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->s_blockgroup_lock);
 443         kfree(sbi);
 444         return;
 445 }
 446
 447 static struct kmem_cache *ext3_inode_cachep;
 448
 449 /*
 450  * Called inside transaction, so use GFP_NOFS
 451  */
 452 static struct inode *ext3_alloc_inode(struct super_block *sb)
 453 {
 454         struct ext3_inode_info *ei;
 455
 456         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 457         if (!ei)
 458                 return NULL;
 459 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 460         ei->i_acl = EXT3_ACL_NOT_CACHED;
 461         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 462 #endif
 463         ei->i_block_alloc_info = NULL;
 464         ei->vfs_inode.i_version = 1;
 465         return &ei->vfs_inode;
 466 }
 467
 468 static void ext3_destroy_inode(struct inode *inode)
 469 {
 470         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 471                 printk("EXT3 Inode %p: orphan list check failed!\n",
 472                         EXT3_I(inode));
 473                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 474                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 475                                 false);
 476                 dump_stack();
 477         }
 478         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 479 }
 480
 481 static void init_once(void *foo)
 482 {
 483         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 484
 485         INIT_LIST_HEAD(&ei->i_orphan);
 486 #ifdef CONFIG_EXT3_FS_XATTR
 487         init_rwsem(&ei->xattr_sem);
 488 #endif
 489         mutex_init(&ei->truncate_mutex);
 490         inode_init_once(&ei->vfs_inode);
 491 }
 492
 493 static int init_inodecache(void)
 494 {
 495         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 496                                              sizeof(struct ext3_inode_info),
 497                                              0, (SLAB_RECLAIM_ACCOUNT|
 498                                                 SLAB_MEM_SPREAD),
 499                                              init_once);
 500         if (ext3_inode_cachep == NULL)
 501                 return -ENOMEM;
 502         return 0;
 503 }
 504
 505 static void destroy_inodecache(void)
 506 {
 507         kmem_cache_destroy(ext3_inode_cachep);
 508 }
 509
 510 static void ext3_clear_inode(struct inode *inode)
 511 {
 512         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 513 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 514         if (EXT3_I(inode)->i_acl &&
 515                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 516                 posix_acl_release(EXT3_I(inode)->i_acl);
 517                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 518         }
 519         if (EXT3_I(inode)->i_default_acl &&
 520                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 521                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 522                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 523         }
 524 #endif
 525         ext3_discard_reservation(inode);
 526         EXT3_I(inode)->i_block_alloc_info = NULL;
 527         if (unlikely(rsv))
 528                 kfree(rsv);
 529 }
 530
 531 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 532 {
 533 #if defined(CONFIG_QUOTA)
 534         struct ext3_sb_info *sbi = EXT3_SB(sb);
 535
 536         if (sbi->s_jquota_fmt)
 537                 seq_printf(seq, ",jqfmt=%s",
 538                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 539
 540         if (sbi->s_qf_names[USRQUOTA])
 541                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 542
 543         if (sbi->s_qf_names[GRPQUOTA])
 544                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 545
 546         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 547                 seq_puts(seq, ",usrquota");
 548
 549         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 550                 seq_puts(seq, ",grpquota");
 551 #endif
 552 }
 553
 554 /*
 555  * Show an option if
 556  *  - it's set to a non-default value OR
 557  *  - if the per-sb default is different from the global default
 558  */
 559 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 560 {
 561         struct super_block *sb = vfs->mnt_sb;
 562         struct ext3_sb_info *sbi = EXT3_SB(sb);
 563         struct ext3_super_block *es = sbi->s_es;
 564         unsigned long def_mount_opts;
 565
 566         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 567
 568         if (sbi->s_sb_block != 1)
 569                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 570         if (test_opt(sb, MINIX_DF))
 571                 seq_puts(seq, ",minixdf");
 572         if (test_opt(sb, GRPID))
 573                 seq_puts(seq, ",grpid");
 574         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 575                 seq_puts(seq, ",nogrpid");
 576         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 577             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 578                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 579         }
 580         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 581             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 582                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 583         }
 584         if (test_opt(sb, ERRORS_RO)) {
 585                 int def_errors = le16_to_cpu(es->s_errors);
 586
 587                 if (def_errors == EXT3_ERRORS_PANIC ||
 588                     def_errors == EXT3_ERRORS_CONTINUE) {
 589                         seq_puts(seq, ",errors=remount-ro");
 590                 }
 591         }
 592         if (test_opt(sb, ERRORS_CONT))
 593                 seq_puts(seq, ",errors=continue");
 594         if (test_opt(sb, ERRORS_PANIC))
 595                 seq_puts(seq, ",errors=panic");
 596         if (test_opt(sb, NO_UID32))
 597                 seq_puts(seq, ",nouid32");
 598         if (test_opt(sb, DEBUG))
 599                 seq_puts(seq, ",debug");
 600         if (test_opt(sb, OLDALLOC))
 601                 seq_puts(seq, ",oldalloc");
 602 #ifdef CONFIG_EXT3_FS_XATTR
 603         if (test_opt(sb, XATTR_USER))
 604                 seq_puts(seq, ",user_xattr");
 605         if (!test_opt(sb, XATTR_USER) &&
 606             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 607                 seq_puts(seq, ",nouser_xattr");
 608         }
 609 #endif
 610 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 611         if (test_opt(sb, POSIX_ACL))
 612                 seq_puts(seq, ",acl");
 613         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 614                 seq_puts(seq, ",noacl");
 615 #endif
 616         if (!test_opt(sb, RESERVATION))
 617                 seq_puts(seq, ",noreservation");
 618         if (sbi->s_commit_interval) {
 619                 seq_printf(seq, ",commit=%u",
 620                            (unsigned) (sbi->s_commit_interval / HZ));
 621         }
 622         if (test_opt(sb, BARRIER))
 623                 seq_puts(seq, ",barrier=1");
 624         if (test_opt(sb, NOBH))
 625                 seq_puts(seq, ",nobh");
 626
 627         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 628                 seq_puts(seq, ",data=journal");
 629         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 630                 seq_puts(seq, ",data=ordered");
 631         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 632                 seq_puts(seq, ",data=writeback");
 633
 634         if (test_opt(sb, DATA_ERR_ABORT))
 635                 seq_puts(seq, ",data_err=abort");
 636
 637         ext3_show_quota_options(seq, sb);
 638
 639         return 0;
 640 }
 641
 642
 643 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 644                 u64 ino, u32 generation)
 645 {
 646         struct inode *inode;
 647
 648         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 649                 return ERR_PTR(-ESTALE);
 650         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 651                 return ERR_PTR(-ESTALE);
 652
 653         /* iget isn't really right if the inode is currently unallocated!!
 654          *
 655          * ext3_read_inode will return a bad_inode if the inode had been
 656          * deleted, so we should be safe.
 657          *
 658          * Currently we don't know the generation for parent directory, so
 659          * a generation of 0 means "accept any"
 660          */
 661         inode = ext3_iget(sb, ino);
 662         if (IS_ERR(inode))
 663                 return ERR_CAST(inode);
 664         if (generation && inode->i_generation != generation) {
 665                 iput(inode);
 666                 return ERR_PTR(-ESTALE);
 667         }
 668
 669         return inode;
 670 }
 671
 672 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 673                 int fh_len, int fh_type)
 674 {
 675         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 676                                     ext3_nfs_get_inode);
 677 }
 678
 679 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 680                 int fh_len, int fh_type)
 681 {
 682         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 683                                     ext3_nfs_get_inode);
 684 }
 685
 686 #ifdef CONFIG_QUOTA
 687 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 688 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 689
 690 static int ext3_dquot_initialize(struct inode *inode, int type);
 691 static int ext3_dquot_drop(struct inode *inode);
 692 static int ext3_write_dquot(struct dquot *dquot);
 693 static int ext3_acquire_dquot(struct dquot *dquot);
 694 static int ext3_release_dquot(struct dquot *dquot);
 695 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 696 static int ext3_write_info(struct super_block *sb, int type);
 697 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 698                                 char *path, int remount);
 699 static int ext3_quota_on_mount(struct super_block *sb, int type);
 700 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 701                                size_t len, loff_t off);
 702 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 703                                 const char *data, size_t len, loff_t off);
 704
 705 static struct dquot_operations ext3_quota_operations = {
 706         .initialize     = ext3_dquot_initialize,
 707         .drop           = ext3_dquot_drop,
 708         .alloc_space    = dquot_alloc_space,
 709         .alloc_inode    = dquot_alloc_inode,
 710         .free_space     = dquot_free_space,
 711         .free_inode     = dquot_free_inode,
 712         .transfer       = dquot_transfer,
 713         .write_dquot    = ext3_write_dquot,
 714         .acquire_dquot  = ext3_acquire_dquot,
 715         .release_dquot  = ext3_release_dquot,
 716         .mark_dirty     = ext3_mark_dquot_dirty,
 717         .write_info     = ext3_write_info,
 718         .alloc_dquot    = dquot_alloc,
 719         .destroy_dquot  = dquot_destroy,
 720 };
 721
 722 static struct quotactl_ops ext3_qctl_operations = {
 723         .quota_on       = ext3_quota_on,
 724         .quota_off      = vfs_quota_off,
 725         .quota_sync     = vfs_quota_sync,
 726         .get_info       = vfs_get_dqinfo,
 727         .set_info       = vfs_set_dqinfo,
 728         .get_dqblk      = vfs_get_dqblk,
 729         .set_dqblk      = vfs_set_dqblk
 730 };
 731 #endif
 732
 733 static const struct super_operations ext3_sops = {
 734         .alloc_inode    = ext3_alloc_inode,
 735         .destroy_inode  = ext3_destroy_inode,
 736         .write_inode    = ext3_write_inode,
 737         .dirty_inode    = ext3_dirty_inode,
 738         .delete_inode   = ext3_delete_inode,
 739         .put_super      = ext3_put_super,
 740         .write_super    = ext3_write_super,
 741         .sync_fs        = ext3_sync_fs,
 742         .write_super_lockfs = ext3_write_super_lockfs,
 743         .unlockfs       = ext3_unlockfs,
 744         .statfs         = ext3_statfs,
 745         .remount_fs     = ext3_remount,
 746         .clear_inode    = ext3_clear_inode,
 747         .show_options   = ext3_show_options,
 748 #ifdef CONFIG_QUOTA
 749         .quota_read     = ext3_quota_read,
 750         .quota_write    = ext3_quota_write,
 751 #endif
 752 };
 753
 754 static const struct export_operations ext3_export_ops = {
 755         .fh_to_dentry = ext3_fh_to_dentry,
 756         .fh_to_parent = ext3_fh_to_parent,
 757         .get_parent = ext3_get_parent,
 758 };
 759
 760 enum {
 761         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 762         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 763         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 764         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 765         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 766         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 767         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 768         Opt_data_err_abort, Opt_data_err_ignore,
 769         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 770         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 771         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 772         Opt_grpquota
 773 };
 774
 775 static const match_table_t tokens = {
 776         {Opt_bsd_df, "bsddf"},
 777         {Opt_minix_df, "minixdf"},
 778         {Opt_grpid, "grpid"},
 779         {Opt_grpid, "bsdgroups"},
 780         {Opt_nogrpid, "nogrpid"},
 781         {Opt_nogrpid, "sysvgroups"},
