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 <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53
  54 /*
  55  * Wrappers for journal_start/end.
  56  *
  57  * The only special thing we need to do here is to make sure that all
  58  * journal_end calls result in the superblock being marked dirty, so
  59  * that sync() will call the filesystem's write_super callback if
  60  * appropriate.
  61  */
  62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (inode->i_sb->s_flags & MS_RDONLY)
  67                 return ERR_PTR(-EROFS);
  68
  69         /* Special case here: if the journal has aborted behind our
  70          * backs (eg. EIO in the commit thread), then we still need to
  71          * take the FS itself readonly cleanly. */
  72         journal = EXT3_JOURNAL(inode);
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(inode->i_sb, __FUNCTION__,
  75                            "Detected aborted journal");
  76                 return ERR_PTR(-EROFS);
  77         }
  78
  79         return journal_start(journal, nblocks);
  80 }
  81
  82 /*
  83  * The only special thing we need to do here is to make sure that all
  84  * journal_stop calls result in the superblock being marked dirty, so
  85  * that sync() will call the filesystem's write_super callback if
  86  * appropriate.
  87  */
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         printk(KERN_ERR "%s: aborting transaction: %s in %s",
 112                caller, errstr, err_fn);
 113
 114         if (bh)
 115                 BUFFER_TRACE(bh, "abort");
 116         journal_abort_handle(handle);
 117         if (!handle->h_err)
 118                 handle->h_err = err;
 119 }
 120
 121 static char error_buf[1024];
 122
 123 /* Deal with the reporting of failure conditions on a filesystem such as
 124  * inconsistencies detected or read IO failures.
 125  *
 126  * On ext2, we can store the error state of the filesystem in the
 127  * superblock.  That is not possible on ext3, because we may have other
 128  * write ordering constraints on the superblock which prevent us from
 129  * writing it out straight away; and given that the journal is about to
 130  * be aborted, we can't rely on the current, or future, transactions to
 131  * write out the superblock safely.
 132  *
 133  * We'll just use the journal_abort() error code to record an error in
 134  * the journal instead.  On recovery, the journal will compain about
 135  * that error until we've noted it down and cleared it.
 136  */
 137
 138 static void ext3_handle_error(struct super_block *sb)
 139 {
 140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 141
 142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 144
 145         if (sb->s_flags & MS_RDONLY)
 146                 return;
 147
 148         if (test_opt (sb, ERRORS_PANIC))
 149                 panic ("EXT3-fs (device %s): panic forced after error\n",
 150                        sb->s_id);
 151         if (test_opt (sb, ERRORS_RO)) {
 152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 153                 sb->s_flags |= MS_RDONLY;
 154         } else {
 155                 journal_t *journal = EXT3_SB(sb)->s_journal;
 156
 157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 158                 if (journal)
 159                         journal_abort(journal, -EIO);
 160         }
 161         ext3_commit_super(sb, es, 1);
 162 }
 163
 164 void ext3_error (struct super_block * sb, const char * function,
 165                  const char * fmt, ...)
 166 {
 167         va_list args;
 168
 169         va_start (args, fmt);
 170         vsprintf (error_buf, fmt, args);
 171         va_end (args);
 172
 173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
 174                 sb->s_id, function, error_buf);
 175
 176         ext3_handle_error(sb);
 177 }
 178
 179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 180 {
 181         char *errstr = NULL;
 182
 183         switch (errno) {
 184         case -EIO:
 185                 errstr = "IO failure";
 186                 break;
 187         case -ENOMEM:
 188                 errstr = "Out of memory";
 189                 break;
 190         case -EROFS:
 191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 192                         errstr = "Journal has aborted";
 193                 else
 194                         errstr = "Readonly filesystem";
 195                 break;
 196         default:
 197                 /* If the caller passed in an extra buffer for unknown
 198                  * errors, textualise them now.  Else we just return
 199                  * NULL. */
 200                 if (nbuf) {
 201                         /* Check for truncated error codes... */
 202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 203                                 errstr = nbuf;
 204                 }
 205                 break;
 206         }
 207
 208         return errstr;
 209 }
 210
 211 /* __ext3_std_error decodes expected errors from journaling functions
 212  * automatically and invokes the appropriate error response.  */
 213
 214 void __ext3_std_error (struct super_block * sb, const char * function,
 215                        int errno)
 216 {
 217         char nbuf[16];
 218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 219
 220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 221                 sb->s_id, function, errstr);
 222
 223         ext3_handle_error(sb);
 224 }
 225
 226 /*
 227  * ext3_abort is a much stronger failure handler than ext3_error.  The
 228  * abort function may be used to deal with unrecoverable failures such
 229  * as journal IO errors or ENOMEM at a critical moment in log management.
 230  *
 231  * We unconditionally force the filesystem into an ABORT|READONLY state,
 232  * unless the error response on the fs has been set to panic in which
 233  * case we take the easy way out and panic immediately.
 234  */
 235
 236 void ext3_abort (struct super_block * sb, const char * function,
 237                  const char * fmt, ...)
 238 {
 239         va_list args;
 240
 241         printk (KERN_CRIT "ext3_abort called.\n");
 242
 243         va_start (args, fmt);
 244         vsprintf (error_buf, fmt, args);
 245         va_end (args);
 246
 247         if (test_opt (sb, ERRORS_PANIC))
 248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
 249                        sb->s_id, function, error_buf);
 250
 251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
 252                 sb->s_id, function, error_buf);
 253
 254         if (sb->s_flags & MS_RDONLY)
 255                 return;
 256
 257         printk (KERN_CRIT "Remounting filesystem read-only\n");
 258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 259         sb->s_flags |= MS_RDONLY;
 260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 262 }
 263
 264 /* Deal with the reporting of failure conditions while running, such as
 265  * inconsistencies in operation or invalid system states.
 266  *
 267  * Use ext3_error() for cases of invalid filesystem states, as that will
 268  * record an error on disk and force a filesystem check on the next boot.
 269  */
 270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 271                             const char * fmt, ...)
 272 {
 273         va_list args;
 274
 275         va_start (args, fmt);
 276         vsprintf (error_buf, fmt, args);
 277         va_end (args);
 278
 279         /* this is to prevent panic from syncing this filesystem */
 280         /* AKPM: is this sufficient? */
 281         sb->s_flags |= MS_RDONLY;
 282         panic ("EXT3-fs panic (device %s): %s: %s\n",
 283                sb->s_id, function, error_buf);
 284 }
 285
 286 void ext3_warning (struct super_block * sb, const char * function,
 287                    const char * fmt, ...)
