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 static const char *ext3_decode_error(struct super_block * sb, int errno,
  54                                      char nbuf[16]);
  55 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  56 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  57 static void ext3_unlockfs(struct super_block *sb);
  58 static void ext3_write_super (struct super_block * sb);
  59 static void ext3_write_super_lockfs(struct super_block *sb);
  60
  61 /*
  62  * Wrappers for journal_start/end.
  63  *
  64  * The only special thing we need to do here is to make sure that all
  65  * journal_end calls result in the superblock being marked dirty, so
  66  * that sync() will call the filesystem's write_super callback if
  67  * appropriate.
  68  */
  69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  70 {
  71         journal_t *journal;
  72
  73         if (sb->s_flags & MS_RDONLY)
  74                 return ERR_PTR(-EROFS);
  75
  76         /* Special case here: if the journal has aborted behind our
  77          * backs (eg. EIO in the commit thread), then we still need to
  78          * take the FS itself readonly cleanly. */
  79         journal = EXT3_SB(sb)->s_journal;
  80         if (is_journal_aborted(journal)) {
  81                 ext3_abort(sb, __FUNCTION__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 /*
  90  * The only special thing we need to do here is to make sure that all
  91  * journal_stop calls result in the superblock being marked dirty, so
  92  * that sync() will call the filesystem's write_super callback if
  93  * appropriate.
  94  */
  95 int __ext3_journal_stop(const char *where, handle_t *handle)
  96 {
  97         struct super_block *sb;
  98         int err;
  99         int rc;
 100
 101         sb = handle->h_transaction->t_journal->j_private;
 102         err = handle->h_err;
 103         rc = journal_stop(handle);
 104
 105         if (!err)
 106                 err = rc;
 107         if (err)
 108                 __ext3_std_error(sb, where, err);
 109         return err;
 110 }
 111
 112 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 113                 struct buffer_head *bh, handle_t *handle, int err)
 114 {
 115         char nbuf[16];
 116         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 117
 118         if (bh)
 119                 BUFFER_TRACE(bh, "abort");
 120
 121         if (!handle->h_err)
 122                 handle->h_err = err;
 123
 124         if (is_handle_aborted(handle))
 125                 return;
 126
 127         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 128                caller, errstr, err_fn);
 129
 130         journal_abort_handle(handle);
 131 }
 132
 133 /* Deal with the reporting of failure conditions on a filesystem such as
 134  * inconsistencies detected or read IO failures.
 135  *
 136  * On ext2, we can store the error state of the filesystem in the
 137  * superblock.  That is not possible on ext3, because we may have other
 138  * write ordering constraints on the superblock which prevent us from
 139  * writing it out straight away; and given that the journal is about to
 140  * be aborted, we can't rely on the current, or future, transactions to
 141  * write out the superblock safely.
 142  *
 143  * We'll just use the journal_abort() error code to record an error in
 144  * the journal instead.  On recovery, the journal will compain about
 145  * that error until we've noted it down and cleared it.
 146  */
 147
 148 static void ext3_handle_error(struct super_block *sb)
 149 {
 150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 151
 152         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 153         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 154
 155         if (sb->s_flags & MS_RDONLY)
 156                 return;
 157
 158         if (test_opt (sb, ERRORS_RO)) {
 159                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 160                 sb->s_flags |= MS_RDONLY;
 161         } else {
 162                 journal_t *journal = EXT3_SB(sb)->s_journal;
 163
 164                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 165                 if (journal)
 166                         journal_abort(journal, -EIO);
 167         }
 168         if (test_opt(sb, ERRORS_PANIC))
 169                 panic("EXT3-fs (device %s): panic forced after error\n",
 170                         sb->s_id);
 171         ext3_commit_super(sb, es, 1);
 172 }
 173
 174 void ext3_error (struct super_block * sb, const char * function,
 175                  const char * fmt, ...)
 176 {
 177         va_list args;
 178
 179         va_start(args, fmt);
 180         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 181         vprintk(fmt, args);
 182         printk("\n");
 183         va_end(args);
 184
 185         ext3_handle_error(sb);
 186 }
 187
 188 static const char *ext3_decode_error(struct super_block * sb, int errno,
 189                                      char nbuf[16])
 190 {
 191         char *errstr = NULL;
 192
 193         switch (errno) {
 194         case -EIO:
 195                 errstr = "IO failure";
 196                 break;
 197         case -ENOMEM:
 198                 errstr = "Out of memory";
 199                 break;
 200         case -EROFS:
 201                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 202                         errstr = "Journal has aborted";
 203                 else
 204                         errstr = "Readonly filesystem";
 205                 break;
 206         default:
 207                 /* If the caller passed in an extra buffer for unknown
 208                  * errors, textualise them now.  Else we just return
 209                  * NULL. */
 210                 if (nbuf) {
 211                         /* Check for truncated error codes... */
 212                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 213                                 errstr = nbuf;
 214                 }
 215                 break;
 216         }
 217
 218         return errstr;
 219 }
 220
 221 /* __ext3_std_error decodes expected errors from journaling functions
 222  * automatically and invokes the appropriate error response.  */
 223
 224 void __ext3_std_error (struct super_block * sb, const char * function,
 225                        int errno)
 226 {
 227         char nbuf[16];
 228         const char *errstr;
 229
 230         /* Special case: if the error is EROFS, and we're not already
 231          * inside a transaction, then there's really no point in logging
 232          * an error. */
 233         if (errno == -EROFS && journal_current_handle() == NULL &&
 234             (sb->s_flags & MS_RDONLY))
 235                 return;
 236
 237         errstr = ext3_decode_error(sb, errno, nbuf);
 238         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 239                 sb->s_id, function, errstr);
 240
 241         ext3_handle_error(sb);
 242 }
 243
 244 /*
 245  * ext3_abort is a much stronger failure handler than ext3_error.  The
 246  * abort function may be used to deal with unrecoverable failures such
 247  * as journal IO errors or ENOMEM at a critical moment in log management.
 248  *
 249  * We unconditionally force the filesystem into an ABORT|READONLY state,
 250  * unless the error response on the fs has been set to panic in which
 251  * case we take the easy way out and panic immediately.
 252  */
 253
 254 void ext3_abort (struct super_block * sb, const char * function,
 255                  const char * fmt, ...)
 256 {
 257         va_list args;
 258
 259         printk (KERN_CRIT "ext3_abort called.\n");
 260
 261         va_start(args, fmt);
 262         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 263         vprintk(fmt, args);
 264         printk("\n");
 265         va_end(args);
 266
 267         if (test_opt(sb, ERRORS_PANIC))
 268                 panic("EXT3-fs panic from previous error\n");
 269
 270         if (sb->s_flags & MS_RDONLY)
 271                 return;
 272
 273         printk(KERN_CRIT "Remounting filesystem read-only\n");
 274         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 275         sb->s_flags |= MS_RDONLY;
 276         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 277         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 278 }
 279
 280 void ext3_warning (struct super_block * sb, const char * function,
 281                    const char * fmt, ...)
 282 {
 283         va_list args;
 284
 285         va_start(args, fmt);
 286         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 287                sb->s_id, function);
 288         vprintk(fmt, args);
 289         printk("\n");
 290         va_end(args);
 291 }
 292
 293 void ext3_update_dynamic_rev(struct super_block *sb)
 294 {
 295         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 296
 297         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 298                 return;
 299
 300         ext3_warning(sb, __FUNCTION__,
 301                      "updating to rev %d because of new feature flag, "
 302                      "running e2fsck is recommended",
 303                      EXT3_DYNAMIC_REV);
 304
 305         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 306         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 307         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 308         /* leave es->s_feature_*compat flags alone */
 309         /* es->s_uuid will be set by e2fsck if empty */
 310
 311         /*
 312          * The rest of the superblock fields should be zero, and if not it
 313          * means they are likely already in use, so leave them alone.  We
 314          * can leave it up to e2fsck to clean up any inconsistencies there.
 315          */
 316 }
 317
 318 /*
 319  * Open the external journal device
 320  */
 321 static struct block_device *ext3_blkdev_get(dev_t dev)
 322 {
 323         struct block_device *bdev;
 324         char b[BDEVNAME_SIZE];
 325
 326         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 327         if (IS_ERR(bdev))
 328                 goto fail;
 329         return bdev;
 330
 331 fail:
 332         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 333                         __bdevname(dev, b), PTR_ERR(bdev));
 334         return NULL;
 335 }
 336
 337 /*
 338  * Release the journal device
 339  */
 340 static int ext3_blkdev_put(struct block_device *bdev)
 341 {
 342         bd_release(bdev);
 343         return blkdev_put(bdev);
 344 }
 345
 346 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 347 {
 348         struct block_device *bdev;
 349         int ret = -ENODEV;
 350
 351         bdev = sbi->journal_bdev;
 352         if (bdev) {
 353                 ret = ext3_blkdev_put(bdev);
 354                 sbi->journal_bdev = NULL;
 355         }
 356         return ret;
 357 }
 358
 359 static inline struct inode *orphan_list_entry(struct list_head *l)
 360 {
 361         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 362 }
 363
 364 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 365 {
 366         struct list_head *l;
 367
 368         printk(KERN_ERR "sb orphan head is %d\n",
 369                le32_to_cpu(sbi->s_es->s_last_orphan));
 370
 371         printk(KERN_ERR "sb_info orphan list:\n");
 372         list_for_each(l, &sbi->s_orphan) {
 373                 struct inode *inode = orphan_list_entry(l);
 374                 printk(KERN_ERR "  "
 375                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 376                        inode->i_sb->s_id, inode->i_ino, inode,
 377                        inode->i_mode, inode->i_nlink,
 378                        NEXT_ORPHAN(inode));
 379         }
 380 }
 381
 382 static void ext3_put_super (struct super_block * sb)
 383 {
 384         struct ext3_sb_info *sbi = EXT3_SB(sb);
 385         struct ext3_super_block *es = sbi->s_es;
 386         int i;
 387
 388         ext3_xattr_put_super(sb);
 389         journal_destroy(sbi->s_journal);
 390         if (!(sb->s_flags & MS_RDONLY)) {
 391                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 392                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 393                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 394                 mark_buffer_dirty(sbi->s_sbh);
 395                 ext3_commit_super(sb, es, 1);
 396         }
 397
 398         for (i = 0; i < sbi->s_gdb_count; i++)
 399                 brelse(sbi->s_group_desc[i]);
 400         kfree(sbi->s_group_desc);
 401         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 402         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 403         percpu_counter_destroy(&sbi->s_dirs_counter);
 404         brelse(sbi->s_sbh);
 405 #ifdef CONFIG_QUOTA
 406         for (i = 0; i < MAXQUOTAS; i++)
 407                 kfree(sbi->s_qf_names[i]);
 408 #endif
 409
 410         /* Debugging code just in case the in-memory inode orphan list
 411          * isn't empty.  The on-disk one can be non-empty if we've
 412          * detected an error and taken the fs readonly, but the
 413          * in-memory list had better be clean by this point. */
 414         if (!list_empty(&sbi->s_orphan))
 415                 dump_orphan_list(sb, sbi);
 416         J_ASSERT(list_empty(&sbi->s_orphan));
 417
 418         invalidate_bdev(sb->s_bdev, 0);
 419         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 420                 /*
 421                  * Invalidate the journal device's buffers.  We don't want them
 422                  * floating about in memory - the physical journal device may
 423                  * hotswapped, and it breaks the `ro-after' testing code.