 782         {Opt_resgid, "resgid=%u"},
 783         {Opt_resuid, "resuid=%u"},
 784         {Opt_sb, "sb=%u"},
 785         {Opt_err_cont, "errors=continue"},
 786         {Opt_err_panic, "errors=panic"},
 787         {Opt_err_ro, "errors=remount-ro"},
 788         {Opt_nouid32, "nouid32"},
 789         {Opt_nocheck, "nocheck"},
 790         {Opt_nocheck, "check=none"},
 791         {Opt_debug, "debug"},
 792         {Opt_oldalloc, "oldalloc"},
 793         {Opt_orlov, "orlov"},
 794         {Opt_user_xattr, "user_xattr"},
 795         {Opt_nouser_xattr, "nouser_xattr"},
 796         {Opt_acl, "acl"},
 797         {Opt_noacl, "noacl"},
 798         {Opt_reservation, "reservation"},
 799         {Opt_noreservation, "noreservation"},
 800         {Opt_noload, "noload"},
 801         {Opt_nobh, "nobh"},
 802         {Opt_bh, "bh"},
 803         {Opt_commit, "commit=%u"},
 804         {Opt_journal_update, "journal=update"},
 805         {Opt_journal_inum, "journal=%u"},
 806         {Opt_journal_dev, "journal_dev=%u"},
 807         {Opt_abort, "abort"},
 808         {Opt_data_journal, "data=journal"},
 809         {Opt_data_ordered, "data=ordered"},
 810         {Opt_data_writeback, "data=writeback"},
 811         {Opt_data_err_abort, "data_err=abort"},
 812         {Opt_data_err_ignore, "data_err=ignore"},
 813         {Opt_offusrjquota, "usrjquota="},
 814         {Opt_usrjquota, "usrjquota=%s"},
 815         {Opt_offgrpjquota, "grpjquota="},
 816         {Opt_grpjquota, "grpjquota=%s"},
 817         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 818         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 819         {Opt_grpquota, "grpquota"},
 820         {Opt_noquota, "noquota"},
 821         {Opt_quota, "quota"},
 822         {Opt_usrquota, "usrquota"},
 823         {Opt_barrier, "barrier=%u"},
 824         {Opt_resize, "resize"},
 825         {Opt_err, NULL},
 826 };
 827
 828 static ext3_fsblk_t get_sb_block(void **data)
 829 {
 830         ext3_fsblk_t    sb_block;
 831         char            *options = (char *) *data;
 832
 833         if (!options || strncmp(options, "sb=", 3) != 0)
 834                 return 1;       /* Default location */
 835         options += 3;
 836         /*todo: use simple_strtoll with >32bit ext3 */
 837         sb_block = simple_strtoul(options, &options, 0);
 838         if (*options && *options != ',') {
 839                 printk("EXT3-fs: Invalid sb specification: %s\n",
 840                        (char *) *data);
 841                 return 1;
 842         }
 843         if (*options == ',')
 844                 options++;
 845         *data = (void *) options;
 846         return sb_block;
 847 }
 848
 849 static int parse_options (char *options, struct super_block *sb,
 850                           unsigned int *inum, unsigned long *journal_devnum,
 851                           ext3_fsblk_t *n_blocks_count, int is_remount)
 852 {
 853         struct ext3_sb_info *sbi = EXT3_SB(sb);
 854         char * p;
 855         substring_t args[MAX_OPT_ARGS];
 856         int data_opt = 0;
 857         int option;
 858 #ifdef CONFIG_QUOTA
 859         int qtype, qfmt;
 860         char *qname;
 861 #endif
 862
 863         if (!options)
 864                 return 1;
 865
 866         while ((p = strsep (&options, ",")) != NULL) {
 867                 int token;
 868                 if (!*p)
 869                         continue;
 870
 871                 token = match_token(p, tokens, args);
 872                 switch (token) {
 873                 case Opt_bsd_df:
 874                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 875                         break;
 876                 case Opt_minix_df:
 877                         set_opt (sbi->s_mount_opt, MINIX_DF);
 878                         break;
 879                 case Opt_grpid:
 880                         set_opt (sbi->s_mount_opt, GRPID);
 881                         break;
 882                 case Opt_nogrpid:
 883                         clear_opt (sbi->s_mount_opt, GRPID);
 884                         break;
 885                 case Opt_resuid:
 886                         if (match_int(&args[0], &option))
 887                                 return 0;
 888                         sbi->s_resuid = option;
 889                         break;
 890                 case Opt_resgid:
 891                         if (match_int(&args[0], &option))
 892                                 return 0;
 893                         sbi->s_resgid = option;
 894                         break;
 895                 case Opt_sb:
 896                         /* handled by get_sb_block() instead of here */
 897                         /* *sb_block = match_int(&args[0]); */
 898                         break;
 899                 case Opt_err_panic:
 900                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 901                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 902                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 903                         break;
 904                 case Opt_err_ro:
 905                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 906                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 907                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 908                         break;
 909                 case Opt_err_cont:
 910                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 911                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 912                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 913                         break;
 914                 case Opt_nouid32:
 915                         set_opt (sbi->s_mount_opt, NO_UID32);
 916                         break;
 917                 case Opt_nocheck:
 918                         clear_opt (sbi->s_mount_opt, CHECK);
 919                         break;
 920                 case Opt_debug:
 921                         set_opt (sbi->s_mount_opt, DEBUG);
 922                         break;
 923                 case Opt_oldalloc:
 924                         set_opt (sbi->s_mount_opt, OLDALLOC);
 925                         break;
 926                 case Opt_orlov:
 927                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 928                         break;
 929 #ifdef CONFIG_EXT3_FS_XATTR
 930                 case Opt_user_xattr:
 931                         set_opt (sbi->s_mount_opt, XATTR_USER);
 932                         break;
 933                 case Opt_nouser_xattr:
 934                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 935                         break;
 936 #else
 937                 case Opt_user_xattr:
 938                 case Opt_nouser_xattr:
 939                         printk("EXT3 (no)user_xattr options not supported\n");
 940                         break;
 941 #endif
 942 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 943                 case Opt_acl:
 944                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 945                         break;
 946                 case Opt_noacl:
 947                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 948                         break;
 949 #else
 950                 case Opt_acl:
 951                 case Opt_noacl:
 952                         printk("EXT3 (no)acl options not supported\n");
 953                         break;
 954 #endif
 955                 case Opt_reservation:
 956                         set_opt(sbi->s_mount_opt, RESERVATION);
 957                         break;
 958                 case Opt_noreservation:
 959                         clear_opt(sbi->s_mount_opt, RESERVATION);
 960                         break;
 961                 case Opt_journal_update:
 962                         /* @@@ FIXME */
 963                         /* Eventually we will want to be able to create
 964                            a journal file here.  For now, only allow the
 965                            user to specify an existing inode to be the
 966                            journal file. */
 967                         if (is_remount) {
 968                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 969                                        "journal on remount\n");
 970                                 return 0;
 971                         }
 972                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 973                         break;
 974                 case Opt_journal_inum:
 975                         if (is_remount) {
 976                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 977                                        "journal on remount\n");
 978                                 return 0;
 979                         }
 980                         if (match_int(&args[0], &option))
 981                                 return 0;
 982                         *inum = option;
 983                         break;
 984                 case Opt_journal_dev:
 985                         if (is_remount) {
 986                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 987                                        "journal on remount\n");
 988                                 return 0;
 989                         }
 990                         if (match_int(&args[0], &option))
 991                                 return 0;
 992                         *journal_devnum = option;
 993                         break;
 994                 case Opt_noload:
 995                         set_opt (sbi->s_mount_opt, NOLOAD);
 996                         break;
 997                 case Opt_commit:
 998                         if (match_int(&args[0], &option))
 999                                 return 0;
1000                         if (option < 0)
1001                                 return 0;
1002                         if (option == 0)
1003                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1004                         sbi->s_commit_interval = HZ * option;
1005                         break;
1006                 case Opt_data_journal:
1007                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1008                         goto datacheck;
1009                 case Opt_data_ordered:
1010                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1011                         goto datacheck;
1012                 case Opt_data_writeback:
1013                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1014                 datacheck:
1015                         if (is_remount) {
1016                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1017                                                 != data_opt) {
1018                                         printk(KERN_ERR
1019                                                 "EXT3-fs: cannot change data "
1020                                                 "mode on remount\n");
1021                                         return 0;
1022                                 }
1023                         } else {
1024                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1025                                 sbi->s_mount_opt |= data_opt;
1026                         }
1027                         break;
1028                 case Opt_data_err_abort:
1029                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1030                         break;
1031                 case Opt_data_err_ignore:
1032                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1033                         break;
1034 #ifdef CONFIG_QUOTA
1035                 case Opt_usrjquota:
1036                         qtype = USRQUOTA;
1037                         goto set_qf_name;
1038                 case Opt_grpjquota:
1039                         qtype = GRPQUOTA;
1040 set_qf_name:
1041                         if (sb_any_quota_loaded(sb) &&
1042                             !sbi->s_qf_names[qtype]) {
1043                                 printk(KERN_ERR
1044                                         "EXT3-fs: Cannot change journaled "
1045                                         "quota options when quota turned on.\n");
1046                                 return 0;
1047                         }
1048                         qname = match_strdup(&args[0]);
1049                         if (!qname) {
1050                                 printk(KERN_ERR
1051                                         "EXT3-fs: not enough memory for "
1052                                         "storing quotafile name.\n");
1053                                 return 0;
1054                         }
1055                         if (sbi->s_qf_names[qtype] &&
1056                             strcmp(sbi->s_qf_names[qtype], qname)) {
1057                                 printk(KERN_ERR
1058                                         "EXT3-fs: %s quota file already "
1059                                         "specified.\n", QTYPE2NAME(qtype));
1060                                 kfree(qname);
1061                                 return 0;
1062                         }
1063                         sbi->s_qf_names[qtype] = qname;
1064                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1065                                 printk(KERN_ERR
1066                                         "EXT3-fs: quotafile must be on "
1067                                         "filesystem root.\n");
1068                                 kfree(sbi->s_qf_names[qtype]);
1069                                 sbi->s_qf_names[qtype] = NULL;
1070                                 return 0;