 288 {
 289         va_list args;
 290
 291         va_start (args, fmt);
 292         vsprintf (error_buf, fmt, args);
 293         va_end (args);
 294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
 295                 sb->s_id, function, error_buf);
 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 = 0;
 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                        le32_to_cpu(NEXT_ORPHAN(inode)));
 384         }
 385 }
 386
 387 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 = le16_to_cpu(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         kfree(sbi->s_debts);
 407         brelse(sbi->s_sbh);
 408 #ifdef CONFIG_QUOTA
 409         for (i = 0; i < MAXQUOTAS; i++) {
 410                 if (sbi->s_qf_names[i])
 411                         kfree(sbi->s_qf_names[i]);
 412         }
 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->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 init_MUTEX(&ei->truncate_sem);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 484                                              init_once, NULL);
 485         if (ext3_inode_cachep == NULL)
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static void destroy_inodecache(void)
 491 {
 492         if (kmem_cache_destroy(ext3_inode_cachep))
 493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 494 }
 495
 496 static void ext3_clear_inode(struct inode *inode)
 497 {
 498 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 499        if (EXT3_I(inode)->i_acl &&
 500            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 501                posix_acl_release(EXT3_I(inode)->i_acl);
 502                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 503        }
 504        if (EXT3_I(inode)->i_default_acl &&
 505            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 506                posix_acl_release(EXT3_I(inode)->i_default_acl);
 507                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 508        }
 509 #endif
 510         if (!is_bad_inode(inode))
 511                 ext3_discard_reservation(inode);
 512 }
 513
 514 #ifdef CONFIG_QUOTA
 515
 516 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 517 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 518
 519 static int ext3_dquot_initialize(struct inode *inode, int type);
 520 static int ext3_dquot_drop(struct inode *inode);
 521 static int ext3_write_dquot(struct dquot *dquot);
 522 static int ext3_acquire_dquot(struct dquot *dquot);
 523 static int ext3_release_dquot(struct dquot *dquot);
 524 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 525 static int ext3_write_info(struct super_block *sb, int type);
 526 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 527 static int ext3_quota_on_mount(struct super_block *sb, int type);
 528 static int ext3_quota_off_mount(struct super_block *sb, int type);
 529
 530 static struct dquot_operations ext3_quota_operations = {
 531         .initialize     = ext3_dquot_initialize,
 532         .drop           = ext3_dquot_drop,
 533         .alloc_space    = dquot_alloc_space,
 534         .alloc_inode    = dquot_alloc_inode,
 535         .free_space     = dquot_free_space,
 536         .free_inode     = dquot_free_inode,
 537         .transfer       = dquot_transfer,
 538         .write_dquot    = ext3_write_dquot,
 539         .acquire_dquot  = ext3_acquire_dquot,
 540         .release_dquot  = ext3_release_dquot,
 541         .mark_dirty     = ext3_mark_dquot_dirty,
 542         .write_info     = ext3_write_info
 543 };
 544
 545 static struct quotactl_ops ext3_qctl_operations = {
 546         .quota_on       = ext3_quota_on,
 547         .quota_off      = vfs_quota_off,
 548         .quota_sync     = vfs_quota_sync,
 549         .get_info       = vfs_get_dqinfo,
 550         .set_info       = vfs_set_dqinfo,
 551         .get_dqblk      = vfs_get_dqblk,
 552         .set_dqblk      = vfs_set_dqblk
 553 };
 554 #endif
 555
 556 static struct super_operations ext3_sops = {
 557         .alloc_inode    = ext3_alloc_inode,
 558         .destroy_inode  = ext3_destroy_inode,
 559         .read_inode     = ext3_read_inode,
 560         .write_inode    = ext3_write_inode,
 561         .dirty_inode    = ext3_dirty_inode,
 562         .delete_inode   = ext3_delete_inode,
 563         .put_super      = ext3_put_super,
 564         .write_super    = ext3_write_super,
 565         .sync_fs        = ext3_sync_fs,
 566         .write_super_lockfs = ext3_write_super_lockfs,
 567         .unlockfs       = ext3_unlockfs,
 568         .statfs         = ext3_statfs,
 569         .remount_fs     = ext3_remount,
 570         .clear_inode    = ext3_clear_inode,
 571 };
 572
 573 struct dentry *ext3_get_parent(struct dentry *child);
 574 static struct export_operations ext3_export_ops = {
 575         .get_parent = ext3_get_parent,
 576 };
 577
 578 enum {
 579         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 580         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 581         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 582         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 583         Opt_reservation, Opt_noreservation, Opt_noload,
 584         Opt_commit, Opt_journal_update, Opt_journal_inum,
 585         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 586         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 587         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 588         Opt_ignore, Opt_err, Opt_resize,
 589 };
 590
 591 static match_table_t tokens = {
 592         {Opt_bsd_df, "bsddf"},
 593         {Opt_minix_df, "minixdf"},
 594         {Opt_grpid, "grpid"},
 595         {Opt_grpid, "bsdgroups"},
 596         {Opt_nogrpid, "nogrpid"},
 597         {Opt_nogrpid, "sysvgroups"},
 598         {Opt_resgid, "resgid=%u"},
 599         {Opt_resuid, "resuid=%u"},
 600         {Opt_sb, "sb=%u"},
 601         {Opt_err_cont, "errors=continue"},
 602         {Opt_err_panic, "errors=panic"},
 603         {Opt_err_ro, "errors=remount-ro"},
 604         {Opt_nouid32, "nouid32"},
 605         {Opt_nocheck, "nocheck"},
 606         {Opt_nocheck, "check=none"},
 607         {Opt_check, "check"},
 608         {Opt_debug, "debug"},
 609         {Opt_oldalloc, "oldalloc"},
 610         {Opt_orlov, "orlov"},
 611         {Opt_user_xattr, "user_xattr"},
 612         {Opt_nouser_xattr, "nouser_xattr"},
 613         {Opt_acl, "acl"},
 614         {Opt_noacl, "noacl"},
 615         {Opt_reservation, "reservation"},
 616         {Opt_noreservation, "noreservation"},
 617         {Opt_noload, "noload"},
 618         {Opt_commit, "commit=%u"},
 619         {Opt_journal_update, "journal=update"},
 620         {Opt_journal_inum, "journal=%u"},
 621         {Opt_abort, "abort"},
 622         {Opt_data_journal, "data=journal"},
 623         {Opt_data_ordered, "data=ordered"},
 624         {Opt_data_writeback, "data=writeback"},
 625         {Opt_offusrjquota, "usrjquota="},
 626         {Opt_usrjquota, "usrjquota=%s"},
 627         {Opt_offgrpjquota, "grpjquota="},
 628         {Opt_grpjquota, "grpjquota=%s"},
 629         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 630         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 631         {Opt_ignore, "grpquota"},
 632         {Opt_ignore, "noquota"},
 633         {Opt_ignore, "quota"},
 634         {Opt_ignore, "usrquota"},
 635         {Opt_err, NULL},
 636         {Opt_resize, "resize"}
 637
 638 };
 639
 640 static unsigned long get_sb_block(void **data)
 641 {
 642         unsigned long   sb_block;
 643         char            *options = (char *) *data;
 644
 645         if (!options || strncmp(options, "sb=", 3) != 0)
 646                 return 1;       /* Default location */
 647         options += 3;
 648         sb_block = simple_strtoul(options, &options, 0);
 649         if (*options && *options != ',') {
 650                 printk("EXT3-fs: Invalid sb specification: %s\n",
 651                        (char *) *data);
 652                 return 1;
 653         }
 654         if (*options == ',')
 655                 options++;
 656         *data = (void *) options;
 657         return sb_block;
 658 }
 659
 660 static int parse_options (char * options, struct super_block *sb,
 661                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 662 {
 663         struct ext3_sb_info *sbi = EXT3_SB(sb);
 664         char * p;
 665         substring_t args[MAX_OPT_ARGS];
 666         int data_opt = 0;
 667         int option;
 668 #ifdef CONFIG_QUOTA
 669         int qtype;
 670 #endif
 671
 672         if (!options)
 673                 return 1;
 674
 675         while ((p = strsep (&options, ",")) != NULL) {
 676                 int token;
 677                 if (!*p)
 678                         continue;
 679
 680                 token = match_token(p, tokens, args);
 681                 switch (token) {
 682                 case Opt_bsd_df:
 683                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 684                         break;
 685                 case Opt_minix_df:
 686                         set_opt (sbi->s_mount_opt, MINIX_DF);
 687                         break;
 688                 case Opt_grpid:
 689                         set_opt (sbi->s_mount_opt, GRPID);
 690                         break;
 691                 case Opt_nogrpid:
 692                         clear_opt (sbi->s_mount_opt, GRPID);
 693                         break;
 694                 case Opt_resuid:
 695                         if (match_int(&args[0], &option))
 696                                 return 0;
 697                         sbi->s_resuid = option;
 698                         break;
 699                 case Opt_resgid:
 700                         if (match_int(&args[0], &option))
 701                                 return 0;
 702                         sbi->s_resgid = option;
 703                         break;
 704                 case Opt_sb:
 705                         /* handled by get_sb_block() instead of here */
 706                         /* *sb_block = match_int(&args[0]); */
 707                         break;
 708                 case Opt_err_panic:
 709                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 710                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 711                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 712                         break;
 713                 case Opt_err_ro:
 714                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 715                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 716                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 717                         break;
 718                 case Opt_err_cont:
 719                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 720                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 721                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 722                         break;
 723                 case Opt_nouid32:
 724                         set_opt (sbi->s_mount_opt, NO_UID32);
 725                         break;
 726                 case Opt_check:
 727 #ifdef CONFIG_EXT3_CHECK
 728                         set_opt (sbi->s_mount_opt, CHECK);
 729 #else
 730                         printk(KERN_ERR
 731                                "EXT3 Check option not supported\n");
 732 #endif
 733                         break;
 734                 case Opt_nocheck:
 735                         clear_opt (sbi->s_mount_opt, CHECK);
 736                         break;
 737                 case Opt_debug:
 738                         set_opt (sbi->s_mount_opt, DEBUG);
 739                         break;
 740                 case Opt_oldalloc:
 741                         set_opt (sbi->s_mount_opt, OLDALLOC);
 742                         break;
 743                 case Opt_orlov:
 744                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 745                         break;
 746 #ifdef CONFIG_EXT3_FS_XATTR
 747                 case Opt_user_xattr:
 748                         set_opt (sbi->s_mount_opt, XATTR_USER);
 749                         break;
 750                 case Opt_nouser_xattr:
 751                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 752                         break;
 753 #else
 754                 case Opt_user_xattr:
 755                 case Opt_nouser_xattr:
 756                         printk("EXT3 (no)user_xattr options not supported\n");
 757                         break;
 758 #endif
 759 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 760                 case Opt_acl:
 761                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 762                         break;
 763                 case Opt_noacl:
 764                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 765                         break;
 766 #else
 767                 case Opt_acl:
 768                 case Opt_noacl:
 769                         printk("EXT3 (no)acl options not supported\n");
 770                         break;
 771 #endif
 772                 case Opt_reservation:
 773                         set_opt(sbi->s_mount_opt, RESERVATION);
 774                         break;
 775                 case Opt_noreservation:
 776                         clear_opt(sbi->s_mount_opt, RESERVATION);
 777                         break;
 778                 case Opt_journal_update:
 779                         /* @@@ FIXME */
 780                         /* Eventually we will want to be able to create
 781                            a journal file here.  For now, only allow the
 782                            user to specify an existing inode to be the
 783                            journal file. */
 784                         if (is_remount) {
 785                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 786                                        "journal on remount\n");
 787                                 return 0;
