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/blkdev.h>
  21 #include <linux/parser.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/statfs.h>
  24 #include <linux/random.h>
  25 #include <linux/mount.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/log2.h>
  29 #include <linux/cleancache.h>
  30 #include <linux/namei.h>
  31
  32 #include <asm/uaccess.h>
  33
  34 #define CREATE_TRACE_POINTS
  35
  36 #include "ext3.h"
  37 #include "xattr.h"
  38 #include "acl.h"
  39 #include "namei.h"
  40
  41 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  42   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  43 #else
  44   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  45 #endif
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static int ext3_commit_super(struct super_block *sb,
  52                                struct ext3_super_block *es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static int ext3_unfreeze(struct super_block *sb);
  64 static int ext3_freeze(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  68  */
  69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  70 {
  71         journal_t *journal;
  72
  73         if (sb->s_flags & MS_RDONLY)
  74                 return ERR_PTR(-EROFS);
  75
  76         /* Special case here: if the journal has aborted behind our
  77          * backs (eg. EIO in the commit thread), then we still need to
  78          * take the FS itself readonly cleanly. */
  79         journal = EXT3_SB(sb)->s_journal;
  80         if (is_journal_aborted(journal)) {
  81                 ext3_abort(sb, __func__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 int __ext3_journal_stop(const char *where, handle_t *handle)
  90 {
  91         struct super_block *sb;
  92         int err;
  93         int rc;
  94
  95         sb = handle->h_transaction->t_journal->j_private;
  96         err = handle->h_err;
  97         rc = journal_stop(handle);
  98
  99         if (!err)
 100                 err = rc;
 101         if (err)
 102                 __ext3_std_error(sb, where, err);
 103         return err;
 104 }
 105
 106 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 107                 struct buffer_head *bh, handle_t *handle, int err)
 108 {
 109         char nbuf[16];
 110         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 111
 112         if (bh)
 113                 BUFFER_TRACE(bh, "abort");
 114
 115         if (!handle->h_err)
 116                 handle->h_err = err;
 117
 118         if (is_handle_aborted(handle))
 119                 return;
 120
 121         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 122                 caller, errstr, err_fn);
 123
 124         journal_abort_handle(handle);
 125 }
 126
 127 void ext3_msg(struct super_block *sb, const char *prefix,
 128                 const char *fmt, ...)
 129 {
 130         struct va_format vaf;
 131         va_list args;
 132
 133         va_start(args, fmt);
 134
 135         vaf.fmt = fmt;
 136         vaf.va = &args;
 137
 138         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 139
 140         va_end(args);
 141 }
 142
 143 /* Deal with the reporting of failure conditions on a filesystem such as
 144  * inconsistencies detected or read IO failures.
 145  *
 146  * On ext2, we can store the error state of the filesystem in the
 147  * superblock.  That is not possible on ext3, because we may have other
 148  * write ordering constraints on the superblock which prevent us from
 149  * writing it out straight away; and given that the journal is about to
 150  * be aborted, we can't rely on the current, or future, transactions to
 151  * write out the superblock safely.
 152  *
 153  * We'll just use the journal_abort() error code to record an error in
 154  * the journal instead.  On recovery, the journal will complain about
 155  * that error until we've noted it down and cleared it.
 156  */
 157
 158 static void ext3_handle_error(struct super_block *sb)
 159 {
 160         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 161
 162         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 163         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 164
 165         if (sb->s_flags & MS_RDONLY)
 166                 return;
 167
 168         if (!test_opt (sb, ERRORS_CONT)) {
 169                 journal_t *journal = EXT3_SB(sb)->s_journal;
 170
 171                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 172                 if (journal)
 173                         journal_abort(journal, -EIO);
 174         }
 175         if (test_opt (sb, ERRORS_RO)) {
 176                 ext3_msg(sb, KERN_CRIT,
 177                         "error: remounting filesystem read-only");
 178                 /*
 179                  * Make sure updated value of ->s_mount_state will be visible
 180                  * before ->s_flags update.
 181                  */
 182                 smp_wmb();
 183                 sb->s_flags |= MS_RDONLY;
 184         }
 185         ext3_commit_super(sb, es, 1);
 186         if (test_opt(sb, ERRORS_PANIC))
 187                 panic("EXT3-fs (%s): panic forced after error\n",
 188                         sb->s_id);
 189 }
 190
 191 void ext3_error(struct super_block *sb, const char *function,
 192                 const char *fmt, ...)
 193 {
 194         struct va_format vaf;
 195         va_list args;
 196
 197         va_start(args, fmt);
 198
 199         vaf.fmt = fmt;
 200         vaf.va = &args;
 201
 202         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 203                sb->s_id, function, &vaf);
 204
 205         va_end(args);
 206
 207         ext3_handle_error(sb);
 208 }
 209
 210 static const char *ext3_decode_error(struct super_block * sb, int errno,
 211                                      char nbuf[16])
 212 {
 213         char *errstr = NULL;
 214
 215         switch (errno) {
 216         case -EIO:
 217                 errstr = "IO failure";
 218                 break;
 219         case -ENOMEM:
 220                 errstr = "Out of memory";
 221                 break;
 222         case -EROFS:
 223                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 224                         errstr = "Journal has aborted";
 225                 else
 226                         errstr = "Readonly filesystem";
 227                 break;
 228         default:
 229                 /* If the caller passed in an extra buffer for unknown
 230                  * errors, textualise them now.  Else we just return
 231                  * NULL. */
 232                 if (nbuf) {
 233                         /* Check for truncated error codes... */
 234                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 235                                 errstr = nbuf;
 236                 }
 237                 break;
 238         }
 239
 240         return errstr;
 241 }
 242
 243 /* __ext3_std_error decodes expected errors from journaling functions
 244  * automatically and invokes the appropriate error response.  */
 245
 246 void __ext3_std_error (struct super_block * sb, const char * function,
 247                        int errno)
 248 {
 249         char nbuf[16];
 250         const char *errstr;
 251
 252         /* Special case: if the error is EROFS, and we're not already
 253          * inside a transaction, then there's really no point in logging
 254          * an error. */
 255         if (errno == -EROFS && journal_current_handle() == NULL &&
 256             (sb->s_flags & MS_RDONLY))
 257                 return;
 258
 259         errstr = ext3_decode_error(sb, errno, nbuf);
 260         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 261
 262         ext3_handle_error(sb);
 263 }
 264
 265 /*
 266  * ext3_abort is a much stronger failure handler than ext3_error.  The
 267  * abort function may be used to deal with unrecoverable failures such
 268  * as journal IO errors or ENOMEM at a critical moment in log management.
 269  *
 270  * We unconditionally force the filesystem into an ABORT|READONLY state,
 271  * unless the error response on the fs has been set to panic in which
 272  * case we take the easy way out and panic immediately.
 273  */
 274
 275 void ext3_abort(struct super_block *sb, const char *function,
 276                  const char *fmt, ...)
 277 {
 278         struct va_format vaf;
 279         va_list args;
 280
 281         va_start(args, fmt);
 282
 283         vaf.fmt = fmt;
 284         vaf.va = &args;
 285
 286         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 287                sb->s_id, function, &vaf);
 288
 289         va_end(args);
 290
 291         if (test_opt(sb, ERRORS_PANIC))
 292                 panic("EXT3-fs: panic from previous error\n");
 293
 294         if (sb->s_flags & MS_RDONLY)
 295                 return;
 296
 297         ext3_msg(sb, KERN_CRIT,
 298                 "error: remounting filesystem read-only");
 299         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 300         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 301         /*
 302          * Make sure updated value of ->s_mount_state will be visible
 303          * before ->s_flags update.
 304          */
 305         smp_wmb();
 306         sb->s_flags |= MS_RDONLY;
 307
 308         if (EXT3_SB(sb)->s_journal)
 309                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 310 }
 311
 312 void ext3_warning(struct super_block *sb, const char *function,
 313                   const char *fmt, ...)
 314 {
 315         struct va_format vaf;
 316         va_list args;
 317
 318         va_start(args, fmt);
 319
 320         vaf.fmt = fmt;
 321         vaf.va = &args;
 322
 323         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 324                sb->s_id, function, &vaf);
 325
 326         va_end(args);
 327 }
 328
 329 void ext3_update_dynamic_rev(struct super_block *sb)
 330 {
 331         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 332
 333         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 334                 return;
 335
 336         ext3_msg(sb, KERN_WARNING,
 337                 "warning: updating to rev %d because of "
 338                 "new feature flag, running e2fsck is recommended",
 339                 EXT3_DYNAMIC_REV);
 340
 341         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 342         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 343         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 344         /* leave es->s_feature_*compat flags alone */
 345         /* es->s_uuid will be set by e2fsck if empty */
 346
 347         /*
 348          * The rest of the superblock fields should be zero, and if not it
 349          * means they are likely already in use, so leave them alone.  We
 350          * can leave it up to e2fsck to clean up any inconsistencies there.
 351          */
 352 }
 353
 354 /*
 355  * Open the external journal device
 356  */
 357 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 358 {
 359         struct block_device *bdev;
 360         char b[BDEVNAME_SIZE];
 361
 362         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 363         if (IS_ERR(bdev))
 364                 goto fail;
 365         return bdev;
 366
 367 fail:
 368         ext3_msg(sb, KERN_ERR, "error: failed to open journal device %s: %ld",
 369                 __bdevname(dev, b), PTR_ERR(bdev));
 370
 371         return NULL;
 372 }
 373
 374 /*
 375  * Release the journal device
 376  */
 377 static void ext3_blkdev_put(struct block_device *bdev)
 378 {
 379         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 380 }
 381
 382 static void ext3_blkdev_remove(struct ext3_sb_info *sbi)
 383 {
 384         struct block_device *bdev;
 385         bdev = sbi->journal_bdev;
 386         if (bdev) {
 387                 ext3_blkdev_put(bdev);
 388                 sbi->journal_bdev = NULL;
 389         }
 390 }
 391
 392 static inline struct inode *orphan_list_entry(struct list_head *l)
 393 {
 394         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 395 }
 396
 397 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 398 {
 399         struct list_head *l;
 400
 401         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 402                le32_to_cpu(sbi->s_es->s_last_orphan));
 403
 404         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 405         list_for_each(l, &sbi->s_orphan) {
 406                 struct inode *inode = orphan_list_entry(l);
 407                 ext3_msg(sb, KERN_ERR, "  "
 408                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 409                        inode->i_sb->s_id, inode->i_ino, inode,
 410                        inode->i_mode, inode->i_nlink,
 411                        NEXT_ORPHAN(inode));
 412         }
 413 }
 414
 415 static void ext3_put_super (struct super_block * sb)
 416 {
 417         struct ext3_sb_info *sbi = EXT3_SB(sb);
 418         struct ext3_super_block *es = sbi->s_es;
 419         int i, err;
 420
 421         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 422         ext3_xattr_put_super(sb);
 423         err = journal_destroy(sbi->s_journal);
 424         sbi->s_journal = NULL;
 425         if (err < 0)
 426                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 427
 428         if (!(sb->s_flags & MS_RDONLY)) {
 429                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 430                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 431                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 432                 mark_buffer_dirty(sbi->s_sbh);
 433                 ext3_commit_super(sb, es, 1);
 434         }
 435
 436         for (i = 0; i < sbi->s_gdb_count; i++)
 437                 brelse(sbi->s_group_desc[i]);
 438         kfree(sbi->s_group_desc);
 439         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 440         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 441         percpu_counter_destroy(&sbi->s_dirs_counter);
 442         brelse(sbi->s_sbh);
 443 #ifdef CONFIG_QUOTA
 444         for (i = 0; i < EXT3_MAXQUOTAS; i++)
 445                 kfree(sbi->s_qf_names[i]);
 446 #endif
 447
 448         /* Debugging code just in case the in-memory inode orphan list
 449          * isn't empty.  The on-disk one can be non-empty if we've
 450          * detected an error and taken the fs readonly, but the
 451          * in-memory list had better be clean by this point. */
 452         if (!list_empty(&sbi->s_orphan))
 453                 dump_orphan_list(sb, sbi);
 454         J_ASSERT(list_empty(&sbi->s_orphan));
 455
 456         invalidate_bdev(sb->s_bdev);
 457         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 458                 /*
 459                  * Invalidate the journal device's buffers.  We don't want them
 460                  * floating about in memory - the physical journal device may
 461                  * hotswapped, and it breaks the `ro-after' testing code.
