Pileus Git - ~andy/linux/blob - fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/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/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  49 #else
  50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  51 #endif
  52
  53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  54                              unsigned long journal_devnum);
  55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  56                                unsigned int);
  57 static int ext3_commit_super(struct super_block *sb,
  58                                struct ext3_super_block *es,
  59                                int sync);
  60 static void ext3_mark_recovery_complete(struct super_block * sb,
  61                                         struct ext3_super_block * es);
  62 static void ext3_clear_journal_err(struct super_block * sb,
  63                                    struct ext3_super_block * es);
  64 static int ext3_sync_fs(struct super_block *sb, int wait);
  65 static const char *ext3_decode_error(struct super_block * sb, int errno,
  66                                      char nbuf[16]);
  67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  69 static int ext3_unfreeze(struct super_block *sb);
  70 static int ext3_freeze(struct super_block *sb);
  71
  72 /*
  73  * Wrappers for journal_start/end.
  74  *
  75  * The only special thing we need to do here is to make sure that all
  76  * journal_end calls result in the superblock being marked dirty, so
  77  * that sync() will call the filesystem's write_super callback if
  78  * appropriate.
  79  */
  80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  81 {
  82         journal_t *journal;
  83
  84         if (sb->s_flags & MS_RDONLY)
  85                 return ERR_PTR(-EROFS);
  86
  87         /* Special case here: if the journal has aborted behind our
  88          * backs (eg. EIO in the commit thread), then we still need to
  89          * take the FS itself readonly cleanly. */
  90         journal = EXT3_SB(sb)->s_journal;
  91         if (is_journal_aborted(journal)) {
  92                 ext3_abort(sb, __func__,
  93                            "Detected aborted journal");
  94                 return ERR_PTR(-EROFS);
  95         }
  96
  97         return journal_start(journal, nblocks);
  98 }
  99
 100 /*
 101  * The only special thing we need to do here is to make sure that all
 102  * journal_stop calls result in the superblock being marked dirty, so
 103  * that sync() will call the filesystem's write_super callback if
 104  * appropriate.
 105  */
 106 int __ext3_journal_stop(const char *where, handle_t *handle)
 107 {
 108         struct super_block *sb;
 109         int err;
 110         int rc;
 111
 112         sb = handle->h_transaction->t_journal->j_private;
 113         err = handle->h_err;
 114         rc = journal_stop(handle);
 115
 116         if (!err)
 117                 err = rc;
 118         if (err)
 119                 __ext3_std_error(sb, where, err);
 120         return err;
 121 }
 122
 123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 124                 struct buffer_head *bh, handle_t *handle, int err)
 125 {
 126         char nbuf[16];
 127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 128
 129         if (bh)
 130                 BUFFER_TRACE(bh, "abort");
 131
 132         if (!handle->h_err)
 133                 handle->h_err = err;
 134
 135         if (is_handle_aborted(handle))
 136                 return;
 137
 138         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 139                caller, errstr, err_fn);
 140
 141         journal_abort_handle(handle);
 142 }
 143
 144 /* Deal with the reporting of failure conditions on a filesystem such as
 145  * inconsistencies detected or read IO failures.
 146  *
 147  * On ext2, we can store the error state of the filesystem in the
 148  * superblock.  That is not possible on ext3, because we may have other
 149  * write ordering constraints on the superblock which prevent us from
 150  * writing it out straight away; and given that the journal is about to
 151  * be aborted, we can't rely on the current, or future, transactions to
 152  * write out the superblock safely.
 153  *
 154  * We'll just use the journal_abort() error code to record an error in
 155  * the journal instead.  On recovery, the journal will compain about
 156  * that error until we've noted it down and cleared it.
 157  */
 158
 159 static void ext3_handle_error(struct super_block *sb)
 160 {
 161         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 162
 163         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 164         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 165
 166         if (sb->s_flags & MS_RDONLY)
 167                 return;
 168
 169         if (!test_opt (sb, ERRORS_CONT)) {
 170                 journal_t *journal = EXT3_SB(sb)->s_journal;
 171
 172                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 173                 if (journal)
 174                         journal_abort(journal, -EIO);
 175         }
 176         if (test_opt (sb, ERRORS_RO)) {
 177                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 178                 sb->s_flags |= MS_RDONLY;
 179         }
 180         ext3_commit_super(sb, es, 1);
 181         if (test_opt(sb, ERRORS_PANIC))
 182                 panic("EXT3-fs (device %s): panic forced after error\n",
 183                         sb->s_id);
 184 }
 185
 186 void ext3_error (struct super_block * sb, const char * function,
 187                  const char * fmt, ...)
 188 {
 189         va_list args;
 190
 191         va_start(args, fmt);
 192         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 193         vprintk(fmt, args);
 194         printk("\n");
 195         va_end(args);
 196
 197         ext3_handle_error(sb);
 198 }
 199
 200 static const char *ext3_decode_error(struct super_block * sb, int errno,
 201                                      char nbuf[16])
 202 {
 203         char *errstr = NULL;
 204
 205         switch (errno) {
 206         case -EIO:
 207                 errstr = "IO failure";
 208                 break;
 209         case -ENOMEM:
 210                 errstr = "Out of memory";
 211                 break;
 212         case -EROFS:
 213                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 214                         errstr = "Journal has aborted";
 215                 else
 216                         errstr = "Readonly filesystem";
 217                 break;
 218         default:
 219                 /* If the caller passed in an extra buffer for unknown
 220                  * errors, textualise them now.  Else we just return
 221                  * NULL. */
 222                 if (nbuf) {
 223                         /* Check for truncated error codes... */
 224                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 225                                 errstr = nbuf;
 226                 }
 227                 break;
 228         }
 229
 230         return errstr;
 231 }
 232
 233 /* __ext3_std_error decodes expected errors from journaling functions
 234  * automatically and invokes the appropriate error response.  */
 235
 236 void __ext3_std_error (struct super_block * sb, const char * function,
 237                        int errno)
 238 {
 239         char nbuf[16];
 240         const char *errstr;
 241
 242         /* Special case: if the error is EROFS, and we're not already
 243          * inside a transaction, then there's really no point in logging
 244          * an error. */
 245         if (errno == -EROFS && journal_current_handle() == NULL &&
 246             (sb->s_flags & MS_RDONLY))
 247                 return;
 248
 249         errstr = ext3_decode_error(sb, errno, nbuf);
 250         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 251                 sb->s_id, function, errstr);
 252
 253         ext3_handle_error(sb);
 254 }
 255
 256 /*
 257  * ext3_abort is a much stronger failure handler than ext3_error.  The
 258  * abort function may be used to deal with unrecoverable failures such
 259  * as journal IO errors or ENOMEM at a critical moment in log management.
 260  *
 261  * We unconditionally force the filesystem into an ABORT|READONLY state,
 262  * unless the error response on the fs has been set to panic in which
 263  * case we take the easy way out and panic immediately.
 264  */
 265
 266 void ext3_abort (struct super_block * sb, const char * function,
 267                  const char * fmt, ...)
 268 {
 269         va_list args;
 270
 271         printk (KERN_CRIT "ext3_abort called.\n");
 272
 273         va_start(args, fmt);
 274         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 275         vprintk(fmt, args);
 276         printk("\n");
 277         va_end(args);
 278
 279         if (test_opt(sb, ERRORS_PANIC))
 280                 panic("EXT3-fs panic from previous error\n");
 281
 282         if (sb->s_flags & MS_RDONLY)
 283                 return;
 284
 285         printk(KERN_CRIT "Remounting filesystem read-only\n");
 286         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 287         sb->s_flags |= MS_RDONLY;
 288         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 289         if (EXT3_SB(sb)->s_journal)
 290                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 291 }
 292
 293 void ext3_warning (struct super_block * sb, const char * function,
 294                    const char * fmt, ...)
 295 {
 296         va_list args;
 297
 298         va_start(args, fmt);
 299         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 300                sb->s_id, function);
 301         vprintk(fmt, args);
 302         printk("\n");
 303         va_end(args);
 304 }
 305
 306 void ext3_update_dynamic_rev(struct super_block *sb)
 307 {
 308         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 309
 310         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 311                 return;
 312
 313         ext3_warning(sb, __func__,
 314                      "updating to rev %d because of new feature flag, "
 315                      "running e2fsck is recommended",
 316                      EXT3_DYNAMIC_REV);
 317
 318         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 319         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 320         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 321         /* leave es->s_feature_*compat flags alone */
 322         /* es->s_uuid will be set by e2fsck if empty */
 323
 324         /*
 325          * The rest of the superblock fields should be zero, and if not it
 326          * means they are likely already in use, so leave them alone.  We
 327          * can leave it up to e2fsck to clean up any inconsistencies there.
 328          */
 329 }
 330
 331 /*
 332  * Open the external journal device
 333  */
 334 static struct block_device *ext3_blkdev_get(dev_t dev)
 335 {
 336         struct block_device *bdev;
 337         char b[BDEVNAME_SIZE];
 338
 339         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 340         if (IS_ERR(bdev))
 341                 goto fail;
 342         return bdev;
 343
 344 fail:
 345         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 346                         __bdevname(dev, b), PTR_ERR(bdev));
 347         return NULL;
 348 }
 349
 350 /*
 351  * Release the journal device
 352  */
 353 static int ext3_blkdev_put(struct block_device *bdev)
 354 {
 355         bd_release(bdev);
 356         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 357 }
 358
 359 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 360 {
 361         struct block_device *bdev;
 362         int ret = -ENODEV;
 363
 364         bdev = sbi->journal_bdev;
 365         if (bdev) {
 366                 ret = ext3_blkdev_put(bdev);
 367                 sbi->journal_bdev = NULL;
 368         }
 369         return ret;
 370 }
 371
 372 static inline struct inode *orphan_list_entry(struct list_head *l)
 373 {
 374         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 375 }
 376
 377 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 378 {
 379         struct list_head *l;
 380
 381         printk(KERN_ERR "sb orphan head is %d\n",
 382                le32_to_cpu(sbi->s_es->s_last_orphan));
 383
 384         printk(KERN_ERR "sb_info orphan list:\n");
 385         list_for_each(l, &sbi->s_orphan) {
 386                 struct inode *inode = orphan_list_entry(l);
 387                 printk(KERN_ERR "  "
 388                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 389                        inode->i_sb->s_id, inode->i_ino, inode,
 390                        inode->i_mode, inode->i_nlink,
 391                        NEXT_ORPHAN(inode));
 392         }
 393 }
 394
 395 static void ext3_put_super (struct super_block * sb)
 396 {
 397         struct ext3_sb_info *sbi = EXT3_SB(sb);
 398         struct ext3_super_block *es = sbi->s_es;
 399         int i, err;
 400
 401         lock_kernel();
 402
 403         ext3_xattr_put_super(sb);
 404         err = journal_destroy(sbi->s_journal);
 405         sbi->s_journal = NULL;
 406         if (err < 0)
 407                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 408
 409         if (!(sb->s_flags & MS_RDONLY)) {
 410                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 411                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 412                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 413                 mark_buffer_dirty(sbi->s_sbh);
 414                 ext3_commit_super(sb, es, 1);
 415         }
 416
 417         for (i = 0; i < sbi->s_gdb_count; i++)
 418                 brelse(sbi->s_group_desc[i]);
 419         kfree(sbi->s_group_desc);
 420         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 421         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 422         percpu_counter_destroy(&sbi->s_dirs_counter);
 423         brelse(sbi->s_sbh);
 424 #ifdef CONFIG_QUOTA
 425         for (i = 0; i < MAXQUOTAS; i++)
 426                 kfree(sbi->s_qf_names[i]);
 427 #endif
 428
 429         /* Debugging code just in case the in-memory inode orphan list
 430          * isn't empty.  The on-disk one can be non-empty if we've
 431          * detected an error and taken the fs readonly, but the
 432          * in-memory list had better be clean by this point. */
 433         if (!list_empty(&sbi->s_orphan))
 434                 dump_orphan_list(sb, sbi);
 435         J_ASSERT(list_empty(&sbi->s_orphan));
 436
 437         invalidate_bdev(sb->s_bdev);
 438         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 439                 /*
 440                  * Invalidate the journal device's buffers.  We don't want them
 441                  * floating about in memory - the physical journal device may
 442                  * hotswapped, and it breaks the `ro-after' testing code.