 788                         }
 789                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 790                         break;
 791                 case Opt_journal_inum:
 792                         if (is_remount) {
 793                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 794                                        "journal on remount\n");
 795                                 return 0;
 796                         }
 797                         if (match_int(&args[0], &option))
 798                                 return 0;
 799                         *inum = option;
 800                         break;
 801                 case Opt_noload:
 802                         set_opt (sbi->s_mount_opt, NOLOAD);
 803                         break;
 804                 case Opt_commit:
 805                         if (match_int(&args[0], &option))
 806                                 return 0;
 807                         if (option < 0)
 808                                 return 0;
 809                         if (option == 0)
 810                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 811                         sbi->s_commit_interval = HZ * option;
 812                         break;
 813                 case Opt_data_journal:
 814                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 815                         goto datacheck;
 816                 case Opt_data_ordered:
 817                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 818                         goto datacheck;
 819                 case Opt_data_writeback:
 820                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 821                 datacheck:
 822                         if (is_remount) {
 823                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 824                                                 != data_opt) {
 825                                         printk(KERN_ERR
 826                                                 "EXT3-fs: cannot change data "
 827                                                 "mode on remount\n");
 828                                         return 0;
 829                                 }
 830                         } else {
 831                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 832                                 sbi->s_mount_opt |= data_opt;
 833                         }
 834                         break;
 835 #ifdef CONFIG_QUOTA
 836                 case Opt_usrjquota:
 837                         qtype = USRQUOTA;
 838                         goto set_qf_name;
 839                 case Opt_grpjquota:
 840                         qtype = GRPQUOTA;
 841 set_qf_name:
 842                         if (sb_any_quota_enabled(sb)) {
 843                                 printk(KERN_ERR
 844                                         "EXT3-fs: Cannot change journalled "
 845                                         "quota options when quota turned on.\n");
 846                                 return 0;
 847                         }
 848                         if (sbi->s_qf_names[qtype]) {
 849                                 printk(KERN_ERR
 850                                         "EXT3-fs: %s quota file already "
 851                                         "specified.\n", QTYPE2NAME(qtype));
 852                                 return 0;
 853                         }
 854                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 855                         if (!sbi->s_qf_names[qtype]) {
 856                                 printk(KERN_ERR
 857                                         "EXT3-fs: not enough memory for "
 858                                         "storing quotafile name.\n");
 859                                 return 0;
 860                         }
 861                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 862                                 printk(KERN_ERR
 863                                         "EXT3-fs: quotafile must be on "
 864                                         "filesystem root.\n");
 865                                 kfree(sbi->s_qf_names[qtype]);
 866                                 sbi->s_qf_names[qtype] = NULL;
 867                                 return 0;
 868                         }
 869                         break;
 870                 case Opt_offusrjquota:
 871                         qtype = USRQUOTA;
 872                         goto clear_qf_name;
 873                 case Opt_offgrpjquota:
 874                         qtype = GRPQUOTA;
 875 clear_qf_name:
 876                         if (sb_any_quota_enabled(sb)) {
 877                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 878                                         "journalled quota options when "
 879                                         "quota turned on.\n");
 880                                 return 0;
 881                         }
 882                         if (sbi->s_qf_names[qtype]) {
 883                                 kfree(sbi->s_qf_names[qtype]);
 884                                 sbi->s_qf_names[qtype] = NULL;
 885                         }
 886                         break;
 887                 case Opt_jqfmt_vfsold:
 888                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 889                         break;
 890                 case Opt_jqfmt_vfsv0:
 891                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 892                         break;
 893 #else
 894                 case Opt_usrjquota:
 895                 case Opt_grpjquota:
 896                 case Opt_offusrjquota:
 897                 case Opt_offgrpjquota:
 898                 case Opt_jqfmt_vfsold:
 899                 case Opt_jqfmt_vfsv0:
 900                         printk(KERN_ERR
 901                                 "EXT3-fs: journalled quota options not "
 902                                 "supported.\n");
 903                         break;
 904 #endif
 905                 case Opt_abort:
 906                         set_opt(sbi->s_mount_opt, ABORT);
 907                         break;
 908                 case Opt_ignore:
 909                         break;
 910                 case Opt_resize:
 911                         if (!n_blocks_count) {
 912                                 printk("EXT3-fs: resize option only available "
 913                                         "for remount\n");
 914                                 return 0;
 915                         }
 916                         match_int(&args[0], &option);
 917                         *n_blocks_count = option;
 918                         break;
 919                 default:
 920                         printk (KERN_ERR
 921                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 922                                 "or missing value\n", p);
 923                         return 0;
 924                 }
 925         }
 926 #ifdef CONFIG_QUOTA
 927         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 928             sbi->s_qf_names[GRPQUOTA])) {
 929                 printk(KERN_ERR
 930                         "EXT3-fs: journalled quota format not specified.\n");
 931                 return 0;
 932         }
 933 #endif
 934
 935         return 1;
 936 }
 937
 938 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 939                             int read_only)
 940 {
 941         struct ext3_sb_info *sbi = EXT3_SB(sb);
 942         int res = 0;
 943
 944         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 945                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 946                         "forcing read-only mode\n");
 947                 res = MS_RDONLY;
 948         }
 949         if (read_only)
 950                 return res;
 951         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 952                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 953                         "running e2fsck is recommended\n");
 954         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 955                 printk (KERN_WARNING
 956                         "EXT3-fs warning: mounting fs with errors, "
 957                         "running e2fsck is recommended\n");
 958         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 959                  le16_to_cpu(es->s_mnt_count) >=
 960                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 961                 printk (KERN_WARNING
 962                         "EXT3-fs warning: maximal mount count reached, "
 963                         "running e2fsck is recommended\n");
 964         else if (le32_to_cpu(es->s_checkinterval) &&
 965                 (le32_to_cpu(es->s_lastcheck) +
 966                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 967                 printk (KERN_WARNING
 968                         "EXT3-fs warning: checktime reached, "
 969                         "running e2fsck is recommended\n");
 970 #if 0
 971                 /* @@@ We _will_ want to clear the valid bit if we find
 972                    inconsistencies, to force a fsck at reboot.  But for
 973                    a plain journaled filesystem we can keep it set as
 974                    valid forever! :) */
 975         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 976 #endif
 977         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 978                 es->s_max_mnt_count =
 979                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 980         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 981         es->s_mtime = cpu_to_le32(get_seconds());
 982         ext3_update_dynamic_rev(sb);
 983         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 984
 985         ext3_commit_super(sb, es, 1);
 986         if (test_opt(sb, DEBUG))
 987                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 988                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 989                         sb->s_blocksize,
 990                         sbi->s_groups_count,
 991                         EXT3_BLOCKS_PER_GROUP(sb),
 992                         EXT3_INODES_PER_GROUP(sb),
 993                         sbi->s_mount_opt);
 994
 995         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 996         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 997                 char b[BDEVNAME_SIZE];
 998
 999                 printk("external journal on %s\n",
1000                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1001         } else {
1002                 printk("internal journal\n");
1003         }
1004 #ifdef CONFIG_EXT3_CHECK
1005         if (test_opt (sb, CHECK)) {
1006                 ext3_check_blocks_bitmap (sb);
1007                 ext3_check_inodes_bitmap (sb);
1008         }
1009 #endif
1010         return res;
1011 }
1012
1013 static int ext3_check_descriptors (struct super_block * sb)
1014 {
1015         struct ext3_sb_info *sbi = EXT3_SB(sb);
1016         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1017         struct ext3_group_desc * gdp = NULL;
1018         int desc_block = 0;
1019         int i;
1020
1021         ext3_debug ("Checking group descriptors");
1022
1023         for (i = 0; i < sbi->s_groups_count; i++)
1024         {
1025                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1026                         gdp = (struct ext3_group_desc *)
1027                                         sbi->s_group_desc[desc_block++]->b_data;
1028                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1029                     le32_to_cpu(gdp->bg_block_bitmap) >=
1030                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1031                 {
1032                         ext3_error (sb, "ext3_check_descriptors",
1033                                     "Block bitmap for group %d"
1034                                     " not in group (block %lu)!",
1035                                     i, (unsigned long)
1036                                         le32_to_cpu(gdp->bg_block_bitmap));
1037                         return 0;
1038                 }
1039                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1040                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1041                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1042                 {
1043                         ext3_error (sb, "ext3_check_descriptors",
1044                                     "Inode bitmap for group %d"
1045                                     " not in group (block %lu)!",
1046                                     i, (unsigned long)
1047                                         le32_to_cpu(gdp->bg_inode_bitmap));
1048                         return 0;
1049                 }
1050                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1051                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1052                     block + EXT3_BLOCKS_PER_GROUP(sb))
1053                 {
1054                         ext3_error (sb, "ext3_check_descriptors",
1055                                     "Inode table for group %d"
1056                                     " not in group (block %lu)!",
1057                                     i, (unsigned long)
1058                                         le32_to_cpu(gdp->bg_inode_table));
1059                         return 0;
1060                 }
1061                 block += EXT3_BLOCKS_PER_GROUP(sb);
1062                 gdp++;
1063         }
1064
1065         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1066         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1067         return 1;
1068 }
1069
1070
1071 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1072  * the superblock) which were deleted from all directories, but held open by
1073  * a process at the time of a crash.  We walk the list and try to delete these
1074  * inodes at recovery time (only with a read-write filesystem).
