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/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/log2.h>
  39
  40 #include <asm/uaccess.h>
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44 #include "namei.h"
  45
  46 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  47   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  48 #else
  49   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  50 #endif
  51
  52 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  53                              unsigned long journal_devnum);
  54 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  55                                unsigned int);
  56 static int ext3_commit_super(struct super_block *sb,
  57                                struct ext3_super_block *es,
  58                                int sync);
  59 static void ext3_mark_recovery_complete(struct super_block * sb,
  60                                         struct ext3_super_block * es);
  61 static void ext3_clear_journal_err(struct super_block * sb,
  62                                    struct ext3_super_block * es);
  63 static int ext3_sync_fs(struct super_block *sb, int wait);
  64 static const char *ext3_decode_error(struct super_block * sb, int errno,
  65                                      char nbuf[16]);
  66 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  67 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  68 static int ext3_unfreeze(struct super_block *sb);
  69 static int ext3_freeze(struct super_block *sb);
  70
  71 /*
  72  * Wrappers for journal_start/end.
  73  *
  74  * The only special thing we need to do here is to make sure that all
  75  * journal_end calls result in the superblock being marked dirty, so
  76  * that sync() will call the filesystem's write_super callback if
  77  * appropriate.
  78  */
  79 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  80 {
  81         journal_t *journal;
  82
  83         if (sb->s_flags & MS_RDONLY)
  84                 return ERR_PTR(-EROFS);
  85
  86         /* Special case here: if the journal has aborted behind our
  87          * backs (eg. EIO in the commit thread), then we still need to
  88          * take the FS itself readonly cleanly. */
  89         journal = EXT3_SB(sb)->s_journal;
  90         if (is_journal_aborted(journal)) {
  91                 ext3_abort(sb, __func__,
  92                            "Detected aborted journal");
  93                 return ERR_PTR(-EROFS);
  94         }
  95
  96         return journal_start(journal, nblocks);
  97 }
  98
  99 /*
 100  * The only special thing we need to do here is to make sure that all
 101  * journal_stop calls result in the superblock being marked dirty, so
 102  * that sync() will call the filesystem's write_super callback if
 103  * appropriate.
 104  */
 105 int __ext3_journal_stop(const char *where, handle_t *handle)
 106 {
 107         struct super_block *sb;
 108         int err;
 109         int rc;
 110
 111         sb = handle->h_transaction->t_journal->j_private;
 112         err = handle->h_err;
 113         rc = journal_stop(handle);
 114
 115         if (!err)
 116                 err = rc;
 117         if (err)
 118                 __ext3_std_error(sb, where, err);
 119         return err;
 120 }
 121
 122 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 123                 struct buffer_head *bh, handle_t *handle, int err)
 124 {
 125         char nbuf[16];
 126         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 127
 128         if (bh)
 129                 BUFFER_TRACE(bh, "abort");
 130
 131         if (!handle->h_err)
 132                 handle->h_err = err;
 133
 134         if (is_handle_aborted(handle))
 135                 return;
 136
 137         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 138                 caller, errstr, err_fn);
 139
 140         journal_abort_handle(handle);
 141 }
 142
 143 void ext3_msg(struct super_block *sb, const char *prefix,
 144                 const char *fmt, ...)
 145 {
 146         struct va_format vaf;
 147         va_list args;
 148
 149         va_start(args, fmt);
 150
 151         vaf.fmt = fmt;
 152         vaf.va = &args;
 153
 154         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 155
 156         va_end(args);
 157 }
 158
 159 /* Deal with the reporting of failure conditions on a filesystem such as
 160  * inconsistencies detected or read IO failures.
 161  *
 162  * On ext2, we can store the error state of the filesystem in the
 163  * superblock.  That is not possible on ext3, because we may have other
 164  * write ordering constraints on the superblock which prevent us from
 165  * writing it out straight away; and given that the journal is about to
 166  * be aborted, we can't rely on the current, or future, transactions to
 167  * write out the superblock safely.
 168  *
 169  * We'll just use the journal_abort() error code to record an error in
 170  * the journal instead.  On recovery, the journal will complain about
 171  * that error until we've noted it down and cleared it.
 172  */
 173
 174 static void ext3_handle_error(struct super_block *sb)
 175 {
 176         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 177
 178         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 179         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 180
 181         if (sb->s_flags & MS_RDONLY)
 182                 return;
 183
 184         if (!test_opt (sb, ERRORS_CONT)) {
 185                 journal_t *journal = EXT3_SB(sb)->s_journal;
 186
 187                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 188                 if (journal)
 189                         journal_abort(journal, -EIO);
 190         }
 191         if (test_opt (sb, ERRORS_RO)) {
 192                 ext3_msg(sb, KERN_CRIT,
 193                         "error: remounting filesystem read-only");
 194                 sb->s_flags |= MS_RDONLY;
 195         }
 196         ext3_commit_super(sb, es, 1);
 197         if (test_opt(sb, ERRORS_PANIC))
 198                 panic("EXT3-fs (%s): panic forced after error\n",
 199                         sb->s_id);
 200 }
 201
 202 void ext3_error(struct super_block *sb, const char *function,
 203                 const char *fmt, ...)
 204 {
 205         struct va_format vaf;
 206         va_list args;
 207
 208         va_start(args, fmt);
 209
 210         vaf.fmt = fmt;
 211         vaf.va = &args;
 212
 213         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 214                sb->s_id, function, &vaf);
 215
 216         va_end(args);
 217
 218         ext3_handle_error(sb);
 219 }
 220
 221 static const char *ext3_decode_error(struct super_block * sb, int errno,
 222                                      char nbuf[16])
 223 {
 224         char *errstr = NULL;
 225
 226         switch (errno) {
 227         case -EIO:
 228                 errstr = "IO failure";
 229                 break;
 230         case -ENOMEM:
 231                 errstr = "Out of memory";
 232                 break;
 233         case -EROFS:
 234                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 235                         errstr = "Journal has aborted";
 236                 else
 237                         errstr = "Readonly filesystem";
 238                 break;
 239         default:
 240                 /* If the caller passed in an extra buffer for unknown
 241                  * errors, textualise them now.  Else we just return
 242                  * NULL. */
 243                 if (nbuf) {
 244                         /* Check for truncated error codes... */
 245                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 246                                 errstr = nbuf;
 247                 }
 248                 break;
 249         }
 250
 251         return errstr;
 252 }
 253
 254 /* __ext3_std_error decodes expected errors from journaling functions
 255  * automatically and invokes the appropriate error response.  */
 256
 257 void __ext3_std_error (struct super_block * sb, const char * function,
 258                        int errno)
 259 {
 260         char nbuf[16];
 261         const char *errstr;
 262
 263         /* Special case: if the error is EROFS, and we're not already
 264          * inside a transaction, then there's really no point in logging
 265          * an error. */
 266         if (errno == -EROFS && journal_current_handle() == NULL &&
 267             (sb->s_flags & MS_RDONLY))
 268                 return;
 269
 270         errstr = ext3_decode_error(sb, errno, nbuf);
 271         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 272
 273         ext3_handle_error(sb);
 274 }
 275
 276 /*
 277  * ext3_abort is a much stronger failure handler than ext3_error.  The
 278  * abort function may be used to deal with unrecoverable failures such
 279  * as journal IO errors or ENOMEM at a critical moment in log management.
 280  *
 281  * We unconditionally force the filesystem into an ABORT|READONLY state,
 282  * unless the error response on the fs has been set to panic in which
 283  * case we take the easy way out and panic immediately.
 284  */
 285
 286 void ext3_abort(struct super_block *sb, const char *function,
 287                  const char *fmt, ...)
 288 {
 289         struct va_format vaf;
 290         va_list args;
 291
 292         va_start(args, fmt);
 293
 294         vaf.fmt = fmt;
 295         vaf.va = &args;
 296
 297         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 298                sb->s_id, function, &vaf);
 299
 300         va_end(args);
 301
 302         if (test_opt(sb, ERRORS_PANIC))
 303                 panic("EXT3-fs: panic from previous error\n");
 304
 305         if (sb->s_flags & MS_RDONLY)
 306                 return;
 307
 308         ext3_msg(sb, KERN_CRIT,
 309                 "error: remounting filesystem read-only");
 310         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 311         sb->s_flags |= MS_RDONLY;
 312         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 313         if (EXT3_SB(sb)->s_journal)
 314                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 315 }
 316
 317 void ext3_warning(struct super_block *sb, const char *function,
 318                   const char *fmt, ...)
 319 {
 320         struct va_format vaf;
 321         va_list args;
 322
 323         va_start(args, fmt);
 324
 325         vaf.fmt = fmt;
 326         vaf.va = &args;
 327
 328         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 329                sb->s_id, function, &vaf);
 330
 331         va_end(args);
 332 }
 333
 334 void ext3_update_dynamic_rev(struct super_block *sb)
 335 {
 336         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 337
 338         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 339                 return;
 340
 341         ext3_msg(sb, KERN_WARNING,
 342                 "warning: updating to rev %d because of "
 343                 "new feature flag, running e2fsck is recommended",
 344                 EXT3_DYNAMIC_REV);
 345
 346         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 347         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 348         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 349         /* leave es->s_feature_*compat flags alone */
 350         /* es->s_uuid will be set by e2fsck if empty */
 351
 352         /*
 353          * The rest of the superblock fields should be zero, and if not it
 354          * means they are likely already in use, so leave them alone.  We
 355          * can leave it up to e2fsck to clean up any inconsistencies there.
 356          */
 357 }
 358
 359 /*
 360  * Open the external journal device
 361  */
 362 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 363 {
 364         struct block_device *bdev;
 365         char b[BDEVNAME_SIZE];
 366
 367         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 368         if (IS_ERR(bdev))
 369                 goto fail;
 370         return bdev;
 371
 372 fail:
 373         ext3_msg(sb, "error: failed to open journal device %s: %ld",
 374                 __bdevname(dev, b), PTR_ERR(bdev));
 375
 376         return NULL;
 377 }
 378
 379 /*
 380  * Release the journal device
 381  */
 382 static int ext3_blkdev_put(struct block_device *bdev)
 383 {
 384         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 385 }
 386
 387 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 388 {
 389         struct block_device *bdev;
 390         int ret = -ENODEV;
 391
 392         bdev = sbi->journal_bdev;
 393         if (bdev) {
 394                 ret = ext3_blkdev_put(bdev);
 395                 sbi->journal_bdev = NULL;
 396         }
 397         return ret;
 398 }
 399
 400 static inline struct inode *orphan_list_entry(struct list_head *l)
 401 {
 402         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 403 }
 404
 405 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 406 {
 407         struct list_head *l;
 408
 409         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 410                le32_to_cpu(sbi->s_es->s_last_orphan));
 411
 412         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 413         list_for_each(l, &sbi->s_orphan) {
 414                 struct inode *inode = orphan_list_entry(l);
 415                 ext3_msg(sb, KERN_ERR, "  "
 416                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 417                        inode->i_sb->s_id, inode->i_ino, inode,
 418                        inode->i_mode, inode->i_nlink,
 419                        NEXT_ORPHAN(inode));
 420         }
 421 }
 422
 423 static void ext3_put_super (struct super_block * sb)
 424 {
 425         struct ext3_sb_info *sbi = EXT3_SB(sb);
 426         struct ext3_super_block *es = sbi->s_es;
 427         int i, err;
 428
 429         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 430         ext3_xattr_put_super(sb);
 431         err = journal_destroy(sbi->s_journal);
 432         sbi->s_journal = NULL;
 433         if (err < 0)
 434                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 435
 436         if (!(sb->s_flags & MS_RDONLY)) {
 437                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 438                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 439                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 440                 mark_buffer_dirty(sbi->s_sbh);
 441                 ext3_commit_super(sb, es, 1);
 442         }
 443
 444         for (i = 0; i < sbi->s_gdb_count; i++)
 445                 brelse(sbi->s_group_desc[i]);
 446         kfree(sbi->s_group_desc);
 447         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 448         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 449         percpu_counter_destroy(&sbi->s_dirs_counter);
 450         brelse(sbi->s_sbh);
 451 #ifdef CONFIG_QUOTA
 452         for (i = 0; i < MAXQUOTAS; i++)
 453                 kfree(sbi->s_qf_names[i]);
 454 #endif
 455
 456         /* Debugging code just in case the in-memory inode orphan list
 457          * isn't empty.  The on-disk one can be non-empty if we've
 458          * detected an error and taken the fs readonly, but the
 459          * in-memory list had better be clean by this point. */
 460         if (!list_empty(&sbi->s_orphan))
 461                 dump_orphan_list(sb, sbi);
 462         J_ASSERT(list_empty(&sbi->s_orphan));
 463
 464         invalidate_bdev(sb->s_bdev);
 465         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 466                 /*
 467                  * Invalidate the journal device's buffers.  We don't want them
 468                  * floating about in memory - the physical journal device may
 469                  * hotswapped, and it breaks the `ro-after' testing code.