 443                  */
 444                 sync_blockdev(sbi->journal_bdev);
 445                 invalidate_bdev(sbi->journal_bdev);
 446                 ext3_blkdev_remove(sbi);
 447         }
 448         sb->s_fs_info = NULL;
 449         kfree(sbi->s_blockgroup_lock);
 450         kfree(sbi);
 451
 452         unlock_kernel();
 453 }
 454
 455 static struct kmem_cache *ext3_inode_cachep;
 456
 457 /*
 458  * Called inside transaction, so use GFP_NOFS
 459  */
 460 static struct inode *ext3_alloc_inode(struct super_block *sb)
 461 {
 462         struct ext3_inode_info *ei;
 463
 464         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 465         if (!ei)
 466                 return NULL;
 467         ei->i_block_alloc_info = NULL;
 468         ei->vfs_inode.i_version = 1;
 469         atomic_set(&ei->i_datasync_tid, 0);
 470         atomic_set(&ei->i_sync_tid, 0);
 471         return &ei->vfs_inode;
 472 }
 473
 474 static void ext3_destroy_inode(struct inode *inode)
 475 {
 476         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 477                 printk("EXT3 Inode %p: orphan list check failed!\n",
 478                         EXT3_I(inode));
 479                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 480                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 481                                 false);
 482                 dump_stack();
 483         }
 484         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 485 }
 486
 487 static void init_once(void *foo)
 488 {
 489         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 490
 491         INIT_LIST_HEAD(&ei->i_orphan);
 492 #ifdef CONFIG_EXT3_FS_XATTR
 493         init_rwsem(&ei->xattr_sem);
 494 #endif
 495         mutex_init(&ei->truncate_mutex);
 496         inode_init_once(&ei->vfs_inode);
 497 }
 498
 499 static int init_inodecache(void)
 500 {
 501         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 502                                              sizeof(struct ext3_inode_info),
 503                                              0, (SLAB_RECLAIM_ACCOUNT|
 504                                                 SLAB_MEM_SPREAD),
 505                                              init_once);
 506         if (ext3_inode_cachep == NULL)
 507                 return -ENOMEM;
 508         return 0;
 509 }
 510
 511 static void destroy_inodecache(void)
 512 {
 513         kmem_cache_destroy(ext3_inode_cachep);
 514 }
 515
 516 static void ext3_clear_inode(struct inode *inode)
 517 {
 518         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 519         ext3_discard_reservation(inode);
 520         EXT3_I(inode)->i_block_alloc_info = NULL;
 521         if (unlikely(rsv))
 522                 kfree(rsv);
 523 }
 524
 525 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 526 {
 527 #if defined(CONFIG_QUOTA)
 528         struct ext3_sb_info *sbi = EXT3_SB(sb);
 529
 530         if (sbi->s_jquota_fmt)
 531                 seq_printf(seq, ",jqfmt=%s",
 532                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 533
 534         if (sbi->s_qf_names[USRQUOTA])
 535                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 536
 537         if (sbi->s_qf_names[GRPQUOTA])
 538                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 539
 540         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 541                 seq_puts(seq, ",usrquota");
 542
 543         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 544                 seq_puts(seq, ",grpquota");
 545 #endif
 546 }
 547
 548 static char *data_mode_string(unsigned long mode)
 549 {
 550         switch (mode) {
 551         case EXT3_MOUNT_JOURNAL_DATA:
 552                 return "journal";
 553         case EXT3_MOUNT_ORDERED_DATA:
 554                 return "ordered";
 555         case EXT3_MOUNT_WRITEBACK_DATA:
 556                 return "writeback";
 557         }
 558         return "unknown";
 559 }
 560
 561 /*
 562  * Show an option if
 563  *  - it's set to a non-default value OR
 564  *  - if the per-sb default is different from the global default
 565  */
 566 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 567 {
 568         struct super_block *sb = vfs->mnt_sb;
 569         struct ext3_sb_info *sbi = EXT3_SB(sb);
 570         struct ext3_super_block *es = sbi->s_es;
 571         unsigned long def_mount_opts;
 572
 573         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 574
 575         if (sbi->s_sb_block != 1)
 576                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 577         if (test_opt(sb, MINIX_DF))
 578                 seq_puts(seq, ",minixdf");
 579         if (test_opt(sb, GRPID))
 580                 seq_puts(seq, ",grpid");
 581         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 582                 seq_puts(seq, ",nogrpid");
 583         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 584             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 585                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 586         }
 587         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 588             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 589                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 590         }
 591         if (test_opt(sb, ERRORS_RO)) {
 592                 int def_errors = le16_to_cpu(es->s_errors);
 593
 594                 if (def_errors == EXT3_ERRORS_PANIC ||
 595                     def_errors == EXT3_ERRORS_CONTINUE) {
 596                         seq_puts(seq, ",errors=remount-ro");
 597                 }
 598         }
 599         if (test_opt(sb, ERRORS_CONT))
 600                 seq_puts(seq, ",errors=continue");
 601         if (test_opt(sb, ERRORS_PANIC))
 602                 seq_puts(seq, ",errors=panic");
 603         if (test_opt(sb, NO_UID32))
 604                 seq_puts(seq, ",nouid32");
 605         if (test_opt(sb, DEBUG))
 606                 seq_puts(seq, ",debug");
 607         if (test_opt(sb, OLDALLOC))
 608                 seq_puts(seq, ",oldalloc");
 609 #ifdef CONFIG_EXT3_FS_XATTR
 610         if (test_opt(sb, XATTR_USER))
 611                 seq_puts(seq, ",user_xattr");
 612         if (!test_opt(sb, XATTR_USER) &&
 613             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 614                 seq_puts(seq, ",nouser_xattr");
 615         }
 616 #endif
 617 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 618         if (test_opt(sb, POSIX_ACL))
 619                 seq_puts(seq, ",acl");
 620         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 621                 seq_puts(seq, ",noacl");
 622 #endif
 623         if (!test_opt(sb, RESERVATION))
 624                 seq_puts(seq, ",noreservation");
 625         if (sbi->s_commit_interval) {
 626                 seq_printf(seq, ",commit=%u",
 627                            (unsigned) (sbi->s_commit_interval / HZ));
 628         }
 629         if (test_opt(sb, BARRIER))
 630                 seq_puts(seq, ",barrier=1");
 631         if (test_opt(sb, NOBH))
 632                 seq_puts(seq, ",nobh");
 633
 634         seq_printf(seq, ",data=%s", data_mode_string(sbi->s_mount_opt &
 635                                                      EXT3_MOUNT_DATA_FLAGS));
 636         if (test_opt(sb, DATA_ERR_ABORT))
 637                 seq_puts(seq, ",data_err=abort");
 638
 639         if (test_opt(sb, NOLOAD))
 640                 seq_puts(seq, ",norecovery");
 641
 642         ext3_show_quota_options(seq, sb);
 643
 644         return 0;
 645 }
 646
 647
 648 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 649                 u64 ino, u32 generation)
 650 {
 651         struct inode *inode;
 652
 653         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 654                 return ERR_PTR(-ESTALE);
 655         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 656                 return ERR_PTR(-ESTALE);
 657
 658         /* iget isn't really right if the inode is currently unallocated!!
 659          *
 660          * ext3_read_inode will return a bad_inode if the inode had been
 661          * deleted, so we should be safe.
 662          *
 663          * Currently we don't know the generation for parent directory, so
 664          * a generation of 0 means "accept any"
 665          */
 666         inode = ext3_iget(sb, ino);
 667         if (IS_ERR(inode))
 668                 return ERR_CAST(inode);
 669         if (generation && inode->i_generation != generation) {
 670                 iput(inode);
 671                 return ERR_PTR(-ESTALE);
 672         }
 673
 674         return inode;
 675 }
 676
 677 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 678                 int fh_len, int fh_type)
 679 {
 680         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 681                                     ext3_nfs_get_inode);
 682 }
 683
 684 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 685                 int fh_len, int fh_type)
 686 {
 687         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 688                                     ext3_nfs_get_inode);
 689 }
 690
 691 /*
 692  * Try to release metadata pages (indirect blocks, directories) which are
 693  * mapped via the block device.  Since these pages could have journal heads
 694  * which would prevent try_to_free_buffers() from freeing them, we must use
 695  * jbd layer's try_to_free_buffers() function to release them.