 424                  */
 425                 sync_blockdev(sbi->journal_bdev);
 426                 invalidate_bdev(sbi->journal_bdev, 0);
 427                 ext3_blkdev_remove(sbi);
 428         }
 429         sb->s_fs_info = NULL;
 430         kfree(sbi);
 431         return;
 432 }
 433
 434 static kmem_cache_t *ext3_inode_cachep;
 435
 436 /*
 437  * Called inside transaction, so use GFP_NOFS
 438  */
 439 static struct inode *ext3_alloc_inode(struct super_block *sb)
 440 {
 441         struct ext3_inode_info *ei;
 442
 443         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 444         if (!ei)
 445                 return NULL;
 446 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 447         ei->i_acl = EXT3_ACL_NOT_CACHED;
 448         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 449 #endif
 450         ei->i_block_alloc_info = NULL;
 451         ei->vfs_inode.i_version = 1;
 452         return &ei->vfs_inode;
 453 }
 454
 455 static void ext3_destroy_inode(struct inode *inode)
 456 {
 457         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 458 }
 459
 460 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 461 {
 462         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 463
 464         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 465             SLAB_CTOR_CONSTRUCTOR) {
 466                 INIT_LIST_HEAD(&ei->i_orphan);
 467 #ifdef CONFIG_EXT3_FS_XATTR
 468                 init_rwsem(&ei->xattr_sem);
 469 #endif
 470                 init_MUTEX(&ei->truncate_sem);
 471                 inode_init_once(&ei->vfs_inode);
 472         }
 473 }
 474
 475 static int init_inodecache(void)
 476 {
 477         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 478                                              sizeof(struct ext3_inode_info),
 479                                              0, SLAB_RECLAIM_ACCOUNT,
 480                                              init_once, NULL);
 481         if (ext3_inode_cachep == NULL)
 482                 return -ENOMEM;
 483         return 0;
 484 }
 485
 486 static void destroy_inodecache(void)
 487 {
 488         if (kmem_cache_destroy(ext3_inode_cachep))
 489                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 490 }
 491
 492 static void ext3_clear_inode(struct inode *inode)
 493 {
 494         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 495 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 496        if (EXT3_I(inode)->i_acl &&
 497            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 498                posix_acl_release(EXT3_I(inode)->i_acl);
 499                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 500        }
 501        if (EXT3_I(inode)->i_default_acl &&
 502            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 503                posix_acl_release(EXT3_I(inode)->i_default_acl);
 504                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 505        }
 506 #endif
 507         ext3_discard_reservation(inode);
 508         EXT3_I(inode)->i_block_alloc_info = NULL;
 509         kfree(rsv);
 510 }
 511
 512 #ifdef CONFIG_QUOTA
 513
 514 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 515 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 516
 517 static int ext3_dquot_initialize(struct inode *inode, int type);
 518 static int ext3_dquot_drop(struct inode *inode);
 519 static int ext3_write_dquot(struct dquot *dquot);
 520 static int ext3_acquire_dquot(struct dquot *dquot);
 521 static int ext3_release_dquot(struct dquot *dquot);
 522 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 523 static int ext3_write_info(struct super_block *sb, int type);
 524 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 525 static int ext3_quota_on_mount(struct super_block *sb, int type);
 526 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 527                                size_t len, loff_t off);
 528 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 529                                 const char *data, size_t len, loff_t off);
 530
 531 static struct dquot_operations ext3_quota_operations = {
 532         .initialize     = ext3_dquot_initialize,
 533         .drop           = ext3_dquot_drop,
 534         .alloc_space    = dquot_alloc_space,
 535         .alloc_inode    = dquot_alloc_inode,
 536         .free_space     = dquot_free_space,
 537         .free_inode     = dquot_free_inode,
 538         .transfer       = dquot_transfer,
 539         .write_dquot    = ext3_write_dquot,
 540         .acquire_dquot  = ext3_acquire_dquot,
 541         .release_dquot  = ext3_release_dquot,
 542         .mark_dirty     = ext3_mark_dquot_dirty,
 543         .write_info     = ext3_write_info
 544 };
 545
 546 static struct quotactl_ops ext3_qctl_operations = {
 547         .quota_on       = ext3_quota_on,
 548         .quota_off      = vfs_quota_off,
 549         .quota_sync     = vfs_quota_sync,
 550         .get_info       = vfs_get_dqinfo,
 551         .set_info       = vfs_set_dqinfo,
 552         .get_dqblk      = vfs_get_dqblk,
 553         .set_dqblk      = vfs_set_dqblk
 554 };
 555 #endif
 556
 557 static struct super_operations ext3_sops = {
 558         .alloc_inode    = ext3_alloc_inode,
 559         .destroy_inode  = ext3_destroy_inode,
 560         .read_inode     = ext3_read_inode,
 561         .write_inode    = ext3_write_inode,
 562         .dirty_inode    = ext3_dirty_inode,
 563         .delete_inode   = ext3_delete_inode,
 564         .put_super      = ext3_put_super,
 565         .write_super    = ext3_write_super,
 566         .sync_fs        = ext3_sync_fs,
 567         .write_super_lockfs = ext3_write_super_lockfs,
 568         .unlockfs       = ext3_unlockfs,
 569         .statfs         = ext3_statfs,
 570         .remount_fs     = ext3_remount,
 571         .clear_inode    = ext3_clear_inode,
 572 #ifdef CONFIG_QUOTA
 573         .quota_read     = ext3_quota_read,
 574         .quota_write    = ext3_quota_write,
 575 #endif
 576 };
 577
 578 struct dentry *ext3_get_parent(struct dentry *child);
 579 static struct export_operations ext3_export_ops = {
 580         .get_parent = ext3_get_parent,
 581 };
 582
 583 enum {
 584         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 585         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 586         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 587         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 588         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 589         Opt_commit, Opt_journal_update, Opt_journal_inum,
 590         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 591         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 592         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 593         Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
 594 };
 595
 596 static match_table_t tokens = {
 597         {Opt_bsd_df, "bsddf"},
 598         {Opt_minix_df, "minixdf"},
 599         {Opt_grpid, "grpid"},
 600         {Opt_grpid, "bsdgroups"},
 601         {Opt_nogrpid, "nogrpid"},
 602         {Opt_nogrpid, "sysvgroups"},
 603         {Opt_resgid, "resgid=%u"},
 604         {Opt_resuid, "resuid=%u"},
 605         {Opt_sb, "sb=%u"},
 606         {Opt_err_cont, "errors=continue"},
 607         {Opt_err_panic, "errors=panic"},
 608         {Opt_err_ro, "errors=remount-ro"},
 609         {Opt_nouid32, "nouid32"},
 610         {Opt_nocheck, "nocheck"},
 611         {Opt_nocheck, "check=none"},
 612         {Opt_check, "check"},
 613         {Opt_debug, "debug"},
 614         {Opt_oldalloc, "oldalloc"},
 615         {Opt_orlov, "orlov"},
 616         {Opt_user_xattr, "user_xattr"},
 617         {Opt_nouser_xattr, "nouser_xattr"},
 618         {Opt_acl, "acl"},
 619         {Opt_noacl, "noacl"},
 620         {Opt_reservation, "reservation"},
 621         {Opt_noreservation, "noreservation"},
 622         {Opt_noload, "noload"},
 623         {Opt_nobh, "nobh"},
 624         {Opt_commit, "commit=%u"},
 625         {Opt_journal_update, "journal=update"},
 626         {Opt_journal_inum, "journal=%u"},
 627         {Opt_abort, "abort"},
 628         {Opt_data_journal, "data=journal"},
 629         {Opt_data_ordered, "data=ordered"},
 630         {Opt_data_writeback, "data=writeback"},
 631         {Opt_offusrjquota, "usrjquota="},
 632         {Opt_usrjquota, "usrjquota=%s"},
 633         {Opt_offgrpjquota, "grpjquota="},
 634         {Opt_grpjquota, "grpjquota=%s"},
 635         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 636         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 637         {Opt_quota, "grpquota"},
 638         {Opt_noquota, "noquota"},
 639         {Opt_quota, "quota"},
 640         {Opt_quota, "usrquota"},
 641         {Opt_barrier, "barrier=%u"},
 642         {Opt_err, NULL},
 643         {Opt_resize, "resize"},
 644 };
 645
 646 static unsigned long get_sb_block(void **data)
 647 {
 648         unsigned long   sb_block;
 649         char            *options = (char *) *data;
 650
 651         if (!options || strncmp(options, "sb=", 3) != 0)
 652                 return 1;       /* Default location */
 653         options += 3;
 654         sb_block = simple_strtoul(options, &options, 0);
 655         if (*options && *options != ',') {
 656                 printk("EXT3-fs: Invalid sb specification: %s\n",
 657                        (char *) *data);
 658                 return 1;
 659         }
 660         if (*options == ',')
 661                 options++;
 662         *data = (void *) options;
 663         return sb_block;
 664 }
 665
 666 static int parse_options (char * options, struct super_block *sb,
 667                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 668 {
 669         struct ext3_sb_info *sbi = EXT3_SB(sb);
 670         char * p;
 671         substring_t args[MAX_OPT_ARGS];
 672         int data_opt = 0;
 673         int option;
 674 #ifdef CONFIG_QUOTA
 675         int qtype;
 676         char *qname;
 677 #endif
 678
 679         if (!options)
 680                 return 1;
 681
 682         while ((p = strsep (&options, ",")) != NULL) {
 683                 int token;
 684                 if (!*p)
 685                         continue;
 686
 687                 token = match_token(p, tokens, args);
 688                 switch (token) {
 689                 case Opt_bsd_df:
 690                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 691                         break;
 692                 case Opt_minix_df:
 693                         set_opt (sbi->s_mount_opt, MINIX_DF);
 694                         break;
 695                 case Opt_grpid:
 696                         set_opt (sbi->s_mount_opt, GRPID);
 697                         break;
 698                 case Opt_nogrpid:
 699                         clear_opt (sbi->s_mount_opt, GRPID);
 700                         break;
 701                 case Opt_resuid:
 702                         if (match_int(&args[0], &option))
 703                                 return 0;
 704                         sbi->s_resuid = option;
 705                         break;
 706                 case Opt_resgid:
 707                         if (match_int(&args[0], &option))
 708                                 return 0;
 709                         sbi->s_resgid = option;
 710                         break;
 711                 case Opt_sb:
 712                         /* handled by get_sb_block() instead of here */
 713                         /* *sb_block = match_int(&args[0]); */
 714                         break;
 715                 case Opt_err_panic:
 716                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 717                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 718                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 719                         break;
 720                 case Opt_err_ro:
