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 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  49 #else
  50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  51 #endif
  52
  53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  54                              unsigned long journal_devnum);
  55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  56                                unsigned int);
  57 static int ext3_commit_super(struct super_block *sb,
  58                                struct ext3_super_block *es,
  59                                int sync);
  60 static void ext3_mark_recovery_complete(struct super_block * sb,
  61                                         struct ext3_super_block * es);
  62 static void ext3_clear_journal_err(struct super_block * sb,
  63                                    struct ext3_super_block * es);
  64 static int ext3_sync_fs(struct super_block *sb, int wait);
  65 static const char *ext3_decode_error(struct super_block * sb, int errno,
  66                                      char nbuf[16]);
  67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  69 static int ext3_unfreeze(struct super_block *sb);
  70 static int ext3_freeze(struct super_block *sb);
  71
  72 /*
  73  * Wrappers for journal_start/end.
  74  *
  75  * The only special thing we need to do here is to make sure that all
  76  * journal_end calls result in the superblock being marked dirty, so
  77  * that sync() will call the filesystem's write_super callback if
  78  * appropriate.
  79  */
  80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  81 {
  82         journal_t *journal;
  83
  84         if (sb->s_flags & MS_RDONLY)
  85                 return ERR_PTR(-EROFS);
  86
  87         /* Special case here: if the journal has aborted behind our
  88          * backs (eg. EIO in the commit thread), then we still need to
  89          * take the FS itself readonly cleanly. */
  90         journal = EXT3_SB(sb)->s_journal;
  91         if (is_journal_aborted(journal)) {
  92                 ext3_abort(sb, __func__,
  93                            "Detected aborted journal");
  94                 return ERR_PTR(-EROFS);
  95         }
  96
  97         return journal_start(journal, nblocks);
  98 }
  99
 100 /*
 101  * The only special thing we need to do here is to make sure that all
 102  * journal_stop calls result in the superblock being marked dirty, so
 103  * that sync() will call the filesystem's write_super callback if
 104  * appropriate.
 105  */
 106 int __ext3_journal_stop(const char *where, handle_t *handle)
 107 {
 108         struct super_block *sb;
 109         int err;
 110         int rc;
 111
 112         sb = handle->h_transaction->t_journal->j_private;
 113         err = handle->h_err;
 114         rc = journal_stop(handle);
 115
 116         if (!err)
 117                 err = rc;
 118         if (err)
 119                 __ext3_std_error(sb, where, err);
 120         return err;
 121 }
 122
 123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 124                 struct buffer_head *bh, handle_t *handle, int err)
 125 {
 126         char nbuf[16];
 127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 128
 129         if (bh)
 130                 BUFFER_TRACE(bh, "abort");
 131
 132         if (!handle->h_err)
 133                 handle->h_err = err;
 134
 135         if (is_handle_aborted(handle))
 136                 return;
 137
 138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 139                 caller, errstr, err_fn);
 140
 141         journal_abort_handle(handle);
 142 }
 143
 144 void ext3_msg(struct super_block *sb, const char *prefix,
 145                 const char *fmt, ...)
 146 {
 147         va_list args;
 148
 149         va_start(args, fmt);
 150         printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
 151         vprintk(fmt, args);
 152         printk("\n");
 153         va_end(args);
 154 }
 155
 156 /* Deal with the reporting of failure conditions on a filesystem such as
 157  * inconsistencies detected or read IO failures.
 158  *
 159  * On ext2, we can store the error state of the filesystem in the
 160  * superblock.  That is not possible on ext3, because we may have other
 161  * write ordering constraints on the superblock which prevent us from
 162  * writing it out straight away; and given that the journal is about to
 163  * be aborted, we can't rely on the current, or future, transactions to
 164  * write out the superblock safely.
 165  *
 166  * We'll just use the journal_abort() error code to record an error in
 167  * the journal instead.  On recovery, the journal will complain about
 168  * that error until we've noted it down and cleared it.
 169  */
 170
 171 static void ext3_handle_error(struct super_block *sb)
 172 {
 173         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 174
 175         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 176         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 177
 178         if (sb->s_flags & MS_RDONLY)
 179                 return;
 180
 181         if (!test_opt (sb, ERRORS_CONT)) {
 182                 journal_t *journal = EXT3_SB(sb)->s_journal;
 183
 184                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 185                 if (journal)
 186                         journal_abort(journal, -EIO);
 187         }
 188         if (test_opt (sb, ERRORS_RO)) {
 189                 ext3_msg(sb, KERN_CRIT,
 190                         "error: remounting filesystem read-only");
 191                 sb->s_flags |= MS_RDONLY;
 192         }
 193         ext3_commit_super(sb, es, 1);
 194         if (test_opt(sb, ERRORS_PANIC))
 195                 panic("EXT3-fs (%s): panic forced after error\n",
 196                         sb->s_id);
 197 }
 198
 199 void ext3_error (struct super_block * sb, const char * function,
 200                  const char * fmt, ...)
 201 {
 202         va_list args;
 203
 204         va_start(args, fmt);
 205         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 206         vprintk(fmt, args);
 207         printk("\n");
 208         va_end(args);
 209
 210         ext3_handle_error(sb);
 211 }
 212
 213 static const char *ext3_decode_error(struct super_block * sb, int errno,
 214                                      char nbuf[16])
 215 {
 216         char *errstr = NULL;
 217
 218         switch (errno) {
 219         case -EIO:
 220                 errstr = "IO failure";
 221                 break;
 222         case -ENOMEM:
 223                 errstr = "Out of memory";
 224                 break;
 225         case -EROFS:
 226                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 227                         errstr = "Journal has aborted";
 228                 else
 229                         errstr = "Readonly filesystem";
 230                 break;
 231         default:
 232                 /* If the caller passed in an extra buffer for unknown
 233                  * errors, textualise them now.  Else we just return
 234                  * NULL. */
 235                 if (nbuf) {
 236                         /* Check for truncated error codes... */
 237                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 238                                 errstr = nbuf;
 239                 }
 240                 break;
 241         }
 242
 243         return errstr;
 244 }
 245
 246 /* __ext3_std_error decodes expected errors from journaling functions
 247  * automatically and invokes the appropriate error response.  */
 248
 249 void __ext3_std_error (struct super_block * sb, const char * function,
 250                        int errno)
 251 {
 252         char nbuf[16];
 253         const char *errstr;
 254
 255         /* Special case: if the error is EROFS, and we're not already
 256          * inside a transaction, then there's really no point in logging
 257          * an error. */
 258         if (errno == -EROFS && journal_current_handle() == NULL &&
 259             (sb->s_flags & MS_RDONLY))
 260                 return;
 261
 262         errstr = ext3_decode_error(sb, errno, nbuf);
 263         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 264
 265         ext3_handle_error(sb);
 266 }
 267
 268 /*
 269  * ext3_abort is a much stronger failure handler than ext3_error.  The
 270  * abort function may be used to deal with unrecoverable failures such
 271  * as journal IO errors or ENOMEM at a critical moment in log management.
 272  *
 273  * We unconditionally force the filesystem into an ABORT|READONLY state,
 274  * unless the error response on the fs has been set to panic in which
 275  * case we take the easy way out and panic immediately.
 276  */
 277
 278 void ext3_abort (struct super_block * sb, const char * function,
 279                  const char * fmt, ...)
 280 {
 281         va_list args;
 282
 283         va_start(args, fmt);
 284         printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
 285         vprintk(fmt, args);
 286         printk("\n");
 287         va_end(args);
 288
 289         if (test_opt(sb, ERRORS_PANIC))
 290                 panic("EXT3-fs: panic from previous error\n");
 291
 292         if (sb->s_flags & MS_RDONLY)
 293                 return;
 294
 295         ext3_msg(sb, KERN_CRIT,
 296                 "error: remounting filesystem read-only");
 297         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 298         sb->s_flags |= MS_RDONLY;
 299         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 300         if (EXT3_SB(sb)->s_journal)
 301                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 302 }
 303
 304 void ext3_warning (struct super_block * sb, const char * function,
 305                    const char * fmt, ...)
 306 {
 307         va_list args;
 308
 309         va_start(args, fmt);
 310         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
 311                sb->s_id, function);
 312         vprintk(fmt, args);
 313         printk("\n");
 314         va_end(args);
 315 }
 316
 317 void ext3_update_dynamic_rev(struct super_block *sb)
 318 {
 319         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 320
 321         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 322                 return;
 323
 324         ext3_msg(sb, KERN_WARNING,
 325                 "warning: updating to rev %d because of "
 326                 "new feature flag, running e2fsck is recommended",
 327                 EXT3_DYNAMIC_REV);
 328
 329         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 330         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 331         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 332         /* leave es->s_feature_*compat flags alone */
 333         /* es->s_uuid will be set by e2fsck if empty */
 334
 335         /*
 336          * The rest of the superblock fields should be zero, and if not it
 337          * means they are likely already in use, so leave them alone.  We
 338          * can leave it up to e2fsck to clean up any inconsistencies there.
 339          */
 340 }
 341
 342 /*
 343  * Open the external journal device
 344  */
 345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 346 {
 347         struct block_device *bdev;
 348         char b[BDEVNAME_SIZE];
 349
 350         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 351         if (IS_ERR(bdev))
 352                 goto fail;
 353         return bdev;
 354
 355 fail:
 356         ext3_msg(sb, "error: failed to open journal device %s: %ld",
 357                 __bdevname(dev, b), PTR_ERR(bdev));
 358
 359         return NULL;
 360 }
 361
 362 /*
 363  * Release the journal device
 364  */
 365 static int ext3_blkdev_put(struct block_device *bdev)
 366 {
 367         bd_release(bdev);
 368         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 369 }
 370
 371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 372 {
 373         struct block_device *bdev;
 374         int ret = -ENODEV;
 375
 376         bdev = sbi->journal_bdev;
 377         if (bdev) {
 378                 ret = ext3_blkdev_put(bdev);
 379                 sbi->journal_bdev = NULL;
 380         }
 381         return ret;
 382 }
 383
 384 static inline struct inode *orphan_list_entry(struct list_head *l)
 385 {
 386         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 387 }
 388
 389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 390 {
 391         struct list_head *l;
 392
 393         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 394                le32_to_cpu(sbi->s_es->s_last_orphan));
 395
 396         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 397         list_for_each(l, &sbi->s_orphan) {
 398                 struct inode *inode = orphan_list_entry(l);
 399                 ext3_msg(sb, KERN_ERR, "  "
 400                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 401                        inode->i_sb->s_id, inode->i_ino, inode,
 402                        inode->i_mode, inode->i_nlink,
 403                        NEXT_ORPHAN(inode));
 404         }
 405 }
 406
 407 static void ext3_put_super (struct super_block * sb)
 408 {
 409         struct ext3_sb_info *sbi = EXT3_SB(sb);
 410         struct ext3_super_block *es = sbi->s_es;
 411         int i, err;
 412
 413         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 414
 415         lock_kernel();
 416
 417         ext3_xattr_put_super(sb);
 418         err = journal_destroy(sbi->s_journal);
 419         sbi->s_journal = NULL;
 420         if (err < 0)
 421                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 422
 423         if (!(sb->s_flags & MS_RDONLY)) {
 424                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 425                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 426                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 427                 mark_buffer_dirty(sbi->s_sbh);
 428                 ext3_commit_super(sb, es, 1);
 429         }
 430
 431         for (i = 0; i < sbi->s_gdb_count; i++)
 432                 brelse(sbi->s_group_desc[i]);
 433         kfree(sbi->s_group_desc);
 434         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 435         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 436         percpu_counter_destroy(&sbi->s_dirs_counter);
 437         brelse(sbi->s_sbh);
 438 #ifdef CONFIG_QUOTA
 439         for (i = 0; i < MAXQUOTAS; i++)
 440                 kfree(sbi->s_qf_names[i]);
 441 #endif
 442
 443         /* Debugging code just in case the in-memory inode orphan list
 444          * isn't empty.  The on-disk one can be non-empty if we've
 445          * detected an error and taken the fs readonly, but the
 446          * in-memory list had better be clean by this point. */
 447         if (!list_empty(&sbi->s_orphan))
 448                 dump_orphan_list(sb, sbi);
 449         J_ASSERT(list_empty(&sbi->s_orphan));
 450
 451         invalidate_bdev(sb->s_bdev);
 452         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 453                 /*
 454                  * Invalidate the journal device's buffers.  We don't want them
 455                  * floating about in memory - the physical journal device may
 456                  * hotswapped, and it breaks the `ro-after' testing code.