 696  */
 697 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 698                                  gfp_t wait)
 699 {
 700         journal_t *journal = EXT3_SB(sb)->s_journal;
 701
 702         WARN_ON(PageChecked(page));
 703         if (!page_has_buffers(page))
 704                 return 0;
 705         if (journal)
 706                 return journal_try_to_free_buffers(journal, page,
 707                                                    wait & ~__GFP_WAIT);
 708         return try_to_free_buffers(page);
 709 }
 710
 711 #ifdef CONFIG_QUOTA
 712 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 713 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 714
 715 static int ext3_write_dquot(struct dquot *dquot);
 716 static int ext3_acquire_dquot(struct dquot *dquot);
 717 static int ext3_release_dquot(struct dquot *dquot);
 718 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 719 static int ext3_write_info(struct super_block *sb, int type);
 720 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 721                                 char *path, int remount);
 722 static int ext3_quota_on_mount(struct super_block *sb, int type);
 723 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 724                                size_t len, loff_t off);
 725 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 726                                 const char *data, size_t len, loff_t off);
 727
 728 static const struct dquot_operations ext3_quota_operations = {
 729         .initialize     = dquot_initialize,
 730         .drop           = dquot_drop,
 731         .alloc_space    = dquot_alloc_space,
 732         .alloc_inode    = dquot_alloc_inode,
 733         .free_space     = dquot_free_space,
 734         .free_inode     = dquot_free_inode,
 735         .transfer       = dquot_transfer,
 736         .write_dquot    = ext3_write_dquot,
 737         .acquire_dquot  = ext3_acquire_dquot,
 738         .release_dquot  = ext3_release_dquot,
 739         .mark_dirty     = ext3_mark_dquot_dirty,
 740         .write_info     = ext3_write_info,
 741         .alloc_dquot    = dquot_alloc,
 742         .destroy_dquot  = dquot_destroy,
 743 };
 744
 745 static const struct quotactl_ops ext3_qctl_operations = {
 746         .quota_on       = ext3_quota_on,
 747         .quota_off      = vfs_quota_off,
 748         .quota_sync     = vfs_quota_sync,
 749         .get_info       = vfs_get_dqinfo,
 750         .set_info       = vfs_set_dqinfo,
 751         .get_dqblk      = vfs_get_dqblk,
 752         .set_dqblk      = vfs_set_dqblk
 753 };
 754 #endif
 755
 756 static const struct super_operations ext3_sops = {
 757         .alloc_inode    = ext3_alloc_inode,
 758         .destroy_inode  = ext3_destroy_inode,
 759         .write_inode    = ext3_write_inode,
 760         .dirty_inode    = ext3_dirty_inode,
 761         .delete_inode   = ext3_delete_inode,
 762         .put_super      = ext3_put_super,
 763         .sync_fs        = ext3_sync_fs,
 764         .freeze_fs      = ext3_freeze,
 765         .unfreeze_fs    = ext3_unfreeze,
 766         .statfs         = ext3_statfs,
 767         .remount_fs     = ext3_remount,
 768         .clear_inode    = ext3_clear_inode,
 769         .show_options   = ext3_show_options,
 770 #ifdef CONFIG_QUOTA
 771         .quota_read     = ext3_quota_read,
 772         .quota_write    = ext3_quota_write,
 773 #endif
 774         .bdev_try_to_free_page = bdev_try_to_free_page,
 775 };
 776
 777 static const struct export_operations ext3_export_ops = {
 778         .fh_to_dentry = ext3_fh_to_dentry,
 779         .fh_to_parent = ext3_fh_to_parent,
 780         .get_parent = ext3_get_parent,
 781 };
 782
 783 enum {
 784         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 785         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 786         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 787         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 788         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 789         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 790         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 791         Opt_data_err_abort, Opt_data_err_ignore,
 792         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 793         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 794         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 795         Opt_grpquota
 796 };
 797
 798 static const match_table_t tokens = {
 799         {Opt_bsd_df, "bsddf"},
 800         {Opt_minix_df, "minixdf"},
 801         {Opt_grpid, "grpid"},
 802         {Opt_grpid, "bsdgroups"},
 803         {Opt_nogrpid, "nogrpid"},
 804         {Opt_nogrpid, "sysvgroups"},
 805         {Opt_resgid, "resgid=%u"},
 806         {Opt_resuid, "resuid=%u"},
 807         {Opt_sb, "sb=%u"},
 808         {Opt_err_cont, "errors=continue"},
 809         {Opt_err_panic, "errors=panic"},
 810         {Opt_err_ro, "errors=remount-ro"},
 811         {Opt_nouid32, "nouid32"},
 812         {Opt_nocheck, "nocheck"},
 813         {Opt_nocheck, "check=none"},
 814         {Opt_debug, "debug"},
 815         {Opt_oldalloc, "oldalloc"},
 816         {Opt_orlov, "orlov"},
 817         {Opt_user_xattr, "user_xattr"},
 818         {Opt_nouser_xattr, "nouser_xattr"},
 819         {Opt_acl, "acl"},
 820         {Opt_noacl, "noacl"},
 821         {Opt_reservation, "reservation"},
 822         {Opt_noreservation, "noreservation"},
 823         {Opt_noload, "noload"},
 824         {Opt_noload, "norecovery"},
 825         {Opt_nobh, "nobh"},
 826         {Opt_bh, "bh"},
 827         {Opt_commit, "commit=%u"},
 828         {Opt_journal_update, "journal=update"},
 829         {Opt_journal_inum, "journal=%u"},
 830         {Opt_journal_dev, "journal_dev=%u"},
 831         {Opt_abort, "abort"},
 832         {Opt_data_journal, "data=journal"},
 833         {Opt_data_ordered, "data=ordered"},
 834         {Opt_data_writeback, "data=writeback"},
 835         {Opt_data_err_abort, "data_err=abort"},
 836         {Opt_data_err_ignore, "data_err=ignore"},
 837         {Opt_offusrjquota, "usrjquota="},
 838         {Opt_usrjquota, "usrjquota=%s"},
 839         {Opt_offgrpjquota, "grpjquota="},
 840         {Opt_grpjquota, "grpjquota=%s"},
 841         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 842         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 843         {Opt_grpquota, "grpquota"},
 844         {Opt_noquota, "noquota"},
 845         {Opt_quota, "quota"},
 846         {Opt_usrquota, "usrquota"},
 847         {Opt_barrier, "barrier=%u"},
 848         {Opt_resize, "resize"},
 849         {Opt_err, NULL},
 850 };
 851
 852 static ext3_fsblk_t get_sb_block(void **data)
 853 {
 854         ext3_fsblk_t    sb_block;
 855         char            *options = (char *) *data;
 856
 857         if (!options || strncmp(options, "sb=", 3) != 0)
 858                 return 1;       /* Default location */
 859         options += 3;
 860         /*todo: use simple_strtoll with >32bit ext3 */
 861         sb_block = simple_strtoul(options, &options, 0);
 862         if (*options && *options != ',') {
 863                 printk("EXT3-fs: Invalid sb specification: %s\n",
 864                        (char *) *data);
 865                 return 1;
 866         }
 867         if (*options == ',')
 868                 options++;
 869         *data = (void *) options;
 870         return sb_block;
 871 }
 872
 873 static int parse_options (char *options, struct super_block *sb,
 874                           unsigned int *inum, unsigned long *journal_devnum,
 875                           ext3_fsblk_t *n_blocks_count, int is_remount)
 876 {
 877         struct ext3_sb_info *sbi = EXT3_SB(sb);
 878         char * p;
 879         substring_t args[MAX_OPT_ARGS];
 880         int data_opt = 0;
 881         int option;
 882 #ifdef CONFIG_QUOTA
 883         int qtype, qfmt;
 884         char *qname;
 885 #endif
 886
 887         if (!options)
 888                 return 1;
 889
 890         while ((p = strsep (&options, ",")) != NULL) {
 891                 int token;
 892                 if (!*p)
 893                         continue;
 894
 895                 token = match_token(p, tokens, args);
 896                 switch (token) {
 897                 case Opt_bsd_df:
 898                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 899                         break;
 900                 case Opt_minix_df:
 901                         set_opt (sbi->s_mount_opt, MINIX_DF);
 902                         break;
 903                 case Opt_grpid:
 904                         set_opt (sbi->s_mount_opt, GRPID);
 905                         break;
 906                 case Opt_nogrpid:
 907                         clear_opt (sbi->s_mount_opt, GRPID);
 908                         break;
 909                 case Opt_resuid:
 910                         if (match_int(&args[0], &option))
 911                                 return 0;
 912                         sbi->s_resuid = option;
 913                         break;
 914                 case Opt_resgid:
 915                         if (match_int(&args[0], &option))
 916                                 return 0;
 917                         sbi->s_resgid = option;
 918                         break;
 919                 case Opt_sb:
 920                         /* handled by get_sb_block() instead of here */
 921                         /* *sb_block = match_int(&args[0]); */
 922                         break;
 923                 case Opt_err_panic:
 924                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 925                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 926                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 927                         break;
 928                 case Opt_err_ro:
 929                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 930                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 931                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 932                         break;
 933                 case Opt_err_cont:
 934                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 935                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 936                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 937                         break;
 938                 case Opt_nouid32:
 939                         set_opt (sbi->s_mount_opt, NO_UID32);
 940                         break;
 941                 case Opt_nocheck:
 942                         clear_opt (sbi->s_mount_opt, CHECK);
 943                         break;
 944                 case Opt_debug:
 945                         set_opt (sbi->s_mount_opt, DEBUG);
 946                         break;
 947                 case Opt_oldalloc:
 948                         set_opt (sbi->s_mount_opt, OLDALLOC);
 949                         break;
 950                 case Opt_orlov:
 951                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 952                         break;
 953 #ifdef CONFIG_EXT3_FS_XATTR
 954                 case Opt_user_xattr:
 955                         set_opt (sbi->s_mount_opt, XATTR_USER);
 956                         break;
 957                 case Opt_nouser_xattr:
 958                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 959                         break;
 960 #else
 961                 case Opt_user_xattr:
 962                 case Opt_nouser_xattr:
 963                         printk("EXT3 (no)user_xattr options not supported\n");
 964                         break;
 965 #endif
 966 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 967                 case Opt_acl:
 968                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 969                         break;
 970                 case Opt_noacl:
 971                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 972                         break;
 973 #else
 974                 case Opt_acl:
 975                 case Opt_noacl:
 976                         printk("EXT3 (no)acl options not supported\n");
 977                         break;
 978 #endif
 979                 case Opt_reservation:
 980                         set_opt(sbi->s_mount_opt, RESERVATION);
 981                         break;
 982                 case Opt_noreservation:
 983                         clear_opt(sbi->s_mount_opt, RESERVATION);
 984                         break;
 985                 case Opt_journal_update:
 986                         /* @@@ FIXME */
 987                         /* Eventually we will want to be able to create
 988                            a journal file here.  For now, only allow the
 989                            user to specify an existing inode to be the
 990                            journal file. */
 991                         if (is_remount) {
 992                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 993                                        "journal on remount\n");
 994                                 return 0;
 995                         }
 996                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 997                         break;
 998                 case Opt_journal_inum:
 999                         if (is_remount) {
1000                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1001                                        "journal on remount\n");
1002                                 return 0;
1003                         }
1004                         if (match_int(&args[0], &option))
1005                                 return 0;
1006                         *inum = option;
1007                         break;
1008                 case Opt_journal_dev:
1009                         if (is_remount) {
1010                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1011                                        "journal on remount\n");
1012                                 return 0;
1013                         }
1014                         if (match_int(&args[0], &option))
1015                                 return 0;
1016                         *journal_devnum = option;
1017                         break;
1018                 case Opt_noload:
1019                         set_opt (sbi->s_mount_opt, NOLOAD);
1020                         break;
1021                 case Opt_commit:
1022                         if (match_int(&args[0], &option))
1023                                 return 0;
1024                         if (option < 0)
1025                                 return 0;
1026                         if (option == 0)
1027                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1028                         sbi->s_commit_interval = HZ * option;
1029                         break;
1030                 case Opt_data_journal:
1031                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1032                         goto datacheck;
1033                 case Opt_data_ordered:
1034                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1035                         goto datacheck;
1036                 case Opt_data_writeback:
1037                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1038                 datacheck:
1039                         if (is_remount) {
1040                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1041                                                 == data_opt)
1042                                         break;
1043                                 printk(KERN_ERR
1044                                         "EXT3-fs (device %s): Cannot change "
1045                                         "data mode on remount. The filesystem "
1046                                         "is mounted in data=%s mode and you "
1047                                         "try to remount it in data=%s mode.\n",
1048                                         sb->s_id,
1049                                         data_mode_string(sbi->s_mount_opt &
1050                                                         EXT3_MOUNT_DATA_FLAGS),
1051                                         data_mode_string(data_opt));
1052                                 return 0;
1053                         } else {
1054                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1055                                 sbi->s_mount_opt |= data_opt;
1056                         }
1057                         break;
1058                 case Opt_data_err_abort:
1059                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1060                         break;
1061                 case Opt_data_err_ignore:
1062                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1063                         break;
1064 #ifdef CONFIG_QUOTA
1065                 case Opt_usrjquota:
1066                         qtype = USRQUOTA;
1067                         goto set_qf_name;
1068                 case Opt_grpjquota:
1069                         qtype = GRPQUOTA;
1070 set_qf_name:
1071                         if (sb_any_quota_loaded(sb) &&
1072                             !sbi->s_qf_names[qtype]) {
1073                                 printk(KERN_ERR
1074                                         "EXT3-fs: Cannot change journaled "
1075                                         "quota options when quota turned on.\n");
1076                                 return 0;
1077                         }
1078                         qname = match_strdup(&args[0]);
1079                         if (!qname) {
1080                                 printk(KERN_ERR
1081                                         "EXT3-fs: not enough memory for "
1082                                         "storing quotafile name.\n");
1083                                 return 0;
1084                         }
1085                         if (sbi->s_qf_names[qtype] &&
1086                             strcmp(sbi->s_qf_names[qtype], qname)) {
1087                                 printk(KERN_ERR
1088                                         "EXT3-fs: %s quota file already "
1089                                         "specified.\n", QTYPE2NAME(qtype));
1090                                 kfree(qname);
1091                                 return 0;
1092                         }
1093                         sbi->s_qf_names[qtype] = qname;
1094                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1095                                 printk(KERN_ERR
1096                                         "EXT3-fs: quotafile must be on "
