fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static void ext3_commit_super (struct super_block * sb,
  52                                struct ext3_super_block * es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static void ext3_unlockfs(struct super_block *sb);
  64 static void ext3_write_super (struct super_block * sb);
  65 static void ext3_write_super_lockfs(struct super_block *sb);
  66
  67 /*
  68  * Wrappers for journal_start/end.
  69  *
  70  * The only special thing we need to do here is to make sure that all
  71  * journal_end calls result in the superblock being marked dirty, so
  72  * that sync() will call the filesystem's write_super callback if
  73  * appropriate.
  74  */
  75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  76 {
  77         journal_t *journal;
  78
  79         if (sb->s_flags & MS_RDONLY)
  80                 return ERR_PTR(-EROFS);
  81
  82         /* Special case here: if the journal has aborted behind our
  83          * backs (eg. EIO in the commit thread), then we still need to
  84          * take the FS itself readonly cleanly. */
  85         journal = EXT3_SB(sb)->s_journal;
  86         if (is_journal_aborted(journal)) {
  87                 ext3_abort(sb, __func__,
  88                            "Detected aborted journal");
  89                 return ERR_PTR(-EROFS);
  90         }
  91
  92         return journal_start(journal, nblocks);
  93 }
  94
  95 /*
  96  * The only special thing we need to do here is to make sure that all
  97  * journal_stop calls result in the superblock being marked dirty, so
  98  * that sync() will call the filesystem's write_super callback if
  99  * appropriate.
 100  */
 101 int __ext3_journal_stop(const char *where, handle_t *handle)
 102 {
 103         struct super_block *sb;
 104         int err;
 105         int rc;
 106
 107         sb = handle->h_transaction->t_journal->j_private;
 108         err = handle->h_err;
 109         rc = journal_stop(handle);
 110
 111         if (!err)
 112                 err = rc;
 113         if (err)
 114                 __ext3_std_error(sb, where, err);
 115         return err;
 116 }
 117
 118 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 119                 struct buffer_head *bh, handle_t *handle, int err)
 120 {
 121         char nbuf[16];
 122         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 123
 124         if (bh)
 125                 BUFFER_TRACE(bh, "abort");
 126
 127         if (!handle->h_err)
 128                 handle->h_err = err;
 129
 130         if (is_handle_aborted(handle))
 131                 return;
 132
 133         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 134                caller, errstr, err_fn);
 135
 136         journal_abort_handle(handle);
 137 }
 138
 139 /* Deal with the reporting of failure conditions on a filesystem such as
 140  * inconsistencies detected or read IO failures.
 141  *
 142  * On ext2, we can store the error state of the filesystem in the
 143  * superblock.  That is not possible on ext3, because we may have other
 144  * write ordering constraints on the superblock which prevent us from
 145  * writing it out straight away; and given that the journal is about to
 146  * be aborted, we can't rely on the current, or future, transactions to
 147  * write out the superblock safely.
 148  *
 149  * We'll just use the journal_abort() error code to record an error in
 150  * the journal instead.  On recovery, the journal will compain about
 151  * that error until we've noted it down and cleared it.
 152  */
 153
 154 static void ext3_handle_error(struct super_block *sb)
 155 {
 156         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 157
 158         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 159         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 160
 161         if (sb->s_flags & MS_RDONLY)
 162                 return;
 163
 164         if (!test_opt (sb, ERRORS_CONT)) {
 165                 journal_t *journal = EXT3_SB(sb)->s_journal;
 166
 167                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 168                 if (journal)
 169                         journal_abort(journal, -EIO);
 170         }
 171         if (test_opt (sb, ERRORS_RO)) {
 172                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 173                 sb->s_flags |= MS_RDONLY;
 174         }
 175         ext3_commit_super(sb, es, 1);
 176         if (test_opt(sb, ERRORS_PANIC))
 177                 panic("EXT3-fs (device %s): panic forced after error\n",
 178                         sb->s_id);
 179 }
 180
 181 void ext3_error (struct super_block * sb, const char * function,
 182                  const char * fmt, ...)
 183 {
 184         va_list args;
 185
 186         va_start(args, fmt);
 187         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 188         vprintk(fmt, args);
 189         printk("\n");
 190         va_end(args);
 191
 192         ext3_handle_error(sb);
 193 }
 194
 195 static const char *ext3_decode_error(struct super_block * sb, int errno,
 196                                      char nbuf[16])
 197 {
 198         char *errstr = NULL;
 199
 200         switch (errno) {
 201         case -EIO:
 202                 errstr = "IO failure";
 203                 break;
 204         case -ENOMEM:
 205                 errstr = "Out of memory";
 206                 break;
 207         case -EROFS:
 208                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 209                         errstr = "Journal has aborted";
 210                 else
 211                         errstr = "Readonly filesystem";
 212                 break;
 213         default:
 214                 /* If the caller passed in an extra buffer for unknown
 215                  * errors, textualise them now.  Else we just return
 216                  * NULL. */
 217                 if (nbuf) {
 218                         /* Check for truncated error codes... */
 219                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 220                                 errstr = nbuf;
 221                 }
 222                 break;
 223         }
 224
 225         return errstr;
 226 }
 227
 228 /* __ext3_std_error decodes expected errors from journaling functions
 229  * automatically and invokes the appropriate error response.  */
 230
 231 void __ext3_std_error (struct super_block * sb, const char * function,
 232                        int errno)
 233 {
 234         char nbuf[16];
 235         const char *errstr;
 236
 237         /* Special case: if the error is EROFS, and we're not already
 238          * inside a transaction, then there's really no point in logging
 239          * an error. */
 240         if (errno == -EROFS && journal_current_handle() == NULL &&
 241             (sb->s_flags & MS_RDONLY))
 242                 return;
 243
 244         errstr = ext3_decode_error(sb, errno, nbuf);
 245         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 246                 sb->s_id, function, errstr);
 247
 248         ext3_handle_error(sb);
 249 }
 250
 251 /*
 252  * ext3_abort is a much stronger failure handler than ext3_error.  The
 253  * abort function may be used to deal with unrecoverable failures such
 254  * as journal IO errors or ENOMEM at a critical moment in log management.
 255  *
 256  * We unconditionally force the filesystem into an ABORT|READONLY state,
 257  * unless the error response on the fs has been set to panic in which
 258  * case we take the easy way out and panic immediately.
 259  */
 260
 261 void ext3_abort (struct super_block * sb, const char * function,
 262                  const char * fmt, ...)
 263 {
 264         va_list args;
 265
 266         printk (KERN_CRIT "ext3_abort called.\n");
 267
 268         va_start(args, fmt);
 269         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 270         vprintk(fmt, args);
 271         printk("\n");
 272         va_end(args);
 273
 274         if (test_opt(sb, ERRORS_PANIC))
 275                 panic("EXT3-fs panic from previous error\n");
 276
 277         if (sb->s_flags & MS_RDONLY)
 278                 return;
 279
 280         printk(KERN_CRIT "Remounting filesystem read-only\n");
 281         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 282         sb->s_flags |= MS_RDONLY;
 283         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 284         if (EXT3_SB(sb)->s_journal)
 285                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 286 }
 287
 288 void ext3_warning (struct super_block * sb, const char * function,
 289                    const char * fmt, ...)
 290 {
 291         va_list args;
 292
 293         va_start(args, fmt);
 294         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 295                sb->s_id, function);
 296         vprintk(fmt, args);
 297         printk("\n");
 298         va_end(args);
 299 }
 300
 301 void ext3_update_dynamic_rev(struct super_block *sb)
 302 {
 303         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 304
 305         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 306                 return;
 307
 308         ext3_warning(sb, __func__,
 309                      "updating to rev %d because of new feature flag, "
 310                      "running e2fsck is recommended",
 311                      EXT3_DYNAMIC_REV);
 312
 313         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 314         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 315         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 316         /* leave es->s_feature_*compat flags alone */
 317         /* es->s_uuid will be set by e2fsck if empty */
 318
 319         /*
 320          * The rest of the superblock fields should be zero, and if not it
 321          * means they are likely already in use, so leave them alone.  We
 322          * can leave it up to e2fsck to clean up any inconsistencies there.
 323          */
 324 }
 325
 326 /*
 327  * Open the external journal device
 328  */
 329 static struct block_device *ext3_blkdev_get(dev_t dev)
 330 {
 331         struct block_device *bdev;
 332         char b[BDEVNAME_SIZE];
 333
 334         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 335         if (IS_ERR(bdev))
 336                 goto fail;
 337         return bdev;
 338
 339 fail:
 340         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 341                         __bdevname(dev, b), PTR_ERR(bdev));
 342         return NULL;
 343 }
 344
 345 /*
 346  * Release the journal device
 347  */
 348 static int ext3_blkdev_put(struct block_device *bdev)
 349 {
 350         bd_release(bdev);
 351         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 352 }
 353
 354 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 355 {
 356         struct block_device *bdev;
 357         int ret = -ENODEV;
 358
 359         bdev = sbi->journal_bdev;
 360         if (bdev) {
 361                 ret = ext3_blkdev_put(bdev);
 362                 sbi->journal_bdev = NULL;
 363         }
 364         return ret;
 365 }
 366
 367 static inline struct inode *orphan_list_entry(struct list_head *l)
 368 {
 369         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 370 }
 371
 372 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 373 {
 374         struct list_head *l;
 375
 376         printk(KERN_ERR "sb orphan head is %d\n",
 377                le32_to_cpu(sbi->s_es->s_last_orphan));
 378
 379         printk(KERN_ERR "sb_info orphan list:\n");
 380         list_for_each(l, &sbi->s_orphan) {
 381                 struct inode *inode = orphan_list_entry(l);
 382                 printk(KERN_ERR "  "
 383                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 384                        inode->i_sb->s_id, inode->i_ino, inode,
 385                        inode->i_mode, inode->i_nlink,
 386                        NEXT_ORPHAN(inode));
 387         }
 388 }
 389
 390 static void ext3_put_super (struct super_block * sb)
 391 {
 392         struct ext3_sb_info *sbi = EXT3_SB(sb);
 393         struct ext3_super_block *es = sbi->s_es;
 394         int i, err;
 395
 396         ext3_xattr_put_super(sb);
 397         err = journal_destroy(sbi->s_journal);
 398         sbi->s_journal = NULL;
 399         if (err < 0)
 400                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 401
 402         if (!(sb->s_flags & MS_RDONLY)) {
 403                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 404                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 405                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 406                 mark_buffer_dirty(sbi->s_sbh);
 407                 ext3_commit_super(sb, es, 1);
 408         }
 409
 410         for (i = 0; i < sbi->s_gdb_count; i++)
 411                 brelse(sbi->s_group_desc[i]);
 412         kfree(sbi->s_group_desc);
 413         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 414         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 415         percpu_counter_destroy(&sbi->s_dirs_counter);
 416         brelse(sbi->s_sbh);
 417 #ifdef CONFIG_QUOTA
 418         for (i = 0; i < MAXQUOTAS; i++)
 419                 kfree(sbi->s_qf_names[i]);
 420 #endif
 421
 422         /* Debugging code just in case the in-memory inode orphan list
 423          * isn't empty.  The on-disk one can be non-empty if we've
 424          * detected an error and taken the fs readonly, but the
 425          * in-memory list had better be clean by this point. */
 426         if (!list_empty(&sbi->s_orphan))
 427                 dump_orphan_list(sb, sbi);
 428         J_ASSERT(list_empty(&sbi->s_orphan));
 429
 430         invalidate_bdev(sb->s_bdev);
 431         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 432                 /*
 433                  * Invalidate the journal device's buffers.  We don't want them
 434                  * floating about in memory - the physical journal device may
 435                  * hotswapped, and it breaks the `ro-after' testing code.