1075  *
1076  * In order to keep the orphan inode chain consistent during traversal (in
1077  * case of crash during recovery), we link each inode into the superblock
1078  * orphan list_head and handle it the same way as an inode deletion during
1079  * normal operation (which journals the operations for us).
1080  *
1081  * We only do an iget() and an iput() on each inode, which is very safe if we
1082  * accidentally point at an in-use or already deleted inode.  The worst that
1083  * can happen in this case is that we get a "bit already cleared" message from
1084  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1085  * e2fsck was run on this filesystem, and it must have already done the orphan
1086  * inode cleanup for us, so we can safely abort without any further action.
1087  */
1088 static void ext3_orphan_cleanup (struct super_block * sb,
1089                                  struct ext3_super_block * es)
1090 {
1091         unsigned int s_flags = sb->s_flags;
1092         int nr_orphans = 0, nr_truncates = 0;
1093 #ifdef CONFIG_QUOTA
1094         int i;
1095 #endif
1096         if (!es->s_last_orphan) {
1097                 jbd_debug(4, "no orphan inodes to clean up\n");
1098                 return;
1099         }
1100
1101         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1102                 if (es->s_last_orphan)
1103                         jbd_debug(1, "Errors on filesystem, "
1104                                   "clearing orphan list.\n");
1105                 es->s_last_orphan = 0;
1106                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1107                 return;
1108         }
1109
1110         if (s_flags & MS_RDONLY) {
1111                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1112                        sb->s_id);
1113                 sb->s_flags &= ~MS_RDONLY;
1114         }
1115 #ifdef CONFIG_QUOTA
1116         /* Needed for iput() to work correctly and not trash data */
1117         sb->s_flags |= MS_ACTIVE;
1118         /* Turn on quotas so that they are updated correctly */
1119         for (i = 0; i < MAXQUOTAS; i++) {
1120                 if (EXT3_SB(sb)->s_qf_names[i]) {
1121                         int ret = ext3_quota_on_mount(sb, i);
1122                         if (ret < 0)
1123                                 printk(KERN_ERR
1124                                         "EXT3-fs: Cannot turn on journalled "
1125                                         "quota: error %d\n", ret);
1126                 }
1127         }
1128 #endif
1129
1130         while (es->s_last_orphan) {
1131                 struct inode *inode;
1132
1133                 if (!(inode =
1134                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1135                         es->s_last_orphan = 0;
1136                         break;
1137                 }
1138
1139                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1140                 DQUOT_INIT(inode);
1141                 if (inode->i_nlink) {
1142                         printk(KERN_DEBUG
1143                                 "%s: truncating inode %ld to %Ld bytes\n",
1144                                 __FUNCTION__, inode->i_ino, inode->i_size);
1145                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1146                                   inode->i_ino, inode->i_size);
1147                         ext3_truncate(inode);
1148                         nr_truncates++;
1149                 } else {
1150                         printk(KERN_DEBUG
1151                                 "%s: deleting unreferenced inode %ld\n",
1152                                 __FUNCTION__, inode->i_ino);
1153                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1154                                   inode->i_ino);
1155                         nr_orphans++;
1156                 }
1157                 iput(inode);  /* The delete magic happens here! */
1158         }
1159
1160 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1161
1162         if (nr_orphans)
1163                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1164                        sb->s_id, PLURAL(nr_orphans));
1165         if (nr_truncates)
1166                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1167                        sb->s_id, PLURAL(nr_truncates));
1168 #ifdef CONFIG_QUOTA
1169         /* Turn quotas off */
1170         for (i = 0; i < MAXQUOTAS; i++) {
1171                 if (sb_dqopt(sb)->files[i])
1172                         ext3_quota_off_mount(sb, i);
1173         }
1174 #endif
1175         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1176 }
1177
1178 #define log2(n) ffz(~(n))
1179
1180 /*
1181  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1182  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1183  * We need to be 1 filesystem block less than the 2^32 sector limit.
1184  */
1185 static loff_t ext3_max_size(int bits)
1186 {
1187         loff_t res = EXT3_NDIR_BLOCKS;
1188         res += 1LL << (bits-2);
1189         res += 1LL << (2*(bits-2));
1190         res += 1LL << (3*(bits-2));
1191         res <<= bits;
1192         if (res > (512LL << 32) - (1 << bits))
1193                 res = (512LL << 32) - (1 << bits);
1194         return res;
1195 }
1196
1197 static unsigned long descriptor_loc(struct super_block *sb,
1198                                     unsigned long logic_sb_block,
1199                                     int nr)
1200 {
1201         struct ext3_sb_info *sbi = EXT3_SB(sb);
1202         unsigned long bg, first_data_block, first_meta_bg;
1203         int has_super = 0;
1204
1205         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1206         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1207
1208         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1209             nr < first_meta_bg)
1210                 return (logic_sb_block + nr + 1);
1211         bg = sbi->s_desc_per_block * nr;
1212         if (ext3_bg_has_super(sb, bg))
1213                 has_super = 1;
1214         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1215 }
1216
1217
1218 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1219 {
1220         struct buffer_head * bh;
1221         struct ext3_super_block *es = 0;
1222         struct ext3_sb_info *sbi;
1223         unsigned long block;
1224         unsigned long sb_block = get_sb_block(&data);
1225         unsigned long logic_sb_block;
1226         unsigned long offset = 0;
1227         unsigned long journal_inum = 0;
1228         unsigned long def_mount_opts;
1229         struct inode *root;
1230         int blocksize;
1231         int hblock;
1232         int db_count;
1233         int i;
1234         int needs_recovery;
1235
1236         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1237         if (!sbi)
1238                 return -ENOMEM;
1239         sb->s_fs_info = sbi;
1240         memset(sbi, 0, sizeof(*sbi));
1241         sbi->s_mount_opt = 0;
1242         sbi->s_resuid = EXT3_DEF_RESUID;
1243         sbi->s_resgid = EXT3_DEF_RESGID;
1244
1245         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1246         if (!blocksize) {
1247                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1248                 goto out_fail;
1249         }
1250
1251         /*
1252          * The ext3 superblock will not be buffer aligned for other than 1kB
1253          * block sizes.  We need to calculate the offset from buffer start.
