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/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38
  39 #include <asm/uaccess.h>
  40
  41 #include "xattr.h"
  42 #include "acl.h"
  43 #include "namei.h"
  44
  45 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  46                              unsigned long journal_devnum);
  47 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  48                                unsigned int);
  49 static void ext4_commit_super (struct super_block * sb,
  50                                struct ext4_super_block * es,
  51                                int sync);
  52 static void ext4_mark_recovery_complete(struct super_block * sb,
  53                                         struct ext4_super_block * es);
  54 static void ext4_clear_journal_err(struct super_block * sb,
  55                                    struct ext4_super_block * es);
  56 static int ext4_sync_fs(struct super_block *sb, int wait);
  57 static const char *ext4_decode_error(struct super_block * sb, int errno,
  58                                      char nbuf[16]);
  59 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  60 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  61 static void ext4_unlockfs(struct super_block *sb);
  62 static void ext4_write_super (struct super_block * sb);
  63 static void ext4_write_super_lockfs(struct super_block *sb);
  64
  65
  66 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  67                                struct ext4_group_desc *bg)
  68 {
  69         return le32_to_cpu(bg->bg_block_bitmap) |
  70                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  71                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  72 }
  73
  74 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  75                                struct ext4_group_desc *bg)
  76 {
  77         return le32_to_cpu(bg->bg_inode_bitmap) |
  78                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  79                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  80 }
  81
  82 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  83                               struct ext4_group_desc *bg)
  84 {
  85         return le32_to_cpu(bg->bg_inode_table) |
  86                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  87                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  88 }
  89
  90 void ext4_block_bitmap_set(struct super_block *sb,
  91                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  92 {
  93         bg->bg_block_bitmap = cpu_to_le32((u32)blk);
  94         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  95                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  96 }
  97
  98 void ext4_inode_bitmap_set(struct super_block *sb,
  99                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 100 {
 101         bg->bg_inode_bitmap  = cpu_to_le32((u32)blk);
 102         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 103                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 104 }
 105
 106 void ext4_inode_table_set(struct super_block *sb,
 107                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 108 {
 109         bg->bg_inode_table = cpu_to_le32((u32)blk);
 110         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 111                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 112 }
 113
 114 /*
 115  * Wrappers for jbd2_journal_start/end.
 116  *
 117  * The only special thing we need to do here is to make sure that all
 118  * journal_end calls result in the superblock being marked dirty, so
 119  * that sync() will call the filesystem's write_super callback if
 120  * appropriate.
 121  */
 122 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 123 {
 124         journal_t *journal;
 125
 126         if (sb->s_flags & MS_RDONLY)
 127                 return ERR_PTR(-EROFS);
 128
 129         /* Special case here: if the journal has aborted behind our
 130          * backs (eg. EIO in the commit thread), then we still need to
 131          * take the FS itself readonly cleanly. */
 132         journal = EXT4_SB(sb)->s_journal;
 133         if (is_journal_aborted(journal)) {
 134                 ext4_abort(sb, __FUNCTION__,
 135                            "Detected aborted journal");
 136                 return ERR_PTR(-EROFS);
 137         }
 138
 139         return jbd2_journal_start(journal, nblocks);
 140 }
 141
 142 /*
 143  * The only special thing we need to do here is to make sure that all
 144  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 145  * that sync() will call the filesystem's write_super callback if
 146  * appropriate.
 147  */
 148 int __ext4_journal_stop(const char *where, handle_t *handle)
 149 {
 150         struct super_block *sb;
 151         int err;
 152         int rc;
 153
 154         sb = handle->h_transaction->t_journal->j_private;
 155         err = handle->h_err;
 156         rc = jbd2_journal_stop(handle);
 157
 158         if (!err)
 159                 err = rc;
 160         if (err)
 161                 __ext4_std_error(sb, where, err);
 162         return err;
 163 }
 164
 165 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 166                 struct buffer_head *bh, handle_t *handle, int err)
 167 {
 168         char nbuf[16];
 169         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 170
 171         if (bh)
 172                 BUFFER_TRACE(bh, "abort");
 173
 174         if (!handle->h_err)
 175                 handle->h_err = err;
 176
 177         if (is_handle_aborted(handle))
 178                 return;
 179
 180         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 181                caller, errstr, err_fn);
 182
 183         jbd2_journal_abort_handle(handle);
 184 }
 185
 186 /* Deal with the reporting of failure conditions on a filesystem such as
 187  * inconsistencies detected or read IO failures.
 188  *
 189  * On ext2, we can store the error state of the filesystem in the
 190  * superblock.  That is not possible on ext4, because we may have other
 191  * write ordering constraints on the superblock which prevent us from
 192  * writing it out straight away; and given that the journal is about to
 193  * be aborted, we can't rely on the current, or future, transactions to
 194  * write out the superblock safely.
 195  *
 196  * We'll just use the jbd2_journal_abort() error code to record an error in
 197  * the journal instead.  On recovery, the journal will compain about
 198  * that error until we've noted it down and cleared it.
 199  */
 200
 201 static void ext4_handle_error(struct super_block *sb)
 202 {
 203         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 204
 205         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 206         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 207
 208         if (sb->s_flags & MS_RDONLY)
 209                 return;
 210
 211         if (!test_opt (sb, ERRORS_CONT)) {
 212                 journal_t *journal = EXT4_SB(sb)->s_journal;
 213
 214                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 215                 if (journal)
 216                         jbd2_journal_abort(journal, -EIO);
 217         }
 218         if (test_opt (sb, ERRORS_RO)) {
 219                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 220                 sb->s_flags |= MS_RDONLY;
 221         }
 222         ext4_commit_super(sb, es, 1);
 223         if (test_opt(sb, ERRORS_PANIC))
 224                 panic("EXT4-fs (device %s): panic forced after error\n",
 225                         sb->s_id);
 226 }
 227
 228 void ext4_error (struct super_block * sb, const char * function,
 229                  const char * fmt, ...)
 230 {
 231         va_list args;
 232
 233         va_start(args, fmt);
 234         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 235         vprintk(fmt, args);
 236         printk("\n");
 237         va_end(args);
 238
 239         ext4_handle_error(sb);
 240 }
 241
 242 static const char *ext4_decode_error(struct super_block * sb, int errno,
 243                                      char nbuf[16])
 244 {
 245         char *errstr = NULL;
 246
 247         switch (errno) {
 248         case -EIO:
 249                 errstr = "IO failure";
 250                 break;
 251         case -ENOMEM:
 252                 errstr = "Out of memory";
 253                 break;
 254         case -EROFS:
 255                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 256                         errstr = "Journal has aborted";
 257                 else
 258                         errstr = "Readonly filesystem";
 259                 break;
 260         default:
 261                 /* If the caller passed in an extra buffer for unknown
 262                  * errors, textualise them now.  Else we just return
 263                  * NULL. */
 264                 if (nbuf) {
 265                         /* Check for truncated error codes... */
 266                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 267                                 errstr = nbuf;
 268                 }
 269                 break;
 270         }
 271
 272         return errstr;
 273 }
 274
 275 /* __ext4_std_error decodes expected errors from journaling functions
 276  * automatically and invokes the appropriate error response.  */
 277
 278 void __ext4_std_error (struct super_block * sb, const char * function,
 279                        int errno)
 280 {
 281         char nbuf[16];
 282         const char *errstr;
 283
 284         /* Special case: if the error is EROFS, and we're not already
 285          * inside a transaction, then there's really no point in logging
 286          * an error. */
 287         if (errno == -EROFS && journal_current_handle() == NULL &&
 288             (sb->s_flags & MS_RDONLY))
 289                 return;
 290
 291         errstr = ext4_decode_error(sb, errno, nbuf);
 292         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 293                 sb->s_id, function, errstr);
 294
 295         ext4_handle_error(sb);
 296 }
 297
 298 /*
 299  * ext4_abort is a much stronger failure handler than ext4_error.  The
 300  * abort function may be used to deal with unrecoverable failures such
 301  * as journal IO errors or ENOMEM at a critical moment in log management.
 302  *
 303  * We unconditionally force the filesystem into an ABORT|READONLY state,
 304  * unless the error response on the fs has been set to panic in which
 305  * case we take the easy way out and panic immediately.
 306  */
 307
 308 void ext4_abort (struct super_block * sb, const char * function,
 309                  const char * fmt, ...)
 310 {
 311         va_list args;
 312
 313         printk (KERN_CRIT "ext4_abort called.\n");
 314
 315         va_start(args, fmt);
 316         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 317         vprintk(fmt, args);
 318         printk("\n");
 319         va_end(args);
 320
 321         if (test_opt(sb, ERRORS_PANIC))
 322                 panic("EXT4-fs panic from previous error\n");
 323
 324         if (sb->s_flags & MS_RDONLY)
 325                 return;
 326
 327         printk(KERN_CRIT "Remounting filesystem read-only\n");
 328         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 329         sb->s_flags |= MS_RDONLY;
 330         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 331         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 332 }
 333
 334 void ext4_warning (struct super_block * sb, const char * function,
 335                    const char * fmt, ...)
 336 {
 337         va_list args;
 338
 339         va_start(args, fmt);
 340         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 341                sb->s_id, function);
 342         vprintk(fmt, args);
 343         printk("\n");
 344         va_end(args);
 345 }
 346
 347 void ext4_update_dynamic_rev(struct super_block *sb)
 348 {
 349         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 350
 351         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 352                 return;
 353
 354         ext4_warning(sb, __FUNCTION__,
 355                      "updating to rev %d because of new feature flag, "
 356                      "running e2fsck is recommended",
 357                      EXT4_DYNAMIC_REV);
 358
 359         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 360         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 361         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 362         /* leave es->s_feature_*compat flags alone */
 363         /* es->s_uuid will be set by e2fsck if empty */
 364
 365         /*
 366          * The rest of the superblock fields should be zero, and if not it
 367          * means they are likely already in use, so leave them alone.  We
 368          * can leave it up to e2fsck to clean up any inconsistencies there.