 462                  */
 463                 sync_blockdev(sbi->journal_bdev);
 464                 invalidate_bdev(sbi->journal_bdev);
 465                 ext3_blkdev_remove(sbi);
 466         }
 467         sb->s_fs_info = NULL;
 468         kfree(sbi->s_blockgroup_lock);
 469         kfree(sbi);
 470 }
 471
 472 static struct kmem_cache *ext3_inode_cachep;
 473
 474 /*
 475  * Called inside transaction, so use GFP_NOFS
 476  */
 477 static struct inode *ext3_alloc_inode(struct super_block *sb)
 478 {
 479         struct ext3_inode_info *ei;
 480
 481         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 482         if (!ei)
 483                 return NULL;
 484         ei->i_block_alloc_info = NULL;
 485         ei->vfs_inode.i_version = 1;
 486         atomic_set(&ei->i_datasync_tid, 0);
 487         atomic_set(&ei->i_sync_tid, 0);
 488 #ifdef CONFIG_QUOTA
 489         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 490 #endif
 491
 492         return &ei->vfs_inode;
 493 }
 494
 495 static int ext3_drop_inode(struct inode *inode)
 496 {
 497         int drop = generic_drop_inode(inode);
 498
 499         trace_ext3_drop_inode(inode, drop);
 500         return drop;
 501 }
 502
 503 static void ext3_i_callback(struct rcu_head *head)
 504 {
 505         struct inode *inode = container_of(head, struct inode, i_rcu);
 506         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 507 }
 508
 509 static void ext3_destroy_inode(struct inode *inode)
 510 {
 511         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 512                 printk("EXT3 Inode %p: orphan list check failed!\n",
 513                         EXT3_I(inode));
 514                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 515                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 516                                 false);
 517                 dump_stack();
 518         }
 519         call_rcu(&inode->i_rcu, ext3_i_callback);
 520 }
 521
 522 static void init_once(void *foo)
 523 {
 524         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 525
 526         INIT_LIST_HEAD(&ei->i_orphan);
 527 #ifdef CONFIG_EXT3_FS_XATTR
 528         init_rwsem(&ei->xattr_sem);
 529 #endif
 530         mutex_init(&ei->truncate_mutex);
 531         inode_init_once(&ei->vfs_inode);
 532 }
 533
 534 static int __init init_inodecache(void)
 535 {
 536         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 537                                              sizeof(struct ext3_inode_info),
 538                                              0, (SLAB_RECLAIM_ACCOUNT|
 539                                                 SLAB_MEM_SPREAD),
 540                                              init_once);
 541         if (ext3_inode_cachep == NULL)
 542                 return -ENOMEM;
 543         return 0;
 544 }
 545
 546 static void destroy_inodecache(void)
 547 {
 548         /*
 549          * Make sure all delayed rcu free inodes are flushed before we
 550          * destroy cache.
 551          */
 552         rcu_barrier();
 553         kmem_cache_destroy(ext3_inode_cachep);
 554 }
 555
 556 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 557 {
 558 #if defined(CONFIG_QUOTA)
 559         struct ext3_sb_info *sbi = EXT3_SB(sb);
 560
 561         if (sbi->s_jquota_fmt) {
 562                 char *fmtname = "";
 563
 564                 switch (sbi->s_jquota_fmt) {
 565                 case QFMT_VFS_OLD:
 566                         fmtname = "vfsold";
 567                         break;
 568                 case QFMT_VFS_V0:
 569                         fmtname = "vfsv0";
 570                         break;
 571                 case QFMT_VFS_V1:
 572                         fmtname = "vfsv1";
 573                         break;
 574                 }
 575                 seq_printf(seq, ",jqfmt=%s", fmtname);
 576         }
 577
 578         if (sbi->s_qf_names[USRQUOTA])
 579                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 580
 581         if (sbi->s_qf_names[GRPQUOTA])
 582                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 583
 584         if (test_opt(sb, USRQUOTA))
 585                 seq_puts(seq, ",usrquota");
 586
 587         if (test_opt(sb, GRPQUOTA))
 588                 seq_puts(seq, ",grpquota");
 589 #endif
 590 }
 591
 592 static char *data_mode_string(unsigned long mode)
 593 {
 594         switch (mode) {
 595         case EXT3_MOUNT_JOURNAL_DATA:
 596                 return "journal";
 597         case EXT3_MOUNT_ORDERED_DATA:
 598                 return "ordered";
 599         case EXT3_MOUNT_WRITEBACK_DATA:
 600                 return "writeback";
 601         }
 602         return "unknown";
 603 }
 604
 605 /*
 606  * Show an option if
 607  *  - it's set to a non-default value OR
 608  *  - if the per-sb default is different from the global default
 609  */
 610 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 611 {
 612         struct super_block *sb = root->d_sb;
 613         struct ext3_sb_info *sbi = EXT3_SB(sb);
 614         struct ext3_super_block *es = sbi->s_es;
 615         unsigned long def_mount_opts;
 616
 617         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 618
 619         if (sbi->s_sb_block != 1)
 620                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 621         if (test_opt(sb, MINIX_DF))
 622                 seq_puts(seq, ",minixdf");
 623         if (test_opt(sb, GRPID))
 624                 seq_puts(seq, ",grpid");
 625         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 626                 seq_puts(seq, ",nogrpid");
 627         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
 628             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 629                 seq_printf(seq, ",resuid=%u",
 630                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
 631         }
 632         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
 633             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 634                 seq_printf(seq, ",resgid=%u",
 635                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
 636         }
 637         if (test_opt(sb, ERRORS_RO)) {
 638                 int def_errors = le16_to_cpu(es->s_errors);
 639
 640                 if (def_errors == EXT3_ERRORS_PANIC ||
 641                     def_errors == EXT3_ERRORS_CONTINUE) {
 642                         seq_puts(seq, ",errors=remount-ro");
 643                 }
 644         }
 645         if (test_opt(sb, ERRORS_CONT))
 646                 seq_puts(seq, ",errors=continue");
 647         if (test_opt(sb, ERRORS_PANIC))
 648                 seq_puts(seq, ",errors=panic");
 649         if (test_opt(sb, NO_UID32))
 650                 seq_puts(seq, ",nouid32");
 651         if (test_opt(sb, DEBUG))
 652                 seq_puts(seq, ",debug");
 653 #ifdef CONFIG_EXT3_FS_XATTR
 654         if (test_opt(sb, XATTR_USER))
 655                 seq_puts(seq, ",user_xattr");
 656         if (!test_opt(sb, XATTR_USER) &&
 657             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 658                 seq_puts(seq, ",nouser_xattr");
 659         }
 660 #endif
 661 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 662         if (test_opt(sb, POSIX_ACL))
 663                 seq_puts(seq, ",acl");
 664         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 665                 seq_puts(seq, ",noacl");
 666 #endif
 667         if (!test_opt(sb, RESERVATION))
 668                 seq_puts(seq, ",noreservation");
 669         if (sbi->s_commit_interval) {
 670                 seq_printf(seq, ",commit=%u",
 671                            (unsigned) (sbi->s_commit_interval / HZ));
 672         }
 673
 674         /*
 675          * Always display barrier state so it's clear what the status is.
 676          */
 677         seq_puts(seq, ",barrier=");
 678         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 679         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 680         if (test_opt(sb, DATA_ERR_ABORT))
 681                 seq_puts(seq, ",data_err=abort");
 682
 683         if (test_opt(sb, NOLOAD))
 684                 seq_puts(seq, ",norecovery");
 685
 686         ext3_show_quota_options(seq, sb);
 687
 688         return 0;
 689 }
 690
 691
 692 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 693                 u64 ino, u32 generation)
 694 {
 695         struct inode *inode;
 696
 697         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 698                 return ERR_PTR(-ESTALE);
 699         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 700                 return ERR_PTR(-ESTALE);
 701
 702         /* iget isn't really right if the inode is currently unallocated!!
 703          *
 704          * ext3_read_inode will return a bad_inode if the inode had been
 705          * deleted, so we should be safe.
 706          *
 707          * Currently we don't know the generation for parent directory, so
 708          * a generation of 0 means "accept any"
 709          */
 710         inode = ext3_iget(sb, ino);
 711         if (IS_ERR(inode))
 712                 return ERR_CAST(inode);
 713         if (generation && inode->i_generation != generation) {
 714                 iput(inode);
 715                 return ERR_PTR(-ESTALE);
 716         }
 717
 718         return inode;
 719 }
 720
 721 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 722                 int fh_len, int fh_type)
 723 {
 724         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 725                                     ext3_nfs_get_inode);
 726 }
 727
 728 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 729                 int fh_len, int fh_type)
 730 {
 731         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 732                                     ext3_nfs_get_inode);
 733 }
 734
 735 /*
 736  * Try to release metadata pages (indirect blocks, directories) which are
 737  * mapped via the block device.  Since these pages could have journal heads
 738  * which would prevent try_to_free_buffers() from freeing them, we must use
 739  * jbd layer's try_to_free_buffers() function to release them.