 436                  */
 437                 sync_blockdev(sbi->journal_bdev);
 438                 invalidate_bdev(sbi->journal_bdev);
 439                 ext3_blkdev_remove(sbi);
 440         }
 441         sb->s_fs_info = NULL;
 442         kfree(sbi);
 443         return;
 444 }
 445
 446 static struct kmem_cache *ext3_inode_cachep;
 447
 448 /*
 449  * Called inside transaction, so use GFP_NOFS
 450  */
 451 static struct inode *ext3_alloc_inode(struct super_block *sb)
 452 {
 453         struct ext3_inode_info *ei;
 454
 455         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 456         if (!ei)
 457                 return NULL;
 458 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 459         ei->i_acl = EXT3_ACL_NOT_CACHED;
 460         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 461 #endif
 462         ei->i_block_alloc_info = NULL;
 463         ei->vfs_inode.i_version = 1;
 464         return &ei->vfs_inode;
 465 }
 466
 467 static void ext3_destroy_inode(struct inode *inode)
 468 {
 469         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 470                 printk("EXT3 Inode %p: orphan list check failed!\n",
 471                         EXT3_I(inode));
 472                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 473                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 474                                 false);
 475                 dump_stack();
 476         }
 477         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 478 }
 479
 480 static void init_once(void *foo)
 481 {
 482         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 483
 484         INIT_LIST_HEAD(&ei->i_orphan);
 485 #ifdef CONFIG_EXT3_FS_XATTR
 486         init_rwsem(&ei->xattr_sem);
 487 #endif
 488         mutex_init(&ei->truncate_mutex);
 489         inode_init_once(&ei->vfs_inode);
 490 }
 491
 492 static int init_inodecache(void)
 493 {
 494         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 495                                              sizeof(struct ext3_inode_info),
 496                                              0, (SLAB_RECLAIM_ACCOUNT|
 497                                                 SLAB_MEM_SPREAD),
 498                                              init_once);
 499         if (ext3_inode_cachep == NULL)
 500                 return -ENOMEM;
 501         return 0;
 502 }
 503
 504 static void destroy_inodecache(void)
 505 {
 506         kmem_cache_destroy(ext3_inode_cachep);
 507 }
 508
 509 static void ext3_clear_inode(struct inode *inode)
 510 {
 511         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 512 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 513         if (EXT3_I(inode)->i_acl &&
 514                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 515                 posix_acl_release(EXT3_I(inode)->i_acl);
 516                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 517         }
 518         if (EXT3_I(inode)->i_default_acl &&
 519                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 520                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 521                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 522         }
 523 #endif
 524         ext3_discard_reservation(inode);
 525         EXT3_I(inode)->i_block_alloc_info = NULL;
 526         if (unlikely(rsv))
 527                 kfree(rsv);
 528 }
 529
 530 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 531 {
 532 #if defined(CONFIG_QUOTA)
 533         struct ext3_sb_info *sbi = EXT3_SB(sb);
 534
 535         if (sbi->s_jquota_fmt)
 536                 seq_printf(seq, ",jqfmt=%s",
 537                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 538
 539         if (sbi->s_qf_names[USRQUOTA])
 540                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 541
 542         if (sbi->s_qf_names[GRPQUOTA])
 543                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 544
 545         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 546                 seq_puts(seq, ",usrquota");
 547
 548         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 549                 seq_puts(seq, ",grpquota");
 550 #endif
 551 }
 552
 553 /*
 554  * Show an option if
 555  *  - it's set to a non-default value OR
 556  *  - if the per-sb default is different from the global default
 557  */
 558 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 559 {
 560         struct super_block *sb = vfs->mnt_sb;
 561         struct ext3_sb_info *sbi = EXT3_SB(sb);
 562         struct ext3_super_block *es = sbi->s_es;
 563         unsigned long def_mount_opts;
 564
 565         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 566
 567         if (sbi->s_sb_block != 1)
 568                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 569         if (test_opt(sb, MINIX_DF))
 570                 seq_puts(seq, ",minixdf");
 571         if (test_opt(sb, GRPID))
 572                 seq_puts(seq, ",grpid");
 573         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 574                 seq_puts(seq, ",nogrpid");
 575         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 576             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 577                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 578         }
 579         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 580             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 581                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 582         }
 583         if (test_opt(sb, ERRORS_RO)) {
 584                 int def_errors = le16_to_cpu(es->s_errors);
 585
 586                 if (def_errors == EXT3_ERRORS_PANIC ||
 587                     def_errors == EXT3_ERRORS_CONTINUE) {
 588                         seq_puts(seq, ",errors=remount-ro");
 589                 }
 590         }
 591         if (test_opt(sb, ERRORS_CONT))
 592                 seq_puts(seq, ",errors=continue");
 593         if (test_opt(sb, ERRORS_PANIC))
 594                 seq_puts(seq, ",errors=panic");
 595         if (test_opt(sb, NO_UID32))
 596                 seq_puts(seq, ",nouid32");
 597         if (test_opt(sb, DEBUG))
 598                 seq_puts(seq, ",debug");
 599         if (test_opt(sb, OLDALLOC))
 600                 seq_puts(seq, ",oldalloc");
 601 #ifdef CONFIG_EXT3_FS_XATTR
 602         if (test_opt(sb, XATTR_USER))
 603                 seq_puts(seq, ",user_xattr");
 604         if (!test_opt(sb, XATTR_USER) &&
 605             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 606                 seq_puts(seq, ",nouser_xattr");
 607         }
 608 #endif
 609 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 610         if (test_opt(sb, POSIX_ACL))
 611                 seq_puts(seq, ",acl");
 612         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 613                 seq_puts(seq, ",noacl");
 614 #endif
 615         if (!test_opt(sb, RESERVATION))
 616                 seq_puts(seq, ",noreservation");
 617         if (sbi->s_commit_interval) {
 618                 seq_printf(seq, ",commit=%u",
 619                            (unsigned) (sbi->s_commit_interval / HZ));
 620         }
 621         if (test_opt(sb, BARRIER))
 622                 seq_puts(seq, ",barrier=1");
 623         if (test_opt(sb, NOBH))
 624                 seq_puts(seq, ",nobh");
 625
 626         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 627                 seq_puts(seq, ",data=journal");
 628         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 629                 seq_puts(seq, ",data=ordered");
 630         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 631                 seq_puts(seq, ",data=writeback");
 632
 633         if (test_opt(sb, DATA_ERR_ABORT))
 634                 seq_puts(seq, ",data_err=abort");
 635
 636         ext3_show_quota_options(seq, sb);
 637
 638         return 0;
 639 }
 640
 641
 642 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 643                 u64 ino, u32 generation)
 644 {
 645         struct inode *inode;
 646
 647         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 648                 return ERR_PTR(-ESTALE);
 649         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 650                 return ERR_PTR(-ESTALE);
 651
 652         /* iget isn't really right if the inode is currently unallocated!!
 653          *
 654          * ext3_read_inode will return a bad_inode if the inode had been
 655          * deleted, so we should be safe.
 656          *
 657          * Currently we don't know the generation for parent directory, so
 658          * a generation of 0 means "accept any"
 659          */
 660         inode = ext3_iget(sb, ino);
 661         if (IS_ERR(inode))
 662                 return ERR_CAST(inode);
 663         if (generation && inode->i_generation != generation) {
 664                 iput(inode);
 665                 return ERR_PTR(-ESTALE);
 666         }
 667
 668         return inode;
 669 }
 670
 671 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 672                 int fh_len, int fh_type)
 673 {
 674         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 675                                     ext3_nfs_get_inode);
 676 }
 677
 678 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 679                 int fh_len, int fh_type)
 680 {
 681         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 682                                     ext3_nfs_get_inode);
 683 }
 684
 685 #ifdef CONFIG_QUOTA
 686 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 687 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 688
 689 static int ext3_dquot_initialize(struct inode *inode, int type);
 690 static int ext3_dquot_drop(struct inode *inode);
 691 static int ext3_write_dquot(struct dquot *dquot);
 692 static int ext3_acquire_dquot(struct dquot *dquot);
 693 static int ext3_release_dquot(struct dquot *dquot);
 694 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 695 static int ext3_write_info(struct super_block *sb, int type);
 696 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 697                                 char *path, int remount);
 698 static int ext3_quota_on_mount(struct super_block *sb, int type);
 699 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 700                                size_t len, loff_t off);
 701 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 702                                 const char *data, size_t len, loff_t off);
 703
 704 static struct dquot_operations ext3_quota_operations = {
 705         .initialize     = ext3_dquot_initialize,
 706         .drop           = ext3_dquot_drop,
 707         .alloc_space    = dquot_alloc_space,
 708         .alloc_inode    = dquot_alloc_inode,
 709         .free_space     = dquot_free_space,
 710         .free_inode     = dquot_free_inode,
 711         .transfer       = dquot_transfer,
 712         .write_dquot    = ext3_write_dquot,
 713         .acquire_dquot  = ext3_acquire_dquot,
 714         .release_dquot  = ext3_release_dquot,
 715         .mark_dirty     = ext3_mark_dquot_dirty,
 716         .write_info     = ext3_write_info,
 717         .alloc_dquot    = dquot_alloc,
 718         .destroy_dquot  = dquot_destroy,
 719 };
 720
 721 static struct quotactl_ops ext3_qctl_operations = {
 722         .quota_on       = ext3_quota_on,
 723         .quota_off      = vfs_quota_off,
 724         .quota_sync     = vfs_quota_sync,
 725         .get_info       = vfs_get_dqinfo,
 726         .set_info       = vfs_set_dqinfo,
 727         .get_dqblk      = vfs_get_dqblk,
 728         .set_dqblk      = vfs_set_dqblk
 729 };
 730 #endif
 731
 732 static const struct super_operations ext3_sops = {
 733         .alloc_inode    = ext3_alloc_inode,
 734         .destroy_inode  = ext3_destroy_inode,
 735         .write_inode    = ext3_write_inode,
 736         .dirty_inode    = ext3_dirty_inode,
 737         .delete_inode   = ext3_delete_inode,
 738         .put_super      = ext3_put_super,
 739         .write_super    = ext3_write_super,
 740         .sync_fs        = ext3_sync_fs,
 741         .write_super_lockfs = ext3_write_super_lockfs,
 742         .unlockfs       = ext3_unlockfs,
 743         .statfs         = ext3_statfs,
 744         .remount_fs     = ext3_remount,
 745         .clear_inode    = ext3_clear_inode,
 746         .show_options   = ext3_show_options,
 747 #ifdef CONFIG_QUOTA
 748         .quota_read     = ext3_quota_read,
 749         .quota_write    = ext3_quota_write,
 750 #endif
 751 };
 752
 753 static const struct export_operations ext3_export_ops = {
 754         .fh_to_dentry = ext3_fh_to_dentry,
 755         .fh_to_parent = ext3_fh_to_parent,
 756         .get_parent = ext3_get_parent,
 757 };
 758
 759 enum {
 760         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 761         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 762         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 763         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 764         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 765         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 766         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 767         Opt_data_err_abort, Opt_data_err_ignore,
 768         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 769         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 770         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 771         Opt_grpquota
 772 };
 773
 774 static const match_table_t tokens = {
 775         {Opt_bsd_df, "bsddf"},
 776         {Opt_minix_df, "minixdf"},
 777         {Opt_grpid, "grpid"},
 778         {Opt_grpid, "bsdgroups"},
 779         {Opt_nogrpid, "nogrpid"},
 780         {Opt_nogrpid, "sysvgroups"},
 781         {Opt_resgid, "resgid=%u"},
 782         {Opt_resuid, "resuid=%u"},
 783         {Opt_sb, "sb=%u"},
 784         {Opt_err_cont, "errors=continue"},
 785         {Opt_err_panic, "errors=panic"},
 786         {Opt_err_ro, "errors=remount-ro"},
 787         {Opt_nouid32, "nouid32"},
 788         {Opt_nocheck, "nocheck"},
 789         {Opt_nocheck, "check=none"},
 790         {Opt_debug, "debug"},
 791         {Opt_oldalloc, "oldalloc"},
 792         {Opt_orlov, "orlov"},
 793         {Opt_user_xattr, "user_xattr"},
 794         {Opt_nouser_xattr, "nouser_xattr"},
 795         {Opt_acl, "acl"},
 796         {Opt_noacl, "noacl"},
 797         {Opt_reservation, "reservation"},
 798         {Opt_noreservation, "noreservation"},
 799         {Opt_noload, "noload"},
 800         {Opt_nobh, "nobh"},
 801         {Opt_bh, "bh"},
 802         {Opt_commit, "commit=%u"},
 803         {Opt_journal_update, "journal=update"},
 804         {Opt_journal_inum, "journal=%u"},
 805         {Opt_journal_dev, "journal_dev=%u"},
 806         {Opt_abort, "abort"},
 807         {Opt_data_journal, "data=journal"},
 808         {Opt_data_ordered, "data=ordered"},
 809         {Opt_data_writeback, "data=writeback"},
 810         {Opt_data_err_abort, "data_err=abort"},
 811         {Opt_data_err_ignore, "data_err=ignore"},
 812         {Opt_offusrjquota, "usrjquota="},
 813         {Opt_usrjquota, "usrjquota=%s"},
 814         {Opt_offgrpjquota, "grpjquota="},
