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/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.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
  40 #include <asm/uaccess.h>
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44 #include "namei.h"
  45
  46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  47                              unsigned long journal_devnum);
  48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  49                                int);
  50 static void ext3_commit_super (struct super_block * sb,
  51                                struct ext3_super_block * es,
  52                                int sync);
  53 static void ext3_mark_recovery_complete(struct super_block * sb,
  54                                         struct ext3_super_block * es);
  55 static void ext3_clear_journal_err(struct super_block * sb,
  56                                    struct ext3_super_block * es);
  57 static int ext3_sync_fs(struct super_block *sb, int wait);
  58 static const char *ext3_decode_error(struct super_block * sb, int errno,
  59                                      char nbuf[16]);
  60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  61 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  62 static void ext3_unlockfs(struct super_block *sb);
  63 static void ext3_write_super (struct super_block * sb);
  64 static void ext3_write_super_lockfs(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  68  *
  69  * The only special thing we need to do here is to make sure that all
  70  * journal_end calls result in the superblock being marked dirty, so
  71  * that sync() will call the filesystem's write_super callback if
  72  * appropriate.
  73  */
  74 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  75 {
  76         journal_t *journal;
  77
  78         if (sb->s_flags & MS_RDONLY)
  79                 return ERR_PTR(-EROFS);
  80
  81         /* Special case here: if the journal has aborted behind our
  82          * backs (eg. EIO in the commit thread), then we still need to
  83          * take the FS itself readonly cleanly. */
  84         journal = EXT3_SB(sb)->s_journal;
  85         if (is_journal_aborted(journal)) {
  86                 ext3_abort(sb, __FUNCTION__,
  87                            "Detected aborted journal");
  88                 return ERR_PTR(-EROFS);
  89         }
  90
  91         return journal_start(journal, nblocks);
  92 }
  93
  94 /*
  95  * The only special thing we need to do here is to make sure that all
  96  * journal_stop calls result in the superblock being marked dirty, so
  97  * that sync() will call the filesystem's write_super callback if
  98  * appropriate.
  99  */
 100 int __ext3_journal_stop(const char *where, handle_t *handle)
 101 {
 102         struct super_block *sb;
 103         int err;
 104         int rc;
 105
 106         sb = handle->h_transaction->t_journal->j_private;
 107         err = handle->h_err;
 108         rc = journal_stop(handle);
 109
 110         if (!err)
 111                 err = rc;
 112         if (err)
 113                 __ext3_std_error(sb, where, err);
 114         return err;
 115 }
 116
 117 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 118                 struct buffer_head *bh, handle_t *handle, int err)
 119 {
 120         char nbuf[16];
 121         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 122
 123         if (bh)
 124                 BUFFER_TRACE(bh, "abort");
 125
 126         if (!handle->h_err)
 127                 handle->h_err = err;
 128
 129         if (is_handle_aborted(handle))
 130                 return;
 131
 132         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 133                caller, errstr, err_fn);
 134
 135         journal_abort_handle(handle);
 136 }
 137
 138 /* Deal with the reporting of failure conditions on a filesystem such as
 139  * inconsistencies detected or read IO failures.
 140  *
 141  * On ext2, we can store the error state of the filesystem in the
 142  * superblock.  That is not possible on ext3, because we may have other
 143  * write ordering constraints on the superblock which prevent us from
 144  * writing it out straight away; and given that the journal is about to
 145  * be aborted, we can't rely on the current, or future, transactions to
 146  * write out the superblock safely.
 147  *
 148  * We'll just use the journal_abort() error code to record an error in
 149  * the journal instead.  On recovery, the journal will compain about
 150  * that error until we've noted it down and cleared it.
 151  */
 152
 153 static void ext3_handle_error(struct super_block *sb)
 154 {
 155         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 156
 157         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 158         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 159
 160         if (sb->s_flags & MS_RDONLY)
 161                 return;
 162
 163         if (test_opt (sb, ERRORS_RO)) {
 164                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 165                 sb->s_flags |= MS_RDONLY;
 166         } else {
 167                 journal_t *journal = EXT3_SB(sb)->s_journal;
 168
 169                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 170                 if (journal)
 171                         journal_abort(journal, -EIO);
 172         }
 173         if (test_opt(sb, ERRORS_PANIC))
 174                 panic("EXT3-fs (device %s): panic forced after error\n",
 175                         sb->s_id);
 176         ext3_commit_super(sb, es, 1);
 177 }
 178
 179 void ext3_error (struct super_block * sb, const char * function,
 180                  const char * fmt, ...)
 181 {
 182         va_list args;
 183
 184         va_start(args, fmt);
 185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 186         vprintk(fmt, args);
 187         printk("\n");
 188         va_end(args);
 189
 190         ext3_handle_error(sb);
 191 }
 192
 193 static const char *ext3_decode_error(struct super_block * sb, int errno,
 194                                      char nbuf[16])
 195 {
 196         char *errstr = NULL;
 197
 198         switch (errno) {
 199         case -EIO:
 200                 errstr = "IO failure";
 201                 break;
 202         case -ENOMEM:
 203                 errstr = "Out of memory";
 204                 break;
 205         case -EROFS:
 206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 207                         errstr = "Journal has aborted";
 208                 else
 209                         errstr = "Readonly filesystem";
 210                 break;
 211         default:
 212                 /* If the caller passed in an extra buffer for unknown
 213                  * errors, textualise them now.  Else we just return
 214                  * NULL. */
 215                 if (nbuf) {
 216                         /* Check for truncated error codes... */
 217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 218                                 errstr = nbuf;
 219                 }
 220                 break;
 221         }
 222
 223         return errstr;
 224 }
 225
 226 /* __ext3_std_error decodes expected errors from journaling functions
 227  * automatically and invokes the appropriate error response.  */
 228
 229 void __ext3_std_error (struct super_block * sb, const char * function,
 230                        int errno)
 231 {
 232         char nbuf[16];
 233         const char *errstr;
 234
 235         /* Special case: if the error is EROFS, and we're not already
 236          * inside a transaction, then there's really no point in logging
 237          * an error. */
 238         if (errno == -EROFS && journal_current_handle() == NULL &&
 239             (sb->s_flags & MS_RDONLY))
 240                 return;
 241
 242         errstr = ext3_decode_error(sb, errno, nbuf);
 243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 244                 sb->s_id, function, errstr);
 245
 246         ext3_handle_error(sb);
 247 }
 248
 249 /*
 250  * ext3_abort is a much stronger failure handler than ext3_error.  The
 251  * abort function may be used to deal with unrecoverable failures such
 252  * as journal IO errors or ENOMEM at a critical moment in log management.
 253  *
 254  * We unconditionally force the filesystem into an ABORT|READONLY state,
 255  * unless the error response on the fs has been set to panic in which
 256  * case we take the easy way out and panic immediately.
 257  */
 258
 259 void ext3_abort (struct super_block * sb, const char * function,
 260                  const char * fmt, ...)
 261 {
 262         va_list args;
 263
 264         printk (KERN_CRIT "ext3_abort called.\n");
 265
 266         va_start(args, fmt);
 267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 268         vprintk(fmt, args);
 269         printk("\n");
 270         va_end(args);
 271
 272         if (test_opt(sb, ERRORS_PANIC))
 273                 panic("EXT3-fs panic from previous error\n");
 274
 275         if (sb->s_flags & MS_RDONLY)
 276                 return;
 277
 278         printk(KERN_CRIT "Remounting filesystem read-only\n");
 279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 280         sb->s_flags |= MS_RDONLY;
 281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 283 }
 284
 285 void ext3_warning (struct super_block * sb, const char * function,
 286                    const char * fmt, ...)
