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