 815         {Opt_grpjquota, "grpjquota=%s"},
 816         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 817         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 818         {Opt_grpquota, "grpquota"},
 819         {Opt_noquota, "noquota"},
 820         {Opt_quota, "quota"},
 821         {Opt_usrquota, "usrquota"},
 822         {Opt_barrier, "barrier=%u"},
 823         {Opt_resize, "resize"},
 824         {Opt_err, NULL},
 825 };
 826
 827 static ext3_fsblk_t get_sb_block(void **data)
 828 {
 829         ext3_fsblk_t    sb_block;
 830         char            *options = (char *) *data;
 831
 832         if (!options || strncmp(options, "sb=", 3) != 0)
 833                 return 1;       /* Default location */
 834         options += 3;
 835         /*todo: use simple_strtoll with >32bit ext3 */
 836         sb_block = simple_strtoul(options, &options, 0);
 837         if (*options && *options != ',') {
 838                 printk("EXT3-fs: Invalid sb specification: %s\n",
 839                        (char *) *data);
 840                 return 1;
 841         }
 842         if (*options == ',')
 843                 options++;
 844         *data = (void *) options;
 845         return sb_block;
 846 }
 847
 848 static int parse_options (char *options, struct super_block *sb,
 849                           unsigned int *inum, unsigned long *journal_devnum,
 850                           ext3_fsblk_t *n_blocks_count, int is_remount)
 851 {
 852         struct ext3_sb_info *sbi = EXT3_SB(sb);
 853         char * p;
 854         substring_t args[MAX_OPT_ARGS];
 855         int data_opt = 0;
 856         int option;
 857 #ifdef CONFIG_QUOTA
 858         int qtype, qfmt;
 859         char *qname;
 860 #endif
 861
 862         if (!options)
 863                 return 1;
 864
 865         while ((p = strsep (&options, ",")) != NULL) {
 866                 int token;
 867                 if (!*p)
 868                         continue;
 869
 870                 token = match_token(p, tokens, args);
 871                 switch (token) {
 872                 case Opt_bsd_df:
 873                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 874                         break;
 875                 case Opt_minix_df:
 876                         set_opt (sbi->s_mount_opt, MINIX_DF);
 877                         break;
 878                 case Opt_grpid:
 879                         set_opt (sbi->s_mount_opt, GRPID);
 880                         break;
 881                 case Opt_nogrpid:
 882                         clear_opt (sbi->s_mount_opt, GRPID);
 883                         break;
 884                 case Opt_resuid:
 885                         if (match_int(&args[0], &option))
 886                                 return 0;
 887                         sbi->s_resuid = option;
 888                         break;
 889                 case Opt_resgid:
 890                         if (match_int(&args[0], &option))
 891                                 return 0;
 892                         sbi->s_resgid = option;
 893                         break;
 894                 case Opt_sb:
 895                         /* handled by get_sb_block() instead of here */
 896                         /* *sb_block = match_int(&args[0]); */
 897                         break;
 898                 case Opt_err_panic:
 899                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 900                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 901                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 902                         break;
 903                 case Opt_err_ro:
 904                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 905                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 906                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 907                         break;
 908                 case Opt_err_cont:
 909                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 910                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 911                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 912                         break;
 913                 case Opt_nouid32:
 914                         set_opt (sbi->s_mount_opt, NO_UID32);
 915                         break;
 916                 case Opt_nocheck:
 917                         clear_opt (sbi->s_mount_opt, CHECK);
 918                         break;
 919                 case Opt_debug:
 920                         set_opt (sbi->s_mount_opt, DEBUG);
 921                         break;
 922                 case Opt_oldalloc:
 923                         set_opt (sbi->s_mount_opt, OLDALLOC);
 924                         break;
 925                 case Opt_orlov:
 926                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 927                         break;
 928 #ifdef CONFIG_EXT3_FS_XATTR
 929                 case Opt_user_xattr:
 930                         set_opt (sbi->s_mount_opt, XATTR_USER);
 931                         break;
 932                 case Opt_nouser_xattr:
 933                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 934                         break;
 935 #else
 936                 case Opt_user_xattr:
 937                 case Opt_nouser_xattr:
 938                         printk("EXT3 (no)user_xattr options not supported\n");
 939                         break;
 940 #endif
 941 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 942                 case Opt_acl:
 943                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 944                         break;
 945                 case Opt_noacl:
 946                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 947                         break;
 948 #else
 949                 case Opt_acl:
 950                 case Opt_noacl:
 951                         printk("EXT3 (no)acl options not supported\n");
 952                         break;
 953 #endif
 954                 case Opt_reservation:
 955                         set_opt(sbi->s_mount_opt, RESERVATION);
 956                         break;
 957                 case Opt_noreservation:
 958                         clear_opt(sbi->s_mount_opt, RESERVATION);
 959                         break;
 960                 case Opt_journal_update:
 961                         /* @@@ FIXME */
 962                         /* Eventually we will want to be able to create
 963                            a journal file here.  For now, only allow the
 964                            user to specify an existing inode to be the
 965                            journal file. */
 966                         if (is_remount) {
 967                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 968                                        "journal on remount\n");
 969                                 return 0;
 970                         }
 971                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 972                         break;
 973                 case Opt_journal_inum:
 974                         if (is_remount) {
 975                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 976                                        "journal on remount\n");
 977                                 return 0;
 978                         }
 979                         if (match_int(&args[0], &option))
 980                                 return 0;
 981                         *inum = option;
 982                         break;
 983                 case Opt_journal_dev:
 984                         if (is_remount) {
 985                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 986                                        "journal on remount\n");
 987                                 return 0;
 988                         }
 989                         if (match_int(&args[0], &option))
 990                                 return 0;
 991                         *journal_devnum = option;
 992                         break;
 993                 case Opt_noload:
 994                         set_opt (sbi->s_mount_opt, NOLOAD);
 995                         break;
 996                 case Opt_commit:
 997                         if (match_int(&args[0], &option))
 998                                 return 0;
 999                         if (option < 0)
1000                                 return 0;
1001                         if (option == 0)
1002                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1003                         sbi->s_commit_interval = HZ * option;
1004                         break;
1005                 case Opt_data_journal:
1006                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1007                         goto datacheck;
1008                 case Opt_data_ordered:
1009                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1010                         goto datacheck;
1011                 case Opt_data_writeback:
1012                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1013                 datacheck:
1014                         if (is_remount) {
1015                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1016                                                 != data_opt) {
1017                                         printk(KERN_ERR
1018                                                 "EXT3-fs: cannot change data "
1019                                                 "mode on remount\n");
1020                                         return 0;
1021                                 }
1022                         } else {
1023                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1024                                 sbi->s_mount_opt |= data_opt;
1025                         }
1026                         break;
1027                 case Opt_data_err_abort:
1028                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1029                         break;
1030                 case Opt_data_err_ignore:
1031                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1032                         break;
1033 #ifdef CONFIG_QUOTA
1034                 case Opt_usrjquota:
1035                         qtype = USRQUOTA;
1036                         goto set_qf_name;
1037                 case Opt_grpjquota:
1038                         qtype = GRPQUOTA;
1039 set_qf_name:
1040                         if (sb_any_quota_loaded(sb) &&
1041                             !sbi->s_qf_names[qtype]) {
1042                                 printk(KERN_ERR
1043                                         "EXT3-fs: Cannot change journaled "
1044                                         "quota options when quota turned on.\n");
1045                                 return 0;
1046                         }
1047                         qname = match_strdup(&args[0]);
1048                         if (!qname) {
1049                                 printk(KERN_ERR
1050                                         "EXT3-fs: not enough memory for "
1051                                         "storing quotafile name.\n");
1052                                 return 0;
1053                         }
1054                         if (sbi->s_qf_names[qtype] &&
1055                             strcmp(sbi->s_qf_names[qtype], qname)) {
1056                                 printk(KERN_ERR
1057                                         "EXT3-fs: %s quota file already "
1058                                         "specified.\n", QTYPE2NAME(qtype));
1059                                 kfree(qname);
1060                                 return 0;
1061                         }
1062                         sbi->s_qf_names[qtype] = qname;
1063                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1064                                 printk(KERN_ERR
1065                                         "EXT3-fs: quotafile must be on "
1066                                         "filesystem root.\n");
1067                                 kfree(sbi->s_qf_names[qtype]);
1068                                 sbi->s_qf_names[qtype] = NULL;
1069                                 return 0;
1070                         }
1071                         set_opt(sbi->s_mount_opt, QUOTA);
1072                         break;
1073                 case Opt_offusrjquota:
1074                         qtype = USRQUOTA;
1075                         goto clear_qf_name;
1076                 case Opt_offgrpjquota:
1077                         qtype = GRPQUOTA;
1078 clear_qf_name:
1079                         if (sb_any_quota_loaded(sb) &&
1080                             sbi->s_qf_names[qtype]) {
1081                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1082                                         "journaled quota options when "
1083                                         "quota turned on.\n");
1084                                 return 0;
1085                         }
1086                         /*
1087                          * The space will be released later when all options
1088                          * are confirmed to be correct
1089                          */
1090                         sbi->s_qf_names[qtype] = NULL;
1091                         break;
1092                 case Opt_jqfmt_vfsold:
1093                         qfmt = QFMT_VFS_OLD;
1094                         goto set_qf_format;
1095                 case Opt_jqfmt_vfsv0:
1096                         qfmt = QFMT_VFS_V0;
1097 set_qf_format:
1098                         if (sb_any_quota_loaded(sb) &&
1099                             sbi->s_jquota_fmt != qfmt) {
1100                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1101                                         "journaled quota options when "
1102                                         "quota turned on.\n");
1103                                 return 0;
1104                         }
1105                         sbi->s_jquota_fmt = qfmt;
1106                         break;
1107                 case Opt_quota:
1108                 case Opt_usrquota:
1109                         set_opt(sbi->s_mount_opt, QUOTA);
1110                         set_opt(sbi->s_mount_opt, USRQUOTA);
1111                         break;
1112                 case Opt_grpquota:
1113                         set_opt(sbi->s_mount_opt, QUOTA);
1114                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1115                         break;
1116                 case Opt_noquota:
1117                         if (sb_any_quota_loaded(sb)) {
1118                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1119                                         "options when quota turned on.\n");
1120                                 return 0;
1121                         }
1122                         clear_opt(sbi->s_mount_opt, QUOTA);
1123                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1124                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1125                         break;
1126 #else
1127                 case Opt_quota:
1128                 case Opt_usrquota:
1129                 case Opt_grpquota:
1130                         printk(KERN_ERR
1131                                 "EXT3-fs: quota options not supported.\n");