 457                  */
 458                 sync_blockdev(sbi->journal_bdev);
 459                 invalidate_bdev(sbi->journal_bdev);
 460                 ext3_blkdev_remove(sbi);
 461         }
 462         sb->s_fs_info = NULL;
 463         kfree(sbi->s_blockgroup_lock);
 464         kfree(sbi);
 465
 466         unlock_kernel();
 467 }
 468
 469 static struct kmem_cache *ext3_inode_cachep;
 470
 471 /*
 472  * Called inside transaction, so use GFP_NOFS
 473  */
 474 static struct inode *ext3_alloc_inode(struct super_block *sb)
 475 {
 476         struct ext3_inode_info *ei;
 477
 478         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 479         if (!ei)
 480                 return NULL;
 481         ei->i_block_alloc_info = NULL;
 482         ei->vfs_inode.i_version = 1;
 483         atomic_set(&ei->i_datasync_tid, 0);
 484         atomic_set(&ei->i_sync_tid, 0);
 485         return &ei->vfs_inode;
 486 }
 487
 488 static void ext3_destroy_inode(struct inode *inode)
 489 {
 490         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 491                 printk("EXT3 Inode %p: orphan list check failed!\n",
 492                         EXT3_I(inode));
 493                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 494                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 495                                 false);
 496                 dump_stack();
 497         }
 498         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 499 }
 500
 501 static void init_once(void *foo)
 502 {
 503         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 504
 505         INIT_LIST_HEAD(&ei->i_orphan);
 506 #ifdef CONFIG_EXT3_FS_XATTR
 507         init_rwsem(&ei->xattr_sem);
 508 #endif
 509         mutex_init(&ei->truncate_mutex);
 510         inode_init_once(&ei->vfs_inode);
 511 }
 512
 513 static int init_inodecache(void)
 514 {
 515         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 516                                              sizeof(struct ext3_inode_info),
 517                                              0, (SLAB_RECLAIM_ACCOUNT|
 518                                                 SLAB_MEM_SPREAD),
 519                                              init_once);
 520         if (ext3_inode_cachep == NULL)
 521                 return -ENOMEM;
 522         return 0;
 523 }
 524
 525 static void destroy_inodecache(void)
 526 {
 527         kmem_cache_destroy(ext3_inode_cachep);
 528 }
 529
 530 static void ext3_clear_inode(struct inode *inode)
 531 {
 532         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 533
 534         dquot_drop(inode);
 535         ext3_discard_reservation(inode);
 536         EXT3_I(inode)->i_block_alloc_info = NULL;
 537         if (unlikely(rsv))
 538                 kfree(rsv);
 539 }
 540
 541 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 542 {
 543 #if defined(CONFIG_QUOTA)
 544         struct ext3_sb_info *sbi = EXT3_SB(sb);
 545
 546         if (sbi->s_jquota_fmt) {
 547                 char *fmtname = "";
 548
 549                 switch (sbi->s_jquota_fmt) {
 550                 case QFMT_VFS_OLD:
 551                         fmtname = "vfsold";
 552                         break;
 553                 case QFMT_VFS_V0:
 554                         fmtname = "vfsv0";
 555                         break;
 556                 case QFMT_VFS_V1:
 557                         fmtname = "vfsv1";
 558                         break;
 559                 }
 560                 seq_printf(seq, ",jqfmt=%s", fmtname);
 561         }
 562
 563         if (sbi->s_qf_names[USRQUOTA])
 564                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 565
 566         if (sbi->s_qf_names[GRPQUOTA])
 567                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 568
 569         if (test_opt(sb, USRQUOTA))
 570                 seq_puts(seq, ",usrquota");
 571
 572         if (test_opt(sb, GRPQUOTA))
 573                 seq_puts(seq, ",grpquota");
 574 #endif
 575 }
 576
 577 static char *data_mode_string(unsigned long mode)
 578 {
 579         switch (mode) {
 580         case EXT3_MOUNT_JOURNAL_DATA:
 581                 return "journal";
 582         case EXT3_MOUNT_ORDERED_DATA:
 583                 return "ordered";
 584         case EXT3_MOUNT_WRITEBACK_DATA:
 585                 return "writeback";
 586         }
 587         return "unknown";
 588 }
 589
 590 /*
 591  * Show an option if
 592  *  - it's set to a non-default value OR
 593  *  - if the per-sb default is different from the global default
 594  */
 595 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 596 {
 597         struct super_block *sb = vfs->mnt_sb;
 598         struct ext3_sb_info *sbi = EXT3_SB(sb);
 599         struct ext3_super_block *es = sbi->s_es;
 600         unsigned long def_mount_opts;
 601
 602         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 603
 604         if (sbi->s_sb_block != 1)
 605                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 606         if (test_opt(sb, MINIX_DF))
 607                 seq_puts(seq, ",minixdf");
 608         if (test_opt(sb, GRPID))
 609                 seq_puts(seq, ",grpid");
 610         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 611                 seq_puts(seq, ",nogrpid");
 612         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 613             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 614                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 615         }
 616         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 617             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 618                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 619         }
 620         if (test_opt(sb, ERRORS_RO)) {
 621                 int def_errors = le16_to_cpu(es->s_errors);
 622
 623                 if (def_errors == EXT3_ERRORS_PANIC ||
 624                     def_errors == EXT3_ERRORS_CONTINUE) {
 625                         seq_puts(seq, ",errors=remount-ro");
 626                 }
 627         }
 628         if (test_opt(sb, ERRORS_CONT))
 629                 seq_puts(seq, ",errors=continue");
 630         if (test_opt(sb, ERRORS_PANIC))
 631                 seq_puts(seq, ",errors=panic");
 632         if (test_opt(sb, NO_UID32))
 633                 seq_puts(seq, ",nouid32");
 634         if (test_opt(sb, DEBUG))
 635                 seq_puts(seq, ",debug");
 636         if (test_opt(sb, OLDALLOC))
 637                 seq_puts(seq, ",oldalloc");
 638 #ifdef CONFIG_EXT3_FS_XATTR
 639         if (test_opt(sb, XATTR_USER))
 640                 seq_puts(seq, ",user_xattr");
 641         if (!test_opt(sb, XATTR_USER) &&
 642             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 643                 seq_puts(seq, ",nouser_xattr");
 644         }
 645 #endif
 646 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 647         if (test_opt(sb, POSIX_ACL))
 648                 seq_puts(seq, ",acl");
 649         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 650                 seq_puts(seq, ",noacl");
 651 #endif
 652         if (!test_opt(sb, RESERVATION))
 653                 seq_puts(seq, ",noreservation");
 654         if (sbi->s_commit_interval) {
 655                 seq_printf(seq, ",commit=%u",
 656                            (unsigned) (sbi->s_commit_interval / HZ));
 657         }
 658
 659         /*
 660          * Always display barrier state so it's clear what the status is.
 661          */
 662         seq_puts(seq, ",barrier=");
 663         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 664         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 665         if (test_opt(sb, DATA_ERR_ABORT))
 666                 seq_puts(seq, ",data_err=abort");
 667
 668         if (test_opt(sb, NOLOAD))
 669                 seq_puts(seq, ",norecovery");
 670
 671         ext3_show_quota_options(seq, sb);
 672
 673         return 0;
 674 }
 675
 676
 677 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 678                 u64 ino, u32 generation)
 679 {
 680         struct inode *inode;
 681
 682         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 683                 return ERR_PTR(-ESTALE);
 684         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 685                 return ERR_PTR(-ESTALE);
 686
 687         /* iget isn't really right if the inode is currently unallocated!!
 688          *
 689          * ext3_read_inode will return a bad_inode if the inode had been
 690          * deleted, so we should be safe.
 691          *
 692          * Currently we don't know the generation for parent directory, so
 693          * a generation of 0 means "accept any"
 694          */
 695         inode = ext3_iget(sb, ino);
 696         if (IS_ERR(inode))
 697                 return ERR_CAST(inode);
 698         if (generation && inode->i_generation != generation) {
 699                 iput(inode);
 700                 return ERR_PTR(-ESTALE);
 701         }
 702
 703         return inode;
 704 }
 705
 706 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 707                 int fh_len, int fh_type)
 708 {
 709         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 710                                     ext3_nfs_get_inode);
 711 }
 712
 713 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 714                 int fh_len, int fh_type)
 715 {
 716         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 717                                     ext3_nfs_get_inode);
 718 }
 719
 720 /*
 721  * Try to release metadata pages (indirect blocks, directories) which are
 722  * mapped via the block device.  Since these pages could have journal heads
 723  * which would prevent try_to_free_buffers() from freeing them, we must use
 724  * jbd layer's try_to_free_buffers() function to release them.
