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