1132                         break;
1133                 case Opt_usrjquota:
1134                 case Opt_grpjquota:
1135                 case Opt_offusrjquota:
1136                 case Opt_offgrpjquota:
1137                 case Opt_jqfmt_vfsold:
1138                 case Opt_jqfmt_vfsv0:
1139                         printk(KERN_ERR
1140                                 "EXT3-fs: journaled quota options not "
1141                                 "supported.\n");
1142                         break;
1143                 case Opt_noquota:
1144                         break;
1145 #endif
1146                 case Opt_abort:
1147                         set_opt(sbi->s_mount_opt, ABORT);
1148                         break;
1149                 case Opt_barrier:
1150                         if (match_int(&args[0], &option))
1151                                 return 0;
1152                         if (option)
1153                                 set_opt(sbi->s_mount_opt, BARRIER);
1154                         else
1155                                 clear_opt(sbi->s_mount_opt, BARRIER);
1156                         break;
1157                 case Opt_ignore:
1158                         break;
1159                 case Opt_resize:
1160                         if (!is_remount) {
1161                                 printk("EXT3-fs: resize option only available "
1162                                         "for remount\n");
1163                                 return 0;
1164                         }
1165                         if (match_int(&args[0], &option) != 0)
1166                                 return 0;
1167                         *n_blocks_count = option;
1168                         break;
1169                 case Opt_nobh:
1170                         set_opt(sbi->s_mount_opt, NOBH);
1171                         break;
1172                 case Opt_bh:
1173                         clear_opt(sbi->s_mount_opt, NOBH);
1174                         break;
1175                 default:
1176                         printk (KERN_ERR
1177                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1178                                 "or missing value\n", p);
1179                         return 0;
1180                 }
1181         }
1182 #ifdef CONFIG_QUOTA
1183         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1184                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1185                      sbi->s_qf_names[USRQUOTA])
1186                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1187
1188                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1189                      sbi->s_qf_names[GRPQUOTA])
1190                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1191
1192                 if ((sbi->s_qf_names[USRQUOTA] &&
1193                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1194                     (sbi->s_qf_names[GRPQUOTA] &&
1195                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1196                         printk(KERN_ERR "EXT3-fs: old and new quota "
1197                                         "format mixing.\n");
1198                         return 0;
1199                 }
1200
1201                 if (!sbi->s_jquota_fmt) {
1202                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1203                                         "not specified.\n");
1204                         return 0;
1205                 }
1206         } else {
1207                 if (sbi->s_jquota_fmt) {
1208                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1209                                         "specified with no journaling "
1210                                         "enabled.\n");
1211                         return 0;
1212                 }
1213         }
1214 #endif
1215         return 1;
1216 }
1217
1218 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1219                             int read_only)
1220 {
1221         struct ext3_sb_info *sbi = EXT3_SB(sb);
1222         int res = 0;
1223
1224         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1225                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1226                         "forcing read-only mode\n");
1227                 res = MS_RDONLY;
1228         }
1229         if (read_only)
1230                 return res;
1231         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1232                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1233                         "running e2fsck is recommended\n");
1234         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1235                 printk (KERN_WARNING
1236                         "EXT3-fs warning: mounting fs with errors, "
1237                         "running e2fsck is recommended\n");
1238         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1239                  le16_to_cpu(es->s_mnt_count) >=
1240                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1241                 printk (KERN_WARNING
1242                         "EXT3-fs warning: maximal mount count reached, "
1243                         "running e2fsck is recommended\n");
1244         else if (le32_to_cpu(es->s_checkinterval) &&
1245                 (le32_to_cpu(es->s_lastcheck) +
1246                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1247                 printk (KERN_WARNING
1248                         "EXT3-fs warning: checktime reached, "
1249                         "running e2fsck is recommended\n");
1250 #if 0
1251                 /* @@@ We _will_ want to clear the valid bit if we find
1252                    inconsistencies, to force a fsck at reboot.  But for
1253                    a plain journaled filesystem we can keep it set as
1254                    valid forever! :) */
1255         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1256 #endif
1257         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1258                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1259         le16_add_cpu(&es->s_mnt_count, 1);
1260         es->s_mtime = cpu_to_le32(get_seconds());
1261         ext3_update_dynamic_rev(sb);
1262         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1263
1264         ext3_commit_super(sb, es, 1);
1265         if (test_opt(sb, DEBUG))
1266                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1267                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1268                         sb->s_blocksize,
1269                         sbi->s_groups_count,
1270                         EXT3_BLOCKS_PER_GROUP(sb),
1271                         EXT3_INODES_PER_GROUP(sb),
1272                         sbi->s_mount_opt);
1273
1274         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1275         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1276                 char b[BDEVNAME_SIZE];
1277
1278                 printk("external journal on %s\n",
1279                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1280         } else {
1281                 printk("internal journal\n");
1282         }
1283         return res;
1284 }
1285
1286 /* Called at mount-time, super-block is locked */
1287 static int ext3_check_descriptors(struct super_block *sb)
1288 {
1289         struct ext3_sb_info *sbi = EXT3_SB(sb);
1290         int i;
1291
1292         ext3_debug ("Checking group descriptors");
1293
1294         for (i = 0; i < sbi->s_groups_count; i++) {
1295                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1296                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1297                 ext3_fsblk_t last_block;
1298
1299                 if (i == sbi->s_groups_count - 1)
1300                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1301                 else
1302                         last_block = first_block +
1303                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1304
1305                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1306                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1307                 {
1308                         ext3_error (sb, "ext3_check_descriptors",
1309                                     "Block bitmap for group %d"
1310                                     " not in group (block %lu)!",
1311                                     i, (unsigned long)
1312                                         le32_to_cpu(gdp->bg_block_bitmap));
1313                         return 0;
1314                 }
1315                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1316                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1317                 {
1318                         ext3_error (sb, "ext3_check_descriptors",
1319                                     "Inode bitmap for group %d"
1320                                     " not in group (block %lu)!",
1321                                     i, (unsigned long)
1322                                         le32_to_cpu(gdp->bg_inode_bitmap));
1323                         return 0;
1324                 }
1325                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1326                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1327                     last_block)
1328                 {
1329                         ext3_error (sb, "ext3_check_descriptors",
1330                                     "Inode table for group %d"
1331                                     " not in group (block %lu)!",
1332                                     i, (unsigned long)
1333                                         le32_to_cpu(gdp->bg_inode_table));
1334                         return 0;
1335                 }
1336         }
1337
1338         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1339         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1340         return 1;
1341 }
1342
1343
1344 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1345  * the superblock) which were deleted from all directories, but held open by
1346  * a process at the time of a crash.  We walk the list and try to delete these
1347  * inodes at recovery time (only with a read-write filesystem).
1348  *
1349  * In order to keep the orphan inode chain consistent during traversal (in
1350  * case of crash during recovery), we link each inode into the superblock
1351  * orphan list_head and handle it the same way as an inode deletion during
1352  * normal operation (which journals the operations for us).
1353  *
1354  * We only do an iget() and an iput() on each inode, which is very safe if we
1355  * accidentally point at an in-use or already deleted inode.  The worst that
1356  * can happen in this case is that we get a "bit already cleared" message from
1357  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1358  * e2fsck was run on this filesystem, and it must have already done the orphan
1359  * inode cleanup for us, so we can safely abort without any further action.
1360  */
1361 static void ext3_orphan_cleanup (struct super_block * sb,
1362                                  struct ext3_super_block * es)
1363 {
1364         unsigned int s_flags = sb->s_flags;
1365         int nr_orphans = 0, nr_truncates = 0;
1366 #ifdef CONFIG_QUOTA
1367         int i;
1368 #endif
1369         if (!es->s_last_orphan) {
1370                 jbd_debug(4, "no orphan inodes to clean up\n");
1371                 return;
1372         }
1373
1374         if (bdev_read_only(sb->s_bdev)) {
1375                 printk(KERN_ERR "EXT3-fs: write access "
1376                         "unavailable, skipping orphan cleanup.\n");
1377                 return;
1378         }
1379
1380         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1381                 if (es->s_last_orphan)
1382                         jbd_debug(1, "Errors on filesystem, "
1383                                   "clearing orphan list.\n");
1384                 es->s_last_orphan = 0;
1385                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1386                 return;
1387         }
1388
1389         if (s_flags & MS_RDONLY) {
1390                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1391                        sb->s_id);
1392                 sb->s_flags &= ~MS_RDONLY;
1393         }
1394 #ifdef CONFIG_QUOTA
1395         /* Needed for iput() to work correctly and not trash data */
1396         sb->s_flags |= MS_ACTIVE;
1397         /* Turn on quotas so that they are updated correctly */
1398         for (i = 0; i < MAXQUOTAS; i++) {
1399                 if (EXT3_SB(sb)->s_qf_names[i]) {
1400                         int ret = ext3_quota_on_mount(sb, i);
1401                         if (ret < 0)
1402                                 printk(KERN_ERR
1403                                         "EXT3-fs: Cannot turn on journaled "
1404                                         "quota: error %d\n", ret);
1405                 }
1406         }
1407 #endif
1408
1409         while (es->s_last_orphan) {
1410                 struct inode *inode;
1411
1412                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1413                 if (IS_ERR(inode)) {
1414                         es->s_last_orphan = 0;
1415                         break;
1416                 }
1417
1418                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1419                 DQUOT_INIT(inode);
1420                 if (inode->i_nlink) {
1421                         printk(KERN_DEBUG
1422                                 "%s: truncating inode %lu to %Ld bytes\n",
1423                                 __func__, inode->i_ino, inode->i_size);
1424                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1425                                   inode->i_ino, inode->i_size);
1426                         ext3_truncate(inode);
1427                         nr_truncates++;
1428                 } else {
1429                         printk(KERN_DEBUG
1430                                 "%s: deleting unreferenced inode %lu\n",
1431                                 __func__, inode->i_ino);
1432                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1433                                   inode->i_ino);
1434                         nr_orphans++;
1435                 }
1436                 iput(inode);  /* The delete magic happens here! */
1437         }
1438
1439 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1440
1441         if (nr_orphans)
1442                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1443                        sb->s_id, PLURAL(nr_orphans));
1444         if (nr_truncates)
1445                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1446                        sb->s_id, PLURAL(nr_truncates));
1447 #ifdef CONFIG_QUOTA
1448         /* Turn quotas off */
1449         for (i = 0; i < MAXQUOTAS; i++) {
1450                 if (sb_dqopt(sb)->files[i])
1451                         vfs_quota_off(sb, i, 0);
1452         }
1453 #endif
1454         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1455 }
1456
1457 /*
1458  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1459  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1460  * We need to be 1 filesystem block less than the 2^32 sector limit.