 725  */
 726 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 727                                  gfp_t wait)
 728 {
 729         journal_t *journal = EXT3_SB(sb)->s_journal;
 730
 731         WARN_ON(PageChecked(page));
 732         if (!page_has_buffers(page))
 733                 return 0;
 734         if (journal)
 735                 return journal_try_to_free_buffers(journal, page,
 736                                                    wait & ~__GFP_WAIT);
 737         return try_to_free_buffers(page);
 738 }
 739
 740 #ifdef CONFIG_QUOTA
 741 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 742 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 743
 744 static int ext3_write_dquot(struct dquot *dquot);
 745 static int ext3_acquire_dquot(struct dquot *dquot);
 746 static int ext3_release_dquot(struct dquot *dquot);
 747 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 748 static int ext3_write_info(struct super_block *sb, int type);
 749 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 750                                 char *path);
 751 static int ext3_quota_on_mount(struct super_block *sb, int type);
 752 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 753                                size_t len, loff_t off);
 754 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 755                                 const char *data, size_t len, loff_t off);
 756
 757 static const struct dquot_operations ext3_quota_operations = {
 758         .write_dquot    = ext3_write_dquot,
 759         .acquire_dquot  = ext3_acquire_dquot,
 760         .release_dquot  = ext3_release_dquot,
 761         .mark_dirty     = ext3_mark_dquot_dirty,
 762         .write_info     = ext3_write_info,
 763         .alloc_dquot    = dquot_alloc,
 764         .destroy_dquot  = dquot_destroy,
 765 };
 766
 767 static const struct quotactl_ops ext3_qctl_operations = {
 768         .quota_on       = ext3_quota_on,
 769         .quota_off      = dquot_quota_off,
 770         .quota_sync     = dquot_quota_sync,
 771         .get_info       = dquot_get_dqinfo,
 772         .set_info       = dquot_set_dqinfo,
 773         .get_dqblk      = dquot_get_dqblk,
 774         .set_dqblk      = dquot_set_dqblk
 775 };
 776 #endif
 777
 778 static const struct super_operations ext3_sops = {
 779         .alloc_inode    = ext3_alloc_inode,
 780         .destroy_inode  = ext3_destroy_inode,
 781         .write_inode    = ext3_write_inode,
 782         .dirty_inode    = ext3_dirty_inode,
 783         .delete_inode   = ext3_delete_inode,
 784         .put_super      = ext3_put_super,
 785         .sync_fs        = ext3_sync_fs,
 786         .freeze_fs      = ext3_freeze,
 787         .unfreeze_fs    = ext3_unfreeze,
 788         .statfs         = ext3_statfs,
 789         .remount_fs     = ext3_remount,
 790         .clear_inode    = ext3_clear_inode,
 791         .show_options   = ext3_show_options,
 792 #ifdef CONFIG_QUOTA
 793         .quota_read     = ext3_quota_read,
 794         .quota_write    = ext3_quota_write,
 795 #endif
 796         .bdev_try_to_free_page = bdev_try_to_free_page,
 797 };
 798
 799 static const struct export_operations ext3_export_ops = {
 800         .fh_to_dentry = ext3_fh_to_dentry,
 801         .fh_to_parent = ext3_fh_to_parent,
 802         .get_parent = ext3_get_parent,
 803 };
 804
 805 enum {
 806         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 807         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 808         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 809         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 810         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 811         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 812         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 813         Opt_data_err_abort, Opt_data_err_ignore,
 814         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 815         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 816         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 817         Opt_resize, Opt_usrquota, Opt_grpquota
 818 };
 819
 820 static const match_table_t tokens = {
 821         {Opt_bsd_df, "bsddf"},
 822         {Opt_minix_df, "minixdf"},
 823         {Opt_grpid, "grpid"},
 824         {Opt_grpid, "bsdgroups"},
 825         {Opt_nogrpid, "nogrpid"},
 826         {Opt_nogrpid, "sysvgroups"},
 827         {Opt_resgid, "resgid=%u"},
 828         {Opt_resuid, "resuid=%u"},
 829         {Opt_sb, "sb=%u"},
 830         {Opt_err_cont, "errors=continue"},
 831         {Opt_err_panic, "errors=panic"},
 832         {Opt_err_ro, "errors=remount-ro"},
 833         {Opt_nouid32, "nouid32"},
 834         {Opt_nocheck, "nocheck"},
 835         {Opt_nocheck, "check=none"},
 836         {Opt_debug, "debug"},
 837         {Opt_oldalloc, "oldalloc"},
 838         {Opt_orlov, "orlov"},
 839         {Opt_user_xattr, "user_xattr"},
 840         {Opt_nouser_xattr, "nouser_xattr"},
 841         {Opt_acl, "acl"},
 842         {Opt_noacl, "noacl"},
 843         {Opt_reservation, "reservation"},
 844         {Opt_noreservation, "noreservation"},
 845         {Opt_noload, "noload"},
 846         {Opt_noload, "norecovery"},
 847         {Opt_nobh, "nobh"},
 848         {Opt_bh, "bh"},
 849         {Opt_commit, "commit=%u"},
 850         {Opt_journal_update, "journal=update"},
 851         {Opt_journal_inum, "journal=%u"},
 852         {Opt_journal_dev, "journal_dev=%u"},
 853         {Opt_abort, "abort"},
 854         {Opt_data_journal, "data=journal"},
 855         {Opt_data_ordered, "data=ordered"},
 856         {Opt_data_writeback, "data=writeback"},
 857         {Opt_data_err_abort, "data_err=abort"},
 858         {Opt_data_err_ignore, "data_err=ignore"},
 859         {Opt_offusrjquota, "usrjquota="},
 860         {Opt_usrjquota, "usrjquota=%s"},
 861         {Opt_offgrpjquota, "grpjquota="},
 862         {Opt_grpjquota, "grpjquota=%s"},
 863         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 864         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 865         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 866         {Opt_grpquota, "grpquota"},
 867         {Opt_noquota, "noquota"},
 868         {Opt_quota, "quota"},
 869         {Opt_usrquota, "usrquota"},
 870         {Opt_barrier, "barrier=%u"},
 871         {Opt_barrier, "barrier"},
 872         {Opt_nobarrier, "nobarrier"},
 873         {Opt_resize, "resize"},
 874         {Opt_err, NULL},
 875 };
 876
 877 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 878 {
 879         ext3_fsblk_t    sb_block;
 880         char            *options = (char *) *data;
 881
 882         if (!options || strncmp(options, "sb=", 3) != 0)
 883                 return 1;       /* Default location */
 884         options += 3;
 885         /*todo: use simple_strtoll with >32bit ext3 */
 886         sb_block = simple_strtoul(options, &options, 0);
 887         if (*options && *options != ',') {
 888                 ext3_msg(sb, "error: invalid sb specification: %s",
 889                        (char *) *data);
 890                 return 1;
 891         }
 892         if (*options == ',')
 893                 options++;
 894         *data = (void *) options;
 895         return sb_block;
 896 }
 897
 898 #ifdef CONFIG_QUOTA
 899 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 900 {
 901         struct ext3_sb_info *sbi = EXT3_SB(sb);
 902         char *qname;
 903
 904         if (sb_any_quota_loaded(sb) &&
 905                 !sbi->s_qf_names[qtype]) {
 906                 ext3_msg(sb, KERN_ERR,
 907                         "Cannot change journaled "
 908                         "quota options when quota turned on");
 909                 return 0;
 910         }
 911         qname = match_strdup(args);
 912         if (!qname) {
 913                 ext3_msg(sb, KERN_ERR,
 914                         "Not enough memory for storing quotafile name");
 915                 return 0;
 916         }
 917         if (sbi->s_qf_names[qtype] &&
 918                 strcmp(sbi->s_qf_names[qtype], qname)) {
 919                 ext3_msg(sb, KERN_ERR,
 920                         "%s quota file already specified", QTYPE2NAME(qtype));
 921                 kfree(qname);
 922                 return 0;
 923         }
 924         sbi->s_qf_names[qtype] = qname;
 925         if (strchr(sbi->s_qf_names[qtype], '/')) {
 926                 ext3_msg(sb, KERN_ERR,
 927                         "quotafile must be on filesystem root");
 928                 kfree(sbi->s_qf_names[qtype]);
 929                 sbi->s_qf_names[qtype] = NULL;
 930                 return 0;
 931         }
 932         set_opt(sbi->s_mount_opt, QUOTA);
 933         return 1;
 934 }
 935
 936 static int clear_qf_name(struct super_block *sb, int qtype) {
 937
 938         struct ext3_sb_info *sbi = EXT3_SB(sb);
 939
 940         if (sb_any_quota_loaded(sb) &&
 941                 sbi->s_qf_names[qtype]) {
 942                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 943                         " when quota turned on");
 944                 return 0;
 945         }
 946         /*
 947          * The space will be released later when all options are confirmed
 948          * to be correct
 949          */
 950         sbi->s_qf_names[qtype] = NULL;
 951         return 1;
 952 }
 953 #endif
 954
 955 static int parse_options (char *options, struct super_block *sb,
 956                           unsigned int *inum, unsigned long *journal_devnum,
 957                           ext3_fsblk_t *n_blocks_count, int is_remount)
 958 {
 959         struct ext3_sb_info *sbi = EXT3_SB(sb);
 960         char * p;
 961         substring_t args[MAX_OPT_ARGS];
 962         int data_opt = 0;
 963         int option;
 964 #ifdef CONFIG_QUOTA
 965         int qfmt;
 966 #endif
 967
 968         if (!options)
 969                 return 1;
 970
 971         while ((p = strsep (&options, ",")) != NULL) {
 972                 int token;
 973                 if (!*p)
 974                         continue;
 975                 /*
 976                  * Initialize args struct so we know whether arg was
 977                  * found; some options take optional arguments.
 978                  */
 979                 args[0].to = args[0].from = 0;
 980                 token = match_token(p, tokens, args);
 981                 switch (token) {
 982                 case Opt_bsd_df:
 983                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 984                         break;
 985                 case Opt_minix_df:
 986                         set_opt (sbi->s_mount_opt, MINIX_DF);
 987                         break;
 988                 case Opt_grpid:
 989                         set_opt (sbi->s_mount_opt, GRPID);
 990                         break;
 991                 case Opt_nogrpid:
 992                         clear_opt (sbi->s_mount_opt, GRPID);
 993                         break;
 994                 case Opt_resuid:
 995                         if (match_int(&args[0], &option))
 996                                 return 0;
 997                         sbi->s_resuid = option;
 998                         break;
 999                 case Opt_resgid:
1000                         if (match_int(&args[0], &option))
1001                                 return 0;
1002                         sbi->s_resgid = option;
1003                         break;
1004                 case Opt_sb:
1005                         /* handled by get_sb_block() instead of here */
1006                         /* *sb_block = match_int(&args[0]); */
1007                         break;
1008                 case Opt_err_panic:
1009                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1010                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1011                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1012                         break;
1013                 case Opt_err_ro:
1014                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1015                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1016                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1017                         break;
1018                 case Opt_err_cont:
1019                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1020                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1021                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1022                         break;
1023                 case Opt_nouid32:
1024                         set_opt (sbi->s_mount_opt, NO_UID32);
1025                         break;
1026                 case Opt_nocheck:
1027                         clear_opt (sbi->s_mount_opt, CHECK);
1028                         break;
1029                 case Opt_debug:
1030                         set_opt (sbi->s_mount_opt, DEBUG);
1031                         break;
1032                 case Opt_oldalloc:
1033                         set_opt (sbi->s_mount_opt, OLDALLOC);
1034                         break;
1035                 case Opt_orlov:
1036                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1037                         break;
1038 #ifdef CONFIG_EXT3_FS_XATTR
1039                 case Opt_user_xattr:
1040                         set_opt (sbi->s_mount_opt, XATTR_USER);
1041                         break;
1042                 case Opt_nouser_xattr:
1043                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1044                         break;
1045 #else
1046                 case Opt_user_xattr:
1047                 case Opt_nouser_xattr:
1048                         ext3_msg(sb, KERN_INFO,
1049                                 "(no)user_xattr options not supported");
1050                         break;
1051 #endif
1052 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1053                 case Opt_acl:
1054                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1055                         break;
1056                 case Opt_noacl:
1057                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1058                         break;
1059 #else
1060                 case Opt_acl:
1061                 case Opt_noacl:
1062                         ext3_msg(sb, KERN_INFO,
1063                                 "(no)acl options not supported");
1064                         break;
1065 #endif
1066                 case Opt_reservation:
1067                         set_opt(sbi->s_mount_opt, RESERVATION);
1068                         break;
1069                 case Opt_noreservation:
1070                         clear_opt(sbi->s_mount_opt, RESERVATION);
1071                         break;
1072                 case Opt_journal_update:
1073                         /* @@@ FIXME */
1074                         /* Eventually we will want to be able to create
1075                            a journal file here.  For now, only allow the
1076                            user to specify an existing inode to be the
1077                            journal file. */
1078                         if (is_remount) {
1079                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1080                                         "journal on remount");
1081                                 return 0;
1082                         }
1083                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1084                         break;
1085                 case Opt_journal_inum:
1086                         if (is_remount) {
1087                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1088                                        "journal on remount");
1089                                 return 0;
1090                         }
1091                         if (match_int(&args[0], &option))
1092                                 return 0;
1093                         *inum = option;
1094                         break;
1095                 case Opt_journal_dev:
1096                         if (is_remount) {
1097                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1098                                        "journal on remount");
1099                                 return 0;
1100                         }
1101                         if (match_int(&args[0], &option))
1102                                 return 0;
1103                         *journal_devnum = option;
1104                         break;
1105                 case Opt_noload:
1106                         set_opt (sbi->s_mount_opt, NOLOAD);
1107                         break;
1108                 case Opt_commit:
1109                         if (match_int(&args[0], &option))
1110                                 return 0;
1111                         if (option < 0)
1112                                 return 0;
1113                         if (option == 0)
1114                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1115                         sbi->s_commit_interval = HZ * option;
1116                         break;
1117                 case Opt_data_journal:
1118                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1119                         goto datacheck;
1120                 case Opt_data_ordered:
1121                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1122                         goto datacheck;
1123                 case Opt_data_writeback:
1124                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1125                 datacheck:
1126                         if (is_remount) {
1127                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1128                                         break;
1129                                 ext3_msg(sb, KERN_ERR,
1130                                         "error: cannot change "
1131                                         "data mode on remount. The filesystem "
1132                                         "is mounted in data=%s mode and you "
1133                                         "try to remount it in data=%s mode.",
1134                                         data_mode_string(test_opt(sb,
1135                                                         DATA_FLAGS)),
1136                                         data_mode_string(data_opt));
1137                                 return 0;
1138                         } else {
1139                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1140                                 sbi->s_mount_opt |= data_opt;
1141                         }
1142                         break;
1143                 case Opt_data_err_abort:
1144                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1145                         break;