 470                  */
 471                 sync_blockdev(sbi->journal_bdev);
 472                 invalidate_bdev(sbi->journal_bdev);
 473                 ext3_blkdev_remove(sbi);
 474         }
 475         sb->s_fs_info = NULL;
 476         kfree(sbi->s_blockgroup_lock);
 477         kfree(sbi);
 478 }
 479
 480 static struct kmem_cache *ext3_inode_cachep;
 481
 482 /*
 483  * Called inside transaction, so use GFP_NOFS
 484  */
 485 static struct inode *ext3_alloc_inode(struct super_block *sb)
 486 {
 487         struct ext3_inode_info *ei;
 488
 489         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 490         if (!ei)
 491                 return NULL;
 492         ei->i_block_alloc_info = NULL;
 493         ei->vfs_inode.i_version = 1;
 494         atomic_set(&ei->i_datasync_tid, 0);
 495         atomic_set(&ei->i_sync_tid, 0);
 496         return &ei->vfs_inode;
 497 }
 498
 499 static void ext3_i_callback(struct rcu_head *head)
 500 {
 501         struct inode *inode = container_of(head, struct inode, i_rcu);
 502         INIT_LIST_HEAD(&inode->i_dentry);
 503         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 504 }
 505
 506 static void ext3_destroy_inode(struct inode *inode)
 507 {
 508         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 509                 printk("EXT3 Inode %p: orphan list check failed!\n",
 510                         EXT3_I(inode));
 511                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 512                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 513                                 false);
 514                 dump_stack();
 515         }
 516         call_rcu(&inode->i_rcu, ext3_i_callback);
 517 }
 518
 519 static void init_once(void *foo)
 520 {
 521         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 522
 523         INIT_LIST_HEAD(&ei->i_orphan);
 524 #ifdef CONFIG_EXT3_FS_XATTR
 525         init_rwsem(&ei->xattr_sem);
 526 #endif
 527         mutex_init(&ei->truncate_mutex);
 528         inode_init_once(&ei->vfs_inode);
 529 }
 530
 531 static int init_inodecache(void)
 532 {
 533         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 534                                              sizeof(struct ext3_inode_info),
 535                                              0, (SLAB_RECLAIM_ACCOUNT|
 536                                                 SLAB_MEM_SPREAD),
 537                                              init_once);
 538         if (ext3_inode_cachep == NULL)
 539                 return -ENOMEM;
 540         return 0;
 541 }
 542
 543 static void destroy_inodecache(void)
 544 {
 545         kmem_cache_destroy(ext3_inode_cachep);
 546 }
 547
 548 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 549 {
 550 #if defined(CONFIG_QUOTA)
 551         struct ext3_sb_info *sbi = EXT3_SB(sb);
 552
 553         if (sbi->s_jquota_fmt) {
 554                 char *fmtname = "";
 555
 556                 switch (sbi->s_jquota_fmt) {
 557                 case QFMT_VFS_OLD:
 558                         fmtname = "vfsold";
 559                         break;
 560                 case QFMT_VFS_V0:
 561                         fmtname = "vfsv0";
 562                         break;
 563                 case QFMT_VFS_V1:
 564                         fmtname = "vfsv1";
 565                         break;
 566                 }
 567                 seq_printf(seq, ",jqfmt=%s", fmtname);
 568         }
 569
 570         if (sbi->s_qf_names[USRQUOTA])
 571                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 572
 573         if (sbi->s_qf_names[GRPQUOTA])
 574                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 575
 576         if (test_opt(sb, USRQUOTA))
 577                 seq_puts(seq, ",usrquota");
 578
 579         if (test_opt(sb, GRPQUOTA))
 580                 seq_puts(seq, ",grpquota");
 581 #endif
 582 }
 583
 584 static char *data_mode_string(unsigned long mode)
 585 {
 586         switch (mode) {
 587         case EXT3_MOUNT_JOURNAL_DATA:
 588                 return "journal";
 589         case EXT3_MOUNT_ORDERED_DATA:
 590                 return "ordered";
 591         case EXT3_MOUNT_WRITEBACK_DATA:
 592                 return "writeback";
 593         }
 594         return "unknown";
 595 }
 596
 597 /*
 598  * Show an option if
 599  *  - it's set to a non-default value OR
 600  *  - if the per-sb default is different from the global default
 601  */
 602 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 603 {
 604         struct super_block *sb = vfs->mnt_sb;
 605         struct ext3_sb_info *sbi = EXT3_SB(sb);
 606         struct ext3_super_block *es = sbi->s_es;
 607         unsigned long def_mount_opts;
 608
 609         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 610
 611         if (sbi->s_sb_block != 1)
 612                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 613         if (test_opt(sb, MINIX_DF))
 614                 seq_puts(seq, ",minixdf");
 615         if (test_opt(sb, GRPID))
 616                 seq_puts(seq, ",grpid");
 617         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 618                 seq_puts(seq, ",nogrpid");
 619         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 620             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 621                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 622         }
 623         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 624             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 625                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 626         }
 627         if (test_opt(sb, ERRORS_RO)) {
 628                 int def_errors = le16_to_cpu(es->s_errors);
 629
 630                 if (def_errors == EXT3_ERRORS_PANIC ||
 631                     def_errors == EXT3_ERRORS_CONTINUE) {
 632                         seq_puts(seq, ",errors=remount-ro");
 633                 }
 634         }
 635         if (test_opt(sb, ERRORS_CONT))
 636                 seq_puts(seq, ",errors=continue");
 637         if (test_opt(sb, ERRORS_PANIC))
 638                 seq_puts(seq, ",errors=panic");
 639         if (test_opt(sb, NO_UID32))
 640                 seq_puts(seq, ",nouid32");
 641         if (test_opt(sb, DEBUG))
 642                 seq_puts(seq, ",debug");
 643         if (test_opt(sb, OLDALLOC))
 644                 seq_puts(seq, ",oldalloc");
 645 #ifdef CONFIG_EXT3_FS_XATTR
 646         if (test_opt(sb, XATTR_USER))
 647                 seq_puts(seq, ",user_xattr");
 648         if (!test_opt(sb, XATTR_USER) &&
 649             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 650                 seq_puts(seq, ",nouser_xattr");
 651         }
 652 #endif
 653 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 654         if (test_opt(sb, POSIX_ACL))
 655                 seq_puts(seq, ",acl");
 656         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 657                 seq_puts(seq, ",noacl");
 658 #endif
 659         if (!test_opt(sb, RESERVATION))
 660                 seq_puts(seq, ",noreservation");
 661         if (sbi->s_commit_interval) {
 662                 seq_printf(seq, ",commit=%u",
 663                            (unsigned) (sbi->s_commit_interval / HZ));
 664         }
 665
 666         /*
 667          * Always display barrier state so it's clear what the status is.
 668          */
 669         seq_puts(seq, ",barrier=");
 670         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 671         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 672         if (test_opt(sb, DATA_ERR_ABORT))
 673                 seq_puts(seq, ",data_err=abort");
 674
 675         if (test_opt(sb, NOLOAD))
 676                 seq_puts(seq, ",norecovery");
 677
 678         ext3_show_quota_options(seq, sb);
 679
 680         return 0;
 681 }
 682
 683
 684 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 685                 u64 ino, u32 generation)
 686 {
 687         struct inode *inode;
 688
 689         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 690                 return ERR_PTR(-ESTALE);
 691         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 692                 return ERR_PTR(-ESTALE);
 693
 694         /* iget isn't really right if the inode is currently unallocated!!
 695          *
 696          * ext3_read_inode will return a bad_inode if the inode had been
 697          * deleted, so we should be safe.
 698          *
 699          * Currently we don't know the generation for parent directory, so
 700          * a generation of 0 means "accept any"
 701          */
 702         inode = ext3_iget(sb, ino);
 703         if (IS_ERR(inode))
 704                 return ERR_CAST(inode);
 705         if (generation && inode->i_generation != generation) {
 706                 iput(inode);
 707                 return ERR_PTR(-ESTALE);
 708         }
 709
 710         return inode;
 711 }
 712
 713 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 714                 int fh_len, int fh_type)
 715 {
 716         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 717                                     ext3_nfs_get_inode);
 718 }
 719
 720 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 721                 int fh_len, int fh_type)
 722 {
 723         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 724                                     ext3_nfs_get_inode);
 725 }
 726
 727 /*
 728  * Try to release metadata pages (indirect blocks, directories) which are
 729  * mapped via the block device.  Since these pages could have journal heads
 730  * which would prevent try_to_free_buffers() from freeing them, we must use
 731  * jbd layer's try_to_free_buffers() function to release them.