1461  */
1462 static loff_t ext3_max_size(int bits)
1463 {
1464         loff_t res = EXT3_NDIR_BLOCKS;
1465         int meta_blocks;
1466         loff_t upper_limit;
1467
1468         /* This is calculated to be the largest file size for a
1469          * dense, file such that the total number of
1470          * sectors in the file, including data and all indirect blocks,
1471          * does not exceed 2^32 -1
1472          * __u32 i_blocks representing the total number of
1473          * 512 bytes blocks of the file
1474          */
1475         upper_limit = (1LL << 32) - 1;
1476
1477         /* total blocks in file system block size */
1478         upper_limit >>= (bits - 9);
1479
1480
1481         /* indirect blocks */
1482         meta_blocks = 1;
1483         /* double indirect blocks */
1484         meta_blocks += 1 + (1LL << (bits-2));
1485         /* tripple indirect blocks */
1486         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1487
1488         upper_limit -= meta_blocks;
1489         upper_limit <<= bits;
1490
1491         res += 1LL << (bits-2);
1492         res += 1LL << (2*(bits-2));
1493         res += 1LL << (3*(bits-2));
1494         res <<= bits;
1495         if (res > upper_limit)
1496                 res = upper_limit;
1497
1498         if (res > MAX_LFS_FILESIZE)
1499                 res = MAX_LFS_FILESIZE;
1500
1501         return res;
1502 }
1503
1504 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1505                                     ext3_fsblk_t logic_sb_block,
1506                                     int nr)
1507 {
1508         struct ext3_sb_info *sbi = EXT3_SB(sb);
1509         unsigned long bg, first_meta_bg;
1510         int has_super = 0;
1511
1512         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1513
1514         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1515             nr < first_meta_bg)
1516                 return (logic_sb_block + nr + 1);
1517         bg = sbi->s_desc_per_block * nr;
1518         if (ext3_bg_has_super(sb, bg))
1519                 has_super = 1;
1520         return (has_super + ext3_group_first_block_no(sb, bg));
1521 }
1522
1523
1524 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1525 {
1526         struct buffer_head * bh;
1527         struct ext3_super_block *es = NULL;
1528         struct ext3_sb_info *sbi;
1529         ext3_fsblk_t block;
1530         ext3_fsblk_t sb_block = get_sb_block(&data);
1531         ext3_fsblk_t logic_sb_block;
1532         unsigned long offset = 0;
1533         unsigned int journal_inum = 0;
1534         unsigned long journal_devnum = 0;
1535         unsigned long def_mount_opts;
1536         struct inode *root;
1537         int blocksize;
1538         int hblock;
1539         int db_count;
1540         int i;
1541         int needs_recovery;
1542         int ret = -EINVAL;
1543         __le32 features;
1544         int err;
1545
1546         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1547         if (!sbi)
1548                 return -ENOMEM;
1549         sb->s_fs_info = sbi;
1550         sbi->s_mount_opt = 0;
1551         sbi->s_resuid = EXT3_DEF_RESUID;
1552         sbi->s_resgid = EXT3_DEF_RESGID;
1553         sbi->s_sb_block = sb_block;
1554
1555         unlock_kernel();
1556
1557         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1558         if (!blocksize) {
1559                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1560                 goto out_fail;
1561         }
1562
1563         /*
1564          * The ext3 superblock will not be buffer aligned for other than 1kB
1565          * block sizes.  We need to calculate the offset from buffer start.
1566          */
1567         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1568                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1569                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1570         } else {
1571                 logic_sb_block = sb_block;
1572         }
1573
1574         if (!(bh = sb_bread(sb, logic_sb_block))) {
1575                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1576                 goto out_fail;
1577         }
1578         /*
1579          * Note: s_es must be initialized as soon as possible because
1580          *       some ext3 macro-instructions depend on its value
1581          */
1582         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1583         sbi->s_es = es;
1584         sb->s_magic = le16_to_cpu(es->s_magic);
1585         if (sb->s_magic != EXT3_SUPER_MAGIC)
1586                 goto cantfind_ext3;
1587
1588         /* Set defaults before we parse the mount options */
1589         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1590         if (def_mount_opts & EXT3_DEFM_DEBUG)
1591                 set_opt(sbi->s_mount_opt, DEBUG);
1592         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1593                 set_opt(sbi->s_mount_opt, GRPID);
1594         if (def_mount_opts & EXT3_DEFM_UID16)
1595                 set_opt(sbi->s_mount_opt, NO_UID32);
1596 #ifdef CONFIG_EXT3_FS_XATTR
1597         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1598                 set_opt(sbi->s_mount_opt, XATTR_USER);
1599 #endif
1600 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1601         if (def_mount_opts & EXT3_DEFM_ACL)
1602                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1603 #endif
1604         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1605                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1606         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1607                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1608         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1609                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1610
1611         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1612                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1613         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1614                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1615         else
1616                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1617
1618         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1619         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1620
1621         set_opt(sbi->s_mount_opt, RESERVATION);
1622
1623         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1624                             NULL, 0))
1625                 goto failed_mount;
1626
1627         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1628                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1629
1630         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1631             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1632              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1633              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1634                 printk(KERN_WARNING
1635                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1636                        "running e2fsck is recommended\n");
1637         /*
1638          * Check feature flags regardless of the revision level, since we
1639          * previously didn't change the revision level when setting the flags,
1640          * so there is a chance incompat flags are set on a rev 0 filesystem.
1641          */
1642         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1643         if (features) {
1644                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1645                        "unsupported optional features (%x).\n",
1646                        sb->s_id, le32_to_cpu(features));
1647                 goto failed_mount;
1648         }
1649         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1650         if (!(sb->s_flags & MS_RDONLY) && features) {
1651                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1652                        "unsupported optional features (%x).\n",
1653                        sb->s_id, le32_to_cpu(features));
1654                 goto failed_mount;
1655         }
1656         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1657
1658         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1659             blocksize > EXT3_MAX_BLOCK_SIZE) {
1660                 printk(KERN_ERR
1661                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1662                        blocksize, sb->s_id);
1663                 goto failed_mount;
1664         }
1665
1666         hblock = bdev_hardsect_size(sb->s_bdev);
1667         if (sb->s_blocksize != blocksize) {
1668                 /*
1669                  * Make sure the blocksize for the filesystem is larger
1670                  * than the hardware sectorsize for the machine.
1671                  */
1672                 if (blocksize < hblock) {
1673                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1674                                "device blocksize %d.\n", blocksize, hblock);
1675                         goto failed_mount;
1676                 }
1677
1678                 brelse (bh);
1679                 if (!sb_set_blocksize(sb, blocksize)) {
1680                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1681                                 blocksize);
1682                         goto out_fail;
1683                 }
1684                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1685                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1686                 bh = sb_bread(sb, logic_sb_block);
1687                 if (!bh) {
1688                         printk(KERN_ERR
1689                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1690                         goto failed_mount;
1691                 }
1692                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1693                 sbi->s_es = es;
1694                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1695                         printk (KERN_ERR
1696                                 "EXT3-fs: Magic mismatch, very weird !\n");
1697                         goto failed_mount;
1698                 }
1699         }
1700
1701         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1702
1703         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1704                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1705                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1706         } else {
1707                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1708                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1709                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1710                     (!is_power_of_2(sbi->s_inode_size)) ||
1711                     (sbi->s_inode_size > blocksize)) {
1712                         printk (KERN_ERR
1713                                 "EXT3-fs: unsupported inode size: %d\n",
1714                                 sbi->s_inode_size);
1715                         goto failed_mount;
1716                 }
1717         }
1718         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1719                                    le32_to_cpu(es->s_log_frag_size);
1720         if (blocksize != sbi->s_frag_size) {
1721                 printk(KERN_ERR
1722                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1723                        sbi->s_frag_size, blocksize);
1724                 goto failed_mount;
1725         }
1726         sbi->s_frags_per_block = 1;
1727         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1728         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1729         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1730         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1731                 goto cantfind_ext3;
1732         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1733         if (sbi->s_inodes_per_block == 0)
1734                 goto cantfind_ext3;
1735         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1736                                         sbi->s_inodes_per_block;
1737         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1738         sbi->s_sbh = bh;
1739         sbi->s_mount_state = le16_to_cpu(es->s_state);
1740         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1741         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1742         for (i=0; i < 4; i++)
1743                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1744         sbi->s_def_hash_version = es->s_def_hash_version;
1745
1746         if (sbi->s_blocks_per_group > blocksize * 8) {
1747                 printk (KERN_ERR
1748                         "EXT3-fs: #blocks per group too big: %lu\n",
1749                         sbi->s_blocks_per_group);
1750                 goto failed_mount;
1751         }
1752         if (sbi->s_frags_per_group > blocksize * 8) {
1753                 printk (KERN_ERR
1754                         "EXT3-fs: #fragments per group too big: %lu\n",
1755                         sbi->s_frags_per_group);
1756                 goto failed_mount;
1757         }
1758         if (sbi->s_inodes_per_group > blocksize * 8) {
1759                 printk (KERN_ERR
1760                         "EXT3-fs: #inodes per group too big: %lu\n",
1761                         sbi->s_inodes_per_group);
1762                 goto failed_mount;
1763         }
1764
1765         if (le32_to_cpu(es->s_blocks_count) >
1766                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1767                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1768                         " too large to mount safely\n", sb->s_id);
1769                 if (sizeof(sector_t) < 8)
1770                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1771                                         "enabled\n");
1772                 goto failed_mount;
1773         }
1774
1775         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1776                 goto cantfind_ext3;
1777         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1778                                le32_to_cpu(es->s_first_data_block) - 1)
1779                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1780         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1781                    EXT3_DESC_PER_BLOCK(sb);
1782         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1783                                     GFP_KERNEL);
1784         if (sbi->s_group_desc == NULL) {
1785                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1786                 goto failed_mount;
1787         }
1788
1789         bgl_lock_init(&sbi->s_blockgroup_lock);
1790
1791         for (i = 0; i < db_count; i++) {
1792                 block = descriptor_loc(sb, logic_sb_block, i);
1793                 sbi->s_group_desc[i] = sb_bread(sb, block);
1794                 if (!sbi->s_group_desc[i]) {
1795                         printk (KERN_ERR "EXT3-fs: "
1796                                 "can't read group descriptor %d\n", i);
1797                         db_count = i;
1798                         goto failed_mount2;
1799                 }
1800         }
1801         if (!ext3_check_descriptors (sb)) {
1802                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1803                 goto failed_mount2;
1804         }
1805         sbi->s_gdb_count = db_count;
1806         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1807         spin_lock_init(&sbi->s_next_gen_lock);
1808
1809         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1810                         ext3_count_free_blocks(sb));
1811         if (!err) {
1812                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1813                                 ext3_count_free_inodes(sb));
1814         }
1815         if (!err) {
1816                 err = percpu_counter_init(&sbi->s_dirs_counter,
1817                                 ext3_count_dirs(sb));
1818         }
1819         if (err) {
1820                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1821                 goto failed_mount3;
1822         }
1823
1824         /* per fileystem reservation list head & lock */
1825         spin_lock_init(&sbi->s_rsv_window_lock);
1826         sbi->s_rsv_window_root = RB_ROOT;
1827         /* Add a single, static dummy reservation to the start of the
1828          * reservation window list --- it gives us a placeholder for
1829          * append-at-start-of-list which makes the allocation logic
1830          * _much_ simpler. */
1831         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1832         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1833         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1834         sbi->s_rsv_window_head.rsv_goal_size = 0;
1835         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1836
1837         /*
1838          * set up enough so that it can read an inode
1839          */
1840         sb->s_op = &ext3_sops;
1841         sb->s_export_op = &ext3_export_ops;
1842         sb->s_xattr = ext3_xattr_handlers;
1843 #ifdef CONFIG_QUOTA
1844         sb->s_qcop = &ext3_qctl_operations;
1845         sb->dq_op = &ext3_quota_operations;
1846 #endif
1847         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1848
1849         sb->s_root = NULL;
1850
1851         needs_recovery = (es->s_last_orphan != 0 ||
1852                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1853                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1854
1855         /*
1856          * The first inode we look at is the journal inode.  Don't try
1857          * root first: it may be modified in the journal!
