fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/parser.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/statfs.h>
  24 #include <linux/random.h>
  25 #include <linux/mount.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/log2.h>
  29 #include <linux/cleancache.h>
  30 #include <linux/namei.h>
  31
  32 #include <asm/uaccess.h>
  33
  34 #define CREATE_TRACE_POINTS
  35
  36 #include "ext3.h"
  37 #include "xattr.h"
  38 #include "acl.h"
  39 #include "namei.h"
  40
  41 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  42   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  43 #else
  44   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  45 #endif
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static int ext3_commit_super(struct super_block *sb,
  52                                struct ext3_super_block *es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static int ext3_unfreeze(struct super_block *sb);
  64 static int ext3_freeze(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  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, __func__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 int __ext3_journal_stop(const char *where, handle_t *handle)
  90 {
  91         struct super_block *sb;
  92         int err;
  93         int rc;
  94
  95         sb = handle->h_transaction->t_journal->j_private;
  96         err = handle->h_err;
  97         rc = journal_stop(handle);
  98
  99         if (!err)
 100                 err = rc;
 101         if (err)
 102                 __ext3_std_error(sb, where, err);
 103         return err;
 104 }
 105
 106 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 107                 struct buffer_head *bh, handle_t *handle, int err)
 108 {
 109         char nbuf[16];
 110         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 111
 112         if (bh)
 113                 BUFFER_TRACE(bh, "abort");
 114
 115         if (!handle->h_err)
 116                 handle->h_err = err;
 117
 118         if (is_handle_aborted(handle))
 119                 return;
 120
 121         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 122                 caller, errstr, err_fn);
 123
 124         journal_abort_handle(handle);
 125 }
 126
 127 void ext3_msg(struct super_block *sb, const char *prefix,
 128                 const char *fmt, ...)
 129 {
 130         struct va_format vaf;
 131         va_list args;
 132
 133         va_start(args, fmt);
 134
 135         vaf.fmt = fmt;
 136         vaf.va = &args;
 137
 138         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 139
 140         va_end(args);
 141 }
 142
 143 /* Deal with the reporting of failure conditions on a filesystem such as
 144  * inconsistencies detected or read IO failures.
 145  *
 146  * On ext2, we can store the error state of the filesystem in the
 147  * superblock.  That is not possible on ext3, because we may have other
 148  * write ordering constraints on the superblock which prevent us from
 149  * writing it out straight away; and given that the journal is about to
 150  * be aborted, we can't rely on the current, or future, transactions to
 151  * write out the superblock safely.
 152  *
 153  * We'll just use the journal_abort() error code to record an error in
 154  * the journal instead.  On recovery, the journal will complain about
 155  * that error until we've noted it down and cleared it.
 156  */
 157
 158 static void ext3_handle_error(struct super_block *sb)
 159 {
 160         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 161
 162         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 163         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 164
 165         if (sb->s_flags & MS_RDONLY)
 166                 return;
 167
 168         if (!test_opt (sb, ERRORS_CONT)) {
 169                 journal_t *journal = EXT3_SB(sb)->s_journal;
 170
 171                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 172                 if (journal)
 173                         journal_abort(journal, -EIO);
 174         }
 175         if (test_opt (sb, ERRORS_RO)) {
 176                 ext3_msg(sb, KERN_CRIT,
 177                         "error: remounting filesystem read-only");
 178                 /*
 179                  * Make sure updated value of ->s_mount_state will be visible
 180                  * before ->s_flags update.
 181                  */
 182                 smp_wmb();
 183                 sb->s_flags |= MS_RDONLY;
 184         }
 185         ext3_commit_super(sb, es, 1);
 186         if (test_opt(sb, ERRORS_PANIC))
 187                 panic("EXT3-fs (%s): panic forced after error\n",
 188                         sb->s_id);
 189 }
 190
 191 void ext3_error(struct super_block *sb, const char *function,
 192                 const char *fmt, ...)
 193 {
 194         struct va_format vaf;
 195         va_list args;
 196
 197         va_start(args, fmt);
 198
 199         vaf.fmt = fmt;
 200         vaf.va = &args;
 201
 202         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 203                sb->s_id, function, &vaf);
 204
 205         va_end(args);
 206
 207         ext3_handle_error(sb);
 208 }
 209
 210 static const char *ext3_decode_error(struct super_block * sb, int errno,
 211                                      char nbuf[16])
 212 {
 213         char *errstr = NULL;
 214
 215         switch (errno) {
 216         case -EIO:
 217                 errstr = "IO failure";
 218                 break;
 219         case -ENOMEM:
 220                 errstr = "Out of memory";
 221                 break;
 222         case -EROFS:
 223                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 224                         errstr = "Journal has aborted";
 225                 else
 226                         errstr = "Readonly filesystem";
 227                 break;
 228         default:
 229                 /* If the caller passed in an extra buffer for unknown
 230                  * errors, textualise them now.  Else we just return
 231                  * NULL. */
 232                 if (nbuf) {
 233                         /* Check for truncated error codes... */
 234                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 235                                 errstr = nbuf;
 236                 }
 237                 break;
 238         }
 239
 240         return errstr;
 241 }
 242
 243 /* __ext3_std_error decodes expected errors from journaling functions
 244  * automatically and invokes the appropriate error response.  */
 245
 246 void __ext3_std_error (struct super_block * sb, const char * function,
 247                        int errno)
 248 {
 249         char nbuf[16];
 250         const char *errstr;
 251
 252         /* Special case: if the error is EROFS, and we're not already
 253          * inside a transaction, then there's really no point in logging
 254          * an error. */
 255         if (errno == -EROFS && journal_current_handle() == NULL &&
 256             (sb->s_flags & MS_RDONLY))
 257                 return;
 258
 259         errstr = ext3_decode_error(sb, errno, nbuf);
 260         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 261
 262         ext3_handle_error(sb);
 263 }
 264
 265 /*
 266  * ext3_abort is a much stronger failure handler than ext3_error.  The
 267  * abort function may be used to deal with unrecoverable failures such
 268  * as journal IO errors or ENOMEM at a critical moment in log management.
 269  *
 270  * We unconditionally force the filesystem into an ABORT|READONLY state,
 271  * unless the error response on the fs has been set to panic in which
 272  * case we take the easy way out and panic immediately.
 273  */
 274
 275 void ext3_abort(struct super_block *sb, const char *function,
 276                  const char *fmt, ...)
 277 {
 278         struct va_format vaf;
 279         va_list args;
 280
 281         va_start(args, fmt);
 282
 283         vaf.fmt = fmt;
 284         vaf.va = &args;
 285
 286         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 287                sb->s_id, function, &vaf);
 288
 289         va_end(args);
 290
 291         if (test_opt(sb, ERRORS_PANIC))
 292                 panic("EXT3-fs: panic from previous error\n");
 293
 294         if (sb->s_flags & MS_RDONLY)
 295                 return;
 296
 297         ext3_msg(sb, KERN_CRIT,
 298                 "error: remounting filesystem read-only");
 299         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 300         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 301         /*
 302          * Make sure updated value of ->s_mount_state will be visible
 303          * before ->s_flags update.
 304          */
 305         smp_wmb();
 306         sb->s_flags |= MS_RDONLY;
 307
 308         if (EXT3_SB(sb)->s_journal)
 309                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 310 }
 311
 312 void ext3_warning(struct super_block *sb, const char *function,
 313                   const char *fmt, ...)
 314 {
 315         struct va_format vaf;
 316         va_list args;
 317
 318         va_start(args, fmt);
 319
 320         vaf.fmt = fmt;
 321         vaf.va = &args;
 322
 323         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 324                sb->s_id, function, &vaf);
 325
 326         va_end(args);
 327 }
 328
 329 void ext3_update_dynamic_rev(struct super_block *sb)
 330 {
 331         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 332
 333         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 334                 return;
 335
 336         ext3_msg(sb, KERN_WARNING,
 337                 "warning: updating to rev %d because of "
 338                 "new feature flag, running e2fsck is recommended",
 339                 EXT3_DYNAMIC_REV);
 340
 341         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 342         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 343         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 344         /* leave es->s_feature_*compat flags alone */
 345         /* es->s_uuid will be set by e2fsck if empty */
 346
 347         /*
 348          * The rest of the superblock fields should be zero, and if not it
 349          * means they are likely already in use, so leave them alone.  We
 350          * can leave it up to e2fsck to clean up any inconsistencies there.