 732  */
 733 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 734                                  gfp_t wait)
 735 {
 736         journal_t *journal = EXT3_SB(sb)->s_journal;
 737
 738         WARN_ON(PageChecked(page));
 739         if (!page_has_buffers(page))
 740                 return 0;
 741         if (journal)
 742                 return journal_try_to_free_buffers(journal, page,
 743                                                    wait & ~__GFP_WAIT);
 744         return try_to_free_buffers(page);
 745 }
 746
 747 #ifdef CONFIG_QUOTA
 748 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 749 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 750
 751 static int ext3_write_dquot(struct dquot *dquot);
 752 static int ext3_acquire_dquot(struct dquot *dquot);
 753 static int ext3_release_dquot(struct dquot *dquot);
 754 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 755 static int ext3_write_info(struct super_block *sb, int type);
 756 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 757                          struct path *path);
 758 static int ext3_quota_on_mount(struct super_block *sb, int type);
 759 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 760                                size_t len, loff_t off);
 761 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 762                                 const char *data, size_t len, loff_t off);
 763
 764 static const struct dquot_operations ext3_quota_operations = {
 765         .write_dquot    = ext3_write_dquot,
 766         .acquire_dquot  = ext3_acquire_dquot,
 767         .release_dquot  = ext3_release_dquot,
 768         .mark_dirty     = ext3_mark_dquot_dirty,
 769         .write_info     = ext3_write_info,
 770         .alloc_dquot    = dquot_alloc,
 771         .destroy_dquot  = dquot_destroy,
 772 };
 773
 774 static const struct quotactl_ops ext3_qctl_operations = {
 775         .quota_on       = ext3_quota_on,
 776         .quota_off      = dquot_quota_off,
 777         .quota_sync     = dquot_quota_sync,
 778         .get_info       = dquot_get_dqinfo,
 779         .set_info       = dquot_set_dqinfo,
 780         .get_dqblk      = dquot_get_dqblk,
 781         .set_dqblk      = dquot_set_dqblk
 782 };
 783 #endif
 784
 785 static const struct super_operations ext3_sops = {
 786         .alloc_inode    = ext3_alloc_inode,
 787         .destroy_inode  = ext3_destroy_inode,
 788         .write_inode    = ext3_write_inode,
 789         .dirty_inode    = ext3_dirty_inode,
 790         .evict_inode    = ext3_evict_inode,
 791         .put_super      = ext3_put_super,
 792         .sync_fs        = ext3_sync_fs,
 793         .freeze_fs      = ext3_freeze,
 794         .unfreeze_fs    = ext3_unfreeze,
 795         .statfs         = ext3_statfs,
 796         .remount_fs     = ext3_remount,
 797         .show_options   = ext3_show_options,
 798 #ifdef CONFIG_QUOTA
 799         .quota_read     = ext3_quota_read,
 800         .quota_write    = ext3_quota_write,
 801 #endif
 802         .bdev_try_to_free_page = bdev_try_to_free_page,
 803 };
 804
 805 static const struct export_operations ext3_export_ops = {
 806         .fh_to_dentry = ext3_fh_to_dentry,
 807         .fh_to_parent = ext3_fh_to_parent,
 808         .get_parent = ext3_get_parent,
 809 };
 810
 811 enum {
 812         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 813         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 814         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 815         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 816         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 817         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 818         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 819         Opt_data_err_abort, Opt_data_err_ignore,
 820         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 821         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 822         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 823         Opt_resize, Opt_usrquota, Opt_grpquota
 824 };
 825
 826 static const match_table_t tokens = {
 827         {Opt_bsd_df, "bsddf"},
 828         {Opt_minix_df, "minixdf"},
 829         {Opt_grpid, "grpid"},
 830         {Opt_grpid, "bsdgroups"},
 831         {Opt_nogrpid, "nogrpid"},
 832         {Opt_nogrpid, "sysvgroups"},
 833         {Opt_resgid, "resgid=%u"},
 834         {Opt_resuid, "resuid=%u"},
 835         {Opt_sb, "sb=%u"},
 836         {Opt_err_cont, "errors=continue"},
 837         {Opt_err_panic, "errors=panic"},
 838         {Opt_err_ro, "errors=remount-ro"},
 839         {Opt_nouid32, "nouid32"},
 840         {Opt_nocheck, "nocheck"},
 841         {Opt_nocheck, "check=none"},
 842         {Opt_debug, "debug"},
 843         {Opt_oldalloc, "oldalloc"},
 844         {Opt_orlov, "orlov"},
 845         {Opt_user_xattr, "user_xattr"},
 846         {Opt_nouser_xattr, "nouser_xattr"},
 847         {Opt_acl, "acl"},
 848         {Opt_noacl, "noacl"},
 849         {Opt_reservation, "reservation"},
 850         {Opt_noreservation, "noreservation"},
 851         {Opt_noload, "noload"},
 852         {Opt_noload, "norecovery"},
 853         {Opt_nobh, "nobh"},
 854         {Opt_bh, "bh"},
 855         {Opt_commit, "commit=%u"},
 856         {Opt_journal_update, "journal=update"},
 857         {Opt_journal_inum, "journal=%u"},
 858         {Opt_journal_dev, "journal_dev=%u"},
 859         {Opt_abort, "abort"},
 860         {Opt_data_journal, "data=journal"},
 861         {Opt_data_ordered, "data=ordered"},
 862         {Opt_data_writeback, "data=writeback"},
 863         {Opt_data_err_abort, "data_err=abort"},
 864         {Opt_data_err_ignore, "data_err=ignore"},
 865         {Opt_offusrjquota, "usrjquota="},
 866         {Opt_usrjquota, "usrjquota=%s"},
 867         {Opt_offgrpjquota, "grpjquota="},
 868         {Opt_grpjquota, "grpjquota=%s"},
 869         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 870         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 871         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 872         {Opt_grpquota, "grpquota"},
 873         {Opt_noquota, "noquota"},
 874         {Opt_quota, "quota"},
 875         {Opt_usrquota, "usrquota"},
 876         {Opt_barrier, "barrier=%u"},
 877         {Opt_barrier, "barrier"},
 878         {Opt_nobarrier, "nobarrier"},
 879         {Opt_resize, "resize"},
 880         {Opt_err, NULL},
 881 };
 882
 883 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 884 {
 885         ext3_fsblk_t    sb_block;
 886         char            *options = (char *) *data;
 887
 888         if (!options || strncmp(options, "sb=", 3) != 0)
 889                 return 1;       /* Default location */
 890         options += 3;
 891         /*todo: use simple_strtoll with >32bit ext3 */
 892         sb_block = simple_strtoul(options, &options, 0);
 893         if (*options && *options != ',') {
 894                 ext3_msg(sb, "error: invalid sb specification: %s",
 895                        (char *) *data);
 896                 return 1;
 897         }
 898         if (*options == ',')
 899                 options++;
 900         *data = (void *) options;
 901         return sb_block;
 902 }
 903
 904 #ifdef CONFIG_QUOTA
 905 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 906 {
 907         struct ext3_sb_info *sbi = EXT3_SB(sb);
 908         char *qname;
 909
 910         if (sb_any_quota_loaded(sb) &&
 911                 !sbi->s_qf_names[qtype]) {
 912                 ext3_msg(sb, KERN_ERR,
 913                         "Cannot change journaled "
 914                         "quota options when quota turned on");
 915                 return 0;
 916         }
 917         qname = match_strdup(args);
 918         if (!qname) {
 919                 ext3_msg(sb, KERN_ERR,
 920                         "Not enough memory for storing quotafile name");
 921                 return 0;
 922         }
 923         if (sbi->s_qf_names[qtype] &&
 924                 strcmp(sbi->s_qf_names[qtype], qname)) {
 925                 ext3_msg(sb, KERN_ERR,
 926                         "%s quota file already specified", QTYPE2NAME(qtype));
 927                 kfree(qname);
 928                 return 0;
 929         }
 930         sbi->s_qf_names[qtype] = qname;
 931         if (strchr(sbi->s_qf_names[qtype], '/')) {
 932                 ext3_msg(sb, KERN_ERR,
 933                         "quotafile must be on filesystem root");
 934                 kfree(sbi->s_qf_names[qtype]);
 935                 sbi->s_qf_names[qtype] = NULL;
 936                 return 0;
 937         }
 938         set_opt(sbi->s_mount_opt, QUOTA);
 939         return 1;
 940 }
 941
 942 static int clear_qf_name(struct super_block *sb, int qtype) {
 943
 944         struct ext3_sb_info *sbi = EXT3_SB(sb);
 945
 946         if (sb_any_quota_loaded(sb) &&
 947                 sbi->s_qf_names[qtype]) {
 948                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 949                         " when quota turned on");
 950                 return 0;
 951         }
 952         /*
 953          * The space will be released later when all options are confirmed
 954          * to be correct
 955          */
 956         sbi->s_qf_names[qtype] = NULL;
 957         return 1;
 958 }
 959 #endif
 960
 961 static int parse_options (char *options, struct super_block *sb,
 962                           unsigned int *inum, unsigned long *journal_devnum,
 963                           ext3_fsblk_t *n_blocks_count, int is_remount)
 964 {
 965         struct ext3_sb_info *sbi = EXT3_SB(sb);
 966         char * p;
 967         substring_t args[MAX_OPT_ARGS];
 968         int data_opt = 0;
 969         int option;
 970 #ifdef CONFIG_QUOTA
 971         int qfmt;
 972 #endif
 973
 974         if (!options)
 975                 return 1;
 976
 977         while ((p = strsep (&options, ",")) != NULL) {
 978                 int token;
 979                 if (!*p)
 980                         continue;
 981                 /*
 982                  * Initialize args struct so we know whether arg was
 983                  * found; some options take optional arguments.
 984                  */
 985                 args[0].to = args[0].from = 0;
 986                 token = match_token(p, tokens, args);
 987                 switch (token) {
 988                 case Opt_bsd_df:
 989                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 990                         break;
 991                 case Opt_minix_df:
 992                         set_opt (sbi->s_mount_opt, MINIX_DF);
 993                         break;
 994                 case Opt_grpid:
 995                         set_opt (sbi->s_mount_opt, GRPID);
 996                         break;
 997                 case Opt_nogrpid:
 998                         clear_opt (sbi->s_mount_opt, GRPID);
 999                         break;
1000                 case Opt_resuid:
1001                         if (match_int(&args[0], &option))
1002                                 return 0;
1003                         sbi->s_resuid = option;
1004                         break;
1005                 case Opt_resgid:
1006                         if (match_int(&args[0], &option))
1007                                 return 0;
1008                         sbi->s_resgid = option;
1009                         break;
1010                 case Opt_sb:
1011                         /* handled by get_sb_block() instead of here */
1012                         /* *sb_block = match_int(&args[0]); */
1013                         break;
1014                 case Opt_err_panic:
1015                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1016                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1017                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1018                         break;
1019                 case Opt_err_ro:
1020                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1021                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1022                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1023                         break;
1024                 case Opt_err_cont:
1025                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1026                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1027                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1028                         break;
1029                 case Opt_nouid32:
1030                         set_opt (sbi->s_mount_opt, NO_UID32);
1031                         break;
1032                 case Opt_nocheck:
1033                         clear_opt (sbi->s_mount_opt, CHECK);
1034                         break;
1035                 case Opt_debug:
1036                         set_opt (sbi->s_mount_opt, DEBUG);
1037                         break;
1038                 case Opt_oldalloc:
1039                         set_opt (sbi->s_mount_opt, OLDALLOC);
1040                         break;
1041                 case Opt_orlov:
1042                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1043                         break;
1044 #ifdef CONFIG_EXT3_FS_XATTR
1045                 case Opt_user_xattr:
1046                         set_opt (sbi->s_mount_opt, XATTR_USER);
1047                         break;
1048                 case Opt_nouser_xattr:
1049                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1050                         break;
1051 #else
1052                 case Opt_user_xattr:
1053                 case Opt_nouser_xattr:
1054                         ext3_msg(sb, KERN_INFO,
1055                                 "(no)user_xattr options not supported");
1056                         break;
1057 #endif
1058 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1059                 case Opt_acl:
1060                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1061                         break;
1062                 case Opt_noacl:
1063                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1064                         break;
1065 #else
1066                 case Opt_acl:
1067                 case Opt_noacl:
1068                         ext3_msg(sb, KERN_INFO,
1069                                 "(no)acl options not supported");
1070                         break;
1071 #endif
1072                 case Opt_reservation:
1073                         set_opt(sbi->s_mount_opt, RESERVATION);
1074                         break;
1075                 case Opt_noreservation:
1076                         clear_opt(sbi->s_mount_opt, RESERVATION);
1077                         break;
1078                 case Opt_journal_update:
1079                         /* @@@ FIXME */
1080                         /* Eventually we will want to be able to create
1081                            a journal file here.  For now, only allow the
1082                            user to specify an existing inode to be the
1083                            journal file. */
1084                         if (is_remount) {
1085                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1086                                         "journal on remount");
1087                                 return 0;
1088                         }
1089                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1090                         break;
1091                 case Opt_journal_inum:
1092                         if (is_remount) {
1093                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1094                                        "journal on remount");
1095                                 return 0;
1096                         }
1097                         if (match_int(&args[0], &option))
1098                                 return 0;
1099                         *inum = option;
1100                         break;
1101                 case Opt_journal_dev:
1102                         if (is_remount) {
1103                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1104                                        "journal on remount");
1105                                 return 0;
1106                         }
1107                         if (match_int(&args[0], &option))
1108                                 return 0;
1109                         *journal_devnum = option;
1110                         break;
1111                 case Opt_noload:
1112                         set_opt (sbi->s_mount_opt, NOLOAD);
1113                         break;
1114                 case Opt_commit:
1115                         if (match_int(&args[0], &option))
1116                                 return 0;
1117                         if (option < 0)
1118                                 return 0;
1119                         if (option == 0)
1120                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1121                         sbi->s_commit_interval = HZ * option;
1122                         break;
1123                 case Opt_data_journal:
1124                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1125                         goto datacheck;
1126                 case Opt_data_ordered:
1127                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1128                         goto datacheck;
1129                 case Opt_data_writeback:
1130                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1131                 datacheck:
1132                         if (is_remount) {
1133                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1134                                         break;
1135                                 ext3_msg(sb, KERN_ERR,
1136                                         "error: cannot change "
1137                                         "data mode on remount. The filesystem "
1138                                         "is mounted in data=%s mode and you "
1139                                         "try to remount it in data=%s mode.",
1140                                         data_mode_string(test_opt(sb,
1141                                                         DATA_FLAGS)),
1142                                         data_mode_string(data_opt));
1143                                 return 0;
1144                         } else {
1145                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1146                                 sbi->s_mount_opt |= data_opt;
1147                         }
1148                         break;
1149                 case Opt_data_err_abort:
1150                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1151                         break;
1152                 case Opt_data_err_ignore:
1153                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1154                         break;