1858          */
1859         if (!test_opt(sb, NOLOAD) &&
1860             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1861                 if (ext3_load_journal(sb, es, journal_devnum))
1862                         goto failed_mount3;
1863         } else if (journal_inum) {
1864                 if (ext3_create_journal(sb, es, journal_inum))
1865                         goto failed_mount3;
1866         } else {
1867                 if (!silent)
1868                         printk (KERN_ERR
1869                                 "ext3: No journal on filesystem on %s\n",
1870                                 sb->s_id);
1871                 goto failed_mount3;
1872         }
1873
1874         /* We have now updated the journal if required, so we can
1875          * validate the data journaling mode. */
1876         switch (test_opt(sb, DATA_FLAGS)) {
1877         case 0:
1878                 /* No mode set, assume a default based on the journal
1879                    capabilities: ORDERED_DATA if the journal can
1880                    cope, else JOURNAL_DATA */
1881                 if (journal_check_available_features
1882                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1883                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1884                 else
1885                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1886                 break;
1887
1888         case EXT3_MOUNT_ORDERED_DATA:
1889         case EXT3_MOUNT_WRITEBACK_DATA:
1890                 if (!journal_check_available_features
1891                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1892                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1893                                "requested data journaling mode\n");
1894                         goto failed_mount4;
1895                 }
1896         default:
1897                 break;
1898         }
1899
1900         if (test_opt(sb, NOBH)) {
1901                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1902                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1903                                 "its supported only with writeback mode\n");
1904                         clear_opt(sbi->s_mount_opt, NOBH);
1905                 }
1906         }
1907         /*
1908          * The journal_load will have done any necessary log recovery,
1909          * so we can safely mount the rest of the filesystem now.
1910          */
1911
1912         root = ext3_iget(sb, EXT3_ROOT_INO);
1913         if (IS_ERR(root)) {
1914                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1915                 ret = PTR_ERR(root);
1916                 goto failed_mount4;
1917         }
1918         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1919                 iput(root);
1920                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1921                 goto failed_mount4;
1922         }
1923         sb->s_root = d_alloc_root(root);
1924         if (!sb->s_root) {
1925                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1926                 iput(root);
1927                 ret = -ENOMEM;
1928                 goto failed_mount4;
1929         }
1930
1931         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1932         /*
1933          * akpm: core read_super() calls in here with the superblock locked.
1934          * That deadlocks, because orphan cleanup needs to lock the superblock
1935          * in numerous places.  Here we just pop the lock - it's relatively
1936          * harmless, because we are now ready to accept write_super() requests,
1937          * and aviro says that's the only reason for hanging onto the
1938          * superblock lock.
1939          */
1940         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1941         ext3_orphan_cleanup(sb, es);
1942         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1943         if (needs_recovery)
1944                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1945         ext3_mark_recovery_complete(sb, es);
1946         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1947                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1948                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1949                 "writeback");
1950
1951         lock_kernel();
1952         return 0;
1953
1954 cantfind_ext3:
1955         if (!silent)
1956                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1957                        sb->s_id);
1958         goto failed_mount;
1959
1960 failed_mount4:
1961         journal_destroy(sbi->s_journal);
1962 failed_mount3:
1963         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1964         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1965         percpu_counter_destroy(&sbi->s_dirs_counter);
1966 failed_mount2:
1967         for (i = 0; i < db_count; i++)
1968                 brelse(sbi->s_group_desc[i]);
1969         kfree(sbi->s_group_desc);
1970 failed_mount:
1971 #ifdef CONFIG_QUOTA
1972         for (i = 0; i < MAXQUOTAS; i++)
1973                 kfree(sbi->s_qf_names[i]);
1974 #endif
1975         ext3_blkdev_remove(sbi);
1976         brelse(bh);
1977 out_fail:
1978         sb->s_fs_info = NULL;
1979         kfree(sbi);
1980         lock_kernel();
1981         return ret;
1982 }
1983
1984 /*
1985  * Setup any per-fs journal parameters now.  We'll do this both on
1986  * initial mount, once the journal has been initialised but before we've
1987  * done any recovery; and again on any subsequent remount.
1988  */
1989 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1990 {
1991         struct ext3_sb_info *sbi = EXT3_SB(sb);
1992
1993         if (sbi->s_commit_interval)
1994                 journal->j_commit_interval = sbi->s_commit_interval;
1995         /* We could also set up an ext3-specific default for the commit
1996          * interval here, but for now we'll just fall back to the jbd
1997          * default. */
1998
1999         spin_lock(&journal->j_state_lock);
2000         if (test_opt(sb, BARRIER))
2001                 journal->j_flags |= JFS_BARRIER;
2002         else
2003                 journal->j_flags &= ~JFS_BARRIER;
2004         if (test_opt(sb, DATA_ERR_ABORT))
2005                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2006         else
2007                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2008         spin_unlock(&journal->j_state_lock);
2009 }
2010
2011 static journal_t *ext3_get_journal(struct super_block *sb,
2012                                    unsigned int journal_inum)
2013 {
2014         struct inode *journal_inode;
2015         journal_t *journal;
2016
2017         /* First, test for the existence of a valid inode on disk.  Bad
2018          * things happen if we iget() an unused inode, as the subsequent
2019          * iput() will try to delete it. */
2020
2021         journal_inode = ext3_iget(sb, journal_inum);
2022         if (IS_ERR(journal_inode)) {
2023                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2024                 return NULL;
2025         }
2026         if (!journal_inode->i_nlink) {
2027                 make_bad_inode(journal_inode);
2028                 iput(journal_inode);
2029                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2030                 return NULL;
2031         }
2032
2033         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2034                   journal_inode, journal_inode->i_size);
2035         if (!S_ISREG(journal_inode->i_mode)) {
2036                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2037                 iput(journal_inode);
2038                 return NULL;
2039         }
2040
2041         journal = journal_init_inode(journal_inode);
2042         if (!journal) {
2043                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2044                 iput(journal_inode);
2045                 return NULL;
2046         }
2047         journal->j_private = sb;
2048         ext3_init_journal_params(sb, journal);
2049         return journal;
2050 }
2051
2052 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2053                                        dev_t j_dev)
2054 {
2055         struct buffer_head * bh;
2056         journal_t *journal;
2057         ext3_fsblk_t start;
2058         ext3_fsblk_t len;
2059         int hblock, blocksize;
2060         ext3_fsblk_t sb_block;
2061         unsigned long offset;
2062         struct ext3_super_block * es;
2063         struct block_device *bdev;
2064
2065         bdev = ext3_blkdev_get(j_dev);
2066         if (bdev == NULL)
2067                 return NULL;
2068
2069         if (bd_claim(bdev, sb)) {
2070                 printk(KERN_ERR
2071                         "EXT3: failed to claim external journal device.\n");
2072                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2073                 return NULL;
2074         }
2075
2076         blocksize = sb->s_blocksize;
2077         hblock = bdev_hardsect_size(bdev);
2078         if (blocksize < hblock) {
2079                 printk(KERN_ERR
2080                         "EXT3-fs: blocksize too small for journal device.\n");
2081                 goto out_bdev;
2082         }
2083
2084         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2085         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2086         set_blocksize(bdev, blocksize);
2087         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2088                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2089                        "external journal\n");
2090                 goto out_bdev;
2091         }
2092
2093         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2094         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2095             !(le32_to_cpu(es->s_feature_incompat) &
2096               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2097                 printk(KERN_ERR "EXT3-fs: external journal has "
2098                                         "bad superblock\n");
2099                 brelse(bh);
2100                 goto out_bdev;
2101         }
2102
2103         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2104                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2105                 brelse(bh);
2106                 goto out_bdev;
2107         }
2108
2109         len = le32_to_cpu(es->s_blocks_count);
2110         start = sb_block + 1;
2111         brelse(bh);     /* we're done with the superblock */
2112
2113         journal = journal_init_dev(bdev, sb->s_bdev,
2114                                         start, len, blocksize);
2115         if (!journal) {
2116                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2117                 goto out_bdev;
2118         }
2119         journal->j_private = sb;
2120         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2121         wait_on_buffer(journal->j_sb_buffer);
2122         if (!buffer_uptodate(journal->j_sb_buffer)) {
2123                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2124                 goto out_journal;
2125         }
2126         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2127                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2128                                         "user (unsupported) - %d\n",
2129                         be32_to_cpu(journal->j_superblock->s_nr_users));
2130                 goto out_journal;
2131         }
2132         EXT3_SB(sb)->journal_bdev = bdev;
2133         ext3_init_journal_params(sb, journal);
2134         return journal;
2135 out_journal:
2136         journal_destroy(journal);
2137 out_bdev:
2138         ext3_blkdev_put(bdev);
2139         return NULL;
2140 }
2141
2142 static int ext3_load_journal(struct super_block *sb,
2143                              struct ext3_super_block *es,
2144                              unsigned long journal_devnum)
2145 {
2146         journal_t *journal;
2147         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2148         dev_t journal_dev;
2149         int err = 0;
2150         int really_read_only;
2151
2152         if (journal_devnum &&
2153             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2154                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2155                         "numbers have changed\n");
2156                 journal_dev = new_decode_dev(journal_devnum);
2157         } else
2158                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2159
2160         really_read_only = bdev_read_only(sb->s_bdev);
2161
2162         /*
2163          * Are we loading a blank journal or performing recovery after a
2164          * crash?  For recovery, we need to check in advance whether we
2165          * can get read-write access to the device.
2166          */
2167
2168         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2169                 if (sb->s_flags & MS_RDONLY) {
2170                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2171                                         "required on readonly filesystem.\n");
2172                         if (really_read_only) {
2173                                 printk(KERN_ERR "EXT3-fs: write access "
2174                                         "unavailable, cannot proceed.\n");
2175                                 return -EROFS;
2176                         }
2177                         printk (KERN_INFO "EXT3-fs: write access will "
2178                                         "be enabled during recovery.\n");
2179                 }
2180         }
2181
2182         if (journal_inum && journal_dev) {
2183                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2184                        "and inode journals!\n");
2185                 return -EINVAL;
2186         }
2187
2188         if (journal_inum) {
2189                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2190                         return -EINVAL;
2191         } else {
2192                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2193                         return -EINVAL;
2194         }
2195
2196         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2197                 err = journal_update_format(journal);
2198                 if (err)  {
2199                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2200                         journal_destroy(journal);
2201                         return err;
2202                 }
2203         }
2204
2205         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2206                 err = journal_wipe(journal, !really_read_only);
2207         if (!err)
2208                 err = journal_load(journal);
2209
2210         if (err) {
2211                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2212                 journal_destroy(journal);
2213                 return err;
2214         }
2215
2216         EXT3_SB(sb)->s_journal = journal;
2217         ext3_clear_journal_err(sb, es);
2218
2219         if (journal_devnum &&
2220             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2221                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2222                 sb->s_dirt = 1;
2223
2224                 /* Make sure we flush the recovery flag to disk. */
2225                 ext3_commit_super(sb, es, 1);
2226         }
2227
2228         return 0;
2229 }
2230
2231 static int ext3_create_journal(struct super_block * sb,
2232                                struct ext3_super_block * es,
2233                                unsigned int journal_inum)
2234 {
2235         journal_t *journal;
2236         int err;
2237
2238         if (sb->s_flags & MS_RDONLY) {
2239                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2240                                 "create journal.\n");
2241                 return -EROFS;
2242         }
2243
2244         journal = ext3_get_journal(sb, journal_inum);
2245         if (!journal)
2246                 return -EINVAL;
2247
2248         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2249                journal_inum);
2250
2251         err = journal_create(journal);
2252         if (err) {
2253                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2254                 journal_destroy(journal);
2255                 return -EIO;
2256         }
2257
2258         EXT3_SB(sb)->s_journal = journal;
2259
2260         ext3_update_dynamic_rev(sb);
2261         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2262         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2263
2264         es->s_journal_inum = cpu_to_le32(journal_inum);
2265         sb->s_dirt = 1;
2266
2267         /* Make sure we flush the recovery flag to disk. */
2268         ext3_commit_super(sb, es, 1);
2269
2270         return 0;
2271 }
2272
2273 static void ext3_commit_super (struct super_block * sb,
2274                                struct ext3_super_block * es,
2275                                int sync)
2276 {
2277         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2278
2279         if (!sbh)
2280                 return;
2281         es->s_wtime = cpu_to_le32(get_seconds());
2282         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2283         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2284         BUFFER_TRACE(sbh, "marking dirty");
2285         mark_buffer_dirty(sbh);
2286         if (sync)
2287                 sync_dirty_buffer(sbh);
2288 }
2289
2290
2291 /*
2292  * Have we just finished recovery?  If so, and if we are mounting (or
2293  * remounting) the filesystem readonly, then we will end up with a
2294  * consistent fs on disk.  Record that fact.