1254          */
1255         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1256                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1257                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1258         } else {
1259                 logic_sb_block = sb_block;
1260         }
1261
1262         if (!(bh = sb_bread(sb, logic_sb_block))) {
1263                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1264                 goto out_fail;
1265         }
1266         /*
1267          * Note: s_es must be initialized as soon as possible because
1268          *       some ext3 macro-instructions depend on its value
1269          */
1270         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1271         sbi->s_es = es;
1272         sb->s_magic = le16_to_cpu(es->s_magic);
1273         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1274                 if (!silent)
1275                         printk(KERN_ERR
1276                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1277                                sb->s_id);
1278                 goto failed_mount;
1279         }
1280
1281         /* Set defaults before we parse the mount options */
1282         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1283         if (def_mount_opts & EXT3_DEFM_DEBUG)
1284                 set_opt(sbi->s_mount_opt, DEBUG);
1285         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1286                 set_opt(sbi->s_mount_opt, GRPID);
1287         if (def_mount_opts & EXT3_DEFM_UID16)
1288                 set_opt(sbi->s_mount_opt, NO_UID32);
1289         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1290                 set_opt(sbi->s_mount_opt, XATTR_USER);
1291         if (def_mount_opts & EXT3_DEFM_ACL)
1292                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1293         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1294                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1295         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1296                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1297         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1298                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1299
1300         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1301                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1302         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1303                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1304
1305         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1306         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1307
1308         set_opt(sbi->s_mount_opt, RESERVATION);
1309
1310         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1311                 goto failed_mount;
1312
1313         sb->s_flags |= MS_ONE_SECOND;
1314         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1315                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1316
1317         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1318             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1319              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1320              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1321                 printk(KERN_WARNING
1322                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1323                        "running e2fsck is recommended\n");
1324         /*
1325          * Check feature flags regardless of the revision level, since we
1326          * previously didn't change the revision level when setting the flags,
1327          * so there is a chance incompat flags are set on a rev 0 filesystem.
1328          */
1329         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1330                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1331                        "unsupported optional features (%x).\n",
1332                        sb->s_id, i);
1333                 goto failed_mount;
1334         }
1335         if (!(sb->s_flags & MS_RDONLY) &&
1336             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1337                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1338                        "unsupported optional features (%x).\n",
1339                        sb->s_id, i);
1340                 goto failed_mount;
1341         }
1342         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1343
1344         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1345             blocksize > EXT3_MAX_BLOCK_SIZE) {
1346                 printk(KERN_ERR
1347                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1348                        blocksize, sb->s_id);
1349                 goto failed_mount;
1350         }
1351
1352         hblock = bdev_hardsect_size(sb->s_bdev);
1353         if (sb->s_blocksize != blocksize) {
1354                 /*
1355                  * Make sure the blocksize for the filesystem is larger
1356                  * than the hardware sectorsize for the machine.
1357                  */
1358                 if (blocksize < hblock) {
1359                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1360                                "device blocksize %d.\n", blocksize, hblock);
1361                         goto failed_mount;
1362                 }
1363
1364                 brelse (bh);
1365                 sb_set_blocksize(sb, blocksize);
1366                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1367                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1368                 bh = sb_bread(sb, logic_sb_block);
1369                 if (!bh) {
1370                         printk(KERN_ERR
1371                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1372                         goto failed_mount;
1373                 }
1374                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1375                 sbi->s_es = es;
1376                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1377                         printk (KERN_ERR
1378                                 "EXT3-fs: Magic mismatch, very weird !\n");
1379                         goto failed_mount;
1380                 }
1381         }
1382
1383         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1384
1385         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1386                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1387                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1388         } else {
1389                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1390                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1391                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1392                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1393                     (sbi->s_inode_size > blocksize)) {
1394                         printk (KERN_ERR
1395                                 "EXT3-fs: unsupported inode size: %d\n",
1396                                 sbi->s_inode_size);
1397                         goto failed_mount;
1398                 }
1399         }
1400         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1401                                    le32_to_cpu(es->s_log_frag_size);
1402         if (blocksize != sbi->s_frag_size) {
1403                 printk(KERN_ERR
1404                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1405                        sbi->s_frag_size, blocksize);
1406                 goto failed_mount;
1407         }
1408         sbi->s_frags_per_block = 1;
1409         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1410         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1411         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1412         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1413         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1414         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1415         sbi->s_sbh = bh;
1416         sbi->s_mount_state = le16_to_cpu(es->s_state);
1417         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1418         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1419         for (i=0; i < 4; i++)
1420                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1421         sbi->s_def_hash_version = es->s_def_hash_version;
1422
1423         if (sbi->s_blocks_per_group > blocksize * 8) {
1424                 printk (KERN_ERR
1425                         "EXT3-fs: #blocks per group too big: %lu\n",
1426                         sbi->s_blocks_per_group);
1427                 goto failed_mount;
1428         }
1429         if (sbi->s_frags_per_group > blocksize * 8) {
1430                 printk (KERN_ERR
1431                         "EXT3-fs: #fragments per group too big: %lu\n",
1432                         sbi->s_frags_per_group);
1433                 goto failed_mount;
1434         }
1435         if (sbi->s_inodes_per_group > blocksize * 8) {
1436                 printk (KERN_ERR
1437                         "EXT3-fs: #inodes per group too big: %lu\n",
1438                         sbi->s_inodes_per_group);
1439                 goto failed_mount;
1440         }
1441
1442         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1443                                le32_to_cpu(es->s_first_data_block) +
1444                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1445                               EXT3_BLOCKS_PER_GROUP(sb);
1446         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1447                    EXT3_DESC_PER_BLOCK(sb);
1448         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1449                                     GFP_KERNEL);
1450         if (sbi->s_group_desc == NULL) {
1451                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1452                 goto failed_mount;
1453         }
1454         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1455                         GFP_KERNEL);
1456         if (!sbi->s_debts) {
1457                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1458                 goto failed_mount2;
1459         }
1460         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1461
1462         percpu_counter_init(&sbi->s_freeblocks_counter);
1463         percpu_counter_init(&sbi->s_freeinodes_counter);
1464         percpu_counter_init(&sbi->s_dirs_counter);
1465         bgl_lock_init(&sbi->s_blockgroup_lock);
1466
1467         for (i = 0; i < db_count; i++) {
1468                 block = descriptor_loc(sb, logic_sb_block, i);
1469                 sbi->s_group_desc[i] = sb_bread(sb, block);
1470                 if (!sbi->s_group_desc[i]) {
1471                         printk (KERN_ERR "EXT3-fs: "
1472                                 "can't read group descriptor %d\n", i);
1473                         db_count = i;
1474                         goto failed_mount2;
1475                 }
1476         }
1477         if (!ext3_check_descriptors (sb)) {
1478                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1479                 goto failed_mount2;
1480         }
1481         sbi->s_gdb_count = db_count;
1482         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1483         spin_lock_init(&sbi->s_next_gen_lock);
1484         /* per fileystem reservation list head & lock */
1485         spin_lock_init(&sbi->s_rsv_window_lock);
1486         INIT_LIST_HEAD(&sbi->s_rsv_window_head.rsv_list);
1487         sbi->s_rsv_window_head.rsv_start = 0;
1488         sbi->s_rsv_window_head.rsv_end = 0;
1489         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1490         atomic_set(&sbi->s_rsv_window_head.rsv_goal_size, 0);
1491
1492         /*
1493          * set up enough so that it can read an inode
1494          */
1495         sb->s_op = &ext3_sops;
1496         sb->s_export_op = &ext3_export_ops;
1497 #ifdef CONFIG_QUOTA
1498         sb->s_qcop = &ext3_qctl_operations;
1499         sb->dq_op = &ext3_quota_operations;
1500 #endif
1501         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1502
1503         sb->s_root = 0;
1504
1505         needs_recovery = (es->s_last_orphan != 0 ||
1506                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1507                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1508
1509         /*
1510          * The first inode we look at is the journal inode.  Don't try
1511          * root first: it may be modified in the journal!