1155 #ifdef CONFIG_QUOTA
1156                 case Opt_usrjquota:
1157                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1158                                 return 0;
1159                         break;
1160                 case Opt_grpjquota:
1161                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1162                                 return 0;
1163                         break;
1164                 case Opt_offusrjquota:
1165                         if (!clear_qf_name(sb, USRQUOTA))
1166                                 return 0;
1167                         break;
1168                 case Opt_offgrpjquota:
1169                         if (!clear_qf_name(sb, GRPQUOTA))
1170                                 return 0;
1171                         break;
1172                 case Opt_jqfmt_vfsold:
1173                         qfmt = QFMT_VFS_OLD;
1174                         goto set_qf_format;
1175                 case Opt_jqfmt_vfsv0:
1176                         qfmt = QFMT_VFS_V0;
1177                         goto set_qf_format;
1178                 case Opt_jqfmt_vfsv1:
1179                         qfmt = QFMT_VFS_V1;
1180 set_qf_format:
1181                         if (sb_any_quota_loaded(sb) &&
1182                             sbi->s_jquota_fmt != qfmt) {
1183                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1184                                         "journaled quota options when "
1185                                         "quota turned on.");
1186                                 return 0;
1187                         }
1188                         sbi->s_jquota_fmt = qfmt;
1189                         break;
1190                 case Opt_quota:
1191                 case Opt_usrquota:
1192                         set_opt(sbi->s_mount_opt, QUOTA);
1193                         set_opt(sbi->s_mount_opt, USRQUOTA);
1194                         break;
1195                 case Opt_grpquota:
1196                         set_opt(sbi->s_mount_opt, QUOTA);
1197                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1198                         break;
1199                 case Opt_noquota:
1200                         if (sb_any_quota_loaded(sb)) {
1201                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1202                                         "quota options when quota turned on.");
1203                                 return 0;
1204                         }
1205                         clear_opt(sbi->s_mount_opt, QUOTA);
1206                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1207                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1208                         break;
1209 #else
1210                 case Opt_quota:
1211                 case Opt_usrquota:
1212                 case Opt_grpquota:
1213                         ext3_msg(sb, KERN_ERR,
1214                                 "error: quota options not supported.");
1215                         break;
1216                 case Opt_usrjquota:
1217                 case Opt_grpjquota:
1218                 case Opt_offusrjquota:
1219                 case Opt_offgrpjquota:
1220                 case Opt_jqfmt_vfsold:
1221                 case Opt_jqfmt_vfsv0:
1222                 case Opt_jqfmt_vfsv1:
1223                         ext3_msg(sb, KERN_ERR,
1224                                 "error: journaled quota options not "
1225                                 "supported.");
1226                         break;
1227                 case Opt_noquota:
1228                         break;
1229 #endif
1230                 case Opt_abort:
1231                         set_opt(sbi->s_mount_opt, ABORT);
1232                         break;
1233                 case Opt_nobarrier:
1234                         clear_opt(sbi->s_mount_opt, BARRIER);
1235                         break;
1236                 case Opt_barrier:
1237                         if (args[0].from) {
1238                                 if (match_int(&args[0], &option))
1239                                         return 0;
1240                         } else
1241                                 option = 1;     /* No argument, default to 1 */
1242                         if (option)
1243                                 set_opt(sbi->s_mount_opt, BARRIER);
1244                         else
1245                                 clear_opt(sbi->s_mount_opt, BARRIER);
1246                         break;
1247                 case Opt_ignore:
1248                         break;
1249                 case Opt_resize:
1250                         if (!is_remount) {
1251                                 ext3_msg(sb, KERN_ERR,
1252                                         "error: resize option only available "
1253                                         "for remount");
1254                                 return 0;
1255                         }
1256                         if (match_int(&args[0], &option) != 0)
1257                                 return 0;
1258                         *n_blocks_count = option;
1259                         break;
1260                 case Opt_nobh:
1261                         ext3_msg(sb, KERN_WARNING,
1262                                 "warning: ignoring deprecated nobh option");
1263                         break;
1264                 case Opt_bh:
1265                         ext3_msg(sb, KERN_WARNING,
1266                                 "warning: ignoring deprecated bh option");
1267                         break;
1268                 default:
1269                         ext3_msg(sb, KERN_ERR,
1270                                 "error: unrecognized mount option \"%s\" "
1271                                 "or missing value", p);
1272                         return 0;
1273                 }
1274         }
1275 #ifdef CONFIG_QUOTA
1276         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1277                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1278                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1279                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1280                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1281
1282                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1283                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1284                                         "format mixing.");
1285                         return 0;
1286                 }
1287
1288                 if (!sbi->s_jquota_fmt) {
1289                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1290                                         "not specified.");
1291                         return 0;
1292                 }
1293         } else {
1294                 if (sbi->s_jquota_fmt) {
1295                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1296                                         "specified with no journaling "
1297                                         "enabled.");
1298                         return 0;
1299                 }
1300         }
1301 #endif
1302         return 1;
1303 }
1304
1305 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1306                             int read_only)
1307 {
1308         struct ext3_sb_info *sbi = EXT3_SB(sb);
1309         int res = 0;
1310
1311         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1312                 ext3_msg(sb, KERN_ERR,
1313                         "error: revision level too high, "
1314                         "forcing read-only mode");
1315                 res = MS_RDONLY;
1316         }
1317         if (read_only)
1318                 return res;
1319         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1320                 ext3_msg(sb, KERN_WARNING,
1321                         "warning: mounting unchecked fs, "
1322                         "running e2fsck is recommended");
1323         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1324                 ext3_msg(sb, KERN_WARNING,
1325                         "warning: mounting fs with errors, "
1326                         "running e2fsck is recommended");
1327         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1328                  le16_to_cpu(es->s_mnt_count) >=
1329                         le16_to_cpu(es->s_max_mnt_count))
1330                 ext3_msg(sb, KERN_WARNING,
1331                         "warning: maximal mount count reached, "
1332                         "running e2fsck is recommended");
1333         else if (le32_to_cpu(es->s_checkinterval) &&
1334                 (le32_to_cpu(es->s_lastcheck) +
1335                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1336                 ext3_msg(sb, KERN_WARNING,
1337                         "warning: checktime reached, "
1338                         "running e2fsck is recommended");
1339 #if 0
1340                 /* @@@ We _will_ want to clear the valid bit if we find
1341                    inconsistencies, to force a fsck at reboot.  But for
1342                    a plain journaled filesystem we can keep it set as
1343                    valid forever! :) */
1344         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1345 #endif
1346         if (!le16_to_cpu(es->s_max_mnt_count))
1347                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1348         le16_add_cpu(&es->s_mnt_count, 1);
1349         es->s_mtime = cpu_to_le32(get_seconds());
1350         ext3_update_dynamic_rev(sb);
1351         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1352
1353         ext3_commit_super(sb, es, 1);
1354         if (test_opt(sb, DEBUG))
1355                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1356                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1357                         sb->s_blocksize,
1358                         sbi->s_groups_count,
1359                         EXT3_BLOCKS_PER_GROUP(sb),
1360                         EXT3_INODES_PER_GROUP(sb),
1361                         sbi->s_mount_opt);
1362
1363         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1364                 char b[BDEVNAME_SIZE];
1365                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1366                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1367         } else {
1368                 ext3_msg(sb, KERN_INFO, "using internal journal");
1369         }
1370         return res;
1371 }
1372
1373 /* Called at mount-time, super-block is locked */
1374 static int ext3_check_descriptors(struct super_block *sb)
1375 {
1376         struct ext3_sb_info *sbi = EXT3_SB(sb);
1377         int i;
1378
1379         ext3_debug ("Checking group descriptors");
1380
1381         for (i = 0; i < sbi->s_groups_count; i++) {
1382                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1383                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1384                 ext3_fsblk_t last_block;
1385
1386                 if (i == sbi->s_groups_count - 1)
1387                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1388                 else
1389                         last_block = first_block +
1390                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1391
1392                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1393                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1394                 {
1395                         ext3_error (sb, "ext3_check_descriptors",
1396                                     "Block bitmap for group %d"
1397                                     " not in group (block %lu)!",
1398                                     i, (unsigned long)
1399                                         le32_to_cpu(gdp->bg_block_bitmap));
1400                         return 0;
1401                 }
1402                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1403                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1404                 {
1405                         ext3_error (sb, "ext3_check_descriptors",
1406                                     "Inode bitmap for group %d"
1407                                     " not in group (block %lu)!",
1408                                     i, (unsigned long)
1409                                         le32_to_cpu(gdp->bg_inode_bitmap));
1410                         return 0;
1411                 }
1412                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1413                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1414                     last_block)
1415                 {
1416                         ext3_error (sb, "ext3_check_descriptors",
1417                                     "Inode table for group %d"
1418                                     " not in group (block %lu)!",
1419                                     i, (unsigned long)
1420                                         le32_to_cpu(gdp->bg_inode_table));
1421                         return 0;
1422                 }
1423         }
1424
1425         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1426         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1427         return 1;
1428 }
1429
1430
1431 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1432  * the superblock) which were deleted from all directories, but held open by
1433  * a process at the time of a crash.  We walk the list and try to delete these
1434  * inodes at recovery time (only with a read-write filesystem).
1435  *
1436  * In order to keep the orphan inode chain consistent during traversal (in
1437  * case of crash during recovery), we link each inode into the superblock
1438  * orphan list_head and handle it the same way as an inode deletion during
1439  * normal operation (which journals the operations for us).
1440  *
1441  * We only do an iget() and an iput() on each inode, which is very safe if we
1442  * accidentally point at an in-use or already deleted inode.  The worst that
1443  * can happen in this case is that we get a "bit already cleared" message from
1444  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1445  * e2fsck was run on this filesystem, and it must have already done the orphan
1446  * inode cleanup for us, so we can safely abort without any further action.
1447  */
1448 static void ext3_orphan_cleanup (struct super_block * sb,
1449                                  struct ext3_super_block * es)
1450 {
1451         unsigned int s_flags = sb->s_flags;
1452         int nr_orphans = 0, nr_truncates = 0;
1453 #ifdef CONFIG_QUOTA
1454         int i;
1455 #endif
1456         if (!es->s_last_orphan) {
1457                 jbd_debug(4, "no orphan inodes to clean up\n");
1458                 return;
1459         }
1460
1461         if (bdev_read_only(sb->s_bdev)) {
1462                 ext3_msg(sb, KERN_ERR, "error: write access "
1463                         "unavailable, skipping orphan cleanup.");
1464                 return;
1465         }
1466
1467         /* Check if feature set allows readwrite operations */
1468         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1469                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1470                          "unknown ROCOMPAT features");
1471                 return;
1472         }
1473
1474         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1475                 if (es->s_last_orphan)
1476                         jbd_debug(1, "Errors on filesystem, "
1477                                   "clearing orphan list.\n");
1478                 es->s_last_orphan = 0;
1479                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1480                 return;
1481         }
1482
1483         if (s_flags & MS_RDONLY) {
1484                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1485                 sb->s_flags &= ~MS_RDONLY;
1486         }
1487 #ifdef CONFIG_QUOTA
1488         /* Needed for iput() to work correctly and not trash data */
1489         sb->s_flags |= MS_ACTIVE;
1490         /* Turn on quotas so that they are updated correctly */
1491         for (i = 0; i < MAXQUOTAS; i++) {
1492                 if (EXT3_SB(sb)->s_qf_names[i]) {
1493                         int ret = ext3_quota_on_mount(sb, i);
1494                         if (ret < 0)
1495                                 ext3_msg(sb, KERN_ERR,
1496                                         "error: cannot turn on journaled "
1497                                         "quota: %d", ret);
1498                 }
1499         }
1500 #endif
1501
1502         while (es->s_last_orphan) {
1503                 struct inode *inode;
1504
1505                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1506                 if (IS_ERR(inode)) {
1507                         es->s_last_orphan = 0;
1508                         break;
1509                 }
1510
1511                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1512                 dquot_initialize(inode);
1513                 if (inode->i_nlink) {
1514                         printk(KERN_DEBUG
1515                                 "%s: truncating inode %lu to %Ld bytes\n",
1516                                 __func__, inode->i_ino, inode->i_size);
1517                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1518                                   inode->i_ino, inode->i_size);
1519                         ext3_truncate(inode);
1520                         nr_truncates++;
1521                 } else {
1522                         printk(KERN_DEBUG
1523                                 "%s: deleting unreferenced inode %lu\n",
1524                                 __func__, inode->i_ino);
1525                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1526                                   inode->i_ino);
1527                         nr_orphans++;
1528                 }
1529                 iput(inode);  /* The delete magic happens here! */
1530         }
1531
1532 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1533
1534         if (nr_orphans)
1535                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1536                        PLURAL(nr_orphans));
1537         if (nr_truncates)
1538                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1539                        PLURAL(nr_truncates));
1540 #ifdef CONFIG_QUOTA
1541         /* Turn quotas off */
1542         for (i = 0; i < MAXQUOTAS; i++) {
1543                 if (sb_dqopt(sb)->files[i])
1544                         dquot_quota_off(sb, i);
1545         }
1546 #endif
1547         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1548 }
1549
1550 /*
1551  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1552  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1553  * We need to be 1 filesystem block less than the 2^32 sector limit.