 369          */
 370 }
 371
 372 /*
 373  * Open the external journal device
 374  */
 375 static struct block_device *ext4_blkdev_get(dev_t dev)
 376 {
 377         struct block_device *bdev;
 378         char b[BDEVNAME_SIZE];
 379
 380         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 381         if (IS_ERR(bdev))
 382                 goto fail;
 383         return bdev;
 384
 385 fail:
 386         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 387                         __bdevname(dev, b), PTR_ERR(bdev));
 388         return NULL;
 389 }
 390
 391 /*
 392  * Release the journal device
 393  */
 394 static int ext4_blkdev_put(struct block_device *bdev)
 395 {
 396         bd_release(bdev);
 397         return blkdev_put(bdev);
 398 }
 399
 400 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 401 {
 402         struct block_device *bdev;
 403         int ret = -ENODEV;
 404
 405         bdev = sbi->journal_bdev;
 406         if (bdev) {
 407                 ret = ext4_blkdev_put(bdev);
 408                 sbi->journal_bdev = NULL;
 409         }
 410         return ret;
 411 }
 412
 413 static inline struct inode *orphan_list_entry(struct list_head *l)
 414 {
 415         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 416 }
 417
 418 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 419 {
 420         struct list_head *l;
 421
 422         printk(KERN_ERR "sb orphan head is %d\n",
 423                le32_to_cpu(sbi->s_es->s_last_orphan));
 424
 425         printk(KERN_ERR "sb_info orphan list:\n");
 426         list_for_each(l, &sbi->s_orphan) {
 427                 struct inode *inode = orphan_list_entry(l);
 428                 printk(KERN_ERR "  "
 429                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 430                        inode->i_sb->s_id, inode->i_ino, inode,
 431                        inode->i_mode, inode->i_nlink,
 432                        NEXT_ORPHAN(inode));
 433         }
 434 }
 435
 436 static void ext4_put_super (struct super_block * sb)
 437 {
 438         struct ext4_sb_info *sbi = EXT4_SB(sb);
 439         struct ext4_super_block *es = sbi->s_es;
 440         int i;
 441
 442         ext4_ext_release(sb);
 443         ext4_xattr_put_super(sb);
 444         jbd2_journal_destroy(sbi->s_journal);
 445         if (!(sb->s_flags & MS_RDONLY)) {
 446                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 447                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 448                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 449                 mark_buffer_dirty(sbi->s_sbh);
 450                 ext4_commit_super(sb, es, 1);
 451         }
 452
 453         for (i = 0; i < sbi->s_gdb_count; i++)
 454                 brelse(sbi->s_group_desc[i]);
 455         kfree(sbi->s_group_desc);
 456         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 457         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 458         percpu_counter_destroy(&sbi->s_dirs_counter);
 459         brelse(sbi->s_sbh);
 460 #ifdef CONFIG_QUOTA
 461         for (i = 0; i < MAXQUOTAS; i++)
 462                 kfree(sbi->s_qf_names[i]);
 463 #endif
 464
 465         /* Debugging code just in case the in-memory inode orphan list
 466          * isn't empty.  The on-disk one can be non-empty if we've
 467          * detected an error and taken the fs readonly, but the
 468          * in-memory list had better be clean by this point. */
 469         if (!list_empty(&sbi->s_orphan))
 470                 dump_orphan_list(sb, sbi);
 471         J_ASSERT(list_empty(&sbi->s_orphan));
 472
 473         invalidate_bdev(sb->s_bdev);
 474         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 475                 /*
 476                  * Invalidate the journal device's buffers.  We don't want them
 477                  * floating about in memory - the physical journal device may
 478                  * hotswapped, and it breaks the `ro-after' testing code.
 479                  */
 480                 sync_blockdev(sbi->journal_bdev);
 481                 invalidate_bdev(sbi->journal_bdev);
 482                 ext4_blkdev_remove(sbi);
 483         }
 484         sb->s_fs_info = NULL;
 485         kfree(sbi);
 486         return;
 487 }
 488
 489 static struct kmem_cache *ext4_inode_cachep;
 490
 491 /*
 492  * Called inside transaction, so use GFP_NOFS
 493  */
 494 static struct inode *ext4_alloc_inode(struct super_block *sb)
 495 {
 496         struct ext4_inode_info *ei;
 497
 498         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 499         if (!ei)
 500                 return NULL;
 501 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 502         ei->i_acl = EXT4_ACL_NOT_CACHED;
 503         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 504 #endif
 505         ei->i_block_alloc_info = NULL;
 506         ei->vfs_inode.i_version = 1;
 507         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 508         return &ei->vfs_inode;
 509 }
 510
 511 static void ext4_destroy_inode(struct inode *inode)
 512 {
 513         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 514 }
 515
 516 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
 517 {
 518         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 519
 520         if (flags & SLAB_CTOR_CONSTRUCTOR) {
 521                 INIT_LIST_HEAD(&ei->i_orphan);
 522 #ifdef CONFIG_EXT4DEV_FS_XATTR
 523                 init_rwsem(&ei->xattr_sem);
 524 #endif
 525                 mutex_init(&ei->truncate_mutex);
 526                 inode_init_once(&ei->vfs_inode);
 527         }
 528 }
 529
 530 static int init_inodecache(void)
 531 {
 532         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 533                                              sizeof(struct ext4_inode_info),
 534                                              0, (SLAB_RECLAIM_ACCOUNT|
 535                                                 SLAB_MEM_SPREAD),
 536                                              init_once, NULL);
 537         if (ext4_inode_cachep == NULL)
 538                 return -ENOMEM;
 539         return 0;
 540 }
 541
 542 static void destroy_inodecache(void)
 543 {
 544         kmem_cache_destroy(ext4_inode_cachep);
 545 }
 546
 547 static void ext4_clear_inode(struct inode *inode)
 548 {
 549         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 550 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 551         if (EXT4_I(inode)->i_acl &&
 552                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 553                 posix_acl_release(EXT4_I(inode)->i_acl);
 554                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 555         }
 556         if (EXT4_I(inode)->i_default_acl &&
 557                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 558                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 559                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 560         }
 561 #endif
 562         ext4_discard_reservation(inode);
 563         EXT4_I(inode)->i_block_alloc_info = NULL;
 564         if (unlikely(rsv))
 565                 kfree(rsv);
 566 }
 567
 568 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 569 {
 570 #if defined(CONFIG_QUOTA)
 571         struct ext4_sb_info *sbi = EXT4_SB(sb);
 572
 573         if (sbi->s_jquota_fmt)
 574                 seq_printf(seq, ",jqfmt=%s",
 575                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 576
 577         if (sbi->s_qf_names[USRQUOTA])
 578                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 579
 580         if (sbi->s_qf_names[GRPQUOTA])
 581                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 582
 583         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 584                 seq_puts(seq, ",usrquota");
 585
 586         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 587                 seq_puts(seq, ",grpquota");
 588 #endif
 589 }
 590
 591 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 592 {
 593         struct super_block *sb = vfs->mnt_sb;
 594
 595         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 596                 seq_puts(seq, ",data=journal");
 597         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 598                 seq_puts(seq, ",data=ordered");
 599         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 600                 seq_puts(seq, ",data=writeback");
 601
 602         ext4_show_quota_options(seq, sb);
 603
 604         return 0;
 605 }
 606
 607
 608 static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
 609 {
 610         __u32 *objp = vobjp;
 611         unsigned long ino = objp[0];
 612         __u32 generation = objp[1];
 613         struct inode *inode;
 614         struct dentry *result;
 615
 616         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 617                 return ERR_PTR(-ESTALE);
 618         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 619                 return ERR_PTR(-ESTALE);
 620
 621         /* iget isn't really right if the inode is currently unallocated!!
 622          *
 623          * ext4_read_inode will return a bad_inode if the inode had been
 624          * deleted, so we should be safe.
 625          *
 626          * Currently we don't know the generation for parent directory, so
 627          * a generation of 0 means "accept any"
 628          */
 629         inode = iget(sb, ino);
 630         if (inode == NULL)
 631                 return ERR_PTR(-ENOMEM);
 632         if (is_bad_inode(inode) ||
 633             (generation && inode->i_generation != generation)) {
 634                 iput(inode);
 635                 return ERR_PTR(-ESTALE);
 636         }
 637         /* now to find a dentry.
 638          * If possible, get a well-connected one
 639          */
 640         result = d_alloc_anon(inode);
 641         if (!result) {
 642                 iput(inode);
 643                 return ERR_PTR(-ENOMEM);
 644         }
 645         return result;
 646 }
 647
 648 #ifdef CONFIG_QUOTA
 649 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 650 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 651
 652 static int ext4_dquot_initialize(struct inode *inode, int type);
 653 static int ext4_dquot_drop(struct inode *inode);
 654 static int ext4_write_dquot(struct dquot *dquot);
 655 static int ext4_acquire_dquot(struct dquot *dquot);
 656 static int ext4_release_dquot(struct dquot *dquot);
 657 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 658 static int ext4_write_info(struct super_block *sb, int type);
 659 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 660 static int ext4_quota_on_mount(struct super_block *sb, int type);
 661 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 662                                size_t len, loff_t off);
 663 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 664                                 const char *data, size_t len, loff_t off);
 665
 666 static struct dquot_operations ext4_quota_operations = {
 667         .initialize     = ext4_dquot_initialize,
 668         .drop           = ext4_dquot_drop,
 669         .alloc_space    = dquot_alloc_space,
 670         .alloc_inode    = dquot_alloc_inode,
 671         .free_space     = dquot_free_space,
 672         .free_inode     = dquot_free_inode,
 673         .transfer       = dquot_transfer,
 674         .write_dquot    = ext4_write_dquot,
 675         .acquire_dquot  = ext4_acquire_dquot,
 676         .release_dquot  = ext4_release_dquot,
 677         .mark_dirty     = ext4_mark_dquot_dirty,
 678         .write_info     = ext4_write_info
 679 };
 680
 681 static struct quotactl_ops ext4_qctl_operations = {
 682         .quota_on       = ext4_quota_on,
 683         .quota_off      = vfs_quota_off,
 684         .quota_sync     = vfs_quota_sync,
 685         .get_info       = vfs_get_dqinfo,
 686         .set_info       = vfs_set_dqinfo,
 687         .get_dqblk      = vfs_get_dqblk,
 688         .set_dqblk      = vfs_set_dqblk
 689 };
 690 #endif
 691
 692 static const struct super_operations ext4_sops = {
 693         .alloc_inode    = ext4_alloc_inode,
 694         .destroy_inode  = ext4_destroy_inode,
 695         .read_inode     = ext4_read_inode,
 696         .write_inode    = ext4_write_inode,
 697         .dirty_inode    = ext4_dirty_inode,
 698         .delete_inode   = ext4_delete_inode,
 699         .put_super      = ext4_put_super,
 700         .write_super    = ext4_write_super,
 701         .sync_fs        = ext4_sync_fs,
 702         .write_super_lockfs = ext4_write_super_lockfs,
 703         .unlockfs       = ext4_unlockfs,
 704         .statfs         = ext4_statfs,
 705         .remount_fs     = ext4_remount,
 706         .clear_inode    = ext4_clear_inode,
 707         .show_options   = ext4_show_options,
 708 #ifdef CONFIG_QUOTA
 709         .quota_read     = ext4_quota_read,
 710         .quota_write    = ext4_quota_write,
 711 #endif
 712 };
 713
 714 static struct export_operations ext4_export_ops = {
 715         .get_parent = ext4_get_parent,
 716         .get_dentry = ext4_get_dentry,
 717 };
 718
 719 enum {
 720         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 721         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 722         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 723         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 724         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 725         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 726         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 727         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 728         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 729         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 730         Opt_grpquota, Opt_extents,
 731 };
 732
 733 static match_table_t tokens = {
 734         {Opt_bsd_df, "bsddf"},
 735         {Opt_minix_df, "minixdf"},
 736         {Opt_grpid, "grpid"},
 737         {Opt_grpid, "bsdgroups"},
 738         {Opt_nogrpid, "nogrpid"},
 739         {Opt_nogrpid, "sysvgroups"},
 740         {Opt_resgid, "resgid=%u"},
 741         {Opt_resuid, "resuid=%u"},
 742         {Opt_sb, "sb=%u"},
 743         {Opt_err_cont, "errors=continue"},
 744         {Opt_err_panic, "errors=panic"},
 745         {Opt_err_ro, "errors=remount-ro"},
 746         {Opt_nouid32, "nouid32"},
 747         {Opt_nocheck, "nocheck"},
 748         {Opt_nocheck, "check=none"},
 749         {Opt_debug, "debug"},
 750         {Opt_oldalloc, "oldalloc"},
 751         {Opt_orlov, "orlov"},
 752         {Opt_user_xattr, "user_xattr"},
 753         {Opt_nouser_xattr, "nouser_xattr"},
 754         {Opt_acl, "acl"},
 755         {Opt_noacl, "noacl"},
 756         {Opt_reservation, "reservation"},
 757         {Opt_noreservation, "noreservation"},
 758         {Opt_noload, "noload"},
 759         {Opt_nobh, "nobh"},
 760         {Opt_bh, "bh"},
 761         {Opt_commit, "commit=%u"},
 762         {Opt_journal_update, "journal=update"},
 763         {Opt_journal_inum, "journal=%u"},
 764         {Opt_journal_dev, "journal_dev=%u"},
 765         {Opt_abort, "abort"},