1512          */
1513         if (!test_opt(sb, NOLOAD) &&
1514             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1515                 if (ext3_load_journal(sb, es))
1516                         goto failed_mount2;
1517         } else if (journal_inum) {
1518                 if (ext3_create_journal(sb, es, journal_inum))
1519                         goto failed_mount2;
1520         } else {
1521                 if (!silent)
1522                         printk (KERN_ERR
1523                                 "ext3: No journal on filesystem on %s\n",
1524                                 sb->s_id);
1525                 goto failed_mount2;
1526         }
1527
1528         /* We have now updated the journal if required, so we can
1529          * validate the data journaling mode. */
1530         switch (test_opt(sb, DATA_FLAGS)) {
1531         case 0:
1532                 /* No mode set, assume a default based on the journal
1533                    capabilities: ORDERED_DATA if the journal can
1534                    cope, else JOURNAL_DATA */
1535                 if (journal_check_available_features
1536                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1537                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1538                 else
1539                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1540                 break;
1541
1542         case EXT3_MOUNT_ORDERED_DATA:
1543         case EXT3_MOUNT_WRITEBACK_DATA:
1544                 if (!journal_check_available_features
1545                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1546                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1547                                "requested data journaling mode\n");
1548                         goto failed_mount3;
1549                 }
1550         default:
1551                 break;
1552         }
1553
1554         /*
1555          * The journal_load will have done any necessary log recovery,
1556          * so we can safely mount the rest of the filesystem now.
1557          */
1558
1559         root = iget(sb, EXT3_ROOT_INO);
1560         sb->s_root = d_alloc_root(root);
1561         if (!sb->s_root) {
1562                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1563                 iput(root);
1564                 goto failed_mount3;
1565         }
1566         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1567                 dput(sb->s_root);
1568                 sb->s_root = NULL;
1569                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1570                 goto failed_mount3;
1571         }
1572
1573         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1574         /*
1575          * akpm: core read_super() calls in here with the superblock locked.
1576          * That deadlocks, because orphan cleanup needs to lock the superblock
1577          * in numerous places.  Here we just pop the lock - it's relatively
1578          * harmless, because we are now ready to accept write_super() requests,
1579          * and aviro says that's the only reason for hanging onto the
1580          * superblock lock.
1581          */
1582         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1583         ext3_orphan_cleanup(sb, es);
1584         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1585         if (needs_recovery)
1586                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1587         ext3_mark_recovery_complete(sb, es);
1588         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1589                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1590                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1591                 "writeback");
1592
1593         percpu_counter_mod(&sbi->s_freeblocks_counter,
1594                 ext3_count_free_blocks(sb));
1595         percpu_counter_mod(&sbi->s_freeinodes_counter,
1596                 ext3_count_free_inodes(sb));
1597         percpu_counter_mod(&sbi->s_dirs_counter,
1598                 ext3_count_dirs(sb));
1599
1600         return 0;
1601
1602 failed_mount3:
1603         journal_destroy(sbi->s_journal);
1604 failed_mount2:
1605         kfree(sbi->s_debts);
1606         for (i = 0; i < db_count; i++)
1607                 brelse(sbi->s_group_desc[i]);
1608         kfree(sbi->s_group_desc);
1609 failed_mount:
1610 #ifdef CONFIG_QUOTA
1611         for (i = 0; i < MAXQUOTAS; i++) {
1612                 if (sbi->s_qf_names[i])
1613                         kfree(sbi->s_qf_names[i]);
1614         }
1615 #endif
1616         ext3_blkdev_remove(sbi);
1617         brelse(bh);
1618 out_fail:
1619         sb->s_fs_info = NULL;
1620         kfree(sbi);
1621         return -EINVAL;
1622 }
1623
1624 /*
1625  * Setup any per-fs journal parameters now.  We'll do this both on
1626  * initial mount, once the journal has been initialised but before we've
1627  * done any recovery; and again on any subsequent remount.
1628  */
1629 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1630                                      journal_t *journal)
1631 {
1632         if (sbi->s_commit_interval)
1633                 journal->j_commit_interval = sbi->s_commit_interval;
1634         /* We could also set up an ext3-specific default for the commit
1635          * interval here, but for now we'll just fall back to the jbd
1636          * default. */
1637 }
1638
1639
1640 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1641 {
1642         struct inode *journal_inode;
1643         journal_t *journal;
1644
1645         /* First, test for the existence of a valid inode on disk.  Bad
1646          * things happen if we iget() an unused inode, as the subsequent
1647          * iput() will try to delete it. */
1648
1649         journal_inode = iget(sb, journal_inum);
1650         if (!journal_inode) {
1651                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1652                 return NULL;
1653         }
1654         if (!journal_inode->i_nlink) {
1655                 make_bad_inode(journal_inode);
1656                 iput(journal_inode);
1657                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1658                 return NULL;
1659         }
1660
1661         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1662                   journal_inode, journal_inode->i_size);
1663         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1664                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1665                 iput(journal_inode);
1666                 return NULL;
1667         }
1668
1669         journal = journal_init_inode(journal_inode);
1670         if (!journal) {
1671                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1672                 iput(journal_inode);
1673         }
1674         journal->j_private = sb;
1675         ext3_init_journal_params(EXT3_SB(sb), journal);
1676         return journal;
1677 }
1678
1679 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1680                                        dev_t j_dev)
1681 {
1682         struct buffer_head * bh;
1683         journal_t *journal;
1684         int start;
1685         int len;
1686         int hblock, blocksize;
1687         unsigned long sb_block;
1688         unsigned long offset;
1689         struct ext3_super_block * es;
1690         struct block_device *bdev;
1691
1692         bdev = ext3_blkdev_get(j_dev);
1693         if (bdev == NULL)
1694                 return NULL;
1695
1696         if (bd_claim(bdev, sb)) {
1697                 printk(KERN_ERR
1698                         "EXT3: failed to claim external journal device.\n");
1699                 blkdev_put(bdev);
1700                 return NULL;
1701         }
1702
1703         blocksize = sb->s_blocksize;
1704         hblock = bdev_hardsect_size(bdev);
1705         if (blocksize < hblock) {
1706                 printk(KERN_ERR
1707                         "EXT3-fs: blocksize too small for journal device.\n");
1708                 goto out_bdev;
1709         }
1710
1711         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1712         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1713         set_blocksize(bdev, blocksize);
1714         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1715                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1716                        "external journal\n");
1717                 goto out_bdev;
1718         }
1719
1720         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1721         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1722             !(le32_to_cpu(es->s_feature_incompat) &
1723               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1724                 printk(KERN_ERR "EXT3-fs: external journal has "
1725                                         "bad superblock\n");
1726                 brelse(bh);
1727                 goto out_bdev;
1728         }
1729
1730         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1731                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1732                 brelse(bh);
1733                 goto out_bdev;
1734         }
1735
1736         len = le32_to_cpu(es->s_blocks_count);
1737         start = sb_block + 1;
1738         brelse(bh);     /* we're done with the superblock */
1739
1740         journal = journal_init_dev(bdev, sb->s_bdev,
1741                                         start, len, blocksize);
1742         if (!journal) {
1743                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1744                 goto out_bdev;
1745         }
1746         journal->j_private = sb;
1747         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1748         wait_on_buffer(journal->j_sb_buffer);
1749         if (!buffer_uptodate(journal->j_sb_buffer)) {
1750                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1751                 goto out_journal;
1752         }
1753         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1754                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1755                                         "user (unsupported) - %d\n",
1756                         ntohl(journal->j_superblock->s_nr_users));
1757                 goto out_journal;
1758         }
1759         EXT3_SB(sb)->journal_bdev = bdev;
1760         ext3_init_journal_params(EXT3_SB(sb), journal);
1761         return journal;
1762 out_journal:
1763         journal_destroy(journal);
1764 out_bdev:
1765         ext3_blkdev_put(bdev);
1766         return NULL;
1767 }
1768
1769 static int ext3_load_journal(struct super_block * sb,
1770                              struct ext3_super_block * es)
1771 {
1772         journal_t *journal;
1773         int journal_inum = le32_to_cpu(es->s_journal_inum);
1774         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1775         int err = 0;
1776         int really_read_only;
1777
1778         really_read_only = bdev_read_only(sb->s_bdev);
1779
1780         /*
1781          * Are we loading a blank journal or performing recovery after a
1782          * crash?  For recovery, we need to check in advance whether we
1783          * can get read-write access to the device.