2295  */
2296 static void ext3_mark_recovery_complete(struct super_block * sb,
2297                                         struct ext3_super_block * es)
2298 {
2299         journal_t *journal = EXT3_SB(sb)->s_journal;
2300
2301         journal_lock_updates(journal);
2302         if (journal_flush(journal) < 0)
2303                 goto out;
2304
2305         lock_super(sb);
2306         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2307             sb->s_flags & MS_RDONLY) {
2308                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2309                 sb->s_dirt = 0;
2310                 ext3_commit_super(sb, es, 1);
2311         }
2312         unlock_super(sb);
2313
2314 out:
2315         journal_unlock_updates(journal);
2316 }
2317
2318 /*
2319  * If we are mounting (or read-write remounting) a filesystem whose journal
2320  * has recorded an error from a previous lifetime, move that error to the
2321  * main filesystem now.
2322  */
2323 static void ext3_clear_journal_err(struct super_block * sb,
2324                                    struct ext3_super_block * es)
2325 {
2326         journal_t *journal;
2327         int j_errno;
2328         const char *errstr;
2329
2330         journal = EXT3_SB(sb)->s_journal;
2331
2332         /*
2333          * Now check for any error status which may have been recorded in the
2334          * journal by a prior ext3_error() or ext3_abort()
2335          */
2336
2337         j_errno = journal_errno(journal);
2338         if (j_errno) {
2339                 char nbuf[16];
2340
2341                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2342                 ext3_warning(sb, __func__, "Filesystem error recorded "
2343                              "from previous mount: %s", errstr);
2344                 ext3_warning(sb, __func__, "Marking fs in need of "
2345                              "filesystem check.");
2346
2347                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2348                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2349                 ext3_commit_super (sb, es, 1);
2350
2351                 journal_clear_err(journal);
2352         }
2353 }
2354
2355 /*
2356  * Force the running and committing transactions to commit,
2357  * and wait on the commit.
2358  */
2359 int ext3_force_commit(struct super_block *sb)
2360 {
2361         journal_t *journal;
2362         int ret;
2363
2364         if (sb->s_flags & MS_RDONLY)
2365                 return 0;
2366
2367         journal = EXT3_SB(sb)->s_journal;
2368         sb->s_dirt = 0;
2369         ret = ext3_journal_force_commit(journal);
2370         return ret;
2371 }
2372
2373 /*
2374  * Ext3 always journals updates to the superblock itself, so we don't
2375  * have to propagate any other updates to the superblock on disk at this
2376  * point.  (We can probably nuke this function altogether, and remove
2377  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2378  */
2379 static void ext3_write_super (struct super_block * sb)
2380 {
2381         if (mutex_trylock(&sb->s_lock) != 0)
2382                 BUG();
2383         sb->s_dirt = 0;
2384 }
2385
2386 static int ext3_sync_fs(struct super_block *sb, int wait)
2387 {
2388         sb->s_dirt = 0;
2389         if (wait)
2390                 ext3_force_commit(sb);
2391         else
2392                 journal_start_commit(EXT3_SB(sb)->s_journal, NULL);
2393
2394         return 0;
2395 }
2396
2397 /*
2398  * LVM calls this function before a (read-only) snapshot is created.  This
2399  * gives us a chance to flush the journal completely and mark the fs clean.
2400  */
2401 static void ext3_write_super_lockfs(struct super_block *sb)
2402 {
2403         sb->s_dirt = 0;
2404
2405         if (!(sb->s_flags & MS_RDONLY)) {
2406                 journal_t *journal = EXT3_SB(sb)->s_journal;
2407
2408                 /* Now we set up the journal barrier. */
2409                 journal_lock_updates(journal);
2410
2411                 /*
2412                  * We don't want to clear needs_recovery flag when we failed
2413                  * to flush the journal.
2414                  */
2415                 if (journal_flush(journal) < 0)
2416                         return;
2417
2418                 /* Journal blocked and flushed, clear needs_recovery flag. */
2419                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2420                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2421         }
2422 }
2423
2424 /*
2425  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2426  * flag here, even though the filesystem is not technically dirty yet.
2427  */
2428 static void ext3_unlockfs(struct super_block *sb)
2429 {
2430         if (!(sb->s_flags & MS_RDONLY)) {
2431                 lock_super(sb);
2432                 /* Reser the needs_recovery flag before the fs is unlocked. */
2433                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2434                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2435                 unlock_super(sb);
2436                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2437         }
2438 }
2439
2440 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2441 {
2442         struct ext3_super_block * es;
2443         struct ext3_sb_info *sbi = EXT3_SB(sb);
2444         ext3_fsblk_t n_blocks_count = 0;
2445         unsigned long old_sb_flags;
2446         struct ext3_mount_options old_opts;
2447         int err;
2448 #ifdef CONFIG_QUOTA
2449         int i;
2450 #endif
2451
2452         /* Store the original options */
2453         old_sb_flags = sb->s_flags;
2454         old_opts.s_mount_opt = sbi->s_mount_opt;
2455         old_opts.s_resuid = sbi->s_resuid;
2456         old_opts.s_resgid = sbi->s_resgid;
2457         old_opts.s_commit_interval = sbi->s_commit_interval;
2458 #ifdef CONFIG_QUOTA
2459         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2460         for (i = 0; i < MAXQUOTAS; i++)
2461                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2462 #endif
2463
2464         /*
2465          * Allow the "check" option to be passed as a remount option.
2466          */
2467         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2468                 err = -EINVAL;
2469                 goto restore_opts;
2470         }
2471
2472         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2473                 ext3_abort(sb, __func__, "Abort forced by user");
2474
2475         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2476                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2477
2478         es = sbi->s_es;
2479
2480         ext3_init_journal_params(sb, sbi->s_journal);
2481
2482         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2483                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2484                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2485                         err = -EROFS;
2486                         goto restore_opts;
2487                 }
2488
2489                 if (*flags & MS_RDONLY) {
2490                         /*
2491                          * First of all, the unconditional stuff we have to do
2492                          * to disable replay of the journal when we next remount
2493                          */
2494                         sb->s_flags |= MS_RDONLY;
2495
2496                         /*
2497                          * OK, test if we are remounting a valid rw partition
2498                          * readonly, and if so set the rdonly flag and then
2499                          * mark the partition as valid again.
2500                          */
2501                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2502                             (sbi->s_mount_state & EXT3_VALID_FS))
2503                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2504
2505                         /*
2506                          * We have to unlock super so that we can wait for
2507                          * transactions.
2508                          */
2509                         unlock_super(sb);
2510                         ext3_mark_recovery_complete(sb, es);
2511                         lock_super(sb);
2512                 } else {
2513                         __le32 ret;
2514                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2515                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2516                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2517                                        "remount RDWR because of unsupported "
2518                                        "optional features (%x).\n",
2519                                        sb->s_id, le32_to_cpu(ret));
2520                                 err = -EROFS;
2521                                 goto restore_opts;
2522                         }
2523
2524                         /*
2525                          * If we have an unprocessed orphan list hanging
2526                          * around from a previously readonly bdev mount,
2527                          * require a full umount/remount for now.
2528                          */
2529                         if (es->s_last_orphan) {
2530                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2531                                        "remount RDWR because of unprocessed "
2532                                        "orphan inode list.  Please "
2533                                        "umount/remount instead.\n",
2534                                        sb->s_id);
2535                                 err = -EINVAL;
2536                                 goto restore_opts;
2537                         }
2538
2539                         /*
2540                          * Mounting a RDONLY partition read-write, so reread
2541                          * and store the current valid flag.  (It may have
2542                          * been changed by e2fsck since we originally mounted
2543                          * the partition.)
2544                          */
2545                         ext3_clear_journal_err(sb, es);
2546                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2547                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2548                                 goto restore_opts;
2549                         if (!ext3_setup_super (sb, es, 0))
2550                                 sb->s_flags &= ~MS_RDONLY;
2551                 }
2552         }
2553 #ifdef CONFIG_QUOTA
2554         /* Release old quota file names */
2555         for (i = 0; i < MAXQUOTAS; i++)
2556                 if (old_opts.s_qf_names[i] &&
2557                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2558                         kfree(old_opts.s_qf_names[i]);
2559 #endif
2560         return 0;
2561 restore_opts:
2562         sb->s_flags = old_sb_flags;
2563         sbi->s_mount_opt = old_opts.s_mount_opt;
2564         sbi->s_resuid = old_opts.s_resuid;
2565         sbi->s_resgid = old_opts.s_resgid;
2566         sbi->s_commit_interval = old_opts.s_commit_interval;
2567 #ifdef CONFIG_QUOTA
2568         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2569         for (i = 0; i < MAXQUOTAS; i++) {
2570                 if (sbi->s_qf_names[i] &&
2571                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2572                         kfree(sbi->s_qf_names[i]);
2573                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2574         }
2575 #endif
2576         return err;
2577 }
2578
2579 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2580 {
2581         struct super_block *sb = dentry->d_sb;
2582         struct ext3_sb_info *sbi = EXT3_SB(sb);
2583         struct ext3_super_block *es = sbi->s_es;
2584         u64 fsid;
2585
2586         if (test_opt(sb, MINIX_DF)) {
2587                 sbi->s_overhead_last = 0;
2588         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2589                 unsigned long ngroups = sbi->s_groups_count, i;
2590                 ext3_fsblk_t overhead = 0;
2591                 smp_rmb();
2592
2593                 /*
2594                  * Compute the overhead (FS structures).  This is constant
2595                  * for a given filesystem unless the number of block groups
2596                  * changes so we cache the previous value until it does.
2597                  */
2598
2599                 /*
2600                  * All of the blocks before first_data_block are
2601                  * overhead
2602                  */
2603                 overhead = le32_to_cpu(es->s_first_data_block);
2604
2605                 /*
2606                  * Add the overhead attributed to the superblock and
2607                  * block group descriptors.  If the sparse superblocks
2608                  * feature is turned on, then not all groups have this.
2609                  */
2610                 for (i = 0; i < ngroups; i++) {
2611                         overhead += ext3_bg_has_super(sb, i) +
2612                                 ext3_bg_num_gdb(sb, i);
2613                         cond_resched();
2614                 }
2615
2616                 /*
2617                  * Every block group has an inode bitmap, a block
2618                  * bitmap, and an inode table.
