fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/log2.h>
  39 #include <linux/cleancache.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 #define CREATE_TRACE_POINTS
  48 #include <trace/events/ext3.h>
  49
  50 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  51   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  52 #else
  53   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  54 #endif
  55
  56 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  57                              unsigned long journal_devnum);
  58 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  59                                unsigned int);
  60 static int ext3_commit_super(struct super_block *sb,
  61                                struct ext3_super_block *es,
  62                                int sync);
  63 static void ext3_mark_recovery_complete(struct super_block * sb,
  64                                         struct ext3_super_block * es);
  65 static void ext3_clear_journal_err(struct super_block * sb,
  66                                    struct ext3_super_block * es);
  67 static int ext3_sync_fs(struct super_block *sb, int wait);
  68 static const char *ext3_decode_error(struct super_block * sb, int errno,
  69                                      char nbuf[16]);
  70 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  71 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  72 static int ext3_unfreeze(struct super_block *sb);
  73 static int ext3_freeze(struct super_block *sb);
  74
  75 /*
  76  * Wrappers for journal_start/end.
  77  *
  78  * The only special thing we need to do here is to make sure that all
  79  * journal_end calls result in the superblock being marked dirty, so
  80  * that sync() will call the filesystem's write_super callback if
  81  * appropriate.
  82  */
  83 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  84 {
  85         journal_t *journal;
  86
  87         if (sb->s_flags & MS_RDONLY)
  88                 return ERR_PTR(-EROFS);
  89
  90         /* Special case here: if the journal has aborted behind our
  91          * backs (eg. EIO in the commit thread), then we still need to
  92          * take the FS itself readonly cleanly. */
  93         journal = EXT3_SB(sb)->s_journal;
  94         if (is_journal_aborted(journal)) {
  95                 ext3_abort(sb, __func__,
  96                            "Detected aborted journal");
  97                 return ERR_PTR(-EROFS);
  98         }
  99
 100         return journal_start(journal, nblocks);
 101 }
 102
 103 /*
 104  * The only special thing we need to do here is to make sure that all
 105  * journal_stop calls result in the superblock being marked dirty, so
 106  * that sync() will call the filesystem's write_super callback if
 107  * appropriate.
 108  */
 109 int __ext3_journal_stop(const char *where, handle_t *handle)
 110 {
 111         struct super_block *sb;
 112         int err;
 113         int rc;
 114
 115         sb = handle->h_transaction->t_journal->j_private;
 116         err = handle->h_err;
 117         rc = journal_stop(handle);
 118
 119         if (!err)
 120                 err = rc;
 121         if (err)
 122                 __ext3_std_error(sb, where, err);
 123         return err;
 124 }
 125
 126 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 127                 struct buffer_head *bh, handle_t *handle, int err)
 128 {
 129         char nbuf[16];
 130         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 131
 132         if (bh)
 133                 BUFFER_TRACE(bh, "abort");
 134
 135         if (!handle->h_err)
 136                 handle->h_err = err;
 137
 138         if (is_handle_aborted(handle))
 139                 return;
 140
 141         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 142                 caller, errstr, err_fn);
 143
 144         journal_abort_handle(handle);
 145 }
 146
 147 void ext3_msg(struct super_block *sb, const char *prefix,
 148                 const char *fmt, ...)
 149 {
 150         struct va_format vaf;
 151         va_list args;
 152
 153         va_start(args, fmt);
 154
 155         vaf.fmt = fmt;
 156         vaf.va = &args;
 157
 158         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 159
 160         va_end(args);
 161 }
 162
 163 /* Deal with the reporting of failure conditions on a filesystem such as
 164  * inconsistencies detected or read IO failures.
 165  *
 166  * On ext2, we can store the error state of the filesystem in the
 167  * superblock.  That is not possible on ext3, because we may have other
 168  * write ordering constraints on the superblock which prevent us from
 169  * writing it out straight away; and given that the journal is about to
 170  * be aborted, we can't rely on the current, or future, transactions to
 171  * write out the superblock safely.
 172  *
 173  * We'll just use the journal_abort() error code to record an error in
 174  * the journal instead.  On recovery, the journal will complain about
 175  * that error until we've noted it down and cleared it.
 176  */
 177
 178 static void ext3_handle_error(struct super_block *sb)
 179 {
 180         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 181
 182         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 183         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 184
 185         if (sb->s_flags & MS_RDONLY)
 186                 return;
 187
 188         if (!test_opt (sb, ERRORS_CONT)) {
 189                 journal_t *journal = EXT3_SB(sb)->s_journal;
 190
 191                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 192                 if (journal)
 193                         journal_abort(journal, -EIO);
 194         }
 195         if (test_opt (sb, ERRORS_RO)) {
 196                 ext3_msg(sb, KERN_CRIT,
 197                         "error: remounting filesystem read-only");
 198                 sb->s_flags |= MS_RDONLY;
 199         }
 200         ext3_commit_super(sb, es, 1);
 201         if (test_opt(sb, ERRORS_PANIC))
 202                 panic("EXT3-fs (%s): panic forced after error\n",
 203                         sb->s_id);
 204 }
 205
 206 void ext3_error(struct super_block *sb, const char *function,
 207                 const char *fmt, ...)
 208 {
 209         struct va_format vaf;
 210         va_list args;
 211
 212         va_start(args, fmt);
 213
 214         vaf.fmt = fmt;
 215         vaf.va = &args;
 216
 217         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 218                sb->s_id, function, &vaf);
 219
 220         va_end(args);
 221
 222         ext3_handle_error(sb);
 223 }
 224
 225 static const char *ext3_decode_error(struct super_block * sb, int errno,
 226                                      char nbuf[16])
 227 {
 228         char *errstr = NULL;
 229
 230         switch (errno) {
 231         case -EIO:
 232                 errstr = "IO failure";
 233                 break;
 234         case -ENOMEM:
 235                 errstr = "Out of memory";
 236                 break;
 237         case -EROFS:
 238                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 239                         errstr = "Journal has aborted";
 240                 else
 241                         errstr = "Readonly filesystem";
 242                 break;
 243         default:
 244                 /* If the caller passed in an extra buffer for unknown
 245                  * errors, textualise them now.  Else we just return
 246                  * NULL. */
 247                 if (nbuf) {
 248                         /* Check for truncated error codes... */
 249                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 250                                 errstr = nbuf;
 251                 }
 252                 break;
 253         }
 254
 255         return errstr;
 256 }
 257
 258 /* __ext3_std_error decodes expected errors from journaling functions
 259  * automatically and invokes the appropriate error response.  */
 260
 261 void __ext3_std_error (struct super_block * sb, const char * function,
 262                        int errno)
 263 {
 264         char nbuf[16];
 265         const char *errstr;
 266
 267         /* Special case: if the error is EROFS, and we're not already
 268          * inside a transaction, then there's really no point in logging
 269          * an error. */
 270         if (errno == -EROFS && journal_current_handle() == NULL &&
 271             (sb->s_flags & MS_RDONLY))
 272                 return;
 273
 274         errstr = ext3_decode_error(sb, errno, nbuf);
 275         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 276
 277         ext3_handle_error(sb);
 278 }
 279
 280 /*
 281  * ext3_abort is a much stronger failure handler than ext3_error.  The
 282  * abort function may be used to deal with unrecoverable failures such
 283  * as journal IO errors or ENOMEM at a critical moment in log management.
 284  *
 285  * We unconditionally force the filesystem into an ABORT|READONLY state,
 286  * unless the error response on the fs has been set to panic in which
 287  * case we take the easy way out and panic immediately.
 288  */
 289
 290 void ext3_abort(struct super_block *sb, const char *function,
 291                  const char *fmt, ...)
 292 {
 293         struct va_format vaf;
 294         va_list args;
 295
 296         va_start(args, fmt);
 297
 298         vaf.fmt = fmt;
 299         vaf.va = &args;
 300
 301         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 302                sb->s_id, function, &vaf);
 303
 304         va_end(args);
 305
 306         if (test_opt(sb, ERRORS_PANIC))
 307                 panic("EXT3-fs: panic from previous error\n");
 308
 309         if (sb->s_flags & MS_RDONLY)
 310                 return;
 311
 312         ext3_msg(sb, KERN_CRIT,
 313                 "error: remounting filesystem read-only");
 314         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 315         sb->s_flags |= MS_RDONLY;
 316         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 317         if (EXT3_SB(sb)->s_journal)
 318                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 319 }
 320
 321 void ext3_warning(struct super_block *sb, const char *function,
 322                   const char *fmt, ...)
 323 {
 324         struct va_format vaf;
 325         va_list args;
 326
 327         va_start(args, fmt);
 328
 329         vaf.fmt = fmt;
 330         vaf.va = &args;
 331
 332         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 333                sb->s_id, function, &vaf);
 334
 335         va_end(args);
 336 }
 337
 338 void ext3_update_dynamic_rev(struct super_block *sb)
 339 {
 340         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 341
 342         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 343                 return;
 344
 345         ext3_msg(sb, KERN_WARNING,
 346                 "warning: updating to rev %d because of "
 347                 "new feature flag, running e2fsck is recommended",
 348                 EXT3_DYNAMIC_REV);
 349
 350         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 351         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 352         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 353         /* leave es->s_feature_*compat flags alone */
 354         /* es->s_uuid will be set by e2fsck if empty */
 355
 356         /*
 357          * The rest of the superblock fields should be zero, and if not it
 358          * means they are likely already in use, so leave them alone.  We
 359          * can leave it up to e2fsck to clean up any inconsistencies there.
 360          */
 361 }
 362
 363 /*
 364  * Open the external journal device
 365  */
 366 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 367 {
 368         struct block_device *bdev;
 369         char b[BDEVNAME_SIZE];
 370
 371         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 372         if (IS_ERR(bdev))
 373                 goto fail;
 374         return bdev;
 375
 376 fail:
 377         ext3_msg(sb, "error: failed to open journal device %s: %ld",
 378                 __bdevname(dev, b), PTR_ERR(bdev));
 379
 380         return NULL;
 381 }
 382
 383 /*
 384  * Release the journal device
 385  */
 386 static int ext3_blkdev_put(struct block_device *bdev)
 387 {
 388         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 389 }
 390
 391 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 392 {
 393         struct block_device *bdev;
 394         int ret = -ENODEV;
 395
 396         bdev = sbi->journal_bdev;
 397         if (bdev) {
 398                 ret = ext3_blkdev_put(bdev);
 399                 sbi->journal_bdev = NULL;
 400         }
 401         return ret;
 402 }
 403
 404 static inline struct inode *orphan_list_entry(struct list_head *l)
 405 {
 406         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 407 }
 408
 409 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 410 {
 411         struct list_head *l;
 412
 413         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 414                le32_to_cpu(sbi->s_es->s_last_orphan));
 415
 416         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 417         list_for_each(l, &sbi->s_orphan) {
 418                 struct inode *inode = orphan_list_entry(l);
 419                 ext3_msg(sb, KERN_ERR, "  "
 420                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 421                        inode->i_sb->s_id, inode->i_ino, inode,
 422                        inode->i_mode, inode->i_nlink,
 423                        NEXT_ORPHAN(inode));
 424         }
 425 }
 426
 427 static void ext3_put_super (struct super_block * sb)
 428 {
 429         struct ext3_sb_info *sbi = EXT3_SB(sb);
 430         struct ext3_super_block *es = sbi->s_es;
 431         int i, err;
 432
 433         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 434         ext3_xattr_put_super(sb);
 435         err = journal_destroy(sbi->s_journal);
 436         sbi->s_journal = NULL;
 437         if (err < 0)
 438                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 439
 440         if (!(sb->s_flags & MS_RDONLY)) {
 441                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 442                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 443                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 444                 mark_buffer_dirty(sbi->s_sbh);
 445                 ext3_commit_super(sb, es, 1);
 446         }
 447
 448         for (i = 0; i < sbi->s_gdb_count; i++)
 449                 brelse(sbi->s_group_desc[i]);
 450         kfree(sbi->s_group_desc);
 451         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 452         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 453         percpu_counter_destroy(&sbi->s_dirs_counter);
 454         brelse(sbi->s_sbh);
 455 #ifdef CONFIG_QUOTA
 456         for (i = 0; i < MAXQUOTAS; i++)
 457                 kfree(sbi->s_qf_names[i]);
 458 #endif
 459
 460         /* Debugging code just in case the in-memory inode orphan list
 461          * isn't empty.  The on-disk one can be non-empty if we've
 462          * detected an error and taken the fs readonly, but the
 463          * in-memory list had better be clean by this point. */
 464         if (!list_empty(&sbi->s_orphan))
 465                 dump_orphan_list(sb, sbi);
 466         J_ASSERT(list_empty(&sbi->s_orphan));
 467
 468         invalidate_bdev(sb->s_bdev);
 469         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 470                 /*
 471                  * Invalidate the journal device's buffers.  We don't want them
 472                  * floating about in memory - the physical journal device may
 473                  * hotswapped, and it breaks the `ro-after' testing code.