 351          */
 352 }
 353
 354 /*
 355  * Open the external journal device
 356  */
 357 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 358 {
 359         struct block_device *bdev;
 360         char b[BDEVNAME_SIZE];
 361
 362         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 363         if (IS_ERR(bdev))
 364                 goto fail;
 365         return bdev;
 366
 367 fail:
 368         ext3_msg(sb, KERN_ERR, "error: failed to open journal device %s: %ld",
 369                 __bdevname(dev, b), PTR_ERR(bdev));
 370
 371         return NULL;
 372 }
 373
 374 /*
 375  * Release the journal device
 376  */
 377 static void ext3_blkdev_put(struct block_device *bdev)
 378 {
 379         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 380 }
 381
 382 static void ext3_blkdev_remove(struct ext3_sb_info *sbi)
 383 {
 384         struct block_device *bdev;
 385         bdev = sbi->journal_bdev;
 386         if (bdev) {
 387                 ext3_blkdev_put(bdev);
 388                 sbi->journal_bdev = NULL;
 389         }
 390 }
 391
 392 static inline struct inode *orphan_list_entry(struct list_head *l)
 393 {
 394         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 395 }
 396
 397 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 398 {
 399         struct list_head *l;
 400
 401         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 402                le32_to_cpu(sbi->s_es->s_last_orphan));
 403
 404         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 405         list_for_each(l, &sbi->s_orphan) {
 406                 struct inode *inode = orphan_list_entry(l);
 407                 ext3_msg(sb, KERN_ERR, "  "
 408                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 409                        inode->i_sb->s_id, inode->i_ino, inode,
 410                        inode->i_mode, inode->i_nlink,
 411                        NEXT_ORPHAN(inode));
 412         }
 413 }
 414
 415 static void ext3_put_super (struct super_block * sb)
 416 {
 417         struct ext3_sb_info *sbi = EXT3_SB(sb);
 418         struct ext3_super_block *es = sbi->s_es;
 419         int i, err;
 420
 421         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 422         ext3_xattr_put_super(sb);
 423         err = journal_destroy(sbi->s_journal);
 424         sbi->s_journal = NULL;
 425         if (err < 0)
 426                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 427
 428         if (!(sb->s_flags & MS_RDONLY)) {
 429                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 430                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 431                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 432                 mark_buffer_dirty(sbi->s_sbh);
 433                 ext3_commit_super(sb, es, 1);
 434         }
 435
 436         for (i = 0; i < sbi->s_gdb_count; i++)
 437                 brelse(sbi->s_group_desc[i]);
 438         kfree(sbi->s_group_desc);
 439         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 440         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 441         percpu_counter_destroy(&sbi->s_dirs_counter);
 442         brelse(sbi->s_sbh);
 443 #ifdef CONFIG_QUOTA
 444         for (i = 0; i < EXT3_MAXQUOTAS; i++)
 445                 kfree(sbi->s_qf_names[i]);
 446 #endif
 447
 448         /* Debugging code just in case the in-memory inode orphan list
 449          * isn't empty.  The on-disk one can be non-empty if we've
 450          * detected an error and taken the fs readonly, but the
 451          * in-memory list had better be clean by this point. */
 452         if (!list_empty(&sbi->s_orphan))
 453                 dump_orphan_list(sb, sbi);
 454         J_ASSERT(list_empty(&sbi->s_orphan));
 455
 456         invalidate_bdev(sb->s_bdev);
 457         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 458                 /*
 459                  * Invalidate the journal device's buffers.  We don't want them
 460                  * floating about in memory - the physical journal device may
 461                  * hotswapped, and it breaks the `ro-after' testing code.
 462                  */
 463                 sync_blockdev(sbi->journal_bdev);
 464                 invalidate_bdev(sbi->journal_bdev);
 465                 ext3_blkdev_remove(sbi);
 466         }
 467         sb->s_fs_info = NULL;
 468         kfree(sbi->s_blockgroup_lock);
 469         mutex_destroy(&sbi->s_orphan_lock);
 470         mutex_destroy(&sbi->s_resize_lock);
 471         kfree(sbi);
 472 }
 473
 474 static struct kmem_cache *ext3_inode_cachep;
 475
 476 /*
 477  * Called inside transaction, so use GFP_NOFS
 478  */
 479 static struct inode *ext3_alloc_inode(struct super_block *sb)
 480 {
 481         struct ext3_inode_info *ei;
 482
 483         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 484         if (!ei)
 485                 return NULL;
 486         ei->i_block_alloc_info = NULL;
 487         ei->vfs_inode.i_version = 1;
 488         atomic_set(&ei->i_datasync_tid, 0);
 489         atomic_set(&ei->i_sync_tid, 0);
 490 #ifdef CONFIG_QUOTA
 491         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 492 #endif
 493
 494         return &ei->vfs_inode;
 495 }
 496
 497 static int ext3_drop_inode(struct inode *inode)
 498 {
 499         int drop = generic_drop_inode(inode);
 500
 501         trace_ext3_drop_inode(inode, drop);
 502         return drop;
 503 }
 504
 505 static void ext3_i_callback(struct rcu_head *head)
 506 {
 507         struct inode *inode = container_of(head, struct inode, i_rcu);
 508         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 509 }
 510
 511 static void ext3_destroy_inode(struct inode *inode)
 512 {
 513         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 514                 printk("EXT3 Inode %p: orphan list check failed!\n",
 515                         EXT3_I(inode));
 516                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 517                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 518                                 false);
 519                 dump_stack();
 520         }
 521         call_rcu(&inode->i_rcu, ext3_i_callback);
 522 }
 523
 524 static void init_once(void *foo)
 525 {
 526         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 527
 528         INIT_LIST_HEAD(&ei->i_orphan);
 529 #ifdef CONFIG_EXT3_FS_XATTR
 530         init_rwsem(&ei->xattr_sem);
 531 #endif
 532         mutex_init(&ei->truncate_mutex);
 533         inode_init_once(&ei->vfs_inode);
 534 }
 535
 536 static int __init init_inodecache(void)
 537 {
 538         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 539                                              sizeof(struct ext3_inode_info),
 540                                              0, (SLAB_RECLAIM_ACCOUNT|
 541                                                 SLAB_MEM_SPREAD),
 542                                              init_once);
 543         if (ext3_inode_cachep == NULL)
 544                 return -ENOMEM;
 545         return 0;
 546 }
 547
 548 static void destroy_inodecache(void)
 549 {
 550         /*
 551          * Make sure all delayed rcu free inodes are flushed before we
 552          * destroy cache.
 553          */
 554         rcu_barrier();
 555         kmem_cache_destroy(ext3_inode_cachep);
 556 }
 557
 558 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 559 {
 560 #if defined(CONFIG_QUOTA)
 561         struct ext3_sb_info *sbi = EXT3_SB(sb);
 562
 563         if (sbi->s_jquota_fmt) {
 564                 char *fmtname = "";
 565
 566                 switch (sbi->s_jquota_fmt) {
 567                 case QFMT_VFS_OLD:
 568                         fmtname = "vfsold";
 569                         break;
 570                 case QFMT_VFS_V0:
 571                         fmtname = "vfsv0";
 572                         break;
 573                 case QFMT_VFS_V1:
 574                         fmtname = "vfsv1";
 575                         break;
 576                 }
 577                 seq_printf(seq, ",jqfmt=%s", fmtname);
 578         }
 579
 580         if (sbi->s_qf_names[USRQUOTA])
 581                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 582
 583         if (sbi->s_qf_names[GRPQUOTA])
 584                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 585
 586         if (test_opt(sb, USRQUOTA))
 587                 seq_puts(seq, ",usrquota");
 588
 589         if (test_opt(sb, GRPQUOTA))
 590                 seq_puts(seq, ",grpquota");
 591 #endif
 592 }
 593
 594 static char *data_mode_string(unsigned long mode)
 595 {
 596         switch (mode) {
 597         case EXT3_MOUNT_JOURNAL_DATA:
 598                 return "journal";
 599         case EXT3_MOUNT_ORDERED_DATA:
 600                 return "ordered";
 601         case EXT3_MOUNT_WRITEBACK_DATA:
 602                 return "writeback";
 603         }
 604         return "unknown";
 605 }
 606
 607 /*
 608  * Show an option if
 609  *  - it's set to a non-default value OR
 610  *  - if the per-sb default is different from the global default
 611  */
 612 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 613 {
 614         struct super_block *sb = root->d_sb;
 615         struct ext3_sb_info *sbi = EXT3_SB(sb);
 616         struct ext3_super_block *es = sbi->s_es;
 617         unsigned long def_mount_opts;
 618
 619         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 620
 621         if (sbi->s_sb_block != 1)
 622                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 623         if (test_opt(sb, MINIX_DF))
 624                 seq_puts(seq, ",minixdf");
 625         if (test_opt(sb, GRPID))
 626                 seq_puts(seq, ",grpid");
 627         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 628                 seq_puts(seq, ",nogrpid");
 629         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
 630             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 631                 seq_printf(seq, ",resuid=%u",
 632                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
 633         }
 634         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
 635             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 636                 seq_printf(seq, ",resgid=%u",
 637                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
 638         }
 639         if (test_opt(sb, ERRORS_RO)) {
 640                 int def_errors = le16_to_cpu(es->s_errors);
 641
 642                 if (def_errors == EXT3_ERRORS_PANIC ||
 643                     def_errors == EXT3_ERRORS_CONTINUE) {
 644                         seq_puts(seq, ",errors=remount-ro");
 645                 }
 646         }
 647         if (test_opt(sb, ERRORS_CONT))
 648                 seq_puts(seq, ",errors=continue");
 649         if (test_opt(sb, ERRORS_PANIC))
 650                 seq_puts(seq, ",errors=panic");
 651         if (test_opt(sb, NO_UID32))
 652                 seq_puts(seq, ",nouid32");
 653         if (test_opt(sb, DEBUG))
 654                 seq_puts(seq, ",debug");
 655 #ifdef CONFIG_EXT3_FS_XATTR
 656         if (test_opt(sb, XATTR_USER))
 657                 seq_puts(seq, ",user_xattr");
 658         if (!test_opt(sb, XATTR_USER) &&
 659             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 660                 seq_puts(seq, ",nouser_xattr");
 661         }
 662 #endif
 663 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 664         if (test_opt(sb, POSIX_ACL))
 665                 seq_puts(seq, ",acl");
 666         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 667                 seq_puts(seq, ",noacl");
 668 #endif
 669         if (!test_opt(sb, RESERVATION))
 670                 seq_puts(seq, ",noreservation");
 671         if (sbi->s_commit_interval) {
 672                 seq_printf(seq, ",commit=%u",
 673                            (unsigned) (sbi->s_commit_interval / HZ));
 674         }
 675
 676         /*
 677          * Always display barrier state so it's clear what the status is.