 766         {Opt_data_journal, "data=journal"},
 767         {Opt_data_ordered, "data=ordered"},
 768         {Opt_data_writeback, "data=writeback"},
 769         {Opt_offusrjquota, "usrjquota="},
 770         {Opt_usrjquota, "usrjquota=%s"},
 771         {Opt_offgrpjquota, "grpjquota="},
 772         {Opt_grpjquota, "grpjquota=%s"},
 773         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 774         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 775         {Opt_grpquota, "grpquota"},
 776         {Opt_noquota, "noquota"},
 777         {Opt_quota, "quota"},
 778         {Opt_usrquota, "usrquota"},
 779         {Opt_barrier, "barrier=%u"},
 780         {Opt_extents, "extents"},
 781         {Opt_err, NULL},
 782         {Opt_resize, "resize"},
 783 };
 784
 785 static ext4_fsblk_t get_sb_block(void **data)
 786 {
 787         ext4_fsblk_t    sb_block;
 788         char            *options = (char *) *data;
 789
 790         if (!options || strncmp(options, "sb=", 3) != 0)
 791                 return 1;       /* Default location */
 792         options += 3;
 793         /*todo: use simple_strtoll with >32bit ext4 */
 794         sb_block = simple_strtoul(options, &options, 0);
 795         if (*options && *options != ',') {
 796                 printk("EXT4-fs: Invalid sb specification: %s\n",
 797                        (char *) *data);
 798                 return 1;
 799         }
 800         if (*options == ',')
 801                 options++;
 802         *data = (void *) options;
 803         return sb_block;
 804 }
 805
 806 static int parse_options (char *options, struct super_block *sb,
 807                           unsigned int *inum, unsigned long *journal_devnum,
 808                           ext4_fsblk_t *n_blocks_count, int is_remount)
 809 {
 810         struct ext4_sb_info *sbi = EXT4_SB(sb);
 811         char * p;
 812         substring_t args[MAX_OPT_ARGS];
 813         int data_opt = 0;
 814         int option;
 815 #ifdef CONFIG_QUOTA
 816         int qtype;
 817         char *qname;
 818 #endif
 819
 820         if (!options)
 821                 return 1;
 822
 823         while ((p = strsep (&options, ",")) != NULL) {
 824                 int token;
 825                 if (!*p)
 826                         continue;
 827
 828                 token = match_token(p, tokens, args);
 829                 switch (token) {
 830                 case Opt_bsd_df:
 831                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 832                         break;
 833                 case Opt_minix_df:
 834                         set_opt (sbi->s_mount_opt, MINIX_DF);
 835                         break;
 836                 case Opt_grpid:
 837                         set_opt (sbi->s_mount_opt, GRPID);
 838                         break;
 839                 case Opt_nogrpid:
 840                         clear_opt (sbi->s_mount_opt, GRPID);
 841                         break;
 842                 case Opt_resuid:
 843                         if (match_int(&args[0], &option))
 844                                 return 0;
 845                         sbi->s_resuid = option;
 846                         break;
 847                 case Opt_resgid:
 848                         if (match_int(&args[0], &option))
 849                                 return 0;
 850                         sbi->s_resgid = option;
 851                         break;
 852                 case Opt_sb:
 853                         /* handled by get_sb_block() instead of here */
 854                         /* *sb_block = match_int(&args[0]); */
 855                         break;
 856                 case Opt_err_panic:
 857                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 858                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 859                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 860                         break;
 861                 case Opt_err_ro:
 862                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 863                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 864                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 865                         break;
 866                 case Opt_err_cont:
 867                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 868                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 869                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 870                         break;
 871                 case Opt_nouid32:
 872                         set_opt (sbi->s_mount_opt, NO_UID32);
 873                         break;
 874                 case Opt_nocheck:
 875                         clear_opt (sbi->s_mount_opt, CHECK);
 876                         break;
 877                 case Opt_debug:
 878                         set_opt (sbi->s_mount_opt, DEBUG);
 879                         break;
 880                 case Opt_oldalloc:
 881                         set_opt (sbi->s_mount_opt, OLDALLOC);
 882                         break;
 883                 case Opt_orlov:
 884                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 885                         break;
 886 #ifdef CONFIG_EXT4DEV_FS_XATTR
 887                 case Opt_user_xattr:
 888                         set_opt (sbi->s_mount_opt, XATTR_USER);
 889                         break;
 890                 case Opt_nouser_xattr:
 891                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 892                         break;
 893 #else
 894                 case Opt_user_xattr:
 895                 case Opt_nouser_xattr:
 896                         printk("EXT4 (no)user_xattr options not supported\n");
 897                         break;
 898 #endif
 899 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 900                 case Opt_acl:
 901                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 902                         break;
 903                 case Opt_noacl:
 904                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 905                         break;
 906 #else
 907                 case Opt_acl:
 908                 case Opt_noacl:
 909                         printk("EXT4 (no)acl options not supported\n");
 910                         break;
 911 #endif
 912                 case Opt_reservation:
 913                         set_opt(sbi->s_mount_opt, RESERVATION);
 914                         break;
 915                 case Opt_noreservation:
 916                         clear_opt(sbi->s_mount_opt, RESERVATION);
 917                         break;
 918                 case Opt_journal_update:
 919                         /* @@@ FIXME */
 920                         /* Eventually we will want to be able to create
 921                            a journal file here.  For now, only allow the
 922                            user to specify an existing inode to be the
 923                            journal file. */
 924                         if (is_remount) {
 925                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 926                                        "journal on remount\n");
 927                                 return 0;
 928                         }
 929                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 930                         break;
 931                 case Opt_journal_inum:
 932                         if (is_remount) {
 933                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 934                                        "journal on remount\n");
 935                                 return 0;
 936                         }
 937                         if (match_int(&args[0], &option))
 938                                 return 0;
 939                         *inum = option;
 940                         break;
 941                 case Opt_journal_dev:
 942                         if (is_remount) {
 943                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 944                                        "journal on remount\n");
 945                                 return 0;
 946                         }
 947                         if (match_int(&args[0], &option))
 948                                 return 0;
 949                         *journal_devnum = option;
 950                         break;
 951                 case Opt_noload:
 952                         set_opt (sbi->s_mount_opt, NOLOAD);
 953                         break;
 954                 case Opt_commit:
 955                         if (match_int(&args[0], &option))
 956                                 return 0;
 957                         if (option < 0)
 958                                 return 0;
 959                         if (option == 0)
 960                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 961                         sbi->s_commit_interval = HZ * option;
 962                         break;
 963                 case Opt_data_journal:
 964                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
 965                         goto datacheck;
 966                 case Opt_data_ordered:
 967                         data_opt = EXT4_MOUNT_ORDERED_DATA;
 968                         goto datacheck;
 969                 case Opt_data_writeback:
 970                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
 971                 datacheck:
 972                         if (is_remount) {
 973                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
 974                                                 != data_opt) {
 975                                         printk(KERN_ERR
 976                                                 "EXT4-fs: cannot change data "
 977                                                 "mode on remount\n");
 978                                         return 0;
 979                                 }
 980                         } else {
 981                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
 982                                 sbi->s_mount_opt |= data_opt;
 983                         }
 984                         break;
 985 #ifdef CONFIG_QUOTA
 986                 case Opt_usrjquota:
 987                         qtype = USRQUOTA;
 988                         goto set_qf_name;
 989                 case Opt_grpjquota:
 990                         qtype = GRPQUOTA;
 991 set_qf_name:
 992                         if (sb_any_quota_enabled(sb)) {
 993                                 printk(KERN_ERR
 994                                         "EXT4-fs: Cannot change journalled "
 995                                         "quota options when quota turned on.\n");
 996                                 return 0;
 997                         }
 998                         qname = match_strdup(&args[0]);
 999                         if (!qname) {
1000                                 printk(KERN_ERR
1001                                         "EXT4-fs: not enough memory for "
1002                                         "storing quotafile name.\n");
1003                                 return 0;
1004                         }
1005                         if (sbi->s_qf_names[qtype] &&
1006                             strcmp(sbi->s_qf_names[qtype], qname)) {
1007                                 printk(KERN_ERR
1008                                         "EXT4-fs: %s quota file already "
1009                                         "specified.\n", QTYPE2NAME(qtype));
1010                                 kfree(qname);
1011                                 return 0;
1012                         }
1013                         sbi->s_qf_names[qtype] = qname;
1014                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1015                                 printk(KERN_ERR
1016                                         "EXT4-fs: quotafile must be on "
1017                                         "filesystem root.\n");
1018                                 kfree(sbi->s_qf_names[qtype]);
1019                                 sbi->s_qf_names[qtype] = NULL;
1020                                 return 0;
1021                         }
1022                         set_opt(sbi->s_mount_opt, QUOTA);
1023                         break;
1024                 case Opt_offusrjquota:
1025                         qtype = USRQUOTA;
1026                         goto clear_qf_name;
1027                 case Opt_offgrpjquota:
1028                         qtype = GRPQUOTA;
1029 clear_qf_name:
1030                         if (sb_any_quota_enabled(sb)) {
1031                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1032                                         "journalled quota options when "
1033                                         "quota turned on.\n");
1034                                 return 0;
1035                         }
1036                         /*
1037                          * The space will be released later when all options
1038                          * are confirmed to be correct
1039                          */
1040                         sbi->s_qf_names[qtype] = NULL;
1041                         break;
1042                 case Opt_jqfmt_vfsold:
1043                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1044                         break;
1045                 case Opt_jqfmt_vfsv0:
1046                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1047                         break;
1048                 case Opt_quota:
1049                 case Opt_usrquota:
1050                         set_opt(sbi->s_mount_opt, QUOTA);
1051                         set_opt(sbi->s_mount_opt, USRQUOTA);
1052                         break;
1053                 case Opt_grpquota:
1054                         set_opt(sbi->s_mount_opt, QUOTA);
1055                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1056                         break;
1057                 case Opt_noquota:
1058                         if (sb_any_quota_enabled(sb)) {
1059                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1060                                         "options when quota turned on.\n");
1061                                 return 0;
1062                         }
1063                         clear_opt(sbi->s_mount_opt, QUOTA);
1064                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1065                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1066                         break;
1067 #else
1068                 case Opt_quota:
1069                 case Opt_usrquota:
1070                 case Opt_grpquota:
1071                 case Opt_usrjquota:
1072                 case Opt_grpjquota:
1073                 case Opt_offusrjquota:
1074                 case Opt_offgrpjquota:
1075                 case Opt_jqfmt_vfsold:
1076                 case Opt_jqfmt_vfsv0:
1077                         printk(KERN_ERR
1078                                 "EXT4-fs: journalled quota options not "
1079                                 "supported.\n");
1080                         break;
1081                 case Opt_noquota:
1082                         break;
1083 #endif