 287 {
 288         va_list args;
 289
 290         va_start(args, fmt);
 291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 292                sb->s_id, function);
 293         vprintk(fmt, args);
 294         printk("\n");
 295         va_end(args);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        NEXT_ORPHAN(inode));
 384         }
 385 }
 386
 387 static void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 408         percpu_counter_destroy(&sbi->s_dirs_counter);
 409         brelse(sbi->s_sbh);
 410 #ifdef CONFIG_QUOTA
 411         for (i = 0; i < MAXQUOTAS; i++)
 412                 kfree(sbi->s_qf_names[i]);
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->i_block_alloc_info = NULL;
 456         ei->vfs_inode.i_version = 1;
 457         return &ei->vfs_inode;
 458 }
 459
 460 static void ext3_destroy_inode(struct inode *inode)
 461 {
 462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 463 }
 464
 465 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 466 {
 467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 468
 469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 470             SLAB_CTOR_CONSTRUCTOR) {
 471                 INIT_LIST_HEAD(&ei->i_orphan);
 472 #ifdef CONFIG_EXT3_FS_XATTR
 473                 init_rwsem(&ei->xattr_sem);
 474 #endif
 475                 mutex_init(&ei->truncate_mutex);
 476                 inode_init_once(&ei->vfs_inode);
 477         }
 478 }
 479
 480 static int init_inodecache(void)
 481 {
 482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 483                                              sizeof(struct ext3_inode_info),
 484                                              0, (SLAB_RECLAIM_ACCOUNT|
 485                                                 SLAB_MEM_SPREAD),
 486                                              init_once, NULL);
 487         if (ext3_inode_cachep == NULL)
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static void destroy_inodecache(void)
 493 {
 494         if (kmem_cache_destroy(ext3_inode_cachep))
 495                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 496 }
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 501 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 502         if (EXT3_I(inode)->i_acl &&
 503                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 504                 posix_acl_release(EXT3_I(inode)->i_acl);
 505                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 506         }
 507         if (EXT3_I(inode)->i_default_acl &&
 508                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 509                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 510                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 511         }
 512 #endif
 513         ext3_discard_reservation(inode);
 514         EXT3_I(inode)->i_block_alloc_info = NULL;
 515         if (unlikely(rsv))
 516                 kfree(rsv);
 517 }
 518
 519 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 520 {
 521 #if defined(CONFIG_QUOTA)
 522         struct ext3_sb_info *sbi = EXT3_SB(sb);
 523
 524         if (sbi->s_jquota_fmt)
 525                 seq_printf(seq, ",jqfmt=%s",
 526                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 527
 528         if (sbi->s_qf_names[USRQUOTA])
 529                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 530
 531         if (sbi->s_qf_names[GRPQUOTA])
 532                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 533
 534         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 535                 seq_puts(seq, ",usrquota");
 536
 537         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 538                 seq_puts(seq, ",grpquota");
 539 #endif
 540 }
 541
 542 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 543 {
 544         struct super_block *sb = vfs->mnt_sb;
 545
 546         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 547                 seq_puts(seq, ",data=journal");
 548         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 549                 seq_puts(seq, ",data=ordered");
 550         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 551                 seq_puts(seq, ",data=writeback");
 552
 553         ext3_show_quota_options(seq, sb);
 554
 555         return 0;
 556 }
 557
 558 #ifdef CONFIG_QUOTA
 559 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 560 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 561
 562 static int ext3_dquot_initialize(struct inode *inode, int type);
 563 static int ext3_dquot_drop(struct inode *inode);
 564 static int ext3_write_dquot(struct dquot *dquot);
 565 static int ext3_acquire_dquot(struct dquot *dquot);
 566 static int ext3_release_dquot(struct dquot *dquot);
 567 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 568 static int ext3_write_info(struct super_block *sb, int type);
 569 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 570 static int ext3_quota_on_mount(struct super_block *sb, int type);
 571 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 572                                size_t len, loff_t off);
 573 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 574                                 const char *data, size_t len, loff_t off);
 575
 576 static struct dquot_operations ext3_quota_operations = {
 577         .initialize     = ext3_dquot_initialize,
 578         .drop           = ext3_dquot_drop,
 579         .alloc_space    = dquot_alloc_space,
 580         .alloc_inode    = dquot_alloc_inode,
 581         .free_space     = dquot_free_space,
 582         .free_inode     = dquot_free_inode,
 583         .transfer       = dquot_transfer,
 584         .write_dquot    = ext3_write_dquot,
 585         .acquire_dquot  = ext3_acquire_dquot,
 586         .release_dquot  = ext3_release_dquot,
 587         .mark_dirty     = ext3_mark_dquot_dirty,
 588         .write_info     = ext3_write_info
 589 };
 590
 591 static struct quotactl_ops ext3_qctl_operations = {
 592         .quota_on       = ext3_quota_on,
 593         .quota_off      = vfs_quota_off,
 594         .quota_sync     = vfs_quota_sync,
 595         .get_info       = vfs_get_dqinfo,
 596         .set_info       = vfs_set_dqinfo,
 597         .get_dqblk      = vfs_get_dqblk,
 598         .set_dqblk      = vfs_set_dqblk
 599 };
 600 #endif
 601
 602 static struct super_operations ext3_sops = {
 603         .alloc_inode    = ext3_alloc_inode,
 604         .destroy_inode  = ext3_destroy_inode,
 605         .read_inode     = ext3_read_inode,
 606         .write_inode    = ext3_write_inode,
 607         .dirty_inode    = ext3_dirty_inode,
 608         .delete_inode   = ext3_delete_inode,
 609         .put_super      = ext3_put_super,
 610         .write_super    = ext3_write_super,
 611         .sync_fs        = ext3_sync_fs,
 612         .write_super_lockfs = ext3_write_super_lockfs,
 613         .unlockfs       = ext3_unlockfs,
 614         .statfs         = ext3_statfs,
 615         .remount_fs     = ext3_remount,
 616         .clear_inode    = ext3_clear_inode,
 617         .show_options   = ext3_show_options,
 618 #ifdef CONFIG_QUOTA
 619         .quota_read     = ext3_quota_read,
 620         .quota_write    = ext3_quota_write,
 621 #endif
 622 };
 623
 624 static struct export_operations ext3_export_ops = {
 625         .get_parent = ext3_get_parent,
 626 };
 627
 628 enum {
 629         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 630         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 631         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 632         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 633         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 634         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 635         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 636         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 637         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 638         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 639         Opt_grpquota
 640 };
 641
 642 static match_table_t tokens = {
 643         {Opt_bsd_df, "bsddf"},
 644         {Opt_minix_df, "minixdf"},
 645         {Opt_grpid, "grpid"},
 646         {Opt_grpid, "bsdgroups"},
 647         {Opt_nogrpid, "nogrpid"},
 648         {Opt_nogrpid, "sysvgroups"},
 649         {Opt_resgid, "resgid=%u"},
 650         {Opt_resuid, "resuid=%u"},
 651         {Opt_sb, "sb=%u"},
 652         {Opt_err_cont, "errors=continue"},
 653         {Opt_err_panic, "errors=panic"},
 654         {Opt_err_ro, "errors=remount-ro"},
 655         {Opt_nouid32, "nouid32"},
 656         {Opt_nocheck, "nocheck"},
 657         {Opt_nocheck, "check=none"},
 658         {Opt_debug, "debug"},
 659         {Opt_oldalloc, "oldalloc"},
 660         {Opt_orlov, "orlov"},
 661         {Opt_user_xattr, "user_xattr"},
 662         {Opt_nouser_xattr, "nouser_xattr"},
 663         {Opt_acl, "acl"},
 664         {Opt_noacl, "noacl"},
 665         {Opt_reservation, "reservation"},
 666         {Opt_noreservation, "noreservation"},
 667         {Opt_noload, "noload"},
 668         {Opt_nobh, "nobh"},
 669         {Opt_commit, "commit=%u"},
 670         {Opt_journal_update, "journal=update"},
 671         {Opt_journal_inum, "journal=%u"},
 672         {Opt_journal_dev, "journal_dev=%u"},
 673         {Opt_abort, "abort"},
 674         {Opt_data_journal, "data=journal"},
 675         {Opt_data_ordered, "data=ordered"},
 676         {Opt_data_writeback, "data=writeback"},
 677         {Opt_offusrjquota, "usrjquota="},
 678         {Opt_usrjquota, "usrjquota=%s"},
 679         {Opt_offgrpjquota, "grpjquota="},
 680         {Opt_grpjquota, "grpjquota=%s"},
 681         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 682         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 683         {Opt_grpquota, "grpquota"},
 684         {Opt_noquota, "noquota"},
 685         {Opt_quota, "quota"},
 686         {Opt_usrquota, "usrquota"},
 687         {Opt_barrier, "barrier=%u"},
 688         {Opt_err, NULL},
 689         {Opt_resize, "resize"},
 690 };
 691
 692 static unsigned long get_sb_block(void **data)
 693 {
 694         unsigned long   sb_block;
 695         char            *options = (char *) *data;
 696
 697         if (!options || strncmp(options, "sb=", 3) != 0)
 698                 return 1;       /* Default location */
 699         options += 3;
 700         sb_block = simple_strtoul(options, &options, 0);
 701         if (*options && *options != ',') {
 702                 printk("EXT3-fs: Invalid sb specification: %s\n",
 703                        (char *) *data);
 704                 return 1;
 705         }
 706         if (*options == ',')
 707                 options++;
 708         *data = (void *) options;
 709         return sb_block;
 710 }
 711
 712 static int parse_options (char *options, struct super_block *sb,
 713                           unsigned long *inum, unsigned long *journal_devnum,
 714                           unsigned long *n_blocks_count, int is_remount)
 715 {
 716         struct ext3_sb_info *sbi = EXT3_SB(sb);
 717         char * p;
 718         substring_t args[MAX_OPT_ARGS];
 719         int data_opt = 0;
 720         int option;
 721 #ifdef CONFIG_QUOTA
 722         int qtype;
 723         char *qname;
 724 #endif
 725
 726         if (!options)
 727                 return 1;
 728
 729         while ((p = strsep (&options, ",")) != NULL) {
 730                 int token;
 731                 if (!*p)
 732                         continue;
 733
 734                 token = match_token(p, tokens, args);
 735                 switch (token) {
 736                 case Opt_bsd_df:
 737                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 738                         break;
 739                 case Opt_minix_df:
 740                         set_opt (sbi->s_mount_opt, MINIX_DF);
 741                         break;
 742                 case Opt_grpid:
 743                         set_opt (sbi->s_mount_opt, GRPID);
 744                         break;
 745                 case Opt_nogrpid:
 746                         clear_opt (sbi->s_mount_opt, GRPID);
 747                         break;
 748                 case Opt_resuid:
 749                         if (match_int(&args[0], &option))
 750                                 return 0;
 751                         sbi->s_resuid = option;
 752                         break;
 753                 case Opt_resgid:
 754                         if (match_int(&args[0], &option))
 755                                 return 0;
 756                         sbi->s_resgid = option;
 757                         break;
 758                 case Opt_sb:
 759                         /* handled by get_sb_block() instead of here */
 760                         /* *sb_block = match_int(&args[0]); */
 761                         break;
 762                 case Opt_err_panic:
 763                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 764                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 765                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 766                         break;
 767                 case Opt_err_ro:
 768                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 769                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 770                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 771                         break;
 772                 case Opt_err_cont:
 773                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 774                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 775                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 776                         break;
 777                 case Opt_nouid32:
 778                         set_opt (sbi->s_mount_opt, NO_UID32);
 779                         break;
 780                 case Opt_nocheck:
 781                         clear_opt (sbi->s_mount_opt, CHECK);
 782                         break;
 783                 case Opt_debug:
 784                         set_opt (sbi->s_mount_opt, DEBUG);
 785                         break;
 786                 case Opt_oldalloc:
 787                         set_opt (sbi->s_mount_opt, OLDALLOC);
 788                         break;
 789                 case Opt_orlov:
 790                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 791                         break;
 792 #ifdef CONFIG_EXT3_FS_XATTR
 793                 case Opt_user_xattr:
 794                         set_opt (sbi->s_mount_opt, XATTR_USER);
 795                         break;
 796                 case Opt_nouser_xattr:
 797                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 798                         break;
 799 #else
 800                 case Opt_user_xattr:
 801                 case Opt_nouser_xattr:
 802                         printk("EXT3 (no)user_xattr options not supported\n");
 803                         break;
 804 #endif
 805 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 806                 case Opt_acl:
 807                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 808                         break;
 809                 case Opt_noacl:
 810                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 811                         break;
 812 #else
 813                 case Opt_acl:
 814                 case Opt_noacl:
 815                         printk("EXT3 (no)acl options not supported\n");
 816                         break;
 817 #endif
 818                 case Opt_reservation:
 819                         set_opt(sbi->s_mount_opt, RESERVATION);
 820                         break;
 821                 case Opt_noreservation:
 822                         clear_opt(sbi->s_mount_opt, RESERVATION);
 823                         break;
 824                 case Opt_journal_update:
 825                         /* @@@ FIXME */
 826                         /* Eventually we will want to be able to create
 827                            a journal file here.  For now, only allow the
 828                            user to specify an existing inode to be the
 829                            journal file. */
 830                         if (is_remount) {
 831                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 832                                        "journal on remount\n");
 833                                 return 0;
 834                         }
 835                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 836                         break;
 837                 case Opt_journal_inum:
 838                         if (is_remount) {
 839                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 840                                        "journal on remount\n");
 841                                 return 0;
 842                         }
 843                         if (match_int(&args[0], &option))
 844                                 return 0;