1071                         }
1072                         set_opt(sbi->s_mount_opt, QUOTA);
1073                         break;
1074                 case Opt_offusrjquota:
1075                         qtype = USRQUOTA;
1076                         goto clear_qf_name;
1077                 case Opt_offgrpjquota:
1078                         qtype = GRPQUOTA;
1079 clear_qf_name:
1080                         if (sb_any_quota_loaded(sb) &&
1081                             sbi->s_qf_names[qtype]) {
1082                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1083                                         "journaled quota options when "
1084                                         "quota turned on.\n");
1085                                 return 0;
1086                         }
1087                         /*
1088                          * The space will be released later when all options
1089                          * are confirmed to be correct
1090                          */
1091                         sbi->s_qf_names[qtype] = NULL;
1092                         break;
1093                 case Opt_jqfmt_vfsold:
1094                         qfmt = QFMT_VFS_OLD;
1095                         goto set_qf_format;
1096                 case Opt_jqfmt_vfsv0:
1097                         qfmt = QFMT_VFS_V0;
1098 set_qf_format:
1099                         if (sb_any_quota_loaded(sb) &&
1100                             sbi->s_jquota_fmt != qfmt) {
1101                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1102                                         "journaled quota options when "
1103                                         "quota turned on.\n");
1104                                 return 0;
1105                         }
1106                         sbi->s_jquota_fmt = qfmt;
1107                         break;
1108                 case Opt_quota:
1109                 case Opt_usrquota:
1110                         set_opt(sbi->s_mount_opt, QUOTA);
1111                         set_opt(sbi->s_mount_opt, USRQUOTA);
1112                         break;
1113                 case Opt_grpquota:
1114                         set_opt(sbi->s_mount_opt, QUOTA);
1115                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1116                         break;
1117                 case Opt_noquota:
1118                         if (sb_any_quota_loaded(sb)) {
1119                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1120                                         "options when quota turned on.\n");
1121                                 return 0;
1122                         }
1123                         clear_opt(sbi->s_mount_opt, QUOTA);
1124                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1125                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1126                         break;
1127 #else
1128                 case Opt_quota:
1129                 case Opt_usrquota:
1130                 case Opt_grpquota:
1131                         printk(KERN_ERR
1132                                 "EXT3-fs: quota options not supported.\n");
1133                         break;
1134                 case Opt_usrjquota:
1135                 case Opt_grpjquota:
1136                 case Opt_offusrjquota:
1137                 case Opt_offgrpjquota:
1138                 case Opt_jqfmt_vfsold:
1139                 case Opt_jqfmt_vfsv0:
1140                         printk(KERN_ERR
1141                                 "EXT3-fs: journaled quota options not "
1142                                 "supported.\n");
1143                         break;
1144                 case Opt_noquota:
1145                         break;
1146 #endif
1147                 case Opt_abort:
1148                         set_opt(sbi->s_mount_opt, ABORT);
1149                         break;
1150                 case Opt_barrier:
1151                         if (match_int(&args[0], &option))
1152                                 return 0;
1153                         if (option)
1154                                 set_opt(sbi->s_mount_opt, BARRIER);
1155                         else
1156                                 clear_opt(sbi->s_mount_opt, BARRIER);
1157                         break;
1158                 case Opt_ignore:
1159                         break;
1160                 case Opt_resize:
1161                         if (!is_remount) {
1162                                 printk("EXT3-fs: resize option only available "
1163                                         "for remount\n");
1164                                 return 0;
1165                         }
1166                         if (match_int(&args[0], &option) != 0)
1167                                 return 0;
1168                         *n_blocks_count = option;
1169                         break;
1170                 case Opt_nobh:
1171                         set_opt(sbi->s_mount_opt, NOBH);
1172                         break;
1173                 case Opt_bh:
1174                         clear_opt(sbi->s_mount_opt, NOBH);
1175                         break;
1176                 default:
1177                         printk (KERN_ERR
1178                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1179                                 "or missing value\n", p);
1180                         return 0;
1181                 }
1182         }
1183 #ifdef CONFIG_QUOTA
1184         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1185                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1186                      sbi->s_qf_names[USRQUOTA])
1187                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1188
1189                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1190                      sbi->s_qf_names[GRPQUOTA])
1191                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1192
1193                 if ((sbi->s_qf_names[USRQUOTA] &&
1194                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1195                     (sbi->s_qf_names[GRPQUOTA] &&
1196                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1197                         printk(KERN_ERR "EXT3-fs: old and new quota "
1198                                         "format mixing.\n");
1199                         return 0;
1200                 }
1201
1202                 if (!sbi->s_jquota_fmt) {
1203                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1204                                         "not specified.\n");
1205                         return 0;
1206                 }
1207         } else {
1208                 if (sbi->s_jquota_fmt) {
1209                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1210                                         "specified with no journaling "
1211                                         "enabled.\n");
1212                         return 0;
1213                 }
1214         }
1215 #endif
1216         return 1;
1217 }
1218
1219 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1220                             int read_only)
1221 {
1222         struct ext3_sb_info *sbi = EXT3_SB(sb);
1223         int res = 0;
1224
1225         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1226                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1227                         "forcing read-only mode\n");
1228                 res = MS_RDONLY;
1229         }
1230         if (read_only)
1231                 return res;
1232         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1233                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1234                         "running e2fsck is recommended\n");
1235         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1236                 printk (KERN_WARNING
1237                         "EXT3-fs warning: mounting fs with errors, "
1238                         "running e2fsck is recommended\n");
1239         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1240                  le16_to_cpu(es->s_mnt_count) >=
1241                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1242                 printk (KERN_WARNING
1243                         "EXT3-fs warning: maximal mount count reached, "
1244                         "running e2fsck is recommended\n");
1245         else if (le32_to_cpu(es->s_checkinterval) &&
1246                 (le32_to_cpu(es->s_lastcheck) +
1247                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1248                 printk (KERN_WARNING
1249                         "EXT3-fs warning: checktime reached, "
1250                         "running e2fsck is recommended\n");
1251 #if 0
1252                 /* @@@ We _will_ want to clear the valid bit if we find
1253                    inconsistencies, to force a fsck at reboot.  But for
1254                    a plain journaled filesystem we can keep it set as
1255                    valid forever! :) */
1256         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1257 #endif
1258         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1259                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1260         le16_add_cpu(&es->s_mnt_count, 1);
1261         es->s_mtime = cpu_to_le32(get_seconds());
1262         ext3_update_dynamic_rev(sb);
1263         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1264
1265         ext3_commit_super(sb, es, 1);
1266         if (test_opt(sb, DEBUG))
1267                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1268                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1269                         sb->s_blocksize,
1270                         sbi->s_groups_count,
1271                         EXT3_BLOCKS_PER_GROUP(sb),
1272                         EXT3_INODES_PER_GROUP(sb),
1273                         sbi->s_mount_opt);
1274
1275         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1276         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1277                 char b[BDEVNAME_SIZE];
1278
1279                 printk("external journal on %s\n",
1280                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1281         } else {
1282                 printk("internal journal\n");
1283         }
1284         return res;
1285 }
1286
1287 /* Called at mount-time, super-block is locked */
1288 static int ext3_check_descriptors(struct super_block *sb)
1289 {
1290         struct ext3_sb_info *sbi = EXT3_SB(sb);
1291         int i;
1292
1293         ext3_debug ("Checking group descriptors");
1294
1295         for (i = 0; i < sbi->s_groups_count; i++) {
1296                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1297                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1298                 ext3_fsblk_t last_block;
1299
1300                 if (i == sbi->s_groups_count - 1)
1301                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1302                 else
1303                         last_block = first_block +
1304                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1305
1306                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1307                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1308                 {
1309                         ext3_error (sb, "ext3_check_descriptors",
1310                                     "Block bitmap for group %d"
1311                                     " not in group (block %lu)!",
1312                                     i, (unsigned long)
1313                                         le32_to_cpu(gdp->bg_block_bitmap));
1314                         return 0;
1315                 }
1316                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1317                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1318                 {
1319                         ext3_error (sb, "ext3_check_descriptors",
1320                                     "Inode bitmap for group %d"
1321                                     " not in group (block %lu)!",
1322                                     i, (unsigned long)
1323                                         le32_to_cpu(gdp->bg_inode_bitmap));
1324                         return 0;
1325                 }
1326                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1327                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1328                     last_block)
1329                 {
1330                         ext3_error (sb, "ext3_check_descriptors",
1331                                     "Inode table for group %d"
1332                                     " not in group (block %lu)!",
1333                                     i, (unsigned long)
1334                                         le32_to_cpu(gdp->bg_inode_table));
1335                         return 0;
1336                 }
1337         }
1338
1339         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1340         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1341         return 1;
1342 }
1343
1344
1345 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1346  * the superblock) which were deleted from all directories, but held open by
1347  * a process at the time of a crash.  We walk the list and try to delete these
1348  * inodes at recovery time (only with a read-write filesystem).
1349  *
1350  * In order to keep the orphan inode chain consistent during traversal (in
1351  * case of crash during recovery), we link each inode into the superblock
1352  * orphan list_head and handle it the same way as an inode deletion during
1353  * normal operation (which journals the operations for us).