1784          */
1785
1786         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1787                 if (sb->s_flags & MS_RDONLY) {
1788                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1789                                         "required on readonly filesystem.\n");
1790                         if (really_read_only) {
1791                                 printk(KERN_ERR "EXT3-fs: write access "
1792                                         "unavailable, cannot proceed.\n");
1793                                 return -EROFS;
1794                         }
1795                         printk (KERN_INFO "EXT3-fs: write access will "
1796                                         "be enabled during recovery.\n");
1797                 }
1798         }
1799
1800         if (journal_inum && journal_dev) {
1801                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1802                        "and inode journals!\n");
1803                 return -EINVAL;
1804         }
1805
1806         if (journal_inum) {
1807                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1808                         return -EINVAL;
1809         } else {
1810                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1811                         return -EINVAL;
1812         }
1813
1814         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1815                 err = journal_update_format(journal);
1816                 if (err)  {
1817                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1818                         journal_destroy(journal);
1819                         return err;
1820                 }
1821         }
1822
1823         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1824                 err = journal_wipe(journal, !really_read_only);
1825         if (!err)
1826                 err = journal_load(journal);
1827
1828         if (err) {
1829                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1830                 journal_destroy(journal);
1831                 return err;
1832         }
1833
1834         EXT3_SB(sb)->s_journal = journal;
1835         ext3_clear_journal_err(sb, es);
1836         return 0;
1837 }
1838
1839 static int ext3_create_journal(struct super_block * sb,
1840                                struct ext3_super_block * es,
1841                                int journal_inum)
1842 {
1843         journal_t *journal;
1844
1845         if (sb->s_flags & MS_RDONLY) {
1846                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1847                                 "create journal.\n");
1848                 return -EROFS;
1849         }
1850
1851         if (!(journal = ext3_get_journal(sb, journal_inum)))
1852                 return -EINVAL;
1853
1854         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1855                journal_inum);
1856
1857         if (journal_create(journal)) {
1858                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1859                 journal_destroy(journal);
1860                 return -EIO;
1861         }
1862
1863         EXT3_SB(sb)->s_journal = journal;
1864
1865         ext3_update_dynamic_rev(sb);
1866         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1867         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1868
1869         es->s_journal_inum = cpu_to_le32(journal_inum);
1870         sb->s_dirt = 1;
1871
1872         /* Make sure we flush the recovery flag to disk. */
1873         ext3_commit_super(sb, es, 1);
1874
1875         return 0;
1876 }
1877
1878 static void ext3_commit_super (struct super_block * sb,
1879                                struct ext3_super_block * es,
1880                                int sync)
1881 {
1882         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1883
1884         if (!sbh)
1885                 return;
1886         es->s_wtime = cpu_to_le32(get_seconds());
1887         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1888         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1889         BUFFER_TRACE(sbh, "marking dirty");
1890         mark_buffer_dirty(sbh);
1891         if (sync)
1892                 sync_dirty_buffer(sbh);
1893 }
1894
1895
1896 /*
1897  * Have we just finished recovery?  If so, and if we are mounting (or
1898  * remounting) the filesystem readonly, then we will end up with a
1899  * consistent fs on disk.  Record that fact.
1900  */
1901 static void ext3_mark_recovery_complete(struct super_block * sb,
1902                                         struct ext3_super_block * es)
1903 {
1904         journal_flush(EXT3_SB(sb)->s_journal);
1905         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1906             sb->s_flags & MS_RDONLY) {
1907                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1908                 sb->s_dirt = 0;
1909                 ext3_commit_super(sb, es, 1);
1910         }
1911 }
1912
1913 /*
1914  * If we are mounting (or read-write remounting) a filesystem whose journal
1915  * has recorded an error from a previous lifetime, move that error to the
1916  * main filesystem now.
1917  */
1918 static void ext3_clear_journal_err(struct super_block * sb,
1919                                    struct ext3_super_block * es)
1920 {
1921         journal_t *journal;
1922         int j_errno;
1923         const char *errstr;
1924
1925         journal = EXT3_SB(sb)->s_journal;
1926
1927         /*
1928          * Now check for any error status which may have been recorded in the
1929          * journal by a prior ext3_error() or ext3_abort()
1930          */
1931
1932         j_errno = journal_errno(journal);
1933         if (j_errno) {
1934                 char nbuf[16];
1935
1936                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1937                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1938                              "from previous mount: %s", errstr);
1939                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1940                              "filesystem check.");
1941
1942                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1943                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1944                 ext3_commit_super (sb, es, 1);
1945
1946                 journal_clear_err(journal);
1947         }
1948 }
1949
1950 /*
1951  * Force the running and committing transactions to commit,
1952  * and wait on the commit.
1953  */
1954 int ext3_force_commit(struct super_block *sb)
1955 {
1956         journal_t *journal;
1957         int ret;
1958
1959         if (sb->s_flags & MS_RDONLY)
1960                 return 0;
1961
1962         journal = EXT3_SB(sb)->s_journal;
1963         sb->s_dirt = 0;
1964         ret = ext3_journal_force_commit(journal);
1965         return ret;
1966 }
1967
1968 /*
1969  * Ext3 always journals updates to the superblock itself, so we don't
1970  * have to propagate any other updates to the superblock on disk at this
1971  * point.  Just start an async writeback to get the buffers on their way
1972  * to the disk.
1973  *
1974  * This implicitly triggers the writebehind on sync().
1975  */
1976
1977 void ext3_write_super (struct super_block * sb)
1978 {
1979         if (down_trylock(&sb->s_lock) == 0)
1980                 BUG();
1981         sb->s_dirt = 0;
1982 }
1983
1984 static int ext3_sync_fs(struct super_block *sb, int wait)
1985 {
1986         tid_t target;
1987
1988         sb->s_dirt = 0;
1989         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1990                 if (wait)
1991                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1992         }
1993         return 0;
1994 }
1995
1996 /*
1997  * LVM calls this function before a (read-only) snapshot is created.  This
1998  * gives us a chance to flush the journal completely and mark the fs clean.
1999  */
2000 void ext3_write_super_lockfs(struct super_block *sb)
2001 {
2002         sb->s_dirt = 0;
2003
2004         if (!(sb->s_flags & MS_RDONLY)) {
2005                 journal_t *journal = EXT3_SB(sb)->s_journal;
2006
2007                 /* Now we set up the journal barrier. */
2008                 journal_lock_updates(journal);
2009                 journal_flush(journal);
2010
2011                 /* Journal blocked and flushed, clear needs_recovery flag. */
2012                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2013                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2014         }
2015 }
2016
2017 /*
2018  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2019  * flag here, even though the filesystem is not technically dirty yet.
2020  */
2021 void ext3_unlockfs(struct super_block *sb)
2022 {
2023         if (!(sb->s_flags & MS_RDONLY)) {
2024                 lock_super(sb);
2025                 /* Reser the needs_recovery flag before the fs is unlocked. */
2026                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2027                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2028                 unlock_super(sb);
2029                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2030         }
2031 }
2032
2033 int ext3_remount (struct super_block * sb, int * flags, char * data)
2034 {
2035         struct ext3_super_block * es;
2036         struct ext3_sb_info *sbi = EXT3_SB(sb);
2037         unsigned long tmp;
2038         unsigned long n_blocks_count = 0;
2039
2040         /*
2041          * Allow the "check" option to be passed as a remount option.
2042          */
2043         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2044                 return -EINVAL;
2045
2046         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2047                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2048
2049         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2050                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2051
2052         es = sbi->s_es;
2053
2054         ext3_init_journal_params(sbi, sbi->s_journal);
2055
2056         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2057                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2058                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2059                         return -EROFS;
2060
2061                 if (*flags & MS_RDONLY) {
2062                         /*
2063                          * First of all, the unconditional stuff we have to do
2064                          * to disable replay of the journal when we next remount
2065                          */
2066                         sb->s_flags |= MS_RDONLY;
2067
2068                         /*
2069                          * OK, test if we are remounting a valid rw partition
2070                          * readonly, and if so set the rdonly flag and then
2071                          * mark the partition as valid again.
2072                          */
2073                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2074                             (sbi->s_mount_state & EXT3_VALID_FS))
2075                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2076
2077                         ext3_mark_recovery_complete(sb, es);
2078                 } else {
2079                         int ret;
2080                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2081                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2082                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2083                                        "remount RDWR because of unsupported "
2084                                        "optional features (%x).\n",
2085                                        sb->s_id, ret);
2086                                 return -EROFS;
2087                         }
2088                         /*
2089                          * Mounting a RDONLY partition read-write, so reread
2090                          * and store the current valid flag.  (It may have
2091                          * been changed by e2fsck since we originally mounted
2092                          * the partition.)
2093                          */
2094                         ext3_clear_journal_err(sb, es);
2095                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2096                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2097                                 return ret;
2098                         if (!ext3_setup_super (sb, es, 0))
2099                                 sb->s_flags &= ~MS_RDONLY;
2100                 }
2101         }
2102         return 0;
2103 }
2104
2105 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2106 {
2107         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2108         unsigned long overhead;
2109         int i;
2110
2111         if (test_opt (sb, MINIX_DF))
2112                 overhead = 0;
2113         else {
2114                 /*
2115                  * Compute the overhead (FS structures)
2116                  */
2117
2118                 /*
2119                  * All of the blocks before first_data_block are
2120                  * overhead
2121                  */
2122                 overhead = le32_to_cpu(es->s_first_data_block);
2123
2124                 /*
2125                  * Add the overhead attributed to the superblock and
2126                  * block group descriptors.  If the sparse superblocks
2127                  * feature is turned on, then not all groups have this.