1146                 case Opt_data_err_ignore:
1147                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1148                         break;
1149 #ifdef CONFIG_QUOTA
1150                 case Opt_usrjquota:
1151                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1152                                 return 0;
1153                         break;
1154                 case Opt_grpjquota:
1155                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1156                                 return 0;
1157                         break;
1158                 case Opt_offusrjquota:
1159                         if (!clear_qf_name(sb, USRQUOTA))
1160                                 return 0;
1161                         break;
1162                 case Opt_offgrpjquota:
1163                         if (!clear_qf_name(sb, GRPQUOTA))
1164                                 return 0;
1165                         break;
1166                 case Opt_jqfmt_vfsold:
1167                         qfmt = QFMT_VFS_OLD;
1168                         goto set_qf_format;
1169                 case Opt_jqfmt_vfsv0:
1170                         qfmt = QFMT_VFS_V0;
1171                         goto set_qf_format;
1172                 case Opt_jqfmt_vfsv1:
1173                         qfmt = QFMT_VFS_V1;
1174 set_qf_format:
1175                         if (sb_any_quota_loaded(sb) &&
1176                             sbi->s_jquota_fmt != qfmt) {
1177                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1178                                         "journaled quota options when "
1179                                         "quota turned on.");
1180                                 return 0;
1181                         }
1182                         sbi->s_jquota_fmt = qfmt;
1183                         break;
1184                 case Opt_quota:
1185                 case Opt_usrquota:
1186                         set_opt(sbi->s_mount_opt, QUOTA);
1187                         set_opt(sbi->s_mount_opt, USRQUOTA);
1188                         break;
1189                 case Opt_grpquota:
1190                         set_opt(sbi->s_mount_opt, QUOTA);
1191                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1192                         break;
1193                 case Opt_noquota:
1194                         if (sb_any_quota_loaded(sb)) {
1195                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1196                                         "quota options when quota turned on.");
1197                                 return 0;
1198                         }
1199                         clear_opt(sbi->s_mount_opt, QUOTA);
1200                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1201                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1202                         break;
1203 #else
1204                 case Opt_quota:
1205                 case Opt_usrquota:
1206                 case Opt_grpquota:
1207                         ext3_msg(sb, KERN_ERR,
1208                                 "error: quota options not supported.");
1209                         break;
1210                 case Opt_usrjquota:
1211                 case Opt_grpjquota:
1212                 case Opt_offusrjquota:
1213                 case Opt_offgrpjquota:
1214                 case Opt_jqfmt_vfsold:
1215                 case Opt_jqfmt_vfsv0:
1216                 case Opt_jqfmt_vfsv1:
1217                         ext3_msg(sb, KERN_ERR,
1218                                 "error: journaled quota options not "
1219                                 "supported.");
1220                         break;
1221                 case Opt_noquota:
1222                         break;
1223 #endif
1224                 case Opt_abort:
1225                         set_opt(sbi->s_mount_opt, ABORT);
1226                         break;
1227                 case Opt_nobarrier:
1228                         clear_opt(sbi->s_mount_opt, BARRIER);
1229                         break;
1230                 case Opt_barrier:
1231                         if (args[0].from) {
1232                                 if (match_int(&args[0], &option))
1233                                         return 0;
1234                         } else
1235                                 option = 1;     /* No argument, default to 1 */
1236                         if (option)
1237                                 set_opt(sbi->s_mount_opt, BARRIER);
1238                         else
1239                                 clear_opt(sbi->s_mount_opt, BARRIER);
1240                         break;
1241                 case Opt_ignore:
1242                         break;
1243                 case Opt_resize:
1244                         if (!is_remount) {
1245                                 ext3_msg(sb, KERN_ERR,
1246                                         "error: resize option only available "
1247                                         "for remount");
1248                                 return 0;
1249                         }
1250                         if (match_int(&args[0], &option) != 0)
1251                                 return 0;
1252                         *n_blocks_count = option;
1253                         break;
1254                 case Opt_nobh:
1255                         ext3_msg(sb, KERN_WARNING,
1256                                 "warning: ignoring deprecated nobh option");
1257                         break;
1258                 case Opt_bh:
1259                         ext3_msg(sb, KERN_WARNING,
1260                                 "warning: ignoring deprecated bh option");
1261                         break;
1262                 default:
1263                         ext3_msg(sb, KERN_ERR,
1264                                 "error: unrecognized mount option \"%s\" "
1265                                 "or missing value", p);
1266                         return 0;
1267                 }
1268         }
1269 #ifdef CONFIG_QUOTA
1270         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1271                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1272                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1273                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1274                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1275
1276                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1277                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1278                                         "format mixing.");
1279                         return 0;
1280                 }
1281
1282                 if (!sbi->s_jquota_fmt) {
1283                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1284                                         "not specified.");
1285                         return 0;
1286                 }
1287         } else {
1288                 if (sbi->s_jquota_fmt) {
1289                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1290                                         "specified with no journaling "
1291                                         "enabled.");
1292                         return 0;
1293                 }
1294         }
1295 #endif
1296         return 1;
1297 }
1298
1299 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1300                             int read_only)
1301 {
1302         struct ext3_sb_info *sbi = EXT3_SB(sb);
1303         int res = 0;
1304
1305         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1306                 ext3_msg(sb, KERN_ERR,
1307                         "error: revision level too high, "
1308                         "forcing read-only mode");
1309                 res = MS_RDONLY;
1310         }
1311         if (read_only)
1312                 return res;
1313         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1314                 ext3_msg(sb, KERN_WARNING,
1315                         "warning: mounting unchecked fs, "
1316                         "running e2fsck is recommended");
1317         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1318                 ext3_msg(sb, KERN_WARNING,
1319                         "warning: mounting fs with errors, "
1320                         "running e2fsck is recommended");
1321         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1322                  le16_to_cpu(es->s_mnt_count) >=
1323                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1324                 ext3_msg(sb, KERN_WARNING,
1325                         "warning: maximal mount count reached, "
1326                         "running e2fsck is recommended");
1327         else if (le32_to_cpu(es->s_checkinterval) &&
1328                 (le32_to_cpu(es->s_lastcheck) +
1329                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1330                 ext3_msg(sb, KERN_WARNING,
1331                         "warning: checktime reached, "
1332                         "running e2fsck is recommended");
1333 #if 0
1334                 /* @@@ We _will_ want to clear the valid bit if we find
1335                    inconsistencies, to force a fsck at reboot.  But for
1336                    a plain journaled filesystem we can keep it set as
1337                    valid forever! :) */
1338         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1339 #endif
1340         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1341                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1342         le16_add_cpu(&es->s_mnt_count, 1);
1343         es->s_mtime = cpu_to_le32(get_seconds());
1344         ext3_update_dynamic_rev(sb);
1345         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1346
1347         ext3_commit_super(sb, es, 1);
1348         if (test_opt(sb, DEBUG))
1349                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1350                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1351                         sb->s_blocksize,
1352                         sbi->s_groups_count,
1353                         EXT3_BLOCKS_PER_GROUP(sb),
1354                         EXT3_INODES_PER_GROUP(sb),
1355                         sbi->s_mount_opt);
1356
1357         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1358                 char b[BDEVNAME_SIZE];
1359                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1360                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1361         } else {
1362                 ext3_msg(sb, KERN_INFO, "using internal journal");
1363         }
1364         return res;
1365 }
1366
1367 /* Called at mount-time, super-block is locked */
1368 static int ext3_check_descriptors(struct super_block *sb)
1369 {
1370         struct ext3_sb_info *sbi = EXT3_SB(sb);
1371         int i;
1372
1373         ext3_debug ("Checking group descriptors");
1374
1375         for (i = 0; i < sbi->s_groups_count; i++) {
1376                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1377                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1378                 ext3_fsblk_t last_block;
1379
1380                 if (i == sbi->s_groups_count - 1)
1381                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1382                 else
1383                         last_block = first_block +
1384                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1385
1386                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1387                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1388                 {
1389                         ext3_error (sb, "ext3_check_descriptors",
1390                                     "Block bitmap for group %d"
1391                                     " not in group (block %lu)!",
1392                                     i, (unsigned long)
1393                                         le32_to_cpu(gdp->bg_block_bitmap));
1394                         return 0;
1395                 }
1396                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1397                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1398                 {
1399                         ext3_error (sb, "ext3_check_descriptors",
1400                                     "Inode bitmap for group %d"
1401                                     " not in group (block %lu)!",
1402                                     i, (unsigned long)
1403                                         le32_to_cpu(gdp->bg_inode_bitmap));
1404                         return 0;
1405                 }
1406                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1407                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1408                     last_block)
1409                 {
1410                         ext3_error (sb, "ext3_check_descriptors",
1411                                     "Inode table for group %d"
1412                                     " not in group (block %lu)!",
1413                                     i, (unsigned long)
1414                                         le32_to_cpu(gdp->bg_inode_table));
1415                         return 0;
1416                 }
1417         }
1418
1419         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1420         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1421         return 1;
1422 }
1423
1424
1425 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1426  * the superblock) which were deleted from all directories, but held open by
1427  * a process at the time of a crash.  We walk the list and try to delete these
1428  * inodes at recovery time (only with a read-write filesystem).
1429  *
1430  * In order to keep the orphan inode chain consistent during traversal (in
1431  * case of crash during recovery), we link each inode into the superblock
1432  * orphan list_head and handle it the same way as an inode deletion during
1433  * normal operation (which journals the operations for us).
1434  *
1435  * We only do an iget() and an iput() on each inode, which is very safe if we
1436  * accidentally point at an in-use or already deleted inode.  The worst that
1437  * can happen in this case is that we get a "bit already cleared" message from
1438  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1439  * e2fsck was run on this filesystem, and it must have already done the orphan
1440  * inode cleanup for us, so we can safely abort without any further action.
1441  */
1442 static void ext3_orphan_cleanup (struct super_block * sb,
1443                                  struct ext3_super_block * es)
1444 {
1445         unsigned int s_flags = sb->s_flags;
1446         int nr_orphans = 0, nr_truncates = 0;
1447 #ifdef CONFIG_QUOTA
1448         int i;
1449 #endif
1450         if (!es->s_last_orphan) {
1451                 jbd_debug(4, "no orphan inodes to clean up\n");
1452                 return;
1453         }
1454
1455         if (bdev_read_only(sb->s_bdev)) {
1456                 ext3_msg(sb, KERN_ERR, "error: write access "
1457                         "unavailable, skipping orphan cleanup.");
1458                 return;
1459         }
1460
1461         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1462                 if (es->s_last_orphan)
1463                         jbd_debug(1, "Errors on filesystem, "
1464                                   "clearing orphan list.\n");
1465                 es->s_last_orphan = 0;
1466                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1467                 return;
1468         }
1469
1470         if (s_flags & MS_RDONLY) {
1471                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1472                 sb->s_flags &= ~MS_RDONLY;
1473         }
1474 #ifdef CONFIG_QUOTA
1475         /* Needed for iput() to work correctly and not trash data */
1476         sb->s_flags |= MS_ACTIVE;
1477         /* Turn on quotas so that they are updated correctly */
1478         for (i = 0; i < MAXQUOTAS; i++) {
1479                 if (EXT3_SB(sb)->s_qf_names[i]) {
1480                         int ret = ext3_quota_on_mount(sb, i);
1481                         if (ret < 0)
1482                                 ext3_msg(sb, KERN_ERR,
1483                                         "error: cannot turn on journaled "
1484                                         "quota: %d", ret);
1485                 }
1486         }
1487 #endif
1488
1489         while (es->s_last_orphan) {
1490                 struct inode *inode;
1491
1492                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1493                 if (IS_ERR(inode)) {
1494                         es->s_last_orphan = 0;
1495                         break;
1496                 }
1497
1498                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1499                 dquot_initialize(inode);
1500                 if (inode->i_nlink) {
1501                         printk(KERN_DEBUG
1502                                 "%s: truncating inode %lu to %Ld bytes\n",
1503                                 __func__, inode->i_ino, inode->i_size);
1504                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1505                                   inode->i_ino, inode->i_size);
1506                         ext3_truncate(inode);
1507                         nr_truncates++;
1508                 } else {
1509                         printk(KERN_DEBUG
1510                                 "%s: deleting unreferenced inode %lu\n",
1511                                 __func__, inode->i_ino);
1512                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1513                                   inode->i_ino);
1514                         nr_orphans++;
1515                 }
1516                 iput(inode);  /* The delete magic happens here! */
1517         }
1518
1519 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1520
1521         if (nr_orphans)
1522                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1523                        PLURAL(nr_orphans));
1524         if (nr_truncates)
1525                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1526                        PLURAL(nr_truncates));
1527 #ifdef CONFIG_QUOTA
1528         /* Turn quotas off */
1529         for (i = 0; i < MAXQUOTAS; i++) {
1530                 if (sb_dqopt(sb)->files[i])
1531                         dquot_quota_off(sb, i);
1532         }
1533 #endif
1534         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1535 }
1536
1537 /*
1538  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1539  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1540  * We need to be 1 filesystem block less than the 2^32 sector limit.