 474                  */
 475                 sync_blockdev(sbi->journal_bdev);
 476                 invalidate_bdev(sbi->journal_bdev);
 477                 ext3_blkdev_remove(sbi);
 478         }
 479         sb->s_fs_info = NULL;
 480         kfree(sbi->s_blockgroup_lock);
 481         kfree(sbi);
 482 }
 483
 484 static struct kmem_cache *ext3_inode_cachep;
 485
 486 /*
 487  * Called inside transaction, so use GFP_NOFS
 488  */
 489 static struct inode *ext3_alloc_inode(struct super_block *sb)
 490 {
 491         struct ext3_inode_info *ei;
 492
 493         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 494         if (!ei)
 495                 return NULL;
 496         ei->i_block_alloc_info = NULL;
 497         ei->vfs_inode.i_version = 1;
 498         atomic_set(&ei->i_datasync_tid, 0);
 499         atomic_set(&ei->i_sync_tid, 0);
 500         return &ei->vfs_inode;
 501 }
 502
 503 static int ext3_drop_inode(struct inode *inode)
 504 {
 505         int drop = generic_drop_inode(inode);
 506
 507         trace_ext3_drop_inode(inode, drop);
 508         return drop;
 509 }
 510
 511 static void ext3_i_callback(struct rcu_head *head)
 512 {
 513         struct inode *inode = container_of(head, struct inode, i_rcu);
 514         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 515 }
 516
 517 static void ext3_destroy_inode(struct inode *inode)
 518 {
 519         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 520                 printk("EXT3 Inode %p: orphan list check failed!\n",
 521                         EXT3_I(inode));
 522                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 523                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 524                                 false);
 525                 dump_stack();
 526         }
 527         call_rcu(&inode->i_rcu, ext3_i_callback);
 528 }
 529
 530 static void init_once(void *foo)
 531 {
 532         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 533
 534         INIT_LIST_HEAD(&ei->i_orphan);
 535 #ifdef CONFIG_EXT3_FS_XATTR
 536         init_rwsem(&ei->xattr_sem);
 537 #endif
 538         mutex_init(&ei->truncate_mutex);
 539         inode_init_once(&ei->vfs_inode);
 540 }
 541
 542 static int init_inodecache(void)
 543 {
 544         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 545                                              sizeof(struct ext3_inode_info),
 546                                              0, (SLAB_RECLAIM_ACCOUNT|
 547                                                 SLAB_MEM_SPREAD),
 548                                              init_once);
 549         if (ext3_inode_cachep == NULL)
 550                 return -ENOMEM;
 551         return 0;
 552 }
 553
 554 static void destroy_inodecache(void)
 555 {
 556         kmem_cache_destroy(ext3_inode_cachep);
 557 }
 558
 559 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 560 {
 561 #if defined(CONFIG_QUOTA)
 562         struct ext3_sb_info *sbi = EXT3_SB(sb);
 563
 564         if (sbi->s_jquota_fmt) {
 565                 char *fmtname = "";
 566
 567                 switch (sbi->s_jquota_fmt) {
 568                 case QFMT_VFS_OLD:
 569                         fmtname = "vfsold";
 570                         break;
 571                 case QFMT_VFS_V0:
 572                         fmtname = "vfsv0";
 573                         break;
 574                 case QFMT_VFS_V1:
 575                         fmtname = "vfsv1";
 576                         break;
 577                 }
 578                 seq_printf(seq, ",jqfmt=%s", fmtname);
 579         }
 580
 581         if (sbi->s_qf_names[USRQUOTA])
 582                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 583
 584         if (sbi->s_qf_names[GRPQUOTA])
 585                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 586
 587         if (test_opt(sb, USRQUOTA))
 588                 seq_puts(seq, ",usrquota");
 589
 590         if (test_opt(sb, GRPQUOTA))
 591                 seq_puts(seq, ",grpquota");
 592 #endif
 593 }
 594
 595 static char *data_mode_string(unsigned long mode)
 596 {
 597         switch (mode) {
 598         case EXT3_MOUNT_JOURNAL_DATA:
 599                 return "journal";
 600         case EXT3_MOUNT_ORDERED_DATA:
 601                 return "ordered";
 602         case EXT3_MOUNT_WRITEBACK_DATA:
 603                 return "writeback";
 604         }
 605         return "unknown";
 606 }
 607
 608 /*
 609  * Show an option if
 610  *  - it's set to a non-default value OR
 611  *  - if the per-sb default is different from the global default
 612  */
 613 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 614 {
 615         struct super_block *sb = root->d_sb;
 616         struct ext3_sb_info *sbi = EXT3_SB(sb);
 617         struct ext3_super_block *es = sbi->s_es;
 618         unsigned long def_mount_opts;
 619
 620         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 621
 622         if (sbi->s_sb_block != 1)
 623                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 624         if (test_opt(sb, MINIX_DF))
 625                 seq_puts(seq, ",minixdf");
 626         if (test_opt(sb, GRPID))
 627                 seq_puts(seq, ",grpid");
 628         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 629                 seq_puts(seq, ",nogrpid");
 630         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 631             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 632                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 633         }
 634         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 635             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 636                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 637         }
 638         if (test_opt(sb, ERRORS_RO)) {
 639                 int def_errors = le16_to_cpu(es->s_errors);
 640
 641                 if (def_errors == EXT3_ERRORS_PANIC ||
 642                     def_errors == EXT3_ERRORS_CONTINUE) {
 643                         seq_puts(seq, ",errors=remount-ro");
 644                 }
 645         }
 646         if (test_opt(sb, ERRORS_CONT))
 647                 seq_puts(seq, ",errors=continue");
 648         if (test_opt(sb, ERRORS_PANIC))
 649                 seq_puts(seq, ",errors=panic");
 650         if (test_opt(sb, NO_UID32))
 651                 seq_puts(seq, ",nouid32");
 652         if (test_opt(sb, DEBUG))
 653                 seq_puts(seq, ",debug");
 654 #ifdef CONFIG_EXT3_FS_XATTR
 655         if (test_opt(sb, XATTR_USER))
 656                 seq_puts(seq, ",user_xattr");
 657         if (!test_opt(sb, XATTR_USER) &&
 658             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 659                 seq_puts(seq, ",nouser_xattr");
 660         }
 661 #endif
 662 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 663         if (test_opt(sb, POSIX_ACL))
 664                 seq_puts(seq, ",acl");
 665         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 666                 seq_puts(seq, ",noacl");
 667 #endif
 668         if (!test_opt(sb, RESERVATION))
 669                 seq_puts(seq, ",noreservation");
 670         if (sbi->s_commit_interval) {
 671                 seq_printf(seq, ",commit=%u",
 672                            (unsigned) (sbi->s_commit_interval / HZ));
 673         }
 674
 675         /*
 676          * Always display barrier state so it's clear what the status is.
 677          */
 678         seq_puts(seq, ",barrier=");
 679         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 680         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 681         if (test_opt(sb, DATA_ERR_ABORT))
 682                 seq_puts(seq, ",data_err=abort");
 683
 684         if (test_opt(sb, NOLOAD))
 685                 seq_puts(seq, ",norecovery");
 686
 687         ext3_show_quota_options(seq, sb);
 688
 689         return 0;
 690 }
 691
 692
 693 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 694                 u64 ino, u32 generation)
 695 {
 696         struct inode *inode;
 697
 698         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 699                 return ERR_PTR(-ESTALE);
 700         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 701                 return ERR_PTR(-ESTALE);
 702
 703         /* iget isn't really right if the inode is currently unallocated!!
 704          *
 705          * ext3_read_inode will return a bad_inode if the inode had been
 706          * deleted, so we should be safe.
 707          *
 708          * Currently we don't know the generation for parent directory, so
 709          * a generation of 0 means "accept any"
 710          */
 711         inode = ext3_iget(sb, ino);
 712         if (IS_ERR(inode))
 713                 return ERR_CAST(inode);
 714         if (generation && inode->i_generation != generation) {
 715                 iput(inode);
 716                 return ERR_PTR(-ESTALE);
 717         }
 718
 719         return inode;
 720 }
 721
 722 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 723                 int fh_len, int fh_type)
 724 {
 725         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 726                                     ext3_nfs_get_inode);
 727 }
 728
 729 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 730                 int fh_len, int fh_type)
 731 {
 732         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 733                                     ext3_nfs_get_inode);
 734 }
 735
 736 /*
 737  * Try to release metadata pages (indirect blocks, directories) which are
 738  * mapped via the block device.  Since these pages could have journal heads
 739  * which would prevent try_to_free_buffers() from freeing them, we must use
 740  * jbd layer's try_to_free_buffers() function to release them.