2619                  */
2620                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2621                 sbi->s_overhead_last = overhead;
2622                 smp_wmb();
2623                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2624         }
2625
2626         buf->f_type = EXT3_SUPER_MAGIC;
2627         buf->f_bsize = sb->s_blocksize;
2628         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2629         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2630         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2631         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2632         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2633                 buf->f_bavail = 0;
2634         buf->f_files = le32_to_cpu(es->s_inodes_count);
2635         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2636         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2637         buf->f_namelen = EXT3_NAME_LEN;
2638         fsid = le64_to_cpup((void *)es->s_uuid) ^
2639                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2640         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2641         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2642         return 0;
2643 }
2644
2645 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2646  * is locked for write. Otherwise the are possible deadlocks:
2647  * Process 1                         Process 2
2648  * ext3_create()                     quota_sync()
2649  *   journal_start()                   write_dquot()
2650  *   DQUOT_INIT()                        down(dqio_mutex)
2651  *     down(dqio_mutex)                    journal_start()
2652  *
2653  */
2654
2655 #ifdef CONFIG_QUOTA
2656
2657 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2658 {
2659         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2660 }
2661
2662 static int ext3_dquot_initialize(struct inode *inode, int type)
2663 {
2664         handle_t *handle;
2665         int ret, err;
2666
2667         /* We may create quota structure so we need to reserve enough blocks */
2668         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2669         if (IS_ERR(handle))
2670                 return PTR_ERR(handle);
2671         ret = dquot_initialize(inode, type);
2672         err = ext3_journal_stop(handle);
2673         if (!ret)
2674                 ret = err;
2675         return ret;
2676 }
2677
2678 static int ext3_dquot_drop(struct inode *inode)
2679 {
2680         handle_t *handle;
2681         int ret, err;
2682
2683         /* We may delete quota structure so we need to reserve enough blocks */
2684         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2685         if (IS_ERR(handle)) {
2686                 /*
2687                  * We call dquot_drop() anyway to at least release references
2688                  * to quota structures so that umount does not hang.
2689                  */
2690                 dquot_drop(inode);
2691                 return PTR_ERR(handle);
2692         }
2693         ret = dquot_drop(inode);
2694         err = ext3_journal_stop(handle);
2695         if (!ret)
2696                 ret = err;
2697         return ret;
2698 }
2699
2700 static int ext3_write_dquot(struct dquot *dquot)
2701 {
2702         int ret, err;
2703         handle_t *handle;
2704         struct inode *inode;
2705
2706         inode = dquot_to_inode(dquot);
2707         handle = ext3_journal_start(inode,
2708                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2709         if (IS_ERR(handle))
2710                 return PTR_ERR(handle);
2711         ret = dquot_commit(dquot);
2712         err = ext3_journal_stop(handle);
2713         if (!ret)
2714                 ret = err;
2715         return ret;
2716 }
2717
2718 static int ext3_acquire_dquot(struct dquot *dquot)
2719 {
2720         int ret, err;
2721         handle_t *handle;
2722
2723         handle = ext3_journal_start(dquot_to_inode(dquot),
2724                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2725         if (IS_ERR(handle))
2726                 return PTR_ERR(handle);
2727         ret = dquot_acquire(dquot);
2728         err = ext3_journal_stop(handle);
2729         if (!ret)
2730                 ret = err;
2731         return ret;
2732 }
2733
2734 static int ext3_release_dquot(struct dquot *dquot)
2735 {
2736         int ret, err;
2737         handle_t *handle;
2738
2739         handle = ext3_journal_start(dquot_to_inode(dquot),
2740                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2741         if (IS_ERR(handle)) {
2742                 /* Release dquot anyway to avoid endless cycle in dqput() */
2743                 dquot_release(dquot);
2744                 return PTR_ERR(handle);
2745         }
2746         ret = dquot_release(dquot);
2747         err = ext3_journal_stop(handle);
2748         if (!ret)
2749                 ret = err;
2750         return ret;
2751 }
2752
2753 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2754 {
2755         /* Are we journaling quotas? */
2756         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2757             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2758                 dquot_mark_dquot_dirty(dquot);
2759                 return ext3_write_dquot(dquot);
2760         } else {
2761                 return dquot_mark_dquot_dirty(dquot);
2762         }
2763 }
2764
2765 static int ext3_write_info(struct super_block *sb, int type)
2766 {
2767         int ret, err;
2768         handle_t *handle;
2769
2770         /* Data block + inode block */
2771         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2772         if (IS_ERR(handle))
2773                 return PTR_ERR(handle);
2774         ret = dquot_commit_info(sb, type);
2775         err = ext3_journal_stop(handle);
2776         if (!ret)
2777                 ret = err;
2778         return ret;
2779 }
2780
2781 /*
2782  * Turn on quotas during mount time - we need to find
2783  * the quota file and such...
2784  */
2785 static int ext3_quota_on_mount(struct super_block *sb, int type)
2786 {
2787         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2788                         EXT3_SB(sb)->s_jquota_fmt, type);
2789 }
2790
2791 /*
2792  * Standard function to be called on quota_on
2793  */
2794 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2795                          char *name, int remount)
2796 {
2797         int err;
2798         struct path path;
2799
2800         if (!test_opt(sb, QUOTA))
2801                 return -EINVAL;
2802         /* When remounting, no checks are needed and in fact, name is NULL */
2803         if (remount)
2804                 return vfs_quota_on(sb, type, format_id, name, remount);
2805
2806         err = kern_path(name, LOOKUP_FOLLOW, &path);
2807         if (err)
2808                 return err;
2809
2810         /* Quotafile not on the same filesystem? */
2811         if (path.mnt->mnt_sb != sb) {
2812                 path_put(&path);
2813                 return -EXDEV;
2814         }
2815         /* Journaling quota? */
2816         if (EXT3_SB(sb)->s_qf_names[type]) {
2817                 /* Quotafile not of fs root? */
2818                 if (path.dentry->d_parent != sb->s_root)
2819                         printk(KERN_WARNING
2820                                 "EXT3-fs: Quota file not on filesystem root. "
2821                                 "Journaled quota will not work.\n");
2822         }
2823
2824         /*
2825          * When we journal data on quota file, we have to flush journal to see
2826          * all updates to the file when we bypass pagecache...
2827          */
2828         if (ext3_should_journal_data(path.dentry->d_inode)) {
2829                 /*
2830                  * We don't need to lock updates but journal_flush() could
2831                  * otherwise be livelocked...
2832                  */
2833                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2834                 err = journal_flush(EXT3_SB(sb)->s_journal);
2835                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2836                 if (err) {
2837                         path_put(&path);
2838                         return err;
2839                 }
2840         }
2841
2842         err = vfs_quota_on_path(sb, type, format_id, &path);
2843         path_put(&path);
2844         return err;
2845 }
2846
2847 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2848  * acquiring the locks... As quota files are never truncated and quota code
2849  * itself serializes the operations (and noone else should touch the files)
2850  * we don't have to be afraid of races */
2851 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2852                                size_t len, loff_t off)
2853 {
2854         struct inode *inode = sb_dqopt(sb)->files[type];
2855         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2856         int err = 0;
2857         int offset = off & (sb->s_blocksize - 1);
2858         int tocopy;
2859         size_t toread;
2860         struct buffer_head *bh;
2861         loff_t i_size = i_size_read(inode);
2862
2863         if (off > i_size)
2864                 return 0;
2865         if (off+len > i_size)
2866                 len = i_size-off;
2867         toread = len;
2868         while (toread > 0) {
2869                 tocopy = sb->s_blocksize - offset < toread ?
2870                                 sb->s_blocksize - offset : toread;
2871                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2872                 if (err)
2873                         return err;
2874                 if (!bh)        /* A hole? */
2875                         memset(data, 0, tocopy);
2876                 else
2877                         memcpy(data, bh->b_data+offset, tocopy);
2878                 brelse(bh);
2879                 offset = 0;
2880                 toread -= tocopy;
2881                 data += tocopy;
2882                 blk++;
2883         }
2884         return len;
2885 }
2886
2887 /* Write to quotafile (we know the transaction is already started and has
2888  * enough credits) */
2889 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2890                                 const char *data, size_t len, loff_t off)
2891 {
2892         struct inode *inode = sb_dqopt(sb)->files[type];
2893         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2894         int err = 0;
2895         int offset = off & (sb->s_blocksize - 1);
2896         int tocopy;
2897         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2898         size_t towrite = len;
2899         struct buffer_head *bh;
2900         handle_t *handle = journal_current_handle();
2901
2902         if (!handle) {
2903                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2904                         " cancelled because transaction is not started.\n",
2905                         (unsigned long long)off, (unsigned long long)len);
2906                 return -EIO;
2907         }
2908         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2909         while (towrite > 0) {
2910                 tocopy = sb->s_blocksize - offset < towrite ?
2911                                 sb->s_blocksize - offset : towrite;
2912                 bh = ext3_bread(handle, inode, blk, 1, &err);
2913                 if (!bh)
2914                         goto out;
2915                 if (journal_quota) {
2916                         err = ext3_journal_get_write_access(handle, bh);
2917                         if (err) {
2918                                 brelse(bh);
2919                                 goto out;
2920                         }
2921                 }
2922                 lock_buffer(bh);
2923                 memcpy(bh->b_data+offset, data, tocopy);
2924                 flush_dcache_page(bh->b_page);
2925                 unlock_buffer(bh);
2926                 if (journal_quota)
2927                         err = ext3_journal_dirty_metadata(handle, bh);
2928                 else {
2929                         /* Always do at least ordered writes for quotas */
2930                         err = ext3_journal_dirty_data(handle, bh);
2931                         mark_buffer_dirty(bh);
2932                 }
2933                 brelse(bh);
2934                 if (err)
2935                         goto out;
2936                 offset = 0;
2937                 towrite -= tocopy;
2938                 data += tocopy;
2939                 blk++;
2940         }
2941 out:
2942         if (len == towrite) {
2943                 mutex_unlock(&inode->i_mutex);
2944                 return err;
2945         }
2946         if (inode->i_size < off+len-towrite) {
2947                 i_size_write(inode, off+len-towrite);
2948                 EXT3_I(inode)->i_disksize = inode->i_size;
2949         }
2950         inode->i_version++;
2951         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2952         ext3_mark_inode_dirty(handle, inode);
2953         mutex_unlock(&inode->i_mutex);
2954         return len - towrite;
2955 }
2956
2957 #endif
2958
2959 static int ext3_get_sb(struct file_system_type *fs_type,
2960         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2961 {
2962         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2963 }
2964
2965 static struct file_system_type ext3_fs_type = {
2966         .owner          = THIS_MODULE,
2967         .name           = "ext3",
2968         .get_sb         = ext3_get_sb,
2969         .kill_sb        = kill_block_super,
2970         .fs_flags       = FS_REQUIRES_DEV,
2971 };
2972
2973 static int __init init_ext3_fs(void)
2974 {
2975         int err = init_ext3_xattr();
2976         if (err)
2977                 return err;
2978         err = init_inodecache();
2979         if (err)
2980                 goto out1;
2981         err = register_filesystem(&ext3_fs_type);
2982         if (err)
2983                 goto out;
2984         return 0;
2985 out:
2986         destroy_inodecache();
2987 out1:
2988         exit_ext3_xattr();
2989         return err;
2990 }
2991
2992 static void __exit exit_ext3_fs(void)
2993 {
2994         unregister_filesystem(&ext3_fs_type);
2995         destroy_inodecache();
2996         exit_ext3_xattr();
2997 }
2998
2999 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3000 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3001 MODULE_LICENSE("GPL");
3002 module_init(init_ext3_fs)
3003 module_exit(exit_ext3_fs)