1541  */
1542 static loff_t ext3_max_size(int bits)
1543 {
1544         loff_t res = EXT3_NDIR_BLOCKS;
1545         int meta_blocks;
1546         loff_t upper_limit;
1547
1548         /* This is calculated to be the largest file size for a
1549          * dense, file such that the total number of
1550          * sectors in the file, including data and all indirect blocks,
1551          * does not exceed 2^32 -1
1552          * __u32 i_blocks representing the total number of
1553          * 512 bytes blocks of the file
1554          */
1555         upper_limit = (1LL << 32) - 1;
1556
1557         /* total blocks in file system block size */
1558         upper_limit >>= (bits - 9);
1559
1560
1561         /* indirect blocks */
1562         meta_blocks = 1;
1563         /* double indirect blocks */
1564         meta_blocks += 1 + (1LL << (bits-2));
1565         /* tripple indirect blocks */
1566         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1567
1568         upper_limit -= meta_blocks;
1569         upper_limit <<= bits;
1570
1571         res += 1LL << (bits-2);
1572         res += 1LL << (2*(bits-2));
1573         res += 1LL << (3*(bits-2));
1574         res <<= bits;
1575         if (res > upper_limit)
1576                 res = upper_limit;
1577
1578         if (res > MAX_LFS_FILESIZE)
1579                 res = MAX_LFS_FILESIZE;
1580
1581         return res;
1582 }
1583
1584 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1585                                     ext3_fsblk_t logic_sb_block,
1586                                     int nr)
1587 {
1588         struct ext3_sb_info *sbi = EXT3_SB(sb);
1589         unsigned long bg, first_meta_bg;
1590         int has_super = 0;
1591
1592         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1593
1594         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1595             nr < first_meta_bg)
1596                 return (logic_sb_block + nr + 1);
1597         bg = sbi->s_desc_per_block * nr;
1598         if (ext3_bg_has_super(sb, bg))
1599                 has_super = 1;
1600         return (has_super + ext3_group_first_block_no(sb, bg));
1601 }
1602
1603
1604 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1605 {
1606         struct buffer_head * bh;
1607         struct ext3_super_block *es = NULL;
1608         struct ext3_sb_info *sbi;
1609         ext3_fsblk_t block;
1610         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1611         ext3_fsblk_t logic_sb_block;
1612         unsigned long offset = 0;
1613         unsigned int journal_inum = 0;
1614         unsigned long journal_devnum = 0;
1615         unsigned long def_mount_opts;
1616         struct inode *root;
1617         int blocksize;
1618         int hblock;
1619         int db_count;
1620         int i;
1621         int needs_recovery;
1622         int ret = -EINVAL;
1623         __le32 features;
1624         int err;
1625
1626         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1627         if (!sbi)
1628                 return -ENOMEM;
1629
1630         sbi->s_blockgroup_lock =
1631                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1632         if (!sbi->s_blockgroup_lock) {
1633                 kfree(sbi);
1634                 return -ENOMEM;
1635         }
1636         sb->s_fs_info = sbi;
1637         sbi->s_mount_opt = 0;
1638         sbi->s_resuid = EXT3_DEF_RESUID;
1639         sbi->s_resgid = EXT3_DEF_RESGID;
1640         sbi->s_sb_block = sb_block;
1641
1642         unlock_kernel();
1643
1644         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1645         if (!blocksize) {
1646                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1647                 goto out_fail;
1648         }
1649
1650         /*
1651          * The ext3 superblock will not be buffer aligned for other than 1kB
1652          * block sizes.  We need to calculate the offset from buffer start.
1653          */
1654         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1655                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1656                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1657         } else {
1658                 logic_sb_block = sb_block;
1659         }
1660
1661         if (!(bh = sb_bread(sb, logic_sb_block))) {
1662                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1663                 goto out_fail;
1664         }
1665         /*
1666          * Note: s_es must be initialized as soon as possible because
1667          *       some ext3 macro-instructions depend on its value
1668          */
1669         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1670         sbi->s_es = es;
1671         sb->s_magic = le16_to_cpu(es->s_magic);
1672         if (sb->s_magic != EXT3_SUPER_MAGIC)
1673                 goto cantfind_ext3;
1674
1675         /* Set defaults before we parse the mount options */
1676         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1677         if (def_mount_opts & EXT3_DEFM_DEBUG)
1678                 set_opt(sbi->s_mount_opt, DEBUG);
1679         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1680                 set_opt(sbi->s_mount_opt, GRPID);
1681         if (def_mount_opts & EXT3_DEFM_UID16)
1682                 set_opt(sbi->s_mount_opt, NO_UID32);
1683 #ifdef CONFIG_EXT3_FS_XATTR
1684         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1685                 set_opt(sbi->s_mount_opt, XATTR_USER);
1686 #endif
1687 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1688         if (def_mount_opts & EXT3_DEFM_ACL)
1689                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1690 #endif
1691         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1692                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1693         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1694                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1695         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1696                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1697
1698         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1699                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1700         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1701                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1702         else
1703                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1704
1705         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1706         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1707
1708         set_opt(sbi->s_mount_opt, RESERVATION);
1709
1710         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1711                             NULL, 0))
1712                 goto failed_mount;
1713
1714         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1715                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1716
1717         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1718             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1719              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1720              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1721                 ext3_msg(sb, KERN_WARNING,
1722                         "warning: feature flags set on rev 0 fs, "
1723                         "running e2fsck is recommended");
1724         /*
1725          * Check feature flags regardless of the revision level, since we
1726          * previously didn't change the revision level when setting the flags,
1727          * so there is a chance incompat flags are set on a rev 0 filesystem.
1728          */
1729         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1730         if (features) {
1731                 ext3_msg(sb, KERN_ERR,
1732                         "error: couldn't mount because of unsupported "
1733                         "optional features (%x)", le32_to_cpu(features));
1734                 goto failed_mount;
1735         }
1736         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1737         if (!(sb->s_flags & MS_RDONLY) && features) {
1738                 ext3_msg(sb, KERN_ERR,
1739                         "error: couldn't mount RDWR because of unsupported "
1740                         "optional features (%x)", le32_to_cpu(features));
1741                 goto failed_mount;
1742         }
1743         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1744
1745         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1746             blocksize > EXT3_MAX_BLOCK_SIZE) {
1747                 ext3_msg(sb, KERN_ERR,
1748                         "error: couldn't mount because of unsupported "
1749                         "filesystem blocksize %d", blocksize);
1750                 goto failed_mount;
1751         }
1752
1753         hblock = bdev_logical_block_size(sb->s_bdev);
1754         if (sb->s_blocksize != blocksize) {
1755                 /*
1756                  * Make sure the blocksize for the filesystem is larger
1757                  * than the hardware sectorsize for the machine.
1758                  */
1759                 if (blocksize < hblock) {
1760                         ext3_msg(sb, KERN_ERR,
1761                                 "error: fsblocksize %d too small for "
1762                                 "hardware sectorsize %d", blocksize, hblock);
1763                         goto failed_mount;
1764                 }
1765
1766                 brelse (bh);
1767                 if (!sb_set_blocksize(sb, blocksize)) {
1768                         ext3_msg(sb, KERN_ERR,
1769                                 "error: bad blocksize %d", blocksize);
1770                         goto out_fail;
1771                 }
1772                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1773                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1774                 bh = sb_bread(sb, logic_sb_block);
1775                 if (!bh) {
1776                         ext3_msg(sb, KERN_ERR,
1777                                "error: can't read superblock on 2nd try");
1778                         goto failed_mount;
1779                 }
1780                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1781                 sbi->s_es = es;
1782                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1783                         ext3_msg(sb, KERN_ERR,
1784                                 "error: magic mismatch");
1785                         goto failed_mount;
1786                 }
1787         }
1788
1789         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1790
1791         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1792                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1793                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1794         } else {
1795                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1796                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1797                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1798                     (!is_power_of_2(sbi->s_inode_size)) ||
1799                     (sbi->s_inode_size > blocksize)) {
1800                         ext3_msg(sb, KERN_ERR,
1801                                 "error: unsupported inode size: %d",
1802                                 sbi->s_inode_size);
1803                         goto failed_mount;
1804                 }
1805         }
1806         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1807                                    le32_to_cpu(es->s_log_frag_size);
1808         if (blocksize != sbi->s_frag_size) {
1809                 ext3_msg(sb, KERN_ERR,
1810                        "error: fragsize %lu != blocksize %u (unsupported)",
1811                        sbi->s_frag_size, blocksize);
1812                 goto failed_mount;
1813         }
1814         sbi->s_frags_per_block = 1;
1815         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1816         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1817         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1818         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1819                 goto cantfind_ext3;
1820         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1821         if (sbi->s_inodes_per_block == 0)
1822                 goto cantfind_ext3;
1823         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1824                                         sbi->s_inodes_per_block;
1825         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1826         sbi->s_sbh = bh;
1827         sbi->s_mount_state = le16_to_cpu(es->s_state);
1828         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1829         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1830         for (i=0; i < 4; i++)
1831                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1832         sbi->s_def_hash_version = es->s_def_hash_version;
1833         i = le32_to_cpu(es->s_flags);
1834         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1835                 sbi->s_hash_unsigned = 3;
1836         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1837 #ifdef __CHAR_UNSIGNED__
1838                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1839                 sbi->s_hash_unsigned = 3;
1840 #else
1841                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1842 #endif
1843         }
1844
1845         if (sbi->s_blocks_per_group > blocksize * 8) {
1846                 ext3_msg(sb, KERN_ERR,
1847                         "#blocks per group too big: %lu",
1848                         sbi->s_blocks_per_group);
1849                 goto failed_mount;
1850         }
1851         if (sbi->s_frags_per_group > blocksize * 8) {
1852                 ext3_msg(sb, KERN_ERR,
1853                         "error: #fragments per group too big: %lu",
1854                         sbi->s_frags_per_group);
1855                 goto failed_mount;
1856         }
1857         if (sbi->s_inodes_per_group > blocksize * 8) {
1858                 ext3_msg(sb, KERN_ERR,
1859                         "error: #inodes per group too big: %lu",
1860                         sbi->s_inodes_per_group);
1861                 goto failed_mount;
1862         }
1863
1864         if (le32_to_cpu(es->s_blocks_count) >
1865                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1866                 ext3_msg(sb, KERN_ERR,
1867                         "error: filesystem is too large to mount safely");
1868                 if (sizeof(sector_t) < 8)
1869                         ext3_msg(sb, KERN_ERR,
1870                                 "error: CONFIG_LBDAF not enabled");
1871                 goto failed_mount;
1872         }
1873
1874         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1875                 goto cantfind_ext3;
1876         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1877                                le32_to_cpu(es->s_first_data_block) - 1)
1878                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1879         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1880                    EXT3_DESC_PER_BLOCK(sb);
1881         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1882                                     GFP_KERNEL);
1883         if (sbi->s_group_desc == NULL) {
1884                 ext3_msg(sb, KERN_ERR,
1885                         "error: not enough memory");
1886                 goto failed_mount;
1887         }
1888
1889         bgl_lock_init(sbi->s_blockgroup_lock);
1890
1891         for (i = 0; i < db_count; i++) {
1892                 block = descriptor_loc(sb, logic_sb_block, i);
1893                 sbi->s_group_desc[i] = sb_bread(sb, block);
1894                 if (!sbi->s_group_desc[i]) {
1895                         ext3_msg(sb, KERN_ERR,
1896                                 "error: can't read group descriptor %d", i);
1897                         db_count = i;
1898                         goto failed_mount2;
1899                 }
1900         }
1901         if (!ext3_check_descriptors (sb)) {
1902                 ext3_msg(sb, KERN_ERR,
1903                         "error: group descriptors corrupted");
1904                 goto failed_mount2;
1905         }
1906         sbi->s_gdb_count = db_count;
1907         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1908         spin_lock_init(&sbi->s_next_gen_lock);
1909
1910         /* per fileystem reservation list head & lock */
1911         spin_lock_init(&sbi->s_rsv_window_lock);
1912         sbi->s_rsv_window_root = RB_ROOT;
1913         /* Add a single, static dummy reservation to the start of the
1914          * reservation window list --- it gives us a placeholder for
1915          * append-at-start-of-list which makes the allocation logic
1916          * _much_ simpler. */
1917         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1918         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1919         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1920         sbi->s_rsv_window_head.rsv_goal_size = 0;
1921         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1922
1923         /*
1924          * set up enough so that it can read an inode
1925          */
1926         sb->s_op = &ext3_sops;
1927         sb->s_export_op = &ext3_export_ops;
1928         sb->s_xattr = ext3_xattr_handlers;
1929 #ifdef CONFIG_QUOTA
1930         sb->s_qcop = &ext3_qctl_operations;
1931         sb->dq_op = &ext3_quota_operations;
1932 #endif
1933         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1934         mutex_init(&sbi->s_orphan_lock);
1935         mutex_init(&sbi->s_resize_lock);
1936
1937         sb->s_root = NULL;
1938
1939         needs_recovery = (es->s_last_orphan != 0 ||
1940                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1941                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1942
1943         /*
1944          * The first inode we look at is the journal inode.  Don't try
1945          * root first: it may be modified in the journal!