1097                                         "filesystem root.\n");
1098                                 kfree(sbi->s_qf_names[qtype]);
1099                                 sbi->s_qf_names[qtype] = NULL;
1100                                 return 0;
1101                         }
1102                         set_opt(sbi->s_mount_opt, QUOTA);
1103                         break;
1104                 case Opt_offusrjquota:
1105                         qtype = USRQUOTA;
1106                         goto clear_qf_name;
1107                 case Opt_offgrpjquota:
1108                         qtype = GRPQUOTA;
1109 clear_qf_name:
1110                         if (sb_any_quota_loaded(sb) &&
1111                             sbi->s_qf_names[qtype]) {
1112                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1113                                         "journaled quota options when "
1114                                         "quota turned on.\n");
1115                                 return 0;
1116                         }
1117                         /*
1118                          * The space will be released later when all options
1119                          * are confirmed to be correct
1120                          */
1121                         sbi->s_qf_names[qtype] = NULL;
1122                         break;
1123                 case Opt_jqfmt_vfsold:
1124                         qfmt = QFMT_VFS_OLD;
1125                         goto set_qf_format;
1126                 case Opt_jqfmt_vfsv0:
1127                         qfmt = QFMT_VFS_V0;
1128 set_qf_format:
1129                         if (sb_any_quota_loaded(sb) &&
1130                             sbi->s_jquota_fmt != qfmt) {
1131                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1132                                         "journaled quota options when "
1133                                         "quota turned on.\n");
1134                                 return 0;
1135                         }
1136                         sbi->s_jquota_fmt = qfmt;
1137                         break;
1138                 case Opt_quota:
1139                 case Opt_usrquota:
1140                         set_opt(sbi->s_mount_opt, QUOTA);
1141                         set_opt(sbi->s_mount_opt, USRQUOTA);
1142                         break;
1143                 case Opt_grpquota:
1144                         set_opt(sbi->s_mount_opt, QUOTA);
1145                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1146                         break;
1147                 case Opt_noquota:
1148                         if (sb_any_quota_loaded(sb)) {
1149                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1150                                         "options when quota turned on.\n");
1151                                 return 0;
1152                         }
1153                         clear_opt(sbi->s_mount_opt, QUOTA);
1154                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1155                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1156                         break;
1157 #else
1158                 case Opt_quota:
1159                 case Opt_usrquota:
1160                 case Opt_grpquota:
1161                         printk(KERN_ERR
1162                                 "EXT3-fs: quota options not supported.\n");
1163                         break;
1164                 case Opt_usrjquota:
1165                 case Opt_grpjquota:
1166                 case Opt_offusrjquota:
1167                 case Opt_offgrpjquota:
1168                 case Opt_jqfmt_vfsold:
1169                 case Opt_jqfmt_vfsv0:
1170                         printk(KERN_ERR
1171                                 "EXT3-fs: journaled quota options not "
1172                                 "supported.\n");
1173                         break;
1174                 case Opt_noquota:
1175                         break;
1176 #endif
1177                 case Opt_abort:
1178                         set_opt(sbi->s_mount_opt, ABORT);
1179                         break;
1180                 case Opt_barrier:
1181                         if (match_int(&args[0], &option))
1182                                 return 0;
1183                         if (option)
1184                                 set_opt(sbi->s_mount_opt, BARRIER);
1185                         else
1186                                 clear_opt(sbi->s_mount_opt, BARRIER);
1187                         break;
1188                 case Opt_ignore:
1189                         break;
1190                 case Opt_resize:
1191                         if (!is_remount) {
1192                                 printk("EXT3-fs: resize option only available "
1193                                         "for remount\n");
1194                                 return 0;
1195                         }
1196                         if (match_int(&args[0], &option) != 0)
1197                                 return 0;
1198                         *n_blocks_count = option;
1199                         break;
1200                 case Opt_nobh:
1201                         set_opt(sbi->s_mount_opt, NOBH);
1202                         break;
1203                 case Opt_bh:
1204                         clear_opt(sbi->s_mount_opt, NOBH);
1205                         break;
1206                 default:
1207                         printk (KERN_ERR
1208                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1209                                 "or missing value\n", p);
1210                         return 0;
1211                 }
1212         }
1213 #ifdef CONFIG_QUOTA
1214         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1215                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1216                      sbi->s_qf_names[USRQUOTA])
1217                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1218
1219                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1220                      sbi->s_qf_names[GRPQUOTA])
1221                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1222
1223                 if ((sbi->s_qf_names[USRQUOTA] &&
1224                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1225                     (sbi->s_qf_names[GRPQUOTA] &&
1226                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1227                         printk(KERN_ERR "EXT3-fs: old and new quota "
1228                                         "format mixing.\n");
1229                         return 0;
1230                 }
1231
1232                 if (!sbi->s_jquota_fmt) {
1233                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1234                                         "not specified.\n");
1235                         return 0;
1236                 }
1237         } else {
1238                 if (sbi->s_jquota_fmt) {
1239                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1240                                         "specified with no journaling "
1241                                         "enabled.\n");
1242                         return 0;
1243                 }
1244         }
1245 #endif
1246         return 1;
1247 }
1248
1249 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1250                             int read_only)
1251 {
1252         struct ext3_sb_info *sbi = EXT3_SB(sb);
1253         int res = 0;
1254
1255         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1256                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1257                         "forcing read-only mode\n");
1258                 res = MS_RDONLY;
1259         }
1260         if (read_only)
1261                 return res;
1262         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1263                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1264                         "running e2fsck is recommended\n");
1265         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1266                 printk (KERN_WARNING
1267                         "EXT3-fs warning: mounting fs with errors, "
1268                         "running e2fsck is recommended\n");
1269         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1270                  le16_to_cpu(es->s_mnt_count) >=
1271                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1272                 printk (KERN_WARNING
1273                         "EXT3-fs warning: maximal mount count reached, "
1274                         "running e2fsck is recommended\n");
1275         else if (le32_to_cpu(es->s_checkinterval) &&
1276                 (le32_to_cpu(es->s_lastcheck) +
1277                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1278                 printk (KERN_WARNING
1279                         "EXT3-fs warning: checktime reached, "
1280                         "running e2fsck is recommended\n");
1281 #if 0
1282                 /* @@@ We _will_ want to clear the valid bit if we find
1283                    inconsistencies, to force a fsck at reboot.  But for
1284                    a plain journaled filesystem we can keep it set as
1285                    valid forever! :) */
1286         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1287 #endif
1288         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1289                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1290         le16_add_cpu(&es->s_mnt_count, 1);
1291         es->s_mtime = cpu_to_le32(get_seconds());
1292         ext3_update_dynamic_rev(sb);
1293         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1294
1295         ext3_commit_super(sb, es, 1);
1296         if (test_opt(sb, DEBUG))
1297                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1298                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1299                         sb->s_blocksize,
1300                         sbi->s_groups_count,
1301                         EXT3_BLOCKS_PER_GROUP(sb),
1302                         EXT3_INODES_PER_GROUP(sb),
1303                         sbi->s_mount_opt);
1304
1305         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1306         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1307                 char b[BDEVNAME_SIZE];
1308
1309                 printk("external journal on %s\n",
1310                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1311         } else {
1312                 printk("internal journal\n");
1313         }
1314         return res;
1315 }
1316
1317 /* Called at mount-time, super-block is locked */
1318 static int ext3_check_descriptors(struct super_block *sb)
1319 {
1320         struct ext3_sb_info *sbi = EXT3_SB(sb);
1321         int i;
1322
1323         ext3_debug ("Checking group descriptors");
1324
1325         for (i = 0; i < sbi->s_groups_count; i++) {
1326                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1327                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1328                 ext3_fsblk_t last_block;
1329
1330                 if (i == sbi->s_groups_count - 1)
1331                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1332                 else
1333                         last_block = first_block +
1334                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1335
1336                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1337                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1338                 {
1339                         ext3_error (sb, "ext3_check_descriptors",
1340                                     "Block bitmap for group %d"
1341                                     " not in group (block %lu)!",
1342                                     i, (unsigned long)
1343                                         le32_to_cpu(gdp->bg_block_bitmap));
1344                         return 0;
1345                 }
1346                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1347                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1348                 {
1349                         ext3_error (sb, "ext3_check_descriptors",
1350                                     "Inode bitmap for group %d"
1351                                     " not in group (block %lu)!",
1352                                     i, (unsigned long)
1353                                         le32_to_cpu(gdp->bg_inode_bitmap));
1354                         return 0;
1355                 }
1356                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1357                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1358                     last_block)
1359                 {
1360                         ext3_error (sb, "ext3_check_descriptors",
1361                                     "Inode table for group %d"
1362                                     " not in group (block %lu)!",
1363                                     i, (unsigned long)
1364                                         le32_to_cpu(gdp->bg_inode_table));
1365                         return 0;
1366                 }
1367         }
1368
1369         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1370         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1371         return 1;
1372 }
1373
1374
1375 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1376  * the superblock) which were deleted from all directories, but held open by
1377  * a process at the time of a crash.  We walk the list and try to delete these
1378  * inodes at recovery time (only with a read-write filesystem).
1379  *
1380  * In order to keep the orphan inode chain consistent during traversal (in
1381  * case of crash during recovery), we link each inode into the superblock
1382  * orphan list_head and handle it the same way as an inode deletion during
1383  * normal operation (which journals the operations for us).
1384  *
1385  * We only do an iget() and an iput() on each inode, which is very safe if we
1386  * accidentally point at an in-use or already deleted inode.  The worst that
1387  * can happen in this case is that we get a "bit already cleared" message from
1388  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1389  * e2fsck was run on this filesystem, and it must have already done the orphan
1390  * inode cleanup for us, so we can safely abort without any further action.
1391  */
1392 static void ext3_orphan_cleanup (struct super_block * sb,
1393                                  struct ext3_super_block * es)
1394 {
1395         unsigned int s_flags = sb->s_flags;
1396         int nr_orphans = 0, nr_truncates = 0;
1397 #ifdef CONFIG_QUOTA
1398         int i;
1399 #endif
1400         if (!es->s_last_orphan) {
1401                 jbd_debug(4, "no orphan inodes to clean up\n");
1402                 return;
1403         }
1404
1405         if (bdev_read_only(sb->s_bdev)) {
1406                 printk(KERN_ERR "EXT3-fs: write access "
1407                         "unavailable, skipping orphan cleanup.\n");
1408                 return;
1409         }
1410
1411         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1412                 if (es->s_last_orphan)
1413                         jbd_debug(1, "Errors on filesystem, "
1414                                   "clearing orphan list.\n");
1415                 es->s_last_orphan = 0;
1416                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1417                 return;
1418         }
1419
1420         if (s_flags & MS_RDONLY) {
1421                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1422                        sb->s_id);
1423                 sb->s_flags &= ~MS_RDONLY;
1424         }
1425 #ifdef CONFIG_QUOTA
1426         /* Needed for iput() to work correctly and not trash data */
1427         sb->s_flags |= MS_ACTIVE;
1428         /* Turn on quotas so that they are updated correctly */
1429         for (i = 0; i < MAXQUOTAS; i++) {
1430                 if (EXT3_SB(sb)->s_qf_names[i]) {
1431                         int ret = ext3_quota_on_mount(sb, i);
1432                         if (ret < 0)
1433                                 printk(KERN_ERR
1434                                         "EXT3-fs: Cannot turn on journaled "
1435                                         "quota: error %d\n", ret);
1436                 }
1437         }
1438 #endif
1439
1440         while (es->s_last_orphan) {
1441                 struct inode *inode;
1442
1443                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1444                 if (IS_ERR(inode)) {
1445                         es->s_last_orphan = 0;
1446                         break;
1447                 }
1448
1449                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1450                 vfs_dq_init(inode);
1451                 if (inode->i_nlink) {
1452                         printk(KERN_DEBUG
1453                                 "%s: truncating inode %lu to %Ld bytes\n",
1454                                 __func__, inode->i_ino, inode->i_size);
1455                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1456                                   inode->i_ino, inode->i_size);
1457                         ext3_truncate(inode);
1458                         nr_truncates++;
1459                 } else {
1460                         printk(KERN_DEBUG
1461                                 "%s: deleting unreferenced inode %lu\n",
1462                                 __func__, inode->i_ino);
1463                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1464                                   inode->i_ino);
1465                         nr_orphans++;
1466                 }
1467                 iput(inode);  /* The delete magic happens here! */
1468         }
1469
1470 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1471
1472         if (nr_orphans)
1473                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1474                        sb->s_id, PLURAL(nr_orphans));
1475         if (nr_truncates)
1476                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1477                        sb->s_id, PLURAL(nr_truncates));
1478 #ifdef CONFIG_QUOTA
1479         /* Turn quotas off */
1480         for (i = 0; i < MAXQUOTAS; i++) {
1481                 if (sb_dqopt(sb)->files[i])
1482                         vfs_quota_off(sb, i, 0);
1483         }
1484 #endif
1485         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1486 }
1487
1488 /*
1489  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1490  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1491  * We need to be 1 filesystem block less than the 2^32 sector limit.