 721                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 722                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 723                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 724                         break;
 725                 case Opt_err_cont:
 726                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 727                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 728                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 729                         break;
 730                 case Opt_nouid32:
 731                         set_opt (sbi->s_mount_opt, NO_UID32);
 732                         break;
 733                 case Opt_check:
 734 #ifdef CONFIG_EXT3_CHECK
 735                         set_opt (sbi->s_mount_opt, CHECK);
 736 #else
 737                         printk(KERN_ERR
 738                                "EXT3 Check option not supported\n");
 739 #endif
 740                         break;
 741                 case Opt_nocheck:
 742                         clear_opt (sbi->s_mount_opt, CHECK);
 743                         break;
 744                 case Opt_debug:
 745                         set_opt (sbi->s_mount_opt, DEBUG);
 746                         break;
 747                 case Opt_oldalloc:
 748                         set_opt (sbi->s_mount_opt, OLDALLOC);
 749                         break;
 750                 case Opt_orlov:
 751                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 752                         break;
 753 #ifdef CONFIG_EXT3_FS_XATTR
 754                 case Opt_user_xattr:
 755                         set_opt (sbi->s_mount_opt, XATTR_USER);
 756                         break;
 757                 case Opt_nouser_xattr:
 758                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 759                         break;
 760 #else
 761                 case Opt_user_xattr:
 762                 case Opt_nouser_xattr:
 763                         printk("EXT3 (no)user_xattr options not supported\n");
 764                         break;
 765 #endif
 766 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 767                 case Opt_acl:
 768                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 769                         break;
 770                 case Opt_noacl:
 771                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 772                         break;
 773 #else
 774                 case Opt_acl:
 775                 case Opt_noacl:
 776                         printk("EXT3 (no)acl options not supported\n");
 777                         break;
 778 #endif
 779                 case Opt_reservation:
 780                         set_opt(sbi->s_mount_opt, RESERVATION);
 781                         break;
 782                 case Opt_noreservation:
 783                         clear_opt(sbi->s_mount_opt, RESERVATION);
 784                         break;
 785                 case Opt_journal_update:
 786                         /* @@@ FIXME */
 787                         /* Eventually we will want to be able to create
 788                            a journal file here.  For now, only allow the
 789                            user to specify an existing inode to be the
 790                            journal file. */
 791                         if (is_remount) {
 792                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 793                                        "journal on remount\n");
 794                                 return 0;
 795                         }
 796                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 797                         break;
 798                 case Opt_journal_inum:
 799                         if (is_remount) {
 800                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 801                                        "journal on remount\n");
 802                                 return 0;
 803                         }
 804                         if (match_int(&args[0], &option))
 805                                 return 0;
 806                         *inum = option;
 807                         break;
 808                 case Opt_noload:
 809                         set_opt (sbi->s_mount_opt, NOLOAD);
 810                         break;
 811                 case Opt_commit:
 812                         if (match_int(&args[0], &option))
 813                                 return 0;
 814                         if (option < 0)
 815                                 return 0;
 816                         if (option == 0)
 817                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 818                         sbi->s_commit_interval = HZ * option;
 819                         break;
 820                 case Opt_data_journal:
 821                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 822                         goto datacheck;
 823                 case Opt_data_ordered:
 824                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 825                         goto datacheck;
 826                 case Opt_data_writeback:
 827                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 828                 datacheck:
 829                         if (is_remount) {
 830                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 831                                                 != data_opt) {
 832                                         printk(KERN_ERR
 833                                                 "EXT3-fs: cannot change data "
 834                                                 "mode on remount\n");
 835                                         return 0;
 836                                 }
 837                         } else {
 838                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 839                                 sbi->s_mount_opt |= data_opt;
 840                         }
 841                         break;
 842 #ifdef CONFIG_QUOTA
 843                 case Opt_usrjquota:
 844                         qtype = USRQUOTA;
 845                         goto set_qf_name;
 846                 case Opt_grpjquota:
 847                         qtype = GRPQUOTA;
 848 set_qf_name:
 849                         if (sb_any_quota_enabled(sb)) {
 850                                 printk(KERN_ERR
 851                                         "EXT3-fs: Cannot change journalled "
 852                                         "quota options when quota turned on.\n");
 853                                 return 0;
 854                         }
 855                         qname = match_strdup(&args[0]);
 856                         if (!qname) {
 857                                 printk(KERN_ERR
 858                                         "EXT3-fs: not enough memory for "
 859                                         "storing quotafile name.\n");
 860                                 return 0;
 861                         }
 862                         if (sbi->s_qf_names[qtype] &&
 863                             strcmp(sbi->s_qf_names[qtype], qname)) {
 864                                 printk(KERN_ERR
 865                                         "EXT3-fs: %s quota file already "
 866                                         "specified.\n", QTYPE2NAME(qtype));
 867                                 kfree(qname);
 868                                 return 0;
 869                         }
 870                         sbi->s_qf_names[qtype] = qname;
 871                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 872                                 printk(KERN_ERR
 873                                         "EXT3-fs: quotafile must be on "
 874                                         "filesystem root.\n");
 875                                 kfree(sbi->s_qf_names[qtype]);
 876                                 sbi->s_qf_names[qtype] = NULL;
 877                                 return 0;
 878                         }
 879                         set_opt(sbi->s_mount_opt, QUOTA);
 880                         break;
 881                 case Opt_offusrjquota:
 882                         qtype = USRQUOTA;
 883                         goto clear_qf_name;
 884                 case Opt_offgrpjquota:
 885                         qtype = GRPQUOTA;
 886 clear_qf_name:
 887                         if (sb_any_quota_enabled(sb)) {
 888                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 889                                         "journalled quota options when "
 890                                         "quota turned on.\n");
 891                                 return 0;
 892                         }
 893                         kfree(sbi->s_qf_names[qtype]);
 894                         sbi->s_qf_names[qtype] = NULL;
 895                         break;
 896                 case Opt_jqfmt_vfsold:
 897                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 898                         break;
 899                 case Opt_jqfmt_vfsv0:
 900                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 901                         break;
 902                 case Opt_quota:
 903                         set_opt(sbi->s_mount_opt, QUOTA);
 904                         break;
 905                 case Opt_noquota:
 906                         if (sb_any_quota_enabled(sb)) {
 907                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
 908                                         "options when quota turned on.\n");
 909                                 return 0;
 910                         }
 911                         clear_opt(sbi->s_mount_opt, QUOTA);
 912                         break;
 913 #else
 914                 case Opt_usrjquota:
 915                 case Opt_grpjquota:
 916                 case Opt_offusrjquota:
 917                 case Opt_offgrpjquota:
 918                 case Opt_jqfmt_vfsold:
 919                 case Opt_jqfmt_vfsv0:
 920                         printk(KERN_ERR
 921                                 "EXT3-fs: journalled quota options not "
 922                                 "supported.\n");
 923                         break;
 924                 case Opt_quota:
 925                 case Opt_noquota:
 926                         break;
 927 #endif
 928                 case Opt_abort:
 929                         set_opt(sbi->s_mount_opt, ABORT);
 930                         break;
 931                 case Opt_barrier:
 932                         if (match_int(&args[0], &option))
 933                                 return 0;
 934                         if (option)
 935                                 set_opt(sbi->s_mount_opt, BARRIER);
 936                         else
 937                                 clear_opt(sbi->s_mount_opt, BARRIER);
 938                         break;
 939                 case Opt_ignore:
 940                         break;
 941                 case Opt_resize:
 942                         if (!n_blocks_count) {
 943                                 printk("EXT3-fs: resize option only available "
 944                                         "for remount\n");
 945                                 return 0;
 946                         }
 947                         if (match_int(&args[0], &option) != 0)
 948                                 return 0;
 949                         *n_blocks_count = option;
 950                         break;
 951                 case Opt_nobh:
 952                         set_opt(sbi->s_mount_opt, NOBH);
 953                         break;
 954                 default:
 955                         printk (KERN_ERR
 956                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 957                                 "or missing value\n", p);
 958                         return 0;
 959                 }
 960         }
 961 #ifdef CONFIG_QUOTA