1946          */
1947         if (!test_opt(sb, NOLOAD) &&
1948             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1949                 if (ext3_load_journal(sb, es, journal_devnum))
1950                         goto failed_mount2;
1951         } else if (journal_inum) {
1952                 if (ext3_create_journal(sb, es, journal_inum))
1953                         goto failed_mount2;
1954         } else {
1955                 if (!silent)
1956                         ext3_msg(sb, KERN_ERR,
1957                                 "error: no journal found. "
1958                                 "mounting ext3 over ext2?");
1959                 goto failed_mount2;
1960         }
1961         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1962                         ext3_count_free_blocks(sb));
1963         if (!err) {
1964                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1965                                 ext3_count_free_inodes(sb));
1966         }
1967         if (!err) {
1968                 err = percpu_counter_init(&sbi->s_dirs_counter,
1969                                 ext3_count_dirs(sb));
1970         }
1971         if (err) {
1972                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1973                 goto failed_mount3;
1974         }
1975
1976         /* We have now updated the journal if required, so we can
1977          * validate the data journaling mode. */
1978         switch (test_opt(sb, DATA_FLAGS)) {
1979         case 0:
1980                 /* No mode set, assume a default based on the journal
1981                    capabilities: ORDERED_DATA if the journal can
1982                    cope, else JOURNAL_DATA */
1983                 if (journal_check_available_features
1984                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1985                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1986                 else
1987                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1988                 break;
1989
1990         case EXT3_MOUNT_ORDERED_DATA:
1991         case EXT3_MOUNT_WRITEBACK_DATA:
1992                 if (!journal_check_available_features
1993                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1994                         ext3_msg(sb, KERN_ERR,
1995                                 "error: journal does not support "
1996                                 "requested data journaling mode");
1997                         goto failed_mount3;
1998                 }
1999         default:
2000                 break;
2001         }
2002
2003         /*
2004          * The journal_load will have done any necessary log recovery,
2005          * so we can safely mount the rest of the filesystem now.
2006          */
2007
2008         root = ext3_iget(sb, EXT3_ROOT_INO);
2009         if (IS_ERR(root)) {
2010                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2011                 ret = PTR_ERR(root);
2012                 goto failed_mount3;
2013         }
2014         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2015                 iput(root);
2016                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2017                 goto failed_mount3;
2018         }
2019         sb->s_root = d_alloc_root(root);
2020         if (!sb->s_root) {
2021                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2022                 iput(root);
2023                 ret = -ENOMEM;
2024                 goto failed_mount3;
2025         }
2026
2027         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2028
2029         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2030         ext3_orphan_cleanup(sb, es);
2031         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2032         if (needs_recovery)
2033                 ext3_msg(sb, KERN_INFO, "recovery complete");
2034         ext3_mark_recovery_complete(sb, es);
2035         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2036                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2037                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2038                 "writeback");
2039
2040         lock_kernel();
2041         return 0;
2042
2043 cantfind_ext3:
2044         if (!silent)
2045                 ext3_msg(sb, KERN_INFO,
2046                         "error: can't find ext3 filesystem on dev %s.",
2047                        sb->s_id);
2048         goto failed_mount;
2049
2050 failed_mount3:
2051         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2052         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2053         percpu_counter_destroy(&sbi->s_dirs_counter);
2054         journal_destroy(sbi->s_journal);
2055 failed_mount2:
2056         for (i = 0; i < db_count; i++)
2057                 brelse(sbi->s_group_desc[i]);
2058         kfree(sbi->s_group_desc);
2059 failed_mount:
2060 #ifdef CONFIG_QUOTA
2061         for (i = 0; i < MAXQUOTAS; i++)
2062                 kfree(sbi->s_qf_names[i]);
2063 #endif
2064         ext3_blkdev_remove(sbi);
2065         brelse(bh);
2066 out_fail:
2067         sb->s_fs_info = NULL;
2068         kfree(sbi->s_blockgroup_lock);
2069         kfree(sbi);
2070         lock_kernel();
2071         return ret;
2072 }
2073
2074 /*
2075  * Setup any per-fs journal parameters now.  We'll do this both on
2076  * initial mount, once the journal has been initialised but before we've
2077  * done any recovery; and again on any subsequent remount.
2078  */
2079 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2080 {
2081         struct ext3_sb_info *sbi = EXT3_SB(sb);
2082
2083         if (sbi->s_commit_interval)
2084                 journal->j_commit_interval = sbi->s_commit_interval;
2085         /* We could also set up an ext3-specific default for the commit
2086          * interval here, but for now we'll just fall back to the jbd
2087          * default. */
2088
2089         spin_lock(&journal->j_state_lock);
2090         if (test_opt(sb, BARRIER))
2091                 journal->j_flags |= JFS_BARRIER;
2092         else
2093                 journal->j_flags &= ~JFS_BARRIER;
2094         if (test_opt(sb, DATA_ERR_ABORT))
2095                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2096         else
2097                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2098         spin_unlock(&journal->j_state_lock);
2099 }
2100
2101 static journal_t *ext3_get_journal(struct super_block *sb,
2102                                    unsigned int journal_inum)
2103 {
2104         struct inode *journal_inode;
2105         journal_t *journal;
2106
2107         /* First, test for the existence of a valid inode on disk.  Bad
2108          * things happen if we iget() an unused inode, as the subsequent
2109          * iput() will try to delete it. */
2110
2111         journal_inode = ext3_iget(sb, journal_inum);
2112         if (IS_ERR(journal_inode)) {
2113                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2114                 return NULL;
2115         }
2116         if (!journal_inode->i_nlink) {
2117                 make_bad_inode(journal_inode);
2118                 iput(journal_inode);
2119                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2120                 return NULL;
2121         }
2122
2123         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2124                   journal_inode, journal_inode->i_size);
2125         if (!S_ISREG(journal_inode->i_mode)) {
2126                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2127                 iput(journal_inode);
2128                 return NULL;
2129         }
2130
2131         journal = journal_init_inode(journal_inode);
2132         if (!journal) {
2133                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2134                 iput(journal_inode);
2135                 return NULL;
2136         }
2137         journal->j_private = sb;
2138         ext3_init_journal_params(sb, journal);
2139         return journal;
2140 }
2141
2142 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2143                                        dev_t j_dev)
2144 {
2145         struct buffer_head * bh;
2146         journal_t *journal;
2147         ext3_fsblk_t start;
2148         ext3_fsblk_t len;
2149         int hblock, blocksize;
2150         ext3_fsblk_t sb_block;
2151         unsigned long offset;
2152         struct ext3_super_block * es;
2153         struct block_device *bdev;
2154
2155         bdev = ext3_blkdev_get(j_dev, sb);
2156         if (bdev == NULL)
2157                 return NULL;
2158
2159         if (bd_claim(bdev, sb)) {
2160                 ext3_msg(sb, KERN_ERR,
2161                         "error: failed to claim external journal device");
2162                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2163                 return NULL;
2164         }
2165
2166         blocksize = sb->s_blocksize;
2167         hblock = bdev_logical_block_size(bdev);
2168         if (blocksize < hblock) {
2169                 ext3_msg(sb, KERN_ERR,
2170                         "error: blocksize too small for journal device");
2171                 goto out_bdev;
2172         }
2173
2174         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2175         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2176         set_blocksize(bdev, blocksize);
2177         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2178                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2179                         "external journal");
2180                 goto out_bdev;
2181         }
2182
2183         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2184         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2185             !(le32_to_cpu(es->s_feature_incompat) &
2186               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2187                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2188                         "bad superblock");
2189                 brelse(bh);
2190                 goto out_bdev;
2191         }
2192
2193         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2194                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2195                 brelse(bh);
2196                 goto out_bdev;
2197         }
2198
2199         len = le32_to_cpu(es->s_blocks_count);
2200         start = sb_block + 1;
2201         brelse(bh);     /* we're done with the superblock */
2202
2203         journal = journal_init_dev(bdev, sb->s_bdev,
2204                                         start, len, blocksize);
2205         if (!journal) {
2206                 ext3_msg(sb, KERN_ERR,
2207                         "error: failed to create device journal");
2208                 goto out_bdev;
2209         }
2210         journal->j_private = sb;
2211         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2212         wait_on_buffer(journal->j_sb_buffer);
2213         if (!buffer_uptodate(journal->j_sb_buffer)) {
2214                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2215                 goto out_journal;
2216         }
2217         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2218                 ext3_msg(sb, KERN_ERR,
2219                         "error: external journal has more than one "
2220                         "user (unsupported) - %d",
2221                         be32_to_cpu(journal->j_superblock->s_nr_users));
2222                 goto out_journal;
2223         }
2224         EXT3_SB(sb)->journal_bdev = bdev;
2225         ext3_init_journal_params(sb, journal);
2226         return journal;
2227 out_journal:
2228         journal_destroy(journal);
2229 out_bdev:
2230         ext3_blkdev_put(bdev);
2231         return NULL;
2232 }
2233
2234 static int ext3_load_journal(struct super_block *sb,
2235                              struct ext3_super_block *es,
2236                              unsigned long journal_devnum)
2237 {
2238         journal_t *journal;
2239         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2240         dev_t journal_dev;
2241         int err = 0;
2242         int really_read_only;
2243
2244         if (journal_devnum &&
2245             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2246                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2247                         "numbers have changed");
2248                 journal_dev = new_decode_dev(journal_devnum);
2249         } else
2250                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2251
2252         really_read_only = bdev_read_only(sb->s_bdev);
2253
2254         /*
2255          * Are we loading a blank journal or performing recovery after a
2256          * crash?  For recovery, we need to check in advance whether we
2257          * can get read-write access to the device.