 845                         *inum = option;
 846                         break;
 847                 case Opt_journal_dev:
 848                         if (is_remount) {
 849                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 850                                        "journal on remount\n");
 851                                 return 0;
 852                         }
 853                         if (match_int(&args[0], &option))
 854                                 return 0;
 855                         *journal_devnum = option;
 856                         break;
 857                 case Opt_noload:
 858                         set_opt (sbi->s_mount_opt, NOLOAD);
 859                         break;
 860                 case Opt_commit:
 861                         if (match_int(&args[0], &option))
 862                                 return 0;
 863                         if (option < 0)
 864                                 return 0;
 865                         if (option == 0)
 866                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 867                         sbi->s_commit_interval = HZ * option;
 868                         break;
 869                 case Opt_data_journal:
 870                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 871                         goto datacheck;
 872                 case Opt_data_ordered:
 873                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 874                         goto datacheck;
 875                 case Opt_data_writeback:
 876                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 877                 datacheck:
 878                         if (is_remount) {
 879                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 880                                                 != data_opt) {
 881                                         printk(KERN_ERR
 882                                                 "EXT3-fs: cannot change data "
 883                                                 "mode on remount\n");
 884                                         return 0;
 885                                 }
 886                         } else {
 887                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 888                                 sbi->s_mount_opt |= data_opt;
 889                         }
 890                         break;
 891 #ifdef CONFIG_QUOTA
 892                 case Opt_usrjquota:
 893                         qtype = USRQUOTA;
 894                         goto set_qf_name;
 895                 case Opt_grpjquota:
 896                         qtype = GRPQUOTA;
 897 set_qf_name:
 898                         if (sb_any_quota_enabled(sb)) {
 899                                 printk(KERN_ERR
 900                                         "EXT3-fs: Cannot change journalled "
 901                                         "quota options when quota turned on.\n");
 902                                 return 0;
 903                         }
 904                         qname = match_strdup(&args[0]);
 905                         if (!qname) {
 906                                 printk(KERN_ERR
 907                                         "EXT3-fs: not enough memory for "
 908                                         "storing quotafile name.\n");
 909                                 return 0;
 910                         }
 911                         if (sbi->s_qf_names[qtype] &&
 912                             strcmp(sbi->s_qf_names[qtype], qname)) {
 913                                 printk(KERN_ERR
 914                                         "EXT3-fs: %s quota file already "
 915                                         "specified.\n", QTYPE2NAME(qtype));
 916                                 kfree(qname);
 917                                 return 0;
 918                         }
 919                         sbi->s_qf_names[qtype] = qname;
 920                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 921                                 printk(KERN_ERR
 922                                         "EXT3-fs: quotafile must be on "
 923                                         "filesystem root.\n");
 924                                 kfree(sbi->s_qf_names[qtype]);
 925                                 sbi->s_qf_names[qtype] = NULL;
 926                                 return 0;
 927                         }
 928                         set_opt(sbi->s_mount_opt, QUOTA);
 929                         break;
 930                 case Opt_offusrjquota:
 931                         qtype = USRQUOTA;
 932                         goto clear_qf_name;
 933                 case Opt_offgrpjquota:
 934                         qtype = GRPQUOTA;
 935 clear_qf_name:
 936                         if (sb_any_quota_enabled(sb)) {
 937                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 938                                         "journalled quota options when "
 939                                         "quota turned on.\n");
 940                                 return 0;
 941                         }
 942                         /*
 943                          * The space will be released later when all options
 944                          * are confirmed to be correct
 945                          */
 946                         sbi->s_qf_names[qtype] = NULL;
 947                         break;
 948                 case Opt_jqfmt_vfsold:
 949                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 950                         break;
 951                 case Opt_jqfmt_vfsv0:
 952                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 953                         break;
 954                 case Opt_quota:
 955                 case Opt_usrquota:
 956                         set_opt(sbi->s_mount_opt, QUOTA);
 957                         set_opt(sbi->s_mount_opt, USRQUOTA);
 958                         break;
 959                 case Opt_grpquota:
 960                         set_opt(sbi->s_mount_opt, QUOTA);
 961                         set_opt(sbi->s_mount_opt, GRPQUOTA);
 962                         break;
 963                 case Opt_noquota:
 964                         if (sb_any_quota_enabled(sb)) {
 965                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
 966                                         "options when quota turned on.\n");
 967                                 return 0;
 968                         }
 969                         clear_opt(sbi->s_mount_opt, QUOTA);
 970                         clear_opt(sbi->s_mount_opt, USRQUOTA);
 971                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
 972                         break;
 973 #else
 974                 case Opt_quota:
 975                 case Opt_usrquota:
 976                 case Opt_grpquota:
 977                 case Opt_usrjquota:
 978                 case Opt_grpjquota:
 979                 case Opt_offusrjquota:
 980                 case Opt_offgrpjquota:
 981                 case Opt_jqfmt_vfsold:
 982                 case Opt_jqfmt_vfsv0:
 983                         printk(KERN_ERR
 984                                 "EXT3-fs: journalled quota options not "
 985                                 "supported.\n");
 986                         break;
 987                 case Opt_noquota:
 988                         break;
 989 #endif
 990                 case Opt_abort:
 991                         set_opt(sbi->s_mount_opt, ABORT);
 992                         break;
 993                 case Opt_barrier:
 994                         if (match_int(&args[0], &option))
 995                                 return 0;
 996                         if (option)
 997                                 set_opt(sbi->s_mount_opt, BARRIER);
 998                         else
 999                                 clear_opt(sbi->s_mount_opt, BARRIER);
1000                         break;
1001                 case Opt_ignore:
1002                         break;
1003                 case Opt_resize:
1004                         if (!is_remount) {
1005                                 printk("EXT3-fs: resize option only available "
1006                                         "for remount\n");
1007                                 return 0;
1008                         }
1009                         if (match_int(&args[0], &option) != 0)
1010                                 return 0;
1011                         *n_blocks_count = option;
1012                         break;
1013                 case Opt_nobh:
1014                         set_opt(sbi->s_mount_opt, NOBH);
1015                         break;
1016                 default:
1017                         printk (KERN_ERR
1018                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1019                                 "or missing value\n", p);
1020                         return 0;
1021                 }
1022         }
1023 #ifdef CONFIG_QUOTA
1024         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1025                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1026                      sbi->s_qf_names[USRQUOTA])
1027                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1028
1029                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1030                      sbi->s_qf_names[GRPQUOTA])
1031                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1032
1033                 if ((sbi->s_qf_names[USRQUOTA] &&
1034                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1035                     (sbi->s_qf_names[GRPQUOTA] &&
1036                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1037                         printk(KERN_ERR "EXT3-fs: old and new quota "
1038                                         "format mixing.\n");
1039                         return 0;
1040                 }
1041
1042                 if (!sbi->s_jquota_fmt) {
1043                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1044                                         "not specified.\n");
1045                         return 0;
1046                 }
1047         } else {
1048                 if (sbi->s_jquota_fmt) {
1049                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1050                                         "specified with no journalling "
1051                                         "enabled.\n");
1052                         return 0;
1053                 }
1054         }
1055 #endif
1056         return 1;
1057 }
1058
1059 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1060                             int read_only)
1061 {
1062         struct ext3_sb_info *sbi = EXT3_SB(sb);
1063         int res = 0;
1064
1065         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1066                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1067                         "forcing read-only mode\n");
1068                 res = MS_RDONLY;
1069         }
1070         if (read_only)
1071                 return res;
1072         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1073                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1074                         "running e2fsck is recommended\n");
1075         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1076                 printk (KERN_WARNING
1077                         "EXT3-fs warning: mounting fs with errors, "
1078                         "running e2fsck is recommended\n");
1079         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1080                  le16_to_cpu(es->s_mnt_count) >=
1081                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1082                 printk (KERN_WARNING
1083                         "EXT3-fs warning: maximal mount count reached, "
1084                         "running e2fsck is recommended\n");
1085         else if (le32_to_cpu(es->s_checkinterval) &&
1086                 (le32_to_cpu(es->s_lastcheck) +
1087                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1088                 printk (KERN_WARNING
1089                         "EXT3-fs warning: checktime reached, "
1090                         "running e2fsck is recommended\n");
1091 #if 0
1092                 /* @@@ We _will_ want to clear the valid bit if we find
1093                    inconsistencies, to force a fsck at reboot.  But for
1094                    a plain journaled filesystem we can keep it set as
1095                    valid forever! :) */
1096         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1097 #endif
1098         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1099                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1100         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1101         es->s_mtime = cpu_to_le32(get_seconds());
1102         ext3_update_dynamic_rev(sb);
1103         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1104
1105         ext3_commit_super(sb, es, 1);
1106         if (test_opt(sb, DEBUG))
1107                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1108                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1109                         sb->s_blocksize,
1110                         sbi->s_groups_count,
1111                         EXT3_BLOCKS_PER_GROUP(sb),
1112                         EXT3_INODES_PER_GROUP(sb),
1113                         sbi->s_mount_opt);
1114
1115         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1116         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1117                 char b[BDEVNAME_SIZE];
1118
1119                 printk("external journal on %s\n",
1120                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1121         } else {
1122                 printk("internal journal\n");
1123         }
1124         return res;
1125 }
1126
1127 /* Called at mount-time, super-block is locked */
1128 static int ext3_check_descriptors (struct super_block * sb)
1129 {
1130         struct ext3_sb_info *sbi = EXT3_SB(sb);
1131         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1132         struct ext3_group_desc * gdp = NULL;
1133         int desc_block = 0;
1134         int i;
1135
1136         ext3_debug ("Checking group descriptors");
1137
1138         for (i = 0; i < sbi->s_groups_count; i++)
1139         {
1140                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1141                         gdp = (struct ext3_group_desc *)
1142                                         sbi->s_group_desc[desc_block++]->b_data;
1143                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1144                     le32_to_cpu(gdp->bg_block_bitmap) >=
1145                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1146                 {
1147                         ext3_error (sb, "ext3_check_descriptors",
1148                                     "Block bitmap for group %d"
1149                                     " not in group (block %lu)!",
1150                                     i, (unsigned long)
1151                                         le32_to_cpu(gdp->bg_block_bitmap));
1152                         return 0;
1153                 }
1154                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1155                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1156                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1157                 {
1158                         ext3_error (sb, "ext3_check_descriptors",
1159                                     "Inode bitmap for group %d"
1160                                     " not in group (block %lu)!",
1161                                     i, (unsigned long)
1162                                         le32_to_cpu(gdp->bg_inode_bitmap));
1163                         return 0;
1164                 }
1165                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1166                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1167                     block + EXT3_BLOCKS_PER_GROUP(sb))
1168                 {
1169                         ext3_error (sb, "ext3_check_descriptors",
1170                                     "Inode table for group %d"
1171                                     " not in group (block %lu)!",
1172                                     i, (unsigned long)
1173                                         le32_to_cpu(gdp->bg_inode_table));
1174                         return 0;
1175                 }
1176                 block += EXT3_BLOCKS_PER_GROUP(sb);
1177                 gdp++;
1178         }
1179
1180         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1181         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1182         return 1;
1183 }
1184
1185
1186 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1187  * the superblock) which were deleted from all directories, but held open by
1188  * a process at the time of a crash.  We walk the list and try to delete these
1189  * inodes at recovery time (only with a read-write filesystem).