 678          */
 679         seq_puts(seq, ",barrier=");
 680         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 681         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 682         if (test_opt(sb, DATA_ERR_ABORT))
 683                 seq_puts(seq, ",data_err=abort");
 684
 685         if (test_opt(sb, NOLOAD))
 686                 seq_puts(seq, ",norecovery");
 687
 688         ext3_show_quota_options(seq, sb);
 689
 690         return 0;
 691 }
 692
 693
 694 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 695                 u64 ino, u32 generation)
 696 {
 697         struct inode *inode;
 698
 699         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 700                 return ERR_PTR(-ESTALE);
 701         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 702                 return ERR_PTR(-ESTALE);
 703
 704         /* iget isn't really right if the inode is currently unallocated!!
 705          *
 706          * ext3_read_inode will return a bad_inode if the inode had been
 707          * deleted, so we should be safe.
 708          *
 709          * Currently we don't know the generation for parent directory, so
 710          * a generation of 0 means "accept any"
 711          */
 712         inode = ext3_iget(sb, ino);
 713         if (IS_ERR(inode))
 714                 return ERR_CAST(inode);
 715         if (generation && inode->i_generation != generation) {
 716                 iput(inode);
 717                 return ERR_PTR(-ESTALE);
 718         }
 719
 720         return inode;
 721 }
 722
 723 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 724                 int fh_len, int fh_type)
 725 {
 726         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 727                                     ext3_nfs_get_inode);
 728 }
 729
 730 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 731                 int fh_len, int fh_type)
 732 {
 733         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 734                                     ext3_nfs_get_inode);
 735 }
 736
 737 /*
 738  * Try to release metadata pages (indirect blocks, directories) which are
 739  * mapped via the block device.  Since these pages could have journal heads
 740  * which would prevent try_to_free_buffers() from freeing them, we must use
 741  * jbd layer's try_to_free_buffers() function to release them.
 742  */
 743 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 744                                  gfp_t wait)
 745 {
 746         journal_t *journal = EXT3_SB(sb)->s_journal;
 747
 748         WARN_ON(PageChecked(page));
 749         if (!page_has_buffers(page))
 750                 return 0;
 751         if (journal)
 752                 return journal_try_to_free_buffers(journal, page,
 753                                                    wait & ~__GFP_WAIT);
 754         return try_to_free_buffers(page);
 755 }
 756
 757 #ifdef CONFIG_QUOTA
 758 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 759 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 760
 761 static int ext3_write_dquot(struct dquot *dquot);
 762 static int ext3_acquire_dquot(struct dquot *dquot);
 763 static int ext3_release_dquot(struct dquot *dquot);
 764 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 765 static int ext3_write_info(struct super_block *sb, int type);
 766 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 767                          struct path *path);
 768 static int ext3_quota_on_mount(struct super_block *sb, int type);
 769 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 770                                size_t len, loff_t off);
 771 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 772                                 const char *data, size_t len, loff_t off);
 773 static struct dquot **ext3_get_dquots(struct inode *inode)
 774 {
 775         return EXT3_I(inode)->i_dquot;
 776 }
 777
 778 static const struct dquot_operations ext3_quota_operations = {
 779         .write_dquot    = ext3_write_dquot,
 780         .acquire_dquot  = ext3_acquire_dquot,
 781         .release_dquot  = ext3_release_dquot,
 782         .mark_dirty     = ext3_mark_dquot_dirty,
 783         .write_info     = ext3_write_info,
 784         .alloc_dquot    = dquot_alloc,
 785         .destroy_dquot  = dquot_destroy,
 786 };
 787
 788 static const struct quotactl_ops ext3_qctl_operations = {
 789         .quota_on       = ext3_quota_on,
 790         .quota_off      = dquot_quota_off,
 791         .quota_sync     = dquot_quota_sync,
 792         .get_info       = dquot_get_dqinfo,
 793         .set_info       = dquot_set_dqinfo,
 794         .get_dqblk      = dquot_get_dqblk,
 795         .set_dqblk      = dquot_set_dqblk
 796 };
 797 #endif
 798
 799 static const struct super_operations ext3_sops = {
 800         .alloc_inode    = ext3_alloc_inode,
 801         .destroy_inode  = ext3_destroy_inode,
 802         .write_inode    = ext3_write_inode,
 803         .dirty_inode    = ext3_dirty_inode,
 804         .drop_inode     = ext3_drop_inode,
 805         .evict_inode    = ext3_evict_inode,
 806         .put_super      = ext3_put_super,
 807         .sync_fs        = ext3_sync_fs,
 808         .freeze_fs      = ext3_freeze,
 809         .unfreeze_fs    = ext3_unfreeze,
 810         .statfs         = ext3_statfs,
 811         .remount_fs     = ext3_remount,
 812         .show_options   = ext3_show_options,
 813 #ifdef CONFIG_QUOTA
 814         .quota_read     = ext3_quota_read,
 815         .quota_write    = ext3_quota_write,
 816         .get_dquots     = ext3_get_dquots,
 817 #endif
 818         .bdev_try_to_free_page = bdev_try_to_free_page,
 819 };
 820
 821 static const struct export_operations ext3_export_ops = {
 822         .fh_to_dentry = ext3_fh_to_dentry,
 823         .fh_to_parent = ext3_fh_to_parent,
 824         .get_parent = ext3_get_parent,
 825 };
 826
 827 enum {
 828         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 829         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 830         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 831         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 832         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 833         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 834         Opt_journal_path,
 835         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 836         Opt_data_err_abort, Opt_data_err_ignore,
 837         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 838         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 839         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 840         Opt_resize, Opt_usrquota, Opt_grpquota
 841 };
 842
 843 static const match_table_t tokens = {
 844         {Opt_bsd_df, "bsddf"},
 845         {Opt_minix_df, "minixdf"},
 846         {Opt_grpid, "grpid"},
 847         {Opt_grpid, "bsdgroups"},
 848         {Opt_nogrpid, "nogrpid"},
 849         {Opt_nogrpid, "sysvgroups"},
 850         {Opt_resgid, "resgid=%u"},
 851         {Opt_resuid, "resuid=%u"},
 852         {Opt_sb, "sb=%u"},
 853         {Opt_err_cont, "errors=continue"},
 854         {Opt_err_panic, "errors=panic"},
 855         {Opt_err_ro, "errors=remount-ro"},
 856         {Opt_nouid32, "nouid32"},
 857         {Opt_nocheck, "nocheck"},
 858         {Opt_nocheck, "check=none"},
 859         {Opt_debug, "debug"},
 860         {Opt_oldalloc, "oldalloc"},
 861         {Opt_orlov, "orlov"},
 862         {Opt_user_xattr, "user_xattr"},
 863         {Opt_nouser_xattr, "nouser_xattr"},
 864         {Opt_acl, "acl"},
 865         {Opt_noacl, "noacl"},
 866         {Opt_reservation, "reservation"},
 867         {Opt_noreservation, "noreservation"},
 868         {Opt_noload, "noload"},
 869         {Opt_noload, "norecovery"},
 870         {Opt_nobh, "nobh"},
 871         {Opt_bh, "bh"},
 872         {Opt_commit, "commit=%u"},
 873         {Opt_journal_update, "journal=update"},
 874         {Opt_journal_inum, "journal=%u"},
 875         {Opt_journal_dev, "journal_dev=%u"},
 876         {Opt_journal_path, "journal_path=%s"},
 877         {Opt_abort, "abort"},
 878         {Opt_data_journal, "data=journal"},
 879         {Opt_data_ordered, "data=ordered"},
 880         {Opt_data_writeback, "data=writeback"},
 881         {Opt_data_err_abort, "data_err=abort"},
 882         {Opt_data_err_ignore, "data_err=ignore"},
 883         {Opt_offusrjquota, "usrjquota="},
 884         {Opt_usrjquota, "usrjquota=%s"},
 885         {Opt_offgrpjquota, "grpjquota="},
 886         {Opt_grpjquota, "grpjquota=%s"},
 887         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 888         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 889         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 890         {Opt_grpquota, "grpquota"},
 891         {Opt_noquota, "noquota"},
 892         {Opt_quota, "quota"},
 893         {Opt_usrquota, "usrquota"},
 894         {Opt_barrier, "barrier=%u"},
 895         {Opt_barrier, "barrier"},
 896         {Opt_nobarrier, "nobarrier"},
 897         {Opt_resize, "resize"},
 898         {Opt_err, NULL},
 899 };
 900
 901 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 902 {
 903         ext3_fsblk_t    sb_block;
 904         char            *options = (char *) *data;
 905
 906         if (!options || strncmp(options, "sb=", 3) != 0)
 907                 return 1;       /* Default location */
 908         options += 3;
 909         /*todo: use simple_strtoll with >32bit ext3 */
 910         sb_block = simple_strtoul(options, &options, 0);
 911         if (*options && *options != ',') {
 912                 ext3_msg(sb, KERN_ERR, "error: invalid sb specification: %s",
 913                        (char *) *data);
 914                 return 1;
 915         }
 916         if (*options == ',')
 917                 options++;
 918         *data = (void *) options;
 919         return sb_block;
 920 }
 921
 922 #ifdef CONFIG_QUOTA
 923 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 924 {
 925         struct ext3_sb_info *sbi = EXT3_SB(sb);
 926         char *qname;
 927
 928         if (sb_any_quota_loaded(sb) &&
 929                 !sbi->s_qf_names[qtype]) {
 930                 ext3_msg(sb, KERN_ERR,
 931                         "Cannot change journaled "
 932                         "quota options when quota turned on");
 933                 return 0;
 934         }
 935         qname = match_strdup(args);
 936         if (!qname) {
 937                 ext3_msg(sb, KERN_ERR,
 938                         "Not enough memory for storing quotafile name");
 939                 return 0;
 940         }
 941         if (sbi->s_qf_names[qtype]) {
 942                 int same = !strcmp(sbi->s_qf_names[qtype], qname);
 943
 944                 kfree(qname);
 945                 if (!same) {
 946                         ext3_msg(sb, KERN_ERR,
 947                                  "%s quota file already specified",
 948                                  QTYPE2NAME(qtype));
 949                 }
 950                 return same;
 951         }
 952         if (strchr(qname, '/')) {
 953                 ext3_msg(sb, KERN_ERR,
 954                         "quotafile must be on filesystem root");
 955                 kfree(qname);
 956                 return 0;
 957         }
 958         sbi->s_qf_names[qtype] = qname;
 959         set_opt(sbi->s_mount_opt, QUOTA);
 960         return 1;
 961 }
 962
 963 static int clear_qf_name(struct super_block *sb, int qtype) {
 964
 965         struct ext3_sb_info *sbi = EXT3_SB(sb);
 966
 967         if (sb_any_quota_loaded(sb) &&
 968                 sbi->s_qf_names[qtype]) {