 741  */
 742 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 743                                  gfp_t wait)
 744 {
 745         journal_t *journal = EXT3_SB(sb)->s_journal;
 746
 747         WARN_ON(PageChecked(page));
 748         if (!page_has_buffers(page))
 749                 return 0;
 750         if (journal)
 751                 return journal_try_to_free_buffers(journal, page,
 752                                                    wait & ~__GFP_WAIT);
 753         return try_to_free_buffers(page);
 754 }
 755
 756 #ifdef CONFIG_QUOTA
 757 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 758 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 759
 760 static int ext3_write_dquot(struct dquot *dquot);
 761 static int ext3_acquire_dquot(struct dquot *dquot);
 762 static int ext3_release_dquot(struct dquot *dquot);
 763 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 764 static int ext3_write_info(struct super_block *sb, int type);
 765 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 766                          struct path *path);
 767 static int ext3_quota_on_mount(struct super_block *sb, int type);
 768 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 769                                size_t len, loff_t off);
 770 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 771                                 const char *data, size_t len, loff_t off);
 772
 773 static const struct dquot_operations ext3_quota_operations = {
 774         .write_dquot    = ext3_write_dquot,
 775         .acquire_dquot  = ext3_acquire_dquot,
 776         .release_dquot  = ext3_release_dquot,
 777         .mark_dirty     = ext3_mark_dquot_dirty,
 778         .write_info     = ext3_write_info,
 779         .alloc_dquot    = dquot_alloc,
 780         .destroy_dquot  = dquot_destroy,
 781 };
 782
 783 static const struct quotactl_ops ext3_qctl_operations = {
 784         .quota_on       = ext3_quota_on,
 785         .quota_off      = dquot_quota_off,
 786         .quota_sync     = dquot_quota_sync,
 787         .get_info       = dquot_get_dqinfo,
 788         .set_info       = dquot_set_dqinfo,
 789         .get_dqblk      = dquot_get_dqblk,
 790         .set_dqblk      = dquot_set_dqblk
 791 };
 792 #endif
 793
 794 static const struct super_operations ext3_sops = {
 795         .alloc_inode    = ext3_alloc_inode,
 796         .destroy_inode  = ext3_destroy_inode,
 797         .write_inode    = ext3_write_inode,
 798         .dirty_inode    = ext3_dirty_inode,
 799         .drop_inode     = ext3_drop_inode,
 800         .evict_inode    = ext3_evict_inode,
 801         .put_super      = ext3_put_super,
 802         .sync_fs        = ext3_sync_fs,
 803         .freeze_fs      = ext3_freeze,
 804         .unfreeze_fs    = ext3_unfreeze,
 805         .statfs         = ext3_statfs,
 806         .remount_fs     = ext3_remount,
 807         .show_options   = ext3_show_options,
 808 #ifdef CONFIG_QUOTA
 809         .quota_read     = ext3_quota_read,
 810         .quota_write    = ext3_quota_write,
 811 #endif
 812         .bdev_try_to_free_page = bdev_try_to_free_page,
 813 };
 814
 815 static const struct export_operations ext3_export_ops = {
 816         .fh_to_dentry = ext3_fh_to_dentry,
 817         .fh_to_parent = ext3_fh_to_parent,
 818         .get_parent = ext3_get_parent,
 819 };
 820
 821 enum {
 822         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 823         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 824         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 825         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 826         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 827         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 828         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 829         Opt_data_err_abort, Opt_data_err_ignore,
 830         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 831         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 832         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 833         Opt_resize, Opt_usrquota, Opt_grpquota
 834 };
 835
 836 static const match_table_t tokens = {
 837         {Opt_bsd_df, "bsddf"},
 838         {Opt_minix_df, "minixdf"},
 839         {Opt_grpid, "grpid"},
 840         {Opt_grpid, "bsdgroups"},
 841         {Opt_nogrpid, "nogrpid"},
 842         {Opt_nogrpid, "sysvgroups"},
 843         {Opt_resgid, "resgid=%u"},
 844         {Opt_resuid, "resuid=%u"},
 845         {Opt_sb, "sb=%u"},
 846         {Opt_err_cont, "errors=continue"},
 847         {Opt_err_panic, "errors=panic"},
 848         {Opt_err_ro, "errors=remount-ro"},
 849         {Opt_nouid32, "nouid32"},
 850         {Opt_nocheck, "nocheck"},
 851         {Opt_nocheck, "check=none"},
 852         {Opt_debug, "debug"},
 853         {Opt_oldalloc, "oldalloc"},
 854         {Opt_orlov, "orlov"},
 855         {Opt_user_xattr, "user_xattr"},
 856         {Opt_nouser_xattr, "nouser_xattr"},
 857         {Opt_acl, "acl"},
 858         {Opt_noacl, "noacl"},
 859         {Opt_reservation, "reservation"},
 860         {Opt_noreservation, "noreservation"},
 861         {Opt_noload, "noload"},
 862         {Opt_noload, "norecovery"},
 863         {Opt_nobh, "nobh"},
 864         {Opt_bh, "bh"},
 865         {Opt_commit, "commit=%u"},
 866         {Opt_journal_update, "journal=update"},
 867         {Opt_journal_inum, "journal=%u"},
 868         {Opt_journal_dev, "journal_dev=%u"},
 869         {Opt_abort, "abort"},
 870         {Opt_data_journal, "data=journal"},
 871         {Opt_data_ordered, "data=ordered"},
 872         {Opt_data_writeback, "data=writeback"},
 873         {Opt_data_err_abort, "data_err=abort"},
 874         {Opt_data_err_ignore, "data_err=ignore"},
 875         {Opt_offusrjquota, "usrjquota="},
 876         {Opt_usrjquota, "usrjquota=%s"},
 877         {Opt_offgrpjquota, "grpjquota="},
 878         {Opt_grpjquota, "grpjquota=%s"},
 879         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 880         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 881         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 882         {Opt_grpquota, "grpquota"},
 883         {Opt_noquota, "noquota"},
 884         {Opt_quota, "quota"},
 885         {Opt_usrquota, "usrquota"},
 886         {Opt_barrier, "barrier=%u"},
 887         {Opt_barrier, "barrier"},
 888         {Opt_nobarrier, "nobarrier"},
 889         {Opt_resize, "resize"},
 890         {Opt_err, NULL},
 891 };
 892
 893 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 894 {
 895         ext3_fsblk_t    sb_block;
 896         char            *options = (char *) *data;
 897
 898         if (!options || strncmp(options, "sb=", 3) != 0)
 899                 return 1;       /* Default location */
 900         options += 3;
 901         /*todo: use simple_strtoll with >32bit ext3 */
 902         sb_block = simple_strtoul(options, &options, 0);
 903         if (*options && *options != ',') {
 904                 ext3_msg(sb, "error: invalid sb specification: %s",
 905                        (char *) *data);
 906                 return 1;
 907         }
 908         if (*options == ',')
 909                 options++;
 910         *data = (void *) options;
 911         return sb_block;
 912 }
 913
 914 #ifdef CONFIG_QUOTA
 915 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 916 {
 917         struct ext3_sb_info *sbi = EXT3_SB(sb);
 918         char *qname;
 919
 920         if (sb_any_quota_loaded(sb) &&
 921                 !sbi->s_qf_names[qtype]) {
 922                 ext3_msg(sb, KERN_ERR,
 923                         "Cannot change journaled "
 924                         "quota options when quota turned on");
 925                 return 0;
 926         }
 927         qname = match_strdup(args);
 928         if (!qname) {
 929                 ext3_msg(sb, KERN_ERR,
 930                         "Not enough memory for storing quotafile name");
 931                 return 0;
 932         }
 933         if (sbi->s_qf_names[qtype] &&
 934                 strcmp(sbi->s_qf_names[qtype], qname)) {
 935                 ext3_msg(sb, KERN_ERR,
 936                         "%s quota file already specified", QTYPE2NAME(qtype));
 937                 kfree(qname);
 938                 return 0;
 939         }
 940         sbi->s_qf_names[qtype] = qname;
 941         if (strchr(sbi->s_qf_names[qtype], '/')) {
 942                 ext3_msg(sb, KERN_ERR,
 943                         "quotafile must be on filesystem root");
 944                 kfree(sbi->s_qf_names[qtype]);
 945                 sbi->s_qf_names[qtype] = NULL;
 946                 return 0;
 947         }
 948         set_opt(sbi->s_mount_opt, QUOTA);
 949         return 1;
 950 }
 951
 952 static int clear_qf_name(struct super_block *sb, int qtype) {
 953
 954         struct ext3_sb_info *sbi = EXT3_SB(sb);
 955
 956         if (sb_any_quota_loaded(sb) &&
 957                 sbi->s_qf_names[qtype]) {
 958                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 959                         " when quota turned on");
 960                 return 0;
 961         }
 962         /*
 963          * The space will be released later when all options are confirmed
 964          * to be correct
 965          */
 966         sbi->s_qf_names[qtype] = NULL;
 967         return 1;
 968 }
 969 #endif
 970
 971 static int parse_options (char *options, struct super_block *sb,
 972                           unsigned int *inum, unsigned long *journal_devnum,
 973                           ext3_fsblk_t *n_blocks_count, int is_remount)
 974 {
 975         struct ext3_sb_info *sbi = EXT3_SB(sb);
 976         char * p;
 977         substring_t args[MAX_OPT_ARGS];
 978         int data_opt = 0;
 979         int option;
 980 #ifdef CONFIG_QUOTA
 981         int qfmt;
 982 #endif
 983
 984         if (!options)
 985                 return 1;
 986
 987         while ((p = strsep (&options, ",")) != NULL) {
 988                 int token;
 989                 if (!*p)
 990                         continue;
 991                 /*
 992                  * Initialize args struct so we know whether arg was
 993                  * found; some options take optional arguments.
 994                  */
 995                 args[0].to = args[0].from = 0;
 996                 token = match_token(p, tokens, args);
 997                 switch (token) {
 998                 case Opt_bsd_df:
 999                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1000                         break;
1001                 case Opt_minix_df:
1002                         set_opt (sbi->s_mount_opt, MINIX_DF);
1003                         break;
1004                 case Opt_grpid:
1005                         set_opt (sbi->s_mount_opt, GRPID);
1006                         break;
1007                 case Opt_nogrpid:
1008                         clear_opt (sbi->s_mount_opt, GRPID);
1009                         break;
1010                 case Opt_resuid:
1011                         if (match_int(&args[0], &option))
1012                                 return 0;
1013                         sbi->s_resuid = option;
1014                         break;
1015                 case Opt_resgid:
1016                         if (match_int(&args[0], &option))
1017                                 return 0;
1018                         sbi->s_resgid = option;
1019                         break;
1020                 case Opt_sb:
1021                         /* handled by get_sb_block() instead of here */
1022                         /* *sb_block = match_int(&args[0]); */
1023                         break;
1024                 case Opt_err_panic:
1025                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1026                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1027                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1028                         break;
1029                 case Opt_err_ro:
1030                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1031                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1032                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1033                         break;
1034                 case Opt_err_cont:
1035                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1036                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1037                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1038                         break;
1039                 case Opt_nouid32:
1040                         set_opt (sbi->s_mount_opt, NO_UID32);
1041                         break;
1042                 case Opt_nocheck:
1043                         clear_opt (sbi->s_mount_opt, CHECK);
1044                         break;
1045                 case Opt_debug:
1046                         set_opt (sbi->s_mount_opt, DEBUG);
1047                         break;
1048                 case Opt_oldalloc:
1049                         ext3_msg(sb, KERN_WARNING,
1050                                 "Ignoring deprecated oldalloc option");
1051                         break;
1052                 case Opt_orlov:
1053                         ext3_msg(sb, KERN_WARNING,
1054                                 "Ignoring deprecated orlov option");
1055                         break;
1056 #ifdef CONFIG_EXT3_FS_XATTR
1057                 case Opt_user_xattr:
1058                         set_opt (sbi->s_mount_opt, XATTR_USER);
1059                         break;
1060                 case Opt_nouser_xattr:
1061                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1062                         break;
1063 #else
1064                 case Opt_user_xattr:
1065                 case Opt_nouser_xattr:
1066                         ext3_msg(sb, KERN_INFO,
1067                                 "(no)user_xattr options not supported");
1068                         break;
1069 #endif
1070 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1071                 case Opt_acl:
1072                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1073                         break;
1074                 case Opt_noacl:
1075                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1076                         break;
1077 #else
1078                 case Opt_acl:
1079                 case Opt_noacl:
1080                         ext3_msg(sb, KERN_INFO,
1081                                 "(no)acl options not supported");
1082                         break;
1083 #endif
1084                 case Opt_reservation:
1085                         set_opt(sbi->s_mount_opt, RESERVATION);
1086                         break;
1087                 case Opt_noreservation:
1088                         clear_opt(sbi->s_mount_opt, RESERVATION);
1089                         break;
1090                 case Opt_journal_update:
1091                         /* @@@ FIXME */
1092                         /* Eventually we will want to be able to create
1093                            a journal file here.  For now, only allow the
1094                            user to specify an existing inode to be the
1095                            journal file. */
1096                         if (is_remount) {
1097                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1098                                         "journal on remount");
1099                                 return 0;
1100                         }
1101                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1102                         break;
1103                 case Opt_journal_inum:
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                         *inum = option;
1112                         break;
1113                 case Opt_journal_dev:
1114                         if (is_remount) {
1115                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1116                                        "journal on remount");
1117                                 return 0;
1118                         }
1119                         if (match_int(&args[0], &option))
1120                                 return 0;
1121                         *journal_devnum = option;
1122                         break;
1123                 case Opt_noload:
1124                         set_opt (sbi->s_mount_opt, NOLOAD);
1125                         break;
1126                 case Opt_commit:
1127                         if (match_int(&args[0], &option))
1128                                 return 0;
1129                         if (option < 0)
1130                                 return 0;
1131                         if (option == 0)
1132                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1133                         sbi->s_commit_interval = HZ * option;
1134                         break;
1135                 case Opt_data_journal:
1136                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1137                         goto datacheck;
1138                 case Opt_data_ordered:
1139                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1140                         goto datacheck;
1141                 case Opt_data_writeback:
1142                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1143                 datacheck:
1144                         if (is_remount) {
1145                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1146                                         break;
1147                                 ext3_msg(sb, KERN_ERR,
1148                                         "error: cannot change "
1149                                         "data mode on remount. The filesystem "
1150                                         "is mounted in data=%s mode and you "
1151                                         "try to remount it in data=%s mode.",
1152                                         data_mode_string(test_opt(sb,
1153                                                         DATA_FLAGS)),
1154                                         data_mode_string(data_opt));
1155                                 return 0;
1156                         } else {
1157                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1158                                 sbi->s_mount_opt |= data_opt;
1159                         }
1160                         break;
1161                 case Opt_data_err_abort:
1162                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1163                         break;
1164                 case Opt_data_err_ignore:
1165                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1166                         break;
1167 #ifdef CONFIG_QUOTA
1168                 case Opt_usrjquota:
1169                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1170                                 return 0;
1171                         break;
1172                 case Opt_grpjquota:
1173                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1174                                 return 0;
1175                         break;
1176                 case Opt_offusrjquota:
1177                         if (!clear_qf_name(sb, USRQUOTA))
1178                                 return 0;
1179                         break;
1180                 case Opt_offgrpjquota:
1181                         if (!clear_qf_name(sb, GRPQUOTA))
1182                                 return 0;
1183                         break;
1184                 case Opt_jqfmt_vfsold:
1185                         qfmt = QFMT_VFS_OLD;
1186                         goto set_qf_format;
1187                 case Opt_jqfmt_vfsv0:
1188                         qfmt = QFMT_VFS_V0;
1189                         goto set_qf_format;
1190                 case Opt_jqfmt_vfsv1:
1191                         qfmt = QFMT_VFS_V1;
1192 set_qf_format:
1193                         if (sb_any_quota_loaded(sb) &&
1194                             sbi->s_jquota_fmt != qfmt) {
1195                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1196                                         "journaled quota options when "
1197                                         "quota turned on.");
1198                                 return 0;
1199                         }
1200                         sbi->s_jquota_fmt = qfmt;
1201                         break;
1202                 case Opt_quota:
1203                 case Opt_usrquota:
1204                         set_opt(sbi->s_mount_opt, QUOTA);
1205                         set_opt(sbi->s_mount_opt, USRQUOTA);
1206                         break;
1207                 case Opt_grpquota:
1208                         set_opt(sbi->s_mount_opt, QUOTA);
1209                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1210                         break;
1211                 case Opt_noquota:
1212                         if (sb_any_quota_loaded(sb)) {
1213                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1214                                         "quota options when quota turned on.");
1215                                 return 0;
1216                         }
1217                         clear_opt(sbi->s_mount_opt, QUOTA);
1218                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1219                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1220                         break;
1221 #else
1222                 case Opt_quota:
1223                 case Opt_usrquota:
1224                 case Opt_grpquota:
1225                         ext3_msg(sb, KERN_ERR,
1226                                 "error: quota options not supported.");
1227                         break;
1228                 case Opt_usrjquota:
1229                 case Opt_grpjquota:
1230                 case Opt_offusrjquota:
1231                 case Opt_offgrpjquota:
1232                 case Opt_jqfmt_vfsold:
1233                 case Opt_jqfmt_vfsv0:
1234                 case Opt_jqfmt_vfsv1:
1235                         ext3_msg(sb, KERN_ERR,
1236                                 "error: journaled quota options not "
1237                                 "supported.");
1238                         break;
1239                 case Opt_noquota:
1240                         break;
1241 #endif
1242                 case Opt_abort:
1243                         set_opt(sbi->s_mount_opt, ABORT);
1244                         break;
1245                 case Opt_nobarrier:
1246                         clear_opt(sbi->s_mount_opt, BARRIER);
1247                         break;
1248                 case Opt_barrier:
1249                         if (args[0].from) {
1250                                 if (match_int(&args[0], &option))
1251                                         return 0;
1252                         } else
1253                                 option = 1;     /* No argument, default to 1 */
1254                         if (option)
1255                                 set_opt(sbi->s_mount_opt, BARRIER);
1256                         else
1257                                 clear_opt(sbi->s_mount_opt, BARRIER);
1258                         break;
1259                 case Opt_ignore:
1260                         break;
1261                 case Opt_resize:
1262                         if (!is_remount) {
1263                                 ext3_msg(sb, KERN_ERR,
1264                                         "error: resize option only available "
1265                                         "for remount");
1266                                 return 0;
1267                         }
1268                         if (match_int(&args[0], &option) != 0)
1269                                 return 0;
1270                         *n_blocks_count = option;
1271                         break;
1272                 case Opt_nobh:
1273                         ext3_msg(sb, KERN_WARNING,
1274                                 "warning: ignoring deprecated nobh option");
1275                         break;
1276                 case Opt_bh:
1277                         ext3_msg(sb, KERN_WARNING,
1278                                 "warning: ignoring deprecated bh option");
1279                         break;
1280                 default:
1281                         ext3_msg(sb, KERN_ERR,
1282                                 "error: unrecognized mount option \"%s\" "
1283                                 "or missing value", p);
1284                         return 0;
1285                 }
1286         }
1287 #ifdef CONFIG_QUOTA
1288         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1289                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1290                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1291                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1292                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1293
1294                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1295                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1296                                         "format mixing.");
1297                         return 0;
1298                 }
1299
1300                 if (!sbi->s_jquota_fmt) {
1301                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1302                                         "not specified.");
1303                         return 0;
1304                 }
1305         } else {
1306                 if (sbi->s_jquota_fmt) {
1307                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1308                                         "specified with no journaling "
1309                                         "enabled.");
1310                         return 0;
1311                 }
1312         }
1313 #endif
1314         return 1;
1315 }
1316
1317 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1318                             int read_only)
1319 {
1320         struct ext3_sb_info *sbi = EXT3_SB(sb);
1321         int res = 0;
1322
1323         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1324                 ext3_msg(sb, KERN_ERR,
1325                         "error: revision level too high, "
1326                         "forcing read-only mode");
1327                 res = MS_RDONLY;
1328         }
1329         if (read_only)
1330                 return res;
1331         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1332                 ext3_msg(sb, KERN_WARNING,
1333                         "warning: mounting unchecked fs, "
1334                         "running e2fsck is recommended");
1335         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1336                 ext3_msg(sb, KERN_WARNING,
1337                         "warning: mounting fs with errors, "
1338                         "running e2fsck is recommended");
1339         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1340                  le16_to_cpu(es->s_mnt_count) >=
1341                         le16_to_cpu(es->s_max_mnt_count))
1342                 ext3_msg(sb, KERN_WARNING,
1343                         "warning: maximal mount count reached, "
1344                         "running e2fsck is recommended");
1345         else if (le32_to_cpu(es->s_checkinterval) &&
1346                 (le32_to_cpu(es->s_lastcheck) +
1347                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1348                 ext3_msg(sb, KERN_WARNING,
1349                         "warning: checktime reached, "
1350                         "running e2fsck is recommended");
1351 #if 0
1352                 /* @@@ We _will_ want to clear the valid bit if we find
1353                    inconsistencies, to force a fsck at reboot.  But for
1354                    a plain journaled filesystem we can keep it set as
1355                    valid forever! :) */
1356         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1357 #endif
1358         if (!le16_to_cpu(es->s_max_mnt_count))
1359                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1360         le16_add_cpu(&es->s_mnt_count, 1);
1361         es->s_mtime = cpu_to_le32(get_seconds());
1362         ext3_update_dynamic_rev(sb);
1363         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1364
1365         ext3_commit_super(sb, es, 1);
1366         if (test_opt(sb, DEBUG))
1367                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1368                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1369                         sb->s_blocksize,
1370                         sbi->s_groups_count,
1371                         EXT3_BLOCKS_PER_GROUP(sb),
1372                         EXT3_INODES_PER_GROUP(sb),
1373                         sbi->s_mount_opt);
1374
1375         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1376                 char b[BDEVNAME_SIZE];
1377                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1378                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1379         } else {
1380                 ext3_msg(sb, KERN_INFO, "using internal journal");
1381         }
1382         cleancache_init_fs(sb);
1383         return res;
1384 }
1385
1386 /* Called at mount-time, super-block is locked */
1387 static int ext3_check_descriptors(struct super_block *sb)
1388 {
1389         struct ext3_sb_info *sbi = EXT3_SB(sb);
1390         int i;
1391
1392         ext3_debug ("Checking group descriptors");
1393
1394         for (i = 0; i < sbi->s_groups_count; i++) {
1395                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1396                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1397                 ext3_fsblk_t last_block;
1398
1399                 if (i == sbi->s_groups_count - 1)
1400                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1401                 else
1402                         last_block = first_block +
1403                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1404
1405                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1406                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1407                 {
1408                         ext3_error (sb, "ext3_check_descriptors",
1409                                     "Block bitmap for group %d"
1410                                     " not in group (block %lu)!",
1411                                     i, (unsigned long)
1412                                         le32_to_cpu(gdp->bg_block_bitmap));
1413                         return 0;
1414                 }
1415                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1416                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1417                 {
1418                         ext3_error (sb, "ext3_check_descriptors",
1419                                     "Inode bitmap for group %d"
1420                                     " not in group (block %lu)!",
1421                                     i, (unsigned long)
1422                                         le32_to_cpu(gdp->bg_inode_bitmap));
1423                         return 0;
1424                 }
1425                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1426                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1427                     last_block)
1428                 {
1429                         ext3_error (sb, "ext3_check_descriptors",
1430                                     "Inode table for group %d"
1431                                     " not in group (block %lu)!",
1432                                     i, (unsigned long)
1433                                         le32_to_cpu(gdp->bg_inode_table));
1434                         return 0;
1435                 }
1436         }
1437
1438         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1439         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1440         return 1;
1441 }
1442
1443
1444 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1445  * the superblock) which were deleted from all directories, but held open by
1446  * a process at the time of a crash.  We walk the list and try to delete these
1447  * inodes at recovery time (only with a read-write filesystem).
1448  *
1449  * In order to keep the orphan inode chain consistent during traversal (in
1450  * case of crash during recovery), we link each inode into the superblock
1451  * orphan list_head and handle it the same way as an inode deletion during
1452  * normal operation (which journals the operations for us).
1453  *
1454  * We only do an iget() and an iput() on each inode, which is very safe if we
1455  * accidentally point at an in-use or already deleted inode.  The worst that
1456  * can happen in this case is that we get a "bit already cleared" message from
1457  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1458  * e2fsck was run on this filesystem, and it must have already done the orphan
1459  * inode cleanup for us, so we can safely abort without any further action.
1460  */
1461 static void ext3_orphan_cleanup (struct super_block * sb,
1462                                  struct ext3_super_block * es)
1463 {
1464         unsigned int s_flags = sb->s_flags;
1465         int nr_orphans = 0, nr_truncates = 0;
1466 #ifdef CONFIG_QUOTA
1467         int i;
1468 #endif
1469         if (!es->s_last_orphan) {
1470                 jbd_debug(4, "no orphan inodes to clean up\n");
1471                 return;
1472         }
1473
1474         if (bdev_read_only(sb->s_bdev)) {
1475                 ext3_msg(sb, KERN_ERR, "error: write access "
1476                         "unavailable, skipping orphan cleanup.");
1477                 return;
1478         }
1479
1480         /* Check if feature set allows readwrite operations */
1481         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1482                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1483                          "unknown ROCOMPAT features");
1484                 return;
1485         }
1486
1487         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1488                 if (es->s_last_orphan)
1489                         jbd_debug(1, "Errors on filesystem, "
1490                                   "clearing orphan list.\n");
1491                 es->s_last_orphan = 0;
1492                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1493                 return;
1494         }
1495
1496         if (s_flags & MS_RDONLY) {
1497                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1498                 sb->s_flags &= ~MS_RDONLY;
1499         }
1500 #ifdef CONFIG_QUOTA
1501         /* Needed for iput() to work correctly and not trash data */
1502         sb->s_flags |= MS_ACTIVE;
1503         /* Turn on quotas so that they are updated correctly */
1504         for (i = 0; i < MAXQUOTAS; i++) {
1505                 if (EXT3_SB(sb)->s_qf_names[i]) {
1506                         int ret = ext3_quota_on_mount(sb, i);
1507                         if (ret < 0)
1508                                 ext3_msg(sb, KERN_ERR,
1509                                         "error: cannot turn on journaled "
1510                                         "quota: %d", ret);
1511                 }
1512         }
1513 #endif
1514
1515         while (es->s_last_orphan) {
1516                 struct inode *inode;
1517
1518                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1519                 if (IS_ERR(inode)) {
1520                         es->s_last_orphan = 0;
1521                         break;
1522                 }
1523
1524                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1525                 dquot_initialize(inode);
1526                 if (inode->i_nlink) {
1527                         printk(KERN_DEBUG
1528                                 "%s: truncating inode %lu to %Ld bytes\n",
1529                                 __func__, inode->i_ino, inode->i_size);
1530                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1531                                   inode->i_ino, inode->i_size);
1532                         ext3_truncate(inode);
1533                         nr_truncates++;
1534                 } else {
1535                         printk(KERN_DEBUG
1536                                 "%s: deleting unreferenced inode %lu\n",
1537                                 __func__, inode->i_ino);
1538                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1539                                   inode->i_ino);
1540                         nr_orphans++;
1541                 }
1542                 iput(inode);  /* The delete magic happens here! */
1543         }
1544
1545 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1546
1547         if (nr_orphans)
1548                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1549                        PLURAL(nr_orphans));
1550         if (nr_truncates)
1551                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1552                        PLURAL(nr_truncates));
1553 #ifdef CONFIG_QUOTA
1554         /* Turn quotas off */
1555         for (i = 0; i < MAXQUOTAS; i++) {
1556                 if (sb_dqopt(sb)->files[i])
1557                         dquot_quota_off(sb, i);
1558         }
1559 #endif
1560         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1561 }
1562
1563 /*
1564  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1565  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1566  * We need to be 1 filesystem block less than the 2^32 sector limit.