1190  *
1191  * In order to keep the orphan inode chain consistent during traversal (in
1192  * case of crash during recovery), we link each inode into the superblock
1193  * orphan list_head and handle it the same way as an inode deletion during
1194  * normal operation (which journals the operations for us).
1195  *
1196  * We only do an iget() and an iput() on each inode, which is very safe if we
1197  * accidentally point at an in-use or already deleted inode.  The worst that
1198  * can happen in this case is that we get a "bit already cleared" message from
1199  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1200  * e2fsck was run on this filesystem, and it must have already done the orphan
1201  * inode cleanup for us, so we can safely abort without any further action.
1202  */
1203 static void ext3_orphan_cleanup (struct super_block * sb,
1204                                  struct ext3_super_block * es)
1205 {
1206         unsigned int s_flags = sb->s_flags;
1207         int nr_orphans = 0, nr_truncates = 0;
1208 #ifdef CONFIG_QUOTA
1209         int i;
1210 #endif
1211         if (!es->s_last_orphan) {
1212                 jbd_debug(4, "no orphan inodes to clean up\n");
1213                 return;
1214         }
1215
1216         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1217                 if (es->s_last_orphan)
1218                         jbd_debug(1, "Errors on filesystem, "
1219                                   "clearing orphan list.\n");
1220                 es->s_last_orphan = 0;
1221                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1222                 return;
1223         }
1224
1225         if (s_flags & MS_RDONLY) {
1226                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1227                        sb->s_id);
1228                 sb->s_flags &= ~MS_RDONLY;
1229         }
1230 #ifdef CONFIG_QUOTA
1231         /* Needed for iput() to work correctly and not trash data */
1232         sb->s_flags |= MS_ACTIVE;
1233         /* Turn on quotas so that they are updated correctly */
1234         for (i = 0; i < MAXQUOTAS; i++) {
1235                 if (EXT3_SB(sb)->s_qf_names[i]) {
1236                         int ret = ext3_quota_on_mount(sb, i);
1237                         if (ret < 0)
1238                                 printk(KERN_ERR
1239                                         "EXT3-fs: Cannot turn on journalled "
1240                                         "quota: error %d\n", ret);
1241                 }
1242         }
1243 #endif
1244
1245         while (es->s_last_orphan) {
1246                 struct inode *inode;
1247
1248                 if (!(inode =
1249                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1250                         es->s_last_orphan = 0;
1251                         break;
1252                 }
1253
1254                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1255                 DQUOT_INIT(inode);
1256                 if (inode->i_nlink) {
1257                         printk(KERN_DEBUG
1258                                 "%s: truncating inode %ld to %Ld bytes\n",
1259                                 __FUNCTION__, inode->i_ino, inode->i_size);
1260                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1261                                   inode->i_ino, inode->i_size);
1262                         ext3_truncate(inode);
1263                         nr_truncates++;
1264                 } else {
1265                         printk(KERN_DEBUG
1266                                 "%s: deleting unreferenced inode %ld\n",
1267                                 __FUNCTION__, inode->i_ino);
1268                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1269                                   inode->i_ino);
1270                         nr_orphans++;
1271                 }
1272                 iput(inode);  /* The delete magic happens here! */
1273         }
1274
1275 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1276
1277         if (nr_orphans)
1278                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1279                        sb->s_id, PLURAL(nr_orphans));
1280         if (nr_truncates)
1281                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1282                        sb->s_id, PLURAL(nr_truncates));
1283 #ifdef CONFIG_QUOTA
1284         /* Turn quotas off */
1285         for (i = 0; i < MAXQUOTAS; i++) {
1286                 if (sb_dqopt(sb)->files[i])
1287                         vfs_quota_off(sb, i);
1288         }
1289 #endif
1290         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1291 }
1292
1293 #define log2(n) ffz(~(n))
1294
1295 /*
1296  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1297  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1298  * We need to be 1 filesystem block less than the 2^32 sector limit.
1299  */
1300 static loff_t ext3_max_size(int bits)
1301 {
1302         loff_t res = EXT3_NDIR_BLOCKS;
1303         /* This constant is calculated to be the largest file size for a
1304          * dense, 4k-blocksize file such that the total number of
1305          * sectors in the file, including data and all indirect blocks,
1306          * does not exceed 2^32. */
1307         const loff_t upper_limit = 0x1ff7fffd000LL;
1308
1309         res += 1LL << (bits-2);
1310         res += 1LL << (2*(bits-2));
1311         res += 1LL << (3*(bits-2));
1312         res <<= bits;
1313         if (res > upper_limit)
1314                 res = upper_limit;
1315         return res;
1316 }
1317
1318 static unsigned long descriptor_loc(struct super_block *sb,
1319                                     unsigned long logic_sb_block,
1320                                     int nr)
1321 {
1322         struct ext3_sb_info *sbi = EXT3_SB(sb);
1323         unsigned long bg, first_data_block, first_meta_bg;
1324         int has_super = 0;
1325
1326         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1327         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1328
1329         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1330             nr < first_meta_bg)
1331                 return (logic_sb_block + nr + 1);
1332         bg = sbi->s_desc_per_block * nr;
1333         if (ext3_bg_has_super(sb, bg))
1334                 has_super = 1;
1335         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1336 }
1337
1338
1339 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1340 {
1341         struct buffer_head * bh;
1342         struct ext3_super_block *es = NULL;
1343         struct ext3_sb_info *sbi;
1344         unsigned long block;
1345         unsigned long sb_block = get_sb_block(&data);
1346         unsigned long logic_sb_block;
1347         unsigned long offset = 0;
1348         unsigned long journal_inum = 0;
1349         unsigned long journal_devnum = 0;
1350         unsigned long def_mount_opts;
1351         struct inode *root;
1352         int blocksize;
1353         int hblock;
1354         int db_count;
1355         int i;
1356         int needs_recovery;
1357         __le32 features;
1358
1359         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1360         if (!sbi)
1361                 return -ENOMEM;
1362         sb->s_fs_info = sbi;
1363         memset(sbi, 0, sizeof(*sbi));
1364         sbi->s_mount_opt = 0;
1365         sbi->s_resuid = EXT3_DEF_RESUID;
1366         sbi->s_resgid = EXT3_DEF_RESGID;
1367
1368         unlock_kernel();
1369
1370         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1371         if (!blocksize) {
1372                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1373                 goto out_fail;
1374         }
1375
1376         /*
1377          * The ext3 superblock will not be buffer aligned for other than 1kB
1378          * block sizes.  We need to calculate the offset from buffer start.