1354  *
1355  * We only do an iget() and an iput() on each inode, which is very safe if we
1356  * accidentally point at an in-use or already deleted inode.  The worst that
1357  * can happen in this case is that we get a "bit already cleared" message from
1358  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1359  * e2fsck was run on this filesystem, and it must have already done the orphan
1360  * inode cleanup for us, so we can safely abort without any further action.
1361  */
1362 static void ext3_orphan_cleanup (struct super_block * sb,
1363                                  struct ext3_super_block * es)
1364 {
1365         unsigned int s_flags = sb->s_flags;
1366         int nr_orphans = 0, nr_truncates = 0;
1367 #ifdef CONFIG_QUOTA
1368         int i;
1369 #endif
1370         if (!es->s_last_orphan) {
1371                 jbd_debug(4, "no orphan inodes to clean up\n");
1372                 return;
1373         }
1374
1375         if (bdev_read_only(sb->s_bdev)) {
1376                 printk(KERN_ERR "EXT3-fs: write access "
1377                         "unavailable, skipping orphan cleanup.\n");
1378                 return;
1379         }
1380
1381         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1382                 if (es->s_last_orphan)
1383                         jbd_debug(1, "Errors on filesystem, "
1384                                   "clearing orphan list.\n");
1385                 es->s_last_orphan = 0;
1386                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1387                 return;
1388         }
1389
1390         if (s_flags & MS_RDONLY) {
1391                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1392                        sb->s_id);
1393                 sb->s_flags &= ~MS_RDONLY;
1394         }
1395 #ifdef CONFIG_QUOTA
1396         /* Needed for iput() to work correctly and not trash data */
1397         sb->s_flags |= MS_ACTIVE;
1398         /* Turn on quotas so that they are updated correctly */
1399         for (i = 0; i < MAXQUOTAS; i++) {
1400                 if (EXT3_SB(sb)->s_qf_names[i]) {
1401                         int ret = ext3_quota_on_mount(sb, i);
1402                         if (ret < 0)
1403                                 printk(KERN_ERR
1404                                         "EXT3-fs: Cannot turn on journaled "
1405                                         "quota: error %d\n", ret);
1406                 }
1407         }
1408 #endif
1409
1410         while (es->s_last_orphan) {
1411                 struct inode *inode;
1412
1413                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1414                 if (IS_ERR(inode)) {
1415                         es->s_last_orphan = 0;
1416                         break;
1417                 }
1418
1419                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1420                 DQUOT_INIT(inode);
1421                 if (inode->i_nlink) {
1422                         printk(KERN_DEBUG
1423                                 "%s: truncating inode %lu to %Ld bytes\n",
1424                                 __func__, inode->i_ino, inode->i_size);
1425                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1426                                   inode->i_ino, inode->i_size);
1427                         ext3_truncate(inode);
1428                         nr_truncates++;
1429                 } else {
1430                         printk(KERN_DEBUG
1431                                 "%s: deleting unreferenced inode %lu\n",
1432                                 __func__, inode->i_ino);
1433                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1434                                   inode->i_ino);
1435                         nr_orphans++;
1436                 }
1437                 iput(inode);  /* The delete magic happens here! */
1438         }
1439
1440 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1441
1442         if (nr_orphans)
1443                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1444                        sb->s_id, PLURAL(nr_orphans));
1445         if (nr_truncates)
1446                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1447                        sb->s_id, PLURAL(nr_truncates));
1448 #ifdef CONFIG_QUOTA
1449         /* Turn quotas off */
1450         for (i = 0; i < MAXQUOTAS; i++) {
1451                 if (sb_dqopt(sb)->files[i])
1452                         vfs_quota_off(sb, i, 0);
1453         }
1454 #endif
1455         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1456 }
1457
1458 /*
1459  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1460  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1461  * We need to be 1 filesystem block less than the 2^32 sector limit.
1462  */
1463 static loff_t ext3_max_size(int bits)
1464 {
1465         loff_t res = EXT3_NDIR_BLOCKS;
1466         int meta_blocks;
1467         loff_t upper_limit;
1468
1469         /* This is calculated to be the largest file size for a
1470          * dense, file such that the total number of
1471          * sectors in the file, including data and all indirect blocks,
1472          * does not exceed 2^32 -1
1473          * __u32 i_blocks representing the total number of
1474          * 512 bytes blocks of the file
1475          */
1476         upper_limit = (1LL << 32) - 1;
1477
1478         /* total blocks in file system block size */
1479         upper_limit >>= (bits - 9);
1480
1481
1482         /* indirect blocks */
1483         meta_blocks = 1;
1484         /* double indirect blocks */
1485         meta_blocks += 1 + (1LL << (bits-2));
1486         /* tripple indirect blocks */
1487         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1488
1489         upper_limit -= meta_blocks;
1490         upper_limit <<= bits;
1491
1492         res += 1LL << (bits-2);
1493         res += 1LL << (2*(bits-2));
1494         res += 1LL << (3*(bits-2));
1495         res <<= bits;
1496         if (res > upper_limit)
1497                 res = upper_limit;
1498
1499         if (res > MAX_LFS_FILESIZE)
1500                 res = MAX_LFS_FILESIZE;
1501
1502         return res;
1503 }
1504
1505 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1506                                     ext3_fsblk_t logic_sb_block,
1507                                     int nr)
1508 {
1509         struct ext3_sb_info *sbi = EXT3_SB(sb);
1510         unsigned long bg, first_meta_bg;
1511         int has_super = 0;
1512
1513         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1514
1515         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1516             nr < first_meta_bg)
1517                 return (logic_sb_block + nr + 1);
1518         bg = sbi->s_desc_per_block * nr;
1519         if (ext3_bg_has_super(sb, bg))
1520                 has_super = 1;
1521         return (has_super + ext3_group_first_block_no(sb, bg));
1522 }
1523
1524
1525 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1526 {
1527         struct buffer_head * bh;
1528         struct ext3_super_block *es = NULL;
1529         struct ext3_sb_info *sbi;
1530         ext3_fsblk_t block;
1531         ext3_fsblk_t sb_block = get_sb_block(&data);
1532         ext3_fsblk_t logic_sb_block;
1533         unsigned long offset = 0;
1534         unsigned int journal_inum = 0;
1535         unsigned long journal_devnum = 0;
1536         unsigned long def_mount_opts;
1537         struct inode *root;
1538         int blocksize;
1539         int hblock;
1540         int db_count;
1541         int i;
1542         int needs_recovery;
1543         int ret = -EINVAL;
1544         __le32 features;
1545         int err;
1546
1547         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1548         if (!sbi)
1549                 return -ENOMEM;
1550
1551         sbi->s_blockgroup_lock =
1552                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1553         if (!sbi->s_blockgroup_lock) {
1554                 kfree(sbi);
1555                 return -ENOMEM;
1556         }
1557         sb->s_fs_info = sbi;
1558         sbi->s_mount_opt = 0;
1559         sbi->s_resuid = EXT3_DEF_RESUID;
1560         sbi->s_resgid = EXT3_DEF_RESGID;
1561         sbi->s_sb_block = sb_block;
1562
1563         unlock_kernel();
1564
1565         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1566         if (!blocksize) {
1567                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1568                 goto out_fail;
1569         }
1570
1571         /*
1572          * The ext3 superblock will not be buffer aligned for other than 1kB
1573          * block sizes.  We need to calculate the offset from buffer start.
1574          */
1575         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1576                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1577                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1578         } else {
1579                 logic_sb_block = sb_block;
1580         }
1581
1582         if (!(bh = sb_bread(sb, logic_sb_block))) {
1583                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1584                 goto out_fail;
1585         }
1586         /*
1587          * Note: s_es must be initialized as soon as possible because
1588          *       some ext3 macro-instructions depend on its value
1589          */
1590         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1591         sbi->s_es = es;
1592         sb->s_magic = le16_to_cpu(es->s_magic);
1593         if (sb->s_magic != EXT3_SUPER_MAGIC)
1594                 goto cantfind_ext3;
1595
1596         /* Set defaults before we parse the mount options */
1597         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1598         if (def_mount_opts & EXT3_DEFM_DEBUG)
1599                 set_opt(sbi->s_mount_opt, DEBUG);
1600         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1601                 set_opt(sbi->s_mount_opt, GRPID);
1602         if (def_mount_opts & EXT3_DEFM_UID16)
1603                 set_opt(sbi->s_mount_opt, NO_UID32);
1604 #ifdef CONFIG_EXT3_FS_XATTR
1605         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1606                 set_opt(sbi->s_mount_opt, XATTR_USER);
1607 #endif
1608 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1609         if (def_mount_opts & EXT3_DEFM_ACL)
1610                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1611 #endif
1612         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1613                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1614         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1615                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1616         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1617                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1618
1619         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1620                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1621         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1622                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1623         else
1624                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1625
1626         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1627         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1628
1629         set_opt(sbi->s_mount_opt, RESERVATION);
1630
1631         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1632                             NULL, 0))
1633                 goto failed_mount;
1634
1635         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1636                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1637
1638         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1639             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1640              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1641              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1642                 printk(KERN_WARNING
1643                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1644                        "running e2fsck is recommended\n");
1645         /*
1646          * Check feature flags regardless of the revision level, since we
1647          * previously didn't change the revision level when setting the flags,
1648          * so there is a chance incompat flags are set on a rev 0 filesystem.
1649          */
1650         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1651         if (features) {
1652                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1653                        "unsupported optional features (%x).\n",
1654                        sb->s_id, le32_to_cpu(features));
1655                 goto failed_mount;
1656         }
1657         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1658         if (!(sb->s_flags & MS_RDONLY) && features) {
1659                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1660                        "unsupported optional features (%x).\n",
1661                        sb->s_id, le32_to_cpu(features));
1662                 goto failed_mount;
1663         }
1664         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1665
1666         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1667             blocksize > EXT3_MAX_BLOCK_SIZE) {
1668                 printk(KERN_ERR
1669                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1670                        blocksize, sb->s_id);
1671                 goto failed_mount;
1672         }
1673
1674         hblock = bdev_hardsect_size(sb->s_bdev);
1675         if (sb->s_blocksize != blocksize) {
1676                 /*
1677                  * Make sure the blocksize for the filesystem is larger
1678                  * than the hardware sectorsize for the machine.