1492  */
1493 static loff_t ext3_max_size(int bits)
1494 {
1495         loff_t res = EXT3_NDIR_BLOCKS;
1496         int meta_blocks;
1497         loff_t upper_limit;
1498
1499         /* This is calculated to be the largest file size for a
1500          * dense, file such that the total number of
1501          * sectors in the file, including data and all indirect blocks,
1502          * does not exceed 2^32 -1
1503          * __u32 i_blocks representing the total number of
1504          * 512 bytes blocks of the file
1505          */
1506         upper_limit = (1LL << 32) - 1;
1507
1508         /* total blocks in file system block size */
1509         upper_limit >>= (bits - 9);
1510
1511
1512         /* indirect blocks */
1513         meta_blocks = 1;
1514         /* double indirect blocks */
1515         meta_blocks += 1 + (1LL << (bits-2));
1516         /* tripple indirect blocks */
1517         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1518
1519         upper_limit -= meta_blocks;
1520         upper_limit <<= bits;
1521
1522         res += 1LL << (bits-2);
1523         res += 1LL << (2*(bits-2));
1524         res += 1LL << (3*(bits-2));
1525         res <<= bits;
1526         if (res > upper_limit)
1527                 res = upper_limit;
1528
1529         if (res > MAX_LFS_FILESIZE)
1530                 res = MAX_LFS_FILESIZE;
1531
1532         return res;
1533 }
1534
1535 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1536                                     ext3_fsblk_t logic_sb_block,
1537                                     int nr)
1538 {
1539         struct ext3_sb_info *sbi = EXT3_SB(sb);
1540         unsigned long bg, first_meta_bg;
1541         int has_super = 0;
1542
1543         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1544
1545         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1546             nr < first_meta_bg)
1547                 return (logic_sb_block + nr + 1);
1548         bg = sbi->s_desc_per_block * nr;
1549         if (ext3_bg_has_super(sb, bg))
1550                 has_super = 1;
1551         return (has_super + ext3_group_first_block_no(sb, bg));
1552 }
1553
1554
1555 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1556 {
1557         struct buffer_head * bh;
1558         struct ext3_super_block *es = NULL;
1559         struct ext3_sb_info *sbi;
1560         ext3_fsblk_t block;
1561         ext3_fsblk_t sb_block = get_sb_block(&data);
1562         ext3_fsblk_t logic_sb_block;
1563         unsigned long offset = 0;
1564         unsigned int journal_inum = 0;
1565         unsigned long journal_devnum = 0;
1566         unsigned long def_mount_opts;
1567         struct inode *root;
1568         int blocksize;
1569         int hblock;
1570         int db_count;
1571         int i;
1572         int needs_recovery;
1573         int ret = -EINVAL;
1574         __le32 features;
1575         int err;
1576
1577         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1578         if (!sbi)
1579                 return -ENOMEM;
1580
1581         sbi->s_blockgroup_lock =
1582                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1583         if (!sbi->s_blockgroup_lock) {
1584                 kfree(sbi);
1585                 return -ENOMEM;
1586         }
1587         sb->s_fs_info = sbi;
1588         sbi->s_mount_opt = 0;
1589         sbi->s_resuid = EXT3_DEF_RESUID;
1590         sbi->s_resgid = EXT3_DEF_RESGID;
1591         sbi->s_sb_block = sb_block;
1592
1593         unlock_kernel();
1594
1595         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1596         if (!blocksize) {
1597                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1598                 goto out_fail;
1599         }
1600
1601         /*
1602          * The ext3 superblock will not be buffer aligned for other than 1kB
1603          * block sizes.  We need to calculate the offset from buffer start.
1604          */
1605         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1606                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1607                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1608         } else {
1609                 logic_sb_block = sb_block;
1610         }
1611
1612         if (!(bh = sb_bread(sb, logic_sb_block))) {
1613                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1614                 goto out_fail;
1615         }
1616         /*
1617          * Note: s_es must be initialized as soon as possible because
1618          *       some ext3 macro-instructions depend on its value
1619          */
1620         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1621         sbi->s_es = es;
1622         sb->s_magic = le16_to_cpu(es->s_magic);
1623         if (sb->s_magic != EXT3_SUPER_MAGIC)
1624                 goto cantfind_ext3;
1625
1626         /* Set defaults before we parse the mount options */
1627         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1628         if (def_mount_opts & EXT3_DEFM_DEBUG)
1629                 set_opt(sbi->s_mount_opt, DEBUG);
1630         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1631                 set_opt(sbi->s_mount_opt, GRPID);
1632         if (def_mount_opts & EXT3_DEFM_UID16)
1633                 set_opt(sbi->s_mount_opt, NO_UID32);
1634 #ifdef CONFIG_EXT3_FS_XATTR
1635         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1636                 set_opt(sbi->s_mount_opt, XATTR_USER);
1637 #endif
1638 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1639         if (def_mount_opts & EXT3_DEFM_ACL)
1640                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1641 #endif
1642         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1643                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1644         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1645                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1646         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1647                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1648
1649         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1650                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1651         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1652                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1653         else
1654                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1655
1656         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1657         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1658
1659         set_opt(sbi->s_mount_opt, RESERVATION);
1660
1661         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1662                             NULL, 0))
1663                 goto failed_mount;
1664
1665         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1666                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1667
1668         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1669             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1670              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1671              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1672                 printk(KERN_WARNING
1673                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1674                        "running e2fsck is recommended\n");
1675         /*
1676          * Check feature flags regardless of the revision level, since we
1677          * previously didn't change the revision level when setting the flags,
1678          * so there is a chance incompat flags are set on a rev 0 filesystem.
1679          */
1680         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1681         if (features) {
1682                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1683                        "unsupported optional features (%x).\n",
1684                        sb->s_id, le32_to_cpu(features));
1685                 goto failed_mount;
1686         }
1687         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1688         if (!(sb->s_flags & MS_RDONLY) && features) {
1689                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1690                        "unsupported optional features (%x).\n",
1691                        sb->s_id, le32_to_cpu(features));
1692                 goto failed_mount;
1693         }
1694         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1695
1696         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1697             blocksize > EXT3_MAX_BLOCK_SIZE) {
1698                 printk(KERN_ERR
1699                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1700                        blocksize, sb->s_id);
1701                 goto failed_mount;
1702         }
1703
1704         hblock = bdev_logical_block_size(sb->s_bdev);
1705         if (sb->s_blocksize != blocksize) {
1706                 /*
1707                  * Make sure the blocksize for the filesystem is larger
1708                  * than the hardware sectorsize for the machine.
1709                  */
1710                 if (blocksize < hblock) {
1711                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1712                                "device blocksize %d.\n", blocksize, hblock);
1713                         goto failed_mount;
1714                 }
1715
1716                 brelse (bh);
1717                 if (!sb_set_blocksize(sb, blocksize)) {
1718                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1719                                 blocksize);
1720                         goto out_fail;
1721                 }
1722                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1723                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1724                 bh = sb_bread(sb, logic_sb_block);
1725                 if (!bh) {
1726                         printk(KERN_ERR
1727                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1728                         goto failed_mount;
1729                 }
1730                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1731                 sbi->s_es = es;
1732                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1733                         printk (KERN_ERR
1734                                 "EXT3-fs: Magic mismatch, very weird !\n");
1735                         goto failed_mount;
1736                 }
1737         }
1738
1739         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1740
1741         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1742                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1743                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1744         } else {
1745                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1746                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1747                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1748                     (!is_power_of_2(sbi->s_inode_size)) ||
1749                     (sbi->s_inode_size > blocksize)) {
1750                         printk (KERN_ERR
1751                                 "EXT3-fs: unsupported inode size: %d\n",
1752                                 sbi->s_inode_size);
1753                         goto failed_mount;
1754                 }
1755         }
1756         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1757                                    le32_to_cpu(es->s_log_frag_size);
1758         if (blocksize != sbi->s_frag_size) {
1759                 printk(KERN_ERR
1760                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1761                        sbi->s_frag_size, blocksize);
1762                 goto failed_mount;
1763         }
1764         sbi->s_frags_per_block = 1;
1765         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1766         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1767         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1768         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1769                 goto cantfind_ext3;
1770         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1771         if (sbi->s_inodes_per_block == 0)
1772                 goto cantfind_ext3;
1773         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1774                                         sbi->s_inodes_per_block;
1775         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1776         sbi->s_sbh = bh;
1777         sbi->s_mount_state = le16_to_cpu(es->s_state);
1778         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1779         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1780         for (i=0; i < 4; i++)
1781                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1782         sbi->s_def_hash_version = es->s_def_hash_version;
1783         i = le32_to_cpu(es->s_flags);
1784         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1785                 sbi->s_hash_unsigned = 3;
1786         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1787 #ifdef __CHAR_UNSIGNED__
1788                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1789                 sbi->s_hash_unsigned = 3;
1790 #else
1791                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1792 #endif
1793         }
1794
1795         if (sbi->s_blocks_per_group > blocksize * 8) {
1796                 printk (KERN_ERR
1797                         "EXT3-fs: #blocks per group too big: %lu\n",
1798                         sbi->s_blocks_per_group);
1799                 goto failed_mount;
1800         }
1801         if (sbi->s_frags_per_group > blocksize * 8) {
1802                 printk (KERN_ERR
1803                         "EXT3-fs: #fragments per group too big: %lu\n",
1804                         sbi->s_frags_per_group);
1805                 goto failed_mount;
1806         }
1807         if (sbi->s_inodes_per_group > blocksize * 8) {
1808                 printk (KERN_ERR
1809                         "EXT3-fs: #inodes per group too big: %lu\n",
1810                         sbi->s_inodes_per_group);
1811                 goto failed_mount;
1812         }
1813
1814         if (le32_to_cpu(es->s_blocks_count) >
1815                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1816                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1817                         " too large to mount safely\n", sb->s_id);
1818                 if (sizeof(sector_t) < 8)
1819                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBDAF not "
1820                                         "enabled\n");
1821                 goto failed_mount;
1822         }
1823
1824         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1825                 goto cantfind_ext3;
1826         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1827                                le32_to_cpu(es->s_first_data_block) - 1)
1828                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1829         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1830                    EXT3_DESC_PER_BLOCK(sb);
1831         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1832                                     GFP_KERNEL);
1833         if (sbi->s_group_desc == NULL) {
1834                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1835                 goto failed_mount;
1836         }
1837
1838         bgl_lock_init(sbi->s_blockgroup_lock);
1839
1840         for (i = 0; i < db_count; i++) {
1841                 block = descriptor_loc(sb, logic_sb_block, i);
1842                 sbi->s_group_desc[i] = sb_bread(sb, block);
1843                 if (!sbi->s_group_desc[i]) {
1844                         printk (KERN_ERR "EXT3-fs: "
1845                                 "can't read group descriptor %d\n", i);
1846                         db_count = i;
1847                         goto failed_mount2;
1848                 }
1849         }
1850         if (!ext3_check_descriptors (sb)) {
1851                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1852                 goto failed_mount2;
1853         }
1854         sbi->s_gdb_count = db_count;
1855         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1856         spin_lock_init(&sbi->s_next_gen_lock);
1857
1858         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1859                         ext3_count_free_blocks(sb));
1860         if (!err) {
1861                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1862                                 ext3_count_free_inodes(sb));
1863         }
1864         if (!err) {
1865                 err = percpu_counter_init(&sbi->s_dirs_counter,
1866                                 ext3_count_dirs(sb));
1867         }
1868         if (err) {
1869                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1870                 goto failed_mount3;
1871         }
1872
1873         /* per fileystem reservation list head & lock */
1874         spin_lock_init(&sbi->s_rsv_window_lock);
1875         sbi->s_rsv_window_root = RB_ROOT;
1876         /* Add a single, static dummy reservation to the start of the
1877          * reservation window list --- it gives us a placeholder for
1878          * append-at-start-of-list which makes the allocation logic
1879          * _much_ simpler. */
1880         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1881         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1882         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1883         sbi->s_rsv_window_head.rsv_goal_size = 0;
1884         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1885
1886         /*
1887          * set up enough so that it can read an inode
1888          */
1889         sb->s_op = &ext3_sops;
1890         sb->s_export_op = &ext3_export_ops;
1891         sb->s_xattr = ext3_xattr_handlers;
1892 #ifdef CONFIG_QUOTA
1893         sb->s_qcop = &ext3_qctl_operations;
1894         sb->dq_op = &ext3_quota_operations;
1895 #endif
1896         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1897
1898         sb->s_root = NULL;
1899
1900         needs_recovery = (es->s_last_orphan != 0 ||
1901                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1902                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1903
1904         /*
1905          * The first inode we look at is the journal inode.  Don't try
1906          * root first: it may be modified in the journal!