1554  */
1555 static loff_t ext3_max_size(int bits)
1556 {
1557         loff_t res = EXT3_NDIR_BLOCKS;
1558         int meta_blocks;
1559         loff_t upper_limit;
1560
1561         /* This is calculated to be the largest file size for a
1562          * dense, file such that the total number of
1563          * sectors in the file, including data and all indirect blocks,
1564          * does not exceed 2^32 -1
1565          * __u32 i_blocks representing the total number of
1566          * 512 bytes blocks of the file
1567          */
1568         upper_limit = (1LL << 32) - 1;
1569
1570         /* total blocks in file system block size */
1571         upper_limit >>= (bits - 9);
1572
1573
1574         /* indirect blocks */
1575         meta_blocks = 1;
1576         /* double indirect blocks */
1577         meta_blocks += 1 + (1LL << (bits-2));
1578         /* tripple indirect blocks */
1579         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1580
1581         upper_limit -= meta_blocks;
1582         upper_limit <<= bits;
1583
1584         res += 1LL << (bits-2);
1585         res += 1LL << (2*(bits-2));
1586         res += 1LL << (3*(bits-2));
1587         res <<= bits;
1588         if (res > upper_limit)
1589                 res = upper_limit;
1590
1591         if (res > MAX_LFS_FILESIZE)
1592                 res = MAX_LFS_FILESIZE;
1593
1594         return res;
1595 }
1596
1597 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1598                                     ext3_fsblk_t logic_sb_block,
1599                                     int nr)
1600 {
1601         struct ext3_sb_info *sbi = EXT3_SB(sb);
1602         unsigned long bg, first_meta_bg;
1603         int has_super = 0;
1604
1605         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1606
1607         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1608             nr < first_meta_bg)
1609                 return (logic_sb_block + nr + 1);
1610         bg = sbi->s_desc_per_block * nr;
1611         if (ext3_bg_has_super(sb, bg))
1612                 has_super = 1;
1613         return (has_super + ext3_group_first_block_no(sb, bg));
1614 }
1615
1616
1617 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1618 {
1619         struct buffer_head * bh;
1620         struct ext3_super_block *es = NULL;
1621         struct ext3_sb_info *sbi;
1622         ext3_fsblk_t block;
1623         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1624         ext3_fsblk_t logic_sb_block;
1625         unsigned long offset = 0;
1626         unsigned int journal_inum = 0;
1627         unsigned long journal_devnum = 0;
1628         unsigned long def_mount_opts;
1629         struct inode *root;
1630         int blocksize;
1631         int hblock;
1632         int db_count;
1633         int i;
1634         int needs_recovery;
1635         int ret = -EINVAL;
1636         __le32 features;
1637         int err;
1638
1639         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1640         if (!sbi)
1641                 return -ENOMEM;
1642
1643         sbi->s_blockgroup_lock =
1644                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1645         if (!sbi->s_blockgroup_lock) {
1646                 kfree(sbi);
1647                 return -ENOMEM;
1648         }
1649         sb->s_fs_info = sbi;
1650         sbi->s_mount_opt = 0;
1651         sbi->s_resuid = EXT3_DEF_RESUID;
1652         sbi->s_resgid = EXT3_DEF_RESGID;
1653         sbi->s_sb_block = sb_block;
1654
1655         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1656         if (!blocksize) {
1657                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1658                 goto out_fail;
1659         }
1660
1661         /*
1662          * The ext3 superblock will not be buffer aligned for other than 1kB
1663          * block sizes.  We need to calculate the offset from buffer start.
1664          */
1665         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1666                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1667                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1668         } else {
1669                 logic_sb_block = sb_block;
1670         }
1671
1672         if (!(bh = sb_bread(sb, logic_sb_block))) {
1673                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1674                 goto out_fail;
1675         }
1676         /*
1677          * Note: s_es must be initialized as soon as possible because
1678          *       some ext3 macro-instructions depend on its value
1679          */
1680         es = (struct ext3_super_block *) (bh->b_data + offset);
1681         sbi->s_es = es;
1682         sb->s_magic = le16_to_cpu(es->s_magic);
1683         if (sb->s_magic != EXT3_SUPER_MAGIC)
1684                 goto cantfind_ext3;
1685
1686         /* Set defaults before we parse the mount options */
1687         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1688         if (def_mount_opts & EXT3_DEFM_DEBUG)
1689                 set_opt(sbi->s_mount_opt, DEBUG);
1690         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1691                 set_opt(sbi->s_mount_opt, GRPID);
1692         if (def_mount_opts & EXT3_DEFM_UID16)
1693                 set_opt(sbi->s_mount_opt, NO_UID32);
1694 #ifdef CONFIG_EXT3_FS_XATTR
1695         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1696                 set_opt(sbi->s_mount_opt, XATTR_USER);
1697 #endif
1698 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1699         if (def_mount_opts & EXT3_DEFM_ACL)
1700                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1701 #endif
1702         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1703                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1704         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1705                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1706         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1707                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1708
1709         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1710                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1711         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1712                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1713         else
1714                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1715
1716         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1717         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1718
1719         set_opt(sbi->s_mount_opt, RESERVATION);
1720
1721         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1722                             NULL, 0))
1723                 goto failed_mount;
1724
1725         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1726                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1727
1728         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1729             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1730              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1731              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1732                 ext3_msg(sb, KERN_WARNING,
1733                         "warning: feature flags set on rev 0 fs, "
1734                         "running e2fsck is recommended");
1735         /*
1736          * Check feature flags regardless of the revision level, since we
1737          * previously didn't change the revision level when setting the flags,
1738          * so there is a chance incompat flags are set on a rev 0 filesystem.
1739          */
1740         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1741         if (features) {
1742                 ext3_msg(sb, KERN_ERR,
1743                         "error: couldn't mount because of unsupported "
1744                         "optional features (%x)", le32_to_cpu(features));
1745                 goto failed_mount;
1746         }
1747         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1748         if (!(sb->s_flags & MS_RDONLY) && features) {
1749                 ext3_msg(sb, KERN_ERR,
1750                         "error: couldn't mount RDWR because of unsupported "
1751                         "optional features (%x)", le32_to_cpu(features));
1752                 goto failed_mount;
1753         }
1754         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1755
1756         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1757             blocksize > EXT3_MAX_BLOCK_SIZE) {
1758                 ext3_msg(sb, KERN_ERR,
1759                         "error: couldn't mount because of unsupported "
1760                         "filesystem blocksize %d", blocksize);
1761                 goto failed_mount;
1762         }
1763
1764         hblock = bdev_logical_block_size(sb->s_bdev);
1765         if (sb->s_blocksize != blocksize) {
1766                 /*
1767                  * Make sure the blocksize for the filesystem is larger
1768                  * than the hardware sectorsize for the machine.
1769                  */
1770                 if (blocksize < hblock) {
1771                         ext3_msg(sb, KERN_ERR,
1772                                 "error: fsblocksize %d too small for "
1773                                 "hardware sectorsize %d", blocksize, hblock);
1774                         goto failed_mount;
1775                 }
1776
1777                 brelse (bh);
1778                 if (!sb_set_blocksize(sb, blocksize)) {
1779                         ext3_msg(sb, KERN_ERR,
1780                                 "error: bad blocksize %d", blocksize);
1781                         goto out_fail;
1782                 }
1783                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1784                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1785                 bh = sb_bread(sb, logic_sb_block);
1786                 if (!bh) {
1787                         ext3_msg(sb, KERN_ERR,
1788                                "error: can't read superblock on 2nd try");
1789                         goto failed_mount;
1790                 }
1791                 es = (struct ext3_super_block *)(bh->b_data + offset);
1792                 sbi->s_es = es;
1793                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1794                         ext3_msg(sb, KERN_ERR,
1795                                 "error: magic mismatch");
1796                         goto failed_mount;
1797                 }
1798         }
1799
1800         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1801
1802         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1803                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1804                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1805         } else {
1806                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1807                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1808                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1809                     (!is_power_of_2(sbi->s_inode_size)) ||
1810                     (sbi->s_inode_size > blocksize)) {
1811                         ext3_msg(sb, KERN_ERR,
1812                                 "error: unsupported inode size: %d",
1813                                 sbi->s_inode_size);
1814                         goto failed_mount;
1815                 }
1816         }
1817         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1818                                    le32_to_cpu(es->s_log_frag_size);
1819         if (blocksize != sbi->s_frag_size) {
1820                 ext3_msg(sb, KERN_ERR,
1821                        "error: fragsize %lu != blocksize %u (unsupported)",
1822                        sbi->s_frag_size, blocksize);
1823                 goto failed_mount;
1824         }
1825         sbi->s_frags_per_block = 1;
1826         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1827         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1828         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1829         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1830                 goto cantfind_ext3;
1831         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1832         if (sbi->s_inodes_per_block == 0)
1833                 goto cantfind_ext3;
1834         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1835                                         sbi->s_inodes_per_block;
1836         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1837         sbi->s_sbh = bh;
1838         sbi->s_mount_state = le16_to_cpu(es->s_state);
1839         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1840         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1841         for (i=0; i < 4; i++)
1842                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1843         sbi->s_def_hash_version = es->s_def_hash_version;
1844         i = le32_to_cpu(es->s_flags);
1845         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1846                 sbi->s_hash_unsigned = 3;
1847         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1848 #ifdef __CHAR_UNSIGNED__
1849                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1850                 sbi->s_hash_unsigned = 3;
1851 #else
1852                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1853 #endif
1854         }
1855
1856         if (sbi->s_blocks_per_group > blocksize * 8) {
1857                 ext3_msg(sb, KERN_ERR,
1858                         "#blocks per group too big: %lu",
1859                         sbi->s_blocks_per_group);
1860                 goto failed_mount;
1861         }
1862         if (sbi->s_frags_per_group > blocksize * 8) {
1863                 ext3_msg(sb, KERN_ERR,
1864                         "error: #fragments per group too big: %lu",
1865                         sbi->s_frags_per_group);
1866                 goto failed_mount;
1867         }
1868         if (sbi->s_inodes_per_group > blocksize * 8) {
1869                 ext3_msg(sb, KERN_ERR,
1870                         "error: #inodes per group too big: %lu",
1871                         sbi->s_inodes_per_group);
1872                 goto failed_mount;
1873         }
1874
1875         err = generic_check_addressable(sb->s_blocksize_bits,
1876                                         le32_to_cpu(es->s_blocks_count));
1877         if (err) {
1878                 ext3_msg(sb, KERN_ERR,
1879                         "error: filesystem is too large to mount safely");
1880                 if (sizeof(sector_t) < 8)
1881                         ext3_msg(sb, KERN_ERR,
1882                                 "error: CONFIG_LBDAF not enabled");
1883                 ret = err;
1884                 goto failed_mount;
1885         }
1886
1887         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1888                 goto cantfind_ext3;
1889         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1890                                le32_to_cpu(es->s_first_data_block) - 1)
1891                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1892         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1893         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1894                                     GFP_KERNEL);
1895         if (sbi->s_group_desc == NULL) {
1896                 ext3_msg(sb, KERN_ERR,
1897                         "error: not enough memory");
1898                 ret = -ENOMEM;
1899                 goto failed_mount;
1900         }
1901
1902         bgl_lock_init(sbi->s_blockgroup_lock);
1903
1904         for (i = 0; i < db_count; i++) {
1905                 block = descriptor_loc(sb, logic_sb_block, i);
1906                 sbi->s_group_desc[i] = sb_bread(sb, block);
1907                 if (!sbi->s_group_desc[i]) {
1908                         ext3_msg(sb, KERN_ERR,
1909                                 "error: can't read group descriptor %d", i);
1910                         db_count = i;
1911                         goto failed_mount2;
1912                 }
1913         }
1914         if (!ext3_check_descriptors (sb)) {
1915                 ext3_msg(sb, KERN_ERR,
1916                         "error: group descriptors corrupted");
1917                 goto failed_mount2;
1918         }
1919         sbi->s_gdb_count = db_count;
1920         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1921         spin_lock_init(&sbi->s_next_gen_lock);
1922
1923         /* per fileystem reservation list head & lock */
1924         spin_lock_init(&sbi->s_rsv_window_lock);
1925         sbi->s_rsv_window_root = RB_ROOT;
1926         /* Add a single, static dummy reservation to the start of the
1927          * reservation window list --- it gives us a placeholder for
1928          * append-at-start-of-list which makes the allocation logic
1929          * _much_ simpler. */
1930         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1931         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1932         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1933         sbi->s_rsv_window_head.rsv_goal_size = 0;
1934         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1935
1936         /*
1937          * set up enough so that it can read an inode
1938          */
1939         sb->s_op = &ext3_sops;
1940         sb->s_export_op = &ext3_export_ops;
1941         sb->s_xattr = ext3_xattr_handlers;
1942 #ifdef CONFIG_QUOTA
1943         sb->s_qcop = &ext3_qctl_operations;
1944         sb->dq_op = &ext3_quota_operations;
1945 #endif
1946         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1947         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1948         mutex_init(&sbi->s_orphan_lock);
1949         mutex_init(&sbi->s_resize_lock);
1950
1951         sb->s_root = NULL;
1952
1953         needs_recovery = (es->s_last_orphan != 0 ||
1954                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1955                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1956
1957         /*
1958          * The first inode we look at is the journal inode.  Don't try
1959          * root first: it may be modified in the journal!