1679                  */
1680                 if (blocksize < hblock) {
1681                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1682                                "device blocksize %d.\n", blocksize, hblock);
1683                         goto failed_mount;
1684                 }
1685
1686                 brelse (bh);
1687                 if (!sb_set_blocksize(sb, blocksize)) {
1688                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1689                                 blocksize);
1690                         goto out_fail;
1691                 }
1692                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1693                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1694                 bh = sb_bread(sb, logic_sb_block);
1695                 if (!bh) {
1696                         printk(KERN_ERR
1697                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1698                         goto failed_mount;
1699                 }
1700                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1701                 sbi->s_es = es;
1702                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1703                         printk (KERN_ERR
1704                                 "EXT3-fs: Magic mismatch, very weird !\n");
1705                         goto failed_mount;
1706                 }
1707         }
1708
1709         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1710
1711         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1712                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1713                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1714         } else {
1715                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1716                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1717                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1718                     (!is_power_of_2(sbi->s_inode_size)) ||
1719                     (sbi->s_inode_size > blocksize)) {
1720                         printk (KERN_ERR
1721                                 "EXT3-fs: unsupported inode size: %d\n",
1722                                 sbi->s_inode_size);
1723                         goto failed_mount;
1724                 }
1725         }
1726         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1727                                    le32_to_cpu(es->s_log_frag_size);
1728         if (blocksize != sbi->s_frag_size) {
1729                 printk(KERN_ERR
1730                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1731                        sbi->s_frag_size, blocksize);
1732                 goto failed_mount;
1733         }
1734         sbi->s_frags_per_block = 1;
1735         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1736         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1737         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1738         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1739                 goto cantfind_ext3;
1740         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1741         if (sbi->s_inodes_per_block == 0)
1742                 goto cantfind_ext3;
1743         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1744                                         sbi->s_inodes_per_block;
1745         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1746         sbi->s_sbh = bh;
1747         sbi->s_mount_state = le16_to_cpu(es->s_state);
1748         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1749         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1750         for (i=0; i < 4; i++)
1751                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1752         sbi->s_def_hash_version = es->s_def_hash_version;
1753
1754         if (sbi->s_blocks_per_group > blocksize * 8) {
1755                 printk (KERN_ERR
1756                         "EXT3-fs: #blocks per group too big: %lu\n",
1757                         sbi->s_blocks_per_group);
1758                 goto failed_mount;
1759         }
1760         if (sbi->s_frags_per_group > blocksize * 8) {
1761                 printk (KERN_ERR
1762                         "EXT3-fs: #fragments per group too big: %lu\n",
1763                         sbi->s_frags_per_group);
1764                 goto failed_mount;
1765         }
1766         if (sbi->s_inodes_per_group > blocksize * 8) {
1767                 printk (KERN_ERR
1768                         "EXT3-fs: #inodes per group too big: %lu\n",
1769                         sbi->s_inodes_per_group);
1770                 goto failed_mount;
1771         }
1772
1773         if (le32_to_cpu(es->s_blocks_count) >
1774                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1775                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1776                         " too large to mount safely\n", sb->s_id);
1777                 if (sizeof(sector_t) < 8)
1778                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1779                                         "enabled\n");
1780                 goto failed_mount;
1781         }
1782
1783         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1784                 goto cantfind_ext3;
1785         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1786                                le32_to_cpu(es->s_first_data_block) - 1)
1787                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1788         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1789                    EXT3_DESC_PER_BLOCK(sb);
1790         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1791                                     GFP_KERNEL);
1792         if (sbi->s_group_desc == NULL) {
1793                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1794                 goto failed_mount;
1795         }
1796
1797         bgl_lock_init(sbi->s_blockgroup_lock);
1798
1799         for (i = 0; i < db_count; i++) {
1800                 block = descriptor_loc(sb, logic_sb_block, i);
1801                 sbi->s_group_desc[i] = sb_bread(sb, block);
1802                 if (!sbi->s_group_desc[i]) {
1803                         printk (KERN_ERR "EXT3-fs: "
1804                                 "can't read group descriptor %d\n", i);
1805                         db_count = i;
1806                         goto failed_mount2;
1807                 }
1808         }
1809         if (!ext3_check_descriptors (sb)) {
1810                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1811                 goto failed_mount2;
1812         }
1813         sbi->s_gdb_count = db_count;
1814         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1815         spin_lock_init(&sbi->s_next_gen_lock);
1816
1817         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1818                         ext3_count_free_blocks(sb));
1819         if (!err) {
1820                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1821                                 ext3_count_free_inodes(sb));
1822         }
1823         if (!err) {
1824                 err = percpu_counter_init(&sbi->s_dirs_counter,
1825                                 ext3_count_dirs(sb));
1826         }
1827         if (err) {
1828                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1829                 goto failed_mount3;
1830         }
1831
1832         /* per fileystem reservation list head & lock */
1833         spin_lock_init(&sbi->s_rsv_window_lock);
1834         sbi->s_rsv_window_root = RB_ROOT;
1835         /* Add a single, static dummy reservation to the start of the
1836          * reservation window list --- it gives us a placeholder for
1837          * append-at-start-of-list which makes the allocation logic
1838          * _much_ simpler. */
1839         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1840         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1841         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1842         sbi->s_rsv_window_head.rsv_goal_size = 0;
1843         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1844
1845         /*
1846          * set up enough so that it can read an inode
1847          */
1848         sb->s_op = &ext3_sops;
1849         sb->s_export_op = &ext3_export_ops;
1850         sb->s_xattr = ext3_xattr_handlers;
1851 #ifdef CONFIG_QUOTA
1852         sb->s_qcop = &ext3_qctl_operations;
1853         sb->dq_op = &ext3_quota_operations;
1854 #endif
1855         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1856
1857         sb->s_root = NULL;
1858
1859         needs_recovery = (es->s_last_orphan != 0 ||
1860                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1861                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1862
1863         /*
1864          * The first inode we look at is the journal inode.  Don't try
1865          * root first: it may be modified in the journal!
1866          */
1867         if (!test_opt(sb, NOLOAD) &&
1868             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1869                 if (ext3_load_journal(sb, es, journal_devnum))
1870                         goto failed_mount3;
1871         } else if (journal_inum) {
1872                 if (ext3_create_journal(sb, es, journal_inum))
1873                         goto failed_mount3;
1874         } else {
1875                 if (!silent)
1876                         printk (KERN_ERR
1877                                 "ext3: No journal on filesystem on %s\n",
1878                                 sb->s_id);
1879                 goto failed_mount3;
1880         }
1881
1882         /* We have now updated the journal if required, so we can
1883          * validate the data journaling mode. */
1884         switch (test_opt(sb, DATA_FLAGS)) {
1885         case 0:
1886                 /* No mode set, assume a default based on the journal
1887                    capabilities: ORDERED_DATA if the journal can
1888                    cope, else JOURNAL_DATA */
1889                 if (journal_check_available_features
1890                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1891                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1892                 else
1893                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1894                 break;
1895
1896         case EXT3_MOUNT_ORDERED_DATA:
1897         case EXT3_MOUNT_WRITEBACK_DATA:
1898                 if (!journal_check_available_features
1899                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1900                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1901                                "requested data journaling mode\n");
1902                         goto failed_mount4;
1903                 }
1904         default:
1905                 break;
1906         }
1907
1908         if (test_opt(sb, NOBH)) {
1909                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1910                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1911                                 "its supported only with writeback mode\n");
1912                         clear_opt(sbi->s_mount_opt, NOBH);
1913                 }
1914         }
1915         /*
1916          * The journal_load will have done any necessary log recovery,
1917          * so we can safely mount the rest of the filesystem now.
1918          */
1919
1920         root = ext3_iget(sb, EXT3_ROOT_INO);
1921         if (IS_ERR(root)) {
1922                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1923                 ret = PTR_ERR(root);
1924                 goto failed_mount4;
1925         }
1926         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1927                 iput(root);
1928                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1929                 goto failed_mount4;
1930         }
1931         sb->s_root = d_alloc_root(root);
1932         if (!sb->s_root) {
1933                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1934                 iput(root);
1935                 ret = -ENOMEM;
1936                 goto failed_mount4;
1937         }
1938
1939         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1940         /*
1941          * akpm: core read_super() calls in here with the superblock locked.
1942          * That deadlocks, because orphan cleanup needs to lock the superblock
1943          * in numerous places.  Here we just pop the lock - it's relatively
1944          * harmless, because we are now ready to accept write_super() requests,
1945          * and aviro says that's the only reason for hanging onto the
1946          * superblock lock.
1947          */
1948         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1949         ext3_orphan_cleanup(sb, es);
1950         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1951         if (needs_recovery)
1952                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1953         ext3_mark_recovery_complete(sb, es);
1954         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1955                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1956                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1957                 "writeback");
1958
1959         lock_kernel();
1960         return 0;
1961
1962 cantfind_ext3:
1963         if (!silent)
1964                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1965                        sb->s_id);
1966         goto failed_mount;
1967
1968 failed_mount4:
1969         journal_destroy(sbi->s_journal);
1970 failed_mount3:
1971         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1972         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1973         percpu_counter_destroy(&sbi->s_dirs_counter);
1974 failed_mount2:
1975         for (i = 0; i < db_count; i++)
1976                 brelse(sbi->s_group_desc[i]);
1977         kfree(sbi->s_group_desc);
1978 failed_mount:
1979 #ifdef CONFIG_QUOTA
1980         for (i = 0; i < MAXQUOTAS; i++)
1981                 kfree(sbi->s_qf_names[i]);
1982 #endif
1983         ext3_blkdev_remove(sbi);
1984         brelse(bh);
1985 out_fail:
1986         sb->s_fs_info = NULL;
1987         kfree(sbi);
1988         lock_kernel();
1989         return ret;
1990 }
1991
1992 /*
1993  * Setup any per-fs journal parameters now.  We'll do this both on
1994  * initial mount, once the journal has been initialised but before we've
1995  * done any recovery; and again on any subsequent remount.