2258          */
2259
2260         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2261                 if (sb->s_flags & MS_RDONLY) {
2262                         ext3_msg(sb, KERN_INFO,
2263                                 "recovery required on readonly filesystem");
2264                         if (really_read_only) {
2265                                 ext3_msg(sb, KERN_ERR, "error: write access "
2266                                         "unavailable, cannot proceed");
2267                                 return -EROFS;
2268                         }
2269                         ext3_msg(sb, KERN_INFO,
2270                                 "write access will be enabled during recovery");
2271                 }
2272         }
2273
2274         if (journal_inum && journal_dev) {
2275                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2276                        "and inode journals");
2277                 return -EINVAL;
2278         }
2279
2280         if (journal_inum) {
2281                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2282                         return -EINVAL;
2283         } else {
2284                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2285                         return -EINVAL;
2286         }
2287
2288         if (!(journal->j_flags & JFS_BARRIER))
2289                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2290
2291         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2292                 err = journal_update_format(journal);
2293                 if (err)  {
2294                         ext3_msg(sb, KERN_ERR, "error updating journal");
2295                         journal_destroy(journal);
2296                         return err;
2297                 }
2298         }
2299
2300         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2301                 err = journal_wipe(journal, !really_read_only);
2302         if (!err)
2303                 err = journal_load(journal);
2304
2305         if (err) {
2306                 ext3_msg(sb, KERN_ERR, "error loading journal");
2307                 journal_destroy(journal);
2308                 return err;
2309         }
2310
2311         EXT3_SB(sb)->s_journal = journal;
2312         ext3_clear_journal_err(sb, es);
2313
2314         if (journal_devnum &&
2315             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2316                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2317
2318                 /* Make sure we flush the recovery flag to disk. */
2319                 ext3_commit_super(sb, es, 1);
2320         }
2321
2322         return 0;
2323 }
2324
2325 static int ext3_create_journal(struct super_block *sb,
2326                                struct ext3_super_block *es,
2327                                unsigned int journal_inum)
2328 {
2329         journal_t *journal;
2330         int err;
2331
2332         if (sb->s_flags & MS_RDONLY) {
2333                 ext3_msg(sb, KERN_ERR,
2334                         "error: readonly filesystem when trying to "
2335                         "create journal");
2336                 return -EROFS;
2337         }
2338
2339         journal = ext3_get_journal(sb, journal_inum);
2340         if (!journal)
2341                 return -EINVAL;
2342
2343         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2344                journal_inum);
2345
2346         err = journal_create(journal);
2347         if (err) {
2348                 ext3_msg(sb, KERN_ERR, "error creating journal");
2349                 journal_destroy(journal);
2350                 return -EIO;
2351         }
2352
2353         EXT3_SB(sb)->s_journal = journal;
2354
2355         ext3_update_dynamic_rev(sb);
2356         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2357         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2358
2359         es->s_journal_inum = cpu_to_le32(journal_inum);
2360
2361         /* Make sure we flush the recovery flag to disk. */
2362         ext3_commit_super(sb, es, 1);
2363
2364         return 0;
2365 }
2366
2367 static int ext3_commit_super(struct super_block *sb,
2368                                struct ext3_super_block *es,
2369                                int sync)
2370 {
2371         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2372         int error = 0;
2373
2374         if (!sbh)
2375                 return error;
2376         /*
2377          * If the file system is mounted read-only, don't update the
2378          * superblock write time.  This avoids updating the superblock
2379          * write time when we are mounting the root file system
2380          * read/only but we need to replay the journal; at that point,
2381          * for people who are east of GMT and who make their clock
2382          * tick in localtime for Windows bug-for-bug compatibility,
2383          * the clock is set in the future, and this will cause e2fsck
2384          * to complain and force a full file system check.
2385          */
2386         if (!(sb->s_flags & MS_RDONLY))
2387                 es->s_wtime = cpu_to_le32(get_seconds());
2388         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2389         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2390         BUFFER_TRACE(sbh, "marking dirty");
2391         mark_buffer_dirty(sbh);
2392         if (sync)
2393                 error = sync_dirty_buffer(sbh);
2394         return error;
2395 }
2396
2397
2398 /*
2399  * Have we just finished recovery?  If so, and if we are mounting (or
2400  * remounting) the filesystem readonly, then we will end up with a
2401  * consistent fs on disk.  Record that fact.
2402  */
2403 static void ext3_mark_recovery_complete(struct super_block * sb,
2404                                         struct ext3_super_block * es)
2405 {
2406         journal_t *journal = EXT3_SB(sb)->s_journal;
2407
2408         journal_lock_updates(journal);
2409         if (journal_flush(journal) < 0)
2410                 goto out;
2411
2412         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2413             sb->s_flags & MS_RDONLY) {
2414                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2415                 ext3_commit_super(sb, es, 1);
2416         }
2417
2418 out:
2419         journal_unlock_updates(journal);
2420 }
2421
2422 /*
2423  * If we are mounting (or read-write remounting) a filesystem whose journal
2424  * has recorded an error from a previous lifetime, move that error to the
2425  * main filesystem now.
2426  */
2427 static void ext3_clear_journal_err(struct super_block *sb,
2428                                    struct ext3_super_block *es)
2429 {
2430         journal_t *journal;
2431         int j_errno;
2432         const char *errstr;
2433
2434         journal = EXT3_SB(sb)->s_journal;
2435
2436         /*
2437          * Now check for any error status which may have been recorded in the
2438          * journal by a prior ext3_error() or ext3_abort()
2439          */
2440
2441         j_errno = journal_errno(journal);
2442         if (j_errno) {
2443                 char nbuf[16];
2444
2445                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2446                 ext3_warning(sb, __func__, "Filesystem error recorded "
2447                              "from previous mount: %s", errstr);
2448                 ext3_warning(sb, __func__, "Marking fs in need of "
2449                              "filesystem check.");
2450
2451                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2452                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2453                 ext3_commit_super (sb, es, 1);
2454
2455                 journal_clear_err(journal);
2456         }
2457 }
2458
2459 /*
2460  * Force the running and committing transactions to commit,
2461  * and wait on the commit.
2462  */
2463 int ext3_force_commit(struct super_block *sb)
2464 {
2465         journal_t *journal;
2466         int ret;
2467
2468         if (sb->s_flags & MS_RDONLY)
2469                 return 0;
2470
2471         journal = EXT3_SB(sb)->s_journal;
2472         ret = ext3_journal_force_commit(journal);
2473         return ret;
2474 }
2475
2476 static int ext3_sync_fs(struct super_block *sb, int wait)
2477 {
2478         tid_t target;
2479
2480         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2481                 if (wait)
2482                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2483         }
2484         return 0;
2485 }
2486
2487 /*
2488  * LVM calls this function before a (read-only) snapshot is created.  This
2489  * gives us a chance to flush the journal completely and mark the fs clean.
2490  */
2491 static int ext3_freeze(struct super_block *sb)
2492 {
2493         int error = 0;
2494         journal_t *journal;
2495
2496         if (!(sb->s_flags & MS_RDONLY)) {
2497                 journal = EXT3_SB(sb)->s_journal;
2498
2499                 /* Now we set up the journal barrier. */
2500                 journal_lock_updates(journal);
2501
2502                 /*
2503                  * We don't want to clear needs_recovery flag when we failed
2504                  * to flush the journal.
2505                  */
2506                 error = journal_flush(journal);
2507                 if (error < 0)
2508                         goto out;
2509
2510                 /* Journal blocked and flushed, clear needs_recovery flag. */
2511                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2512                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2513                 if (error)
2514                         goto out;
2515         }
2516         return 0;
2517
2518 out:
2519         journal_unlock_updates(journal);
2520         return error;
2521 }
2522
2523 /*
2524  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2525  * flag here, even though the filesystem is not technically dirty yet.
2526  */
2527 static int ext3_unfreeze(struct super_block *sb)
2528 {
2529         if (!(sb->s_flags & MS_RDONLY)) {
2530                 lock_super(sb);
2531                 /* Reser the needs_recovery flag before the fs is unlocked. */
2532                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2533                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2534                 unlock_super(sb);
2535                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2536         }
2537         return 0;
2538 }
2539
2540 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2541 {
2542         struct ext3_super_block * es;
2543         struct ext3_sb_info *sbi = EXT3_SB(sb);
2544         ext3_fsblk_t n_blocks_count = 0;
2545         unsigned long old_sb_flags;
2546         struct ext3_mount_options old_opts;
2547         int enable_quota = 0;
2548         int err;
2549 #ifdef CONFIG_QUOTA
2550         int i;
2551 #endif
2552
2553         lock_kernel();
2554
2555         /* Store the original options */
2556         lock_super(sb);
2557         old_sb_flags = sb->s_flags;
2558         old_opts.s_mount_opt = sbi->s_mount_opt;
2559         old_opts.s_resuid = sbi->s_resuid;
2560         old_opts.s_resgid = sbi->s_resgid;
2561         old_opts.s_commit_interval = sbi->s_commit_interval;
2562 #ifdef CONFIG_QUOTA
2563         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2564         for (i = 0; i < MAXQUOTAS; i++)
2565                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2566 #endif
2567
2568         /*
2569          * Allow the "check" option to be passed as a remount option.
2570          */
2571         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2572                 err = -EINVAL;
2573                 goto restore_opts;
2574         }
2575
2576         if (test_opt(sb, ABORT))
2577                 ext3_abort(sb, __func__, "Abort forced by user");
2578
2579         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2580                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2581
2582         es = sbi->s_es;
2583
2584         ext3_init_journal_params(sb, sbi->s_journal);
2585
2586         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2587                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2588                 if (test_opt(sb, ABORT)) {
2589                         err = -EROFS;
2590                         goto restore_opts;
2591                 }
2592
2593                 if (*flags & MS_RDONLY) {
2594                         err = dquot_suspend(sb, -1);
2595                         if (err < 0)
2596                                 goto restore_opts;
2597
2598                         /*
2599                          * First of all, the unconditional stuff we have to do
2600                          * to disable replay of the journal when we next remount
2601                          */
2602                         sb->s_flags |= MS_RDONLY;
2603
2604                         /*
2605                          * OK, test if we are remounting a valid rw partition
2606                          * readonly, and if so set the rdonly flag and then
2607                          * mark the partition as valid again.
2608                          */
2609                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2610                             (sbi->s_mount_state & EXT3_VALID_FS))
2611                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2612
2613                         ext3_mark_recovery_complete(sb, es);
2614                 } else {
2615                         __le32 ret;
2616                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2617                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2618                                 ext3_msg(sb, KERN_WARNING,
2619                                         "warning: couldn't remount RDWR "
2620                                         "because of unsupported optional "
2621                                         "features (%x)", le32_to_cpu(ret));
2622                                 err = -EROFS;
2623                                 goto restore_opts;
2624                         }
2625
2626                         /*
2627                          * If we have an unprocessed orphan list hanging
2628                          * around from a previously readonly bdev mount,
2629                          * require a full umount/remount for now.
2630                          */
2631                         if (es->s_last_orphan) {
2632                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2633                                        "remount RDWR because of unprocessed "
2634                                        "orphan inode list.  Please "
2635                                        "umount/remount instead.");
2636                                 err = -EINVAL;
2637                                 goto restore_opts;
2638                         }
2639
2640                         /*
2641                          * Mounting a RDONLY partition read-write, so reread
2642                          * and store the current valid flag.  (It may have
2643                          * been changed by e2fsck since we originally mounted
2644                          * the partition.)
2645                          */
2646                         ext3_clear_journal_err(sb, es);
2647                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2648                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2649                                 goto restore_opts;
2650                         if (!ext3_setup_super (sb, es, 0))
2651                                 sb->s_flags &= ~MS_RDONLY;
2652                         enable_quota = 1;
2653                 }
2654         }
2655 #ifdef CONFIG_QUOTA
2656         /* Release old quota file names */
2657         for (i = 0; i < MAXQUOTAS; i++)
2658                 if (old_opts.s_qf_names[i] &&
2659                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2660                         kfree(old_opts.s_qf_names[i]);
2661 #endif
2662         unlock_super(sb);
2663         unlock_kernel();
2664
2665         if (enable_quota)
2666                 dquot_resume(sb, -1);
2667         return 0;
2668 restore_opts:
2669         sb->s_flags = old_sb_flags;
2670         sbi->s_mount_opt = old_opts.s_mount_opt;
2671         sbi->s_resuid = old_opts.s_resuid;
2672         sbi->s_resgid = old_opts.s_resgid;
2673         sbi->s_commit_interval = old_opts.s_commit_interval;
2674 #ifdef CONFIG_QUOTA
2675         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2676         for (i = 0; i < MAXQUOTAS; i++) {
2677                 if (sbi->s_qf_names[i] &&
2678                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2679                         kfree(sbi->s_qf_names[i]);
2680                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2681         }
2682 #endif
2683         unlock_super(sb);
2684         unlock_kernel();
2685         return err;
2686 }
2687
2688 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2689 {
2690         struct super_block *sb = dentry->d_sb;
2691         struct ext3_sb_info *sbi = EXT3_SB(sb);
2692         struct ext3_super_block *es = sbi->s_es;
2693         u64 fsid;
2694
2695         if (test_opt(sb, MINIX_DF)) {
2696                 sbi->s_overhead_last = 0;
2697         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2698                 unsigned long ngroups = sbi->s_groups_count, i;
2699                 ext3_fsblk_t overhead = 0;
2700                 smp_rmb();
2701
2702                 /*
2703                  * Compute the overhead (FS structures).  This is constant
2704                  * for a given filesystem unless the number of block groups
2705                  * changes so we cache the previous value until it does.