1567  */
1568 static loff_t ext3_max_size(int bits)
1569 {
1570         loff_t res = EXT3_NDIR_BLOCKS;
1571         int meta_blocks;
1572         loff_t upper_limit;
1573
1574         /* This is calculated to be the largest file size for a
1575          * dense, file such that the total number of
1576          * sectors in the file, including data and all indirect blocks,
1577          * does not exceed 2^32 -1
1578          * __u32 i_blocks representing the total number of
1579          * 512 bytes blocks of the file
1580          */
1581         upper_limit = (1LL << 32) - 1;
1582
1583         /* total blocks in file system block size */
1584         upper_limit >>= (bits - 9);
1585
1586
1587         /* indirect blocks */
1588         meta_blocks = 1;
1589         /* double indirect blocks */
1590         meta_blocks += 1 + (1LL << (bits-2));
1591         /* tripple indirect blocks */
1592         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1593
1594         upper_limit -= meta_blocks;
1595         upper_limit <<= bits;
1596
1597         res += 1LL << (bits-2);
1598         res += 1LL << (2*(bits-2));
1599         res += 1LL << (3*(bits-2));
1600         res <<= bits;
1601         if (res > upper_limit)
1602                 res = upper_limit;
1603
1604         if (res > MAX_LFS_FILESIZE)
1605                 res = MAX_LFS_FILESIZE;
1606
1607         return res;
1608 }
1609
1610 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1611                                     ext3_fsblk_t logic_sb_block,
1612                                     int nr)
1613 {
1614         struct ext3_sb_info *sbi = EXT3_SB(sb);
1615         unsigned long bg, first_meta_bg;
1616         int has_super = 0;
1617
1618         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1619
1620         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1621             nr < first_meta_bg)
1622                 return (logic_sb_block + nr + 1);
1623         bg = sbi->s_desc_per_block * nr;
1624         if (ext3_bg_has_super(sb, bg))
1625                 has_super = 1;
1626         return (has_super + ext3_group_first_block_no(sb, bg));
1627 }
1628
1629
1630 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1631 {
1632         struct buffer_head * bh;
1633         struct ext3_super_block *es = NULL;
1634         struct ext3_sb_info *sbi;
1635         ext3_fsblk_t block;
1636         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1637         ext3_fsblk_t logic_sb_block;
1638         unsigned long offset = 0;
1639         unsigned int journal_inum = 0;
1640         unsigned long journal_devnum = 0;
1641         unsigned long def_mount_opts;
1642         struct inode *root;
1643         int blocksize;
1644         int hblock;
1645         int db_count;
1646         int i;
1647         int needs_recovery;
1648         int ret = -EINVAL;
1649         __le32 features;
1650         int err;
1651
1652         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1653         if (!sbi)
1654                 return -ENOMEM;
1655
1656         sbi->s_blockgroup_lock =
1657                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1658         if (!sbi->s_blockgroup_lock) {
1659                 kfree(sbi);
1660                 return -ENOMEM;
1661         }
1662         sb->s_fs_info = sbi;
1663         sbi->s_mount_opt = 0;
1664         sbi->s_resuid = EXT3_DEF_RESUID;
1665         sbi->s_resgid = EXT3_DEF_RESGID;
1666         sbi->s_sb_block = sb_block;
1667
1668         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1669         if (!blocksize) {
1670                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1671                 goto out_fail;
1672         }
1673
1674         /*
1675          * The ext3 superblock will not be buffer aligned for other than 1kB
1676          * block sizes.  We need to calculate the offset from buffer start.
1677          */
1678         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1679                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1680                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1681         } else {
1682                 logic_sb_block = sb_block;
1683         }
1684
1685         if (!(bh = sb_bread(sb, logic_sb_block))) {
1686                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1687                 goto out_fail;
1688         }
1689         /*
1690          * Note: s_es must be initialized as soon as possible because
1691          *       some ext3 macro-instructions depend on its value
1692          */
1693         es = (struct ext3_super_block *) (bh->b_data + offset);
1694         sbi->s_es = es;
1695         sb->s_magic = le16_to_cpu(es->s_magic);
1696         if (sb->s_magic != EXT3_SUPER_MAGIC)
1697                 goto cantfind_ext3;
1698
1699         /* Set defaults before we parse the mount options */
1700         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1701         if (def_mount_opts & EXT3_DEFM_DEBUG)
1702                 set_opt(sbi->s_mount_opt, DEBUG);
1703         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1704                 set_opt(sbi->s_mount_opt, GRPID);
1705         if (def_mount_opts & EXT3_DEFM_UID16)
1706                 set_opt(sbi->s_mount_opt, NO_UID32);
1707 #ifdef CONFIG_EXT3_FS_XATTR
1708         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1709                 set_opt(sbi->s_mount_opt, XATTR_USER);
1710 #endif
1711 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1712         if (def_mount_opts & EXT3_DEFM_ACL)
1713                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1714 #endif
1715         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1716                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1717         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1718                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1719         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1720                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1721
1722         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1723                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1724         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1725                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1726         else
1727                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1728
1729         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1730         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1731
1732         /* enable barriers by default */
1733         set_opt(sbi->s_mount_opt, BARRIER);
1734         set_opt(sbi->s_mount_opt, RESERVATION);
1735
1736         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1737                             NULL, 0))
1738                 goto failed_mount;
1739
1740         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1741                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1742
1743         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1744             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1745              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1746              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1747                 ext3_msg(sb, KERN_WARNING,
1748                         "warning: feature flags set on rev 0 fs, "
1749                         "running e2fsck is recommended");
1750         /*
1751          * Check feature flags regardless of the revision level, since we
1752          * previously didn't change the revision level when setting the flags,
1753          * so there is a chance incompat flags are set on a rev 0 filesystem.
1754          */
1755         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1756         if (features) {
1757                 ext3_msg(sb, KERN_ERR,
1758                         "error: couldn't mount because of unsupported "
1759                         "optional features (%x)", le32_to_cpu(features));
1760                 goto failed_mount;
1761         }
1762         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1763         if (!(sb->s_flags & MS_RDONLY) && features) {
1764                 ext3_msg(sb, KERN_ERR,
1765                         "error: couldn't mount RDWR because of unsupported "
1766                         "optional features (%x)", le32_to_cpu(features));
1767                 goto failed_mount;
1768         }
1769         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1770
1771         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1772             blocksize > EXT3_MAX_BLOCK_SIZE) {
1773                 ext3_msg(sb, KERN_ERR,
1774                         "error: couldn't mount because of unsupported "
1775                         "filesystem blocksize %d", blocksize);
1776                 goto failed_mount;
1777         }
1778
1779         hblock = bdev_logical_block_size(sb->s_bdev);
1780         if (sb->s_blocksize != blocksize) {
1781                 /*
1782                  * Make sure the blocksize for the filesystem is larger
1783                  * than the hardware sectorsize for the machine.
1784                  */
1785                 if (blocksize < hblock) {
1786                         ext3_msg(sb, KERN_ERR,
1787                                 "error: fsblocksize %d too small for "
1788                                 "hardware sectorsize %d", blocksize, hblock);
1789                         goto failed_mount;
1790                 }
1791
1792                 brelse (bh);
1793                 if (!sb_set_blocksize(sb, blocksize)) {
1794                         ext3_msg(sb, KERN_ERR,
1795                                 "error: bad blocksize %d", blocksize);
1796                         goto out_fail;
1797                 }
1798                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1799                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1800                 bh = sb_bread(sb, logic_sb_block);
1801                 if (!bh) {
1802                         ext3_msg(sb, KERN_ERR,
1803                                "error: can't read superblock on 2nd try");
1804                         goto failed_mount;
1805                 }
1806                 es = (struct ext3_super_block *)(bh->b_data + offset);
1807                 sbi->s_es = es;
1808                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1809                         ext3_msg(sb, KERN_ERR,
1810                                 "error: magic mismatch");
1811                         goto failed_mount;
1812                 }
1813         }
1814
1815         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1816
1817         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1818                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1819                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1820         } else {
1821                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1822                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1823                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1824                     (!is_power_of_2(sbi->s_inode_size)) ||
1825                     (sbi->s_inode_size > blocksize)) {
1826                         ext3_msg(sb, KERN_ERR,
1827                                 "error: unsupported inode size: %d",
1828                                 sbi->s_inode_size);
1829                         goto failed_mount;
1830                 }
1831         }
1832         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1833                                    le32_to_cpu(es->s_log_frag_size);
1834         if (blocksize != sbi->s_frag_size) {
1835                 ext3_msg(sb, KERN_ERR,
1836                        "error: fragsize %lu != blocksize %u (unsupported)",
1837                        sbi->s_frag_size, blocksize);
1838                 goto failed_mount;
1839         }
1840         sbi->s_frags_per_block = 1;
1841         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1842         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1843         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1844         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1845                 goto cantfind_ext3;
1846         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1847         if (sbi->s_inodes_per_block == 0)
1848                 goto cantfind_ext3;
1849         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1850                                         sbi->s_inodes_per_block;
1851         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1852         sbi->s_sbh = bh;
1853         sbi->s_mount_state = le16_to_cpu(es->s_state);
1854         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1855         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1856         for (i=0; i < 4; i++)
1857                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1858         sbi->s_def_hash_version = es->s_def_hash_version;
1859         i = le32_to_cpu(es->s_flags);
1860         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1861                 sbi->s_hash_unsigned = 3;
1862         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1863 #ifdef __CHAR_UNSIGNED__
1864                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1865                 sbi->s_hash_unsigned = 3;
1866 #else
1867                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1868 #endif
1869         }
1870
1871         if (sbi->s_blocks_per_group > blocksize * 8) {
1872                 ext3_msg(sb, KERN_ERR,
1873                         "#blocks per group too big: %lu",
1874                         sbi->s_blocks_per_group);
1875                 goto failed_mount;
1876         }
1877         if (sbi->s_frags_per_group > blocksize * 8) {
1878                 ext3_msg(sb, KERN_ERR,
1879                         "error: #fragments per group too big: %lu",
1880                         sbi->s_frags_per_group);
1881                 goto failed_mount;
1882         }
1883         if (sbi->s_inodes_per_group > blocksize * 8) {
1884                 ext3_msg(sb, KERN_ERR,
1885                         "error: #inodes per group too big: %lu",
1886                         sbi->s_inodes_per_group);
1887                 goto failed_mount;
1888         }
1889
1890         err = generic_check_addressable(sb->s_blocksize_bits,
1891                                         le32_to_cpu(es->s_blocks_count));
1892         if (err) {
1893                 ext3_msg(sb, KERN_ERR,
1894                         "error: filesystem is too large to mount safely");
1895                 if (sizeof(sector_t) < 8)
1896                         ext3_msg(sb, KERN_ERR,
1897                                 "error: CONFIG_LBDAF not enabled");
1898                 ret = err;
1899                 goto failed_mount;
1900         }
1901
1902         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1903                 goto cantfind_ext3;
1904         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1905                                le32_to_cpu(es->s_first_data_block) - 1)
1906                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1907         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1908         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1909                                     GFP_KERNEL);
1910         if (sbi->s_group_desc == NULL) {
1911                 ext3_msg(sb, KERN_ERR,
1912                         "error: not enough memory");
1913                 ret = -ENOMEM;
1914                 goto failed_mount;
1915         }
1916
1917         bgl_lock_init(sbi->s_blockgroup_lock);
1918
1919         for (i = 0; i < db_count; i++) {
1920                 block = descriptor_loc(sb, logic_sb_block, i);
1921                 sbi->s_group_desc[i] = sb_bread(sb, block);
1922                 if (!sbi->s_group_desc[i]) {
1923                         ext3_msg(sb, KERN_ERR,
1924                                 "error: can't read group descriptor %d", i);
1925                         db_count = i;
1926                         goto failed_mount2;
1927                 }
1928         }
1929         if (!ext3_check_descriptors (sb)) {
1930                 ext3_msg(sb, KERN_ERR,
1931                         "error: group descriptors corrupted");
1932                 goto failed_mount2;
1933         }
1934         sbi->s_gdb_count = db_count;
1935         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1936         spin_lock_init(&sbi->s_next_gen_lock);
1937
1938         /* per fileystem reservation list head & lock */
1939         spin_lock_init(&sbi->s_rsv_window_lock);
1940         sbi->s_rsv_window_root = RB_ROOT;
1941         /* Add a single, static dummy reservation to the start of the
1942          * reservation window list --- it gives us a placeholder for
1943          * append-at-start-of-list which makes the allocation logic
1944          * _much_ simpler. */
1945         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1946         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1947         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1948         sbi->s_rsv_window_head.rsv_goal_size = 0;
1949         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1950
1951         /*
1952          * set up enough so that it can read an inode
1953          */
1954         sb->s_op = &ext3_sops;
1955         sb->s_export_op = &ext3_export_ops;
1956         sb->s_xattr = ext3_xattr_handlers;
1957 #ifdef CONFIG_QUOTA
1958         sb->s_qcop = &ext3_qctl_operations;
1959         sb->dq_op = &ext3_quota_operations;
1960 #endif
1961         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1962         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1963         mutex_init(&sbi->s_orphan_lock);
1964         mutex_init(&sbi->s_resize_lock);
1965
1966         sb->s_root = NULL;
1967
1968         needs_recovery = (es->s_last_orphan != 0 ||
1969                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1970                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1971
1972         /*
1973          * The first inode we look at is the journal inode.  Don't try
1974          * root first: it may be modified in the journal!