1996  */
1997 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1998 {
1999         struct ext3_sb_info *sbi = EXT3_SB(sb);
2000
2001         if (sbi->s_commit_interval)
2002                 journal->j_commit_interval = sbi->s_commit_interval;
2003         /* We could also set up an ext3-specific default for the commit
2004          * interval here, but for now we'll just fall back to the jbd
2005          * default. */
2006
2007         spin_lock(&journal->j_state_lock);
2008         if (test_opt(sb, BARRIER))
2009                 journal->j_flags |= JFS_BARRIER;
2010         else
2011                 journal->j_flags &= ~JFS_BARRIER;
2012         if (test_opt(sb, DATA_ERR_ABORT))
2013                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2014         else
2015                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2016         spin_unlock(&journal->j_state_lock);
2017 }
2018
2019 static journal_t *ext3_get_journal(struct super_block *sb,
2020                                    unsigned int journal_inum)
2021 {
2022         struct inode *journal_inode;
2023         journal_t *journal;
2024
2025         /* First, test for the existence of a valid inode on disk.  Bad
2026          * things happen if we iget() an unused inode, as the subsequent
2027          * iput() will try to delete it. */
2028
2029         journal_inode = ext3_iget(sb, journal_inum);
2030         if (IS_ERR(journal_inode)) {
2031                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2032                 return NULL;
2033         }
2034         if (!journal_inode->i_nlink) {
2035                 make_bad_inode(journal_inode);
2036                 iput(journal_inode);
2037                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2038                 return NULL;
2039         }
2040
2041         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2042                   journal_inode, journal_inode->i_size);
2043         if (!S_ISREG(journal_inode->i_mode)) {
2044                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2045                 iput(journal_inode);
2046                 return NULL;
2047         }
2048
2049         journal = journal_init_inode(journal_inode);
2050         if (!journal) {
2051                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2052                 iput(journal_inode);
2053                 return NULL;
2054         }
2055         journal->j_private = sb;
2056         ext3_init_journal_params(sb, journal);
2057         return journal;
2058 }
2059
2060 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2061                                        dev_t j_dev)
2062 {
2063         struct buffer_head * bh;
2064         journal_t *journal;
2065         ext3_fsblk_t start;
2066         ext3_fsblk_t len;
2067         int hblock, blocksize;
2068         ext3_fsblk_t sb_block;
2069         unsigned long offset;
2070         struct ext3_super_block * es;
2071         struct block_device *bdev;
2072
2073         bdev = ext3_blkdev_get(j_dev);
2074         if (bdev == NULL)
2075                 return NULL;
2076
2077         if (bd_claim(bdev, sb)) {
2078                 printk(KERN_ERR
2079                         "EXT3: failed to claim external journal device.\n");
2080                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2081                 return NULL;
2082         }
2083
2084         blocksize = sb->s_blocksize;
2085         hblock = bdev_hardsect_size(bdev);
2086         if (blocksize < hblock) {
2087                 printk(KERN_ERR
2088                         "EXT3-fs: blocksize too small for journal device.\n");
2089                 goto out_bdev;
2090         }
2091
2092         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2093         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2094         set_blocksize(bdev, blocksize);
2095         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2096                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2097                        "external journal\n");
2098                 goto out_bdev;
2099         }
2100
2101         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2102         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2103             !(le32_to_cpu(es->s_feature_incompat) &
2104               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2105                 printk(KERN_ERR "EXT3-fs: external journal has "
2106                                         "bad superblock\n");
2107                 brelse(bh);
2108                 goto out_bdev;
2109         }
2110
2111         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2112                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2113                 brelse(bh);
2114                 goto out_bdev;
2115         }
2116
2117         len = le32_to_cpu(es->s_blocks_count);
2118         start = sb_block + 1;
2119         brelse(bh);     /* we're done with the superblock */
2120
2121         journal = journal_init_dev(bdev, sb->s_bdev,
2122                                         start, len, blocksize);
2123         if (!journal) {
2124                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2125                 goto out_bdev;
2126         }
2127         journal->j_private = sb;
2128         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2129         wait_on_buffer(journal->j_sb_buffer);
2130         if (!buffer_uptodate(journal->j_sb_buffer)) {
2131                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2132                 goto out_journal;
2133         }
2134         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2135                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2136                                         "user (unsupported) - %d\n",
2137                         be32_to_cpu(journal->j_superblock->s_nr_users));
2138                 goto out_journal;
2139         }
2140         EXT3_SB(sb)->journal_bdev = bdev;
2141         ext3_init_journal_params(sb, journal);
2142         return journal;
2143 out_journal:
2144         journal_destroy(journal);
2145 out_bdev:
2146         ext3_blkdev_put(bdev);
2147         return NULL;
2148 }
2149
2150 static int ext3_load_journal(struct super_block *sb,
2151                              struct ext3_super_block *es,
2152                              unsigned long journal_devnum)
2153 {
2154         journal_t *journal;
2155         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2156         dev_t journal_dev;
2157         int err = 0;
2158         int really_read_only;
2159
2160         if (journal_devnum &&
2161             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2162                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2163                         "numbers have changed\n");
2164                 journal_dev = new_decode_dev(journal_devnum);
2165         } else
2166                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2167
2168         really_read_only = bdev_read_only(sb->s_bdev);
2169
2170         /*
2171          * Are we loading a blank journal or performing recovery after a
2172          * crash?  For recovery, we need to check in advance whether we
2173          * can get read-write access to the device.
2174          */
2175
2176         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2177                 if (sb->s_flags & MS_RDONLY) {
2178                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2179                                         "required on readonly filesystem.\n");
2180                         if (really_read_only) {
2181                                 printk(KERN_ERR "EXT3-fs: write access "
2182                                         "unavailable, cannot proceed.\n");
2183                                 return -EROFS;
2184                         }
2185                         printk (KERN_INFO "EXT3-fs: write access will "
2186                                         "be enabled during recovery.\n");
2187                 }
2188         }
2189
2190         if (journal_inum && journal_dev) {
2191                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2192                        "and inode journals!\n");
2193                 return -EINVAL;
2194         }
2195
2196         if (journal_inum) {
2197                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2198                         return -EINVAL;
2199         } else {
2200                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2201                         return -EINVAL;
2202         }
2203
2204         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2205                 err = journal_update_format(journal);
2206                 if (err)  {
2207                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2208                         journal_destroy(journal);
2209                         return err;
2210                 }
2211         }
2212
2213         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2214                 err = journal_wipe(journal, !really_read_only);
2215         if (!err)
2216                 err = journal_load(journal);
2217
2218         if (err) {
2219                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2220                 journal_destroy(journal);
2221                 return err;
2222         }
2223
2224         EXT3_SB(sb)->s_journal = journal;
2225         ext3_clear_journal_err(sb, es);
2226
2227         if (journal_devnum &&
2228             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2229                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2230                 sb->s_dirt = 1;
2231
2232                 /* Make sure we flush the recovery flag to disk. */
2233                 ext3_commit_super(sb, es, 1);
2234         }
2235
2236         return 0;
2237 }
2238
2239 static int ext3_create_journal(struct super_block * sb,
2240                                struct ext3_super_block * es,
2241                                unsigned int journal_inum)
2242 {
2243         journal_t *journal;
2244         int err;
2245
2246         if (sb->s_flags & MS_RDONLY) {
2247                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2248                                 "create journal.\n");
2249                 return -EROFS;
2250         }
2251
2252         journal = ext3_get_journal(sb, journal_inum);
2253         if (!journal)
2254                 return -EINVAL;
2255
2256         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2257                journal_inum);
2258
2259         err = journal_create(journal);
2260         if (err) {
2261                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2262                 journal_destroy(journal);
2263                 return -EIO;
2264         }
2265
2266         EXT3_SB(sb)->s_journal = journal;
2267
2268         ext3_update_dynamic_rev(sb);
2269         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2270         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2271
2272         es->s_journal_inum = cpu_to_le32(journal_inum);
2273         sb->s_dirt = 1;
2274
2275         /* Make sure we flush the recovery flag to disk. */
2276         ext3_commit_super(sb, es, 1);
2277
2278         return 0;
2279 }
2280
2281 static void ext3_commit_super (struct super_block * sb,
2282                                struct ext3_super_block * es,
2283                                int sync)
2284 {
2285         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2286
2287         if (!sbh)
2288                 return;
2289         es->s_wtime = cpu_to_le32(get_seconds());
2290         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2291         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2292         BUFFER_TRACE(sbh, "marking dirty");
2293         mark_buffer_dirty(sbh);
2294         if (sync)
2295                 sync_dirty_buffer(sbh);
2296 }
2297
2298
2299 /*
2300  * Have we just finished recovery?  If so, and if we are mounting (or
2301  * remounting) the filesystem readonly, then we will end up with a
2302  * consistent fs on disk.  Record that fact.
2303  */
2304 static void ext3_mark_recovery_complete(struct super_block * sb,
2305                                         struct ext3_super_block * es)
2306 {
2307         journal_t *journal = EXT3_SB(sb)->s_journal;
2308
2309         journal_lock_updates(journal);
2310         if (journal_flush(journal) < 0)
2311                 goto out;
2312
2313         lock_super(sb);
2314         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2315             sb->s_flags & MS_RDONLY) {
2316                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2317                 sb->s_dirt = 0;
2318                 ext3_commit_super(sb, es, 1);
2319         }
2320         unlock_super(sb);
2321
2322 out:
2323         journal_unlock_updates(journal);
2324 }
2325
2326 /*
2327  * If we are mounting (or read-write remounting) a filesystem whose journal
2328  * has recorded an error from a previous lifetime, move that error to the
2329  * main filesystem now.
2330  */
2331 static void ext3_clear_journal_err(struct super_block * sb,
2332                                    struct ext3_super_block * es)
2333 {
2334         journal_t *journal;
2335         int j_errno;
2336         const char *errstr;
2337
2338         journal = EXT3_SB(sb)->s_journal;
2339
2340         /*
2341          * Now check for any error status which may have been recorded in the
2342          * journal by a prior ext3_error() or ext3_abort()
2343          */
2344
2345         j_errno = journal_errno(journal);
2346         if (j_errno) {
2347                 char nbuf[16];
2348
2349                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2350                 ext3_warning(sb, __func__, "Filesystem error recorded "
2351                              "from previous mount: %s", errstr);
2352                 ext3_warning(sb, __func__, "Marking fs in need of "
2353                              "filesystem check.");
2354
2355                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2356                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2357                 ext3_commit_super (sb, es, 1);
2358
2359                 journal_clear_err(journal);
2360         }
2361 }
2362
2363 /*
2364  * Force the running and committing transactions to commit,
2365  * and wait on the commit.