1379          */
1380         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1381                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1382                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1383         } else {
1384                 logic_sb_block = sb_block;
1385         }
1386
1387         if (!(bh = sb_bread(sb, logic_sb_block))) {
1388                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1389                 goto out_fail;
1390         }
1391         /*
1392          * Note: s_es must be initialized as soon as possible because
1393          *       some ext3 macro-instructions depend on its value
1394          */
1395         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1396         sbi->s_es = es;
1397         sb->s_magic = le16_to_cpu(es->s_magic);
1398         if (sb->s_magic != EXT3_SUPER_MAGIC)
1399                 goto cantfind_ext3;
1400
1401         /* Set defaults before we parse the mount options */
1402         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1403         if (def_mount_opts & EXT3_DEFM_DEBUG)
1404                 set_opt(sbi->s_mount_opt, DEBUG);
1405         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1406                 set_opt(sbi->s_mount_opt, GRPID);
1407         if (def_mount_opts & EXT3_DEFM_UID16)
1408                 set_opt(sbi->s_mount_opt, NO_UID32);
1409         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1410                 set_opt(sbi->s_mount_opt, XATTR_USER);
1411         if (def_mount_opts & EXT3_DEFM_ACL)
1412                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1413         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1414                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1415         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1416                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1417         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1418                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1419
1420         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1421                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1422         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1423                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1424
1425         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1426         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1427
1428         set_opt(sbi->s_mount_opt, RESERVATION);
1429
1430         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1431                             NULL, 0))
1432                 goto failed_mount;
1433
1434         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1435                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1436
1437         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1438             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1439              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1440              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1441                 printk(KERN_WARNING
1442                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1443                        "running e2fsck is recommended\n");
1444         /*
1445          * Check feature flags regardless of the revision level, since we
1446          * previously didn't change the revision level when setting the flags,
1447          * so there is a chance incompat flags are set on a rev 0 filesystem.
1448          */
1449         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1450         if (features) {
1451                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1452                        "unsupported optional features (%x).\n",
1453                        sb->s_id, le32_to_cpu(features));
1454                 goto failed_mount;
1455         }
1456         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1457         if (!(sb->s_flags & MS_RDONLY) && features) {
1458                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1459                        "unsupported optional features (%x).\n",
1460                        sb->s_id, le32_to_cpu(features));
1461                 goto failed_mount;
1462         }
1463         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1464
1465         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1466             blocksize > EXT3_MAX_BLOCK_SIZE) {
1467                 printk(KERN_ERR
1468                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1469                        blocksize, sb->s_id);
1470                 goto failed_mount;
1471         }
1472
1473         hblock = bdev_hardsect_size(sb->s_bdev);
1474         if (sb->s_blocksize != blocksize) {
1475                 /*
1476                  * Make sure the blocksize for the filesystem is larger
1477                  * than the hardware sectorsize for the machine.
1478                  */
1479                 if (blocksize < hblock) {
1480                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1481                                "device blocksize %d.\n", blocksize, hblock);
1482                         goto failed_mount;
1483                 }
1484
1485                 brelse (bh);
1486                 sb_set_blocksize(sb, blocksize);
1487                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1488                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1489                 bh = sb_bread(sb, logic_sb_block);
1490                 if (!bh) {
1491                         printk(KERN_ERR
1492                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1493                         goto failed_mount;
1494                 }
1495                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1496                 sbi->s_es = es;
1497                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1498                         printk (KERN_ERR
1499                                 "EXT3-fs: Magic mismatch, very weird !\n");
1500                         goto failed_mount;
1501                 }
1502         }
1503
1504         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1505
1506         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1507                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1508                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1509         } else {
1510                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1511                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1512                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1513                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1514                     (sbi->s_inode_size > blocksize)) {
1515                         printk (KERN_ERR
1516                                 "EXT3-fs: unsupported inode size: %d\n",
1517                                 sbi->s_inode_size);
1518                         goto failed_mount;
1519                 }
1520         }
1521         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1522                                    le32_to_cpu(es->s_log_frag_size);
1523         if (blocksize != sbi->s_frag_size) {
1524                 printk(KERN_ERR
1525                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1526                        sbi->s_frag_size, blocksize);
1527                 goto failed_mount;
1528         }
1529         sbi->s_frags_per_block = 1;
1530         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1531         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1532         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1533         if (EXT3_INODE_SIZE(sb) == 0)
1534                 goto cantfind_ext3;
1535         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1536         if (sbi->s_inodes_per_block == 0)
1537                 goto cantfind_ext3;
1538         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1539                                         sbi->s_inodes_per_block;
1540         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1541         sbi->s_sbh = bh;
1542         sbi->s_mount_state = le16_to_cpu(es->s_state);
1543         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1544         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1545         for (i=0; i < 4; i++)
1546                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1547         sbi->s_def_hash_version = es->s_def_hash_version;
1548
1549         if (sbi->s_blocks_per_group > blocksize * 8) {
1550                 printk (KERN_ERR
1551                         "EXT3-fs: #blocks per group too big: %lu\n",
1552                         sbi->s_blocks_per_group);
1553                 goto failed_mount;
1554         }
1555         if (sbi->s_frags_per_group > blocksize * 8) {
1556                 printk (KERN_ERR
1557                         "EXT3-fs: #fragments per group too big: %lu\n",
1558                         sbi->s_frags_per_group);
1559                 goto failed_mount;
1560         }
1561         if (sbi->s_inodes_per_group > blocksize * 8) {
1562                 printk (KERN_ERR
1563                         "EXT3-fs: #inodes per group too big: %lu\n",
1564                         sbi->s_inodes_per_group);
1565                 goto failed_mount;
1566         }
1567
1568         if (le32_to_cpu(es->s_blocks_count) >
1569                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1570                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1571                         " too large to mount safely\n", sb->s_id);
1572                 if (sizeof(sector_t) < 8)
1573                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1574                                         "enabled\n");
1575                 goto failed_mount;
1576         }
1577
1578         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1579                 goto cantfind_ext3;
1580         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1581                                le32_to_cpu(es->s_first_data_block) +
1582                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1583                               EXT3_BLOCKS_PER_GROUP(sb);
1584         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1585                    EXT3_DESC_PER_BLOCK(sb);
1586         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1587                                     GFP_KERNEL);
1588         if (sbi->s_group_desc == NULL) {
1589                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1590                 goto failed_mount;
1591         }
1592
1593         bgl_lock_init(&sbi->s_blockgroup_lock);
1594
1595         for (i = 0; i < db_count; i++) {
1596                 block = descriptor_loc(sb, logic_sb_block, i);
1597                 sbi->s_group_desc[i] = sb_bread(sb, block);
1598                 if (!sbi->s_group_desc[i]) {
1599                         printk (KERN_ERR "EXT3-fs: "
1600                                 "can't read group descriptor %d\n", i);
1601                         db_count = i;
1602                         goto failed_mount2;
1603                 }
1604         }
1605         if (!ext3_check_descriptors (sb)) {
1606                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1607                 goto failed_mount2;
1608         }
1609         sbi->s_gdb_count = db_count;
1610         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1611         spin_lock_init(&sbi->s_next_gen_lock);
1612
1613         percpu_counter_init(&sbi->s_freeblocks_counter,
1614                 ext3_count_free_blocks(sb));
1615         percpu_counter_init(&sbi->s_freeinodes_counter,
1616                 ext3_count_free_inodes(sb));
1617         percpu_counter_init(&sbi->s_dirs_counter,
1618                 ext3_count_dirs(sb));
1619
1620         /* per fileystem reservation list head & lock */
1621         spin_lock_init(&sbi->s_rsv_window_lock);
1622         sbi->s_rsv_window_root = RB_ROOT;
1623         /* Add a single, static dummy reservation to the start of the
1624          * reservation window list --- it gives us a placeholder for
1625          * append-at-start-of-list which makes the allocation logic
1626          * _much_ simpler. */
1627         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1628         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1629         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1630         sbi->s_rsv_window_head.rsv_goal_size = 0;
1631         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1632
1633         /*
1634          * set up enough so that it can read an inode
1635          */
1636         sb->s_op = &ext3_sops;
1637         sb->s_export_op = &ext3_export_ops;
1638         sb->s_xattr = ext3_xattr_handlers;
1639 #ifdef CONFIG_QUOTA
1640         sb->s_qcop = &ext3_qctl_operations;
1641         sb->dq_op = &ext3_quota_operations;
1642 #endif
1643         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1644
1645         sb->s_root = NULL;
1646
1647         needs_recovery = (es->s_last_orphan != 0 ||
1648                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1649                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1650
1651         /*
1652          * The first inode we look at is the journal inode.  Don't try
1653          * root first: it may be modified in the journal!
1654          */
1655         if (!test_opt(sb, NOLOAD) &&
1656             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1657                 if (ext3_load_journal(sb, es, journal_devnum))
1658                         goto failed_mount3;
1659         } else if (journal_inum) {
1660                 if (ext3_create_journal(sb, es, journal_inum))
1661                         goto failed_mount3;
1662         } else {
1663                 if (!silent)
1664                         printk (KERN_ERR
1665                                 "ext3: No journal on filesystem on %s\n",
1666                                 sb->s_id);
1667                 goto failed_mount3;
1668         }
1669
1670         /* We have now updated the journal if required, so we can
1671          * validate the data journaling mode. */
1672         switch (test_opt(sb, DATA_FLAGS)) {
1673         case 0:
1674                 /* No mode set, assume a default based on the journal
1675                    capabilities: ORDERED_DATA if the journal can
1676                    cope, else JOURNAL_DATA */
1677                 if (journal_check_available_features
1678                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1679                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1680                 else
1681                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1682                 break;
1683
1684         case EXT3_MOUNT_ORDERED_DATA:
1685         case EXT3_MOUNT_WRITEBACK_DATA:
1686                 if (!journal_check_available_features
1687                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1688                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1689                                "requested data journaling mode\n");
1690                         goto failed_mount4;
1691                 }
1692         default:
1693                 break;
1694         }
1695
1696         if (test_opt(sb, NOBH)) {
1697                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1698                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1699                                 "its supported only with writeback mode\n");
1700                         clear_opt(sbi->s_mount_opt, NOBH);
1701                 }
1702         }
1703         /*
1704          * The journal_load will have done any necessary log recovery,
1705          * so we can safely mount the rest of the filesystem now.
