fs/ext3/super.c

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