 962         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 963             sbi->s_qf_names[GRPQUOTA])) {
 964                 printk(KERN_ERR
 965                         "EXT3-fs: journalled quota format not specified.\n");
 966                 return 0;
 967         }
 968 #endif
 969
 970         return 1;
 971 }
 972
 973 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 974                             int read_only)
 975 {
 976         struct ext3_sb_info *sbi = EXT3_SB(sb);
 977         int res = 0;
 978
 979         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 980                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 981                         "forcing read-only mode\n");
 982                 res = MS_RDONLY;
 983         }
 984         if (read_only)
 985                 return res;
 986         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 987                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 988                         "running e2fsck is recommended\n");
 989         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 990                 printk (KERN_WARNING
 991                         "EXT3-fs warning: mounting fs with errors, "
 992                         "running e2fsck is recommended\n");
 993         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 994                  le16_to_cpu(es->s_mnt_count) >=
 995                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 996                 printk (KERN_WARNING
 997                         "EXT3-fs warning: maximal mount count reached, "
 998                         "running e2fsck is recommended\n");
 999         else if (le32_to_cpu(es->s_checkinterval) &&
1000                 (le32_to_cpu(es->s_lastcheck) +
1001                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1002                 printk (KERN_WARNING
1003                         "EXT3-fs warning: checktime reached, "
1004                         "running e2fsck is recommended\n");
1005 #if 0
1006                 /* @@@ We _will_ want to clear the valid bit if we find
1007                    inconsistencies, to force a fsck at reboot.  But for
1008                    a plain journaled filesystem we can keep it set as
1009                    valid forever! :) */
1010         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1011 #endif
1012         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1013                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1014         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1015         es->s_mtime = cpu_to_le32(get_seconds());
1016         ext3_update_dynamic_rev(sb);
1017         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1018
1019         ext3_commit_super(sb, es, 1);
1020         if (test_opt(sb, DEBUG))
1021                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1022                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1023                         sb->s_blocksize,
1024                         sbi->s_groups_count,
1025                         EXT3_BLOCKS_PER_GROUP(sb),
1026                         EXT3_INODES_PER_GROUP(sb),
1027                         sbi->s_mount_opt);
1028
1029         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1030         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1031                 char b[BDEVNAME_SIZE];
1032
1033                 printk("external journal on %s\n",
1034                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1035         } else {
1036                 printk("internal journal\n");
1037         }
1038 #ifdef CONFIG_EXT3_CHECK
1039         if (test_opt (sb, CHECK)) {
1040                 ext3_check_blocks_bitmap (sb);
1041                 ext3_check_inodes_bitmap (sb);
1042         }
1043 #endif
1044         return res;
1045 }
1046
1047 /* Called at mount-time, super-block is locked */
1048 static int ext3_check_descriptors (struct super_block * sb)
1049 {
1050         struct ext3_sb_info *sbi = EXT3_SB(sb);
1051         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1052         struct ext3_group_desc * gdp = NULL;
1053         int desc_block = 0;
1054         int i;
1055
1056         ext3_debug ("Checking group descriptors");
1057
1058         for (i = 0; i < sbi->s_groups_count; i++)
1059         {
1060                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1061                         gdp = (struct ext3_group_desc *)
1062                                         sbi->s_group_desc[desc_block++]->b_data;
1063                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1064                     le32_to_cpu(gdp->bg_block_bitmap) >=
1065                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1066                 {
1067                         ext3_error (sb, "ext3_check_descriptors",
1068                                     "Block bitmap for group %d"
1069                                     " not in group (block %lu)!",
1070                                     i, (unsigned long)
1071                                         le32_to_cpu(gdp->bg_block_bitmap));
1072                         return 0;
1073                 }
1074                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1075                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1076                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1077                 {
1078                         ext3_error (sb, "ext3_check_descriptors",
1079                                     "Inode bitmap for group %d"
1080                                     " not in group (block %lu)!",
1081                                     i, (unsigned long)
1082                                         le32_to_cpu(gdp->bg_inode_bitmap));
1083                         return 0;
1084                 }
1085                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1086                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1087                     block + EXT3_BLOCKS_PER_GROUP(sb))
1088                 {
1089                         ext3_error (sb, "ext3_check_descriptors",
1090                                     "Inode table for group %d"
1091                                     " not in group (block %lu)!",
1092                                     i, (unsigned long)
1093                                         le32_to_cpu(gdp->bg_inode_table));
1094                         return 0;
1095                 }
1096                 block += EXT3_BLOCKS_PER_GROUP(sb);
1097                 gdp++;
1098         }
1099
1100         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1101         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1102         return 1;
1103 }
1104
1105
1106 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1107  * the superblock) which were deleted from all directories, but held open by
1108  * a process at the time of a crash.  We walk the list and try to delete these
1109  * inodes at recovery time (only with a read-write filesystem).
1110  *
1111  * In order to keep the orphan inode chain consistent during traversal (in
1112  * case of crash during recovery), we link each inode into the superblock
1113  * orphan list_head and handle it the same way as an inode deletion during
1114  * normal operation (which journals the operations for us).
1115  *
1116  * We only do an iget() and an iput() on each inode, which is very safe if we
1117  * accidentally point at an in-use or already deleted inode.  The worst that
1118  * can happen in this case is that we get a "bit already cleared" message from
1119  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1120  * e2fsck was run on this filesystem, and it must have already done the orphan
1121  * inode cleanup for us, so we can safely abort without any further action.
1122  */
1123 static void ext3_orphan_cleanup (struct super_block * sb,
1124                                  struct ext3_super_block * es)
1125 {
1126         unsigned int s_flags = sb->s_flags;
1127         int nr_orphans = 0, nr_truncates = 0;
1128 #ifdef CONFIG_QUOTA
1129         int i;
1130 #endif
1131         if (!es->s_last_orphan) {
1132                 jbd_debug(4, "no orphan inodes to clean up\n");
1133                 return;
1134         }
1135
1136         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1137                 if (es->s_last_orphan)
1138                         jbd_debug(1, "Errors on filesystem, "
1139                                   "clearing orphan list.\n");
1140                 es->s_last_orphan = 0;
1141                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1142                 return;
1143         }
1144
1145         if (s_flags & MS_RDONLY) {
1146                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1147                        sb->s_id);
1148                 sb->s_flags &= ~MS_RDONLY;
1149         }
1150 #ifdef CONFIG_QUOTA
1151         /* Needed for iput() to work correctly and not trash data */
1152         sb->s_flags |= MS_ACTIVE;
1153         /* Turn on quotas so that they are updated correctly */
1154         for (i = 0; i < MAXQUOTAS; i++) {
1155                 if (EXT3_SB(sb)->s_qf_names[i]) {
1156                         int ret = ext3_quota_on_mount(sb, i);
1157                         if (ret < 0)
1158                                 printk(KERN_ERR
1159                                         "EXT3-fs: Cannot turn on journalled "
1160                                         "quota: error %d\n", ret);
1161                 }
1162         }
1163 #endif
1164
1165         while (es->s_last_orphan) {
1166                 struct inode *inode;
1167
1168                 if (!(inode =
1169                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1170                         es->s_last_orphan = 0;
1171                         break;
1172                 }
1173
1174                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1175                 DQUOT_INIT(inode);
1176                 if (inode->i_nlink) {
1177                         printk(KERN_DEBUG
1178                                 "%s: truncating inode %ld to %Ld bytes\n",
1179                                 __FUNCTION__, inode->i_ino, inode->i_size);
1180                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1181                                   inode->i_ino, inode->i_size);
1182                         ext3_truncate(inode);
1183                         nr_truncates++;
1184                 } else {
1185                         printk(KERN_DEBUG
1186                                 "%s: deleting unreferenced inode %ld\n",
1187                                 __FUNCTION__, inode->i_ino);
1188                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1189                                   inode->i_ino);
1190                         nr_orphans++;
1191                 }
1192                 iput(inode);  /* The delete magic happens here! */
1193         }
1194
1195 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1196
1197         if (nr_orphans)
1198                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1199                        sb->s_id, PLURAL(nr_orphans));
1200         if (nr_truncates)
1201                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1202                        sb->s_id, PLURAL(nr_truncates));
1203 #ifdef CONFIG_QUOTA
1204         /* Turn quotas off */
1205         for (i = 0; i < MAXQUOTAS; i++) {
1206                 if (sb_dqopt(sb)->files[i])
1207                         vfs_quota_off(sb, i);
1208         }
1209 #endif
1210         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1211 }
1212
1213 #define log2(n) ffz(~(n))
1214
1215 /*
1216  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1217  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1218  * We need to be 1 filesystem block less than the 2^32 sector limit.