 740  */
 741 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 742                                  gfp_t wait)
 743 {
 744         journal_t *journal = EXT3_SB(sb)->s_journal;
 745
 746         WARN_ON(PageChecked(page));
 747         if (!page_has_buffers(page))
 748                 return 0;
 749         if (journal)
 750                 return journal_try_to_free_buffers(journal, page,
 751                                                    wait & ~__GFP_WAIT);
 752         return try_to_free_buffers(page);
 753 }
 754
 755 #ifdef CONFIG_QUOTA
 756 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 757 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 758
 759 static int ext3_write_dquot(struct dquot *dquot);
 760 static int ext3_acquire_dquot(struct dquot *dquot);
 761 static int ext3_release_dquot(struct dquot *dquot);
 762 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 763 static int ext3_write_info(struct super_block *sb, int type);
 764 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 765                          struct path *path);
 766 static int ext3_quota_on_mount(struct super_block *sb, int type);
 767 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 768                                size_t len, loff_t off);
 769 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 770                                 const char *data, size_t len, loff_t off);
 771 static struct dquot **ext3_get_dquots(struct inode *inode)
 772 {
 773         return EXT3_I(inode)->i_dquot;
 774 }
 775
 776 static const struct dquot_operations ext3_quota_operations = {
 777         .write_dquot    = ext3_write_dquot,
 778         .acquire_dquot  = ext3_acquire_dquot,
 779         .release_dquot  = ext3_release_dquot,
 780         .mark_dirty     = ext3_mark_dquot_dirty,
 781         .write_info     = ext3_write_info,
 782         .alloc_dquot    = dquot_alloc,
 783         .destroy_dquot  = dquot_destroy,
 784 };
 785
 786 static const struct quotactl_ops ext3_qctl_operations = {
 787         .quota_on       = ext3_quota_on,
 788         .quota_off      = dquot_quota_off,
 789         .quota_sync     = dquot_quota_sync,
 790         .get_info       = dquot_get_dqinfo,
 791         .set_info       = dquot_set_dqinfo,
 792         .get_dqblk      = dquot_get_dqblk,
 793         .set_dqblk      = dquot_set_dqblk
 794 };
 795 #endif
 796
 797 static const struct super_operations ext3_sops = {
 798         .alloc_inode    = ext3_alloc_inode,
 799         .destroy_inode  = ext3_destroy_inode,
 800         .write_inode    = ext3_write_inode,
 801         .dirty_inode    = ext3_dirty_inode,
 802         .drop_inode     = ext3_drop_inode,
 803         .evict_inode    = ext3_evict_inode,
 804         .put_super      = ext3_put_super,
 805         .sync_fs        = ext3_sync_fs,
 806         .freeze_fs      = ext3_freeze,
 807         .unfreeze_fs    = ext3_unfreeze,
 808         .statfs         = ext3_statfs,
 809         .remount_fs     = ext3_remount,
 810         .show_options   = ext3_show_options,
 811 #ifdef CONFIG_QUOTA
 812         .quota_read     = ext3_quota_read,
 813         .quota_write    = ext3_quota_write,
 814         .get_dquots     = ext3_get_dquots,
 815 #endif
 816         .bdev_try_to_free_page = bdev_try_to_free_page,
 817 };
 818
 819 static const struct export_operations ext3_export_ops = {
 820         .fh_to_dentry = ext3_fh_to_dentry,
 821         .fh_to_parent = ext3_fh_to_parent,
 822         .get_parent = ext3_get_parent,
 823 };
 824
 825 enum {
 826         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 827         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 828         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 829         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 830         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 831         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 832         Opt_journal_path,
 833         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 834         Opt_data_err_abort, Opt_data_err_ignore,
 835         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 836         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 837         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 838         Opt_resize, Opt_usrquota, Opt_grpquota
 839 };
 840
 841 static const match_table_t tokens = {
 842         {Opt_bsd_df, "bsddf"},
 843         {Opt_minix_df, "minixdf"},
 844         {Opt_grpid, "grpid"},
 845         {Opt_grpid, "bsdgroups"},
 846         {Opt_nogrpid, "nogrpid"},
 847         {Opt_nogrpid, "sysvgroups"},
 848         {Opt_resgid, "resgid=%u"},
 849         {Opt_resuid, "resuid=%u"},
 850         {Opt_sb, "sb=%u"},
 851         {Opt_err_cont, "errors=continue"},
 852         {Opt_err_panic, "errors=panic"},
 853         {Opt_err_ro, "errors=remount-ro"},
 854         {Opt_nouid32, "nouid32"},
 855         {Opt_nocheck, "nocheck"},
 856         {Opt_nocheck, "check=none"},
 857         {Opt_debug, "debug"},
 858         {Opt_oldalloc, "oldalloc"},
 859         {Opt_orlov, "orlov"},
 860         {Opt_user_xattr, "user_xattr"},
 861         {Opt_nouser_xattr, "nouser_xattr"},
 862         {Opt_acl, "acl"},
 863         {Opt_noacl, "noacl"},
 864         {Opt_reservation, "reservation"},
 865         {Opt_noreservation, "noreservation"},
 866         {Opt_noload, "noload"},
 867         {Opt_noload, "norecovery"},
 868         {Opt_nobh, "nobh"},
 869         {Opt_bh, "bh"},
 870         {Opt_commit, "commit=%u"},
 871         {Opt_journal_update, "journal=update"},
 872         {Opt_journal_inum, "journal=%u"},
 873         {Opt_journal_dev, "journal_dev=%u"},
 874         {Opt_journal_path, "journal_path=%s"},
 875         {Opt_abort, "abort"},
 876         {Opt_data_journal, "data=journal"},
 877         {Opt_data_ordered, "data=ordered"},
 878         {Opt_data_writeback, "data=writeback"},
 879         {Opt_data_err_abort, "data_err=abort"},
 880         {Opt_data_err_ignore, "data_err=ignore"},
 881         {Opt_offusrjquota, "usrjquota="},
 882         {Opt_usrjquota, "usrjquota=%s"},
 883         {Opt_offgrpjquota, "grpjquota="},
 884         {Opt_grpjquota, "grpjquota=%s"},
 885         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 886         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 887         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 888         {Opt_grpquota, "grpquota"},
 889         {Opt_noquota, "noquota"},
 890         {Opt_quota, "quota"},
 891         {Opt_usrquota, "usrquota"},
 892         {Opt_barrier, "barrier=%u"},
 893         {Opt_barrier, "barrier"},
 894         {Opt_nobarrier, "nobarrier"},
 895         {Opt_resize, "resize"},
 896         {Opt_err, NULL},
 897 };
 898
 899 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 900 {
 901         ext3_fsblk_t    sb_block;
 902         char            *options = (char *) *data;
 903
 904         if (!options || strncmp(options, "sb=", 3) != 0)
 905                 return 1;       /* Default location */
 906         options += 3;
 907         /*todo: use simple_strtoll with >32bit ext3 */
 908         sb_block = simple_strtoul(options, &options, 0);
 909         if (*options && *options != ',') {
 910                 ext3_msg(sb, KERN_ERR, "error: invalid sb specification: %s",
 911                        (char *) *data);
 912                 return 1;
 913         }
 914         if (*options == ',')
 915                 options++;
 916         *data = (void *) options;
 917         return sb_block;
 918 }
 919
 920 #ifdef CONFIG_QUOTA
 921 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 922 {
 923         struct ext3_sb_info *sbi = EXT3_SB(sb);
 924         char *qname;
 925
 926         if (sb_any_quota_loaded(sb) &&
 927                 !sbi->s_qf_names[qtype]) {
 928                 ext3_msg(sb, KERN_ERR,
 929                         "Cannot change journaled "
 930                         "quota options when quota turned on");
 931                 return 0;
 932         }
 933         qname = match_strdup(args);
 934         if (!qname) {
 935                 ext3_msg(sb, KERN_ERR,
 936                         "Not enough memory for storing quotafile name");
 937                 return 0;
 938         }
 939         if (sbi->s_qf_names[qtype]) {
 940                 int same = !strcmp(sbi->s_qf_names[qtype], qname);
 941
 942                 kfree(qname);
 943                 if (!same) {
 944                         ext3_msg(sb, KERN_ERR,
 945                                  "%s quota file already specified",
 946                                  QTYPE2NAME(qtype));
 947                 }
 948                 return same;
 949         }
 950         if (strchr(qname, '/')) {
 951                 ext3_msg(sb, KERN_ERR,
 952                         "quotafile must be on filesystem root");
 953                 kfree(qname);
 954                 return 0;
 955         }
 956         sbi->s_qf_names[qtype] = qname;
 957         set_opt(sbi->s_mount_opt, QUOTA);
 958         return 1;
 959 }
 960
 961 static int clear_qf_name(struct super_block *sb, int qtype) {
 962
 963         struct ext3_sb_info *sbi = EXT3_SB(sb);
 964
 965         if (sb_any_quota_loaded(sb) &&
 966                 sbi->s_qf_names[qtype]) {
 967                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 968                         " when quota turned on");
 969                 return 0;
 970         }
 971         if (sbi->s_qf_names[qtype]) {
 972                 kfree(sbi->s_qf_names[qtype]);
 973                 sbi->s_qf_names[qtype] = NULL;
 974         }
 975         return 1;
 976 }
 977 #endif
 978
 979 static int parse_options (char *options, struct super_block *sb,
 980                           unsigned int *inum, unsigned long *journal_devnum,
 981                           ext3_fsblk_t *n_blocks_count, int is_remount)
 982 {
 983         struct ext3_sb_info *sbi = EXT3_SB(sb);
 984         char * p;
 985         substring_t args[MAX_OPT_ARGS];
 986         int data_opt = 0;
 987         int option;
 988         kuid_t uid;
 989         kgid_t gid;
 990         char *journal_path;
 991         struct inode *journal_inode;
 992         struct path path;
 993         int error;
 994
 995 #ifdef CONFIG_QUOTA
 996         int qfmt;
 997 #endif
 998
 999         if (!options)
1000                 return 1;
1001
1002         while ((p = strsep (&options, ",")) != NULL) {
1003                 int token;
1004                 if (!*p)
1005                         continue;
1006                 /*
1007                  * Initialize args struct so we know whether arg was
1008                  * found; some options take optional arguments.
1009                  */
1010                 args[0].to = args[0].from = NULL;
1011                 token = match_token(p, tokens, args);
1012                 switch (token) {
1013                 case Opt_bsd_df:
1014                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1015                         break;
1016                 case Opt_minix_df:
1017                         set_opt (sbi->s_mount_opt, MINIX_DF);
1018                         break;
1019                 case Opt_grpid:
1020                         set_opt (sbi->s_mount_opt, GRPID);
1021                         break;
1022                 case Opt_nogrpid:
1023                         clear_opt (sbi->s_mount_opt, GRPID);
1024                         break;
1025                 case Opt_resuid:
1026                         if (match_int(&args[0], &option))
1027                                 return 0;
1028                         uid = make_kuid(current_user_ns(), option);
1029                         if (!uid_valid(uid)) {
1030                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1031                                 return 0;
1032
1033                         }
1034                         sbi->s_resuid = uid;
1035                         break;
1036                 case Opt_resgid:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         gid = make_kgid(current_user_ns(), option);
1040                         if (!gid_valid(gid)) {
1041                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1042                                 return 0;
1043                         }
1044                         sbi->s_resgid = gid;
1045                         break;
1046                 case Opt_sb:
1047                         /* handled by get_sb_block() instead of here */
1048                         /* *sb_block = match_int(&args[0]); */
1049                         break;
1050                 case Opt_err_panic:
1051                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1052                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1053                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1054                         break;
1055                 case Opt_err_ro:
1056                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1057                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1058                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1059                         break;
1060                 case Opt_err_cont:
1061                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1062                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1063                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1064                         break;
1065                 case Opt_nouid32:
1066                         set_opt (sbi->s_mount_opt, NO_UID32);
1067                         break;
1068                 case Opt_nocheck:
1069                         clear_opt (sbi->s_mount_opt, CHECK);
1070                         break;
1071                 case Opt_debug:
1072                         set_opt (sbi->s_mount_opt, DEBUG);
1073                         break;
1074                 case Opt_oldalloc:
1075                         ext3_msg(sb, KERN_WARNING,
1076                                 "Ignoring deprecated oldalloc option");
1077                         break;
1078                 case Opt_orlov:
1079                         ext3_msg(sb, KERN_WARNING,
1080                                 "Ignoring deprecated orlov option");
1081                         break;
1082 #ifdef CONFIG_EXT3_FS_XATTR
1083                 case Opt_user_xattr:
1084                         set_opt (sbi->s_mount_opt, XATTR_USER);
1085                         break;
1086                 case Opt_nouser_xattr:
1087                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1088                         break;
1089 #else
1090                 case Opt_user_xattr:
1091                 case Opt_nouser_xattr:
1092                         ext3_msg(sb, KERN_INFO,
1093                                 "(no)user_xattr options not supported");
1094                         break;
1095 #endif
1096 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1097                 case Opt_acl:
1098                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1099                         break;
1100                 case Opt_noacl:
1101                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1102                         break;
1103 #else
1104                 case Opt_acl:
1105                 case Opt_noacl:
1106                         ext3_msg(sb, KERN_INFO,
1107                                 "(no)acl options not supported");
1108                         break;
1109 #endif
1110                 case Opt_reservation:
1111                         set_opt(sbi->s_mount_opt, RESERVATION);
1112                         break;
1113                 case Opt_noreservation:
1114                         clear_opt(sbi->s_mount_opt, RESERVATION);
1115                         break;
1116                 case Opt_journal_update:
1117                         /* @@@ FIXME */
1118                         /* Eventually we will want to be able to create
1119                            a journal file here.  For now, only allow the
1120                            user to specify an existing inode to be the
1121                            journal file. */
1122                         if (is_remount) {
1123                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1124                                         "journal on remount");
1125                                 return 0;
1126                         }
1127                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1128                         break;
1129                 case Opt_journal_inum:
1130                         if (is_remount) {
1131                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1132                                        "journal on remount");
1133                                 return 0;
1134                         }
1135                         if (match_int(&args[0], &option))
1136                                 return 0;
1137                         *inum = option;
1138                         break;
1139                 case Opt_journal_dev:
1140                         if (is_remount) {
1141                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1142                                        "journal on remount");
1143                                 return 0;
1144                         }
1145                         if (match_int(&args[0], &option))
1146                                 return 0;
1147                         *journal_devnum = option;
1148                         break;
1149                 case Opt_journal_path:
1150                         if (is_remount) {
1151                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1152                                        "journal on remount");
1153                                 return 0;
1154                         }
1155
1156                         journal_path = match_strdup(&args[0]);
1157                         if (!journal_path) {
1158                                 ext3_msg(sb, KERN_ERR, "error: could not dup "
1159                                         "journal device string");
1160                                 return 0;
1161                         }
1162
1163                         error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1164                         if (error) {
1165                                 ext3_msg(sb, KERN_ERR, "error: could not find "
1166                                         "journal device path: error %d", error);
1167                                 kfree(journal_path);
1168                                 return 0;
1169                         }
1170
1171                         journal_inode = path.dentry->d_inode;
1172                         if (!S_ISBLK(journal_inode->i_mode)) {
1173                                 ext3_msg(sb, KERN_ERR, "error: journal path %s "
1174                                         "is not a block device", journal_path);
1175                                 path_put(&path);
1176                                 kfree(journal_path);
1177                                 return 0;
1178                         }
1179
1180                         *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1181                         path_put(&path);
1182                         kfree(journal_path);
1183                         break;
1184                 case Opt_noload:
1185                         set_opt (sbi->s_mount_opt, NOLOAD);
1186                         break;
1187                 case Opt_commit:
1188                         if (match_int(&args[0], &option))
1189                                 return 0;
1190                         if (option < 0)
1191                                 return 0;
1192                         if (option == 0)
1193                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1194                         sbi->s_commit_interval = HZ * option;
1195                         break;
1196                 case Opt_data_journal:
1197                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1198                         goto datacheck;
1199                 case Opt_data_ordered:
1200                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1201                         goto datacheck;