1706          */
1707
1708         root = iget(sb, EXT3_ROOT_INO);
1709         sb->s_root = d_alloc_root(root);
1710         if (!sb->s_root) {
1711                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1712                 iput(root);
1713                 goto failed_mount4;
1714         }
1715         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1716                 dput(sb->s_root);
1717                 sb->s_root = NULL;
1718                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1719                 goto failed_mount4;
1720         }
1721
1722         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1723         /*
1724          * akpm: core read_super() calls in here with the superblock locked.
1725          * That deadlocks, because orphan cleanup needs to lock the superblock
1726          * in numerous places.  Here we just pop the lock - it's relatively
1727          * harmless, because we are now ready to accept write_super() requests,
1728          * and aviro says that's the only reason for hanging onto the
1729          * superblock lock.
1730          */
1731         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1732         ext3_orphan_cleanup(sb, es);
1733         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1734         if (needs_recovery)
1735                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1736         ext3_mark_recovery_complete(sb, es);
1737         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1738                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1739                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1740                 "writeback");
1741
1742         lock_kernel();
1743         return 0;
1744
1745 cantfind_ext3:
1746         if (!silent)
1747                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1748                        sb->s_id);
1749         goto failed_mount;
1750
1751 failed_mount4:
1752         journal_destroy(sbi->s_journal);
1753 failed_mount3:
1754         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1755         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1756         percpu_counter_destroy(&sbi->s_dirs_counter);
1757 failed_mount2:
1758         for (i = 0; i < db_count; i++)
1759                 brelse(sbi->s_group_desc[i]);
1760         kfree(sbi->s_group_desc);
1761 failed_mount:
1762 #ifdef CONFIG_QUOTA
1763         for (i = 0; i < MAXQUOTAS; i++)
1764                 kfree(sbi->s_qf_names[i]);
1765 #endif
1766         ext3_blkdev_remove(sbi);
1767         brelse(bh);
1768 out_fail:
1769         sb->s_fs_info = NULL;
1770         kfree(sbi);
1771         lock_kernel();
1772         return -EINVAL;
1773 }
1774
1775 /*
1776  * Setup any per-fs journal parameters now.  We'll do this both on
1777  * initial mount, once the journal has been initialised but before we've
1778  * done any recovery; and again on any subsequent remount.
1779  */
1780 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1781 {
1782         struct ext3_sb_info *sbi = EXT3_SB(sb);
1783
1784         if (sbi->s_commit_interval)
1785                 journal->j_commit_interval = sbi->s_commit_interval;
1786         /* We could also set up an ext3-specific default for the commit
1787          * interval here, but for now we'll just fall back to the jbd
1788          * default. */
1789
1790         spin_lock(&journal->j_state_lock);
1791         if (test_opt(sb, BARRIER))
1792                 journal->j_flags |= JFS_BARRIER;
1793         else
1794                 journal->j_flags &= ~JFS_BARRIER;
1795         spin_unlock(&journal->j_state_lock);
1796 }
1797
1798 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1799 {
1800         struct inode *journal_inode;
1801         journal_t *journal;
1802
1803         /* First, test for the existence of a valid inode on disk.  Bad
1804          * things happen if we iget() an unused inode, as the subsequent
1805          * iput() will try to delete it. */
1806
1807         journal_inode = iget(sb, journal_inum);
1808         if (!journal_inode) {
1809                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1810                 return NULL;
1811         }
1812         if (!journal_inode->i_nlink) {
1813                 make_bad_inode(journal_inode);
1814                 iput(journal_inode);
1815                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1816                 return NULL;
1817         }
1818
1819         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1820                   journal_inode, journal_inode->i_size);
1821         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1822                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1823                 iput(journal_inode);
1824                 return NULL;
1825         }
1826
1827         journal = journal_init_inode(journal_inode);
1828         if (!journal) {
1829                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1830                 iput(journal_inode);
1831                 return NULL;
1832         }
1833         journal->j_private = sb;
1834         ext3_init_journal_params(sb, journal);
1835         return journal;
1836 }
1837
1838 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1839                                        dev_t j_dev)
1840 {
1841         struct buffer_head * bh;
1842         journal_t *journal;
1843         int start;
1844         int len;
1845         int hblock, blocksize;
1846         unsigned long sb_block;
1847         unsigned long offset;
1848         struct ext3_super_block * es;
1849         struct block_device *bdev;
1850
1851         bdev = ext3_blkdev_get(j_dev);
1852         if (bdev == NULL)
1853                 return NULL;
1854
1855         if (bd_claim(bdev, sb)) {
1856                 printk(KERN_ERR
1857                         "EXT3: failed to claim external journal device.\n");
1858                 blkdev_put(bdev);
1859                 return NULL;
1860         }
1861
1862         blocksize = sb->s_blocksize;
1863         hblock = bdev_hardsect_size(bdev);
1864         if (blocksize < hblock) {
1865                 printk(KERN_ERR
1866                         "EXT3-fs: blocksize too small for journal device.\n");
1867                 goto out_bdev;
1868         }
1869
1870         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1871         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1872         set_blocksize(bdev, blocksize);
1873         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1874                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1875                        "external journal\n");
1876                 goto out_bdev;
1877         }
1878
1879         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1880         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1881             !(le32_to_cpu(es->s_feature_incompat) &
1882               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1883                 printk(KERN_ERR "EXT3-fs: external journal has "
1884                                         "bad superblock\n");
1885                 brelse(bh);
1886                 goto out_bdev;
1887         }
1888
1889         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1890                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1891                 brelse(bh);
1892                 goto out_bdev;
1893         }
1894
1895         len = le32_to_cpu(es->s_blocks_count);
1896         start = sb_block + 1;
1897         brelse(bh);     /* we're done with the superblock */
1898
1899         journal = journal_init_dev(bdev, sb->s_bdev,
1900                                         start, len, blocksize);
1901         if (!journal) {
1902                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1903                 goto out_bdev;
1904         }
1905         journal->j_private = sb;
1906         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1907         wait_on_buffer(journal->j_sb_buffer);
1908         if (!buffer_uptodate(journal->j_sb_buffer)) {
1909                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1910                 goto out_journal;
1911         }
1912         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1913                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1914                                         "user (unsupported) - %d\n",
1915                         be32_to_cpu(journal->j_superblock->s_nr_users));
1916                 goto out_journal;
1917         }
1918         EXT3_SB(sb)->journal_bdev = bdev;
1919         ext3_init_journal_params(sb, journal);
1920         return journal;
1921 out_journal:
1922         journal_destroy(journal);
1923 out_bdev:
1924         ext3_blkdev_put(bdev);
1925         return NULL;
1926 }
1927
1928 static int ext3_load_journal(struct super_block *sb,
1929                              struct ext3_super_block *es,
1930                              unsigned long journal_devnum)
1931 {
1932         journal_t *journal;
1933         int journal_inum = le32_to_cpu(es->s_journal_inum);
1934         dev_t journal_dev;
1935         int err = 0;
1936         int really_read_only;
1937
1938         if (journal_devnum &&
1939             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1940                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1941                         "numbers have changed\n");
1942                 journal_dev = new_decode_dev(journal_devnum);
1943         } else
1944                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1945
1946         really_read_only = bdev_read_only(sb->s_bdev);
1947
1948         /*
1949          * Are we loading a blank journal or performing recovery after a
1950          * crash?  For recovery, we need to check in advance whether we
1951          * can get read-write access to the device.
1952          */
1953
1954         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1955                 if (sb->s_flags & MS_RDONLY) {
1956                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1957                                         "required on readonly filesystem.\n");
1958                         if (really_read_only) {
1959                                 printk(KERN_ERR "EXT3-fs: write access "
1960                                         "unavailable, cannot proceed.\n");
1961                                 return -EROFS;
1962                         }
1963                         printk (KERN_INFO "EXT3-fs: write access will "
1964                                         "be enabled during recovery.\n");
1965                 }
1966         }
1967
1968         if (journal_inum && journal_dev) {
1969                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1970                        "and inode journals!\n");
1971                 return -EINVAL;
1972         }
1973
1974         if (journal_inum) {
1975                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1976                         return -EINVAL;
1977         } else {
1978                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1979                         return -EINVAL;
1980         }
1981
1982         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1983                 err = journal_update_format(journal);
1984                 if (err)  {
1985                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1986                         journal_destroy(journal);
1987                         return err;
1988                 }
1989         }
1990
1991         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1992                 err = journal_wipe(journal, !really_read_only);
1993         if (!err)
1994                 err = journal_load(journal);
1995
1996         if (err) {
1997                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1998                 journal_destroy(journal);
1999                 return err;
2000         }
2001
2002         EXT3_SB(sb)->s_journal = journal;
2003         ext3_clear_journal_err(sb, es);
2004
2005         if (journal_devnum &&
2006             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2007                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2008                 sb->s_dirt = 1;
2009
2010                 /* Make sure we flush the recovery flag to disk. */
2011                 ext3_commit_super(sb, es, 1);
2012         }
2013
2014         return 0;
2015 }
2016
2017 static int ext3_create_journal(struct super_block * sb,
2018                                struct ext3_super_block * es,
2019                                int journal_inum)
2020 {
2021         journal_t *journal;
2022
2023         if (sb->s_flags & MS_RDONLY) {
2024                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2025                                 "create journal.\n");
2026                 return -EROFS;
2027         }
2028
2029         if (!(journal = ext3_get_journal(sb, journal_inum)))
2030                 return -EINVAL;
2031
2032         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2033                journal_inum);
2034
2035         if (journal_create(journal)) {
2036                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2037                 journal_destroy(journal);
2038                 return -EIO;
2039         }
2040
2041         EXT3_SB(sb)->s_journal = journal;
2042
2043         ext3_update_dynamic_rev(sb);
2044         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2045         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2046
2047         es->s_journal_inum = cpu_to_le32(journal_inum);
2048         sb->s_dirt = 1;
2049
2050         /* Make sure we flush the recovery flag to disk. */
2051         ext3_commit_super(sb, es, 1);
2052
2053         return 0;
2054 }
2055
2056 static void ext3_commit_super (struct super_block * sb,
2057                                struct ext3_super_block * es,
2058                                int sync)
2059 {
2060         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2061
2062         if (!sbh)
2063                 return;
2064         es->s_wtime = cpu_to_le32(get_seconds());
2065         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2066         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2067         BUFFER_TRACE(sbh, "marking dirty");
2068         mark_buffer_dirty(sbh);
2069         if (sync)
2070                 sync_dirty_buffer(sbh);
2071 }
2072
2073
2074 /*
2075  * Have we just finished recovery?  If so, and if we are mounting (or
2076  * remounting) the filesystem readonly, then we will end up with a
2077  * consistent fs on disk.  Record that fact.