1219  */
1220 static loff_t ext3_max_size(int bits)
1221 {
1222         loff_t res = EXT3_NDIR_BLOCKS;
1223         /* This constant is calculated to be the largest file size for a
1224          * dense, 4k-blocksize file such that the total number of
1225          * sectors in the file, including data and all indirect blocks,
1226          * does not exceed 2^32. */
1227         const loff_t upper_limit = 0x1ff7fffd000LL;
1228
1229         res += 1LL << (bits-2);
1230         res += 1LL << (2*(bits-2));
1231         res += 1LL << (3*(bits-2));
1232         res <<= bits;
1233         if (res > upper_limit)
1234                 res = upper_limit;
1235         return res;
1236 }
1237
1238 static unsigned long descriptor_loc(struct super_block *sb,
1239                                     unsigned long logic_sb_block,
1240                                     int nr)
1241 {
1242         struct ext3_sb_info *sbi = EXT3_SB(sb);
1243         unsigned long bg, first_data_block, first_meta_bg;
1244         int has_super = 0;
1245
1246         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1247         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1248
1249         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1250             nr < first_meta_bg)
1251                 return (logic_sb_block + nr + 1);
1252         bg = sbi->s_desc_per_block * nr;
1253         if (ext3_bg_has_super(sb, bg))
1254                 has_super = 1;
1255         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1256 }
1257
1258
1259 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1260 {
1261         struct buffer_head * bh;
1262         struct ext3_super_block *es = NULL;
1263         struct ext3_sb_info *sbi;
1264         unsigned long block;
1265         unsigned long sb_block = get_sb_block(&data);
1266         unsigned long logic_sb_block;
1267         unsigned long offset = 0;
1268         unsigned long journal_inum = 0;
1269         unsigned long def_mount_opts;
1270         struct inode *root;
1271         int blocksize;
1272         int hblock;
1273         int db_count;
1274         int i;
1275         int needs_recovery;
1276         __le32 features;
1277
1278         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1279         if (!sbi)
1280                 return -ENOMEM;
1281         sb->s_fs_info = sbi;
1282         memset(sbi, 0, sizeof(*sbi));
1283         sbi->s_mount_opt = 0;
1284         sbi->s_resuid = EXT3_DEF_RESUID;
1285         sbi->s_resgid = EXT3_DEF_RESGID;
1286
1287         unlock_kernel();
1288
1289         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1290         if (!blocksize) {
1291                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1292                 goto out_fail;
1293         }
1294
1295         /*
1296          * The ext3 superblock will not be buffer aligned for other than 1kB
1297          * block sizes.  We need to calculate the offset from buffer start.
1298          */
1299         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1300                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1301                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1302         } else {
1303                 logic_sb_block = sb_block;
1304         }
1305
1306         if (!(bh = sb_bread(sb, logic_sb_block))) {
1307                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1308                 goto out_fail;
1309         }
1310         /*
1311          * Note: s_es must be initialized as soon as possible because
1312          *       some ext3 macro-instructions depend on its value
1313          */
1314         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1315         sbi->s_es = es;
1316         sb->s_magic = le16_to_cpu(es->s_magic);
1317         if (sb->s_magic != EXT3_SUPER_MAGIC)
1318                 goto cantfind_ext3;
1319
1320         /* Set defaults before we parse the mount options */
1321         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1322         if (def_mount_opts & EXT3_DEFM_DEBUG)
1323                 set_opt(sbi->s_mount_opt, DEBUG);
1324         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1325                 set_opt(sbi->s_mount_opt, GRPID);
1326         if (def_mount_opts & EXT3_DEFM_UID16)
1327                 set_opt(sbi->s_mount_opt, NO_UID32);
1328         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1329                 set_opt(sbi->s_mount_opt, XATTR_USER);
1330         if (def_mount_opts & EXT3_DEFM_ACL)
1331                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1332         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1333                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1334         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1335                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1336         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1337                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1338
1339         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1340                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1341         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1342                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1343
1344         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1345         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1346
1347         set_opt(sbi->s_mount_opt, RESERVATION);
1348
1349         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1350                 goto failed_mount;
1351
1352         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1353                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1354
1355         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1356             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1357              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1358              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1359                 printk(KERN_WARNING
1360                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1361                        "running e2fsck is recommended\n");
1362         /*
1363          * Check feature flags regardless of the revision level, since we
1364          * previously didn't change the revision level when setting the flags,
1365          * so there is a chance incompat flags are set on a rev 0 filesystem.
1366          */
1367         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1368         if (features) {
1369                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1370                        "unsupported optional features (%x).\n",
1371                        sb->s_id, le32_to_cpu(features));
1372                 goto failed_mount;
1373         }
1374         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1375         if (!(sb->s_flags & MS_RDONLY) && features) {
1376                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1377                        "unsupported optional features (%x).\n",
1378                        sb->s_id, le32_to_cpu(features));
1379                 goto failed_mount;
1380         }
1381         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1382
1383         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1384             blocksize > EXT3_MAX_BLOCK_SIZE) {
1385                 printk(KERN_ERR
1386                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1387                        blocksize, sb->s_id);
1388                 goto failed_mount;
1389         }
1390
1391         hblock = bdev_hardsect_size(sb->s_bdev);
1392         if (sb->s_blocksize != blocksize) {
1393                 /*
1394                  * Make sure the blocksize for the filesystem is larger
1395                  * than the hardware sectorsize for the machine.
1396                  */
1397                 if (blocksize < hblock) {
1398                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1399                                "device blocksize %d.\n", blocksize, hblock);
1400                         goto failed_mount;
1401                 }
1402
1403                 brelse (bh);
1404                 sb_set_blocksize(sb, blocksize);
1405                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1406                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1407                 bh = sb_bread(sb, logic_sb_block);
1408                 if (!bh) {
1409                         printk(KERN_ERR
1410                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1411                         goto failed_mount;
1412                 }
1413                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1414                 sbi->s_es = es;
1415                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1416                         printk (KERN_ERR
1417                                 "EXT3-fs: Magic mismatch, very weird !\n");
1418                         goto failed_mount;
1419                 }
1420         }
1421
1422         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1423
1424         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1425                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1426                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1427         } else {
1428                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1429                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1430                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1431                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1432                     (sbi->s_inode_size > blocksize)) {
1433                         printk (KERN_ERR
1434                                 "EXT3-fs: unsupported inode size: %d\n",
1435                                 sbi->s_inode_size);
1436                         goto failed_mount;
1437                 }
1438         }
1439         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1440                                    le32_to_cpu(es->s_log_frag_size);
1441         if (blocksize != sbi->s_frag_size) {
1442                 printk(KERN_ERR
1443                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1444                        sbi->s_frag_size, blocksize);
1445                 goto failed_mount;
1446         }
1447         sbi->s_frags_per_block = 1;
1448         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1449         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1450         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1451         if (EXT3_INODE_SIZE(sb) == 0)
1452                 goto cantfind_ext3;
1453         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1454         if (sbi->s_inodes_per_block == 0)
1455                 goto cantfind_ext3;
1456         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1457                                         sbi->s_inodes_per_block;
1458         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1459         sbi->s_sbh = bh;
1460         sbi->s_mount_state = le16_to_cpu(es->s_state);
1461         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1462         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1463         for (i=0; i < 4; i++)
1464                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1465         sbi->s_def_hash_version = es->s_def_hash_version;
1466
1467         if (sbi->s_blocks_per_group > blocksize * 8) {
1468                 printk (KERN_ERR
1469                         "EXT3-fs: #blocks per group too big: %lu\n",
1470                         sbi->s_blocks_per_group);
1471                 goto failed_mount;
1472         }
1473         if (sbi->s_frags_per_group > blocksize * 8) {
1474                 printk (KERN_ERR
1475                         "EXT3-fs: #fragments per group too big: %lu\n",
1476                         sbi->s_frags_per_group);
1477                 goto failed_mount;
1478         }
1479         if (sbi->s_inodes_per_group > blocksize * 8) {
1480                 printk (KERN_ERR
1481                         "EXT3-fs: #inodes per group too big: %lu\n",
1482                         sbi->s_inodes_per_group);
1483                 goto failed_mount;
1484         }
1485
1486         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1487                 goto cantfind_ext3;
1488         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1489                                le32_to_cpu(es->s_first_data_block) +
1490                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1491                               EXT3_BLOCKS_PER_GROUP(sb);
1492         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1493                    EXT3_DESC_PER_BLOCK(sb);
1494         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1495                                     GFP_KERNEL);
1496         if (sbi->s_group_desc == NULL) {
1497                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1498                 goto failed_mount;
1499         }
1500
1501         percpu_counter_init(&sbi->s_freeblocks_counter);
1502         percpu_counter_init(&sbi->s_freeinodes_counter);
1503         percpu_counter_init(&sbi->s_dirs_counter);
1504         bgl_lock_init(&sbi->s_blockgroup_lock);
1505
1506         for (i = 0; i < db_count; i++) {
1507                 block = descriptor_loc(sb, logic_sb_block, i);
1508                 sbi->s_group_desc[i] = sb_bread(sb, block);
1509                 if (!sbi->s_group_desc[i]) {
1510                         printk (KERN_ERR "EXT3-fs: "
1511                                 "can't read group descriptor %d\n", i);
1512                         db_count = i;
1513                         goto failed_mount2;
1514                 }
1515         }
1516         if (!ext3_check_descriptors (sb)) {
1517                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1518                 goto failed_mount2;
1519         }
1520         sbi->s_gdb_count = db_count;
1521         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1522         spin_lock_init(&sbi->s_next_gen_lock);
1523         /* per fileystem reservation list head & lock */
1524         spin_lock_init(&sbi->s_rsv_window_lock);
1525         sbi->s_rsv_window_root = RB_ROOT;
1526         /* Add a single, static dummy reservation to the start of the
1527          * reservation window list --- it gives us a placeholder for
1528          * append-at-start-of-list which makes the allocation logic
1529          * _much_ simpler. */
1530         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1531         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1532         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1533         sbi->s_rsv_window_head.rsv_goal_size = 0;
1534         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1535
1536         /*
1537          * set up enough so that it can read an inode
1538          */
1539         sb->s_op = &ext3_sops;
1540         sb->s_export_op = &ext3_export_ops;
1541         sb->s_xattr = ext3_xattr_handlers;
1542 #ifdef CONFIG_QUOTA
1543         sb->s_qcop = &ext3_qctl_operations;
1544         sb->dq_op = &ext3_quota_operations;
1545 #endif
1546         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1547
1548         sb->s_root = NULL;
1549
1550         needs_recovery = (es->s_last_orphan != 0 ||
1551                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1552                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1553
1554         /*
1555          * The first inode we look at is the journal inode.  Don't try
1556          * root first: it may be modified in the journal!