 969                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 970                         " when quota turned on");
 971                 return 0;
 972         }
 973         if (sbi->s_qf_names[qtype]) {
 974                 kfree(sbi->s_qf_names[qtype]);
 975                 sbi->s_qf_names[qtype] = NULL;
 976         }
 977         return 1;
 978 }
 979 #endif
 980
 981 static int parse_options (char *options, struct super_block *sb,
 982                           unsigned int *inum, unsigned long *journal_devnum,
 983                           ext3_fsblk_t *n_blocks_count, int is_remount)
 984 {
 985         struct ext3_sb_info *sbi = EXT3_SB(sb);
 986         char * p;
 987         substring_t args[MAX_OPT_ARGS];
 988         int data_opt = 0;
 989         int option;
 990         kuid_t uid;
 991         kgid_t gid;
 992         char *journal_path;
 993         struct inode *journal_inode;
 994         struct path path;
 995         int error;
 996
 997 #ifdef CONFIG_QUOTA
 998         int qfmt;
 999 #endif
1000
1001         if (!options)
1002                 return 1;
1003
1004         while ((p = strsep (&options, ",")) != NULL) {
1005                 int token;
1006                 if (!*p)
1007                         continue;
1008                 /*
1009                  * Initialize args struct so we know whether arg was
1010                  * found; some options take optional arguments.
1011                  */
1012                 args[0].to = args[0].from = NULL;
1013                 token = match_token(p, tokens, args);
1014                 switch (token) {
1015                 case Opt_bsd_df:
1016                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1017                         break;
1018                 case Opt_minix_df:
1019                         set_opt (sbi->s_mount_opt, MINIX_DF);
1020                         break;
1021                 case Opt_grpid:
1022                         set_opt (sbi->s_mount_opt, GRPID);
1023                         break;
1024                 case Opt_nogrpid:
1025                         clear_opt (sbi->s_mount_opt, GRPID);
1026                         break;
1027                 case Opt_resuid:
1028                         if (match_int(&args[0], &option))
1029                                 return 0;
1030                         uid = make_kuid(current_user_ns(), option);
1031                         if (!uid_valid(uid)) {
1032                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1033                                 return 0;
1034
1035                         }
1036                         sbi->s_resuid = uid;
1037                         break;
1038                 case Opt_resgid:
1039                         if (match_int(&args[0], &option))
1040                                 return 0;
1041                         gid = make_kgid(current_user_ns(), option);
1042                         if (!gid_valid(gid)) {
1043                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1044                                 return 0;
1045                         }
1046                         sbi->s_resgid = gid;
1047                         break;
1048                 case Opt_sb:
1049                         /* handled by get_sb_block() instead of here */
1050                         /* *sb_block = match_int(&args[0]); */
1051                         break;
1052                 case Opt_err_panic:
1053                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1054                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1055                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1056                         break;
1057                 case Opt_err_ro:
1058                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1059                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1060                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1061                         break;
1062                 case Opt_err_cont:
1063                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1064                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1065                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1066                         break;
1067                 case Opt_nouid32:
1068                         set_opt (sbi->s_mount_opt, NO_UID32);
1069                         break;
1070                 case Opt_nocheck:
1071                         clear_opt (sbi->s_mount_opt, CHECK);
1072                         break;
1073                 case Opt_debug:
1074                         set_opt (sbi->s_mount_opt, DEBUG);
1075                         break;
1076                 case Opt_oldalloc:
1077                         ext3_msg(sb, KERN_WARNING,
1078                                 "Ignoring deprecated oldalloc option");
1079                         break;
1080                 case Opt_orlov:
1081                         ext3_msg(sb, KERN_WARNING,
1082                                 "Ignoring deprecated orlov option");
1083                         break;
1084 #ifdef CONFIG_EXT3_FS_XATTR
1085                 case Opt_user_xattr:
1086                         set_opt (sbi->s_mount_opt, XATTR_USER);
1087                         break;
1088                 case Opt_nouser_xattr:
1089                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1090                         break;
1091 #else
1092                 case Opt_user_xattr:
1093                 case Opt_nouser_xattr:
1094                         ext3_msg(sb, KERN_INFO,
1095                                 "(no)user_xattr options not supported");
1096                         break;
1097 #endif
1098 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1099                 case Opt_acl:
1100                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1101                         break;
1102                 case Opt_noacl:
1103                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1104                         break;
1105 #else
1106                 case Opt_acl:
1107                 case Opt_noacl:
1108                         ext3_msg(sb, KERN_INFO,
1109                                 "(no)acl options not supported");
1110                         break;
1111 #endif
1112                 case Opt_reservation:
1113                         set_opt(sbi->s_mount_opt, RESERVATION);
1114                         break;
1115                 case Opt_noreservation:
1116                         clear_opt(sbi->s_mount_opt, RESERVATION);
1117                         break;
1118                 case Opt_journal_update:
1119                         /* @@@ FIXME */
1120                         /* Eventually we will want to be able to create
1121                            a journal file here.  For now, only allow the
1122                            user to specify an existing inode to be the
1123                            journal file. */
1124                         if (is_remount) {
1125                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1126                                         "journal on remount");
1127                                 return 0;
1128                         }
1129                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1130                         break;
1131                 case Opt_journal_inum:
1132                         if (is_remount) {
1133                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1134                                        "journal on remount");
1135                                 return 0;
1136                         }
1137                         if (match_int(&args[0], &option))
1138                                 return 0;
1139                         *inum = option;
1140                         break;
1141                 case Opt_journal_dev:
1142                         if (is_remount) {
1143                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1144                                        "journal on remount");
1145                                 return 0;
1146                         }
1147                         if (match_int(&args[0], &option))
1148                                 return 0;
1149                         *journal_devnum = option;
1150                         break;
1151                 case Opt_journal_path:
1152                         if (is_remount) {
1153                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1154                                        "journal on remount");
1155                                 return 0;
1156                         }
1157
1158                         journal_path = match_strdup(&args[0]);
1159                         if (!journal_path) {
1160                                 ext3_msg(sb, KERN_ERR, "error: could not dup "
1161                                         "journal device string");
1162                                 return 0;
1163                         }
1164
1165                         error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1166                         if (error) {
1167                                 ext3_msg(sb, KERN_ERR, "error: could not find "
1168                                         "journal device path: error %d", error);
1169                                 kfree(journal_path);
1170                                 return 0;
1171                         }
1172
1173                         journal_inode = path.dentry->d_inode;
1174                         if (!S_ISBLK(journal_inode->i_mode)) {
1175                                 ext3_msg(sb, KERN_ERR, "error: journal path %s "
1176                                         "is not a block device", journal_path);
1177                                 path_put(&path);
1178                                 kfree(journal_path);
1179                                 return 0;
1180                         }
1181
1182                         *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1183                         path_put(&path);
1184                         kfree(journal_path);
1185                         break;
1186                 case Opt_noload:
1187                         set_opt (sbi->s_mount_opt, NOLOAD);
1188                         break;
1189                 case Opt_commit:
1190                         if (match_int(&args[0], &option))
1191                                 return 0;
1192                         if (option < 0)
1193                                 return 0;
1194                         if (option == 0)
1195                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1196                         sbi->s_commit_interval = HZ * option;
1197                         break;
1198                 case Opt_data_journal:
1199                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1200                         goto datacheck;
1201                 case Opt_data_ordered:
1202                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1203                         goto datacheck;
1204                 case Opt_data_writeback:
1205                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1206                 datacheck:
1207                         if (is_remount) {
1208                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1209                                         break;
1210                                 ext3_msg(sb, KERN_ERR,
1211                                         "error: cannot change "
1212                                         "data mode on remount. The filesystem "
1213                                         "is mounted in data=%s mode and you "