1202                 case Opt_data_writeback:
1203                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1204                 datacheck:
1205                         if (is_remount) {
1206                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1207                                         break;
1208                                 ext3_msg(sb, KERN_ERR,
1209                                         "error: cannot change "
1210                                         "data mode on remount. The filesystem "
1211                                         "is mounted in data=%s mode and you "
1212                                         "try to remount it in data=%s mode.",
1213                                         data_mode_string(test_opt(sb,
1214                                                         DATA_FLAGS)),
1215                                         data_mode_string(data_opt));
1216                                 return 0;
1217                         } else {
1218                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1219                                 sbi->s_mount_opt |= data_opt;
1220                         }
1221                         break;
1222                 case Opt_data_err_abort:
1223                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1224                         break;
1225                 case Opt_data_err_ignore:
1226                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1227                         break;
1228 #ifdef CONFIG_QUOTA
1229                 case Opt_usrjquota:
1230                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1231                                 return 0;
1232                         break;
1233                 case Opt_grpjquota:
1234                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1235                                 return 0;
1236                         break;
1237                 case Opt_offusrjquota:
1238                         if (!clear_qf_name(sb, USRQUOTA))
1239                                 return 0;
1240                         break;
1241                 case Opt_offgrpjquota:
1242                         if (!clear_qf_name(sb, GRPQUOTA))
1243                                 return 0;
1244                         break;
1245                 case Opt_jqfmt_vfsold:
1246                         qfmt = QFMT_VFS_OLD;
1247                         goto set_qf_format;
1248                 case Opt_jqfmt_vfsv0:
1249                         qfmt = QFMT_VFS_V0;
1250                         goto set_qf_format;
1251                 case Opt_jqfmt_vfsv1:
1252                         qfmt = QFMT_VFS_V1;
1253 set_qf_format:
1254                         if (sb_any_quota_loaded(sb) &&
1255                             sbi->s_jquota_fmt != qfmt) {
1256                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1257                                         "journaled quota options when "
1258                                         "quota turned on.");
1259                                 return 0;
1260                         }
1261                         sbi->s_jquota_fmt = qfmt;
1262                         break;
1263                 case Opt_quota:
1264                 case Opt_usrquota:
1265                         set_opt(sbi->s_mount_opt, QUOTA);
1266                         set_opt(sbi->s_mount_opt, USRQUOTA);
1267                         break;
1268                 case Opt_grpquota:
1269                         set_opt(sbi->s_mount_opt, QUOTA);
1270                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1271                         break;
1272                 case Opt_noquota:
1273                         if (sb_any_quota_loaded(sb)) {
1274                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1275                                         "quota options when quota turned on.");
1276                                 return 0;
1277                         }
1278                         clear_opt(sbi->s_mount_opt, QUOTA);
1279                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1280                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1281                         break;
1282 #else
1283                 case Opt_quota:
1284                 case Opt_usrquota:
1285                 case Opt_grpquota:
1286                         ext3_msg(sb, KERN_ERR,
1287                                 "error: quota options not supported.");
1288                         break;
1289                 case Opt_usrjquota:
1290                 case Opt_grpjquota:
1291                 case Opt_offusrjquota:
1292                 case Opt_offgrpjquota:
1293                 case Opt_jqfmt_vfsold:
1294                 case Opt_jqfmt_vfsv0:
1295                 case Opt_jqfmt_vfsv1:
1296                         ext3_msg(sb, KERN_ERR,
1297                                 "error: journaled quota options not "
1298                                 "supported.");
1299                         break;
1300                 case Opt_noquota:
1301                         break;
1302 #endif
1303                 case Opt_abort:
1304                         set_opt(sbi->s_mount_opt, ABORT);
1305                         break;
1306                 case Opt_nobarrier:
1307                         clear_opt(sbi->s_mount_opt, BARRIER);
1308                         break;
1309                 case Opt_barrier:
1310                         if (args[0].from) {
1311                                 if (match_int(&args[0], &option))
1312                                         return 0;
1313                         } else
1314                                 option = 1;     /* No argument, default to 1 */
1315                         if (option)
1316                                 set_opt(sbi->s_mount_opt, BARRIER);
1317                         else
1318                                 clear_opt(sbi->s_mount_opt, BARRIER);
1319                         break;
1320                 case Opt_ignore:
1321                         break;
1322                 case Opt_resize:
1323                         if (!is_remount) {
1324                                 ext3_msg(sb, KERN_ERR,
1325                                         "error: resize option only available "
1326                                         "for remount");
1327                                 return 0;
1328                         }
1329                         if (match_int(&args[0], &option) != 0)
1330                                 return 0;
1331                         *n_blocks_count = option;
1332                         break;
1333                 case Opt_nobh:
1334                         ext3_msg(sb, KERN_WARNING,
1335                                 "warning: ignoring deprecated nobh option");
1336                         break;
1337                 case Opt_bh:
1338                         ext3_msg(sb, KERN_WARNING,
1339                                 "warning: ignoring deprecated bh option");
1340                         break;
1341                 default:
1342                         ext3_msg(sb, KERN_ERR,
1343                                 "error: unrecognized mount option \"%s\" "
1344                                 "or missing value", p);
1345                         return 0;
1346                 }
1347         }
1348 #ifdef CONFIG_QUOTA
1349         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1350                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1351                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1352                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1353                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1354
1355                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1356                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1357                                         "format mixing.");
1358                         return 0;
1359                 }
1360
1361                 if (!sbi->s_jquota_fmt) {
1362                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1363                                         "not specified.");
1364                         return 0;
1365                 }
1366         }
1367 #endif
1368         return 1;
1369 }
1370
1371 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1372                             int read_only)
1373 {
1374         struct ext3_sb_info *sbi = EXT3_SB(sb);
1375         int res = 0;
1376
1377         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1378                 ext3_msg(sb, KERN_ERR,
1379                         "error: revision level too high, "
1380                         "forcing read-only mode");
1381                 res = MS_RDONLY;
1382         }
1383         if (read_only)
1384                 return res;
1385         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1386                 ext3_msg(sb, KERN_WARNING,
1387                         "warning: mounting unchecked fs, "
1388                         "running e2fsck is recommended");
1389         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1390                 ext3_msg(sb, KERN_WARNING,
1391                         "warning: mounting fs with errors, "
1392                         "running e2fsck is recommended");
1393         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1394                  le16_to_cpu(es->s_mnt_count) >=
1395                         le16_to_cpu(es->s_max_mnt_count))
1396                 ext3_msg(sb, KERN_WARNING,
1397                         "warning: maximal mount count reached, "
1398                         "running e2fsck is recommended");
1399         else if (le32_to_cpu(es->s_checkinterval) &&
1400                 (le32_to_cpu(es->s_lastcheck) +
1401                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1402                 ext3_msg(sb, KERN_WARNING,
1403                         "warning: checktime reached, "
1404                         "running e2fsck is recommended");
1405 #if 0
1406                 /* @@@ We _will_ want to clear the valid bit if we find
1407                    inconsistencies, to force a fsck at reboot.  But for
1408                    a plain journaled filesystem we can keep it set as
1409                    valid forever! :) */
1410         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1411 #endif
1412         if (!le16_to_cpu(es->s_max_mnt_count))
1413                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1414         le16_add_cpu(&es->s_mnt_count, 1);
1415         es->s_mtime = cpu_to_le32(get_seconds());
1416         ext3_update_dynamic_rev(sb);
1417         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1418
1419         ext3_commit_super(sb, es, 1);
1420         if (test_opt(sb, DEBUG))
1421                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1422                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1423                         sb->s_blocksize,
1424                         sbi->s_groups_count,
1425                         EXT3_BLOCKS_PER_GROUP(sb),
1426                         EXT3_INODES_PER_GROUP(sb),
1427                         sbi->s_mount_opt);
1428
1429         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1430                 char b[BDEVNAME_SIZE];
1431                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1432                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1433         } else {
1434                 ext3_msg(sb, KERN_INFO, "using internal journal");
1435         }
1436         cleancache_init_fs(sb);
1437         return res;
1438 }
1439
1440 /* Called at mount-time, super-block is locked */
1441 static int ext3_check_descriptors(struct super_block *sb)
1442 {
1443         struct ext3_sb_info *sbi = EXT3_SB(sb);
1444         int i;
1445
1446         ext3_debug ("Checking group descriptors");
1447
1448         for (i = 0; i < sbi->s_groups_count; i++) {
1449                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1450                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1451                 ext3_fsblk_t last_block;
1452
1453                 if (i == sbi->s_groups_count - 1)
1454                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1455                 else
1456                         last_block = first_block +
1457                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1458
1459                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1460                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1461                 {
1462                         ext3_error (sb, "ext3_check_descriptors",
1463                                     "Block bitmap for group %d"
1464                                     " not in group (block %lu)!",
1465                                     i, (unsigned long)
1466                                         le32_to_cpu(gdp->bg_block_bitmap));
1467                         return 0;
1468                 }
1469                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1470                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1471                 {
1472                         ext3_error (sb, "ext3_check_descriptors",
1473                                     "Inode bitmap for group %d"
1474                                     " not in group (block %lu)!",
1475                                     i, (unsigned long)
1476                                         le32_to_cpu(gdp->bg_inode_bitmap));
1477                         return 0;
1478                 }
1479                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1480                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1481                     last_block)
1482                 {
1483                         ext3_error (sb, "ext3_check_descriptors",
1484                                     "Inode table for group %d"
1485                                     " not in group (block %lu)!",
1486                                     i, (unsigned long)
1487                                         le32_to_cpu(gdp->bg_inode_table));
1488                         return 0;
1489                 }
1490         }
1491
1492         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1493         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1494         return 1;
1495 }
1496
1497
1498 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1499  * the superblock) which were deleted from all directories, but held open by
1500  * a process at the time of a crash.  We walk the list and try to delete these
1501  * inodes at recovery time (only with a read-write filesystem).
1502  *
1503  * In order to keep the orphan inode chain consistent during traversal (in
1504  * case of crash during recovery), we link each inode into the superblock
1505  * orphan list_head and handle it the same way as an inode deletion during
1506  * normal operation (which journals the operations for us).
1507  *
1508  * We only do an iget() and an iput() on each inode, which is very safe if we
1509  * accidentally point at an in-use or already deleted inode.  The worst that
1510  * can happen in this case is that we get a "bit already cleared" message from
1511  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1512  * e2fsck was run on this filesystem, and it must have already done the orphan
1513  * inode cleanup for us, so we can safely abort without any further action.
1514  */
1515 static void ext3_orphan_cleanup (struct super_block * sb,
1516                                  struct ext3_super_block * es)
1517 {
1518         unsigned int s_flags = sb->s_flags;
1519         int nr_orphans = 0, nr_truncates = 0;
1520 #ifdef CONFIG_QUOTA
1521         int i;
1522 #endif
1523         if (!es->s_last_orphan) {
1524                 jbd_debug(4, "no orphan inodes to clean up\n");
1525                 return;
1526         }
1527
1528         if (bdev_read_only(sb->s_bdev)) {
1529                 ext3_msg(sb, KERN_ERR, "error: write access "
1530                         "unavailable, skipping orphan cleanup.");
1531                 return;
1532         }
1533
1534         /* Check if feature set allows readwrite operations */
1535         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1536                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1537                          "unknown ROCOMPAT features");
1538                 return;
1539         }
1540
1541         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1542                 /* don't clear list on RO mount w/ errors */
1543                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
1544                         jbd_debug(1, "Errors on filesystem, "
1545                                   "clearing orphan list.\n");
1546                         es->s_last_orphan = 0;
1547                 }
1548                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1549                 return;
1550         }
1551
1552         if (s_flags & MS_RDONLY) {
1553                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1554                 sb->s_flags &= ~MS_RDONLY;
1555         }
1556 #ifdef CONFIG_QUOTA
1557         /* Needed for iput() to work correctly and not trash data */
1558         sb->s_flags |= MS_ACTIVE;
1559         /* Turn on quotas so that they are updated correctly */
1560         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
1561                 if (EXT3_SB(sb)->s_qf_names[i]) {
1562                         int ret = ext3_quota_on_mount(sb, i);
1563                         if (ret < 0)
1564                                 ext3_msg(sb, KERN_ERR,
1565                                         "error: cannot turn on journaled "
1566                                         "quota: %d", ret);
1567                 }
1568         }
1569 #endif
1570
1571         while (es->s_last_orphan) {
1572                 struct inode *inode;
1573
1574                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1575                 if (IS_ERR(inode)) {
1576                         es->s_last_orphan = 0;
1577                         break;
1578                 }
1579
1580                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1581                 dquot_initialize(inode);
1582                 if (inode->i_nlink) {
1583                         printk(KERN_DEBUG
1584                                 "%s: truncating inode %lu to %Ld bytes\n",
1585                                 __func__, inode->i_ino, inode->i_size);
1586                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1587                                   inode->i_ino, inode->i_size);
1588                         ext3_truncate(inode);
1589                         nr_truncates++;
1590                 } else {
1591                         printk(KERN_DEBUG
1592                                 "%s: deleting unreferenced inode %lu\n",
1593                                 __func__, inode->i_ino);
1594                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1595                                   inode->i_ino);
1596                         nr_orphans++;
1597                 }
1598                 iput(inode);  /* The delete magic happens here! */
1599         }
1600
1601 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1602
1603         if (nr_orphans)
1604                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1605                        PLURAL(nr_orphans));
1606         if (nr_truncates)
1607                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1608                        PLURAL(nr_truncates));
1609 #ifdef CONFIG_QUOTA
1610         /* Turn quotas off */
1611         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
1612                 if (sb_dqopt(sb)->files[i])
1613                         dquot_quota_off(sb, i);
1614         }
1615 #endif
1616         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1617 }
1618
1619 /*
1620  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1621  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1622  * We need to be 1 filesystem block less than the 2^32 sector limit.