2366  */
2367 int ext3_force_commit(struct super_block *sb)
2368 {
2369         journal_t *journal;
2370         int ret;
2371
2372         if (sb->s_flags & MS_RDONLY)
2373                 return 0;
2374
2375         journal = EXT3_SB(sb)->s_journal;
2376         sb->s_dirt = 0;
2377         ret = ext3_journal_force_commit(journal);
2378         return ret;
2379 }
2380
2381 /*
2382  * Ext3 always journals updates to the superblock itself, so we don't
2383  * have to propagate any other updates to the superblock on disk at this
2384  * point.  (We can probably nuke this function altogether, and remove
2385  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2386  */
2387 static void ext3_write_super (struct super_block * sb)
2388 {
2389         if (mutex_trylock(&sb->s_lock) != 0)
2390                 BUG();
2391         sb->s_dirt = 0;
2392 }
2393
2394 static int ext3_sync_fs(struct super_block *sb, int wait)
2395 {
2396         sb->s_dirt = 0;
2397         if (wait)
2398                 ext3_force_commit(sb);
2399         else
2400                 journal_start_commit(EXT3_SB(sb)->s_journal, NULL);
2401
2402         return 0;
2403 }
2404
2405 /*
2406  * LVM calls this function before a (read-only) snapshot is created.  This
2407  * gives us a chance to flush the journal completely and mark the fs clean.
2408  */
2409 static void ext3_write_super_lockfs(struct super_block *sb)
2410 {
2411         sb->s_dirt = 0;
2412
2413         if (!(sb->s_flags & MS_RDONLY)) {
2414                 journal_t *journal = EXT3_SB(sb)->s_journal;
2415
2416                 /* Now we set up the journal barrier. */
2417                 journal_lock_updates(journal);
2418
2419                 /*
2420                  * We don't want to clear needs_recovery flag when we failed
2421                  * to flush the journal.
2422                  */
2423                 if (journal_flush(journal) < 0)
2424                         return;
2425
2426                 /* Journal blocked and flushed, clear needs_recovery flag. */
2427                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2428                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2429         }
2430 }
2431
2432 /*
2433  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2434  * flag here, even though the filesystem is not technically dirty yet.
2435  */
2436 static void ext3_unlockfs(struct super_block *sb)
2437 {
2438         if (!(sb->s_flags & MS_RDONLY)) {
2439                 lock_super(sb);
2440                 /* Reser the needs_recovery flag before the fs is unlocked. */
2441                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2442                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2443                 unlock_super(sb);
2444                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2445         }
2446 }
2447
2448 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2449 {
2450         struct ext3_super_block * es;
2451         struct ext3_sb_info *sbi = EXT3_SB(sb);
2452         ext3_fsblk_t n_blocks_count = 0;
2453         unsigned long old_sb_flags;
2454         struct ext3_mount_options old_opts;
2455         int err;
2456 #ifdef CONFIG_QUOTA
2457         int i;
2458 #endif
2459
2460         /* Store the original options */
2461         old_sb_flags = sb->s_flags;
2462         old_opts.s_mount_opt = sbi->s_mount_opt;
2463         old_opts.s_resuid = sbi->s_resuid;
2464         old_opts.s_resgid = sbi->s_resgid;
2465         old_opts.s_commit_interval = sbi->s_commit_interval;
2466 #ifdef CONFIG_QUOTA
2467         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2468         for (i = 0; i < MAXQUOTAS; i++)
2469                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2470 #endif
2471
2472         /*
2473          * Allow the "check" option to be passed as a remount option.
2474          */
2475         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2476                 err = -EINVAL;
2477                 goto restore_opts;
2478         }
2479
2480         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2481                 ext3_abort(sb, __func__, "Abort forced by user");
2482
2483         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2484                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2485
2486         es = sbi->s_es;
2487
2488         ext3_init_journal_params(sb, sbi->s_journal);
2489
2490         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2491                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2492                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2493                         err = -EROFS;
2494                         goto restore_opts;
2495                 }
2496
2497                 if (*flags & MS_RDONLY) {
2498                         /*
2499                          * First of all, the unconditional stuff we have to do
2500                          * to disable replay of the journal when we next remount
2501                          */
2502                         sb->s_flags |= MS_RDONLY;
2503
2504                         /*
2505                          * OK, test if we are remounting a valid rw partition
2506                          * readonly, and if so set the rdonly flag and then
2507                          * mark the partition as valid again.
2508                          */
2509                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2510                             (sbi->s_mount_state & EXT3_VALID_FS))
2511                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2512
2513                         /*
2514                          * We have to unlock super so that we can wait for
2515                          * transactions.
2516                          */
2517                         unlock_super(sb);
2518                         ext3_mark_recovery_complete(sb, es);
2519                         lock_super(sb);
2520                 } else {
2521                         __le32 ret;
2522                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2523                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2524                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2525                                        "remount RDWR because of unsupported "
2526                                        "optional features (%x).\n",
2527                                        sb->s_id, le32_to_cpu(ret));
2528                                 err = -EROFS;
2529                                 goto restore_opts;
2530                         }
2531
2532                         /*
2533                          * If we have an unprocessed orphan list hanging
2534                          * around from a previously readonly bdev mount,
2535                          * require a full umount/remount for now.
2536                          */
2537                         if (es->s_last_orphan) {
2538                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2539                                        "remount RDWR because of unprocessed "
2540                                        "orphan inode list.  Please "
2541                                        "umount/remount instead.\n",
2542                                        sb->s_id);
2543                                 err = -EINVAL;
2544                                 goto restore_opts;
2545                         }
2546
2547                         /*
2548                          * Mounting a RDONLY partition read-write, so reread
2549                          * and store the current valid flag.  (It may have
2550                          * been changed by e2fsck since we originally mounted
2551                          * the partition.)
2552                          */
2553                         ext3_clear_journal_err(sb, es);
2554                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2555                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2556                                 goto restore_opts;
2557                         if (!ext3_setup_super (sb, es, 0))
2558                                 sb->s_flags &= ~MS_RDONLY;
2559                 }
2560         }
2561 #ifdef CONFIG_QUOTA
2562         /* Release old quota file names */
2563         for (i = 0; i < MAXQUOTAS; i++)
2564                 if (old_opts.s_qf_names[i] &&
2565                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2566                         kfree(old_opts.s_qf_names[i]);
2567 #endif
2568         return 0;
2569 restore_opts:
2570         sb->s_flags = old_sb_flags;
2571         sbi->s_mount_opt = old_opts.s_mount_opt;
2572         sbi->s_resuid = old_opts.s_resuid;
2573         sbi->s_resgid = old_opts.s_resgid;
2574         sbi->s_commit_interval = old_opts.s_commit_interval;
2575 #ifdef CONFIG_QUOTA
2576         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2577         for (i = 0; i < MAXQUOTAS; i++) {
2578                 if (sbi->s_qf_names[i] &&
2579                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2580                         kfree(sbi->s_qf_names[i]);
2581                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2582         }
2583 #endif
2584         return err;
2585 }
2586
2587 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2588 {
2589         struct super_block *sb = dentry->d_sb;
2590         struct ext3_sb_info *sbi = EXT3_SB(sb);
2591         struct ext3_super_block *es = sbi->s_es;
2592         u64 fsid;
2593
2594         if (test_opt(sb, MINIX_DF)) {
2595                 sbi->s_overhead_last = 0;
2596         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2597                 unsigned long ngroups = sbi->s_groups_count, i;
2598                 ext3_fsblk_t overhead = 0;
2599                 smp_rmb();
2600
2601                 /*
2602                  * Compute the overhead (FS structures).  This is constant
2603                  * for a given filesystem unless the number of block groups
2604                  * changes so we cache the previous value until it does.
2605                  */
2606
2607                 /*
2608                  * All of the blocks before first_data_block are
2609                  * overhead
2610                  */
2611                 overhead = le32_to_cpu(es->s_first_data_block);
2612
2613                 /*
2614                  * Add the overhead attributed to the superblock and
2615                  * block group descriptors.  If the sparse superblocks
2616                  * feature is turned on, then not all groups have this.
2617                  */
2618                 for (i = 0; i < ngroups; i++) {
2619                         overhead += ext3_bg_has_super(sb, i) +
2620                                 ext3_bg_num_gdb(sb, i);
2621                         cond_resched();
2622                 }
2623
2624                 /*
2625                  * Every block group has an inode bitmap, a block
2626                  * bitmap, and an inode table.
2627                  */
2628                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2629                 sbi->s_overhead_last = overhead;
2630                 smp_wmb();
2631                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2632         }
2633
2634         buf->f_type = EXT3_SUPER_MAGIC;
2635         buf->f_bsize = sb->s_blocksize;
2636         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2637         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2638         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2639         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2640         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2641                 buf->f_bavail = 0;
2642         buf->f_files = le32_to_cpu(es->s_inodes_count);
2643         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2644         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2645         buf->f_namelen = EXT3_NAME_LEN;
2646         fsid = le64_to_cpup((void *)es->s_uuid) ^
2647                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2648         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2649         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2650         return 0;
2651 }
2652
2653 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2654  * is locked for write. Otherwise the are possible deadlocks:
2655  * Process 1                         Process 2
2656  * ext3_create()                     quota_sync()
2657  *   journal_start()                   write_dquot()
2658  *   DQUOT_INIT()                        down(dqio_mutex)
2659  *     down(dqio_mutex)                    journal_start()
2660  *
2661  */
2662
2663 #ifdef CONFIG_QUOTA
2664
2665 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2666 {
2667         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2668 }
2669
2670 static int ext3_dquot_initialize(struct inode *inode, int type)
2671 {
2672         handle_t *handle;
2673         int ret, err;
2674
2675         /* We may create quota structure so we need to reserve enough blocks */
2676         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2677         if (IS_ERR(handle))
2678                 return PTR_ERR(handle);
2679         ret = dquot_initialize(inode, type);
2680         err = ext3_journal_stop(handle);
2681         if (!ret)
2682                 ret = err;
2683         return ret;
2684 }
2685
2686 static int ext3_dquot_drop(struct inode *inode)
2687 {
2688         handle_t *handle;
2689         int ret, err;
2690
2691         /* We may delete quota structure so we need to reserve enough blocks */
2692         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2693         if (IS_ERR(handle)) {
2694                 /*
2695                  * We call dquot_drop() anyway to at least release references
2696                  * to quota structures so that umount does not hang.