1975          */
1976         if (!test_opt(sb, NOLOAD) &&
1977             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1978                 if (ext3_load_journal(sb, es, journal_devnum))
1979                         goto failed_mount2;
1980         } else if (journal_inum) {
1981                 if (ext3_create_journal(sb, es, journal_inum))
1982                         goto failed_mount2;
1983         } else {
1984                 if (!silent)
1985                         ext3_msg(sb, KERN_ERR,
1986                                 "error: no journal found. "
1987                                 "mounting ext3 over ext2?");
1988                 goto failed_mount2;
1989         }
1990         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1991                         ext3_count_free_blocks(sb));
1992         if (!err) {
1993                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1994                                 ext3_count_free_inodes(sb));
1995         }
1996         if (!err) {
1997                 err = percpu_counter_init(&sbi->s_dirs_counter,
1998                                 ext3_count_dirs(sb));
1999         }
2000         if (err) {
2001                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2002                 ret = err;
2003                 goto failed_mount3;
2004         }
2005
2006         /* We have now updated the journal if required, so we can
2007          * validate the data journaling mode. */
2008         switch (test_opt(sb, DATA_FLAGS)) {
2009         case 0:
2010                 /* No mode set, assume a default based on the journal
2011                    capabilities: ORDERED_DATA if the journal can
2012                    cope, else JOURNAL_DATA */
2013                 if (journal_check_available_features
2014                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2015                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2016                 else
2017                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2018                 break;
2019
2020         case EXT3_MOUNT_ORDERED_DATA:
2021         case EXT3_MOUNT_WRITEBACK_DATA:
2022                 if (!journal_check_available_features
2023                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2024                         ext3_msg(sb, KERN_ERR,
2025                                 "error: journal does not support "
2026                                 "requested data journaling mode");
2027                         goto failed_mount3;
2028                 }
2029         default:
2030                 break;
2031         }
2032
2033         /*
2034          * The journal_load will have done any necessary log recovery,
2035          * so we can safely mount the rest of the filesystem now.
2036          */
2037
2038         root = ext3_iget(sb, EXT3_ROOT_INO);
2039         if (IS_ERR(root)) {
2040                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2041                 ret = PTR_ERR(root);
2042                 goto failed_mount3;
2043         }
2044         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2045                 iput(root);
2046                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2047                 goto failed_mount3;
2048         }
2049         sb->s_root = d_alloc_root(root);
2050         if (!sb->s_root) {
2051                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2052                 iput(root);
2053                 ret = -ENOMEM;
2054                 goto failed_mount3;
2055         }
2056
2057         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2058
2059         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2060         ext3_orphan_cleanup(sb, es);
2061         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2062         if (needs_recovery) {
2063                 ext3_mark_recovery_complete(sb, es);
2064                 ext3_msg(sb, KERN_INFO, "recovery complete");
2065         }
2066         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2067                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2068                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2069                 "writeback");
2070
2071         return 0;
2072
2073 cantfind_ext3:
2074         if (!silent)
2075                 ext3_msg(sb, KERN_INFO,
2076                         "error: can't find ext3 filesystem on dev %s.",
2077                        sb->s_id);
2078         goto failed_mount;
2079
2080 failed_mount3:
2081         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2082         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2083         percpu_counter_destroy(&sbi->s_dirs_counter);
2084         journal_destroy(sbi->s_journal);
2085 failed_mount2:
2086         for (i = 0; i < db_count; i++)
2087                 brelse(sbi->s_group_desc[i]);
2088         kfree(sbi->s_group_desc);
2089 failed_mount:
2090 #ifdef CONFIG_QUOTA
2091         for (i = 0; i < MAXQUOTAS; i++)
2092                 kfree(sbi->s_qf_names[i]);
2093 #endif
2094         ext3_blkdev_remove(sbi);
2095         brelse(bh);
2096 out_fail:
2097         sb->s_fs_info = NULL;
2098         kfree(sbi->s_blockgroup_lock);
2099         kfree(sbi);
2100         return ret;
2101 }
2102
2103 /*
2104  * Setup any per-fs journal parameters now.  We'll do this both on
2105  * initial mount, once the journal has been initialised but before we've
2106  * done any recovery; and again on any subsequent remount.
2107  */
2108 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2109 {
2110         struct ext3_sb_info *sbi = EXT3_SB(sb);
2111
2112         if (sbi->s_commit_interval)
2113                 journal->j_commit_interval = sbi->s_commit_interval;
2114         /* We could also set up an ext3-specific default for the commit
2115          * interval here, but for now we'll just fall back to the jbd
2116          * default. */
2117
2118         spin_lock(&journal->j_state_lock);
2119         if (test_opt(sb, BARRIER))
2120                 journal->j_flags |= JFS_BARRIER;
2121         else
2122                 journal->j_flags &= ~JFS_BARRIER;
2123         if (test_opt(sb, DATA_ERR_ABORT))
2124                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2125         else
2126                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2127         spin_unlock(&journal->j_state_lock);
2128 }
2129
2130 static journal_t *ext3_get_journal(struct super_block *sb,
2131                                    unsigned int journal_inum)
2132 {
2133         struct inode *journal_inode;
2134         journal_t *journal;
2135
2136         /* First, test for the existence of a valid inode on disk.  Bad
2137          * things happen if we iget() an unused inode, as the subsequent
2138          * iput() will try to delete it. */
2139
2140         journal_inode = ext3_iget(sb, journal_inum);
2141         if (IS_ERR(journal_inode)) {
2142                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2143                 return NULL;
2144         }
2145         if (!journal_inode->i_nlink) {
2146                 make_bad_inode(journal_inode);
2147                 iput(journal_inode);
2148                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2149                 return NULL;
2150         }
2151
2152         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2153                   journal_inode, journal_inode->i_size);
2154         if (!S_ISREG(journal_inode->i_mode)) {
2155                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2156                 iput(journal_inode);
2157                 return NULL;
2158         }
2159
2160         journal = journal_init_inode(journal_inode);
2161         if (!journal) {
2162                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2163                 iput(journal_inode);
2164                 return NULL;
2165         }
2166         journal->j_private = sb;
2167         ext3_init_journal_params(sb, journal);
2168         return journal;
2169 }
2170
2171 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2172                                        dev_t j_dev)
2173 {
2174         struct buffer_head * bh;
2175         journal_t *journal;
2176         ext3_fsblk_t start;
2177         ext3_fsblk_t len;
2178         int hblock, blocksize;
2179         ext3_fsblk_t sb_block;
2180         unsigned long offset;
2181         struct ext3_super_block * es;
2182         struct block_device *bdev;
2183
2184         bdev = ext3_blkdev_get(j_dev, sb);
2185         if (bdev == NULL)
2186                 return NULL;
2187
2188         blocksize = sb->s_blocksize;
2189         hblock = bdev_logical_block_size(bdev);
2190         if (blocksize < hblock) {
2191                 ext3_msg(sb, KERN_ERR,
2192                         "error: blocksize too small for journal device");
2193                 goto out_bdev;
2194         }
2195
2196         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2197         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2198         set_blocksize(bdev, blocksize);
2199         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2200                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2201                         "external journal");
2202                 goto out_bdev;
2203         }
2204
2205         es = (struct ext3_super_block *) (bh->b_data + offset);
2206         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2207             !(le32_to_cpu(es->s_feature_incompat) &
2208               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2209                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2210                         "bad superblock");
2211                 brelse(bh);
2212                 goto out_bdev;
2213         }
2214
2215         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2216                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2217                 brelse(bh);
2218                 goto out_bdev;
2219         }
2220
2221         len = le32_to_cpu(es->s_blocks_count);
2222         start = sb_block + 1;
2223         brelse(bh);     /* we're done with the superblock */
2224
2225         journal = journal_init_dev(bdev, sb->s_bdev,
2226                                         start, len, blocksize);
2227         if (!journal) {
2228                 ext3_msg(sb, KERN_ERR,
2229                         "error: failed to create device journal");
2230                 goto out_bdev;
2231         }
2232         journal->j_private = sb;
2233         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2234                 if (bh_submit_read(journal->j_sb_buffer)) {
2235                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2236                         goto out_journal;
2237                 }
2238         }
2239         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2240                 ext3_msg(sb, KERN_ERR,
2241                         "error: external journal has more than one "
2242                         "user (unsupported) - %d",
2243                         be32_to_cpu(journal->j_superblock->s_nr_users));
2244                 goto out_journal;
2245         }
2246         EXT3_SB(sb)->journal_bdev = bdev;
2247         ext3_init_journal_params(sb, journal);
2248         return journal;
2249 out_journal:
2250         journal_destroy(journal);
2251 out_bdev:
2252         ext3_blkdev_put(bdev);
2253         return NULL;
2254 }
2255
2256 static int ext3_load_journal(struct super_block *sb,
2257                              struct ext3_super_block *es,
2258                              unsigned long journal_devnum)
2259 {
2260         journal_t *journal;
2261         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2262         dev_t journal_dev;
2263         int err = 0;
2264         int really_read_only;
2265
2266         if (journal_devnum &&
2267             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2268                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2269                         "numbers have changed");
2270                 journal_dev = new_decode_dev(journal_devnum);
2271         } else
2272                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2273
2274         really_read_only = bdev_read_only(sb->s_bdev);
2275
2276         /*
2277          * Are we loading a blank journal or performing recovery after a
2278          * crash?  For recovery, we need to check in advance whether we
2279          * can get read-write access to the device.