1623  */
1624 static loff_t ext3_max_size(int bits)
1625 {
1626         loff_t res = EXT3_NDIR_BLOCKS;
1627         int meta_blocks;
1628         loff_t upper_limit;
1629
1630         /* This is calculated to be the largest file size for a
1631          * dense, file such that the total number of
1632          * sectors in the file, including data and all indirect blocks,
1633          * does not exceed 2^32 -1
1634          * __u32 i_blocks representing the total number of
1635          * 512 bytes blocks of the file
1636          */
1637         upper_limit = (1LL << 32) - 1;
1638
1639         /* total blocks in file system block size */
1640         upper_limit >>= (bits - 9);
1641
1642
1643         /* indirect blocks */
1644         meta_blocks = 1;
1645         /* double indirect blocks */
1646         meta_blocks += 1 + (1LL << (bits-2));
1647         /* tripple indirect blocks */
1648         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1649
1650         upper_limit -= meta_blocks;
1651         upper_limit <<= bits;
1652
1653         res += 1LL << (bits-2);
1654         res += 1LL << (2*(bits-2));
1655         res += 1LL << (3*(bits-2));
1656         res <<= bits;
1657         if (res > upper_limit)
1658                 res = upper_limit;
1659
1660         if (res > MAX_LFS_FILESIZE)
1661                 res = MAX_LFS_FILESIZE;
1662
1663         return res;
1664 }
1665
1666 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1667                                     ext3_fsblk_t logic_sb_block,
1668                                     int nr)
1669 {
1670         struct ext3_sb_info *sbi = EXT3_SB(sb);
1671         unsigned long bg, first_meta_bg;
1672         int has_super = 0;
1673
1674         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1675
1676         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1677             nr < first_meta_bg)
1678                 return (logic_sb_block + nr + 1);
1679         bg = sbi->s_desc_per_block * nr;
1680         if (ext3_bg_has_super(sb, bg))
1681                 has_super = 1;
1682         return (has_super + ext3_group_first_block_no(sb, bg));
1683 }
1684
1685
1686 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1687 {
1688         struct buffer_head * bh;
1689         struct ext3_super_block *es = NULL;
1690         struct ext3_sb_info *sbi;
1691         ext3_fsblk_t block;
1692         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1693         ext3_fsblk_t logic_sb_block;
1694         unsigned long offset = 0;
1695         unsigned int journal_inum = 0;
1696         unsigned long journal_devnum = 0;
1697         unsigned long def_mount_opts;
1698         struct inode *root;
1699         int blocksize;
1700         int hblock;
1701         int db_count;
1702         int i;
1703         int needs_recovery;
1704         int ret = -EINVAL;
1705         __le32 features;
1706         int err;
1707
1708         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1709         if (!sbi)
1710                 return -ENOMEM;
1711
1712         sbi->s_blockgroup_lock =
1713                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1714         if (!sbi->s_blockgroup_lock) {
1715                 kfree(sbi);
1716                 return -ENOMEM;
1717         }
1718         sb->s_fs_info = sbi;
1719         sbi->s_sb_block = sb_block;
1720
1721         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1722         if (!blocksize) {
1723                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1724                 goto out_fail;
1725         }
1726
1727         /*
1728          * The ext3 superblock will not be buffer aligned for other than 1kB
1729          * block sizes.  We need to calculate the offset from buffer start.
1730          */
1731         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1732                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1733                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1734         } else {
1735                 logic_sb_block = sb_block;
1736         }
1737
1738         if (!(bh = sb_bread(sb, logic_sb_block))) {
1739                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1740                 goto out_fail;
1741         }
1742         /*
1743          * Note: s_es must be initialized as soon as possible because
1744          *       some ext3 macro-instructions depend on its value
1745          */
1746         es = (struct ext3_super_block *) (bh->b_data + offset);
1747         sbi->s_es = es;
1748         sb->s_magic = le16_to_cpu(es->s_magic);
1749         if (sb->s_magic != EXT3_SUPER_MAGIC)
1750                 goto cantfind_ext3;
1751
1752         /* Set defaults before we parse the mount options */
1753         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1754         if (def_mount_opts & EXT3_DEFM_DEBUG)
1755                 set_opt(sbi->s_mount_opt, DEBUG);
1756         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1757                 set_opt(sbi->s_mount_opt, GRPID);
1758         if (def_mount_opts & EXT3_DEFM_UID16)
1759                 set_opt(sbi->s_mount_opt, NO_UID32);
1760 #ifdef CONFIG_EXT3_FS_XATTR
1761         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1762                 set_opt(sbi->s_mount_opt, XATTR_USER);
1763 #endif
1764 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1765         if (def_mount_opts & EXT3_DEFM_ACL)
1766                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1767 #endif
1768         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1769                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1770         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1771                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1772         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1773                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1774
1775         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1776                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1777         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1778                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1779         else
1780                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1781
1782         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1783         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1784
1785         /* enable barriers by default */
1786         set_opt(sbi->s_mount_opt, BARRIER);
1787         set_opt(sbi->s_mount_opt, RESERVATION);
1788
1789         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1790                             NULL, 0))
1791                 goto failed_mount;
1792
1793         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1794                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1795
1796         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1797             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1798              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1799              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1800                 ext3_msg(sb, KERN_WARNING,
1801                         "warning: feature flags set on rev 0 fs, "
1802                         "running e2fsck is recommended");
1803         /*
1804          * Check feature flags regardless of the revision level, since we
1805          * previously didn't change the revision level when setting the flags,
1806          * so there is a chance incompat flags are set on a rev 0 filesystem.
1807          */
1808         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1809         if (features) {
1810                 ext3_msg(sb, KERN_ERR,
1811                         "error: couldn't mount because of unsupported "
1812                         "optional features (%x)", le32_to_cpu(features));
1813                 goto failed_mount;
1814         }
1815         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1816         if (!(sb->s_flags & MS_RDONLY) && features) {
1817                 ext3_msg(sb, KERN_ERR,
1818                         "error: couldn't mount RDWR because of unsupported "
1819                         "optional features (%x)", le32_to_cpu(features));
1820                 goto failed_mount;
1821         }
1822         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1823
1824         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1825             blocksize > EXT3_MAX_BLOCK_SIZE) {
1826                 ext3_msg(sb, KERN_ERR,
1827                         "error: couldn't mount because of unsupported "
1828                         "filesystem blocksize %d", blocksize);
1829                 goto failed_mount;
1830         }
1831
1832         hblock = bdev_logical_block_size(sb->s_bdev);
1833         if (sb->s_blocksize != blocksize) {
1834                 /*
1835                  * Make sure the blocksize for the filesystem is larger
1836                  * than the hardware sectorsize for the machine.
1837                  */
1838                 if (blocksize < hblock) {
1839                         ext3_msg(sb, KERN_ERR,
1840                                 "error: fsblocksize %d too small for "
1841                                 "hardware sectorsize %d", blocksize, hblock);
1842                         goto failed_mount;
1843                 }
1844
1845                 brelse (bh);
1846                 if (!sb_set_blocksize(sb, blocksize)) {
1847                         ext3_msg(sb, KERN_ERR,
1848                                 "error: bad blocksize %d", blocksize);
1849                         goto out_fail;
1850                 }
1851                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1852                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1853                 bh = sb_bread(sb, logic_sb_block);
1854                 if (!bh) {
1855                         ext3_msg(sb, KERN_ERR,
1856                                "error: can't read superblock on 2nd try");
1857                         goto failed_mount;
1858                 }
1859                 es = (struct ext3_super_block *)(bh->b_data + offset);
1860                 sbi->s_es = es;
1861                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1862                         ext3_msg(sb, KERN_ERR,
1863                                 "error: magic mismatch");
1864                         goto failed_mount;
1865                 }
1866         }
1867
1868         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1869
1870         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1871                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1872                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1873         } else {
1874                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1875                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1876                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1877                     (!is_power_of_2(sbi->s_inode_size)) ||
1878                     (sbi->s_inode_size > blocksize)) {
1879                         ext3_msg(sb, KERN_ERR,
1880                                 "error: unsupported inode size: %d",
1881                                 sbi->s_inode_size);
1882                         goto failed_mount;
1883                 }
1884         }
1885         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1886                                    le32_to_cpu(es->s_log_frag_size);
1887         if (blocksize != sbi->s_frag_size) {
1888                 ext3_msg(sb, KERN_ERR,
1889                        "error: fragsize %lu != blocksize %u (unsupported)",
1890                        sbi->s_frag_size, blocksize);
1891                 goto failed_mount;
1892         }
1893         sbi->s_frags_per_block = 1;
1894         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1895         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1896         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1897         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1898                 goto cantfind_ext3;
1899         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1900         if (sbi->s_inodes_per_block == 0)
1901                 goto cantfind_ext3;
1902         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1903                                         sbi->s_inodes_per_block;
1904         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1905         sbi->s_sbh = bh;
1906         sbi->s_mount_state = le16_to_cpu(es->s_state);
1907         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1908         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1909         for (i=0; i < 4; i++)
1910                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1911         sbi->s_def_hash_version = es->s_def_hash_version;
1912         i = le32_to_cpu(es->s_flags);
1913         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1914                 sbi->s_hash_unsigned = 3;
1915         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1916 #ifdef __CHAR_UNSIGNED__
1917                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1918                 sbi->s_hash_unsigned = 3;
1919 #else
1920                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1921 #endif
1922         }
1923
1924         if (sbi->s_blocks_per_group > blocksize * 8) {
1925                 ext3_msg(sb, KERN_ERR,
1926                         "#blocks per group too big: %lu",
1927                         sbi->s_blocks_per_group);
1928                 goto failed_mount;
1929         }
1930         if (sbi->s_frags_per_group > blocksize * 8) {
1931                 ext3_msg(sb, KERN_ERR,
1932                         "error: #fragments per group too big: %lu",
1933                         sbi->s_frags_per_group);
1934                 goto failed_mount;
1935         }
1936         if (sbi->s_inodes_per_group > blocksize * 8) {
1937                 ext3_msg(sb, KERN_ERR,
1938                         "error: #inodes per group too big: %lu",
1939                         sbi->s_inodes_per_group);
1940                 goto failed_mount;
1941         }
1942
1943         err = generic_check_addressable(sb->s_blocksize_bits,
1944                                         le32_to_cpu(es->s_blocks_count));
1945         if (err) {
1946                 ext3_msg(sb, KERN_ERR,
1947                         "error: filesystem is too large to mount safely");
1948                 if (sizeof(sector_t) < 8)
1949                         ext3_msg(sb, KERN_ERR,
1950                                 "error: CONFIG_LBDAF not enabled");
1951                 ret = err;
1952                 goto failed_mount;
1953         }
1954
1955         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1956                 goto cantfind_ext3;
1957         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1958                                le32_to_cpu(es->s_first_data_block) - 1)
1959                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1960         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1961         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1962                                     GFP_KERNEL);
1963         if (sbi->s_group_desc == NULL) {
1964                 ext3_msg(sb, KERN_ERR,
1965                         "error: not enough memory");
1966                 ret = -ENOMEM;
1967                 goto failed_mount;
1968         }
1969
1970         bgl_lock_init(sbi->s_blockgroup_lock);
1971
1972         for (i = 0; i < db_count; i++) {
1973                 block = descriptor_loc(sb, logic_sb_block, i);
1974                 sbi->s_group_desc[i] = sb_bread(sb, block);
1975                 if (!sbi->s_group_desc[i]) {
1976                         ext3_msg(sb, KERN_ERR,
1977                                 "error: can't read group descriptor %d", i);
1978                         db_count = i;
1979                         goto failed_mount2;
1980                 }
1981         }
1982         if (!ext3_check_descriptors (sb)) {
1983                 ext3_msg(sb, KERN_ERR,
1984                         "error: group descriptors corrupted");
1985                 goto failed_mount2;
1986         }
1987         sbi->s_gdb_count = db_count;
1988         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1989         spin_lock_init(&sbi->s_next_gen_lock);
1990
1991         /* per fileystem reservation list head & lock */
1992         spin_lock_init(&sbi->s_rsv_window_lock);
1993         sbi->s_rsv_window_root = RB_ROOT;
1994         /* Add a single, static dummy reservation to the start of the
1995          * reservation window list --- it gives us a placeholder for
1996          * append-at-start-of-list which makes the allocation logic
1997          * _much_ simpler. */
1998         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1999         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
2000         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2001         sbi->s_rsv_window_head.rsv_goal_size = 0;
2002         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
2003
2004         /*
2005          * set up enough so that it can read an inode
2006          */
2007         sb->s_op = &ext3_sops;
2008         sb->s_export_op = &ext3_export_ops;
2009         sb->s_xattr = ext3_xattr_handlers;
2010 #ifdef CONFIG_QUOTA
2011         sb->s_qcop = &ext3_qctl_operations;
2012         sb->dq_op = &ext3_quota_operations;
2013         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2014 #endif
2015         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
2016         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2017         mutex_init(&sbi->s_orphan_lock);
2018         mutex_init(&sbi->s_resize_lock);
2019
2020         sb->s_root = NULL;
2021
2022         needs_recovery = (es->s_last_orphan != 0 ||
2023                           EXT3_HAS_INCOMPAT_FEATURE(sb,
2024                                     EXT3_FEATURE_INCOMPAT_RECOVER));
2025
2026         /*
2027          * The first inode we look at is the journal inode.  Don't try
2028          * root first: it may be modified in the journal!