1084                 case Opt_abort:
1085                         set_opt(sbi->s_mount_opt, ABORT);
1086                         break;
1087                 case Opt_barrier:
1088                         if (match_int(&args[0], &option))
1089                                 return 0;
1090                         if (option)
1091                                 set_opt(sbi->s_mount_opt, BARRIER);
1092                         else
1093                                 clear_opt(sbi->s_mount_opt, BARRIER);
1094                         break;
1095                 case Opt_ignore:
1096                         break;
1097                 case Opt_resize:
1098                         if (!is_remount) {
1099                                 printk("EXT4-fs: resize option only available "
1100                                         "for remount\n");
1101                                 return 0;
1102                         }
1103                         if (match_int(&args[0], &option) != 0)
1104                                 return 0;
1105                         *n_blocks_count = option;
1106                         break;
1107                 case Opt_nobh:
1108                         set_opt(sbi->s_mount_opt, NOBH);
1109                         break;
1110                 case Opt_bh:
1111                         clear_opt(sbi->s_mount_opt, NOBH);
1112                         break;
1113                 case Opt_extents:
1114                         set_opt (sbi->s_mount_opt, EXTENTS);
1115                         break;
1116                 default:
1117                         printk (KERN_ERR
1118                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1119                                 "or missing value\n", p);
1120                         return 0;
1121                 }
1122         }
1123 #ifdef CONFIG_QUOTA
1124         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1125                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1126                      sbi->s_qf_names[USRQUOTA])
1127                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1128
1129                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1130                      sbi->s_qf_names[GRPQUOTA])
1131                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1132
1133                 if ((sbi->s_qf_names[USRQUOTA] &&
1134                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1135                     (sbi->s_qf_names[GRPQUOTA] &&
1136                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1137                         printk(KERN_ERR "EXT4-fs: old and new quota "
1138                                         "format mixing.\n");
1139                         return 0;
1140                 }
1141
1142                 if (!sbi->s_jquota_fmt) {
1143                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1144                                         "not specified.\n");
1145                         return 0;
1146                 }
1147         } else {
1148                 if (sbi->s_jquota_fmt) {
1149                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1150                                         "specified with no journalling "
1151                                         "enabled.\n");
1152                         return 0;
1153                 }
1154         }
1155 #endif
1156         return 1;
1157 }
1158
1159 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1160                             int read_only)
1161 {
1162         struct ext4_sb_info *sbi = EXT4_SB(sb);
1163         int res = 0;
1164
1165         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1166                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1167                         "forcing read-only mode\n");
1168                 res = MS_RDONLY;
1169         }
1170         if (read_only)
1171                 return res;
1172         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1173                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1174                         "running e2fsck is recommended\n");
1175         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1176                 printk (KERN_WARNING
1177                         "EXT4-fs warning: mounting fs with errors, "
1178                         "running e2fsck is recommended\n");
1179         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1180                  le16_to_cpu(es->s_mnt_count) >=
1181                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1182                 printk (KERN_WARNING
1183                         "EXT4-fs warning: maximal mount count reached, "
1184                         "running e2fsck is recommended\n");
1185         else if (le32_to_cpu(es->s_checkinterval) &&
1186                 (le32_to_cpu(es->s_lastcheck) +
1187                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1188                 printk (KERN_WARNING
1189                         "EXT4-fs warning: checktime reached, "
1190                         "running e2fsck is recommended\n");
1191 #if 0
1192                 /* @@@ We _will_ want to clear the valid bit if we find
1193                  * inconsistencies, to force a fsck at reboot.  But for
1194                  * a plain journaled filesystem we can keep it set as
1195                  * valid forever! :)
1196                  */
1197         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1198 #endif
1199         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1200                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1201         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1202         es->s_mtime = cpu_to_le32(get_seconds());
1203         ext4_update_dynamic_rev(sb);
1204         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1205
1206         ext4_commit_super(sb, es, 1);
1207         if (test_opt(sb, DEBUG))
1208                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1209                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1210                         sb->s_blocksize,
1211                         sbi->s_groups_count,
1212                         EXT4_BLOCKS_PER_GROUP(sb),
1213                         EXT4_INODES_PER_GROUP(sb),
1214                         sbi->s_mount_opt);
1215
1216         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1217         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1218                 char b[BDEVNAME_SIZE];
1219
1220                 printk("external journal on %s\n",
1221                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1222         } else {
1223                 printk("internal journal\n");
1224         }
1225         return res;
1226 }
1227
1228 /* Called at mount-time, super-block is locked */
1229 static int ext4_check_descriptors (struct super_block * sb)
1230 {
1231         struct ext4_sb_info *sbi = EXT4_SB(sb);
1232         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1233         ext4_fsblk_t last_block;
1234         ext4_fsblk_t block_bitmap;
1235         ext4_fsblk_t inode_bitmap;
1236         ext4_fsblk_t inode_table;
1237         struct ext4_group_desc * gdp = NULL;
1238         int desc_block = 0;
1239         int i;
1240
1241         ext4_debug ("Checking group descriptors");
1242
1243         for (i = 0; i < sbi->s_groups_count; i++)
1244         {
1245                 if (i == sbi->s_groups_count - 1)
1246                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1247                 else
1248                         last_block = first_block +
1249                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1250
1251                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1252                         gdp = (struct ext4_group_desc *)
1253                                         sbi->s_group_desc[desc_block++]->b_data;
1254                 block_bitmap = ext4_block_bitmap(sb, gdp);
1255                 if (block_bitmap < first_block || block_bitmap > last_block)
1256                 {
1257                         ext4_error (sb, "ext4_check_descriptors",
1258                                     "Block bitmap for group %d"
1259                                     " not in group (block %llu)!",
1260                                     i, block_bitmap);
1261                         return 0;
1262                 }
1263                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1264                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1265                 {
1266                         ext4_error (sb, "ext4_check_descriptors",
1267                                     "Inode bitmap for group %d"
1268                                     " not in group (block %llu)!",
1269                                     i, inode_bitmap);
1270                         return 0;
1271                 }
1272                 inode_table = ext4_inode_table(sb, gdp);
1273                 if (inode_table < first_block ||
1274                     inode_table + sbi->s_itb_per_group > last_block)
1275                 {
1276                         ext4_error (sb, "ext4_check_descriptors",
1277                                     "Inode table for group %d"
1278                                     " not in group (block %llu)!",
1279                                     i, inode_table);
1280                         return 0;
1281                 }
1282                 first_block += EXT4_BLOCKS_PER_GROUP(sb);
1283                 gdp = (struct ext4_group_desc *)
1284                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1285         }
1286
1287         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1288         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1289         return 1;
1290 }
1291
1292
1293 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1294  * the superblock) which were deleted from all directories, but held open by
1295  * a process at the time of a crash.  We walk the list and try to delete these
1296  * inodes at recovery time (only with a read-write filesystem).
1297  *
1298  * In order to keep the orphan inode chain consistent during traversal (in
1299  * case of crash during recovery), we link each inode into the superblock
1300  * orphan list_head and handle it the same way as an inode deletion during
1301  * normal operation (which journals the operations for us).
1302  *
1303  * We only do an iget() and an iput() on each inode, which is very safe if we
1304  * accidentally point at an in-use or already deleted inode.  The worst that
1305  * can happen in this case is that we get a "bit already cleared" message from
1306  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1307  * e2fsck was run on this filesystem, and it must have already done the orphan
1308  * inode cleanup for us, so we can safely abort without any further action.
1309  */
1310 static void ext4_orphan_cleanup (struct super_block * sb,
1311                                  struct ext4_super_block * es)
1312 {
1313         unsigned int s_flags = sb->s_flags;
1314         int nr_orphans = 0, nr_truncates = 0;
1315 #ifdef CONFIG_QUOTA
1316         int i;
1317 #endif
1318         if (!es->s_last_orphan) {
1319                 jbd_debug(4, "no orphan inodes to clean up\n");
1320                 return;
1321         }
1322
1323         if (bdev_read_only(sb->s_bdev)) {
1324                 printk(KERN_ERR "EXT4-fs: write access "
1325                         "unavailable, skipping orphan cleanup.\n");
1326                 return;
1327         }
1328
1329         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1330                 if (es->s_last_orphan)
1331                         jbd_debug(1, "Errors on filesystem, "
1332                                   "clearing orphan list.\n");
1333                 es->s_last_orphan = 0;
1334                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1335                 return;
1336         }
1337
1338         if (s_flags & MS_RDONLY) {
1339                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1340                        sb->s_id);
1341                 sb->s_flags &= ~MS_RDONLY;
1342         }
1343 #ifdef CONFIG_QUOTA
1344         /* Needed for iput() to work correctly and not trash data */
1345         sb->s_flags |= MS_ACTIVE;
1346         /* Turn on quotas so that they are updated correctly */
1347         for (i = 0; i < MAXQUOTAS; i++) {
1348                 if (EXT4_SB(sb)->s_qf_names[i]) {
1349                         int ret = ext4_quota_on_mount(sb, i);
1350                         if (ret < 0)
1351                                 printk(KERN_ERR
1352                                         "EXT4-fs: Cannot turn on journalled "
1353                                         "quota: error %d\n", ret);
1354                 }
1355         }
1356 #endif
1357
1358         while (es->s_last_orphan) {
1359                 struct inode *inode;
1360
1361                 if (!(inode =
1362                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1363                         es->s_last_orphan = 0;
1364                         break;
1365                 }
1366
1367                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1368                 DQUOT_INIT(inode);
1369                 if (inode->i_nlink) {
1370                         printk(KERN_DEBUG
1371                                 "%s: truncating inode %lu to %Ld bytes\n",
1372                                 __FUNCTION__, inode->i_ino, inode->i_size);
1373                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1374                                   inode->i_ino, inode->i_size);
1375                         ext4_truncate(inode);
1376                         nr_truncates++;
1377                 } else {
1378                         printk(KERN_DEBUG
1379                                 "%s: deleting unreferenced inode %lu\n",
1380                                 __FUNCTION__, inode->i_ino);
1381                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1382                                   inode->i_ino);
1383                         nr_orphans++;
1384                 }
1385                 iput(inode);  /* The delete magic happens here! */
1386         }
1387
1388 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1389
1390         if (nr_orphans)
1391                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1392                        sb->s_id, PLURAL(nr_orphans));
1393         if (nr_truncates)
1394                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1395                        sb->s_id, PLURAL(nr_truncates));
1396 #ifdef CONFIG_QUOTA
1397         /* Turn quotas off */
1398         for (i = 0; i < MAXQUOTAS; i++) {
1399                 if (sb_dqopt(sb)->files[i])
1400                         vfs_quota_off(sb, i);
1401         }
1402 #endif
1403         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1404 }
1405
1406 #define log2(n) ffz(~(n))
1407
1408 /*
1409  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1410  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1411  * We need to be 1 filesystem block less than the 2^32 sector limit.