1960          */
1961         if (!test_opt(sb, NOLOAD) &&
1962             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1963                 if (ext3_load_journal(sb, es, journal_devnum))
1964                         goto failed_mount2;
1965         } else if (journal_inum) {
1966                 if (ext3_create_journal(sb, es, journal_inum))
1967                         goto failed_mount2;
1968         } else {
1969                 if (!silent)
1970                         ext3_msg(sb, KERN_ERR,
1971                                 "error: no journal found. "
1972                                 "mounting ext3 over ext2?");
1973                 goto failed_mount2;
1974         }
1975         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1976                         ext3_count_free_blocks(sb));
1977         if (!err) {
1978                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1979                                 ext3_count_free_inodes(sb));
1980         }
1981         if (!err) {
1982                 err = percpu_counter_init(&sbi->s_dirs_counter,
1983                                 ext3_count_dirs(sb));
1984         }
1985         if (err) {
1986                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1987                 ret = err;
1988                 goto failed_mount3;
1989         }
1990
1991         /* We have now updated the journal if required, so we can
1992          * validate the data journaling mode. */
1993         switch (test_opt(sb, DATA_FLAGS)) {
1994         case 0:
1995                 /* No mode set, assume a default based on the journal
1996                    capabilities: ORDERED_DATA if the journal can
1997                    cope, else JOURNAL_DATA */
1998                 if (journal_check_available_features
1999                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2000                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2001                 else
2002                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2003                 break;
2004
2005         case EXT3_MOUNT_ORDERED_DATA:
2006         case EXT3_MOUNT_WRITEBACK_DATA:
2007                 if (!journal_check_available_features
2008                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2009                         ext3_msg(sb, KERN_ERR,
2010                                 "error: journal does not support "
2011                                 "requested data journaling mode");
2012                         goto failed_mount3;
2013                 }
2014         default:
2015                 break;
2016         }
2017
2018         /*
2019          * The journal_load will have done any necessary log recovery,
2020          * so we can safely mount the rest of the filesystem now.
2021          */
2022
2023         root = ext3_iget(sb, EXT3_ROOT_INO);
2024         if (IS_ERR(root)) {
2025                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2026                 ret = PTR_ERR(root);
2027                 goto failed_mount3;
2028         }
2029         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2030                 iput(root);
2031                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2032                 goto failed_mount3;
2033         }
2034         sb->s_root = d_alloc_root(root);
2035         if (!sb->s_root) {
2036                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2037                 iput(root);
2038                 ret = -ENOMEM;
2039                 goto failed_mount3;
2040         }
2041
2042         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2043
2044         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2045         ext3_orphan_cleanup(sb, es);
2046         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2047         if (needs_recovery)
2048                 ext3_msg(sb, KERN_INFO, "recovery complete");
2049         ext3_mark_recovery_complete(sb, es);
2050         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2051                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2052                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2053                 "writeback");
2054
2055         return 0;
2056
2057 cantfind_ext3:
2058         if (!silent)
2059                 ext3_msg(sb, KERN_INFO,
2060                         "error: can't find ext3 filesystem on dev %s.",
2061                        sb->s_id);
2062         goto failed_mount;
2063
2064 failed_mount3:
2065         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2066         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2067         percpu_counter_destroy(&sbi->s_dirs_counter);
2068         journal_destroy(sbi->s_journal);
2069 failed_mount2:
2070         for (i = 0; i < db_count; i++)
2071                 brelse(sbi->s_group_desc[i]);
2072         kfree(sbi->s_group_desc);
2073 failed_mount:
2074 #ifdef CONFIG_QUOTA
2075         for (i = 0; i < MAXQUOTAS; i++)
2076                 kfree(sbi->s_qf_names[i]);
2077 #endif
2078         ext3_blkdev_remove(sbi);
2079         brelse(bh);
2080 out_fail:
2081         sb->s_fs_info = NULL;
2082         kfree(sbi->s_blockgroup_lock);
2083         kfree(sbi);
2084         return ret;
2085 }
2086
2087 /*
2088  * Setup any per-fs journal parameters now.  We'll do this both on
2089  * initial mount, once the journal has been initialised but before we've
2090  * done any recovery; and again on any subsequent remount.
2091  */
2092 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2093 {
2094         struct ext3_sb_info *sbi = EXT3_SB(sb);
2095
2096         if (sbi->s_commit_interval)
2097                 journal->j_commit_interval = sbi->s_commit_interval;
2098         /* We could also set up an ext3-specific default for the commit
2099          * interval here, but for now we'll just fall back to the jbd
2100          * default. */
2101
2102         spin_lock(&journal->j_state_lock);
2103         if (test_opt(sb, BARRIER))
2104                 journal->j_flags |= JFS_BARRIER;
2105         else
2106                 journal->j_flags &= ~JFS_BARRIER;
2107         if (test_opt(sb, DATA_ERR_ABORT))
2108                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2109         else
2110                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2111         spin_unlock(&journal->j_state_lock);
2112 }
2113
2114 static journal_t *ext3_get_journal(struct super_block *sb,
2115                                    unsigned int journal_inum)
2116 {
2117         struct inode *journal_inode;
2118         journal_t *journal;
2119
2120         /* First, test for the existence of a valid inode on disk.  Bad
2121          * things happen if we iget() an unused inode, as the subsequent
2122          * iput() will try to delete it. */
2123
2124         journal_inode = ext3_iget(sb, journal_inum);
2125         if (IS_ERR(journal_inode)) {
2126                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2127                 return NULL;
2128         }
2129         if (!journal_inode->i_nlink) {
2130                 make_bad_inode(journal_inode);
2131                 iput(journal_inode);
2132                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2133                 return NULL;
2134         }
2135
2136         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2137                   journal_inode, journal_inode->i_size);
2138         if (!S_ISREG(journal_inode->i_mode)) {
2139                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2140                 iput(journal_inode);
2141                 return NULL;
2142         }
2143
2144         journal = journal_init_inode(journal_inode);
2145         if (!journal) {
2146                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2147                 iput(journal_inode);
2148                 return NULL;
2149         }
2150         journal->j_private = sb;
2151         ext3_init_journal_params(sb, journal);
2152         return journal;
2153 }
2154
2155 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2156                                        dev_t j_dev)
2157 {
2158         struct buffer_head * bh;
2159         journal_t *journal;
2160         ext3_fsblk_t start;
2161         ext3_fsblk_t len;
2162         int hblock, blocksize;
2163         ext3_fsblk_t sb_block;
2164         unsigned long offset;
2165         struct ext3_super_block * es;
2166         struct block_device *bdev;
2167
2168         bdev = ext3_blkdev_get(j_dev, sb);
2169         if (bdev == NULL)
2170                 return NULL;
2171
2172         blocksize = sb->s_blocksize;
2173         hblock = bdev_logical_block_size(bdev);
2174         if (blocksize < hblock) {
2175                 ext3_msg(sb, KERN_ERR,
2176                         "error: blocksize too small for journal device");
2177                 goto out_bdev;
2178         }
2179
2180         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2181         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2182         set_blocksize(bdev, blocksize);
2183         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2184                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2185                         "external journal");
2186                 goto out_bdev;
2187         }
2188
2189         es = (struct ext3_super_block *) (bh->b_data + offset);
2190         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2191             !(le32_to_cpu(es->s_feature_incompat) &
2192               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2193                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2194                         "bad superblock");
2195                 brelse(bh);
2196                 goto out_bdev;
2197         }
2198
2199         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2200                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2201                 brelse(bh);
2202                 goto out_bdev;
2203         }
2204
2205         len = le32_to_cpu(es->s_blocks_count);
2206         start = sb_block + 1;
2207         brelse(bh);     /* we're done with the superblock */
2208
2209         journal = journal_init_dev(bdev, sb->s_bdev,
2210                                         start, len, blocksize);
2211         if (!journal) {
2212                 ext3_msg(sb, KERN_ERR,
2213                         "error: failed to create device journal");
2214                 goto out_bdev;
2215         }
2216         journal->j_private = sb;
2217         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2218         wait_on_buffer(journal->j_sb_buffer);
2219         if (!buffer_uptodate(journal->j_sb_buffer)) {
2220                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2221                 goto out_journal;
2222         }
2223         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2224                 ext3_msg(sb, KERN_ERR,
2225                         "error: external journal has more than one "
2226                         "user (unsupported) - %d",
2227                         be32_to_cpu(journal->j_superblock->s_nr_users));
2228                 goto out_journal;
2229         }
2230         EXT3_SB(sb)->journal_bdev = bdev;
2231         ext3_init_journal_params(sb, journal);
2232         return journal;
2233 out_journal:
2234         journal_destroy(journal);
2235 out_bdev:
2236         ext3_blkdev_put(bdev);
2237         return NULL;
2238 }
2239
2240 static int ext3_load_journal(struct super_block *sb,
2241                              struct ext3_super_block *es,
2242                              unsigned long journal_devnum)
2243 {
2244         journal_t *journal;
2245         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2246         dev_t journal_dev;
2247         int err = 0;
2248         int really_read_only;
2249
2250         if (journal_devnum &&
2251             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2252                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2253                         "numbers have changed");
2254                 journal_dev = new_decode_dev(journal_devnum);
2255         } else
2256                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2257
2258         really_read_only = bdev_read_only(sb->s_bdev);
2259
2260         /*
2261          * Are we loading a blank journal or performing recovery after a
2262          * crash?  For recovery, we need to check in advance whether we
2263          * can get read-write access to the device.
