];
3183         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3184         int err = 0;
3185         int offset = off & (sb->s_blocksize - 1);
3186         int tocopy;
3187         size_t toread;
3188         struct buffer_head *bh;
3189         loff_t i_size = i_size_read(inode);
3190
3191         if (off > i_size)
3192                 return 0;
3193         if (off+len > i_size)
3194                 len = i_size-off;
3195         toread = len;
3196         while (toread > 0) {
3197                 tocopy = sb->s_blocksize - offset < toread ?
3198                                 sb->s_blocksize - offset : toread;
3199                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3200                 if (err)
3201                         return err;
3202                 if (!bh)        /* A hole? */
3203                         memset(data, 0, tocopy);
3204                 else
3205                         memcpy(data, bh->b_data+offset, tocopy);
3206                 brelse(bh);
3207                 offset = 0;
3208                 toread -= tocopy;
3209                 data += tocopy;
3210                 blk++;
3211         }
3212         return len;
3213 }
3214
3215 /* Write to quotafile (we know the transaction is already started and has
3216  * enough credits) */
3217 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3218                                 const char *data, size_t len, loff_t off)
3219 {
3220         struct inode *inode = sb_dqopt(sb)->files[type];
3221         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3222         int err = 0;
3223         int offset = off & (sb->s_blocksize - 1);
3224         int tocopy;
3225         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3226         size_t towrite = len;
3227         struct buffer_head *bh;
3228         handle_t *handle = journal_current_handle();
3229
3230         if (!handle) {
3231                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3232                         " cancelled because transaction is not started.\n",
3233                         (unsigned long long)off, (unsigned long long)len);
3234                 return -EIO;
3235         }
3236         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3237         while (towrite > 0) {
3238                 tocopy = sb->s_blocksize - offset < towrite ?
3239                                 sb->s_blocksize - offset : towrite;
3240                 bh = ext4_bread(handle, inode, blk, 1, &err);
3241                 if (!bh)
3242                         goto out;
3243                 if (journal_quota) {
3244                         err = ext4_journal_get_write_access(handle, bh);
3245                         if (err) {
3246                                 brelse(bh);
3247                                 goto out;
3248                         }
3249                 }
3250                 lock_buffer(bh);
3251                 memcpy(bh->b_data+offset, data, tocopy);
3252                 flush_dcache_page(bh->b_page);
3253                 unlock_buffer(bh);
3254                 if (journal_quota)
3255                         err = ext4_journal_dirty_metadata(handle, bh);
3256                 else {
3257                         /* Always do at least ordered writes for quotas */
3258                         err = ext4_journal_dirty_data(handle, bh);
3259                         mark_buffer_dirty(bh);
3260                 }
3261                 brelse(bh);
3262                 if (err)
3263                         goto out;
3264                 offset = 0;
3265                 towrite -= tocopy;
3266                 data += tocopy;
3267                 blk++;
3268         }
3269 out:
3270         if (len == towrite)
3271                 return err;
3272         if (inode->i_size < off+len-towrite) {
3273                 i_size_write(inode, off+len-towrite);
3274                 EXT4_I(inode)->i_disksize = inode->i_size;
3275         }
3276         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3277         ext4_mark_inode_dirty(handle, inode);
3278         mutex_unlock(&inode->i_mutex);
3279         return len - towrite;
3280 }
3281
3282 #endif
3283
3284 static int ext4_get_sb(struct file_system_type *fs_type,
3285         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3286 {
3287         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3288 }
3289
3290 static struct file_system_type ext4dev_fs_type = {
3291         .owner          = THIS_MODULE,
3292         .name           = "ext4dev",
3293         .get_sb         = ext4_get_sb,
3294         .kill_sb        = kill_block_super,
3295         .fs_flags       = FS_REQUIRES_DEV,
3296 };
3297
3298 static int __init init_ext4_fs(void)
3299 {
3300         int err;
3301
3302         err = init_ext4_mballoc();
3303         if (err)
3304                 return err;
3305
3306         err = init_ext4_xattr();
3307         if (err)
3308                 goto out2;
3309         err = init_inodecache();
3310         if (err)
3311                 goto out1;
3312         err = register_filesystem(&ext4dev_fs_type);
3313         if (err)
3314                 goto out;
3315         return 0;
3316 out:
3317         destroy_inodecache();
3318 out1:
3319         exit_ext4_xattr();
3320 out2:
3321         exit_ext4_mballoc();
3322         return err;
3323 }
3324
3325 static void __exit exit_ext4_fs(void)
3326 {
3327         unregister_filesystem(&ext4dev_fs_type);
3328         destroy_inodecache();
3329         exit_ext4_xattr();
3330         exit_ext4_mballoc();
3331 }
3332
3333 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3334 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3335 MODULE_LICENSE("GPL");
3336 module_init(init_ext4_fs)
3337 module_exit(exit_ext4_fs)