1412  */
1413 static loff_t ext4_max_size(int bits)
1414 {
1415         loff_t res = EXT4_NDIR_BLOCKS;
1416         /* This constant is calculated to be the largest file size for a
1417          * dense, 4k-blocksize file such that the total number of
1418          * sectors in the file, including data and all indirect blocks,
1419          * does not exceed 2^32. */
1420         const loff_t upper_limit = 0x1ff7fffd000LL;
1421
1422         res += 1LL << (bits-2);
1423         res += 1LL << (2*(bits-2));
1424         res += 1LL << (3*(bits-2));
1425         res <<= bits;
1426         if (res > upper_limit)
1427                 res = upper_limit;
1428         return res;
1429 }
1430
1431 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1432                                 ext4_fsblk_t logical_sb_block, int nr)
1433 {
1434         struct ext4_sb_info *sbi = EXT4_SB(sb);
1435         unsigned long bg, first_meta_bg;
1436         int has_super = 0;
1437
1438         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1439
1440         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1441             nr < first_meta_bg)
1442                 return logical_sb_block + nr + 1;
1443         bg = sbi->s_desc_per_block * nr;
1444         if (ext4_bg_has_super(sb, bg))
1445                 has_super = 1;
1446         return (has_super + ext4_group_first_block_no(sb, bg));
1447 }
1448
1449
1450 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1451 {
1452         struct buffer_head * bh;
1453         struct ext4_super_block *es = NULL;
1454         struct ext4_sb_info *sbi;
1455         ext4_fsblk_t block;
1456         ext4_fsblk_t sb_block = get_sb_block(&data);
1457         ext4_fsblk_t logical_sb_block;
1458         unsigned long offset = 0;
1459         unsigned int journal_inum = 0;
1460         unsigned long journal_devnum = 0;
1461         unsigned long def_mount_opts;
1462         struct inode *root;
1463         int blocksize;
1464         int hblock;
1465         int db_count;
1466         int i;
1467         int needs_recovery;
1468         __le32 features;
1469         __u64 blocks_count;
1470
1471         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1472         if (!sbi)
1473                 return -ENOMEM;
1474         sb->s_fs_info = sbi;
1475         sbi->s_mount_opt = 0;
1476         sbi->s_resuid = EXT4_DEF_RESUID;
1477         sbi->s_resgid = EXT4_DEF_RESGID;
1478
1479         unlock_kernel();
1480
1481         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1482         if (!blocksize) {
1483                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1484                 goto out_fail;
1485         }
1486
1487         /*
1488          * The ext4 superblock will not be buffer aligned for other than 1kB
1489          * block sizes.  We need to calculate the offset from buffer start.
1490          */
1491         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1492                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1493                 offset = do_div(logical_sb_block, blocksize);
1494         } else {
1495                 logical_sb_block = sb_block;
1496         }
1497
1498         if (!(bh = sb_bread(sb, logical_sb_block))) {
1499                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1500                 goto out_fail;
1501         }
1502         /*
1503          * Note: s_es must be initialized as soon as possible because
1504          *       some ext4 macro-instructions depend on its value
1505          */
1506         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1507         sbi->s_es = es;
1508         sb->s_magic = le16_to_cpu(es->s_magic);
1509         if (sb->s_magic != EXT4_SUPER_MAGIC)
1510                 goto cantfind_ext4;
1511
1512         /* Set defaults before we parse the mount options */
1513         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1514         if (def_mount_opts & EXT4_DEFM_DEBUG)
1515                 set_opt(sbi->s_mount_opt, DEBUG);
1516         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1517                 set_opt(sbi->s_mount_opt, GRPID);
1518         if (def_mount_opts & EXT4_DEFM_UID16)
1519                 set_opt(sbi->s_mount_opt, NO_UID32);
1520 #ifdef CONFIG_EXT4DEV_FS_XATTR
1521         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1522                 set_opt(sbi->s_mount_opt, XATTR_USER);
1523 #endif
1524 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1525         if (def_mount_opts & EXT4_DEFM_ACL)
1526                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1527 #endif
1528         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1529                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1530         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1531                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1532         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1533                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1534
1535         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1536                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1537         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1538                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1539         else
1540                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1541
1542         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1543         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1544
1545         set_opt(sbi->s_mount_opt, RESERVATION);
1546
1547         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1548                             NULL, 0))
1549                 goto failed_mount;
1550
1551         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1552                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1553
1554         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1555             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1556              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1557              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1558                 printk(KERN_WARNING
1559                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1560                        "running e2fsck is recommended\n");
1561         /*
1562          * Check feature flags regardless of the revision level, since we
1563          * previously didn't change the revision level when setting the flags,
1564          * so there is a chance incompat flags are set on a rev 0 filesystem.
1565          */
1566         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1567         if (features) {
1568                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1569                        "unsupported optional features (%x).\n",
1570                        sb->s_id, le32_to_cpu(features));
1571                 goto failed_mount;
1572         }
1573         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1574         if (!(sb->s_flags & MS_RDONLY) && features) {
1575                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1576                        "unsupported optional features (%x).\n",
1577                        sb->s_id, le32_to_cpu(features));
1578                 goto failed_mount;
1579         }
1580         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1581
1582         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1583             blocksize > EXT4_MAX_BLOCK_SIZE) {
1584                 printk(KERN_ERR
1585                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1586                        blocksize, sb->s_id);
1587                 goto failed_mount;
1588         }
1589
1590         hblock = bdev_hardsect_size(sb->s_bdev);
1591         if (sb->s_blocksize != blocksize) {
1592                 /*
1593                  * Make sure the blocksize for the filesystem is larger
1594                  * than the hardware sectorsize for the machine.
1595                  */
1596                 if (blocksize < hblock) {
1597                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1598                                "device blocksize %d.\n", blocksize, hblock);
1599                         goto failed_mount;
1600                 }
1601
1602                 brelse (bh);
1603                 sb_set_blocksize(sb, blocksize);
1604                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1605                 offset = do_div(logical_sb_block, blocksize);
1606                 bh = sb_bread(sb, logical_sb_block);
1607                 if (!bh) {
1608                         printk(KERN_ERR
1609                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1610                         goto failed_mount;
1611                 }
1612                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1613                 sbi->s_es = es;
1614                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1615                         printk (KERN_ERR
1616                                 "EXT4-fs: Magic mismatch, very weird !\n");
1617                         goto failed_mount;
1618                 }
1619         }
1620
1621         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1622
1623         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1624                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1625                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1626         } else {
1627                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1628                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1629                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1630                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1631                     (sbi->s_inode_size > blocksize)) {
1632                         printk (KERN_ERR
1633                                 "EXT4-fs: unsupported inode size: %d\n",
1634                                 sbi->s_inode_size);
1635                         goto failed_mount;
1636                 }
1637         }
1638         sbi->s_frag_size = EXT4_MIN_FRAG_SIZE <<
1639                                    le32_to_cpu(es->s_log_frag_size);
1640         if (blocksize != sbi->s_frag_size) {
1641                 printk(KERN_ERR
1642                        "EXT4-fs: fragsize %lu != blocksize %u (unsupported)\n",
1643                        sbi->s_frag_size, blocksize);
1644                 goto failed_mount;
1645         }
1646         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1647         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1648                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1649                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1650                     sbi->s_desc_size & (sbi->s_desc_size - 1)) {
1651                         printk(KERN_ERR
1652                                "EXT4-fs: unsupported descriptor size %lu\n",
1653                                sbi->s_desc_size);
1654                         goto failed_mount;
1655                 }
1656         } else
1657                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1658         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1659         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1660         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1661         if (EXT4_INODE_SIZE(sb) == 0)
1662                 goto cantfind_ext4;
1663         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1664         if (sbi->s_inodes_per_block == 0)
1665                 goto cantfind_ext4;
1666         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1667                                         sbi->s_inodes_per_block;
1668         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1669         sbi->s_sbh = bh;
1670         sbi->s_mount_state = le16_to_cpu(es->s_state);
1671         sbi->s_addr_per_block_bits = log2(EXT4_ADDR_PER_BLOCK(sb));
1672         sbi->s_desc_per_block_bits = log2(EXT4_DESC_PER_BLOCK(sb));
1673         for (i=0; i < 4; i++)
1674                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1675         sbi->s_def_hash_version = es->s_def_hash_version;
1676
1677         if (sbi->s_blocks_per_group > blocksize * 8) {
1678                 printk (KERN_ERR
1679                         "EXT4-fs: #blocks per group too big: %lu\n",
1680                         sbi->s_blocks_per_group);
1681                 goto failed_mount;
1682         }
1683         if (sbi->s_frags_per_group > blocksize * 8) {
1684                 printk (KERN_ERR
1685                         "EXT4-fs: #fragments per group too big: %lu\n",
1686                         sbi->s_frags_per_group);
1687                 goto failed_mount;
1688         }
1689         if (sbi->s_inodes_per_group > blocksize * 8) {
1690                 printk (KERN_ERR
1691                         "EXT4-fs: #inodes per group too big: %lu\n",
1692                         sbi->s_inodes_per_group);
1693                 goto failed_mount;
1694         }
1695
1696         if (ext4_blocks_count(es) >
1697                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1698                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1699                         " too large to mount safely\n", sb->s_id);
1700                 if (sizeof(sector_t) < 8)
1701                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1702                                         "enabled\n");
1703                 goto failed_mount;
1704         }
1705
1706         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1707                 goto cantfind_ext4;
1708         blocks_count = (ext4_blocks_count(es) -
1709                         le32_to_cpu(es->s_first_data_block) +
1710                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1711         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1712         sbi->s_groups_count = blocks_count;
1713         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1714                    EXT4_DESC_PER_BLOCK(sb);
1715         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1716                                     GFP_KERNEL);
1717         if (sbi->s_group_desc == NULL) {
1718                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1719                 goto failed_mount;
1720         }
1721
1722         bgl_lock_init(&sbi->s_blockgroup_lock);
1723
1724         for (i = 0; i < db_count; i++) {
1725                 block = descriptor_loc(sb, logical_sb_block, i);
1726                 sbi->s_group_desc[i] = sb_bread(sb, block);
1727                 if (!sbi->s_group_desc[i]) {
1728                         printk (KERN_ERR "EXT4-fs: "
1729                                 "can't read group descriptor %d\n", i);
1730                         db_count = i;
1731                         goto failed_mount2;
1732                 }
1733         }
1734         if (!ext4_check_descriptors (sb)) {
1735                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1736                 goto failed_mount2;
1737         }
1738         sbi->s_gdb_count = db_count;
1739         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1740         spin_lock_init(&sbi->s_next_gen_lock);
1741
1742         percpu_counter_init(&sbi->s_freeblocks_counter,
1743                 ext4_count_free_blocks(sb));
1744         percpu_counter_init(&sbi->s_freeinodes_counter,
1745                 ext4_count_free_inodes(sb));
1746         percpu_counter_init(&sbi->s_dirs_counter,
1747                 ext4_count_dirs(sb));
1748
1749         /* per fileystem reservation list head & lock */
1750         spin_lock_init(&sbi->s_rsv_window_lock);
1751         sbi->s_rsv_window_root = RB_ROOT;
1752         /* Add a single, static dummy reservation to the start of the
1753          * reservation window list --- it gives us a placeholder for
1754          * append-at-start-of-list which makes the allocation logic
1755          * _much_ simpler. */
1756         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1757         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1758         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1759         sbi->s_rsv_window_head.rsv_goal_size = 0;
1760         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1761
1762         /*
1763          * set up enough so that it can read an inode
1764          */
1765         sb->s_op = &ext4_sops;
1766         sb->s_export_op = &ext4_export_ops;
1767         sb->s_xattr = ext4_xattr_handlers;
1768 #ifdef CONFIG_QUOTA
1769         sb->s_qcop = &ext4_qctl_operations;
1770         sb->dq_op = &ext4_quota_operations;
1771 #endif
1772         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1773
1774         sb->s_root = NULL;
1775
1776         needs_recovery = (es->s_last_orphan != 0 ||
1777                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1778                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1779
1780         /*
1781          * The first inode we look at is the journal inode.  Don't try
1782          * root first: it may be modified in the journal!