2264          */
2265
2266         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2267                 if (sb->s_flags & MS_RDONLY) {
2268                         ext3_msg(sb, KERN_INFO,
2269                                 "recovery required on readonly filesystem");
2270                         if (really_read_only) {
2271                                 ext3_msg(sb, KERN_ERR, "error: write access "
2272                                         "unavailable, cannot proceed");
2273                                 return -EROFS;
2274                         }
2275                         ext3_msg(sb, KERN_INFO,
2276                                 "write access will be enabled during recovery");
2277                 }
2278         }
2279
2280         if (journal_inum && journal_dev) {
2281                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2282                        "and inode journals");
2283                 return -EINVAL;
2284         }
2285
2286         if (journal_inum) {
2287                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2288                         return -EINVAL;
2289         } else {
2290                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2291                         return -EINVAL;
2292         }
2293
2294         if (!(journal->j_flags & JFS_BARRIER))
2295                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2296
2297         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2298                 err = journal_update_format(journal);
2299                 if (err)  {
2300                         ext3_msg(sb, KERN_ERR, "error updating journal");
2301                         journal_destroy(journal);
2302                         return err;
2303                 }
2304         }
2305
2306         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2307                 err = journal_wipe(journal, !really_read_only);
2308         if (!err)
2309                 err = journal_load(journal);
2310
2311         if (err) {
2312                 ext3_msg(sb, KERN_ERR, "error loading journal");
2313                 journal_destroy(journal);
2314                 return err;
2315         }
2316
2317         EXT3_SB(sb)->s_journal = journal;
2318         ext3_clear_journal_err(sb, es);
2319
2320         if (!really_read_only && journal_devnum &&
2321             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2322                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2323
2324                 /* Make sure we flush the recovery flag to disk. */
2325                 ext3_commit_super(sb, es, 1);
2326         }
2327
2328         return 0;
2329 }
2330
2331 static int ext3_create_journal(struct super_block *sb,
2332                                struct ext3_super_block *es,
2333                                unsigned int journal_inum)
2334 {
2335         journal_t *journal;
2336         int err;
2337
2338         if (sb->s_flags & MS_RDONLY) {
2339                 ext3_msg(sb, KERN_ERR,
2340                         "error: readonly filesystem when trying to "
2341                         "create journal");
2342                 return -EROFS;
2343         }
2344
2345         journal = ext3_get_journal(sb, journal_inum);
2346         if (!journal)
2347                 return -EINVAL;
2348
2349         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2350                journal_inum);
2351
2352         err = journal_create(journal);
2353         if (err) {
2354                 ext3_msg(sb, KERN_ERR, "error creating journal");
2355                 journal_destroy(journal);
2356                 return -EIO;
2357         }
2358
2359         EXT3_SB(sb)->s_journal = journal;
2360
2361         ext3_update_dynamic_rev(sb);
2362         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2363         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2364
2365         es->s_journal_inum = cpu_to_le32(journal_inum);
2366
2367         /* Make sure we flush the recovery flag to disk. */
2368         ext3_commit_super(sb, es, 1);
2369
2370         return 0;
2371 }
2372
2373 static int ext3_commit_super(struct super_block *sb,
2374                                struct ext3_super_block *es,
2375                                int sync)
2376 {
2377         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2378         int error = 0;
2379
2380         if (!sbh)
2381                 return error;
2382
2383         if (buffer_write_io_error(sbh)) {
2384                 /*
2385                  * Oh, dear.  A previous attempt to write the
2386                  * superblock failed.  This could happen because the
2387                  * USB device was yanked out.  Or it could happen to
2388                  * be a transient write error and maybe the block will
2389                  * be remapped.  Nothing we can do but to retry the
2390                  * write and hope for the best.
2391                  */
2392                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2393                        "superblock detected");
2394                 clear_buffer_write_io_error(sbh);
2395                 set_buffer_uptodate(sbh);
2396         }
2397         /*
2398          * If the file system is mounted read-only, don't update the
2399          * superblock write time.  This avoids updating the superblock
2400          * write time when we are mounting the root file system
2401          * read/only but we need to replay the journal; at that point,
2402          * for people who are east of GMT and who make their clock
2403          * tick in localtime for Windows bug-for-bug compatibility,
2404          * the clock is set in the future, and this will cause e2fsck
2405          * to complain and force a full file system check.
2406          */
2407         if (!(sb->s_flags & MS_RDONLY))
2408                 es->s_wtime = cpu_to_le32(get_seconds());
2409         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2410         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2411         BUFFER_TRACE(sbh, "marking dirty");
2412         mark_buffer_dirty(sbh);
2413         if (sync) {
2414                 error = sync_dirty_buffer(sbh);
2415                 if (buffer_write_io_error(sbh)) {
2416                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2417                                "superblock");
2418                         clear_buffer_write_io_error(sbh);
2419                         set_buffer_uptodate(sbh);
2420                 }
2421         }
2422         return error;
2423 }
2424
2425
2426 /*
2427  * Have we just finished recovery?  If so, and if we are mounting (or
2428  * remounting) the filesystem readonly, then we will end up with a
2429  * consistent fs on disk.  Record that fact.
2430  */
2431 static void ext3_mark_recovery_complete(struct super_block * sb,
2432                                         struct ext3_super_block * es)
2433 {
2434         journal_t *journal = EXT3_SB(sb)->s_journal;
2435
2436         journal_lock_updates(journal);
2437         if (journal_flush(journal) < 0)
2438                 goto out;
2439
2440         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2441             sb->s_flags & MS_RDONLY) {
2442                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2443                 ext3_commit_super(sb, es, 1);
2444         }
2445
2446 out:
2447         journal_unlock_updates(journal);
2448 }
2449
2450 /*
2451  * If we are mounting (or read-write remounting) a filesystem whose journal
2452  * has recorded an error from a previous lifetime, move that error to the
2453  * main filesystem now.
2454  */
2455 static void ext3_clear_journal_err(struct super_block *sb,
2456                                    struct ext3_super_block *es)
2457 {
2458         journal_t *journal;
2459         int j_errno;
2460         const char *errstr;
2461
2462         journal = EXT3_SB(sb)->s_journal;
2463
2464         /*
2465          * Now check for any error status which may have been recorded in the
2466          * journal by a prior ext3_error() or ext3_abort()
2467          */
2468
2469         j_errno = journal_errno(journal);
2470         if (j_errno) {
2471                 char nbuf[16];
2472
2473                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2474                 ext3_warning(sb, __func__, "Filesystem error recorded "
2475                              "from previous mount: %s", errstr);
2476                 ext3_warning(sb, __func__, "Marking fs in need of "
2477                              "filesystem check.");
2478
2479                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2480                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2481                 ext3_commit_super (sb, es, 1);
2482
2483                 journal_clear_err(journal);
2484         }
2485 }
2486
2487 /*
2488  * Force the running and committing transactions to commit,
2489  * and wait on the commit.
2490  */
2491 int ext3_force_commit(struct super_block *sb)
2492 {
2493         journal_t *journal;
2494         int ret;
2495
2496         if (sb->s_flags & MS_RDONLY)
2497                 return 0;
2498
2499         journal = EXT3_SB(sb)->s_journal;
2500         ret = ext3_journal_force_commit(journal);
2501         return ret;
2502 }
2503
2504 static int ext3_sync_fs(struct super_block *sb, int wait)
2505 {
2506         tid_t target;
2507
2508         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2509                 if (wait)
2510                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2511         }
2512         return 0;
2513 }
2514
2515 /*
2516  * LVM calls this function before a (read-only) snapshot is created.  This
2517  * gives us a chance to flush the journal completely and mark the fs clean.
2518  */
2519 static int ext3_freeze(struct super_block *sb)
2520 {
2521         int error = 0;
2522         journal_t *journal;
2523
2524         if (!(sb->s_flags & MS_RDONLY)) {
2525                 journal = EXT3_SB(sb)->s_journal;
2526
2527                 /* Now we set up the journal barrier. */
2528                 journal_lock_updates(journal);
2529
2530                 /*
2531                  * We don't want to clear needs_recovery flag when we failed
2532                  * to flush the journal.
2533                  */
2534                 error = journal_flush(journal);
2535                 if (error < 0)
2536                         goto out;
2537
2538                 /* Journal blocked and flushed, clear needs_recovery flag. */
2539                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2540                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2541                 if (error)
2542                         goto out;
2543         }
2544         return 0;
2545
2546 out:
2547         journal_unlock_updates(journal);
2548         return error;
2549 }
2550
2551 /*
2552  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2553  * flag here, even though the filesystem is not technically dirty yet.
2554  */
2555 static int ext3_unfreeze(struct super_block *sb)
2556 {
2557         if (!(sb->s_flags & MS_RDONLY)) {
2558                 lock_super(sb);
2559                 /* Reser the needs_recovery flag before the fs is unlocked. */
2560                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2561                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2562                 unlock_super(sb);
2563                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2564         }
2565         return 0;
2566 }
2567
2568 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2569 {
2570         struct ext3_super_block * es;
2571         struct ext3_sb_info *sbi = EXT3_SB(sb);
2572         ext3_fsblk_t n_blocks_count = 0;
2573         unsigned long old_sb_flags;
2574         struct ext3_mount_options old_opts;
2575         int enable_quota = 0;
2576         int err;
2577 #ifdef CONFIG_QUOTA
2578         int i;
2579 #endif
2580
2581         /* Store the original options */
2582         lock_super(sb);
2583         old_sb_flags = sb->s_flags;
2584         old_opts.s_mount_opt = sbi->s_mount_opt;
2585         old_opts.s_resuid = sbi->s_resuid;
2586         old_opts.s_resgid = sbi->s_resgid;
2587         old_opts.s_commit_interval = sbi->s_commit_interval;
2588 #ifdef CONFIG_QUOTA
2589         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2590         for (i = 0; i < MAXQUOTAS; i++)
2591                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2592 #endif
2593
2594         /*
2595          * Allow the "check" option to be passed as a remount option.
2596          */
2597         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2598                 err = -EINVAL;
2599                 goto restore_opts;
2600         }
2601
2602         if (test_opt(sb, ABORT))
2603                 ext3_abort(sb, __func__, "Abort forced by user");
2604
2605         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2606                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2607
2608         es = sbi->s_es;
2609
2610         ext3_init_journal_params(sb, sbi->s_journal);
2611
2612         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2613                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2614                 if (test_opt(sb, ABORT)) {
2615                         err = -EROFS;
2616                         goto restore_opts;
2617                 }
2618
2619                 if (*flags & MS_RDONLY) {
2620                         err = dquot_suspend(sb, -1);
2621                         if (err < 0)
2622                                 goto restore_opts;
2623
2624                         /*
2625                          * First of all, the unconditional stuff we have to do
2626                          * to disable replay of the journal when we next remount
2627                          */
2628                         sb->s_flags |= MS_RDONLY;
2629
2630                         /*
2631                          * OK, test if we are remounting a valid rw partition
2632                          * readonly, and if so set the rdonly flag and then
2633                          * mark the partition as valid again.
2634                          */
2635                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2636                             (sbi->s_mount_state & EXT3_VALID_FS))
2637                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2638
2639                         ext3_mark_recovery_complete(sb, es);
2640                 } else {
2641                         __le32 ret;
2642                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2643                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2644                                 ext3_msg(sb, KERN_WARNING,
2645                                         "warning: couldn't remount RDWR "
2646                                         "because of unsupported optional "
2647                                         "features (%x)", le32_to_cpu(ret));
2648                                 err = -EROFS;
2649                                 goto restore_opts;
2650                         }
2651
2652                         /*
2653                          * If we have an unprocessed orphan list hanging
2654                          * around from a previously readonly bdev mount,
2655                          * require a full umount/remount for now.
2656                          */
2657                         if (es->s_last_orphan) {
2658                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2659                                        "remount RDWR because of unprocessed "
2660                                        "orphan inode list.  Please "
2661                                        "umount/remount instead.");
2662                                 err = -EINVAL;
2663                                 goto restore_opts;
2664                         }
2665
2666                         /*
2667                          * Mounting a RDONLY partition read-write, so reread
2668                          * and store the current valid flag.  (It may have
2669                          * been changed by e2fsck since we originally mounted
2670                          * the partition.)