1214                                         "try to remount it in data=%s mode.",
1215                                         data_mode_string(test_opt(sb,
1216                                                         DATA_FLAGS)),
1217                                         data_mode_string(data_opt));
1218                                 return 0;
1219                         } else {
1220                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1221                                 sbi->s_mount_opt |= data_opt;
1222                         }
1223                         break;
1224                 case Opt_data_err_abort:
1225                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1226                         break;
1227                 case Opt_data_err_ignore:
1228                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1229                         break;
1230 #ifdef CONFIG_QUOTA
1231                 case Opt_usrjquota:
1232                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1233                                 return 0;
1234                         break;
1235                 case Opt_grpjquota:
1236                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1237                                 return 0;
1238                         break;
1239                 case Opt_offusrjquota:
1240                         if (!clear_qf_name(sb, USRQUOTA))
1241                                 return 0;
1242                         break;
1243                 case Opt_offgrpjquota:
1244                         if (!clear_qf_name(sb, GRPQUOTA))
1245                                 return 0;
1246                         break;
1247                 case Opt_jqfmt_vfsold:
1248                         qfmt = QFMT_VFS_OLD;
1249                         goto set_qf_format;
1250                 case Opt_jqfmt_vfsv0:
1251                         qfmt = QFMT_VFS_V0;
1252                         goto set_qf_format;
1253                 case Opt_jqfmt_vfsv1:
1254                         qfmt = QFMT_VFS_V1;
1255 set_qf_format:
1256                         if (sb_any_quota_loaded(sb) &&
1257                             sbi->s_jquota_fmt != qfmt) {
1258                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1259                                         "journaled quota options when "
1260                                         "quota turned on.");
1261                                 return 0;
1262                         }
1263                         sbi->s_jquota_fmt = qfmt;
1264                         break;
1265                 case Opt_quota:
1266                 case Opt_usrquota:
1267                         set_opt(sbi->s_mount_opt, QUOTA);
1268                         set_opt(sbi->s_mount_opt, USRQUOTA);
1269                         break;
1270                 case Opt_grpquota:
1271                         set_opt(sbi->s_mount_opt, QUOTA);
1272                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1273                         break;
1274                 case Opt_noquota:
1275                         if (sb_any_quota_loaded(sb)) {
1276                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1277                                         "quota options when quota turned on.");
1278                                 return 0;
1279                         }
1280                         clear_opt(sbi->s_mount_opt, QUOTA);
1281                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1282                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1283                         break;
1284 #else
1285                 case Opt_quota:
1286                 case Opt_usrquota:
1287                 case Opt_grpquota:
1288                         ext3_msg(sb, KERN_ERR,
1289                                 "error: quota options not supported.");
1290                         break;
1291                 case Opt_usrjquota:
1292                 case Opt_grpjquota:
1293                 case Opt_offusrjquota:
1294                 case Opt_offgrpjquota:
1295                 case Opt_jqfmt_vfsold:
1296                 case Opt_jqfmt_vfsv0:
1297                 case Opt_jqfmt_vfsv1:
1298                         ext3_msg(sb, KERN_ERR,
1299                                 "error: journaled quota options not "
1300                                 "supported.");
1301                         break;
1302                 case Opt_noquota:
1303                         break;
1304 #endif
1305                 case Opt_abort:
1306                         set_opt(sbi->s_mount_opt, ABORT);
1307                         break;
1308                 case Opt_nobarrier:
1309                         clear_opt(sbi->s_mount_opt, BARRIER);
1310                         break;
1311                 case Opt_barrier:
1312                         if (args[0].from) {
1313                                 if (match_int(&args[0], &option))
1314                                         return 0;
1315                         } else
1316                                 option = 1;     /* No argument, default to 1 */
1317                         if (option)
1318                                 set_opt(sbi->s_mount_opt, BARRIER);
1319                         else
1320                                 clear_opt(sbi->s_mount_opt, BARRIER);
1321                         break;
1322                 case Opt_ignore:
1323                         break;
1324                 case Opt_resize:
1325                         if (!is_remount) {
1326                                 ext3_msg(sb, KERN_ERR,
1327                                         "error: resize option only available "
1328                                         "for remount");
1329                                 return 0;
1330                         }
1331                         if (match_int(&args[0], &option) != 0)
1332                                 return 0;
1333                         *n_blocks_count = option;
1334                         break;
1335                 case Opt_nobh:
1336                         ext3_msg(sb, KERN_WARNING,
1337                                 "warning: ignoring deprecated nobh option");
1338                         break;
1339                 case Opt_bh:
1340                         ext3_msg(sb, KERN_WARNING,
1341                                 "warning: ignoring deprecated bh option");
1342                         break;
1343                 default:
1344                         ext3_msg(sb, KERN_ERR,
1345                                 "error: unrecognized mount option \"%s\" "
1346                                 "or missing value", p);
1347                         return 0;
1348                 }
1349         }
1350 #ifdef CONFIG_QUOTA
1351         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1352                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1353                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1354                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1355                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1356
1357                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1358                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1359                                         "format mixing.");
1360                         return 0;
1361                 }
1362
1363                 if (!sbi->s_jquota_fmt) {
1364                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1365                                         "not specified.");
1366                         return 0;
1367                 }
1368         }
1369 #endif
1370         return 1;
1371 }
1372
1373 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1374                             int read_only)
1375 {
1376         struct ext3_sb_info *sbi = EXT3_SB(sb);
1377         int res = 0;
1378
1379         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1380                 ext3_msg(sb, KERN_ERR,
1381                         "error: revision level too high, "
1382                         "forcing read-only mode");
1383                 res = MS_RDONLY;
1384         }
1385         if (read_only)
1386                 return res;
1387         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1388                 ext3_msg(sb, KERN_WARNING,
1389                         "warning: mounting unchecked fs, "
1390                         "running e2fsck is recommended");
1391         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1392                 ext3_msg(sb, KERN_WARNING,
1393                         "warning: mounting fs with errors, "
1394                         "running e2fsck is recommended");
1395         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1396                  le16_to_cpu(es->s_mnt_count) >=
1397                         le16_to_cpu(es->s_max_mnt_count))
1398                 ext3_msg(sb, KERN_WARNING,
1399                         "warning: maximal mount count reached, "
1400                         "running e2fsck is recommended");
1401         else if (le32_to_cpu(es->s_checkinterval) &&
1402                 (le32_to_cpu(es->s_lastcheck) +
1403                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1404                 ext3_msg(sb, KERN_WARNING,
1405                         "warning: checktime reached, "
1406                         "running e2fsck is recommended");
1407 #if 0
1408                 /* @@@ We _will_ want to clear the valid bit if we find
1409                    inconsistencies, to force a fsck at reboot.  But for
1410                    a plain journaled filesystem we can keep it set as
1411                    valid forever! :) */
1412         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1413 #endif
1414         if (!le16_to_cpu(es->s_max_mnt_count))
1415                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1416         le16_add_cpu(&es->s_mnt_count, 1);
1417         es->s_mtime = cpu_to_le32(get_seconds());
1418         ext3_update_dynamic_rev(sb);
1419         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1420
1421         ext3_commit_super(sb, es, 1);
1422         if (test_opt(sb, DEBUG))
1423                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1424                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1425                         sb->s_blocksize,
1426                         sbi->s_groups_count,
1427                         EXT3_BLOCKS_PER_GROUP(sb),
1428                         EXT3_INODES_PER_GROUP(sb),
1429                         sbi->s_mount_opt);
1430
1431         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1432                 char b[BDEVNAME_SIZE];
1433                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1434                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1435         } else {
1436                 ext3_msg(sb, KERN_INFO, "using internal journal");
1437         }
1438         cleancache_init_fs(sb);
1439         return res;
1440 }
1441
1442 /* Called at mount-time, super-block is locked */
1443 static int ext3_check_descriptors(struct super_block *sb)
1444 {
1445         struct ext3_sb_info *sbi = EXT3_SB(sb);
1446         int i;
1447
1448         ext3_debug ("Checking group descriptors");
1449
1450         for (i = 0; i < sbi->s_groups_count; i++) {
1451                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1452                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1453                 ext3_fsblk_t last_block;
1454
1455                 if (i == sbi->s_groups_count - 1)
1456                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1457                 else