1907          */
1908         if (!test_opt(sb, NOLOAD) &&
1909             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1910                 if (ext3_load_journal(sb, es, journal_devnum))
1911                         goto failed_mount3;
1912         } else if (journal_inum) {
1913                 if (ext3_create_journal(sb, es, journal_inum))
1914                         goto failed_mount3;
1915         } else {
1916                 if (!silent)
1917                         printk (KERN_ERR
1918                                 "ext3: No journal on filesystem on %s\n",
1919                                 sb->s_id);
1920                 goto failed_mount3;
1921         }
1922
1923         /* We have now updated the journal if required, so we can
1924          * validate the data journaling mode. */
1925         switch (test_opt(sb, DATA_FLAGS)) {
1926         case 0:
1927                 /* No mode set, assume a default based on the journal
1928                    capabilities: ORDERED_DATA if the journal can
1929                    cope, else JOURNAL_DATA */
1930                 if (journal_check_available_features
1931                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1932                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1933                 else
1934                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1935                 break;
1936
1937         case EXT3_MOUNT_ORDERED_DATA:
1938         case EXT3_MOUNT_WRITEBACK_DATA:
1939                 if (!journal_check_available_features
1940                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1941                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1942                                "requested data journaling mode\n");
1943                         goto failed_mount4;
1944                 }
1945         default:
1946                 break;
1947         }
1948
1949         if (test_opt(sb, NOBH)) {
1950                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1951                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1952                                 "its supported only with writeback mode\n");
1953                         clear_opt(sbi->s_mount_opt, NOBH);
1954                 }
1955         }
1956         /*
1957          * The journal_load will have done any necessary log recovery,
1958          * so we can safely mount the rest of the filesystem now.
1959          */
1960
1961         root = ext3_iget(sb, EXT3_ROOT_INO);
1962         if (IS_ERR(root)) {
1963                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1964                 ret = PTR_ERR(root);
1965                 goto failed_mount4;
1966         }
1967         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1968                 iput(root);
1969                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1970                 goto failed_mount4;
1971         }
1972         sb->s_root = d_alloc_root(root);
1973         if (!sb->s_root) {
1974                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1975                 iput(root);
1976                 ret = -ENOMEM;
1977                 goto failed_mount4;
1978         }
1979
1980         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1981         /*
1982          * akpm: core read_super() calls in here with the superblock locked.
1983          * That deadlocks, because orphan cleanup needs to lock the superblock
1984          * in numerous places.  Here we just pop the lock - it's relatively
1985          * harmless, because we are now ready to accept write_super() requests,
1986          * and aviro says that's the only reason for hanging onto the
1987          * superblock lock.
1988          */
1989         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1990         ext3_orphan_cleanup(sb, es);
1991         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1992         if (needs_recovery)
1993                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1994         ext3_mark_recovery_complete(sb, es);
1995         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1996                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1997                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1998                 "writeback");
1999
2000         lock_kernel();
2001         return 0;
2002
2003 cantfind_ext3:
2004         if (!silent)
2005                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
2006                        sb->s_id);
2007         goto failed_mount;
2008
2009 failed_mount4:
2010         journal_destroy(sbi->s_journal);
2011 failed_mount3:
2012         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2013         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2014         percpu_counter_destroy(&sbi->s_dirs_counter);
2015 failed_mount2:
2016         for (i = 0; i < db_count; i++)
2017                 brelse(sbi->s_group_desc[i]);
2018         kfree(sbi->s_group_desc);
2019 failed_mount:
2020 #ifdef CONFIG_QUOTA
2021         for (i = 0; i < MAXQUOTAS; i++)
2022                 kfree(sbi->s_qf_names[i]);
2023 #endif
2024         ext3_blkdev_remove(sbi);
2025         brelse(bh);
2026 out_fail:
2027         sb->s_fs_info = NULL;
2028         kfree(sbi->s_blockgroup_lock);
2029         kfree(sbi);
2030         lock_kernel();
2031         return ret;
2032 }
2033
2034 /*
2035  * Setup any per-fs journal parameters now.  We'll do this both on
2036  * initial mount, once the journal has been initialised but before we've
2037  * done any recovery; and again on any subsequent remount.
2038  */
2039 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2040 {
2041         struct ext3_sb_info *sbi = EXT3_SB(sb);
2042
2043         if (sbi->s_commit_interval)
2044                 journal->j_commit_interval = sbi->s_commit_interval;
2045         /* We could also set up an ext3-specific default for the commit
2046          * interval here, but for now we'll just fall back to the jbd
2047          * default. */
2048
2049         spin_lock(&journal->j_state_lock);
2050         if (test_opt(sb, BARRIER))
2051                 journal->j_flags |= JFS_BARRIER;
2052         else
2053                 journal->j_flags &= ~JFS_BARRIER;
2054         if (test_opt(sb, DATA_ERR_ABORT))
2055                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2056         else
2057                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2058         spin_unlock(&journal->j_state_lock);
2059 }
2060
2061 static journal_t *ext3_get_journal(struct super_block *sb,
2062                                    unsigned int journal_inum)
2063 {
2064         struct inode *journal_inode;
2065         journal_t *journal;
2066
2067         /* First, test for the existence of a valid inode on disk.  Bad
2068          * things happen if we iget() an unused inode, as the subsequent
2069          * iput() will try to delete it. */
2070
2071         journal_inode = ext3_iget(sb, journal_inum);
2072         if (IS_ERR(journal_inode)) {
2073                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2074                 return NULL;
2075         }
2076         if (!journal_inode->i_nlink) {
2077                 make_bad_inode(journal_inode);
2078                 iput(journal_inode);
2079                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2080                 return NULL;
2081         }
2082
2083         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2084                   journal_inode, journal_inode->i_size);
2085         if (!S_ISREG(journal_inode->i_mode)) {
2086                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2087                 iput(journal_inode);
2088                 return NULL;
2089         }
2090
2091         journal = journal_init_inode(journal_inode);
2092         if (!journal) {
2093                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2094                 iput(journal_inode);
2095                 return NULL;
2096         }
2097         journal->j_private = sb;
2098         ext3_init_journal_params(sb, journal);
2099         return journal;
2100 }
2101
2102 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2103                                        dev_t j_dev)
2104 {
2105         struct buffer_head * bh;
2106         journal_t *journal;
2107         ext3_fsblk_t start;
2108         ext3_fsblk_t len;
2109         int hblock, blocksize;
2110         ext3_fsblk_t sb_block;
2111         unsigned long offset;
2112         struct ext3_super_block * es;
2113         struct block_device *bdev;
2114
2115         bdev = ext3_blkdev_get(j_dev);
2116         if (bdev == NULL)
2117                 return NULL;
2118
2119         if (bd_claim(bdev, sb)) {
2120                 printk(KERN_ERR
2121                         "EXT3: failed to claim external journal device.\n");
2122                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2123                 return NULL;
2124         }
2125
2126         blocksize = sb->s_blocksize;
2127         hblock = bdev_logical_block_size(bdev);
2128         if (blocksize < hblock) {
2129                 printk(KERN_ERR
2130                         "EXT3-fs: blocksize too small for journal device.\n");
2131                 goto out_bdev;
2132         }
2133
2134         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2135         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2136         set_blocksize(bdev, blocksize);
2137         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2138                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2139                        "external journal\n");
2140                 goto out_bdev;
2141         }
2142
2143         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2144         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2145             !(le32_to_cpu(es->s_feature_incompat) &
2146               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2147                 printk(KERN_ERR "EXT3-fs: external journal has "
2148                                         "bad superblock\n");
2149                 brelse(bh);
2150                 goto out_bdev;
2151         }
2152
2153         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2154                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2155                 brelse(bh);
2156                 goto out_bdev;
2157         }
2158
2159         len = le32_to_cpu(es->s_blocks_count);
2160         start = sb_block + 1;
2161         brelse(bh);     /* we're done with the superblock */
2162
2163         journal = journal_init_dev(bdev, sb->s_bdev,
2164                                         start, len, blocksize);
2165         if (!journal) {
2166                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2167                 goto out_bdev;
2168         }
2169         journal->j_private = sb;
2170         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2171         wait_on_buffer(journal->j_sb_buffer);
2172         if (!buffer_uptodate(journal->j_sb_buffer)) {
2173                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2174                 goto out_journal;
2175         }
2176         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2177                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2178                                         "user (unsupported) - %d\n",
2179                         be32_to_cpu(journal->j_superblock->s_nr_users));
2180                 goto out_journal;
2181         }
2182         EXT3_SB(sb)->journal_bdev = bdev;
2183         ext3_init_journal_params(sb, journal);
2184         return journal;
2185 out_journal:
2186         journal_destroy(journal);
2187 out_bdev:
2188         ext3_blkdev_put(bdev);
2189         return NULL;
2190 }
2191
2192 static int ext3_load_journal(struct super_block *sb,
2193                              struct ext3_super_block *es,
2194                              unsigned long journal_devnum)
2195 {
2196         journal_t *journal;
2197         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2198         dev_t journal_dev;
2199         int err = 0;
2200         int really_read_only;
2201
2202         if (journal_devnum &&
2203             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2204                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2205                         "numbers have changed\n");
2206                 journal_dev = new_decode_dev(journal_devnum);
2207         } else
2208                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2209
2210         really_read_only = bdev_read_only(sb->s_bdev);
2211
2212         /*
2213          * Are we loading a blank journal or performing recovery after a
2214          * crash?  For recovery, we need to check in advance whether we
2215          * can get read-write access to the device.