2706                  */
2707
2708                 /*
2709                  * All of the blocks before first_data_block are
2710                  * overhead
2711                  */
2712                 overhead = le32_to_cpu(es->s_first_data_block);
2713
2714                 /*
2715                  * Add the overhead attributed to the superblock and
2716                  * block group descriptors.  If the sparse superblocks
2717                  * feature is turned on, then not all groups have this.
2718                  */
2719                 for (i = 0; i < ngroups; i++) {
2720                         overhead += ext3_bg_has_super(sb, i) +
2721                                 ext3_bg_num_gdb(sb, i);
2722                         cond_resched();
2723                 }
2724
2725                 /*
2726                  * Every block group has an inode bitmap, a block
2727                  * bitmap, and an inode table.
2728                  */
2729                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2730                 sbi->s_overhead_last = overhead;
2731                 smp_wmb();
2732                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2733         }
2734
2735         buf->f_type = EXT3_SUPER_MAGIC;
2736         buf->f_bsize = sb->s_blocksize;
2737         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2738         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2739         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2740         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2741                 buf->f_bavail = 0;
2742         buf->f_files = le32_to_cpu(es->s_inodes_count);
2743         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2744         buf->f_namelen = EXT3_NAME_LEN;
2745         fsid = le64_to_cpup((void *)es->s_uuid) ^
2746                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2747         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2748         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2749         return 0;
2750 }
2751
2752 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2753  * is locked for write. Otherwise the are possible deadlocks:
2754  * Process 1                         Process 2
2755  * ext3_create()                     quota_sync()
2756  *   journal_start()                   write_dquot()
2757  *   dquot_initialize()                       down(dqio_mutex)
2758  *     down(dqio_mutex)                    journal_start()
2759  *
2760  */
2761
2762 #ifdef CONFIG_QUOTA
2763
2764 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2765 {
2766         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2767 }
2768
2769 static int ext3_write_dquot(struct dquot *dquot)
2770 {
2771         int ret, err;
2772         handle_t *handle;
2773         struct inode *inode;
2774
2775         inode = dquot_to_inode(dquot);
2776         handle = ext3_journal_start(inode,
2777                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2778         if (IS_ERR(handle))
2779                 return PTR_ERR(handle);
2780         ret = dquot_commit(dquot);
2781         err = ext3_journal_stop(handle);
2782         if (!ret)
2783                 ret = err;
2784         return ret;
2785 }
2786
2787 static int ext3_acquire_dquot(struct dquot *dquot)
2788 {
2789         int ret, err;
2790         handle_t *handle;
2791
2792         handle = ext3_journal_start(dquot_to_inode(dquot),
2793                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2794         if (IS_ERR(handle))
2795                 return PTR_ERR(handle);
2796         ret = dquot_acquire(dquot);
2797         err = ext3_journal_stop(handle);
2798         if (!ret)
2799                 ret = err;
2800         return ret;
2801 }
2802
2803 static int ext3_release_dquot(struct dquot *dquot)
2804 {
2805         int ret, err;
2806         handle_t *handle;
2807
2808         handle = ext3_journal_start(dquot_to_inode(dquot),
2809                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2810         if (IS_ERR(handle)) {
2811                 /* Release dquot anyway to avoid endless cycle in dqput() */
2812                 dquot_release(dquot);
2813                 return PTR_ERR(handle);
2814         }
2815         ret = dquot_release(dquot);
2816         err = ext3_journal_stop(handle);
2817         if (!ret)
2818                 ret = err;
2819         return ret;
2820 }
2821
2822 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2823 {
2824         /* Are we journaling quotas? */
2825         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2826             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2827                 dquot_mark_dquot_dirty(dquot);
2828                 return ext3_write_dquot(dquot);
2829         } else {
2830                 return dquot_mark_dquot_dirty(dquot);
2831         }
2832 }
2833
2834 static int ext3_write_info(struct super_block *sb, int type)
2835 {
2836         int ret, err;
2837         handle_t *handle;
2838
2839         /* Data block + inode block */
2840         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2841         if (IS_ERR(handle))
2842                 return PTR_ERR(handle);
2843         ret = dquot_commit_info(sb, type);
2844         err = ext3_journal_stop(handle);
2845         if (!ret)
2846                 ret = err;
2847         return ret;
2848 }
2849
2850 /*
2851  * Turn on quotas during mount time - we need to find
2852  * the quota file and such...
2853  */
2854 static int ext3_quota_on_mount(struct super_block *sb, int type)
2855 {
2856         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2857                                         EXT3_SB(sb)->s_jquota_fmt, type);
2858 }
2859
2860 /*
2861  * Standard function to be called on quota_on
2862  */
2863 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2864                          char *name)
2865 {
2866         int err;
2867         struct path path;
2868
2869         if (!test_opt(sb, QUOTA))
2870                 return -EINVAL;
2871
2872         err = kern_path(name, LOOKUP_FOLLOW, &path);
2873         if (err)
2874                 return err;
2875
2876         /* Quotafile not on the same filesystem? */
2877         if (path.mnt->mnt_sb != sb) {
2878                 path_put(&path);
2879                 return -EXDEV;
2880         }
2881         /* Journaling quota? */
2882         if (EXT3_SB(sb)->s_qf_names[type]) {
2883                 /* Quotafile not of fs root? */
2884                 if (path.dentry->d_parent != sb->s_root)
2885                         ext3_msg(sb, KERN_WARNING,
2886                                 "warning: Quota file not on filesystem root. "
2887                                 "Journaled quota will not work.");
2888         }
2889
2890         /*
2891          * When we journal data on quota file, we have to flush journal to see
2892          * all updates to the file when we bypass pagecache...
2893          */
2894         if (ext3_should_journal_data(path.dentry->d_inode)) {
2895                 /*
2896                  * We don't need to lock updates but journal_flush() could
2897                  * otherwise be livelocked...
2898                  */
2899                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2900                 err = journal_flush(EXT3_SB(sb)->s_journal);
2901                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2902                 if (err) {
2903                         path_put(&path);
2904                         return err;
2905                 }
2906         }
2907
2908         err = dquot_quota_on_path(sb, type, format_id, &path);
2909         path_put(&path);
2910         return err;
2911 }
2912
2913 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2914  * acquiring the locks... As quota files are never truncated and quota code
2915  * itself serializes the operations (and noone else should touch the files)
2916  * we don't have to be afraid of races */
2917 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2918                                size_t len, loff_t off)
2919 {
2920         struct inode *inode = sb_dqopt(sb)->files[type];
2921         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2922         int err = 0;
2923         int offset = off & (sb->s_blocksize - 1);
2924         int tocopy;
2925         size_t toread;
2926         struct buffer_head *bh;
2927         loff_t i_size = i_size_read(inode);
2928
2929         if (off > i_size)
2930                 return 0;
2931         if (off+len > i_size)
2932                 len = i_size-off;
2933         toread = len;
2934         while (toread > 0) {
2935                 tocopy = sb->s_blocksize - offset < toread ?
2936                                 sb->s_blocksize - offset : toread;
2937                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2938                 if (err)
2939                         return err;
2940                 if (!bh)        /* A hole? */
2941                         memset(data, 0, tocopy);
2942                 else
2943                         memcpy(data, bh->b_data+offset, tocopy);
2944                 brelse(bh);
2945                 offset = 0;
2946                 toread -= tocopy;
2947                 data += tocopy;
2948                 blk++;
2949         }
2950         return len;
2951 }
2952
2953 /* Write to quotafile (we know the transaction is already started and has
2954  * enough credits) */
2955 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2956                                 const char *data, size_t len, loff_t off)
2957 {
2958         struct inode *inode = sb_dqopt(sb)->files[type];
2959         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2960         int err = 0;
2961         int offset = off & (sb->s_blocksize - 1);
2962         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2963         struct buffer_head *bh;
2964         handle_t *handle = journal_current_handle();
2965
2966         if (!handle) {
2967                 ext3_msg(sb, KERN_WARNING,
2968                         "warning: quota write (off=%llu, len=%llu)"
2969                         " cancelled because transaction is not started.",
2970                         (unsigned long long)off, (unsigned long long)len);
2971                 return -EIO;
2972         }
2973
2974         /*
2975          * Since we account only one data block in transaction credits,
2976          * then it is impossible to cross a block boundary.
2977          */
2978         if (sb->s_blocksize - offset < len) {
2979                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2980                         " cancelled because not block aligned",
2981                         (unsigned long long)off, (unsigned long long)len);
2982                 return -EIO;
2983         }
2984         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2985         bh = ext3_bread(handle, inode, blk, 1, &err);
2986         if (!bh)
2987                 goto out;
2988         if (journal_quota) {
2989                 err = ext3_journal_get_write_access(handle, bh);
2990                 if (err) {
2991                         brelse(bh);
2992                         goto out;
2993                 }
2994         }
2995         lock_buffer(bh);
2996         memcpy(bh->b_data+offset, data, len);
2997         flush_dcache_page(bh->b_page);
2998         unlock_buffer(bh);
2999         if (journal_quota)
3000                 err = ext3_journal_dirty_metadata(handle, bh);
3001         else {
3002                 /* Always do at least ordered writes for quotas */
3003                 err = ext3_journal_dirty_data(handle, bh);
3004                 mark_buffer_dirty(bh);
3005         }
3006         brelse(bh);
3007 out:
3008         if (err) {
3009                 mutex_unlock(&inode->i_mutex);
3010                 return err;
3011         }
3012         if (inode->i_size < off + len) {
3013                 i_size_write(inode, off + len);
3014                 EXT3_I(inode)->i_disksize = inode->i_size;
3015         }
3016         inode->i_version++;
3017         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3018         ext3_mark_inode_dirty(handle, inode);
3019         mutex_unlock(&inode->i_mutex);
3020         return len;
3021 }
3022
3023 #endif
3024
3025 static int ext3_get_sb(struct file_system_type *fs_type,
3026         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3027 {
3028         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3029 }
3030
3031 static struct file_system_type ext3_fs_type = {
3032         .owner          = THIS_MODULE,
3033         .name           = "ext3",
3034         .get_sb         = ext3_get_sb,
3035         .kill_sb        = kill_block_super,
3036         .fs_flags       = FS_REQUIRES_DEV,
3037 };
3038
3039 static int __init init_ext3_fs(void)
3040 {
3041         int err = init_ext3_xattr();
3042         if (err)
3043                 return err;
3044         err = init_inodecache();
3045         if (err)
3046                 goto out1;
3047         err = register_filesystem(&ext3_fs_type);
3048         if (err)
3049                 goto out;
3050         return 0;
3051 out:
3052         destroy_inodecache();
3053 out1:
3054         exit_ext3_xattr();
3055         return err;
3056 }
3057
3058 static void __exit exit_ext3_fs(void)
3059 {
3060         unregister_filesystem(&ext3_fs_type);
3061         destroy_inodecache();
3062         exit_ext3_xattr();
3063 }
3064
3065 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3066 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3067 MODULE_LICENSE("GPL");
3068 module_init(init_ext3_fs)
3069 module_exit(exit_ext3_fs)