2029          */
2030         if (!test_opt(sb, NOLOAD) &&
2031             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
2032                 if (ext3_load_journal(sb, es, journal_devnum))
2033                         goto failed_mount2;
2034         } else if (journal_inum) {
2035                 if (ext3_create_journal(sb, es, journal_inum))
2036                         goto failed_mount2;
2037         } else {
2038                 if (!silent)
2039                         ext3_msg(sb, KERN_ERR,
2040                                 "error: no journal found. "
2041                                 "mounting ext3 over ext2?");
2042                 goto failed_mount2;
2043         }
2044         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2045                         ext3_count_free_blocks(sb), GFP_KERNEL);
2046         if (!err) {
2047                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2048                                 ext3_count_free_inodes(sb), GFP_KERNEL);
2049         }
2050         if (!err) {
2051                 err = percpu_counter_init(&sbi->s_dirs_counter,
2052                                 ext3_count_dirs(sb), GFP_KERNEL);
2053         }
2054         if (err) {
2055                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2056                 ret = err;
2057                 goto failed_mount3;
2058         }
2059
2060         /* We have now updated the journal if required, so we can
2061          * validate the data journaling mode. */
2062         switch (test_opt(sb, DATA_FLAGS)) {
2063         case 0:
2064                 /* No mode set, assume a default based on the journal
2065                    capabilities: ORDERED_DATA if the journal can
2066                    cope, else JOURNAL_DATA */
2067                 if (journal_check_available_features
2068                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2069                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2070                 else
2071                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2072                 break;
2073
2074         case EXT3_MOUNT_ORDERED_DATA:
2075         case EXT3_MOUNT_WRITEBACK_DATA:
2076                 if (!journal_check_available_features
2077                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2078                         ext3_msg(sb, KERN_ERR,
2079                                 "error: journal does not support "
2080                                 "requested data journaling mode");
2081                         goto failed_mount3;
2082                 }
2083         default:
2084                 break;
2085         }
2086
2087         /*
2088          * The journal_load will have done any necessary log recovery,
2089          * so we can safely mount the rest of the filesystem now.
2090          */
2091
2092         root = ext3_iget(sb, EXT3_ROOT_INO);
2093         if (IS_ERR(root)) {
2094                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2095                 ret = PTR_ERR(root);
2096                 goto failed_mount3;
2097         }
2098         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2099                 iput(root);
2100                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2101                 goto failed_mount3;
2102         }
2103         sb->s_root = d_make_root(root);
2104         if (!sb->s_root) {
2105                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2106                 ret = -ENOMEM;
2107                 goto failed_mount3;
2108         }
2109
2110         if (ext3_setup_super(sb, es, sb->s_flags & MS_RDONLY))
2111                 sb->s_flags |= MS_RDONLY;
2112
2113         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2114         ext3_orphan_cleanup(sb, es);
2115         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2116         if (needs_recovery) {
2117                 ext3_mark_recovery_complete(sb, es);
2118                 ext3_msg(sb, KERN_INFO, "recovery complete");
2119         }
2120         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2121                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2122                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2123                 "writeback");
2124
2125         return 0;
2126
2127 cantfind_ext3:
2128         if (!silent)
2129                 ext3_msg(sb, KERN_INFO,
2130                         "error: can't find ext3 filesystem on dev %s.",
2131                        sb->s_id);
2132         goto failed_mount;
2133
2134 failed_mount3:
2135         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2136         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2137         percpu_counter_destroy(&sbi->s_dirs_counter);
2138         journal_destroy(sbi->s_journal);
2139 failed_mount2:
2140         for (i = 0; i < db_count; i++)
2141                 brelse(sbi->s_group_desc[i]);
2142         kfree(sbi->s_group_desc);
2143 failed_mount:
2144 #ifdef CONFIG_QUOTA
2145         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2146                 kfree(sbi->s_qf_names[i]);
2147 #endif
2148         ext3_blkdev_remove(sbi);
2149         brelse(bh);
2150 out_fail:
2151         sb->s_fs_info = NULL;
2152         kfree(sbi->s_blockgroup_lock);
2153         kfree(sbi);
2154         return ret;
2155 }
2156
2157 /*
2158  * Setup any per-fs journal parameters now.  We'll do this both on
2159  * initial mount, once the journal has been initialised but before we've
2160  * done any recovery; and again on any subsequent remount.
2161  */
2162 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2163 {
2164         struct ext3_sb_info *sbi = EXT3_SB(sb);
2165
2166         if (sbi->s_commit_interval)
2167                 journal->j_commit_interval = sbi->s_commit_interval;
2168         /* We could also set up an ext3-specific default for the commit
2169          * interval here, but for now we'll just fall back to the jbd
2170          * default. */
2171
2172         spin_lock(&journal->j_state_lock);
2173         if (test_opt(sb, BARRIER))
2174                 journal->j_flags |= JFS_BARRIER;
2175         else
2176                 journal->j_flags &= ~JFS_BARRIER;
2177         if (test_opt(sb, DATA_ERR_ABORT))
2178                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2179         else
2180                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2181         spin_unlock(&journal->j_state_lock);
2182 }
2183
2184 static journal_t *ext3_get_journal(struct super_block *sb,
2185                                    unsigned int journal_inum)
2186 {
2187         struct inode *journal_inode;
2188         journal_t *journal;
2189
2190         /* First, test for the existence of a valid inode on disk.  Bad
2191          * things happen if we iget() an unused inode, as the subsequent
2192          * iput() will try to delete it. */
2193
2194         journal_inode = ext3_iget(sb, journal_inum);
2195         if (IS_ERR(journal_inode)) {
2196                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2197                 return NULL;
2198         }
2199         if (!journal_inode->i_nlink) {
2200                 make_bad_inode(journal_inode);
2201                 iput(journal_inode);
2202                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2203                 return NULL;
2204         }
2205
2206         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2207                   journal_inode, journal_inode->i_size);
2208         if (!S_ISREG(journal_inode->i_mode)) {
2209                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2210                 iput(journal_inode);
2211                 return NULL;
2212         }
2213
2214         journal = journal_init_inode(journal_inode);
2215         if (!journal) {
2216                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2217                 iput(journal_inode);
2218                 return NULL;
2219         }
2220         journal->j_private = sb;
2221         ext3_init_journal_params(sb, journal);
2222         return journal;
2223 }
2224
2225 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2226                                        dev_t j_dev)
2227 {
2228         struct buffer_head * bh;
2229         journal_t *journal;
2230         ext3_fsblk_t start;
2231         ext3_fsblk_t len;
2232         int hblock, blocksize;
2233         ext3_fsblk_t sb_block;
2234         unsigned long offset;
2235         struct ext3_super_block * es;
2236         struct block_device *bdev;
2237
2238         bdev = ext3_blkdev_get(j_dev, sb);
2239         if (bdev == NULL)
2240                 return NULL;
2241
2242         blocksize = sb->s_blocksize;
2243         hblock = bdev_logical_block_size(bdev);
2244         if (blocksize < hblock) {
2245                 ext3_msg(sb, KERN_ERR,
2246                         "error: blocksize too small for journal device");
2247                 goto out_bdev;
2248         }
2249
2250         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2251         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2252         set_blocksize(bdev, blocksize);
2253         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2254                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2255                         "external journal");
2256                 goto out_bdev;
2257         }
2258
2259         es = (struct ext3_super_block *) (bh->b_data + offset);
2260         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2261             !(le32_to_cpu(es->s_feature_incompat) &
2262               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2263                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2264                         "bad superblock");
2265                 brelse(bh);
2266                 goto out_bdev;
2267         }
2268
2269         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2270                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2271                 brelse(bh);
2272                 goto out_bdev;
2273         }
2274
2275         len = le32_to_cpu(es->s_blocks_count);
2276         start = sb_block + 1;
2277         brelse(bh);     /* we're done with the superblock */
2278
2279         journal = journal_init_dev(bdev, sb->s_bdev,
2280                                         start, len, blocksize);
2281         if (!journal) {
2282                 ext3_msg(sb, KERN_ERR,
2283                         "error: failed to create device journal");
2284                 goto out_bdev;
2285         }
2286         journal->j_private = sb;
2287         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2288                 if (bh_submit_read(journal->j_sb_buffer)) {
2289                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2290                         goto out_journal;
2291                 }
2292         }
2293         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2294                 ext3_msg(sb, KERN_ERR,
2295                         "error: external journal has more than one "
2296                         "user (unsupported) - %d",
2297                         be32_to_cpu(journal->j_superblock->s_nr_users));
2298                 goto out_journal;
2299         }
2300         EXT3_SB(sb)->journal_bdev = bdev;
2301         ext3_init_journal_params(sb, journal);
2302         return journal;
2303 out_journal:
2304         journal_destroy(journal);
2305 out_bdev:
2306         ext3_blkdev_put(bdev);
2307         return NULL;
2308 }
2309
2310 static int ext3_load_journal(struct super_block *sb,
2311                              struct ext3_super_block *es,
2312                              unsigned long journal_devnum)
2313 {
2314         journal_t *journal;
2315         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2316         dev_t journal_dev;
2317         int err = 0;
2318         int really_read_only;
2319
2320         if (journal_devnum &&
2321             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2322                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2323                         "numbers have changed");
2324                 journal_dev = new_decode_dev(journal_devnum);
2325         } else
2326                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2327
2328         really_read_only = bdev_read_only(sb->s_bdev);
2329
2330         /*
2331          * Are we loading a blank journal or performing recovery after a
2332          * crash?  For recovery, we need to check in advance whether we
2333          * can get read-write access to the device.