2078  */
2079 static void ext3_mark_recovery_complete(struct super_block * sb,
2080                                         struct ext3_super_block * es)
2081 {
2082         journal_t *journal = EXT3_SB(sb)->s_journal;
2083
2084         journal_lock_updates(journal);
2085         journal_flush(journal);
2086         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2087             sb->s_flags & MS_RDONLY) {
2088                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2089                 sb->s_dirt = 0;
2090                 ext3_commit_super(sb, es, 1);
2091         }
2092         journal_unlock_updates(journal);
2093 }
2094
2095 /*
2096  * If we are mounting (or read-write remounting) a filesystem whose journal
2097  * has recorded an error from a previous lifetime, move that error to the
2098  * main filesystem now.
2099  */
2100 static void ext3_clear_journal_err(struct super_block * sb,
2101                                    struct ext3_super_block * es)
2102 {
2103         journal_t *journal;
2104         int j_errno;
2105         const char *errstr;
2106
2107         journal = EXT3_SB(sb)->s_journal;
2108
2109         /*
2110          * Now check for any error status which may have been recorded in the
2111          * journal by a prior ext3_error() or ext3_abort()
2112          */
2113
2114         j_errno = journal_errno(journal);
2115         if (j_errno) {
2116                 char nbuf[16];
2117
2118                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2119                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2120                              "from previous mount: %s", errstr);
2121                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2122                              "filesystem check.");
2123
2124                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2125                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2126                 ext3_commit_super (sb, es, 1);
2127
2128                 journal_clear_err(journal);
2129         }
2130 }
2131
2132 /*
2133  * Force the running and committing transactions to commit,
2134  * and wait on the commit.
2135  */
2136 int ext3_force_commit(struct super_block *sb)
2137 {
2138         journal_t *journal;
2139         int ret;
2140
2141         if (sb->s_flags & MS_RDONLY)
2142                 return 0;
2143
2144         journal = EXT3_SB(sb)->s_journal;
2145         sb->s_dirt = 0;
2146         ret = ext3_journal_force_commit(journal);
2147         return ret;
2148 }
2149
2150 /*
2151  * Ext3 always journals updates to the superblock itself, so we don't
2152  * have to propagate any other updates to the superblock on disk at this
2153  * point.  Just start an async writeback to get the buffers on their way
2154  * to the disk.
2155  *
2156  * This implicitly triggers the writebehind on sync().
2157  */
2158
2159 static void ext3_write_super (struct super_block * sb)
2160 {
2161         if (mutex_trylock(&sb->s_lock) != 0)
2162                 BUG();
2163         sb->s_dirt = 0;
2164 }
2165
2166 static int ext3_sync_fs(struct super_block *sb, int wait)
2167 {
2168         tid_t target;
2169
2170         sb->s_dirt = 0;
2171         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2172                 if (wait)
2173                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2174         }
2175         return 0;
2176 }
2177
2178 /*
2179  * LVM calls this function before a (read-only) snapshot is created.  This
2180  * gives us a chance to flush the journal completely and mark the fs clean.
2181  */
2182 static void ext3_write_super_lockfs(struct super_block *sb)
2183 {
2184         sb->s_dirt = 0;
2185
2186         if (!(sb->s_flags & MS_RDONLY)) {
2187                 journal_t *journal = EXT3_SB(sb)->s_journal;
2188
2189                 /* Now we set up the journal barrier. */
2190                 journal_lock_updates(journal);
2191                 journal_flush(journal);
2192
2193                 /* Journal blocked and flushed, clear needs_recovery flag. */
2194                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2195                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2196         }
2197 }
2198
2199 /*
2200  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2201  * flag here, even though the filesystem is not technically dirty yet.
2202  */
2203 static void ext3_unlockfs(struct super_block *sb)
2204 {
2205         if (!(sb->s_flags & MS_RDONLY)) {
2206                 lock_super(sb);
2207                 /* Reser the needs_recovery flag before the fs is unlocked. */
2208                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2209                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2210                 unlock_super(sb);
2211                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2212         }
2213 }
2214
2215 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2216 {
2217         struct ext3_super_block * es;
2218         struct ext3_sb_info *sbi = EXT3_SB(sb);
2219         unsigned long n_blocks_count = 0;
2220         unsigned long old_sb_flags;
2221         struct ext3_mount_options old_opts;
2222         int err;
2223 #ifdef CONFIG_QUOTA
2224         int i;
2225 #endif
2226
2227         /* Store the original options */
2228         old_sb_flags = sb->s_flags;
2229         old_opts.s_mount_opt = sbi->s_mount_opt;
2230         old_opts.s_resuid = sbi->s_resuid;
2231         old_opts.s_resgid = sbi->s_resgid;
2232         old_opts.s_commit_interval = sbi->s_commit_interval;
2233 #ifdef CONFIG_QUOTA
2234         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2235         for (i = 0; i < MAXQUOTAS; i++)
2236                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2237 #endif
2238
2239         /*
2240          * Allow the "check" option to be passed as a remount option.
2241          */
2242         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2243                 err = -EINVAL;
2244                 goto restore_opts;
2245         }
2246
2247         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2248                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2249
2250         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2251                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2252
2253         es = sbi->s_es;
2254
2255         ext3_init_journal_params(sb, sbi->s_journal);
2256
2257         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2258                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2259                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2260                         err = -EROFS;
2261                         goto restore_opts;
2262                 }
2263
2264                 if (*flags & MS_RDONLY) {
2265                         /*
2266                          * First of all, the unconditional stuff we have to do
2267                          * to disable replay of the journal when we next remount
2268                          */
2269                         sb->s_flags |= MS_RDONLY;
2270
2271                         /*
2272                          * OK, test if we are remounting a valid rw partition
2273                          * readonly, and if so set the rdonly flag and then
2274                          * mark the partition as valid again.
2275                          */
2276                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2277                             (sbi->s_mount_state & EXT3_VALID_FS))
2278                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2279
2280                         ext3_mark_recovery_complete(sb, es);
2281                 } else {
2282                         __le32 ret;
2283                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2284                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2285                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2286                                        "remount RDWR because of unsupported "
2287                                        "optional features (%x).\n",
2288                                        sb->s_id, le32_to_cpu(ret));
2289                                 err = -EROFS;
2290                                 goto restore_opts;
2291                         }
2292                         /*
2293                          * Mounting a RDONLY partition read-write, so reread
2294                          * and store the current valid flag.  (It may have
2295                          * been changed by e2fsck since we originally mounted
2296                          * the partition.)
2297                          */
2298                         ext3_clear_journal_err(sb, es);
2299                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2300                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2301                                 err = ret;
2302                                 goto restore_opts;
2303                         }
2304                         if (!ext3_setup_super (sb, es, 0))
2305                                 sb->s_flags &= ~MS_RDONLY;
2306                 }
2307         }
2308 #ifdef CONFIG_QUOTA
2309         /* Release old quota file names */
2310         for (i = 0; i < MAXQUOTAS; i++)
2311                 if (old_opts.s_qf_names[i] &&
2312                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2313                         kfree(old_opts.s_qf_names[i]);
2314 #endif
2315         return 0;
2316 restore_opts:
2317         sb->s_flags = old_sb_flags;
2318         sbi->s_mount_opt = old_opts.s_mount_opt;
2319         sbi->s_resuid = old_opts.s_resuid;
2320         sbi->s_resgid = old_opts.s_resgid;
2321         sbi->s_commit_interval = old_opts.s_commit_interval;
2322 #ifdef CONFIG_QUOTA
2323         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2324         for (i = 0; i < MAXQUOTAS; i++) {
2325                 if (sbi->s_qf_names[i] &&
2326                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2327                         kfree(sbi->s_qf_names[i]);
2328                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2329         }
2330 #endif
2331         return err;
2332 }
2333
2334 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2335 {
2336         struct super_block *sb = dentry->d_sb;
2337         struct ext3_sb_info *sbi = EXT3_SB(sb);
2338         struct ext3_super_block *es = sbi->s_es;
2339         unsigned long overhead;
2340         int i;
2341
2342         if (test_opt (sb, MINIX_DF))
2343                 overhead = 0;
2344         else {
2345                 unsigned long ngroups;
2346                 ngroups = EXT3_SB(sb)->s_groups_count;
2347                 smp_rmb();
2348
2349                 /*
2350                  * Compute the overhead (FS structures)
2351                  */
2352
2353                 /*
2354                  * All of the blocks before first_data_block are
2355                  * overhead
2356                  */
2357                 overhead = le32_to_cpu(es->s_first_data_block);
2358
2359                 /*
2360                  * Add the overhead attributed to the superblock and
2361                  * block group descriptors.  If the sparse superblocks
2362                  * feature is turned on, then not all groups have this.
2363                  */
2364                 for (i = 0; i < ngroups; i++) {
2365                         overhead += ext3_bg_has_super(sb, i) +
2366                                 ext3_bg_num_gdb(sb, i);
2367                         cond_resched();
2368                 }
2369
2370                 /*
2371                  * Every block group has an inode bitmap, a block
2372                  * bitmap, and an inode table.