2671                          */
2672                         ext3_clear_journal_err(sb, es);
2673                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2674                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2675                                 goto restore_opts;
2676                         if (!ext3_setup_super (sb, es, 0))
2677                                 sb->s_flags &= ~MS_RDONLY;
2678                         enable_quota = 1;
2679                 }
2680         }
2681 #ifdef CONFIG_QUOTA
2682         /* Release old quota file names */
2683         for (i = 0; i < MAXQUOTAS; i++)
2684                 if (old_opts.s_qf_names[i] &&
2685                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2686                         kfree(old_opts.s_qf_names[i]);
2687 #endif
2688         unlock_super(sb);
2689
2690         if (enable_quota)
2691                 dquot_resume(sb, -1);
2692         return 0;
2693 restore_opts:
2694         sb->s_flags = old_sb_flags;
2695         sbi->s_mount_opt = old_opts.s_mount_opt;
2696         sbi->s_resuid = old_opts.s_resuid;
2697         sbi->s_resgid = old_opts.s_resgid;
2698         sbi->s_commit_interval = old_opts.s_commit_interval;
2699 #ifdef CONFIG_QUOTA
2700         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2701         for (i = 0; i < MAXQUOTAS; i++) {
2702                 if (sbi->s_qf_names[i] &&
2703                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2704                         kfree(sbi->s_qf_names[i]);
2705                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2706         }
2707 #endif
2708         unlock_super(sb);
2709         return err;
2710 }
2711
2712 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2713 {
2714         struct super_block *sb = dentry->d_sb;
2715         struct ext3_sb_info *sbi = EXT3_SB(sb);
2716         struct ext3_super_block *es = sbi->s_es;
2717         u64 fsid;
2718
2719         if (test_opt(sb, MINIX_DF)) {
2720                 sbi->s_overhead_last = 0;
2721         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2722                 unsigned long ngroups = sbi->s_groups_count, i;
2723                 ext3_fsblk_t overhead = 0;
2724                 smp_rmb();
2725
2726                 /*
2727                  * Compute the overhead (FS structures).  This is constant
2728                  * for a given filesystem unless the number of block groups
2729                  * changes so we cache the previous value until it does.
2730                  */
2731
2732                 /*
2733                  * All of the blocks before first_data_block are
2734                  * overhead
2735                  */
2736                 overhead = le32_to_cpu(es->s_first_data_block);
2737
2738                 /*
2739                  * Add the overhead attributed to the superblock and
2740                  * block group descriptors.  If the sparse superblocks
2741                  * feature is turned on, then not all groups have this.
2742                  */
2743                 for (i = 0; i < ngroups; i++) {
2744                         overhead += ext3_bg_has_super(sb, i) +
2745                                 ext3_bg_num_gdb(sb, i);
2746                         cond_resched();
2747                 }
2748
2749                 /*
2750                  * Every block group has an inode bitmap, a block
2751                  * bitmap, and an inode table.
2752                  */
2753                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2754                 sbi->s_overhead_last = overhead;
2755                 smp_wmb();
2756                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2757         }
2758
2759         buf->f_type = EXT3_SUPER_MAGIC;
2760         buf->f_bsize = sb->s_blocksize;
2761         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2762         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2763         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2764         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2765                 buf->f_bavail = 0;
2766         buf->f_files = le32_to_cpu(es->s_inodes_count);
2767         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2768         buf->f_namelen = EXT3_NAME_LEN;
2769         fsid = le64_to_cpup((void *)es->s_uuid) ^
2770                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2771         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2772         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2773         return 0;
2774 }
2775
2776 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2777  * is locked for write. Otherwise the are possible deadlocks:
2778  * Process 1                         Process 2
2779  * ext3_create()                     quota_sync()
2780  *   journal_start()                   write_dquot()
2781  *   dquot_initialize()                       down(dqio_mutex)
2782  *     down(dqio_mutex)                    journal_start()
2783  *
2784  */
2785
2786 #ifdef CONFIG_QUOTA
2787
2788 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2789 {
2790         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2791 }
2792
2793 static int ext3_write_dquot(struct dquot *dquot)
2794 {
2795         int ret, err;
2796         handle_t *handle;
2797         struct inode *inode;
2798
2799         inode = dquot_to_inode(dquot);
2800         handle = ext3_journal_start(inode,
2801                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2802         if (IS_ERR(handle))
2803                 return PTR_ERR(handle);
2804         ret = dquot_commit(dquot);
2805         err = ext3_journal_stop(handle);
2806         if (!ret)
2807                 ret = err;
2808         return ret;
2809 }
2810
2811 static int ext3_acquire_dquot(struct dquot *dquot)
2812 {
2813         int ret, err;
2814         handle_t *handle;
2815
2816         handle = ext3_journal_start(dquot_to_inode(dquot),
2817                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2818         if (IS_ERR(handle))
2819                 return PTR_ERR(handle);
2820         ret = dquot_acquire(dquot);
2821         err = ext3_journal_stop(handle);
2822         if (!ret)
2823                 ret = err;
2824         return ret;
2825 }
2826
2827 static int ext3_release_dquot(struct dquot *dquot)
2828 {
2829         int ret, err;
2830         handle_t *handle;
2831
2832         handle = ext3_journal_start(dquot_to_inode(dquot),
2833                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2834         if (IS_ERR(handle)) {
2835                 /* Release dquot anyway to avoid endless cycle in dqput() */
2836                 dquot_release(dquot);
2837                 return PTR_ERR(handle);
2838         }
2839         ret = dquot_release(dquot);
2840         err = ext3_journal_stop(handle);
2841         if (!ret)
2842                 ret = err;
2843         return ret;
2844 }
2845
2846 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2847 {
2848         /* Are we journaling quotas? */
2849         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2850             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2851                 dquot_mark_dquot_dirty(dquot);
2852                 return ext3_write_dquot(dquot);
2853         } else {
2854                 return dquot_mark_dquot_dirty(dquot);
2855         }
2856 }
2857
2858 static int ext3_write_info(struct super_block *sb, int type)
2859 {
2860         int ret, err;
2861         handle_t *handle;
2862
2863         /* Data block + inode block */
2864         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2865         if (IS_ERR(handle))
2866                 return PTR_ERR(handle);
2867         ret = dquot_commit_info(sb, type);
2868         err = ext3_journal_stop(handle);
2869         if (!ret)
2870                 ret = err;
2871         return ret;
2872 }
2873
2874 /*
2875  * Turn on quotas during mount time - we need to find
2876  * the quota file and such...
2877  */
2878 static int ext3_quota_on_mount(struct super_block *sb, int type)
2879 {
2880         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2881                                         EXT3_SB(sb)->s_jquota_fmt, type);
2882 }
2883
2884 /*
2885  * Standard function to be called on quota_on
2886  */
2887 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2888                          struct path *path)
2889 {
2890         int err;
2891
2892         if (!test_opt(sb, QUOTA))
2893                 return -EINVAL;
2894
2895         /* Quotafile not on the same filesystem? */
2896         if (path->mnt->mnt_sb != sb)
2897                 return -EXDEV;
2898         /* Journaling quota? */
2899         if (EXT3_SB(sb)->s_qf_names[type]) {
2900                 /* Quotafile not of fs root? */
2901                 if (path->dentry->d_parent != sb->s_root)
2902                         ext3_msg(sb, KERN_WARNING,
2903                                 "warning: Quota file not on filesystem root. "
2904                                 "Journaled quota will not work.");
2905         }
2906
2907         /*
2908          * When we journal data on quota file, we have to flush journal to see
2909          * all updates to the file when we bypass pagecache...
2910          */
2911         if (ext3_should_journal_data(path->dentry->d_inode)) {
2912                 /*
2913                  * We don't need to lock updates but journal_flush() could
2914                  * otherwise be livelocked...
2915                  */
2916                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2917                 err = journal_flush(EXT3_SB(sb)->s_journal);
2918                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2919                 if (err)
2920                         return err;
2921         }
2922
2923         return dquot_quota_on(sb, type, format_id, path);
2924 }
2925
2926 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2927  * acquiring the locks... As quota files are never truncated and quota code
2928  * itself serializes the operations (and noone else should touch the files)
2929  * we don't have to be afraid of races */
2930 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2931                                size_t len, loff_t off)
2932 {
2933         struct inode *inode = sb_dqopt(sb)->files[type];
2934         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2935         int err = 0;
2936         int offset = off & (sb->s_blocksize - 1);
2937         int tocopy;
2938         size_t toread;
2939         struct buffer_head *bh;
2940         loff_t i_size = i_size_read(inode);
2941
2942         if (off > i_size)
2943                 return 0;
2944         if (off+len > i_size)
2945                 len = i_size-off;
2946         toread = len;
2947         while (toread > 0) {
2948                 tocopy = sb->s_blocksize - offset < toread ?
2949                                 sb->s_blocksize - offset : toread;
2950                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2951                 if (err)
2952                         return err;
2953                 if (!bh)        /* A hole? */
2954                         memset(data, 0, tocopy);
2955                 else
2956                         memcpy(data, bh->b_data+offset, tocopy);
2957                 brelse(bh);
2958                 offset = 0;
2959                 toread -= tocopy;
2960                 data += tocopy;
2961                 blk++;
2962         }
2963         return len;
2964 }
2965
2966 /* Write to quotafile (we know the transaction is already started and has
2967  * enough credits) */
2968 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2969                                 const char *data, size_t len, loff_t off)
2970 {
2971         struct inode *inode = sb_dqopt(sb)->files[type];
2972         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2973         int err = 0;
2974         int offset = off & (sb->s_blocksize - 1);
2975         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2976         struct buffer_head *bh;
2977         handle_t *handle = journal_current_handle();
2978
2979         if (!handle) {
2980                 ext3_msg(sb, KERN_WARNING,
2981                         "warning: quota write (off=%llu, len=%llu)"
2982                         " cancelled because transaction is not started.",
2983                         (unsigned long long)off, (unsigned long long)len);
2984                 return -EIO;
2985         }
2986
2987         /*
2988          * Since we account only one data block in transaction credits,
2989          * then it is impossible to cross a block boundary.
2990          */
2991         if (sb->s_blocksize - offset < len) {
2992                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2993                         " cancelled because not block aligned",
2994                         (unsigned long long)off, (unsigned long long)len);
2995                 return -EIO;
2996         }
2997         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2998         bh = ext3_bread(handle, inode, blk, 1, &err);
2999         if (!bh)
3000                 goto out;
3001         if (journal_quota) {
3002                 err = ext3_journal_get_write_access(handle, bh);
3003                 if (err) {
3004                         brelse(bh);
3005                         goto out;
3006                 }
3007         }
3008         lock_buffer(bh);
3009         memcpy(bh->b_data+offset, data, len);
3010         flush_dcache_page(bh->b_page);
3011         unlock_buffer(bh);
3012         if (journal_quota)
3013                 err = ext3_journal_dirty_metadata(handle, bh);
3014         else {
3015                 /* Always do at least ordered writes for quotas */
3016                 err = ext3_journal_dirty_data(handle, bh);
3017                 mark_buffer_dirty(bh);
3018         }
3019         brelse(bh);
3020 out:
3021         if (err) {
3022                 mutex_unlock(&inode->i_mutex);
3023                 return err;
3024         }
3025         if (inode->i_size < off + len) {
3026                 i_size_write(inode, off + len);
3027                 EXT3_I(inode)->i_disksize = inode->i_size;
3028         }
3029         inode->i_version++;
3030         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3031         ext3_mark_inode_dirty(handle, inode);
3032         mutex_unlock(&inode->i_mutex);
3033         return len;
3034 }
3035
3036 #endif
3037
3038 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3039         int flags, const char *dev_name, void *data)
3040 {
3041         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3042 }
3043
3044 static struct file_system_type ext3_fs_type = {
3045         .owner          = THIS_MODULE,
3046         .name           = "ext3",
3047         .mount          = ext3_mount,
3048         .kill_sb        = kill_block_super,
3049         .fs_flags       = FS_REQUIRES_DEV,
3050 };
3051
3052 static int __init init_ext3_fs(void)
3053 {
3054         int err = init_ext3_xattr();
3055         if (err)
3056                 return err;
3057         err = init_inodecache();
3058         if (err)
3059                 goto out1;
3060         err = register_filesystem(&ext3_fs_type);
3061         if (err)
3062                 goto out;
3063         return 0;
3064 out:
3065         destroy_inodecache();
3066 out1:
3067         exit_ext3_xattr();
3068         return err;
3069 }
3070
3071 static void __exit exit_ext3_fs(void)
3072 {
3073         unregister_filesystem(&ext3_fs_type);
3074         destroy_inodecache();
3075         exit_ext3_xattr();
3076 }
3077
3078 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3079 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3080 MODULE_LICENSE("GPL");
3081 module_init(init_ext3_fs)
3082 module_exit(exit_ext3_fs)