1557          */
1558         if (!test_opt(sb, NOLOAD) &&
1559             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1560                 if (ext3_load_journal(sb, es))
1561                         goto failed_mount2;
1562         } else if (journal_inum) {
1563                 if (ext3_create_journal(sb, es, journal_inum))
1564                         goto failed_mount2;
1565         } else {
1566                 if (!silent)
1567                         printk (KERN_ERR
1568                                 "ext3: No journal on filesystem on %s\n",
1569                                 sb->s_id);
1570                 goto failed_mount2;
1571         }
1572
1573         /* We have now updated the journal if required, so we can
1574          * validate the data journaling mode. */
1575         switch (test_opt(sb, DATA_FLAGS)) {
1576         case 0:
1577                 /* No mode set, assume a default based on the journal
1578                    capabilities: ORDERED_DATA if the journal can
1579                    cope, else JOURNAL_DATA */
1580                 if (journal_check_available_features
1581                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1582                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1583                 else
1584                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1585                 break;
1586
1587         case EXT3_MOUNT_ORDERED_DATA:
1588         case EXT3_MOUNT_WRITEBACK_DATA:
1589                 if (!journal_check_available_features
1590                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1591                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1592                                "requested data journaling mode\n");
1593                         goto failed_mount3;
1594                 }
1595         default:
1596                 break;
1597         }
1598
1599         if (test_opt(sb, NOBH)) {
1600                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1601                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1602                                 "since filesystem blocksize doesn't match "
1603                                 "pagesize\n");
1604                         clear_opt(sbi->s_mount_opt, NOBH);
1605                 }
1606                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1607                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1608                                 "its supported only with writeback mode\n");
1609                         clear_opt(sbi->s_mount_opt, NOBH);
1610                 }
1611         }
1612         /*
1613          * The journal_load will have done any necessary log recovery,
1614          * so we can safely mount the rest of the filesystem now.
1615          */
1616
1617         root = iget(sb, EXT3_ROOT_INO);
1618         sb->s_root = d_alloc_root(root);
1619         if (!sb->s_root) {
1620                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1621                 iput(root);
1622                 goto failed_mount3;
1623         }
1624         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1625                 dput(sb->s_root);
1626                 sb->s_root = NULL;
1627                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1628                 goto failed_mount3;
1629         }
1630
1631         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1632         /*
1633          * akpm: core read_super() calls in here with the superblock locked.
1634          * That deadlocks, because orphan cleanup needs to lock the superblock
1635          * in numerous places.  Here we just pop the lock - it's relatively
1636          * harmless, because we are now ready to accept write_super() requests,
1637          * and aviro says that's the only reason for hanging onto the
1638          * superblock lock.
1639          */
1640         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1641         ext3_orphan_cleanup(sb, es);
1642         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1643         if (needs_recovery)
1644                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1645         ext3_mark_recovery_complete(sb, es);
1646         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1647                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1648                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1649                 "writeback");
1650
1651         percpu_counter_mod(&sbi->s_freeblocks_counter,
1652                 ext3_count_free_blocks(sb));
1653         percpu_counter_mod(&sbi->s_freeinodes_counter,
1654                 ext3_count_free_inodes(sb));
1655         percpu_counter_mod(&sbi->s_dirs_counter,
1656                 ext3_count_dirs(sb));
1657
1658         lock_kernel();
1659         return 0;
1660
1661 cantfind_ext3:
1662         if (!silent)
1663                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1664                        sb->s_id);
1665         goto failed_mount;
1666
1667 failed_mount3:
1668         journal_destroy(sbi->s_journal);
1669 failed_mount2:
1670         for (i = 0; i < db_count; i++)
1671                 brelse(sbi->s_group_desc[i]);
1672         kfree(sbi->s_group_desc);
1673 failed_mount:
1674 #ifdef CONFIG_QUOTA
1675         for (i = 0; i < MAXQUOTAS; i++)
1676                 kfree(sbi->s_qf_names[i]);
1677 #endif
1678         ext3_blkdev_remove(sbi);
1679         brelse(bh);
1680 out_fail:
1681         sb->s_fs_info = NULL;
1682         kfree(sbi);
1683         lock_kernel();
1684         return -EINVAL;
1685 }
1686
1687 /*
1688  * Setup any per-fs journal parameters now.  We'll do this both on
1689  * initial mount, once the journal has been initialised but before we've
1690  * done any recovery; and again on any subsequent remount.
1691  */
1692 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1693 {
1694         struct ext3_sb_info *sbi = EXT3_SB(sb);
1695
1696         if (sbi->s_commit_interval)
1697                 journal->j_commit_interval = sbi->s_commit_interval;
1698         /* We could also set up an ext3-specific default for the commit
1699          * interval here, but for now we'll just fall back to the jbd
1700          * default. */
1701
1702         spin_lock(&journal->j_state_lock);
1703         if (test_opt(sb, BARRIER))
1704                 journal->j_flags |= JFS_BARRIER;
1705         else
1706                 journal->j_flags &= ~JFS_BARRIER;
1707         spin_unlock(&journal->j_state_lock);
1708 }
1709
1710 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1711 {
1712         struct inode *journal_inode;
1713         journal_t *journal;
1714
1715         /* First, test for the existence of a valid inode on disk.  Bad
1716          * things happen if we iget() an unused inode, as the subsequent
1717          * iput() will try to delete it. */
1718
1719         journal_inode = iget(sb, journal_inum);
1720         if (!journal_inode) {
1721                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1722                 return NULL;
1723         }
1724         if (!journal_inode->i_nlink) {
1725                 make_bad_inode(journal_inode);
1726                 iput(journal_inode);
1727                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1728                 return NULL;
1729         }
1730
1731         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1732                   journal_inode, journal_inode->i_size);
1733         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1734                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1735                 iput(journal_inode);
1736                 return NULL;
1737         }
1738
1739         journal = journal_init_inode(journal_inode);
1740         if (!journal) {
1741                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1742                 iput(journal_inode);
1743                 return NULL;
1744         }
1745         journal->j_private = sb;
1746         ext3_init_journal_params(sb, journal);
1747         return journal;
1748 }
1749
1750 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1751                                        dev_t j_dev)
1752 {
1753         struct buffer_head * bh;
1754         journal_t *journal;
1755         int start;
1756         int len;
1757         int hblock, blocksize;
1758         unsigned long sb_block;
1759         unsigned long offset;
1760         struct ext3_super_block * es;
1761         struct block_device *bdev;
1762
1763         bdev = ext3_blkdev_get(j_dev);
1764         if (bdev == NULL)
1765                 return NULL;
1766
1767         if (bd_claim(bdev, sb)) {
1768                 printk(KERN_ERR
1769                         "EXT3: failed to claim external journal device.\n");
1770                 blkdev_put(bdev);
1771                 return NULL;
1772         }
1773
1774         blocksize = sb->s_blocksize;
1775         hblock = bdev_hardsect_size(bdev);
1776         if (blocksize < hblock) {
1777                 printk(KERN_ERR
1778                         "EXT3-fs: blocksize too small for journal device.\n");
1779                 goto out_bdev;
1780         }
1781
1782         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1783         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1784         set_blocksize(bdev, blocksize);
1785         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1786                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1787                        "external journal\n");
1788                 goto out_bdev;
1789         }
1790
1791         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1792         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1793             !(le32_to_cpu(es->s_feature_incompat) &
1794               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1795                 printk(KERN_ERR "EXT3-fs: external journal has "
1796                                         "bad superblock\n");
1797                 brelse(bh);
1798                 goto out_bdev;
1799         }
1800
1801         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1802                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1803                 brelse(bh);
1804                 goto out_bdev;
1805         }
1806
1807         len = le32_to_cpu(es->s_blocks_count);
1808         start = sb_block + 1;
1809         brelse(bh);     /* we're done with the superblock */
1810
1811         journal = journal_init_dev(bdev, sb->s_bdev,
1812                                         start, len, blocksize);
1813         if (!journal) {
1814                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1815                 goto out_bdev;
1816         }
1817         journal->j_private = sb;
1818         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1819         wait_on_buffer(journal->j_sb_buffer);
1820         if (!buffer_uptodate(journal->j_sb_buffer)) {
1821                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1822                 goto out_journal;
1823         }
1824         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1825                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1826                                         "user (unsupported) - %d\n",
1827                         be32_to_cpu(journal->j_superblock->s_nr_users));
1828                 goto out_journal;
1829         }
1830         EXT3_SB(sb)->journal_bdev = bdev;
1831         ext3_init_journal_params(sb, journal);
1832         return journal;
1833 out_journal:
1834         journal_destroy(journal);
1835 out_bdev:
1836         ext3_blkdev_put(bdev);
1837         return NULL;
1838 }
1839
1840 static int ext3_load_journal(struct super_block * sb,
1841                              struct ext3_super_block * es)
1842 {
1843         journal_t *journal;
1844         int journal_inum = le32_to_cpu(es->s_journal_inum);
1845         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1846         int err = 0;
1847         int really_read_only;
1848
1849         really_read_only = bdev_read_only(sb->s_bdev);
1850
1851         /*
1852          * Are we loading a blank journal or performing recovery after a
1853          * crash?  For recovery, we need to check in advance whether we
1854          * can get read-write access to the device.