2334          */
2335
2336         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2337                 if (sb->s_flags & MS_RDONLY) {
2338                         ext3_msg(sb, KERN_INFO,
2339                                 "recovery required on readonly filesystem");
2340                         if (really_read_only) {
2341                                 ext3_msg(sb, KERN_ERR, "error: write access "
2342                                         "unavailable, cannot proceed");
2343                                 return -EROFS;
2344                         }
2345                         ext3_msg(sb, KERN_INFO,
2346                                 "write access will be enabled during recovery");
2347                 }
2348         }
2349
2350         if (journal_inum && journal_dev) {
2351                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2352                        "and inode journals");
2353                 return -EINVAL;
2354         }
2355
2356         if (journal_inum) {
2357                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2358                         return -EINVAL;
2359         } else {
2360                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2361                         return -EINVAL;
2362         }
2363
2364         if (!(journal->j_flags & JFS_BARRIER))
2365                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2366
2367         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2368                 err = journal_update_format(journal);
2369                 if (err)  {
2370                         ext3_msg(sb, KERN_ERR, "error updating journal");
2371                         journal_destroy(journal);
2372                         return err;
2373                 }
2374         }
2375
2376         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2377                 err = journal_wipe(journal, !really_read_only);
2378         if (!err)
2379                 err = journal_load(journal);
2380
2381         if (err) {
2382                 ext3_msg(sb, KERN_ERR, "error loading journal");
2383                 journal_destroy(journal);
2384                 return err;
2385         }
2386
2387         EXT3_SB(sb)->s_journal = journal;
2388         ext3_clear_journal_err(sb, es);
2389
2390         if (!really_read_only && journal_devnum &&
2391             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2392                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2393
2394                 /* Make sure we flush the recovery flag to disk. */
2395                 ext3_commit_super(sb, es, 1);
2396         }
2397
2398         return 0;
2399 }
2400
2401 static int ext3_create_journal(struct super_block *sb,
2402                                struct ext3_super_block *es,
2403                                unsigned int journal_inum)
2404 {
2405         journal_t *journal;
2406         int err;
2407
2408         if (sb->s_flags & MS_RDONLY) {
2409                 ext3_msg(sb, KERN_ERR,
2410                         "error: readonly filesystem when trying to "
2411                         "create journal");
2412                 return -EROFS;
2413         }
2414
2415         journal = ext3_get_journal(sb, journal_inum);
2416         if (!journal)
2417                 return -EINVAL;
2418
2419         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2420                journal_inum);
2421
2422         err = journal_create(journal);
2423         if (err) {
2424                 ext3_msg(sb, KERN_ERR, "error creating journal");
2425                 journal_destroy(journal);
2426                 return -EIO;
2427         }
2428
2429         EXT3_SB(sb)->s_journal = journal;
2430
2431         ext3_update_dynamic_rev(sb);
2432         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2433         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2434
2435         es->s_journal_inum = cpu_to_le32(journal_inum);
2436
2437         /* Make sure we flush the recovery flag to disk. */
2438         ext3_commit_super(sb, es, 1);
2439
2440         return 0;
2441 }
2442
2443 static int ext3_commit_super(struct super_block *sb,
2444                                struct ext3_super_block *es,
2445                                int sync)
2446 {
2447         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2448         int error = 0;
2449
2450         if (!sbh)
2451                 return error;
2452
2453         if (buffer_write_io_error(sbh)) {
2454                 /*
2455                  * Oh, dear.  A previous attempt to write the
2456                  * superblock failed.  This could happen because the
2457                  * USB device was yanked out.  Or it could happen to
2458                  * be a transient write error and maybe the block will
2459                  * be remapped.  Nothing we can do but to retry the
2460                  * write and hope for the best.
2461                  */
2462                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2463                        "superblock detected");
2464                 clear_buffer_write_io_error(sbh);
2465                 set_buffer_uptodate(sbh);
2466         }
2467         /*
2468          * If the file system is mounted read-only, don't update the
2469          * superblock write time.  This avoids updating the superblock
2470          * write time when we are mounting the root file system
2471          * read/only but we need to replay the journal; at that point,
2472          * for people who are east of GMT and who make their clock
2473          * tick in localtime for Windows bug-for-bug compatibility,
2474          * the clock is set in the future, and this will cause e2fsck
2475          * to complain and force a full file system check.
2476          */
2477         if (!(sb->s_flags & MS_RDONLY))
2478                 es->s_wtime = cpu_to_le32(get_seconds());
2479         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2480         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2481         BUFFER_TRACE(sbh, "marking dirty");
2482         mark_buffer_dirty(sbh);
2483         if (sync) {
2484                 error = sync_dirty_buffer(sbh);
2485                 if (buffer_write_io_error(sbh)) {
2486                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2487                                "superblock");
2488                         clear_buffer_write_io_error(sbh);
2489                         set_buffer_uptodate(sbh);
2490                 }
2491         }
2492         return error;
2493 }
2494
2495
2496 /*
2497  * Have we just finished recovery?  If so, and if we are mounting (or
2498  * remounting) the filesystem readonly, then we will end up with a
2499  * consistent fs on disk.  Record that fact.
2500  */
2501 static void ext3_mark_recovery_complete(struct super_block * sb,
2502                                         struct ext3_super_block * es)
2503 {
2504         journal_t *journal = EXT3_SB(sb)->s_journal;
2505
2506         journal_lock_updates(journal);
2507         if (journal_flush(journal) < 0)
2508                 goto out;
2509
2510         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2511             sb->s_flags & MS_RDONLY) {
2512                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2513                 ext3_commit_super(sb, es, 1);
2514         }
2515
2516 out:
2517         journal_unlock_updates(journal);
2518 }
2519
2520 /*
2521  * If we are mounting (or read-write remounting) a filesystem whose journal
2522  * has recorded an error from a previous lifetime, move that error to the
2523  * main filesystem now.
2524  */
2525 static void ext3_clear_journal_err(struct super_block *sb,
2526                                    struct ext3_super_block *es)
2527 {
2528         journal_t *journal;
2529         int j_errno;
2530         const char *errstr;
2531
2532         journal = EXT3_SB(sb)->s_journal;
2533
2534         /*
2535          * Now check for any error status which may have been recorded in the
2536          * journal by a prior ext3_error() or ext3_abort()
2537          */
2538
2539         j_errno = journal_errno(journal);
2540         if (j_errno) {
2541                 char nbuf[16];
2542
2543                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2544                 ext3_warning(sb, __func__, "Filesystem error recorded "
2545                              "from previous mount: %s", errstr);
2546                 ext3_warning(sb, __func__, "Marking fs in need of "
2547                              "filesystem check.");
2548
2549                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2550                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2551                 ext3_commit_super (sb, es, 1);
2552
2553                 journal_clear_err(journal);
2554         }
2555 }
2556
2557 /*
2558  * Force the running and committing transactions to commit,
2559  * and wait on the commit.
2560  */
2561 int ext3_force_commit(struct super_block *sb)
2562 {
2563         journal_t *journal;
2564         int ret;
2565
2566         if (sb->s_flags & MS_RDONLY)
2567                 return 0;
2568
2569         journal = EXT3_SB(sb)->s_journal;
2570         ret = ext3_journal_force_commit(journal);
2571         return ret;
2572 }
2573
2574 static int ext3_sync_fs(struct super_block *sb, int wait)
2575 {
2576         tid_t target;
2577
2578         trace_ext3_sync_fs(sb, wait);
2579         /*
2580          * Writeback quota in non-journalled quota case - journalled quota has
2581          * no dirty dquots
2582          */
2583         dquot_writeback_dquots(sb, -1);
2584         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2585                 if (wait)
2586                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2587         }
2588         return 0;
2589 }
2590
2591 /*
2592  * LVM calls this function before a (read-only) snapshot is created.  This
2593  * gives us a chance to flush the journal completely and mark the fs clean.
2594  */
2595 static int ext3_freeze(struct super_block *sb)
2596 {
2597         int error = 0;
2598         journal_t *journal;
2599
2600         if (!(sb->s_flags & MS_RDONLY)) {
2601                 journal = EXT3_SB(sb)->s_journal;
2602
2603                 /* Now we set up the journal barrier. */
2604                 journal_lock_updates(journal);
2605
2606                 /*
2607                  * We don't want to clear needs_recovery flag when we failed
2608                  * to flush the journal.
2609                  */
2610                 error = journal_flush(journal);
2611                 if (error < 0)
2612                         goto out;
2613
2614                 /* Journal blocked and flushed, clear needs_recovery flag. */
2615                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2616                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2617                 if (error)
2618                         goto out;
2619         }
2620         return 0;
2621
2622 out:
2623         journal_unlock_updates(journal);
2624         return error;
2625 }
2626
2627 /*
2628  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2629  * flag here, even though the filesystem is not technically dirty yet.
2630  */
2631 static int ext3_unfreeze(struct super_block *sb)
2632 {
2633         if (!(sb->s_flags & MS_RDONLY)) {
2634                 /* Reser the needs_recovery flag before the fs is unlocked. */
2635                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2636                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2637                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2638         }
2639         return 0;
2640 }
2641
2642 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2643 {
2644         struct ext3_super_block * es;
2645         struct ext3_sb_info *sbi = EXT3_SB(sb);
2646         ext3_fsblk_t n_blocks_count = 0;
2647         unsigned long old_sb_flags;
2648         struct ext3_mount_options old_opts;
2649         int enable_quota = 0;
2650         int err;
2651 #ifdef CONFIG_QUOTA
2652         int i;
2653 #endif
2654
2655         sync_filesystem(sb);
2656
2657         /* Store the original options */
2658         old_sb_flags = sb->s_flags;
2659         old_opts.s_mount_opt = sbi->s_mount_opt;
2660         old_opts.s_resuid = sbi->s_resuid;
2661         old_opts.s_resgid = sbi->s_resgid;
2662         old_opts.s_commit_interval = sbi->s_commit_interval;
2663 #ifdef CONFIG_QUOTA
2664         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2665         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2666                 if (sbi->s_qf_names[i]) {
2667                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
2668                                                          GFP_KERNEL);
2669                         if (!old_opts.s_qf_names[i]) {
2670                                 int j;
2671
2672                                 for (j = 0; j < i; j++)
2673                                         kfree(old_opts.s_qf_names[j]);
2674                                 return -ENOMEM;
2675                         }
2676                 } else
2677                         old_opts.s_qf_names[i] = NULL;
2678 #endif
2679
2680         /*
2681          * Allow the "check" option to be passed as a remount option.
2682          */
2683         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2684                 err = -EINVAL;
2685                 goto restore_opts;
2686         }
2687
2688         if (test_opt(sb, ABORT))
2689                 ext3_abort(sb, __func__, "Abort forced by user");
2690
2691         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2692                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2693
2694         es = sbi->s_es;
2695
2696         ext3_init_journal_params(sb, sbi->s_journal);
2697
2698         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2699                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2700                 if (test_opt(sb, ABORT)) {
2701                         err = -EROFS;
2702                         goto restore_opts;
2703                 }
2704
2705                 if (*flags & MS_RDONLY) {
2706                         err = dquot_suspend(sb, -1);
2707                         if (err < 0)
2708                                 goto restore_opts;
2709
2710                         /*
2711                          * First of all, the unconditional stuff we have to do
2712                          * to disable replay of the journal when we next remount
2713                          */
2714                         sb->s_flags |= MS_RDONLY;
2715
2716                         /*
2717                          * OK, test if we are remounting a valid rw partition
2718                          * readonly, and if so set the rdonly flag and then
2719                          * mark the partition as valid again.
2720                          */
2721                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2722                             (sbi->s_mount_state & EXT3_VALID_FS))
2723                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2724
2725                         ext3_mark_recovery_complete(sb, es);
2726                 } else {
2727                         __le32 ret;
2728                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2729                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2730                                 ext3_msg(sb, KERN_WARNING,
2731                                         "warning: couldn't remount RDWR "
2732                                         "because of unsupported optional "
2733                                         "features (%x)", le32_to_cpu(ret));
2734                                 err = -EROFS;
2735                                 goto restore_opts;
2736                         }
2737
2738                         /*
2739                          * If we have an unprocessed orphan list hanging
2740                          * around from a previously readonly bdev mount,
2741                          * require a full umount & mount for now.
2742                          */
2743                         if (es->s_last_orphan) {
2744                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2745                                        "remount RDWR because of unprocessed "
2746                                        "orphan inode list.  Please "
2747                                        "umount & mount instead.");
2748                                 err = -EINVAL;
2749                                 goto restore_opts;
2750                         }
2751
2752                         /*
2753                          * Mounting a RDONLY partition read-write, so reread
2754                          * and store the current valid flag.  (It may have
2755                          * been changed by e2fsck since we originally mounted
2756                          * the partition.)