1458                         last_block = first_block +
1459                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1460
1461                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1462                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1463                 {
1464                         ext3_error (sb, "ext3_check_descriptors",
1465                                     "Block bitmap for group %d"
1466                                     " not in group (block %lu)!",
1467                                     i, (unsigned long)
1468                                         le32_to_cpu(gdp->bg_block_bitmap));
1469                         return 0;
1470                 }
1471                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1472                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1473                 {
1474                         ext3_error (sb, "ext3_check_descriptors",
1475                                     "Inode bitmap for group %d"
1476                                     " not in group (block %lu)!",
1477                                     i, (unsigned long)
1478                                         le32_to_cpu(gdp->bg_inode_bitmap));
1479                         return 0;
1480                 }
1481                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1482                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1483                     last_block)
1484                 {
1485                         ext3_error (sb, "ext3_check_descriptors",
1486                                     "Inode table for group %d"
1487                                     " not in group (block %lu)!",
1488                                     i, (unsigned long)
1489                                         le32_to_cpu(gdp->bg_inode_table));
1490                         return 0;
1491                 }
1492         }
1493
1494         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1495         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1496         return 1;
1497 }
1498
1499
1500 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1501  * the superblock) which were deleted from all directories, but held open by
1502  * a process at the time of a crash.  We walk the list and try to delete these
1503  * inodes at recovery time (only with a read-write filesystem).
1504  *
1505  * In order to keep the orphan inode chain consistent during traversal (in
1506  * case of crash during recovery), we link each inode into the superblock
1507  * orphan list_head and handle it the same way as an inode deletion during
1508  * normal operation (which journals the operations for us).
1509  *
1510  * We only do an iget() and an iput() on each inode, which is very safe if we
1511  * accidentally point at an in-use or already deleted inode.  The worst that
1512  * can happen in this case is that we get a "bit already cleared" message from
1513  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1514  * e2fsck was run on this filesystem, and it must have already done the orphan
1515  * inode cleanup for us, so we can safely abort without any further action.
1516  */
1517 static void ext3_orphan_cleanup (struct super_block * sb,
1518                                  struct ext3_super_block * es)
1519 {
1520         unsigned int s_flags = sb->s_flags;
1521         int nr_orphans = 0, nr_truncates = 0;
1522 #ifdef CONFIG_QUOTA
1523         int i;
1524 #endif
1525         if (!es->s_last_orphan) {
1526                 jbd_debug(4, "no orphan inodes to clean up\n");
1527                 return;
1528         }
1529
1530         if (bdev_read_only(sb->s_bdev)) {
1531                 ext3_msg(sb, KERN_ERR, "error: write access "
1532                         "unavailable, skipping orphan cleanup.");
1533                 return;
1534         }
1535
1536         /* Check if feature set allows readwrite operations */
1537         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1538                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1539                          "unknown ROCOMPAT features");
1540                 return;
1541         }
1542
1543         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1544                 /* don't clear list on RO mount w/ errors */
1545                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
1546                         jbd_debug(1, "Errors on filesystem, "
1547                                   "clearing orphan list.\n");
1548                         es->s_last_orphan = 0;
1549                 }
1550                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1551                 return;
1552         }
1553
1554         if (s_flags & MS_RDONLY) {
1555                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1556                 sb->s_flags &= ~MS_RDONLY;
1557         }
1558 #ifdef CONFIG_QUOTA
1559         /* Needed for iput() to work correctly and not trash data */
1560         sb->s_flags |= MS_ACTIVE;
1561         /* Turn on quotas so that they are updated correctly */
1562         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
1563                 if (EXT3_SB(sb)->s_qf_names[i]) {
1564                         int ret = ext3_quota_on_mount(sb, i);
1565                         if (ret < 0)
1566                                 ext3_msg(sb, KERN_ERR,
1567                                         "error: cannot turn on journaled "
1568                                         "quota: %d", ret);
1569                 }
1570         }
1571 #endif
1572
1573         while (es->s_last_orphan) {
1574                 struct inode *inode;
1575
1576                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1577                 if (IS_ERR(inode)) {
1578                         es->s_last_orphan = 0;
1579                         break;
1580                 }
1581
1582                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1583                 dquot_initialize(inode);
1584                 if (inode->i_nlink) {
1585                         printk(KERN_DEBUG
1586                                 "%s: truncating inode %lu to %Ld bytes\n",
1587                                 __func__, inode->i_ino, inode->i_size);
1588                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1589                                   inode->i_ino, inode->i_size);
1590                         ext3_truncate(inode);
1591                         nr_truncates++;
1592                 } else {
1593                         printk(KERN_DEBUG
1594                                 "%s: deleting unreferenced inode %lu\n",
1595                                 __func__, inode->i_ino);
1596                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1597                                   inode->i_ino);
1598                         nr_orphans++;
1599                 }
1600                 iput(inode);  /* The delete magic happens here! */
1601         }
1602
1603 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1604
1605         if (nr_orphans)
1606                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1607                        PLURAL(nr_orphans));
1608         if (nr_truncates)
1609                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1610                        PLURAL(nr_truncates));
1611 #ifdef CONFIG_QUOTA
1612         /* Turn quotas off */
1613         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
1614                 if (sb_dqopt(sb)->files[i])
1615                         dquot_quota_off(sb, i);
1616         }
1617 #endif
1618         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1619 }
1620
1621 /*
1622  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1623  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1624  * We need to be 1 filesystem block less than the 2^32 sector limit.
1625  */
1626 static loff_t ext3_max_size(int bits)
1627 {
1628         loff_t res = EXT3_NDIR_BLOCKS;
1629         int meta_blocks;
1630         loff_t upper_limit;
1631
1632         /* This is calculated to be the largest file size for a
1633          * dense, file such that the total number of
1634          * sectors in the file, including data and all indirect blocks,
1635          * does not exceed 2^32 -1
1636          * __u32 i_blocks representing the total number of
1637          * 512 bytes blocks of the file
1638          */
1639         upper_limit = (1LL << 32) - 1;
1640
1641         /* total blocks in file system block size */
1642         upper_limit >>= (bits - 9);
1643
1644
1645         /* indirect blocks */
1646         meta_blocks = 1;
1647         /* double indirect blocks */
1648         meta_blocks += 1 + (1LL << (bits-2));
1649         /* tripple indirect blocks */
1650         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1651
1652         upper_limit -= meta_blocks;
1653         upper_limit <<= bits;
1654
1655         res += 1LL << (bits-2);
1656         res += 1LL << (2*(bits-2));
1657         res += 1LL << (3*(bits-2));
1658         res <<= bits;
1659         if (res > upper_limit)
1660                 res = upper_limit;
1661
1662         if (res > MAX_LFS_FILESIZE)
1663                 res = MAX_LFS_FILESIZE;
1664
1665         return res;
1666 }
1667
1668 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1669                                     ext3_fsblk_t logic_sb_block,
1670                                     int nr)
1671 {
1672         struct ext3_sb_info *sbi = EXT3_SB(sb);
1673         unsigned long bg, first_meta_bg;
1674         int has_super = 0;
1675
1676         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1677
1678         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1679             nr < first_meta_bg)
1680                 return (logic_sb_block + nr + 1);
1681         bg = sbi->s_desc_per_block * nr;
1682         if (ext3_bg_has_super(sb, bg))
1683                 has_super = 1;
1684         return (has_super + ext3_group_first_block_no(sb, bg));
1685 }
1686
1687
1688 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1689 {
1690         struct buffer_head * bh;
1691         struct ext3_super_block *es = NULL;
1692         struct ext3_sb_info *sbi;
1693         ext3_fsblk_t block;
1694         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1695         ext3_fsblk_t logic_sb_block;
1696         unsigned long offset = 0;
1697         unsigned int journal_inum = 0;
1698         unsigned long journal_devnum = 0;
1699         unsigned long def_mount_opts;
1700         struct inode *root;
1701         int blocksize;
1702         int hblock;
1703         int db_count;
1704         int i;
1705         int needs_recovery;
1706         int ret = -EINVAL;
1707         __le32 features;
1708         int err;
1709
1710         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1711         if (!sbi)
1712                 return -ENOMEM;
1713
1714         sbi->s_blockgroup_lock =
1715                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1716         if (!sbi->s_blockgroup_lock) {
1717                 kfree(sbi);
1718                 return -ENOMEM;
1719         }
1720         sb->s_fs_info = sbi;
1721         sbi->s_sb_block = sb_block;
1722
1723         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1724         if (!blocksize) {
1725                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1726                 goto out_fail;
1727         }
1728
1729         /*
1730          * The ext3 superblock will not be buffer aligned for other than 1kB
1731          * block sizes.  We need to calculate the offset from buffer start.