2280          */
2281
2282         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2283                 if (sb->s_flags & MS_RDONLY) {
2284                         ext3_msg(sb, KERN_INFO,
2285                                 "recovery required on readonly filesystem");
2286                         if (really_read_only) {
2287                                 ext3_msg(sb, KERN_ERR, "error: write access "
2288                                         "unavailable, cannot proceed");
2289                                 return -EROFS;
2290                         }
2291                         ext3_msg(sb, KERN_INFO,
2292                                 "write access will be enabled during recovery");
2293                 }
2294         }
2295
2296         if (journal_inum && journal_dev) {
2297                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2298                        "and inode journals");
2299                 return -EINVAL;
2300         }
2301
2302         if (journal_inum) {
2303                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2304                         return -EINVAL;
2305         } else {
2306                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2307                         return -EINVAL;
2308         }
2309
2310         if (!(journal->j_flags & JFS_BARRIER))
2311                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2312
2313         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2314                 err = journal_update_format(journal);
2315                 if (err)  {
2316                         ext3_msg(sb, KERN_ERR, "error updating journal");
2317                         journal_destroy(journal);
2318                         return err;
2319                 }
2320         }
2321
2322         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2323                 err = journal_wipe(journal, !really_read_only);
2324         if (!err)
2325                 err = journal_load(journal);
2326
2327         if (err) {
2328                 ext3_msg(sb, KERN_ERR, "error loading journal");
2329                 journal_destroy(journal);
2330                 return err;
2331         }
2332
2333         EXT3_SB(sb)->s_journal = journal;
2334         ext3_clear_journal_err(sb, es);
2335
2336         if (!really_read_only && journal_devnum &&
2337             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2338                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2339
2340                 /* Make sure we flush the recovery flag to disk. */
2341                 ext3_commit_super(sb, es, 1);
2342         }
2343
2344         return 0;
2345 }
2346
2347 static int ext3_create_journal(struct super_block *sb,
2348                                struct ext3_super_block *es,
2349                                unsigned int journal_inum)
2350 {
2351         journal_t *journal;
2352         int err;
2353
2354         if (sb->s_flags & MS_RDONLY) {
2355                 ext3_msg(sb, KERN_ERR,
2356                         "error: readonly filesystem when trying to "
2357                         "create journal");
2358                 return -EROFS;
2359         }
2360
2361         journal = ext3_get_journal(sb, journal_inum);
2362         if (!journal)
2363                 return -EINVAL;
2364
2365         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2366                journal_inum);
2367
2368         err = journal_create(journal);
2369         if (err) {
2370                 ext3_msg(sb, KERN_ERR, "error creating journal");
2371                 journal_destroy(journal);
2372                 return -EIO;
2373         }
2374
2375         EXT3_SB(sb)->s_journal = journal;
2376
2377         ext3_update_dynamic_rev(sb);
2378         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2379         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2380
2381         es->s_journal_inum = cpu_to_le32(journal_inum);
2382
2383         /* Make sure we flush the recovery flag to disk. */
2384         ext3_commit_super(sb, es, 1);
2385
2386         return 0;
2387 }
2388
2389 static int ext3_commit_super(struct super_block *sb,
2390                                struct ext3_super_block *es,
2391                                int sync)
2392 {
2393         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2394         int error = 0;
2395
2396         if (!sbh)
2397                 return error;
2398
2399         if (buffer_write_io_error(sbh)) {
2400                 /*
2401                  * Oh, dear.  A previous attempt to write the
2402                  * superblock failed.  This could happen because the
2403                  * USB device was yanked out.  Or it could happen to
2404                  * be a transient write error and maybe the block will
2405                  * be remapped.  Nothing we can do but to retry the
2406                  * write and hope for the best.
2407                  */
2408                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2409                        "superblock detected");
2410                 clear_buffer_write_io_error(sbh);
2411                 set_buffer_uptodate(sbh);
2412         }
2413         /*
2414          * If the file system is mounted read-only, don't update the
2415          * superblock write time.  This avoids updating the superblock
2416          * write time when we are mounting the root file system
2417          * read/only but we need to replay the journal; at that point,
2418          * for people who are east of GMT and who make their clock
2419          * tick in localtime for Windows bug-for-bug compatibility,
2420          * the clock is set in the future, and this will cause e2fsck
2421          * to complain and force a full file system check.
2422          */
2423         if (!(sb->s_flags & MS_RDONLY))
2424                 es->s_wtime = cpu_to_le32(get_seconds());
2425         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2426         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2427         BUFFER_TRACE(sbh, "marking dirty");
2428         mark_buffer_dirty(sbh);
2429         if (sync) {
2430                 error = sync_dirty_buffer(sbh);
2431                 if (buffer_write_io_error(sbh)) {
2432                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2433                                "superblock");
2434                         clear_buffer_write_io_error(sbh);
2435                         set_buffer_uptodate(sbh);
2436                 }
2437         }
2438         return error;
2439 }
2440
2441
2442 /*
2443  * Have we just finished recovery?  If so, and if we are mounting (or
2444  * remounting) the filesystem readonly, then we will end up with a
2445  * consistent fs on disk.  Record that fact.
2446  */
2447 static void ext3_mark_recovery_complete(struct super_block * sb,
2448                                         struct ext3_super_block * es)
2449 {
2450         journal_t *journal = EXT3_SB(sb)->s_journal;
2451
2452         journal_lock_updates(journal);
2453         if (journal_flush(journal) < 0)
2454                 goto out;
2455
2456         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2457             sb->s_flags & MS_RDONLY) {
2458                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2459                 ext3_commit_super(sb, es, 1);
2460         }
2461
2462 out:
2463         journal_unlock_updates(journal);
2464 }
2465
2466 /*
2467  * If we are mounting (or read-write remounting) a filesystem whose journal
2468  * has recorded an error from a previous lifetime, move that error to the
2469  * main filesystem now.
2470  */
2471 static void ext3_clear_journal_err(struct super_block *sb,
2472                                    struct ext3_super_block *es)
2473 {
2474         journal_t *journal;
2475         int j_errno;
2476         const char *errstr;
2477
2478         journal = EXT3_SB(sb)->s_journal;
2479
2480         /*
2481          * Now check for any error status which may have been recorded in the
2482          * journal by a prior ext3_error() or ext3_abort()
2483          */
2484
2485         j_errno = journal_errno(journal);
2486         if (j_errno) {
2487                 char nbuf[16];
2488
2489                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2490                 ext3_warning(sb, __func__, "Filesystem error recorded "
2491                              "from previous mount: %s", errstr);
2492                 ext3_warning(sb, __func__, "Marking fs in need of "
2493                              "filesystem check.");
2494
2495                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2496                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2497                 ext3_commit_super (sb, es, 1);
2498
2499                 journal_clear_err(journal);
2500         }
2501 }
2502
2503 /*
2504  * Force the running and committing transactions to commit,
2505  * and wait on the commit.
2506  */
2507 int ext3_force_commit(struct super_block *sb)
2508 {
2509         journal_t *journal;
2510         int ret;
2511
2512         if (sb->s_flags & MS_RDONLY)
2513                 return 0;
2514
2515         journal = EXT3_SB(sb)->s_journal;
2516         ret = ext3_journal_force_commit(journal);
2517         return ret;
2518 }
2519
2520 static int ext3_sync_fs(struct super_block *sb, int wait)
2521 {
2522         tid_t target;
2523
2524         trace_ext3_sync_fs(sb, wait);
2525         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2526                 if (wait)
2527                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2528         }
2529         return 0;
2530 }
2531
2532 /*
2533  * LVM calls this function before a (read-only) snapshot is created.  This
2534  * gives us a chance to flush the journal completely and mark the fs clean.
2535  */
2536 static int ext3_freeze(struct super_block *sb)
2537 {
2538         int error = 0;
2539         journal_t *journal;
2540
2541         if (!(sb->s_flags & MS_RDONLY)) {
2542                 journal = EXT3_SB(sb)->s_journal;
2543
2544                 /* Now we set up the journal barrier. */
2545                 journal_lock_updates(journal);
2546
2547                 /*
2548                  * We don't want to clear needs_recovery flag when we failed
2549                  * to flush the journal.
2550                  */
2551                 error = journal_flush(journal);
2552                 if (error < 0)
2553                         goto out;
2554
2555                 /* Journal blocked and flushed, clear needs_recovery flag. */
2556                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2557                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2558                 if (error)
2559                         goto out;
2560         }
2561         return 0;
2562
2563 out:
2564         journal_unlock_updates(journal);
2565         return error;
2566 }
2567
2568 /*
2569  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2570  * flag here, even though the filesystem is not technically dirty yet.
2571  */
2572 static int ext3_unfreeze(struct super_block *sb)
2573 {
2574         if (!(sb->s_flags & MS_RDONLY)) {
2575                 lock_super(sb);
2576                 /* Reser the needs_recovery flag before the fs is unlocked. */
2577                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2578                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2579                 unlock_super(sb);
2580                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2581         }
2582         return 0;
2583 }
2584
2585 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2586 {
2587         struct ext3_super_block * es;
2588         struct ext3_sb_info *sbi = EXT3_SB(sb);
2589         ext3_fsblk_t n_blocks_count = 0;
2590         unsigned long old_sb_flags;
2591         struct ext3_mount_options old_opts;
2592         int enable_quota = 0;
2593         int err;
2594 #ifdef CONFIG_QUOTA
2595         int i;
2596 #endif
2597
2598         /* Store the original options */
2599         lock_super(sb);
2600         old_sb_flags = sb->s_flags;
2601         old_opts.s_mount_opt = sbi->s_mount_opt;
2602         old_opts.s_resuid = sbi->s_resuid;
2603         old_opts.s_resgid = sbi->s_resgid;
2604         old_opts.s_commit_interval = sbi->s_commit_interval;
2605 #ifdef CONFIG_QUOTA
2606         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2607         for (i = 0; i < MAXQUOTAS; i++)
2608                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2609 #endif
2610
2611         /*
2612          * Allow the "check" option to be passed as a remount option.
2613          */
2614         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2615                 err = -EINVAL;
2616                 goto restore_opts;
2617         }
2618
2619         if (test_opt(sb, ABORT))
2620                 ext3_abort(sb, __func__, "Abort forced by user");
2621
2622         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2623                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2624
2625         es = sbi->s_es;
2626
2627         ext3_init_journal_params(sb, sbi->s_journal);
2628
2629         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2630                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2631                 if (test_opt(sb, ABORT)) {
2632                         err = -EROFS;
2633                         goto restore_opts;
2634                 }
2635
2636                 if (*flags & MS_RDONLY) {
2637                         err = dquot_suspend(sb, -1);
2638                         if (err < 0)
2639                                 goto restore_opts;
2640
2641                         /*
2642                          * First of all, the unconditional stuff we have to do
2643                          * to disable replay of the journal when we next remount
2644                          */
2645                         sb->s_flags |= MS_RDONLY;
2646
2647                         /*
2648                          * OK, test if we are remounting a valid rw partition
2649                          * readonly, and if so set the rdonly flag and then
2650                          * mark the partition as valid again.
2651                          */
2652                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2653                             (sbi->s_mount_state & EXT3_VALID_FS))
2654                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2655
2656                         ext3_mark_recovery_complete(sb, es);
2657                 } else {
2658                         __le32 ret;
2659                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2660                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2661                                 ext3_msg(sb, KERN_WARNING,
2662                                         "warning: couldn't remount RDWR "
2663                                         "because of unsupported optional "
2664                                         "features (%x)", le32_to_cpu(ret));
2665                                 err = -EROFS;
2666                                 goto restore_opts;
2667                         }
2668
2669                         /*
2670                          * If we have an unprocessed orphan list hanging
2671                          * around from a previously readonly bdev mount,
2672                          * require a full umount & mount for now.
2673                          */
2674                         if (es->s_last_orphan) {
2675                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2676                                        "remount RDWR because of unprocessed "
2677                                        "orphan inode list.  Please "
2678                                        "umount & mount instead.");
2679                                 err = -EINVAL;
2680                                 goto restore_opts;
2681                         }
2682
2683                         /*
2684                          * Mounting a RDONLY partition read-write, so reread
2685                          * and store the current valid flag.  (It may have
2686                          * been changed by e2fsck since we originally mounted
2687                          * the partition.)