1855          */
1856
1857         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1858                 if (sb->s_flags & MS_RDONLY) {
1859                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1860                                         "required on readonly filesystem.\n");
1861                         if (really_read_only) {
1862                                 printk(KERN_ERR "EXT3-fs: write access "
1863                                         "unavailable, cannot proceed.\n");
1864                                 return -EROFS;
1865                         }
1866                         printk (KERN_INFO "EXT3-fs: write access will "
1867                                         "be enabled during recovery.\n");
1868                 }
1869         }
1870
1871         if (journal_inum && journal_dev) {
1872                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1873                        "and inode journals!\n");
1874                 return -EINVAL;
1875         }
1876
1877         if (journal_inum) {
1878                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1879                         return -EINVAL;
1880         } else {
1881                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1882                         return -EINVAL;
1883         }
1884
1885         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1886                 err = journal_update_format(journal);
1887                 if (err)  {
1888                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1889                         journal_destroy(journal);
1890                         return err;
1891                 }
1892         }
1893
1894         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1895                 err = journal_wipe(journal, !really_read_only);
1896         if (!err)
1897                 err = journal_load(journal);
1898
1899         if (err) {
1900                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1901                 journal_destroy(journal);
1902                 return err;
1903         }
1904
1905         EXT3_SB(sb)->s_journal = journal;
1906         ext3_clear_journal_err(sb, es);
1907         return 0;
1908 }
1909
1910 static int ext3_create_journal(struct super_block * sb,
1911                                struct ext3_super_block * es,
1912                                int journal_inum)
1913 {
1914         journal_t *journal;
1915
1916         if (sb->s_flags & MS_RDONLY) {
1917                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1918                                 "create journal.\n");
1919                 return -EROFS;
1920         }
1921
1922         if (!(journal = ext3_get_journal(sb, journal_inum)))
1923                 return -EINVAL;
1924
1925         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1926                journal_inum);
1927
1928         if (journal_create(journal)) {
1929                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1930                 journal_destroy(journal);
1931                 return -EIO;
1932         }
1933
1934         EXT3_SB(sb)->s_journal = journal;
1935
1936         ext3_update_dynamic_rev(sb);
1937         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1938         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1939
1940         es->s_journal_inum = cpu_to_le32(journal_inum);
1941         sb->s_dirt = 1;
1942
1943         /* Make sure we flush the recovery flag to disk. */
1944         ext3_commit_super(sb, es, 1);
1945
1946         return 0;
1947 }
1948
1949 static void ext3_commit_super (struct super_block * sb,
1950                                struct ext3_super_block * es,
1951                                int sync)
1952 {
1953         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1954
1955         if (!sbh)
1956                 return;
1957         es->s_wtime = cpu_to_le32(get_seconds());
1958         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1959         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1960         BUFFER_TRACE(sbh, "marking dirty");
1961         mark_buffer_dirty(sbh);
1962         if (sync)
1963                 sync_dirty_buffer(sbh);
1964 }
1965
1966
1967 /*
1968  * Have we just finished recovery?  If so, and if we are mounting (or
1969  * remounting) the filesystem readonly, then we will end up with a
1970  * consistent fs on disk.  Record that fact.
1971  */
1972 static void ext3_mark_recovery_complete(struct super_block * sb,
1973                                         struct ext3_super_block * es)
1974 {
1975         journal_t *journal = EXT3_SB(sb)->s_journal;
1976
1977         journal_lock_updates(journal);
1978         journal_flush(journal);
1979         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1980             sb->s_flags & MS_RDONLY) {
1981                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1982                 sb->s_dirt = 0;
1983                 ext3_commit_super(sb, es, 1);
1984         }
1985         journal_unlock_updates(journal);
1986 }
1987
1988 /*
1989  * If we are mounting (or read-write remounting) a filesystem whose journal
1990  * has recorded an error from a previous lifetime, move that error to the
1991  * main filesystem now.
1992  */
1993 static void ext3_clear_journal_err(struct super_block * sb,
1994                                    struct ext3_super_block * es)
1995 {
1996         journal_t *journal;
1997         int j_errno;
1998         const char *errstr;
1999
2000         journal = EXT3_SB(sb)->s_journal;
2001
2002         /*
2003          * Now check for any error status which may have been recorded in the
2004          * journal by a prior ext3_error() or ext3_abort()
2005          */
2006
2007         j_errno = journal_errno(journal);
2008         if (j_errno) {
2009                 char nbuf[16];
2010
2011                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2012                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2013                              "from previous mount: %s", errstr);
2014                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2015                              "filesystem check.");
2016
2017                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2018                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2019                 ext3_commit_super (sb, es, 1);
2020
2021                 journal_clear_err(journal);
2022         }
2023 }
2024
2025 /*
2026  * Force the running and committing transactions to commit,
2027  * and wait on the commit.
2028  */
2029 int ext3_force_commit(struct super_block *sb)
2030 {
2031         journal_t *journal;
2032         int ret;
2033
2034         if (sb->s_flags & MS_RDONLY)
2035                 return 0;
2036
2037         journal = EXT3_SB(sb)->s_journal;
2038         sb->s_dirt = 0;
2039         ret = ext3_journal_force_commit(journal);
2040         return ret;
2041 }
2042
2043 /*
2044  * Ext3 always journals updates to the superblock itself, so we don't
2045  * have to propagate any other updates to the superblock on disk at this
2046  * point.  Just start an async writeback to get the buffers on their way
2047  * to the disk.
2048  *
2049  * This implicitly triggers the writebehind on sync().
2050  */
2051
2052 static void ext3_write_super (struct super_block * sb)
2053 {
2054         if (down_trylock(&sb->s_lock) == 0)
2055                 BUG();
2056         sb->s_dirt = 0;
2057 }
2058
2059 static int ext3_sync_fs(struct super_block *sb, int wait)
2060 {
2061         tid_t target;
2062
2063         sb->s_dirt = 0;
2064         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2065                 if (wait)
2066                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2067         }
2068         return 0;
2069 }
2070
2071 /*
2072  * LVM calls this function before a (read-only) snapshot is created.  This
2073  * gives us a chance to flush the journal completely and mark the fs clean.
2074  */
2075 static void ext3_write_super_lockfs(struct super_block *sb)
2076 {
2077         sb->s_dirt = 0;
2078
2079         if (!(sb->s_flags & MS_RDONLY)) {
2080                 journal_t *journal = EXT3_SB(sb)->s_journal;
2081
2082                 /* Now we set up the journal barrier. */
2083                 journal_lock_updates(journal);
2084                 journal_flush(journal);
2085
2086                 /* Journal blocked and flushed, clear needs_recovery flag. */
2087                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2088                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2089         }
2090 }
2091
2092 /*
2093  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2094  * flag here, even though the filesystem is not technically dirty yet.
2095  */
2096 static void ext3_unlockfs(struct super_block *sb)
2097 {
2098         if (!(sb->s_flags & MS_RDONLY)) {
2099                 lock_super(sb);
2100                 /* Reser the needs_recovery flag before the fs is unlocked. */
2101                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2102                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2103                 unlock_super(sb);
2104                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2105         }
2106 }
2107
2108 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2109 {
2110         struct ext3_super_block * es;
2111         struct ext3_sb_info *sbi = EXT3_SB(sb);
2112         unsigned long tmp;
2113         unsigned long n_blocks_count = 0;
2114
2115         /*
2116          * Allow the "check" option to be passed as a remount option.
2117          */
2118         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2119                 return -EINVAL;
2120
2121         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2122                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2123
2124         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2125                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2126
2127         es = sbi->s_es;
2128
2129         ext3_init_journal_params(sb, sbi->s_journal);
2130
2131         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2132                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2133                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2134                         return -EROFS;
2135
2136                 if (*flags & MS_RDONLY) {
2137                         /*
2138                          * First of all, the unconditional stuff we have to do
2139                          * to disable replay of the journal when we next remount
2140                          */
2141                         sb->s_flags |= MS_RDONLY;
2142
2143                         /*
2144                          * OK, test if we are remounting a valid rw partition
2145                          * readonly, and if so set the rdonly flag and then
2146                          * mark the partition as valid again.
2147                          */
2148                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2149                             (sbi->s_mount_state & EXT3_VALID_FS))
2150                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2151
2152                         ext3_mark_recovery_complete(sb, es);
2153                 } else {
2154                         __le32 ret;
2155                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2156                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2157                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2158                                        "remount RDWR because of unsupported "
2159                                        "optional features (%x).\n",
2160                                        sb->s_id, le32_to_cpu(ret));
2161                                 return -EROFS;
2162                         }
2163                         /*
2164                          * Mounting a RDONLY partition read-write, so reread
2165                          * and store the current valid flag.  (It may have
2166                          * been changed by e2fsck since we originally mounted
2167                          * the partition.)
2168                          */
2169                         ext3_clear_journal_err(sb, es);
2170                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2171                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2172                                 return ret;
2173                         if (!ext3_setup_super (sb, es, 0))
2174                                 sb->s_flags &= ~MS_RDONLY;
2175                 }
2176         }
2177         return 0;
2178 }
2179
2180 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2181 {
2182         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2183         unsigned long overhead;
2184         int i;
2185
2186         if (test_opt (sb, MINIX_DF))
2187                 overhead = 0;
2188         else {
2189                 unsigned long ngroups;
2190                 ngroups = EXT3_SB(sb)->s_groups_count;
2191                 smp_rmb();
2192
2193                 /*
2194                  * Compute the overhead (FS structures)
2195                  */
2196
2197                 /*
2198                  * All of the blocks before first_data_block are
2199                  * overhead
2200                  */
2201                 overhead = le32_to_cpu(es->s_first_data_block);
2202
2203                 /*
2204                  * Add the overhead attributed to the superblock and
2205                  * block group descriptors.  If the sparse superblocks
2206                  * feature is turned on, then not all groups have this.
2207                  */
2208                 for (i = 0; i < ngroups; i++) {
2209                         overhead += ext3_bg_has_super(sb, i) +
2210                                 ext3_bg_num_gdb(sb, i);
2211                         cond_resched();
2212                 }
2213
2214                 /*
2215                  * Every block group has an inode bitmap, a block
2216                  * bitmap, and an inode table.