2373                  */
2374                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2375         }
2376
2377         buf->f_type = EXT3_SUPER_MAGIC;
2378         buf->f_bsize = sb->s_blocksize;
2379         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2380         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2381         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2382         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2383                 buf->f_bavail = 0;
2384         buf->f_files = le32_to_cpu(es->s_inodes_count);
2385         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2386         buf->f_namelen = EXT3_NAME_LEN;
2387         return 0;
2388 }
2389
2390 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2391  * is locked for write. Otherwise the are possible deadlocks:
2392  * Process 1                         Process 2
2393  * ext3_create()                     quota_sync()
2394  *   journal_start()                   write_dquot()
2395  *   DQUOT_INIT()                        down(dqio_mutex)
2396  *     down(dqio_mutex)                    journal_start()
2397  *
2398  */
2399
2400 #ifdef CONFIG_QUOTA
2401
2402 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2403 {
2404         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2405 }
2406
2407 static int ext3_dquot_initialize(struct inode *inode, int type)
2408 {
2409         handle_t *handle;
2410         int ret, err;
2411
2412         /* We may create quota structure so we need to reserve enough blocks */
2413         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2414         if (IS_ERR(handle))
2415                 return PTR_ERR(handle);
2416         ret = dquot_initialize(inode, type);
2417         err = ext3_journal_stop(handle);
2418         if (!ret)
2419                 ret = err;
2420         return ret;
2421 }
2422
2423 static int ext3_dquot_drop(struct inode *inode)
2424 {
2425         handle_t *handle;
2426         int ret, err;
2427
2428         /* We may delete quota structure so we need to reserve enough blocks */
2429         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2430         if (IS_ERR(handle))
2431                 return PTR_ERR(handle);
2432         ret = dquot_drop(inode);
2433         err = ext3_journal_stop(handle);
2434         if (!ret)
2435                 ret = err;
2436         return ret;
2437 }
2438
2439 static int ext3_write_dquot(struct dquot *dquot)
2440 {
2441         int ret, err;
2442         handle_t *handle;
2443         struct inode *inode;
2444
2445         inode = dquot_to_inode(dquot);
2446         handle = ext3_journal_start(inode,
2447                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2448         if (IS_ERR(handle))
2449                 return PTR_ERR(handle);
2450         ret = dquot_commit(dquot);
2451         err = ext3_journal_stop(handle);
2452         if (!ret)
2453                 ret = err;
2454         return ret;
2455 }
2456
2457 static int ext3_acquire_dquot(struct dquot *dquot)
2458 {
2459         int ret, err;
2460         handle_t *handle;
2461
2462         handle = ext3_journal_start(dquot_to_inode(dquot),
2463                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2464         if (IS_ERR(handle))
2465                 return PTR_ERR(handle);
2466         ret = dquot_acquire(dquot);
2467         err = ext3_journal_stop(handle);
2468         if (!ret)
2469                 ret = err;
2470         return ret;
2471 }
2472
2473 static int ext3_release_dquot(struct dquot *dquot)
2474 {
2475         int ret, err;
2476         handle_t *handle;
2477
2478         handle = ext3_journal_start(dquot_to_inode(dquot),
2479                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2480         if (IS_ERR(handle))
2481                 return PTR_ERR(handle);
2482         ret = dquot_release(dquot);
2483         err = ext3_journal_stop(handle);
2484         if (!ret)
2485                 ret = err;
2486         return ret;
2487 }
2488
2489 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2490 {
2491         /* Are we journalling quotas? */
2492         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2493             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2494                 dquot_mark_dquot_dirty(dquot);
2495                 return ext3_write_dquot(dquot);
2496         } else {
2497                 return dquot_mark_dquot_dirty(dquot);
2498         }
2499 }
2500
2501 static int ext3_write_info(struct super_block *sb, int type)
2502 {
2503         int ret, err;
2504         handle_t *handle;
2505
2506         /* Data block + inode block */
2507         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2508         if (IS_ERR(handle))
2509                 return PTR_ERR(handle);
2510         ret = dquot_commit_info(sb, type);
2511         err = ext3_journal_stop(handle);
2512         if (!ret)
2513                 ret = err;
2514         return ret;
2515 }
2516
2517 /*
2518  * Turn on quotas during mount time - we need to find
2519  * the quota file and such...
2520  */
2521 static int ext3_quota_on_mount(struct super_block *sb, int type)
2522 {
2523         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2524                         EXT3_SB(sb)->s_jquota_fmt, type);
2525 }
2526
2527 /*
2528  * Standard function to be called on quota_on
2529  */
2530 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2531                          char *path)
2532 {
2533         int err;
2534         struct nameidata nd;
2535
2536         if (!test_opt(sb, QUOTA))
2537                 return -EINVAL;
2538         /* Not journalling quota? */
2539         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2540             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2541                 return vfs_quota_on(sb, type, format_id, path);
2542         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2543         if (err)
2544                 return err;
2545         /* Quotafile not on the same filesystem? */
2546         if (nd.mnt->mnt_sb != sb) {
2547                 path_release(&nd);
2548                 return -EXDEV;
2549         }
2550         /* Quotafile not of fs root? */
2551         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2552                 printk(KERN_WARNING
2553                         "EXT3-fs: Quota file not on filesystem root. "
2554                         "Journalled quota will not work.\n");
2555         path_release(&nd);
2556         return vfs_quota_on(sb, type, format_id, path);
2557 }
2558
2559 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2560  * acquiring the locks... As quota files are never truncated and quota code
2561  * itself serializes the operations (and noone else should touch the files)
2562  * we don't have to be afraid of races */
2563 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2564                                size_t len, loff_t off)
2565 {
2566         struct inode *inode = sb_dqopt(sb)->files[type];
2567         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2568         int err = 0;
2569         int offset = off & (sb->s_blocksize - 1);
2570         int tocopy;
2571         size_t toread;
2572         struct buffer_head *bh;
2573         loff_t i_size = i_size_read(inode);
2574
2575         if (off > i_size)
2576                 return 0;
2577         if (off+len > i_size)
2578                 len = i_size-off;
2579         toread = len;
2580         while (toread > 0) {
2581                 tocopy = sb->s_blocksize - offset < toread ?
2582                                 sb->s_blocksize - offset : toread;
2583                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2584                 if (err)
2585                         return err;
2586                 if (!bh)        /* A hole? */
2587                         memset(data, 0, tocopy);
2588                 else
2589                         memcpy(data, bh->b_data+offset, tocopy);
2590                 brelse(bh);
2591                 offset = 0;
2592                 toread -= tocopy;
2593                 data += tocopy;
2594                 blk++;
2595         }
2596         return len;
2597 }
2598
2599 /* Write to quotafile (we know the transaction is already started and has
2600  * enough credits) */
2601 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2602                                 const char *data, size_t len, loff_t off)
2603 {
2604         struct inode *inode = sb_dqopt(sb)->files[type];
2605         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2606         int err = 0;
2607         int offset = off & (sb->s_blocksize - 1);
2608         int tocopy;
2609         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2610         size_t towrite = len;
2611         struct buffer_head *bh;
2612         handle_t *handle = journal_current_handle();
2613
2614         mutex_lock(&inode->i_mutex);
2615         while (towrite > 0) {
2616                 tocopy = sb->s_blocksize - offset < towrite ?
2617                                 sb->s_blocksize - offset : towrite;
2618                 bh = ext3_bread(handle, inode, blk, 1, &err);
2619                 if (!bh)
2620                         goto out;
2621                 if (journal_quota) {
2622                         err = ext3_journal_get_write_access(handle, bh);
2623                         if (err) {
2624                                 brelse(bh);
2625                                 goto out;
2626                         }
2627                 }
2628                 lock_buffer(bh);
2629                 memcpy(bh->b_data+offset, data, tocopy);
2630                 flush_dcache_page(bh->b_page);
2631                 unlock_buffer(bh);
2632                 if (journal_quota)
2633                         err = ext3_journal_dirty_metadata(handle, bh);
2634                 else {
2635                         /* Always do at least ordered writes for quotas */
2636                         err = ext3_journal_dirty_data(handle, bh);
2637                         mark_buffer_dirty(bh);
2638                 }
2639                 brelse(bh);
2640                 if (err)
2641                         goto out;
2642                 offset = 0;
2643                 towrite -= tocopy;
2644                 data += tocopy;
2645                 blk++;
2646         }
2647 out:
2648         if (len == towrite)
2649                 return err;
2650         if (inode->i_size < off+len-towrite) {
2651                 i_size_write(inode, off+len-towrite);
2652                 EXT3_I(inode)->i_disksize = inode->i_size;
2653         }
2654         inode->i_version++;
2655         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2656         ext3_mark_inode_dirty(handle, inode);
2657         mutex_unlock(&inode->i_mutex);
2658         return len - towrite;
2659 }
2660
2661 #endif
2662
2663 static int ext3_get_sb(struct file_system_type *fs_type,
2664         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2665 {
2666         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2667 }
2668
2669 static struct file_system_type ext3_fs_type = {
2670         .owner          = THIS_MODULE,
2671         .name           = "ext3",
2672         .get_sb         = ext3_get_sb,
2673         .kill_sb        = kill_block_super,
2674         .fs_flags       = FS_REQUIRES_DEV,
2675 };
2676
2677 static int __init init_ext3_fs(void)
2678 {
2679         int err = init_ext3_xattr();
2680         if (err)
2681                 return err;
2682         err = init_inodecache();
2683         if (err)
2684                 goto out1;
2685         err = register_filesystem(&ext3_fs_type);
2686         if (err)
2687                 goto out;
2688         return 0;
2689 out:
2690         destroy_inodecache();
2691 out1:
2692         exit_ext3_xattr();
2693         return err;
2694 }
2695
2696 static void __exit exit_ext3_fs(void)
2697 {
2698         unregister_filesystem(&ext3_fs_type);
2699         destroy_inodecache();
2700         exit_ext3_xattr();
2701 }
2702
2703 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2704 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2705 MODULE_LICENSE("GPL");
2706 module_init(init_ext3_fs)
2707 module_exit(exit_ext3_fs)