1783          */
1784         if (!test_opt(sb, NOLOAD) &&
1785             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1786                 if (ext4_load_journal(sb, es, journal_devnum))
1787                         goto failed_mount3;
1788         } else if (journal_inum) {
1789                 if (ext4_create_journal(sb, es, journal_inum))
1790                         goto failed_mount3;
1791         } else {
1792                 if (!silent)
1793                         printk (KERN_ERR
1794                                 "ext4: No journal on filesystem on %s\n",
1795                                 sb->s_id);
1796                 goto failed_mount3;
1797         }
1798
1799         /* We have now updated the journal if required, so we can
1800          * validate the data journaling mode. */
1801         switch (test_opt(sb, DATA_FLAGS)) {
1802         case 0:
1803                 /* No mode set, assume a default based on the journal
1804                  * capabilities: ORDERED_DATA if the journal can
1805                  * cope, else JOURNAL_DATA
1806                  */
1807                 if (jbd2_journal_check_available_features
1808                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
1809                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1810                 else
1811                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1812                 break;
1813
1814         case EXT4_MOUNT_ORDERED_DATA:
1815         case EXT4_MOUNT_WRITEBACK_DATA:
1816                 if (!jbd2_journal_check_available_features
1817                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
1818                         printk(KERN_ERR "EXT4-fs: Journal does not support "
1819                                "requested data journaling mode\n");
1820                         goto failed_mount4;
1821                 }
1822         default:
1823                 break;
1824         }
1825
1826         if (test_opt(sb, NOBH)) {
1827                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
1828                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
1829                                 "its supported only with writeback mode\n");
1830                         clear_opt(sbi->s_mount_opt, NOBH);
1831                 }
1832         }
1833         /*
1834          * The jbd2_journal_load will have done any necessary log recovery,
1835          * so we can safely mount the rest of the filesystem now.
1836          */
1837
1838         root = iget(sb, EXT4_ROOT_INO);
1839         sb->s_root = d_alloc_root(root);
1840         if (!sb->s_root) {
1841                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
1842                 iput(root);
1843                 goto failed_mount4;
1844         }
1845         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1846                 dput(sb->s_root);
1847                 sb->s_root = NULL;
1848                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
1849                 goto failed_mount4;
1850         }
1851
1852         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1853         /*
1854          * akpm: core read_super() calls in here with the superblock locked.
1855          * That deadlocks, because orphan cleanup needs to lock the superblock
1856          * in numerous places.  Here we just pop the lock - it's relatively
1857          * harmless, because we are now ready to accept write_super() requests,
1858          * and aviro says that's the only reason for hanging onto the
1859          * superblock lock.
1860          */
1861         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
1862         ext4_orphan_cleanup(sb, es);
1863         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
1864         if (needs_recovery)
1865                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
1866         ext4_mark_recovery_complete(sb, es);
1867         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
1868                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
1869                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
1870                 "writeback");
1871
1872         ext4_ext_init(sb);
1873
1874         lock_kernel();
1875         return 0;
1876
1877 cantfind_ext4:
1878         if (!silent)
1879                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
1880                        sb->s_id);
1881         goto failed_mount;
1882
1883 failed_mount4:
1884         jbd2_journal_destroy(sbi->s_journal);
1885 failed_mount3:
1886         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1887         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1888         percpu_counter_destroy(&sbi->s_dirs_counter);
1889 failed_mount2:
1890         for (i = 0; i < db_count; i++)
1891                 brelse(sbi->s_group_desc[i]);
1892         kfree(sbi->s_group_desc);
1893 failed_mount:
1894 #ifdef CONFIG_QUOTA
1895         for (i = 0; i < MAXQUOTAS; i++)
1896                 kfree(sbi->s_qf_names[i]);
1897 #endif
1898         ext4_blkdev_remove(sbi);
1899         brelse(bh);
1900 out_fail:
1901         sb->s_fs_info = NULL;
1902         kfree(sbi);
1903         lock_kernel();
1904         return -EINVAL;
1905 }
1906
1907 /*
1908  * Setup any per-fs journal parameters now.  We'll do this both on
1909  * initial mount, once the journal has been initialised but before we've
1910  * done any recovery; and again on any subsequent remount.
1911  */
1912 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
1913 {
1914         struct ext4_sb_info *sbi = EXT4_SB(sb);
1915
1916         if (sbi->s_commit_interval)
1917                 journal->j_commit_interval = sbi->s_commit_interval;
1918         /* We could also set up an ext4-specific default for the commit
1919          * interval here, but for now we'll just fall back to the jbd
1920          * default. */
1921
1922         spin_lock(&journal->j_state_lock);
1923         if (test_opt(sb, BARRIER))
1924                 journal->j_flags |= JBD2_BARRIER;
1925         else
1926                 journal->j_flags &= ~JBD2_BARRIER;
1927         spin_unlock(&journal->j_state_lock);
1928 }
1929
1930 static journal_t *ext4_get_journal(struct super_block *sb,
1931                                    unsigned int journal_inum)
1932 {
1933         struct inode *journal_inode;
1934         journal_t *journal;
1935
1936         /* First, test for the existence of a valid inode on disk.  Bad
1937          * things happen if we iget() an unused inode, as the subsequent
1938          * iput() will try to delete it. */
1939
1940         journal_inode = iget(sb, journal_inum);
1941         if (!journal_inode) {
1942                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
1943                 return NULL;
1944         }
1945         if (!journal_inode->i_nlink) {
1946                 make_bad_inode(journal_inode);
1947                 iput(journal_inode);
1948                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
1949                 return NULL;
1950         }
1951
1952         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1953                   journal_inode, journal_inode->i_size);
1954         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1955                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
1956                 iput(journal_inode);
1957                 return NULL;
1958         }
1959
1960         journal = jbd2_journal_init_inode(journal_inode);
1961         if (!journal) {
1962                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
1963                 iput(journal_inode);
1964                 return NULL;
1965         }
1966         journal->j_private = sb;
1967         ext4_init_journal_params(sb, journal);
1968         return journal;
1969 }
1970
1971 static journal_t *ext4_get_dev_journal(struct super_block *sb,
1972                                        dev_t j_dev)
1973 {
1974         struct buffer_head * bh;
1975         journal_t *journal;
1976         ext4_fsblk_t start;
1977         ext4_fsblk_t len;
1978         int hblock, blocksize;
1979         ext4_fsblk_t sb_block;
1980         unsigned long offset;
1981         struct ext4_super_block * es;
1982         struct block_device *bdev;
1983
1984         bdev = ext4_blkdev_get(j_dev);
1985         if (bdev == NULL)
1986                 return NULL;
1987
1988         if (bd_claim(bdev, sb)) {
1989                 printk(KERN_ERR
1990                         "EXT4: failed to claim external journal device.\n");
1991                 blkdev_put(bdev);
1992                 return NULL;
1993         }
1994
1995         blocksize = sb->s_blocksize;
1996         hblock = bdev_hardsect_size(bdev);
1997         if (blocksize < hblock) {
1998                 printk(KERN_ERR
1999                         "EXT4-fs: blocksize too small for journal device.\n");
2000                 goto out_bdev;
2001         }
2002
2003         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2004         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2005         set_blocksize(bdev, blocksize);
2006         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2007                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2008                        "external journal\n");
2009                 goto out_bdev;
2010         }
2011
2012         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2013         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2014             !(le32_to_cpu(es->s_feature_incompat) &
2015               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2016                 printk(KERN_ERR "EXT4-fs: external journal has "
2017                                         "bad superblock\n");
2018                 brelse(bh);
2019                 goto out_bdev;
2020         }
2021
2022         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2023                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2024                 brelse(bh);
2025                 goto out_bdev;
2026         }
2027
2028         len = ext4_blocks_count(es);
2029         start = sb_block + 1;
2030         brelse(bh);     /* we're done with the superblock */
2031
2032         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2033                                         start, len, blocksize);
2034         if (!journal) {
2035                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2036                 goto out_bdev;
2037         }
2038         journal->j_private = sb;
2039         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2040         wait_on_buffer(journal->j_sb_buffer);
2041         if (!buffer_uptodate(journal->j_sb_buffer)) {
2042                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2043                 goto out_journal;
2044         }
2045         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2046                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2047                                         "user (unsupported) - %d\n",
2048                         be32_to_cpu(journal->j_superblock->s_nr_users));
2049                 goto out_journal;
2050         }
2051         EXT4_SB(sb)->journal_bdev = bdev;
2052         ext4_init_journal_params(sb, journal);
2053         return journal;
2054 out_journal:
2055         jbd2_journal_destroy(journal);
2056 out_bdev:
2057         ext4_blkdev_put(bdev);
2058         return NULL;
2059 }
2060
2061 static int ext4_load_journal(struct super_block *sb,
2062                              struct ext4_super_block *es,
2063                              unsigned long journal_devnum)
2064 {
2065         journal_t *journal;
2066         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2067         dev_t journal_dev;
2068         int err = 0;
2069         int really_read_only;
2070
2071         if (journal_devnum &&
2072             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2073                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2074                         "numbers have changed\n");
2075                 journal_dev = new_decode_dev(journal_devnum);
2076         } else
2077                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2078
2079         really_read_only = bdev_read_only(sb->s_bdev);
2080
2081         /*
2082          * Are we loading a blank journal or performing recovery after a
2083          * crash?  For recovery, we need to check in advance whether we
2084          * can get read-write access to the device.