2216          */
2217
2218         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2219                 if (sb->s_flags & MS_RDONLY) {
2220                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2221                                         "required on readonly filesystem.\n");
2222                         if (really_read_only) {
2223                                 printk(KERN_ERR "EXT3-fs: write access "
2224                                         "unavailable, cannot proceed.\n");
2225                                 return -EROFS;
2226                         }
2227                         printk (KERN_INFO "EXT3-fs: write access will "
2228                                         "be enabled during recovery.\n");
2229                 }
2230         }
2231
2232         if (journal_inum && journal_dev) {
2233                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2234                        "and inode journals!\n");
2235                 return -EINVAL;
2236         }
2237
2238         if (journal_inum) {
2239                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2240                         return -EINVAL;
2241         } else {
2242                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2243                         return -EINVAL;
2244         }
2245
2246         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2247                 err = journal_update_format(journal);
2248                 if (err)  {
2249                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2250                         journal_destroy(journal);
2251                         return err;
2252                 }
2253         }
2254
2255         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2256                 err = journal_wipe(journal, !really_read_only);
2257         if (!err)
2258                 err = journal_load(journal);
2259
2260         if (err) {
2261                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2262                 journal_destroy(journal);
2263                 return err;
2264         }
2265
2266         EXT3_SB(sb)->s_journal = journal;
2267         ext3_clear_journal_err(sb, es);
2268
2269         if (journal_devnum &&
2270             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2271                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2272
2273                 /* Make sure we flush the recovery flag to disk. */
2274                 ext3_commit_super(sb, es, 1);
2275         }
2276
2277         return 0;
2278 }
2279
2280 static int ext3_create_journal(struct super_block * sb,
2281                                struct ext3_super_block * es,
2282                                unsigned int journal_inum)
2283 {
2284         journal_t *journal;
2285         int err;
2286
2287         if (sb->s_flags & MS_RDONLY) {
2288                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2289                                 "create journal.\n");
2290                 return -EROFS;
2291         }
2292
2293         journal = ext3_get_journal(sb, journal_inum);
2294         if (!journal)
2295                 return -EINVAL;
2296
2297         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2298                journal_inum);
2299
2300         err = journal_create(journal);
2301         if (err) {
2302                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2303                 journal_destroy(journal);
2304                 return -EIO;
2305         }
2306
2307         EXT3_SB(sb)->s_journal = journal;
2308
2309         ext3_update_dynamic_rev(sb);
2310         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2311         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2312
2313         es->s_journal_inum = cpu_to_le32(journal_inum);
2314
2315         /* Make sure we flush the recovery flag to disk. */
2316         ext3_commit_super(sb, es, 1);
2317
2318         return 0;
2319 }
2320
2321 static int ext3_commit_super(struct super_block *sb,
2322                                struct ext3_super_block *es,
2323                                int sync)
2324 {
2325         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2326         int error = 0;
2327
2328         if (!sbh)
2329                 return error;
2330         /*
2331          * If the file system is mounted read-only, don't update the
2332          * superblock write time.  This avoids updating the superblock
2333          * write time when we are mounting the root file system
2334          * read/only but we need to replay the journal; at that point,
2335          * for people who are east of GMT and who make their clock
2336          * tick in localtime for Windows bug-for-bug compatibility,
2337          * the clock is set in the future, and this will cause e2fsck
2338          * to complain and force a full file system check.
2339          */
2340         if (!(sb->s_flags & MS_RDONLY))
2341                 es->s_wtime = cpu_to_le32(get_seconds());
2342         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2343         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2344         BUFFER_TRACE(sbh, "marking dirty");
2345         mark_buffer_dirty(sbh);
2346         if (sync)
2347                 error = sync_dirty_buffer(sbh);
2348         return error;
2349 }
2350
2351
2352 /*
2353  * Have we just finished recovery?  If so, and if we are mounting (or
2354  * remounting) the filesystem readonly, then we will end up with a
2355  * consistent fs on disk.  Record that fact.
2356  */
2357 static void ext3_mark_recovery_complete(struct super_block * sb,
2358                                         struct ext3_super_block * es)
2359 {
2360         journal_t *journal = EXT3_SB(sb)->s_journal;
2361
2362         journal_lock_updates(journal);
2363         if (journal_flush(journal) < 0)
2364                 goto out;
2365
2366         lock_super(sb);
2367         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2368             sb->s_flags & MS_RDONLY) {
2369                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2370                 ext3_commit_super(sb, es, 1);
2371         }
2372         unlock_super(sb);
2373
2374 out:
2375         journal_unlock_updates(journal);
2376 }
2377
2378 /*
2379  * If we are mounting (or read-write remounting) a filesystem whose journal
2380  * has recorded an error from a previous lifetime, move that error to the
2381  * main filesystem now.
2382  */
2383 static void ext3_clear_journal_err(struct super_block * sb,
2384                                    struct ext3_super_block * es)
2385 {
2386         journal_t *journal;
2387         int j_errno;
2388         const char *errstr;
2389
2390         journal = EXT3_SB(sb)->s_journal;
2391
2392         /*
2393          * Now check for any error status which may have been recorded in the
2394          * journal by a prior ext3_error() or ext3_abort()
2395          */
2396
2397         j_errno = journal_errno(journal);
2398         if (j_errno) {
2399                 char nbuf[16];
2400
2401                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2402                 ext3_warning(sb, __func__, "Filesystem error recorded "
2403                              "from previous mount: %s", errstr);
2404                 ext3_warning(sb, __func__, "Marking fs in need of "
2405                              "filesystem check.");
2406
2407                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2408                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2409                 ext3_commit_super (sb, es, 1);
2410
2411                 journal_clear_err(journal);
2412         }
2413 }
2414
2415 /*
2416  * Force the running and committing transactions to commit,
2417  * and wait on the commit.
2418  */
2419 int ext3_force_commit(struct super_block *sb)
2420 {
2421         journal_t *journal;
2422         int ret;
2423
2424         if (sb->s_flags & MS_RDONLY)
2425                 return 0;
2426
2427         journal = EXT3_SB(sb)->s_journal;
2428         ret = ext3_journal_force_commit(journal);
2429         return ret;
2430 }
2431
2432 static int ext3_sync_fs(struct super_block *sb, int wait)
2433 {
2434         tid_t target;
2435
2436         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2437                 if (wait)
2438                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2439         }
2440         return 0;
2441 }
2442
2443 /*
2444  * LVM calls this function before a (read-only) snapshot is created.  This
2445  * gives us a chance to flush the journal completely and mark the fs clean.
2446  */
2447 static int ext3_freeze(struct super_block *sb)
2448 {
2449         int error = 0;
2450         journal_t *journal;
2451
2452         if (!(sb->s_flags & MS_RDONLY)) {
2453                 journal = EXT3_SB(sb)->s_journal;
2454
2455                 /* Now we set up the journal barrier. */
2456                 journal_lock_updates(journal);
2457
2458                 /*
2459                  * We don't want to clear needs_recovery flag when we failed
2460                  * to flush the journal.
2461                  */
2462                 error = journal_flush(journal);
2463                 if (error < 0)
2464                         goto out;
2465
2466                 /* Journal blocked and flushed, clear needs_recovery flag. */
2467                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2468                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2469                 if (error)
2470                         goto out;
2471         }
2472         return 0;
2473
2474 out:
2475         journal_unlock_updates(journal);
2476         return error;
2477 }
2478
2479 /*
2480  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2481  * flag here, even though the filesystem is not technically dirty yet.
2482  */
2483 static int ext3_unfreeze(struct super_block *sb)
2484 {
2485         if (!(sb->s_flags & MS_RDONLY)) {
2486                 lock_super(sb);
2487                 /* Reser the needs_recovery flag before the fs is unlocked. */
2488                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2489                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2490                 unlock_super(sb);
2491                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2492         }
2493         return 0;
2494 }
2495
2496 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2497 {
2498         struct ext3_super_block * es;
2499         struct ext3_sb_info *sbi = EXT3_SB(sb);
2500         ext3_fsblk_t n_blocks_count = 0;
2501         unsigned long old_sb_flags;
2502         struct ext3_mount_options old_opts;
2503         int err;
2504 #ifdef CONFIG_QUOTA
2505         int i;
2506 #endif
2507
2508         lock_kernel();
2509
2510         /* Store the original options */
2511         lock_super(sb);
2512         old_sb_flags = sb->s_flags;
2513         old_opts.s_mount_opt = sbi->s_mount_opt;
2514         old_opts.s_resuid = sbi->s_resuid;
2515         old_opts.s_resgid = sbi->s_resgid;
2516         old_opts.s_commit_interval = sbi->s_commit_interval;
2517 #ifdef CONFIG_QUOTA
2518         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2519         for (i = 0; i < MAXQUOTAS; i++)
2520                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2521 #endif
2522
2523         /*
2524          * Allow the "check" option to be passed as a remount option.
2525          */
2526         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2527                 err = -EINVAL;
2528                 goto restore_opts;
2529         }
2530
2531         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2532                 ext3_abort(sb, __func__, "Abort forced by user");
2533
2534         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2535                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2536
2537         es = sbi->s_es;
2538
2539         ext3_init_journal_params(sb, sbi->s_journal);
2540
2541         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2542                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2543                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2544                         err = -EROFS;
2545                         goto restore_opts;
2546                 }
2547
2548                 if (*flags & MS_RDONLY) {
2549                         /*
2550                          * First of all, the unconditional stuff we have to do
2551                          * to disable replay of the journal when we next remount
2552                          */
2553                         sb->s_flags |= MS_RDONLY;
2554
2555                         /*
2556                          * OK, test if we are remounting a valid rw partition
2557                          * readonly, and if so set the rdonly flag and then
2558                          * mark the partition as valid again.
2559                          */
2560                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2561                             (sbi->s_mount_state & EXT3_VALID_FS))
2562                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2563
2564                         /*
2565                          * We have to unlock super so that we can wait for
2566                          * transactions.
2567                          */
2568                         unlock_super(sb);
2569                         ext3_mark_recovery_complete(sb, es);
2570                         lock_super(sb);
2571                 } else {
2572                         __le32 ret;
2573                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2574                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2575                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2576                                        "remount RDWR because of unsupported "
2577                                        "optional features (%x).\n",
2578                                        sb->s_id, le32_to_cpu(ret));
2579                                 err = -EROFS;
2580                                 goto restore_opts;
2581                         }
2582
2583                         /*
2584                          * If we have an unprocessed orphan list hanging
2585                          * around from a previously readonly bdev mount,
2586                          * require a full umount/remount for now.
2587                          */
2588                         if (es->s_last_orphan) {
2589                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2590                                        "remount RDWR because of unprocessed "
2591                                        "orphan inode list.  Please "
2592                                        "umount/remount instead.\n",
2593                                        sb->s_id);
2594                                 err = -EINVAL;
2595                                 goto restore_opts;
2596                         }
2597
2598                         /*
2599                          * Mounting a RDONLY partition read-write, so reread
2600                          * and store the current valid flag.  (It may have
2601                          * been changed by e2fsck since we originally mounted
2602                          * the partition.)
2603                          */
2604                         ext3_clear_journal_err(sb, es);
2605                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2606                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2607                                 goto restore_opts;
2608                         if (!ext3_setup_super (sb, es, 0))
2609                                 sb->s_flags &= ~MS_RDONLY;
2610                 }
2611         }
2612 #ifdef CONFIG_QUOTA
2613         /* Release old quota file names */
2614         for (i = 0; i < MAXQUOTAS; i++)
2615                 if (old_opts.s_qf_names[i] &&
2616                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2617                         kfree(old_opts.s_qf_names[i]);
2618 #endif
2619         unlock_super(sb);
2620         unlock_kernel();
2621         return 0;
2622 restore_opts:
2623         sb->s_flags = old_sb_flags;
2624         sbi->s_mount_opt = old_opts.s_mount_opt;
2625         sbi->s_resuid = old_opts.s_resuid;
2626         sbi->s_resgid = old_opts.s_resgid;
2627         sbi->s_commit_interval = old_opts.s_commit_interval;
2628 #ifdef CONFIG_QUOTA
2629         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2630         for (i = 0; i < MAXQUOTAS; i++) {
2631                 if (sbi->s_qf_names[i] &&
2632                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2633                         kfree(sbi->s_qf_names[i]);
2634                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2635         }
2636 #endif
2637         unlock_super(sb);
2638         unlock_kernel();
2639         return err;
2640 }
2641
2642 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2643 {
2644         struct super_block *sb = dentry->d_sb;
2645         struct ext3_sb_info *sbi = EXT3_SB(sb);
2646         struct ext3_super_block *es = sbi->s_es;
2647         u64 fsid;
2648
2649         if (test_opt(sb, MINIX_DF)) {
2650                 sbi->s_overhead_last = 0;
2651         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2652                 unsigned long ngroups = sbi->s_groups_count, i;
2653                 ext3_fsblk_t overhead = 0;
2654                 smp_rmb();
2655
2656                 /*
2657                  * Compute the overhead (FS structures).  This is constant
2658                  * for a given filesystem unless the number of block groups
2659                  * changes so we cache the previous value until it does.