2217                  */
2218                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2219         }
2220
2221         buf->f_type = EXT3_SUPER_MAGIC;
2222         buf->f_bsize = sb->s_blocksize;
2223         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2224         buf->f_bfree = ext3_count_free_blocks (sb);
2225         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2226         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2227                 buf->f_bavail = 0;
2228         buf->f_files = le32_to_cpu(es->s_inodes_count);
2229         buf->f_ffree = ext3_count_free_inodes (sb);
2230         buf->f_namelen = EXT3_NAME_LEN;
2231         return 0;
2232 }
2233
2234 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2235  * is locked for write. Otherwise the are possible deadlocks:
2236  * Process 1                         Process 2
2237  * ext3_create()                     quota_sync()
2238  *   journal_start()                   write_dquot()
2239  *   DQUOT_INIT()                        down(dqio_sem)
2240  *     down(dqio_sem)                    journal_start()
2241  *
2242  */
2243
2244 #ifdef CONFIG_QUOTA
2245
2246 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2247 {
2248         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2249 }
2250
2251 static int ext3_dquot_initialize(struct inode *inode, int type)
2252 {
2253         handle_t *handle;
2254         int ret, err;
2255
2256         /* We may create quota structure so we need to reserve enough blocks */
2257         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2258         if (IS_ERR(handle))
2259                 return PTR_ERR(handle);
2260         ret = dquot_initialize(inode, type);
2261         err = ext3_journal_stop(handle);
2262         if (!ret)
2263                 ret = err;
2264         return ret;
2265 }
2266
2267 static int ext3_dquot_drop(struct inode *inode)
2268 {
2269         handle_t *handle;
2270         int ret, err;
2271
2272         /* We may delete quota structure so we need to reserve enough blocks */
2273         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2274         if (IS_ERR(handle))
2275                 return PTR_ERR(handle);
2276         ret = dquot_drop(inode);
2277         err = ext3_journal_stop(handle);
2278         if (!ret)
2279                 ret = err;
2280         return ret;
2281 }
2282
2283 static int ext3_write_dquot(struct dquot *dquot)
2284 {
2285         int ret, err;
2286         handle_t *handle;
2287         struct inode *inode;
2288
2289         inode = dquot_to_inode(dquot);
2290         handle = ext3_journal_start(inode,
2291                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2292         if (IS_ERR(handle))
2293                 return PTR_ERR(handle);
2294         ret = dquot_commit(dquot);
2295         err = ext3_journal_stop(handle);
2296         if (!ret)
2297                 ret = err;
2298         return ret;
2299 }
2300
2301 static int ext3_acquire_dquot(struct dquot *dquot)
2302 {
2303         int ret, err;
2304         handle_t *handle;
2305
2306         handle = ext3_journal_start(dquot_to_inode(dquot),
2307                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2308         if (IS_ERR(handle))
2309                 return PTR_ERR(handle);
2310         ret = dquot_acquire(dquot);
2311         err = ext3_journal_stop(handle);
2312         if (!ret)
2313                 ret = err;
2314         return ret;
2315 }
2316
2317 static int ext3_release_dquot(struct dquot *dquot)
2318 {
2319         int ret, err;
2320         handle_t *handle;
2321
2322         handle = ext3_journal_start(dquot_to_inode(dquot),
2323                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2324         if (IS_ERR(handle))
2325                 return PTR_ERR(handle);
2326         ret = dquot_release(dquot);
2327         err = ext3_journal_stop(handle);
2328         if (!ret)
2329                 ret = err;
2330         return ret;
2331 }
2332
2333 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2334 {
2335         /* Are we journalling quotas? */
2336         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2337             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2338                 dquot_mark_dquot_dirty(dquot);
2339                 return ext3_write_dquot(dquot);
2340         } else {
2341                 return dquot_mark_dquot_dirty(dquot);
2342         }
2343 }
2344
2345 static int ext3_write_info(struct super_block *sb, int type)
2346 {
2347         int ret, err;
2348         handle_t *handle;
2349
2350         /* Data block + inode block */
2351         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2352         if (IS_ERR(handle))
2353                 return PTR_ERR(handle);
2354         ret = dquot_commit_info(sb, type);
2355         err = ext3_journal_stop(handle);
2356         if (!ret)
2357                 ret = err;
2358         return ret;
2359 }
2360
2361 /*
2362  * Turn on quotas during mount time - we need to find
2363  * the quota file and such...
2364  */
2365 static int ext3_quota_on_mount(struct super_block *sb, int type)
2366 {
2367         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2368                         EXT3_SB(sb)->s_jquota_fmt, type);
2369 }
2370
2371 /*
2372  * Standard function to be called on quota_on
2373  */
2374 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2375                          char *path)
2376 {
2377         int err;
2378         struct nameidata nd;
2379
2380         if (!test_opt(sb, QUOTA))
2381                 return -EINVAL;
2382         /* Not journalling quota? */
2383         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2384             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2385                 return vfs_quota_on(sb, type, format_id, path);
2386         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2387         if (err)
2388                 return err;
2389         /* Quotafile not on the same filesystem? */
2390         if (nd.mnt->mnt_sb != sb) {
2391                 path_release(&nd);
2392                 return -EXDEV;
2393         }
2394         /* Quotafile not of fs root? */
2395         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2396                 printk(KERN_WARNING
2397                         "EXT3-fs: Quota file not on filesystem root. "
2398                         "Journalled quota will not work.\n");
2399         path_release(&nd);
2400         return vfs_quota_on(sb, type, format_id, path);
2401 }
2402
2403 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2404  * acquiring the locks... As quota files are never truncated and quota code
2405  * itself serializes the operations (and noone else should touch the files)
2406  * we don't have to be afraid of races */
2407 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2408                                size_t len, loff_t off)
2409 {
2410         struct inode *inode = sb_dqopt(sb)->files[type];
2411         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2412         int err = 0;
2413         int offset = off & (sb->s_blocksize - 1);
2414         int tocopy;
2415         size_t toread;
2416         struct buffer_head *bh;
2417         loff_t i_size = i_size_read(inode);
2418
2419         if (off > i_size)
2420                 return 0;
2421         if (off+len > i_size)
2422                 len = i_size-off;
2423         toread = len;
2424         while (toread > 0) {
2425                 tocopy = sb->s_blocksize - offset < toread ?
2426                                 sb->s_blocksize - offset : toread;
2427                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2428                 if (err)
2429                         return err;
2430                 if (!bh)        /* A hole? */
2431                         memset(data, 0, tocopy);
2432                 else
2433                         memcpy(data, bh->b_data+offset, tocopy);
2434                 brelse(bh);
2435                 offset = 0;
2436                 toread -= tocopy;
2437                 data += tocopy;
2438                 blk++;
2439         }
2440         return len;
2441 }
2442
2443 /* Write to quotafile (we know the transaction is already started and has
2444  * enough credits) */
2445 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2446                                 const char *data, size_t len, loff_t off)
2447 {
2448         struct inode *inode = sb_dqopt(sb)->files[type];
2449         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2450         int err = 0;
2451         int offset = off & (sb->s_blocksize - 1);
2452         int tocopy;
2453         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2454         size_t towrite = len;
2455         struct buffer_head *bh;
2456         handle_t *handle = journal_current_handle();
2457
2458         down(&inode->i_sem);
2459         while (towrite > 0) {
2460                 tocopy = sb->s_blocksize - offset < towrite ?
2461                                 sb->s_blocksize - offset : towrite;
2462                 bh = ext3_bread(handle, inode, blk, 1, &err);
2463                 if (!bh)
2464                         goto out;
2465                 if (journal_quota) {
2466                         err = ext3_journal_get_write_access(handle, bh);
2467                         if (err) {
2468                                 brelse(bh);
2469                                 goto out;
2470                         }
2471                 }
2472                 lock_buffer(bh);
2473                 memcpy(bh->b_data+offset, data, tocopy);
2474                 flush_dcache_page(bh->b_page);
2475                 unlock_buffer(bh);
2476                 if (journal_quota)
2477                         err = ext3_journal_dirty_metadata(handle, bh);
2478                 else {
2479                         /* Always do at least ordered writes for quotas */
2480                         err = ext3_journal_dirty_data(handle, bh);
2481                         mark_buffer_dirty(bh);
2482                 }
2483                 brelse(bh);
2484                 if (err)
2485                         goto out;
2486                 offset = 0;
2487                 towrite -= tocopy;
2488                 data += tocopy;
2489                 blk++;
2490         }
2491 out:
2492         if (len == towrite)
2493                 return err;
2494         if (inode->i_size < off+len-towrite) {
2495                 i_size_write(inode, off+len-towrite);
2496                 EXT3_I(inode)->i_disksize = inode->i_size;
2497         }
2498         inode->i_version++;
2499         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2500         ext3_mark_inode_dirty(handle, inode);
2501         up(&inode->i_sem);
2502         return len - towrite;
2503 }
2504
2505 #endif
2506
2507 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2508         int flags, const char *dev_name, void *data)
2509 {
2510         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2511 }
2512
2513 static struct file_system_type ext3_fs_type = {
2514         .owner          = THIS_MODULE,
2515         .name           = "ext3",
2516         .get_sb         = ext3_get_sb,
2517         .kill_sb        = kill_block_super,
2518         .fs_flags       = FS_REQUIRES_DEV,
2519 };
2520
2521 static int __init init_ext3_fs(void)
2522 {
2523         int err = init_ext3_xattr();
2524         if (err)
2525                 return err;
2526         err = init_inodecache();
2527         if (err)
2528                 goto out1;
2529         err = register_filesystem(&ext3_fs_type);
2530         if (err)
2531                 goto out;
2532         return 0;
2533 out:
2534         destroy_inodecache();
2535 out1:
2536         exit_ext3_xattr();
2537         return err;
2538 }
2539
2540 static void __exit exit_ext3_fs(void)
2541 {
2542         unregister_filesystem(&ext3_fs_type);
2543         destroy_inodecache();
2544         exit_ext3_xattr();
2545 }
2546
2547 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2548 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2549 MODULE_LICENSE("GPL");
2550 module_init(init_ext3_fs)
2551 module_exit(exit_ext3_fs)