2085          */
2086
2087         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2088                 if (sb->s_flags & MS_RDONLY) {
2089                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2090                                         "required on readonly filesystem.\n");
2091                         if (really_read_only) {
2092                                 printk(KERN_ERR "EXT4-fs: write access "
2093                                         "unavailable, cannot proceed.\n");
2094                                 return -EROFS;
2095                         }
2096                         printk (KERN_INFO "EXT4-fs: write access will "
2097                                         "be enabled during recovery.\n");
2098                 }
2099         }
2100
2101         if (journal_inum && journal_dev) {
2102                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2103                        "and inode journals!\n");
2104                 return -EINVAL;
2105         }
2106
2107         if (journal_inum) {
2108                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2109                         return -EINVAL;
2110         } else {
2111                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2112                         return -EINVAL;
2113         }
2114
2115         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2116                 err = jbd2_journal_update_format(journal);
2117                 if (err)  {
2118                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2119                         jbd2_journal_destroy(journal);
2120                         return err;
2121                 }
2122         }
2123
2124         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2125                 err = jbd2_journal_wipe(journal, !really_read_only);
2126         if (!err)
2127                 err = jbd2_journal_load(journal);
2128
2129         if (err) {
2130                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2131                 jbd2_journal_destroy(journal);
2132                 return err;
2133         }
2134
2135         EXT4_SB(sb)->s_journal = journal;
2136         ext4_clear_journal_err(sb, es);
2137
2138         if (journal_devnum &&
2139             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2140                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2141                 sb->s_dirt = 1;
2142
2143                 /* Make sure we flush the recovery flag to disk. */
2144                 ext4_commit_super(sb, es, 1);
2145         }
2146
2147         return 0;
2148 }
2149
2150 static int ext4_create_journal(struct super_block * sb,
2151                                struct ext4_super_block * es,
2152                                unsigned int journal_inum)
2153 {
2154         journal_t *journal;
2155
2156         if (sb->s_flags & MS_RDONLY) {
2157                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2158                                 "create journal.\n");
2159                 return -EROFS;
2160         }
2161
2162         if (!(journal = ext4_get_journal(sb, journal_inum)))
2163                 return -EINVAL;
2164
2165         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2166                journal_inum);
2167
2168         if (jbd2_journal_create(journal)) {
2169                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2170                 jbd2_journal_destroy(journal);
2171                 return -EIO;
2172         }
2173
2174         EXT4_SB(sb)->s_journal = journal;
2175
2176         ext4_update_dynamic_rev(sb);
2177         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2178         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2179
2180         es->s_journal_inum = cpu_to_le32(journal_inum);
2181         sb->s_dirt = 1;
2182
2183         /* Make sure we flush the recovery flag to disk. */
2184         ext4_commit_super(sb, es, 1);
2185
2186         return 0;
2187 }
2188
2189 static void ext4_commit_super (struct super_block * sb,
2190                                struct ext4_super_block * es,
2191                                int sync)
2192 {
2193         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2194
2195         if (!sbh)
2196                 return;
2197         es->s_wtime = cpu_to_le32(get_seconds());
2198         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2199         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2200         BUFFER_TRACE(sbh, "marking dirty");
2201         mark_buffer_dirty(sbh);
2202         if (sync)
2203                 sync_dirty_buffer(sbh);
2204 }
2205
2206
2207 /*
2208  * Have we just finished recovery?  If so, and if we are mounting (or
2209  * remounting) the filesystem readonly, then we will end up with a
2210  * consistent fs on disk.  Record that fact.
2211  */
2212 static void ext4_mark_recovery_complete(struct super_block * sb,
2213                                         struct ext4_super_block * es)
2214 {
2215         journal_t *journal = EXT4_SB(sb)->s_journal;
2216
2217         jbd2_journal_lock_updates(journal);
2218         jbd2_journal_flush(journal);
2219         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2220             sb->s_flags & MS_RDONLY) {
2221                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2222                 sb->s_dirt = 0;
2223                 ext4_commit_super(sb, es, 1);
2224         }
2225         jbd2_journal_unlock_updates(journal);
2226 }
2227
2228 /*
2229  * If we are mounting (or read-write remounting) a filesystem whose journal
2230  * has recorded an error from a previous lifetime, move that error to the
2231  * main filesystem now.
2232  */
2233 static void ext4_clear_journal_err(struct super_block * sb,
2234                                    struct ext4_super_block * es)
2235 {
2236         journal_t *journal;
2237         int j_errno;
2238         const char *errstr;
2239
2240         journal = EXT4_SB(sb)->s_journal;
2241
2242         /*
2243          * Now check for any error status which may have been recorded in the
2244          * journal by a prior ext4_error() or ext4_abort()
2245          */
2246
2247         j_errno = jbd2_journal_errno(journal);
2248         if (j_errno) {
2249                 char nbuf[16];
2250
2251                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2252                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2253                              "from previous mount: %s", errstr);
2254                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2255                              "filesystem check.");
2256
2257                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2258                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2259                 ext4_commit_super (sb, es, 1);
2260
2261                 jbd2_journal_clear_err(journal);
2262         }
2263 }
2264
2265 /*
2266  * Force the running and committing transactions to commit,
2267  * and wait on the commit.
2268  */
2269 int ext4_force_commit(struct super_block *sb)
2270 {
2271         journal_t *journal;
2272         int ret;
2273
2274         if (sb->s_flags & MS_RDONLY)
2275                 return 0;
2276
2277         journal = EXT4_SB(sb)->s_journal;
2278         sb->s_dirt = 0;
2279         ret = ext4_journal_force_commit(journal);
2280         return ret;
2281 }
2282
2283 /*
2284  * Ext4 always journals updates to the superblock itself, so we don't
2285  * have to propagate any other updates to the superblock on disk at this
2286  * point.  Just start an async writeback to get the buffers on their way
2287  * to the disk.
2288  *
2289  * This implicitly triggers the writebehind on sync().
2290  */
2291
2292 static void ext4_write_super (struct super_block * sb)
2293 {
2294         if (mutex_trylock(&sb->s_lock) != 0)
2295                 BUG();
2296         sb->s_dirt = 0;
2297 }
2298
2299 static int ext4_sync_fs(struct super_block *sb, int wait)
2300 {
2301         tid_t target;
2302
2303         sb->s_dirt = 0;
2304         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2305                 if (wait)
2306                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2307         }
2308         return 0;
2309 }
2310
2311 /*
2312  * LVM calls this function before a (read-only) snapshot is created.  This
2313  * gives us a chance to flush the journal completely and mark the fs clean.
2314  */
2315 static void ext4_write_super_lockfs(struct super_block *sb)
2316 {
2317         sb->s_dirt = 0;
2318
2319         if (!(sb->s_flags & MS_RDONLY)) {
2320                 journal_t *journal = EXT4_SB(sb)->s_journal;
2321
2322                 /* Now we set up the journal barrier. */
2323                 jbd2_journal_lock_updates(journal);
2324                 jbd2_journal_flush(journal);
2325
2326                 /* Journal blocked and flushed, clear needs_recovery flag. */
2327                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2328                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2329         }
2330 }
2331
2332 /*
2333  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2334  * flag here, even though the filesystem is not technically dirty yet.
2335  */
2336 static void ext4_unlockfs(struct super_block *sb)
2337 {
2338         if (!(sb->s_flags & MS_RDONLY)) {
2339                 lock_super(sb);
2340                 /* Reser the needs_recovery flag before the fs is unlocked. */
2341                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2342                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2343                 unlock_super(sb);
2344                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2345         }
2346 }
2347
2348 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2349 {
2350         struct ext4_super_block * es;
2351         struct ext4_sb_info *sbi = EXT4_SB(sb);
2352         ext4_fsblk_t n_blocks_count = 0;
2353         unsigned long old_sb_flags;
2354         struct ext4_mount_options old_opts;
2355         int err;
2356 #ifdef CONFIG_QUOTA
2357         int i;
2358 #endif
2359
2360         /* Store the original options */
2361         old_sb_flags = sb->s_flags;
2362         old_opts.s_mount_opt = sbi->s_mount_opt;
2363         old_opts.s_resuid = sbi->s_resuid;
2364         old_opts.s_resgid = sbi->s_resgid;
2365         old_opts.s_commit_interval = sbi->s_commit_interval;
2366 #ifdef CONFIG_QUOTA
2367         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2368         for (i = 0; i < MAXQUOTAS; i++)
2369                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2370 #endif
2371
2372         /*
2373          * Allow the "check" option to be passed as a remount option.
2374          */
2375         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2376                 err = -EINVAL;
2377                 goto restore_opts;
2378         }
2379
2380         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2381                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2382
2383         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2384                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2385
2386         es = sbi->s_es;
2387
2388         ext4_init_journal_params(sb, sbi->s_journal);
2389
2390         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2391                 n_blocks_count > ext4_blocks_count(es)) {
2392                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2393                         err = -EROFS;
2394                         goto restore_opts;
2395                 }
2396
2397                 if (*flags & MS_RDONLY) {
2398                         /*
2399                          * First of all, the unconditional stuff we have to do
2400                          * to disable replay of the journal when we next remount
2401                          */
2402                         sb->s_flags |= MS_RDONLY;
2403
2404                         /*
2405                          * OK, test if we are remounting a valid rw partition
2406                          * readonly, and if so set the rdonly flag and then
2407                          * mark the partition as valid again.
2408                          */
2409                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2410                             (sbi->s_mount_state & EXT4_VALID_FS))
2411                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2412
2413                         ext4_mark_recovery_complete(sb, es);
2414                 } else {
2415                         __le32 ret;
2416                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2417                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2418                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2419                                        "remount RDWR because of unsupported "
2420                                        "optional features (%x).\n",
2421                                        sb->s_id, le32_to_cpu(ret));
2422                                 err = -EROFS;
2423                                 goto restore_opts;
2424                         }
2425
2426                         /*
2427                          * If we have an unprocessed orphan list hanging
2428                          * around from a previously readonly bdev mount,
2429                          * require a full umount/remount for now.
2430                          */
2431                         if (es->s_last_orphan) {
2432                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2433                                        "remount RDWR because of unprocessed "
2434                                        "orphan inode list.  Please "
2435                                        "umount/remount instead.\n",
2436                                        sb->s_id);
2437                                 err = -EINVAL;
2438                                 goto restore_opts;
2439                         }
2440
2441                         /*
2442                          * Mounting a RDONLY partition read-write, so reread
2443                          * and store the current valid flag.  (It may have
2444                          * been changed by e2fsck since we originally mounted
2445                          * the partition.)
2446                          */
2447                         ext4_clear_journal_err(sb, es);
2448                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2449                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2450                                 goto restore_opts;
2451                         if (!ext4_setup_super (sb, es, 0))
2452                                 sb->s_flags &= ~MS_RDONLY;
2453                 }
2454         }
2455 #ifdef CONFIG_QUOTA
2456         /* Release old quota file names */
2457         for (i = 0; i < MAXQUOTAS; i++)
2458                 if (old_opts.s_qf_names[i] &&
2459                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2460                         kfree(old_opts.s_qf_names[i]);
2461 #endif
2462         return 0;
2463 restore_opts:
2464         sb->s_flags = old_sb_flags;
2465         sbi->s_mount_opt = old_opts.s_mount_opt;
2466         sbi->s_resuid = old_opts.s_resuid;
2467         sbi->s_resgid = old_opts.s_resgid;
2468         sbi->s_commit_interval = old_opts.s_commit_interval;
2469 #ifdef CONFIG_QUOTA
2470         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2471         for (i = 0; i < MAXQUOTAS; i++) {
2472                 if (sbi->s_qf_names[i] &&
2473                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2474                         kfree(sbi->s_qf_names[i]);
2475                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2476         }
2477 #endif
2478         return err;
2479 }
2480
2481 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2482 {
2483         struct super_block *sb = dentry->d_sb;
2484         struct ext4_sb_info *sbi = EXT4_SB(sb);
2485         struct ext4_super_block *es = sbi->s_es;
2486         ext4_fsblk_t overhead;
2487         int i;
2488         u64 fsid;
2489
2490         if (test_opt (sb, MINIX_DF))
2491                 overhead = 0;
2492         else {
2493                 unsigned long ngroups;
2494                 ngroups = EXT4_SB(sb)->s_groups_count;
2495                 smp_rmb();
2496
2497                 /*
2498                  * Compute the overhead (FS structures)
2499                  */
2500
2501                 /*
2502                  * All of the blocks before first_data_block are
2503                  * overhead
2504                  */
2505                 overhead = le32_to_cpu(es->s_first_data_block);
2506
2507                 /*
2508                  * Add the overhead attributed to the superblock and
2509                  * block group descriptors.  If the sparse superblocks
2510                  * feature is turned on, then not all groups have this.