2128                  */
2129                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2130                         overhead += ext3_bg_has_super(sb, i) +
2131                                 ext3_bg_num_gdb(sb, i);
2132
2133                 /*
2134                  * Every block group has an inode bitmap, a block
2135                  * bitmap, and an inode table.
2136                  */
2137                 overhead += (EXT3_SB(sb)->s_groups_count *
2138                              (2 + EXT3_SB(sb)->s_itb_per_group));
2139         }
2140
2141         buf->f_type = EXT3_SUPER_MAGIC;
2142         buf->f_bsize = sb->s_blocksize;
2143         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2144         buf->f_bfree = ext3_count_free_blocks (sb);
2145         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2146         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2147                 buf->f_bavail = 0;
2148         buf->f_files = le32_to_cpu(es->s_inodes_count);
2149         buf->f_ffree = ext3_count_free_inodes (sb);
2150         buf->f_namelen = EXT3_NAME_LEN;
2151         return 0;
2152 }
2153
2154 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2155  * is locked for write. Otherwise the are possible deadlocks:
2156  * Process 1                         Process 2
2157  * ext3_create()                     quota_sync()
2158  *   journal_start()                   write_dquot()
2159  *   DQUOT_INIT()                        down(dqio_sem)
2160  *     down(dqio_sem)                    journal_start()
2161  *
2162  */
2163
2164 #ifdef CONFIG_QUOTA
2165
2166 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2167 {
2168         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2169 }
2170
2171 static int ext3_dquot_initialize(struct inode *inode, int type)
2172 {
2173         handle_t *handle;
2174         int ret, err;
2175
2176         /* We may create quota structure so we need to reserve enough blocks */
2177         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2178         if (IS_ERR(handle))
2179                 return PTR_ERR(handle);
2180         ret = dquot_initialize(inode, type);
2181         err = ext3_journal_stop(handle);
2182         if (!ret)
2183                 ret = err;
2184         return ret;
2185 }
2186
2187 static int ext3_dquot_drop(struct inode *inode)
2188 {
2189         handle_t *handle;
2190         int ret, err;
2191
2192         /* We may delete quota structure so we need to reserve enough blocks */
2193         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2194         if (IS_ERR(handle))
2195                 return PTR_ERR(handle);
2196         ret = dquot_drop(inode);
2197         err = ext3_journal_stop(handle);
2198         if (!ret)
2199                 ret = err;
2200         return ret;
2201 }
2202
2203 static int ext3_write_dquot(struct dquot *dquot)
2204 {
2205         int ret, err;
2206         handle_t *handle;
2207
2208         handle = ext3_journal_start(dquot_to_inode(dquot),
2209                                         EXT3_QUOTA_TRANS_BLOCKS);
2210         if (IS_ERR(handle))
2211                 return PTR_ERR(handle);
2212         ret = dquot_commit(dquot);
2213         err = ext3_journal_stop(handle);
2214         if (!ret)
2215                 ret = err;
2216         return ret;
2217 }
2218
2219 static int ext3_acquire_dquot(struct dquot *dquot)
2220 {
2221         int ret, err;
2222         handle_t *handle;
2223
2224         handle = ext3_journal_start(dquot_to_inode(dquot),
2225                                         EXT3_QUOTA_INIT_BLOCKS);
2226         if (IS_ERR(handle))
2227                 return PTR_ERR(handle);
2228         ret = dquot_acquire(dquot);
2229         err = ext3_journal_stop(handle);
2230         if (!ret)
2231                 ret = err;
2232         return ret;
2233 }
2234
2235 static int ext3_release_dquot(struct dquot *dquot)
2236 {
2237         int ret, err;
2238         handle_t *handle;
2239
2240         handle = ext3_journal_start(dquot_to_inode(dquot),
2241                                         EXT3_QUOTA_INIT_BLOCKS);
2242         if (IS_ERR(handle))
2243                 return PTR_ERR(handle);
2244         ret = dquot_release(dquot);
2245         err = ext3_journal_stop(handle);
2246         if (!ret)
2247                 ret = err;
2248         return ret;
2249 }
2250
2251 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2252 {
2253         /* Are we journalling quotas? */
2254         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2255             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2256                 dquot_mark_dquot_dirty(dquot);
2257                 return ext3_write_dquot(dquot);
2258         } else {
2259                 return dquot_mark_dquot_dirty(dquot);
2260         }
2261 }
2262
2263 static int ext3_write_info(struct super_block *sb, int type)
2264 {
2265         int ret, err;
2266         handle_t *handle;
2267
2268         /* Data block + inode block */
2269         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2270         if (IS_ERR(handle))
2271                 return PTR_ERR(handle);
2272         ret = dquot_commit_info(sb, type);
2273         err = ext3_journal_stop(handle);
2274         if (!ret)
2275                 ret = err;
2276         return ret;
2277 }
2278
2279 /*
2280  * Turn on quotas during mount time - we need to find
2281  * the quota file and such...
2282  */
2283 static int ext3_quota_on_mount(struct super_block *sb, int type)
2284 {
2285         int err;
2286         struct dentry *dentry;
2287         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2288                              .hash = 0,
2289                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2290
2291         dentry = lookup_hash(&name, sb->s_root);
2292         if (IS_ERR(dentry))
2293                 return PTR_ERR(dentry);
2294         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2295         if (err)
2296                 dput(dentry);
2297         /* We keep the dentry reference if everything went ok - we drop it
2298          * on quota_off time */
2299         return err;
2300 }
2301
2302 /* Turn quotas off during mount time */
2303 static int ext3_quota_off_mount(struct super_block *sb, int type)
2304 {
2305         int err;
2306         struct dentry *dentry;
2307
2308         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2309         err = vfs_quota_off_mount(sb, type);
2310         /* We invalidate dentry - it has at least wrong hash... */
2311         d_invalidate(dentry);
2312         dput(dentry);
2313         return err;
2314 }
2315
2316 /*
2317  * Standard function to be called on quota_on
2318  */
2319 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2320                          char *path)
2321 {
2322         int err;
2323         struct nameidata nd;
2324
2325         /* Not journalling quota? */
2326         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2327             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2328                 return vfs_quota_on(sb, type, format_id, path);
2329         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2330         if (err)
2331                 return err;
2332         /* Quotafile not on the same filesystem? */
2333         if (nd.mnt->mnt_sb != sb)
2334                 return -EXDEV;
2335         /* Quotafile not of fs root? */
2336         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2337                 printk(KERN_WARNING
2338                         "EXT3-fs: Quota file not on filesystem root. "
2339                         "Journalled quota will not work.\n");
2340         if (!ext3_should_journal_data(nd.dentry->d_inode))
2341                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2342                         "data-journalling. Journalled quota will not work.\n");
2343         path_release(&nd);
2344         return vfs_quota_on(sb, type, format_id, path);
2345 }
2346
2347 #endif
2348
2349 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2350         int flags, const char *dev_name, void *data)
2351 {
2352         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2353 }
2354
2355 static struct file_system_type ext3_fs_type = {
2356         .owner          = THIS_MODULE,
2357         .name           = "ext3",
2358         .get_sb         = ext3_get_sb,
2359         .kill_sb        = kill_block_super,
2360         .fs_flags       = FS_REQUIRES_DEV,
2361 };
2362
2363 static int __init init_ext3_fs(void)
2364 {
2365         int err = init_ext3_xattr();
2366         if (err)
2367                 return err;
2368         err = init_inodecache();
2369         if (err)
2370                 goto out1;
2371         err = register_filesystem(&ext3_fs_type);
2372         if (err)
2373                 goto out;
2374         return 0;
2375 out:
2376         destroy_inodecache();
2377 out1:
2378         exit_ext3_xattr();
2379         return err;
2380 }
2381
2382 static void __exit exit_ext3_fs(void)
2383 {
2384         unregister_filesystem(&ext3_fs_type);
2385         destroy_inodecache();
2386         exit_ext3_xattr();
2387 }
2388
2389 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2390 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2391 MODULE_LICENSE("GPL");
2392 module_init(init_ext3_fs)
2393 module_exit(exit_ext3_fs)