2688                          */
2689                         ext3_clear_journal_err(sb, es);
2690                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2691                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2692                                 goto restore_opts;
2693                         if (!ext3_setup_super (sb, es, 0))
2694                                 sb->s_flags &= ~MS_RDONLY;
2695                         enable_quota = 1;
2696                 }
2697         }
2698 #ifdef CONFIG_QUOTA
2699         /* Release old quota file names */
2700         for (i = 0; i < MAXQUOTAS; i++)
2701                 if (old_opts.s_qf_names[i] &&
2702                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2703                         kfree(old_opts.s_qf_names[i]);
2704 #endif
2705         unlock_super(sb);
2706
2707         if (enable_quota)
2708                 dquot_resume(sb, -1);
2709         return 0;
2710 restore_opts:
2711         sb->s_flags = old_sb_flags;
2712         sbi->s_mount_opt = old_opts.s_mount_opt;
2713         sbi->s_resuid = old_opts.s_resuid;
2714         sbi->s_resgid = old_opts.s_resgid;
2715         sbi->s_commit_interval = old_opts.s_commit_interval;
2716 #ifdef CONFIG_QUOTA
2717         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2718         for (i = 0; i < MAXQUOTAS; i++) {
2719                 if (sbi->s_qf_names[i] &&
2720                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2721                         kfree(sbi->s_qf_names[i]);
2722                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2723         }
2724 #endif
2725         unlock_super(sb);
2726         return err;
2727 }
2728
2729 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2730 {
2731         struct super_block *sb = dentry->d_sb;
2732         struct ext3_sb_info *sbi = EXT3_SB(sb);
2733         struct ext3_super_block *es = sbi->s_es;
2734         u64 fsid;
2735
2736         if (test_opt(sb, MINIX_DF)) {
2737                 sbi->s_overhead_last = 0;
2738         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2739                 unsigned long ngroups = sbi->s_groups_count, i;
2740                 ext3_fsblk_t overhead = 0;
2741                 smp_rmb();
2742
2743                 /*
2744                  * Compute the overhead (FS structures).  This is constant
2745                  * for a given filesystem unless the number of block groups
2746                  * changes so we cache the previous value until it does.
2747                  */
2748
2749                 /*
2750                  * All of the blocks before first_data_block are
2751                  * overhead
2752                  */
2753                 overhead = le32_to_cpu(es->s_first_data_block);
2754
2755                 /*
2756                  * Add the overhead attributed to the superblock and
2757                  * block group descriptors.  If the sparse superblocks
2758                  * feature is turned on, then not all groups have this.
2759                  */
2760                 for (i = 0; i < ngroups; i++) {
2761                         overhead += ext3_bg_has_super(sb, i) +
2762                                 ext3_bg_num_gdb(sb, i);
2763                         cond_resched();
2764                 }
2765
2766                 /*
2767                  * Every block group has an inode bitmap, a block
2768                  * bitmap, and an inode table.
2769                  */
2770                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2771                 sbi->s_overhead_last = overhead;
2772                 smp_wmb();
2773                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2774         }
2775
2776         buf->f_type = EXT3_SUPER_MAGIC;
2777         buf->f_bsize = sb->s_blocksize;
2778         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2779         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2780         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2781         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2782                 buf->f_bavail = 0;
2783         buf->f_files = le32_to_cpu(es->s_inodes_count);
2784         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2785         buf->f_namelen = EXT3_NAME_LEN;
2786         fsid = le64_to_cpup((void *)es->s_uuid) ^
2787                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2788         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2789         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2790         return 0;
2791 }
2792
2793 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2794  * is locked for write. Otherwise the are possible deadlocks:
2795  * Process 1                         Process 2
2796  * ext3_create()                     quota_sync()
2797  *   journal_start()                   write_dquot()
2798  *   dquot_initialize()                       down(dqio_mutex)
2799  *     down(dqio_mutex)                    journal_start()
2800  *
2801  */
2802
2803 #ifdef CONFIG_QUOTA
2804
2805 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2806 {
2807         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2808 }
2809
2810 static int ext3_write_dquot(struct dquot *dquot)
2811 {
2812         int ret, err;
2813         handle_t *handle;
2814         struct inode *inode;
2815
2816         inode = dquot_to_inode(dquot);
2817         handle = ext3_journal_start(inode,
2818                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2819         if (IS_ERR(handle))
2820                 return PTR_ERR(handle);
2821         ret = dquot_commit(dquot);
2822         err = ext3_journal_stop(handle);
2823         if (!ret)
2824                 ret = err;
2825         return ret;
2826 }
2827
2828 static int ext3_acquire_dquot(struct dquot *dquot)
2829 {
2830         int ret, err;
2831         handle_t *handle;
2832
2833         handle = ext3_journal_start(dquot_to_inode(dquot),
2834                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2835         if (IS_ERR(handle))
2836                 return PTR_ERR(handle);
2837         ret = dquot_acquire(dquot);
2838         err = ext3_journal_stop(handle);
2839         if (!ret)
2840                 ret = err;
2841         return ret;
2842 }
2843
2844 static int ext3_release_dquot(struct dquot *dquot)
2845 {
2846         int ret, err;
2847         handle_t *handle;
2848
2849         handle = ext3_journal_start(dquot_to_inode(dquot),
2850                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2851         if (IS_ERR(handle)) {
2852                 /* Release dquot anyway to avoid endless cycle in dqput() */
2853                 dquot_release(dquot);
2854                 return PTR_ERR(handle);
2855         }
2856         ret = dquot_release(dquot);
2857         err = ext3_journal_stop(handle);
2858         if (!ret)
2859                 ret = err;
2860         return ret;
2861 }
2862
2863 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2864 {
2865         /* Are we journaling quotas? */
2866         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2867             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2868                 dquot_mark_dquot_dirty(dquot);
2869                 return ext3_write_dquot(dquot);
2870         } else {
2871                 return dquot_mark_dquot_dirty(dquot);
2872         }
2873 }
2874
2875 static int ext3_write_info(struct super_block *sb, int type)
2876 {
2877         int ret, err;
2878         handle_t *handle;
2879
2880         /* Data block + inode block */
2881         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2882         if (IS_ERR(handle))
2883                 return PTR_ERR(handle);
2884         ret = dquot_commit_info(sb, type);
2885         err = ext3_journal_stop(handle);
2886         if (!ret)
2887                 ret = err;
2888         return ret;
2889 }
2890
2891 /*
2892  * Turn on quotas during mount time - we need to find
2893  * the quota file and such...
2894  */
2895 static int ext3_quota_on_mount(struct super_block *sb, int type)
2896 {
2897         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2898                                         EXT3_SB(sb)->s_jquota_fmt, type);
2899 }
2900
2901 /*
2902  * Standard function to be called on quota_on
2903  */
2904 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2905                          struct path *path)
2906 {
2907         int err;
2908
2909         if (!test_opt(sb, QUOTA))
2910                 return -EINVAL;
2911
2912         /* Quotafile not on the same filesystem? */
2913         if (path->dentry->d_sb != sb)
2914                 return -EXDEV;
2915         /* Journaling quota? */
2916         if (EXT3_SB(sb)->s_qf_names[type]) {
2917                 /* Quotafile not of fs root? */
2918                 if (path->dentry->d_parent != sb->s_root)
2919                         ext3_msg(sb, KERN_WARNING,
2920                                 "warning: Quota file not on filesystem root. "
2921                                 "Journaled quota will not work.");
2922         }
2923
2924         /*
2925          * When we journal data on quota file, we have to flush journal to see
2926          * all updates to the file when we bypass pagecache...
2927          */
2928         if (ext3_should_journal_data(path->dentry->d_inode)) {
2929                 /*
2930                  * We don't need to lock updates but journal_flush() could
2931                  * otherwise be livelocked...
2932                  */
2933                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2934                 err = journal_flush(EXT3_SB(sb)->s_journal);
2935                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2936                 if (err)
2937                         return err;
2938         }
2939
2940         return dquot_quota_on(sb, type, format_id, path);
2941 }
2942
2943 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2944  * acquiring the locks... As quota files are never truncated and quota code
2945  * itself serializes the operations (and no one else should touch the files)
2946  * we don't have to be afraid of races */
2947 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2948                                size_t len, loff_t off)
2949 {
2950         struct inode *inode = sb_dqopt(sb)->files[type];
2951         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2952         int err = 0;
2953         int offset = off & (sb->s_blocksize - 1);
2954         int tocopy;
2955         size_t toread;
2956         struct buffer_head *bh;
2957         loff_t i_size = i_size_read(inode);
2958
2959         if (off > i_size)
2960                 return 0;
2961         if (off+len > i_size)
2962                 len = i_size-off;
2963         toread = len;
2964         while (toread > 0) {
2965                 tocopy = sb->s_blocksize - offset < toread ?
2966                                 sb->s_blocksize - offset : toread;
2967                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2968                 if (err)
2969                         return err;
2970                 if (!bh)        /* A hole? */
2971                         memset(data, 0, tocopy);
2972                 else
2973                         memcpy(data, bh->b_data+offset, tocopy);
2974                 brelse(bh);
2975                 offset = 0;
2976                 toread -= tocopy;
2977                 data += tocopy;
2978                 blk++;
2979         }
2980         return len;
2981 }
2982
2983 /* Write to quotafile (we know the transaction is already started and has
2984  * enough credits) */
2985 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2986                                 const char *data, size_t len, loff_t off)
2987 {
2988         struct inode *inode = sb_dqopt(sb)->files[type];
2989         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2990         int err = 0;
2991         int offset = off & (sb->s_blocksize - 1);
2992         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2993         struct buffer_head *bh;
2994         handle_t *handle = journal_current_handle();
2995
2996         if (!handle) {
2997                 ext3_msg(sb, KERN_WARNING,
2998                         "warning: quota write (off=%llu, len=%llu)"
2999                         " cancelled because transaction is not started.",
3000                         (unsigned long long)off, (unsigned long long)len);
3001                 return -EIO;
3002         }
3003
3004         /*
3005          * Since we account only one data block in transaction credits,
3006          * then it is impossible to cross a block boundary.
3007          */
3008         if (sb->s_blocksize - offset < len) {
3009                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3010                         " cancelled because not block aligned",
3011                         (unsigned long long)off, (unsigned long long)len);
3012                 return -EIO;
3013         }
3014         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3015         bh = ext3_bread(handle, inode, blk, 1, &err);
3016         if (!bh)
3017                 goto out;
3018         if (journal_quota) {
3019                 err = ext3_journal_get_write_access(handle, bh);
3020                 if (err) {
3021                         brelse(bh);
3022                         goto out;
3023                 }
3024         }
3025         lock_buffer(bh);
3026         memcpy(bh->b_data+offset, data, len);
3027         flush_dcache_page(bh->b_page);
3028         unlock_buffer(bh);
3029         if (journal_quota)
3030                 err = ext3_journal_dirty_metadata(handle, bh);
3031         else {
3032                 /* Always do at least ordered writes for quotas */
3033                 err = ext3_journal_dirty_data(handle, bh);
3034                 mark_buffer_dirty(bh);
3035         }
3036         brelse(bh);
3037 out:
3038         if (err) {
3039                 mutex_unlock(&inode->i_mutex);
3040                 return err;
3041         }
3042         if (inode->i_size < off + len) {
3043                 i_size_write(inode, off + len);
3044                 EXT3_I(inode)->i_disksize = inode->i_size;
3045         }
3046         inode->i_version++;
3047         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3048         ext3_mark_inode_dirty(handle, inode);
3049         mutex_unlock(&inode->i_mutex);
3050         return len;
3051 }
3052
3053 #endif
3054
3055 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3056         int flags, const char *dev_name, void *data)
3057 {
3058         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3059 }
3060
3061 static struct file_system_type ext3_fs_type = {
3062         .owner          = THIS_MODULE,
3063         .name           = "ext3",
3064         .mount          = ext3_mount,
3065         .kill_sb        = kill_block_super,
3066         .fs_flags       = FS_REQUIRES_DEV,
3067 };
3068
3069 static int __init init_ext3_fs(void)
3070 {
3071         int err = init_ext3_xattr();
3072         if (err)
3073                 return err;
3074         err = init_inodecache();
3075         if (err)
3076                 goto out1;
3077         err = register_filesystem(&ext3_fs_type);
3078         if (err)
3079                 goto out;
3080         return 0;
3081 out:
3082         destroy_inodecache();
3083 out1:
3084         exit_ext3_xattr();
3085         return err;
3086 }
3087
3088 static void __exit exit_ext3_fs(void)
3089 {
3090         unregister_filesystem(&ext3_fs_type);
3091         destroy_inodecache();
3092         exit_ext3_xattr();
3093 }
3094
3095 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3096 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3097 MODULE_LICENSE("GPL");
3098 module_init(init_ext3_fs)
3099 module_exit(exit_ext3_fs)