1732          */
1733         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1734                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1735                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1736         } else {
1737                 logic_sb_block = sb_block;
1738         }
1739
1740         if (!(bh = sb_bread(sb, logic_sb_block))) {
1741                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1742                 goto out_fail;
1743         }
1744         /*
1745          * Note: s_es must be initialized as soon as possible because
1746          *       some ext3 macro-instructions depend on its value
1747          */
1748         es = (struct ext3_super_block *) (bh->b_data + offset);
1749         sbi->s_es = es;
1750         sb->s_magic = le16_to_cpu(es->s_magic);
1751         if (sb->s_magic != EXT3_SUPER_MAGIC)
1752                 goto cantfind_ext3;
1753
1754         /* Set defaults before we parse the mount options */
1755         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1756         if (def_mount_opts & EXT3_DEFM_DEBUG)
1757                 set_opt(sbi->s_mount_opt, DEBUG);
1758         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1759                 set_opt(sbi->s_mount_opt, GRPID);
1760         if (def_mount_opts & EXT3_DEFM_UID16)
1761                 set_opt(sbi->s_mount_opt, NO_UID32);
1762 #ifdef CONFIG_EXT3_FS_XATTR
1763         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1764                 set_opt(sbi->s_mount_opt, XATTR_USER);
1765 #endif
1766 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1767         if (def_mount_opts & EXT3_DEFM_ACL)
1768                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1769 #endif
1770         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1771                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1772         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1773                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1774         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1775                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1776
1777         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1778                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1779         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1780                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1781         else
1782                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1783
1784         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1785         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1786
1787         /* enable barriers by default */
1788         set_opt(sbi->s_mount_opt, BARRIER);
1789         set_opt(sbi->s_mount_opt, RESERVATION);
1790
1791         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1792                             NULL, 0))
1793                 goto failed_mount;
1794
1795         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1796                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1797
1798         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1799             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1800              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1801              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1802                 ext3_msg(sb, KERN_WARNING,
1803                         "warning: feature flags set on rev 0 fs, "
1804                         "running e2fsck is recommended");
1805         /*
1806          * Check feature flags regardless of the revision level, since we
1807          * previously didn't change the revision level when setting the flags,
1808          * so there is a chance incompat flags are set on a rev 0 filesystem.
1809          */
1810         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1811         if (features) {
1812                 ext3_msg(sb, KERN_ERR,
1813                         "error: couldn't mount because of unsupported "
1814                         "optional features (%x)", le32_to_cpu(features));
1815                 goto failed_mount;
1816         }
1817         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1818         if (!(sb->s_flags & MS_RDONLY) && features) {
1819                 ext3_msg(sb, KERN_ERR,
1820                         "error: couldn't mount RDWR because of unsupported "
1821                         "optional features (%x)", le32_to_cpu(features));
1822                 goto failed_mount;
1823         }
1824         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1825
1826         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1827             blocksize > EXT3_MAX_BLOCK_SIZE) {
1828                 ext3_msg(sb, KERN_ERR,
1829                         "error: couldn't mount because of unsupported "
1830                         "filesystem blocksize %d", blocksize);
1831                 goto failed_mount;
1832         }
1833
1834         hblock = bdev_logical_block_size(sb->s_bdev);
1835         if (sb->s_blocksize != blocksize) {
1836                 /*
1837                  * Make sure the blocksize for the filesystem is larger
1838                  * than the hardware sectorsize for the machine.
1839                  */
1840                 if (blocksize < hblock) {
1841                         ext3_msg(sb, KERN_ERR,
1842                                 "error: fsblocksize %d too small for "
1843                                 "hardware sectorsize %d", blocksize, hblock);
1844                         goto failed_mount;
1845                 }
1846
1847                 brelse (bh);
1848                 if (!sb_set_blocksize(sb, blocksize)) {
1849                         ext3_msg(sb, KERN_ERR,
1850                                 "error: bad blocksize %d", blocksize);
1851                         goto out_fail;
1852                 }
1853                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1854                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1855                 bh = sb_bread(sb, logic_sb_block);
1856                 if (!bh) {
1857                         ext3_msg(sb, KERN_ERR,
1858                                "error: can't read superblock on 2nd try");
1859                         goto failed_mount;
1860                 }
1861                 es = (struct ext3_super_block *)(bh->b_data + offset);
1862                 sbi->s_es = es;
1863                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1864                         ext3_msg(sb, KERN_ERR,
1865                                 "error: magic mismatch");
1866                         goto failed_mount;
1867                 }
1868         }
1869
1870         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1871
1872         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1873                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1874                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1875         } else {
1876                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1877                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1878                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1879                     (!is_power_of_2(sbi->s_inode_size)) ||
1880                     (sbi->s_inode_size > blocksize)) {
1881                         ext3_msg(sb, KERN_ERR,
1882                                 "error: unsupported inode size: %d",
1883                                 sbi->s_inode_size);
1884                         goto failed_mount;
1885                 }
1886         }
1887         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1888                                    le32_to_cpu(es->s_log_frag_size);
1889         if (blocksize != sbi->s_frag_size) {
1890                 ext3_msg(sb, KERN_ERR,
1891                        "error: fragsize %lu != blocksize %u (unsupported)",
1892                        sbi->s_frag_size, blocksize);
1893                 goto failed_mount;
1894         }
1895         sbi->s_frags_per_block = 1;
1896         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1897         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1898         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1899         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1900                 goto cantfind_ext3;
1901         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1902         if (sbi->s_inodes_per_block == 0)
1903                 goto cantfind_ext3;
1904         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1905                                         sbi->s_inodes_per_block;
1906         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1907         sbi->s_sbh = bh;
1908         sbi->s_mount_state = le16_to_cpu(es->s_state);
1909         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1910         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1911         for (i=0; i < 4; i++)
1912                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1913         sbi->s_def_hash_version = es->s_def_hash_version;
1914         i = le32_to_cpu(es->s_flags);
1915         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1916                 sbi->s_hash_unsigned = 3;
1917         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1918 #ifdef __CHAR_UNSIGNED__
1919                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1920                 sbi->s_hash_unsigned = 3;
1921 #else
1922                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1923 #endif
1924         }
1925
1926         if (sbi->s_blocks_per_group > blocksize * 8) {
1927                 ext3_msg(sb, KERN_ERR,
1928                         "#blocks per group too big: %lu",
1929                         sbi->s_blocks_per_group);
1930                 goto failed_mount;
1931         }
1932         if (sbi->s_frags_per_group > blocksize * 8) {
1933                 ext3_msg(sb, KERN_ERR,
1934                         "error: #fragments per group too big: %lu",
1935                         sbi->s_frags_per_group);
1936                 goto failed_mount;
1937         }
1938         if (sbi->s_inodes_per_group > blocksize * 8) {
1939                 ext3_msg(sb, KERN_ERR,
1940                         "error: #inodes per group too big: %lu",
1941                         sbi->s_inodes_per_group);
1942                 goto failed_mount;
1943         }
1944
1945         err = generic_check_addressable(sb->s_blocksize_bits,
1946                                         le32_to_cpu(es->s_blocks_count));
1947         if (err) {
1948                 ext3_msg(sb, KERN_ERR,
1949                         "error: filesystem is too large to mount safely");
1950                 if (sizeof(sector_t) < 8)
1951                         ext3_msg(sb, KERN_ERR,
1952                                 "error: CONFIG_LBDAF not enabled");
1953                 ret = err;
1954                 goto failed_mount;