2511                  */
2512                 for (i = 0; i < ngroups; i++) {
2513                         overhead += ext4_bg_has_super(sb, i) +
2514                                 ext4_bg_num_gdb(sb, i);
2515                         cond_resched();
2516                 }
2517
2518                 /*
2519                  * Every block group has an inode bitmap, a block
2520                  * bitmap, and an inode table.
2521                  */
2522                 overhead += (ngroups * (2 + EXT4_SB(sb)->s_itb_per_group));
2523         }
2524
2525         buf->f_type = EXT4_SUPER_MAGIC;
2526         buf->f_bsize = sb->s_blocksize;
2527         buf->f_blocks = ext4_blocks_count(es) - overhead;
2528         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2529         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2530         if (buf->f_bfree < ext4_r_blocks_count(es))
2531                 buf->f_bavail = 0;
2532         buf->f_files = le32_to_cpu(es->s_inodes_count);
2533         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2534         buf->f_namelen = EXT4_NAME_LEN;
2535         fsid = le64_to_cpup((void *)es->s_uuid) ^
2536                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2537         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2538         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2539         return 0;
2540 }
2541
2542 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2543  * is locked for write. Otherwise the are possible deadlocks:
2544  * Process 1                         Process 2
2545  * ext4_create()                     quota_sync()
2546  *   jbd2_journal_start()                   write_dquot()
2547  *   DQUOT_INIT()                        down(dqio_mutex)
2548  *     down(dqio_mutex)                    jbd2_journal_start()
2549  *
2550  */
2551
2552 #ifdef CONFIG_QUOTA
2553
2554 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2555 {
2556         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2557 }
2558
2559 static int ext4_dquot_initialize(struct inode *inode, int type)
2560 {
2561         handle_t *handle;
2562         int ret, err;
2563
2564         /* We may create quota structure so we need to reserve enough blocks */
2565         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2566         if (IS_ERR(handle))
2567                 return PTR_ERR(handle);
2568         ret = dquot_initialize(inode, type);
2569         err = ext4_journal_stop(handle);
2570         if (!ret)
2571                 ret = err;
2572         return ret;
2573 }
2574
2575 static int ext4_dquot_drop(struct inode *inode)
2576 {
2577         handle_t *handle;
2578         int ret, err;
2579
2580         /* We may delete quota structure so we need to reserve enough blocks */
2581         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2582         if (IS_ERR(handle))
2583                 return PTR_ERR(handle);
2584         ret = dquot_drop(inode);
2585         err = ext4_journal_stop(handle);
2586         if (!ret)
2587                 ret = err;
2588         return ret;
2589 }
2590
2591 static int ext4_write_dquot(struct dquot *dquot)
2592 {
2593         int ret, err;
2594         handle_t *handle;
2595         struct inode *inode;
2596
2597         inode = dquot_to_inode(dquot);
2598         handle = ext4_journal_start(inode,
2599                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2600         if (IS_ERR(handle))
2601                 return PTR_ERR(handle);
2602         ret = dquot_commit(dquot);
2603         err = ext4_journal_stop(handle);
2604         if (!ret)
2605                 ret = err;
2606         return ret;
2607 }
2608
2609 static int ext4_acquire_dquot(struct dquot *dquot)
2610 {
2611         int ret, err;
2612         handle_t *handle;
2613
2614         handle = ext4_journal_start(dquot_to_inode(dquot),
2615                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2616         if (IS_ERR(handle))
2617                 return PTR_ERR(handle);
2618         ret = dquot_acquire(dquot);
2619         err = ext4_journal_stop(handle);
2620         if (!ret)
2621                 ret = err;
2622         return ret;
2623 }
2624
2625 static int ext4_release_dquot(struct dquot *dquot)
2626 {
2627         int ret, err;
2628         handle_t *handle;
2629
2630         handle = ext4_journal_start(dquot_to_inode(dquot),
2631                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2632         if (IS_ERR(handle))
2633                 return PTR_ERR(handle);
2634         ret = dquot_release(dquot);
2635         err = ext4_journal_stop(handle);
2636         if (!ret)
2637                 ret = err;
2638         return ret;
2639 }
2640
2641 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2642 {
2643         /* Are we journalling quotas? */
2644         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2645             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2646                 dquot_mark_dquot_dirty(dquot);
2647                 return ext4_write_dquot(dquot);
2648         } else {
2649                 return dquot_mark_dquot_dirty(dquot);
2650         }
2651 }
2652
2653 static int ext4_write_info(struct super_block *sb, int type)
2654 {
2655         int ret, err;
2656         handle_t *handle;
2657
2658         /* Data block + inode block */
2659         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2660         if (IS_ERR(handle))
2661                 return PTR_ERR(handle);
2662         ret = dquot_commit_info(sb, type);
2663         err = ext4_journal_stop(handle);
2664         if (!ret)
2665                 ret = err;
2666         return ret;
2667 }
2668
2669 /*
2670  * Turn on quotas during mount time - we need to find
2671  * the quota file and such...
2672  */
2673 static int ext4_quota_on_mount(struct super_block *sb, int type)
2674 {
2675         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2676                         EXT4_SB(sb)->s_jquota_fmt, type);
2677 }
2678
2679 /*
2680  * Standard function to be called on quota_on
2681  */
2682 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2683                          char *path)
2684 {
2685         int err;
2686         struct nameidata nd;
2687
2688         if (!test_opt(sb, QUOTA))
2689                 return -EINVAL;
2690         /* Not journalling quota? */
2691         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2692             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2693                 return vfs_quota_on(sb, type, format_id, path);
2694         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2695         if (err)
2696                 return err;
2697         /* Quotafile not on the same filesystem? */
2698         if (nd.mnt->mnt_sb != sb) {
2699                 path_release(&nd);
2700                 return -EXDEV;
2701         }
2702         /* Quotafile not of fs root? */
2703         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2704                 printk(KERN_WARNING
2705                         "EXT4-fs: Quota file not on filesystem root. "
2706                         "Journalled quota will not work.\n");
2707         path_release(&nd);
2708         return vfs_quota_on(sb, type, format_id, path);
2709 }
2710
2711 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2712  * acquiring the locks... As quota files are never truncated and quota code
2713  * itself serializes the operations (and noone else should touch the files)
2714  * we don't have to be afraid of races */
2715 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2716                                size_t len, loff_t off)
2717 {
2718         struct inode *inode = sb_dqopt(sb)->files[type];
2719         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2720         int err = 0;
2721         int offset = off & (sb->s_blocksize - 1);
2722         int tocopy;
2723         size_t toread;
2724         struct buffer_head *bh;
2725         loff_t i_size = i_size_read(inode);
2726
2727         if (off > i_size)
2728                 return 0;
2729         if (off+len > i_size)
2730                 len = i_size-off;
2731         toread = len;
2732         while (toread > 0) {
2733                 tocopy = sb->s_blocksize - offset < toread ?
2734                                 sb->s_blocksize - offset : toread;
2735                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2736                 if (err)
2737                         return err;
2738                 if (!bh)        /* A hole? */
2739                         memset(data, 0, tocopy);
2740                 else
2741                         memcpy(data, bh->b_data+offset, tocopy);
2742                 brelse(bh);
2743                 offset = 0;
2744                 toread -= tocopy;
2745                 data += tocopy;
2746                 blk++;
2747         }
2748         return len;
2749 }
2750
2751 /* Write to quotafile (we know the transaction is already started and has
2752  * enough credits) */
2753 static ssize_t ext4_quota_write(struct super_block *sb, int type,
2754                                 const char *data, size_t len, loff_t off)
2755 {
2756         struct inode *inode = sb_dqopt(sb)->files[type];
2757         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2758         int err = 0;
2759         int offset = off & (sb->s_blocksize - 1);
2760         int tocopy;
2761         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
2762         size_t towrite = len;
2763         struct buffer_head *bh;
2764         handle_t *handle = journal_current_handle();
2765
2766         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2767         while (towrite > 0) {
2768                 tocopy = sb->s_blocksize - offset < towrite ?
2769                                 sb->s_blocksize - offset : towrite;
2770                 bh = ext4_bread(handle, inode, blk, 1, &err);
2771                 if (!bh)
2772                         goto out;
2773                 if (journal_quota) {
2774                         err = ext4_journal_get_write_access(handle, bh);
2775                         if (err) {
2776                                 brelse(bh);
2777                                 goto out;
2778                         }
2779                 }
2780                 lock_buffer(bh);
2781                 memcpy(bh->b_data+offset, data, tocopy);
2782                 flush_dcache_page(bh->b_page);
2783                 unlock_buffer(bh);
2784                 if (journal_quota)
2785                         err = ext4_journal_dirty_metadata(handle, bh);
2786                 else {
2787                         /* Always do at least ordered writes for quotas */
2788                         err = ext4_journal_dirty_data(handle, bh);
2789                         mark_buffer_dirty(bh);
2790                 }
2791                 brelse(bh);
2792                 if (err)
2793                         goto out;
2794                 offset = 0;
2795                 towrite -= tocopy;
2796                 data += tocopy;
2797                 blk++;
2798         }
2799 out:
2800         if (len == towrite)
2801                 return err;
2802         if (inode->i_size < off+len-towrite) {
2803                 i_size_write(inode, off+len-towrite);
2804                 EXT4_I(inode)->i_disksize = inode->i_size;
2805         }
2806         inode->i_version++;
2807         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2808         ext4_mark_inode_dirty(handle, inode);
2809         mutex_unlock(&inode->i_mutex);
2810         return len - towrite;
2811 }
2812
2813 #endif
2814
2815 static int ext4_get_sb(struct file_system_type *fs_type,
2816         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2817 {
2818         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
2819 }
2820
2821 static struct file_system_type ext4dev_fs_type = {
2822         .owner          = THIS_MODULE,
2823         .name           = "ext4dev",
2824         .get_sb         = ext4_get_sb,
2825         .kill_sb        = kill_block_super,
2826         .fs_flags       = FS_REQUIRES_DEV,
2827 };
2828
2829 static int __init init_ext4_fs(void)
2830 {
2831         int err = init_ext4_xattr();
2832         if (err)
2833                 return err;
2834         err = init_inodecache();
2835         if (err)
2836                 goto out1;
2837         err = register_filesystem(&ext4dev_fs_type);
2838         if (err)
2839                 goto out;
2840         return 0;
2841 out:
2842         destroy_inodecache();
2843 out1:
2844         exit_ext4_xattr();
2845         return err;
2846 }
2847
2848 static void __exit exit_ext4_fs(void)
2849 {
2850         unregister_filesystem(&ext4dev_fs_type);
2851         destroy_inodecache();
2852         exit_ext4_xattr();
2853 }
2854
2855 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2856 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
2857 MODULE_LICENSE("GPL");
2858 module_init(init_ext4_fs)
2859 module_exit(exit_ext4_fs)