2660                  */
2661
2662                 /*
2663                  * All of the blocks before first_data_block are
2664                  * overhead
2665                  */
2666                 overhead = le32_to_cpu(es->s_first_data_block);
2667
2668                 /*
2669                  * Add the overhead attributed to the superblock and
2670                  * block group descriptors.  If the sparse superblocks
2671                  * feature is turned on, then not all groups have this.
2672                  */
2673                 for (i = 0; i < ngroups; i++) {
2674                         overhead += ext3_bg_has_super(sb, i) +
2675                                 ext3_bg_num_gdb(sb, i);
2676                         cond_resched();
2677                 }
2678
2679                 /*
2680                  * Every block group has an inode bitmap, a block
2681                  * bitmap, and an inode table.
2682                  */
2683                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2684                 sbi->s_overhead_last = overhead;
2685                 smp_wmb();
2686                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2687         }
2688
2689         buf->f_type = EXT3_SUPER_MAGIC;
2690         buf->f_bsize = sb->s_blocksize;
2691         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2692         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2693         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2694         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2695                 buf->f_bavail = 0;
2696         buf->f_files = le32_to_cpu(es->s_inodes_count);
2697         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2698         buf->f_namelen = EXT3_NAME_LEN;
2699         fsid = le64_to_cpup((void *)es->s_uuid) ^
2700                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2701         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2702         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2703         return 0;
2704 }
2705
2706 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2707  * is locked for write. Otherwise the are possible deadlocks:
2708  * Process 1                         Process 2
2709  * ext3_create()                     quota_sync()
2710  *   journal_start()                   write_dquot()
2711  *   vfs_dq_init()                       down(dqio_mutex)
2712  *     down(dqio_mutex)                    journal_start()
2713  *
2714  */
2715
2716 #ifdef CONFIG_QUOTA
2717
2718 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2719 {
2720         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2721 }
2722
2723 static int ext3_write_dquot(struct dquot *dquot)
2724 {
2725         int ret, err;
2726         handle_t *handle;
2727         struct inode *inode;
2728
2729         inode = dquot_to_inode(dquot);
2730         handle = ext3_journal_start(inode,
2731                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2732         if (IS_ERR(handle))
2733                 return PTR_ERR(handle);
2734         ret = dquot_commit(dquot);
2735         err = ext3_journal_stop(handle);
2736         if (!ret)
2737                 ret = err;
2738         return ret;
2739 }
2740
2741 static int ext3_acquire_dquot(struct dquot *dquot)
2742 {
2743         int ret, err;
2744         handle_t *handle;
2745
2746         handle = ext3_journal_start(dquot_to_inode(dquot),
2747                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2748         if (IS_ERR(handle))
2749                 return PTR_ERR(handle);
2750         ret = dquot_acquire(dquot);
2751         err = ext3_journal_stop(handle);
2752         if (!ret)
2753                 ret = err;
2754         return ret;
2755 }
2756
2757 static int ext3_release_dquot(struct dquot *dquot)
2758 {
2759         int ret, err;
2760         handle_t *handle;
2761
2762         handle = ext3_journal_start(dquot_to_inode(dquot),
2763                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2764         if (IS_ERR(handle)) {
2765                 /* Release dquot anyway to avoid endless cycle in dqput() */
2766                 dquot_release(dquot);
2767                 return PTR_ERR(handle);
2768         }
2769         ret = dquot_release(dquot);
2770         err = ext3_journal_stop(handle);
2771         if (!ret)
2772                 ret = err;
2773         return ret;
2774 }
2775
2776 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2777 {
2778         /* Are we journaling quotas? */
2779         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2780             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2781                 dquot_mark_dquot_dirty(dquot);
2782                 return ext3_write_dquot(dquot);
2783         } else {
2784                 return dquot_mark_dquot_dirty(dquot);
2785         }
2786 }
2787
2788 static int ext3_write_info(struct super_block *sb, int type)
2789 {
2790         int ret, err;
2791         handle_t *handle;
2792
2793         /* Data block + inode block */
2794         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2795         if (IS_ERR(handle))
2796                 return PTR_ERR(handle);
2797         ret = dquot_commit_info(sb, type);
2798         err = ext3_journal_stop(handle);
2799         if (!ret)
2800                 ret = err;
2801         return ret;
2802 }
2803
2804 /*
2805  * Turn on quotas during mount time - we need to find
2806  * the quota file and such...
2807  */
2808 static int ext3_quota_on_mount(struct super_block *sb, int type)
2809 {
2810         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2811                         EXT3_SB(sb)->s_jquota_fmt, type);
2812 }
2813
2814 /*
2815  * Standard function to be called on quota_on
2816  */
2817 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2818                          char *name, int remount)
2819 {
2820         int err;
2821         struct path path;
2822
2823         if (!test_opt(sb, QUOTA))
2824                 return -EINVAL;
2825         /* When remounting, no checks are needed and in fact, name is NULL */
2826         if (remount)
2827                 return vfs_quota_on(sb, type, format_id, name, remount);
2828
2829         err = kern_path(name, LOOKUP_FOLLOW, &path);
2830         if (err)
2831                 return err;
2832
2833         /* Quotafile not on the same filesystem? */
2834         if (path.mnt->mnt_sb != sb) {
2835                 path_put(&path);
2836                 return -EXDEV;
2837         }
2838         /* Journaling quota? */
2839         if (EXT3_SB(sb)->s_qf_names[type]) {
2840                 /* Quotafile not of fs root? */
2841                 if (path.dentry->d_parent != sb->s_root)
2842                         printk(KERN_WARNING
2843                                 "EXT3-fs: Quota file not on filesystem root. "
2844                                 "Journaled quota will not work.\n");
2845         }
2846
2847         /*
2848          * When we journal data on quota file, we have to flush journal to see
2849          * all updates to the file when we bypass pagecache...
2850          */
2851         if (ext3_should_journal_data(path.dentry->d_inode)) {
2852                 /*
2853                  * We don't need to lock updates but journal_flush() could
2854                  * otherwise be livelocked...
2855                  */
2856                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2857                 err = journal_flush(EXT3_SB(sb)->s_journal);
2858                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2859                 if (err) {
2860                         path_put(&path);
2861                         return err;
2862                 }
2863         }
2864
2865         err = vfs_quota_on_path(sb, type, format_id, &path);
2866         path_put(&path);
2867         return err;
2868 }
2869
2870 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2871  * acquiring the locks... As quota files are never truncated and quota code
2872  * itself serializes the operations (and noone else should touch the files)
2873  * we don't have to be afraid of races */
2874 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2875                                size_t len, loff_t off)
2876 {
2877         struct inode *inode = sb_dqopt(sb)->files[type];
2878         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2879         int err = 0;
2880         int offset = off & (sb->s_blocksize - 1);
2881         int tocopy;
2882         size_t toread;
2883         struct buffer_head *bh;
2884         loff_t i_size = i_size_read(inode);
2885
2886         if (off > i_size)
2887                 return 0;
2888         if (off+len > i_size)
2889                 len = i_size-off;
2890         toread = len;
2891         while (toread > 0) {
2892                 tocopy = sb->s_blocksize - offset < toread ?
2893                                 sb->s_blocksize - offset : toread;
2894                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2895                 if (err)
2896                         return err;
2897                 if (!bh)        /* A hole? */
2898                         memset(data, 0, tocopy);
2899                 else
2900                         memcpy(data, bh->b_data+offset, tocopy);
2901                 brelse(bh);
2902                 offset = 0;
2903                 toread -= tocopy;
2904                 data += tocopy;
2905                 blk++;
2906         }
2907         return len;
2908 }
2909
2910 /* Write to quotafile (we know the transaction is already started and has
2911  * enough credits) */
2912 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2913                                 const char *data, size_t len, loff_t off)
2914 {
2915         struct inode *inode = sb_dqopt(sb)->files[type];
2916         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2917         int err = 0;
2918         int offset = off & (sb->s_blocksize - 1);
2919         int tocopy;
2920         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2921         size_t towrite = len;
2922         struct buffer_head *bh;
2923         handle_t *handle = journal_current_handle();
2924
2925         if (!handle) {
2926                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2927                         " cancelled because transaction is not started.\n",
2928                         (unsigned long long)off, (unsigned long long)len);
2929                 return -EIO;
2930         }
2931         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2932         while (towrite > 0) {
2933                 tocopy = sb->s_blocksize - offset < towrite ?
2934                                 sb->s_blocksize - offset : towrite;
2935                 bh = ext3_bread(handle, inode, blk, 1, &err);
2936                 if (!bh)
2937                         goto out;
2938                 if (journal_quota) {
2939                         err = ext3_journal_get_write_access(handle, bh);
2940                         if (err) {
2941                                 brelse(bh);
2942                                 goto out;
2943                         }
2944                 }
2945                 lock_buffer(bh);
2946                 memcpy(bh->b_data+offset, data, tocopy);
2947                 flush_dcache_page(bh->b_page);
2948                 unlock_buffer(bh);
2949                 if (journal_quota)
2950                         err = ext3_journal_dirty_metadata(handle, bh);
2951                 else {
2952                         /* Always do at least ordered writes for quotas */
2953                         err = ext3_journal_dirty_data(handle, bh);
2954                         mark_buffer_dirty(bh);
2955                 }
2956                 brelse(bh);
2957                 if (err)
2958                         goto out;
2959                 offset = 0;
2960                 towrite -= tocopy;
2961                 data += tocopy;
2962                 blk++;
2963         }
2964 out:
2965         if (len == towrite) {
2966                 mutex_unlock(&inode->i_mutex);
2967                 return err;
2968         }
2969         if (inode->i_size < off+len-towrite) {
2970                 i_size_write(inode, off+len-towrite);
2971                 EXT3_I(inode)->i_disksize = inode->i_size;
2972         }
2973         inode->i_version++;
2974         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2975         ext3_mark_inode_dirty(handle, inode);
2976         mutex_unlock(&inode->i_mutex);
2977         return len - towrite;
2978 }
2979
2980 #endif
2981
2982 static int ext3_get_sb(struct file_system_type *fs_type,
2983         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2984 {
2985         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2986 }
2987
2988 static struct file_system_type ext3_fs_type = {
2989         .owner          = THIS_MODULE,
2990         .name           = "ext3",
2991         .get_sb         = ext3_get_sb,
2992         .kill_sb        = kill_block_super,
2993         .fs_flags       = FS_REQUIRES_DEV,
2994 };
2995
2996 static int __init init_ext3_fs(void)
2997 {
2998         int err = init_ext3_xattr();
2999         if (err)
3000                 return err;
3001         err = init_inodecache();
3002         if (err)
3003                 goto out1;
3004         err = register_filesystem(&ext3_fs_type);
3005         if (err)
3006                 goto out;
3007         return 0;
3008 out:
3009         destroy_inodecache();
3010 out1:
3011         exit_ext3_xattr();
3012         return err;
3013 }
3014
3015 static void __exit exit_ext3_fs(void)
3016 {
3017         unregister_filesystem(&ext3_fs_type);
3018         destroy_inodecache();
3019         exit_ext3_xattr();
3020 }
3021
3022 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3023 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3024 MODULE_LICENSE("GPL");
3025 module_init(init_ext3_fs)
3026 module_exit(exit_ext3_fs)