2757                          */
2758                         ext3_clear_journal_err(sb, es);
2759                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2760                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2761                                 goto restore_opts;
2762                         if (!ext3_setup_super (sb, es, 0))
2763                                 sb->s_flags &= ~MS_RDONLY;
2764                         enable_quota = 1;
2765                 }
2766         }
2767 #ifdef CONFIG_QUOTA
2768         /* Release old quota file names */
2769         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2770                 kfree(old_opts.s_qf_names[i]);
2771 #endif
2772         if (enable_quota)
2773                 dquot_resume(sb, -1);
2774         return 0;
2775 restore_opts:
2776         sb->s_flags = old_sb_flags;
2777         sbi->s_mount_opt = old_opts.s_mount_opt;
2778         sbi->s_resuid = old_opts.s_resuid;
2779         sbi->s_resgid = old_opts.s_resgid;
2780         sbi->s_commit_interval = old_opts.s_commit_interval;
2781 #ifdef CONFIG_QUOTA
2782         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2783         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
2784                 kfree(sbi->s_qf_names[i]);
2785                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2786         }
2787 #endif
2788         return err;
2789 }
2790
2791 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2792 {
2793         struct super_block *sb = dentry->d_sb;
2794         struct ext3_sb_info *sbi = EXT3_SB(sb);
2795         struct ext3_super_block *es = sbi->s_es;
2796         u64 fsid;
2797
2798         if (test_opt(sb, MINIX_DF)) {
2799                 sbi->s_overhead_last = 0;
2800         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2801                 unsigned long ngroups = sbi->s_groups_count, i;
2802                 ext3_fsblk_t overhead = 0;
2803                 smp_rmb();
2804
2805                 /*
2806                  * Compute the overhead (FS structures).  This is constant
2807                  * for a given filesystem unless the number of block groups
2808                  * changes so we cache the previous value until it does.
2809                  */
2810
2811                 /*
2812                  * All of the blocks before first_data_block are
2813                  * overhead
2814                  */
2815                 overhead = le32_to_cpu(es->s_first_data_block);
2816
2817                 /*
2818                  * Add the overhead attributed to the superblock and
2819                  * block group descriptors.  If the sparse superblocks
2820                  * feature is turned on, then not all groups have this.
2821                  */
2822                 for (i = 0; i < ngroups; i++) {
2823                         overhead += ext3_bg_has_super(sb, i) +
2824                                 ext3_bg_num_gdb(sb, i);
2825                         cond_resched();
2826                 }
2827
2828                 /*
2829                  * Every block group has an inode bitmap, a block
2830                  * bitmap, and an inode table.
2831                  */
2832                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2833
2834                 /* Add the internal journal blocks as well */
2835                 if (sbi->s_journal && !sbi->journal_bdev)
2836                         overhead += sbi->s_journal->j_maxlen;
2837
2838                 sbi->s_overhead_last = overhead;
2839                 smp_wmb();
2840                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2841         }
2842
2843         buf->f_type = EXT3_SUPER_MAGIC;
2844         buf->f_bsize = sb->s_blocksize;
2845         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2846         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2847         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2848         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2849                 buf->f_bavail = 0;
2850         buf->f_files = le32_to_cpu(es->s_inodes_count);
2851         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2852         buf->f_namelen = EXT3_NAME_LEN;
2853         fsid = le64_to_cpup((void *)es->s_uuid) ^
2854                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2855         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2856         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2857         return 0;
2858 }
2859
2860 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2861  * is locked for write. Otherwise the are possible deadlocks:
2862  * Process 1                         Process 2
2863  * ext3_create()                     quota_sync()
2864  *   journal_start()                   write_dquot()
2865  *   dquot_initialize()                       down(dqio_mutex)
2866  *     down(dqio_mutex)                    journal_start()
2867  *
2868  */
2869
2870 #ifdef CONFIG_QUOTA
2871
2872 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2873 {
2874         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
2875 }
2876
2877 static int ext3_write_dquot(struct dquot *dquot)
2878 {
2879         int ret, err;
2880         handle_t *handle;
2881         struct inode *inode;
2882
2883         inode = dquot_to_inode(dquot);
2884         handle = ext3_journal_start(inode,
2885                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2886         if (IS_ERR(handle))
2887                 return PTR_ERR(handle);
2888         ret = dquot_commit(dquot);
2889         err = ext3_journal_stop(handle);
2890         if (!ret)
2891                 ret = err;
2892         return ret;
2893 }
2894
2895 static int ext3_acquire_dquot(struct dquot *dquot)
2896 {
2897         int ret, err;
2898         handle_t *handle;
2899
2900         handle = ext3_journal_start(dquot_to_inode(dquot),
2901                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2902         if (IS_ERR(handle))
2903                 return PTR_ERR(handle);
2904         ret = dquot_acquire(dquot);
2905         err = ext3_journal_stop(handle);
2906         if (!ret)
2907                 ret = err;
2908         return ret;
2909 }
2910
2911 static int ext3_release_dquot(struct dquot *dquot)
2912 {
2913         int ret, err;
2914         handle_t *handle;
2915
2916         handle = ext3_journal_start(dquot_to_inode(dquot),
2917                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2918         if (IS_ERR(handle)) {
2919                 /* Release dquot anyway to avoid endless cycle in dqput() */
2920                 dquot_release(dquot);
2921                 return PTR_ERR(handle);
2922         }
2923         ret = dquot_release(dquot);
2924         err = ext3_journal_stop(handle);
2925         if (!ret)
2926                 ret = err;
2927         return ret;
2928 }
2929
2930 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2931 {
2932         /* Are we journaling quotas? */
2933         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2934             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2935                 dquot_mark_dquot_dirty(dquot);
2936                 return ext3_write_dquot(dquot);
2937         } else {
2938                 return dquot_mark_dquot_dirty(dquot);
2939         }
2940 }
2941
2942 static int ext3_write_info(struct super_block *sb, int type)
2943 {
2944         int ret, err;
2945         handle_t *handle;
2946
2947         /* Data block + inode block */
2948         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2949         if (IS_ERR(handle))
2950                 return PTR_ERR(handle);
2951         ret = dquot_commit_info(sb, type);
2952         err = ext3_journal_stop(handle);
2953         if (!ret)
2954                 ret = err;
2955         return ret;
2956 }
2957
2958 /*
2959  * Turn on quotas during mount time - we need to find
2960  * the quota file and such...
2961  */
2962 static int ext3_quota_on_mount(struct super_block *sb, int type)
2963 {
2964         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2965                                         EXT3_SB(sb)->s_jquota_fmt, type);
2966 }
2967
2968 /*
2969  * Standard function to be called on quota_on
2970  */
2971 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2972                          struct path *path)
2973 {
2974         int err;
2975
2976         if (!test_opt(sb, QUOTA))
2977                 return -EINVAL;
2978
2979         /* Quotafile not on the same filesystem? */
2980         if (path->dentry->d_sb != sb)
2981                 return -EXDEV;
2982         /* Journaling quota? */
2983         if (EXT3_SB(sb)->s_qf_names[type]) {
2984                 /* Quotafile not of fs root? */
2985                 if (path->dentry->d_parent != sb->s_root)
2986                         ext3_msg(sb, KERN_WARNING,
2987                                 "warning: Quota file not on filesystem root. "
2988                                 "Journaled quota will not work.");
2989         }
2990
2991         /*
2992          * When we journal data on quota file, we have to flush journal to see
2993          * all updates to the file when we bypass pagecache...
2994          */
2995         if (ext3_should_journal_data(path->dentry->d_inode)) {
2996                 /*
2997                  * We don't need to lock updates but journal_flush() could
2998                  * otherwise be livelocked...
2999                  */
3000                 journal_lock_updates(EXT3_SB(sb)->s_journal);
3001                 err = journal_flush(EXT3_SB(sb)->s_journal);
3002                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
3003                 if (err)
3004                         return err;
3005         }
3006
3007         return dquot_quota_on(sb, type, format_id, path);
3008 }
3009
3010 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3011  * acquiring the locks... As quota files are never truncated and quota code
3012  * itself serializes the operations (and no one else should touch the files)
3013  * we don't have to be afraid of races */
3014 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
3015                                size_t len, loff_t off)
3016 {
3017         struct inode *inode = sb_dqopt(sb)->files[type];
3018         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3019         int err = 0;
3020         int offset = off & (sb->s_blocksize - 1);
3021         int tocopy;
3022         size_t toread;
3023         struct buffer_head *bh;
3024         loff_t i_size = i_size_read(inode);
3025
3026         if (off > i_size)
3027                 return 0;
3028         if (off+len > i_size)
3029                 len = i_size-off;
3030         toread = len;
3031         while (toread > 0) {
3032                 tocopy = sb->s_blocksize - offset < toread ?
3033                                 sb->s_blocksize - offset : toread;
3034                 bh = ext3_bread(NULL, inode, blk, 0, &err);
3035                 if (err)
3036                         return err;
3037                 if (!bh)        /* A hole? */
3038                         memset(data, 0, tocopy);
3039                 else
3040                         memcpy(data, bh->b_data+offset, tocopy);
3041                 brelse(bh);
3042                 offset = 0;
3043                 toread -= tocopy;
3044                 data += tocopy;
3045                 blk++;
3046         }
3047         return len;
3048 }
3049
3050 /* Write to quotafile (we know the transaction is already started and has
3051  * enough credits) */
3052 static ssize_t ext3_quota_write(struct super_block *sb, int type,
3053                                 const char *data, size_t len, loff_t off)
3054 {
3055         struct inode *inode = sb_dqopt(sb)->files[type];
3056         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3057         int err = 0;
3058         int offset = off & (sb->s_blocksize - 1);
3059         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
3060         struct buffer_head *bh;
3061         handle_t *handle = journal_current_handle();
3062
3063         if (!handle) {
3064                 ext3_msg(sb, KERN_WARNING,
3065                         "warning: quota write (off=%llu, len=%llu)"
3066                         " cancelled because transaction is not started.",
3067                         (unsigned long long)off, (unsigned long long)len);
3068                 return -EIO;
3069         }
3070
3071         /*
3072          * Since we account only one data block in transaction credits,
3073          * then it is impossible to cross a block boundary.
3074          */
3075         if (sb->s_blocksize - offset < len) {
3076                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3077                         " cancelled because not block aligned",
3078                         (unsigned long long)off, (unsigned long long)len);
3079                 return -EIO;
3080         }
3081         bh = ext3_bread(handle, inode, blk, 1, &err);
3082         if (!bh)
3083                 goto out;
3084         if (journal_quota) {
3085                 err = ext3_journal_get_write_access(handle, bh);
3086                 if (err) {
3087                         brelse(bh);
3088                         goto out;
3089                 }
3090         }
3091         lock_buffer(bh);
3092         memcpy(bh->b_data+offset, data, len);
3093         flush_dcache_page(bh->b_page);
3094         unlock_buffer(bh);
3095         if (journal_quota)
3096                 err = ext3_journal_dirty_metadata(handle, bh);
3097         else {
3098                 /* Always do at least ordered writes for quotas */
3099                 err = ext3_journal_dirty_data(handle, bh);
3100                 mark_buffer_dirty(bh);
3101         }
3102         brelse(bh);
3103 out:
3104         if (err)
3105                 return err;
3106         if (inode->i_size < off + len) {
3107                 i_size_write(inode, off + len);
3108                 EXT3_I(inode)->i_disksize = inode->i_size;
3109         }
3110         inode->i_version++;
3111         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3112         ext3_mark_inode_dirty(handle, inode);
3113         return len;
3114 }
3115
3116 #endif
3117
3118 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3119         int flags, const char *dev_name, void *data)
3120 {
3121         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3122 }
3123
3124 static struct file_system_type ext3_fs_type = {
3125         .owner          = THIS_MODULE,
3126         .name           = "ext3",
3127         .mount          = ext3_mount,
3128         .kill_sb        = kill_block_super,
3129         .fs_flags       = FS_REQUIRES_DEV,
3130 };
3131 MODULE_ALIAS_FS("ext3");
3132
3133 static int __init init_ext3_fs(void)
3134 {
3135         int err = init_ext3_xattr();
3136         if (err)
3137                 return err;
3138         err = init_inodecache();
3139         if (err)
3140                 goto out1;
3141         err = register_filesystem(&ext3_fs_type);
3142         if (err)
3143                 goto out;
3144         return 0;
3145 out:
3146         destroy_inodecache();
3147 out1:
3148         exit_ext3_xattr();
3149         return err;
3150 }
3151
3152 static void __exit exit_ext3_fs(void)
3153 {
3154         unregister_filesystem(&ext3_fs_type);
3155         destroy_inodecache();
3156         exit_ext3_xattr();
3157 }
3158
3159 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3160 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3161 MODULE_LICENSE("GPL");
3162 module_init(init_ext3_fs)
3163 module_exit(exit_ext3_fs)