2697                  */
2698                 dquot_drop(inode);
2699                 return PTR_ERR(handle);
2700         }
2701         ret = dquot_drop(inode);
2702         err = ext3_journal_stop(handle);
2703         if (!ret)
2704                 ret = err;
2705         return ret;
2706 }
2707
2708 static int ext3_write_dquot(struct dquot *dquot)
2709 {
2710         int ret, err;
2711         handle_t *handle;
2712         struct inode *inode;
2713
2714         inode = dquot_to_inode(dquot);
2715         handle = ext3_journal_start(inode,
2716                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2717         if (IS_ERR(handle))
2718                 return PTR_ERR(handle);
2719         ret = dquot_commit(dquot);
2720         err = ext3_journal_stop(handle);
2721         if (!ret)
2722                 ret = err;
2723         return ret;
2724 }
2725
2726 static int ext3_acquire_dquot(struct dquot *dquot)
2727 {
2728         int ret, err;
2729         handle_t *handle;
2730
2731         handle = ext3_journal_start(dquot_to_inode(dquot),
2732                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2733         if (IS_ERR(handle))
2734                 return PTR_ERR(handle);
2735         ret = dquot_acquire(dquot);
2736         err = ext3_journal_stop(handle);
2737         if (!ret)
2738                 ret = err;
2739         return ret;
2740 }
2741
2742 static int ext3_release_dquot(struct dquot *dquot)
2743 {
2744         int ret, err;
2745         handle_t *handle;
2746
2747         handle = ext3_journal_start(dquot_to_inode(dquot),
2748                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2749         if (IS_ERR(handle)) {
2750                 /* Release dquot anyway to avoid endless cycle in dqput() */
2751                 dquot_release(dquot);
2752                 return PTR_ERR(handle);
2753         }
2754         ret = dquot_release(dquot);
2755         err = ext3_journal_stop(handle);
2756         if (!ret)
2757                 ret = err;
2758         return ret;
2759 }
2760
2761 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2762 {
2763         /* Are we journaling quotas? */
2764         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2765             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2766                 dquot_mark_dquot_dirty(dquot);
2767                 return ext3_write_dquot(dquot);
2768         } else {
2769                 return dquot_mark_dquot_dirty(dquot);
2770         }
2771 }
2772
2773 static int ext3_write_info(struct super_block *sb, int type)
2774 {
2775         int ret, err;
2776         handle_t *handle;
2777
2778         /* Data block + inode block */
2779         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2780         if (IS_ERR(handle))
2781                 return PTR_ERR(handle);
2782         ret = dquot_commit_info(sb, type);
2783         err = ext3_journal_stop(handle);
2784         if (!ret)
2785                 ret = err;
2786         return ret;
2787 }
2788
2789 /*
2790  * Turn on quotas during mount time - we need to find
2791  * the quota file and such...
2792  */
2793 static int ext3_quota_on_mount(struct super_block *sb, int type)
2794 {
2795         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2796                         EXT3_SB(sb)->s_jquota_fmt, type);
2797 }
2798
2799 /*
2800  * Standard function to be called on quota_on
2801  */
2802 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2803                          char *name, int remount)
2804 {
2805         int err;
2806         struct path path;
2807
2808         if (!test_opt(sb, QUOTA))
2809                 return -EINVAL;
2810         /* When remounting, no checks are needed and in fact, name is NULL */
2811         if (remount)
2812                 return vfs_quota_on(sb, type, format_id, name, remount);
2813
2814         err = kern_path(name, LOOKUP_FOLLOW, &path);
2815         if (err)
2816                 return err;
2817
2818         /* Quotafile not on the same filesystem? */
2819         if (path.mnt->mnt_sb != sb) {
2820                 path_put(&path);
2821                 return -EXDEV;
2822         }
2823         /* Journaling quota? */
2824         if (EXT3_SB(sb)->s_qf_names[type]) {
2825                 /* Quotafile not of fs root? */
2826                 if (path.dentry->d_parent != sb->s_root)
2827                         printk(KERN_WARNING
2828                                 "EXT3-fs: Quota file not on filesystem root. "
2829                                 "Journaled quota will not work.\n");
2830         }
2831
2832         /*
2833          * When we journal data on quota file, we have to flush journal to see
2834          * all updates to the file when we bypass pagecache...
2835          */
2836         if (ext3_should_journal_data(path.dentry->d_inode)) {
2837                 /*
2838                  * We don't need to lock updates but journal_flush() could
2839                  * otherwise be livelocked...
2840                  */
2841                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2842                 err = journal_flush(EXT3_SB(sb)->s_journal);
2843                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2844                 if (err) {
2845                         path_put(&path);
2846                         return err;
2847                 }
2848         }
2849
2850         err = vfs_quota_on_path(sb, type, format_id, &path);
2851         path_put(&path);
2852         return err;
2853 }
2854
2855 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2856  * acquiring the locks... As quota files are never truncated and quota code
2857  * itself serializes the operations (and noone else should touch the files)
2858  * we don't have to be afraid of races */
2859 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2860                                size_t len, loff_t off)
2861 {
2862         struct inode *inode = sb_dqopt(sb)->files[type];
2863         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2864         int err = 0;
2865         int offset = off & (sb->s_blocksize - 1);
2866         int tocopy;
2867         size_t toread;
2868         struct buffer_head *bh;
2869         loff_t i_size = i_size_read(inode);
2870
2871         if (off > i_size)
2872                 return 0;
2873         if (off+len > i_size)
2874                 len = i_size-off;
2875         toread = len;
2876         while (toread > 0) {
2877                 tocopy = sb->s_blocksize - offset < toread ?
2878                                 sb->s_blocksize - offset : toread;
2879                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2880                 if (err)
2881                         return err;
2882                 if (!bh)        /* A hole? */
2883                         memset(data, 0, tocopy);
2884                 else
2885                         memcpy(data, bh->b_data+offset, tocopy);
2886                 brelse(bh);
2887                 offset = 0;
2888                 toread -= tocopy;
2889                 data += tocopy;
2890                 blk++;
2891         }
2892         return len;
2893 }
2894
2895 /* Write to quotafile (we know the transaction is already started and has
2896  * enough credits) */
2897 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2898                                 const char *data, size_t len, loff_t off)
2899 {
2900         struct inode *inode = sb_dqopt(sb)->files[type];
2901         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2902         int err = 0;
2903         int offset = off & (sb->s_blocksize - 1);
2904         int tocopy;
2905         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2906         size_t towrite = len;
2907         struct buffer_head *bh;
2908         handle_t *handle = journal_current_handle();
2909
2910         if (!handle) {
2911                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2912                         " cancelled because transaction is not started.\n",
2913                         (unsigned long long)off, (unsigned long long)len);
2914                 return -EIO;
2915         }
2916         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2917         while (towrite > 0) {
2918                 tocopy = sb->s_blocksize - offset < towrite ?
2919                                 sb->s_blocksize - offset : towrite;
2920                 bh = ext3_bread(handle, inode, blk, 1, &err);
2921                 if (!bh)
2922                         goto out;
2923                 if (journal_quota) {
2924                         err = ext3_journal_get_write_access(handle, bh);
2925                         if (err) {
2926                                 brelse(bh);
2927                                 goto out;
2928                         }
2929                 }
2930                 lock_buffer(bh);
2931                 memcpy(bh->b_data+offset, data, tocopy);
2932                 flush_dcache_page(bh->b_page);
2933                 unlock_buffer(bh);
2934                 if (journal_quota)
2935                         err = ext3_journal_dirty_metadata(handle, bh);
2936                 else {
2937                         /* Always do at least ordered writes for quotas */
2938                         err = ext3_journal_dirty_data(handle, bh);
2939                         mark_buffer_dirty(bh);
2940                 }
2941                 brelse(bh);
2942                 if (err)
2943                         goto out;
2944                 offset = 0;
2945                 towrite -= tocopy;
2946                 data += tocopy;
2947                 blk++;
2948         }
2949 out:
2950         if (len == towrite) {
2951                 mutex_unlock(&inode->i_mutex);
2952                 return err;
2953         }
2954         if (inode->i_size < off+len-towrite) {
2955                 i_size_write(inode, off+len-towrite);
2956                 EXT3_I(inode)->i_disksize = inode->i_size;
2957         }
2958         inode->i_version++;
2959         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2960         ext3_mark_inode_dirty(handle, inode);
2961         mutex_unlock(&inode->i_mutex);
2962         return len - towrite;
2963 }
2964
2965 #endif
2966
2967 static int ext3_get_sb(struct file_system_type *fs_type,
2968         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2969 {
2970         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2971 }
2972
2973 static struct file_system_type ext3_fs_type = {
2974         .owner          = THIS_MODULE,
2975         .name           = "ext3",
2976         .get_sb         = ext3_get_sb,
2977         .kill_sb        = kill_block_super,
2978         .fs_flags       = FS_REQUIRES_DEV,
2979 };
2980
2981 static int __init init_ext3_fs(void)
2982 {
2983         int err = init_ext3_xattr();
2984         if (err)
2985                 return err;
2986         err = init_inodecache();
2987         if (err)
2988                 goto out1;
2989         err = register_filesystem(&ext3_fs_type);
2990         if (err)
2991                 goto out;
2992         return 0;
2993 out:
2994         destroy_inodecache();
2995 out1:
2996         exit_ext3_xattr();
2997         return err;
2998 }
2999
3000 static void __exit exit_ext3_fs(void)
3001 {
3002         unregister_filesystem(&ext3_fs_type);
3003         destroy_inodecache();
3004         exit_ext3_xattr();
3005 }
3006
3007 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3008 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3009 MODULE_LICENSE("GPL");
3010 module_init(init_ext3_fs)
3011 module_exit(exit_ext3_fs)