1955         }
1956
1957         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1958                 goto cantfind_ext3;
1959         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1960                                le32_to_cpu(es->s_first_data_block) - 1)
1961                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1962         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1963         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1964                                     GFP_KERNEL);
1965         if (sbi->s_group_desc == NULL) {
1966                 ext3_msg(sb, KERN_ERR,
1967                         "error: not enough memory");
1968                 ret = -ENOMEM;
1969                 goto failed_mount;
1970         }
1971
1972         bgl_lock_init(sbi->s_blockgroup_lock);
1973
1974         for (i = 0; i < db_count; i++) {
1975                 block = descriptor_loc(sb, logic_sb_block, i);
1976                 sbi->s_group_desc[i] = sb_bread(sb, block);
1977                 if (!sbi->s_group_desc[i]) {
1978                         ext3_msg(sb, KERN_ERR,
1979                                 "error: can't read group descriptor %d", i);
1980                         db_count = i;
1981                         goto failed_mount2;
1982                 }
1983         }
1984         if (!ext3_check_descriptors (sb)) {
1985                 ext3_msg(sb, KERN_ERR,
1986                         "error: group descriptors corrupted");
1987                 goto failed_mount2;
1988         }
1989         sbi->s_gdb_count = db_count;
1990         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1991         spin_lock_init(&sbi->s_next_gen_lock);
1992
1993         /* per fileystem reservation list head & lock */
1994         spin_lock_init(&sbi->s_rsv_window_lock);
1995         sbi->s_rsv_window_root = RB_ROOT;
1996         /* Add a single, static dummy reservation to the start of the
1997          * reservation window list --- it gives us a placeholder for
1998          * append-at-start-of-list which makes the allocation logic
1999          * _much_ simpler. */
2000         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
2001         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
2002         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2003         sbi->s_rsv_window_head.rsv_goal_size = 0;
2004         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
2005
2006         /*
2007          * set up enough so that it can read an inode
2008          */
2009         sb->s_op = &ext3_sops;
2010         sb->s_export_op = &ext3_export_ops;
2011         sb->s_xattr = ext3_xattr_handlers;
2012 #ifdef CONFIG_QUOTA
2013         sb->s_qcop = &ext3_qctl_operations;
2014         sb->dq_op = &ext3_quota_operations;
2015         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2016 #endif
2017         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
2018         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2019         mutex_init(&sbi->s_orphan_lock);
2020         mutex_init(&sbi->s_resize_lock);
2021
2022         sb->s_root = NULL;
2023
2024         needs_recovery = (es->s_last_orphan != 0 ||
2025                           EXT3_HAS_INCOMPAT_FEATURE(sb,
2026                                     EXT3_FEATURE_INCOMPAT_RECOVER));
2027
2028         /*
2029          * The first inode we look at is the journal inode.  Don't try
2030          * root first: it may be modified in the journal!
2031          */
2032         if (!test_opt(sb, NOLOAD) &&
2033             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
2034                 if (ext3_load_journal(sb, es, journal_devnum))
2035                         goto failed_mount2;
2036         } else if (journal_inum) {
2037                 if (ext3_create_journal(sb, es, journal_inum))
2038                         goto failed_mount2;
2039         } else {
2040                 if (!silent)
2041                         ext3_msg(sb, KERN_ERR,
2042                                 "error: no journal found. "
2043                                 "mounting ext3 over ext2?");
2044                 goto failed_mount2;
2045         }
2046         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2047                         ext3_count_free_blocks(sb), GFP_KERNEL);
2048         if (!err) {
2049                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2050                                 ext3_count_free_inodes(sb), GFP_KERNEL);
2051         }
2052         if (!err) {
2053                 err = percpu_counter_init(&sbi->s_dirs_counter,
2054                                 ext3_count_dirs(sb), GFP_KERNEL);
2055         }
2056         if (err) {
2057                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2058                 ret = err;
2059                 goto failed_mount3;
2060         }
2061
2062         /* We have now updated the journal if required, so we can
2063          * validate the data journaling mode. */
2064         switch (test_opt(sb, DATA_FLAGS)) {
2065         case 0:
2066                 /* No mode set, assume a default based on the journal
2067                    capabilities: ORDERED_DATA if the journal can
2068                    cope, else JOURNAL_DATA */
2069                 if (journal_check_available_features
2070                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2071                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2072                 else
2073                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2074                 break;
2075
2076         case EXT3_MOUNT_ORDERED_DATA:
2077         case EXT3_MOUNT_WRITEBACK_DATA:
2078                 if (!journal_check_available_features
2079                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2080                         ext3_msg(sb, KERN_ERR,
2081                                 "error: journal does not support "
2082                                 "requested data journaling mode");
2083                         goto failed_mount3;
2084                 }
2085         default:
2086                 break;
2087         }
2088
2089         /*
2090          * The journal_load will have done any necessary log recovery,
2091          * so we can safely mount the rest of the filesystem now.
2092          */
2093
2094         root = ext3_iget(sb, EXT3_ROOT_INO);
2095         if (IS_ERR(root)) {
2096                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2097                 ret = PTR_ERR(root);
2098                 goto failed_mount3;
2099         }
2100         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2101                 iput(root);
2102                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2103                 goto failed_mount3;
2104         }
2105         sb->s_root = d_make_root(root);
2106         if (!sb->s_root) {
2107                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2108                 ret = -ENOMEM;
2109                 goto failed_mount3;
2110         }
2111
2112         if (ext3_setup_super(sb, es, sb->s_flags & MS_RDONLY))
2113                 sb->s_flags |= MS_RDONLY;
2114
2115         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2116         ext3_orphan_cleanup(sb, es);
2117         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2118         if (needs_recovery) {
2119                 ext3_mark_recovery_complete(sb, es);
2120                 ext3_msg(sb, KERN_INFO, "recovery complete");
2121         }
2122         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2123                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2124                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2125                 "writeback");
2126
2127         return 0;
2128
2129 cantfind_ext3:
2130         if (!silent)
2131                 ext3_msg(sb, KERN_INFO,
2132                         "error: can't find ext3 filesystem on dev %s.",
2133                        sb->s_id);
2134         goto failed_mount;
2135
2136 failed_mount3:
2137         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2138         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2139         percpu_counter_destroy(&sbi->s_dirs_counter);
2140         journal_destroy(sbi->s_journal);
2141 failed_mount2:
2142         for (i = 0; i < db_count; i++)
2143                 brelse(sbi->s_group_desc[i]);
2144         kfree(sbi->s_group_desc);
2145 failed_mount:
2146 #ifdef CONFIG_QUOTA
2147         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2148                 kfree(sbi->s_qf_names[i]);
2149 #endif
2150         ext3_blkdev_remove(sbi);
2151         brelse(bh);
2152 out_fail:
2153         sb->s_fs_info = NULL;
2154         kfree(sbi->s_blockgroup_lock);
2155         kfree(sbi);
2156         return ret;
2157 }
2158
2159 /*
2160  * Setup any per-fs journal parameters now.  We'll do this both on
2161  * initial mount, once the journal has been initialised but before we've
2162  * done any recovery; and again on any subsequent remount.
2163  */
2164 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2165 {
2166         struct ext3_sb_info *sbi = EXT3_SB(sb);
2167
2168         if (sbi->s_commit_interval)
2169                 journal->j_commit_interval = sbi->s_commit_interval;
2170         /* We could also set up an ext3-specific default for the commit
2171          * interval here, but for now we'll just fall back to the jbd
2172          * default. */
2173
2174         spin_lock(&journal->j_state_lock);
2175         if (test_opt(sb, BARRIER))
2176                 journal->j_flags |= JFS_BARRIER;
2177         else
2178                 journal->j_flags &= ~JFS_BARRIER;
2179         if (test_opt(sb, DATA_ERR_ABORT))
2180                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2181         else
2182                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2183         spin_unlock(&journal->j_state_lock);
2184 }
2185
2186 static journal_t *ext3_get_journal(struct super_block *sb,
2187                                    unsigned int journal_inum)
2188 {
2189         struct inode *journal_inode;
2190         journal_t *journal;
2191
2192         /* First, test for the existence of a valid inode on disk.  Bad
2193          * things happen if we iget() an unused inode, as the subsequent
2194          * iput() will try to delete it. */
2195
2196         journal_inode = ext3_iget(sb, journal_inum);
2197         if (IS_ERR(journal_inode)) {
2198                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2199                 return NULL;
2200         }
2201         if (!journal_inode->i_nlink) {
2202                 make_bad_inode(journal_inode);
2203                 iput(journal_inode);
2204                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2205                 return NULL;
2206         }
2207
2208         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2209                   journal_inode, journal_inode->i_size);
2210         if (!S_ISREG(journal_inode->i_mode)) {
2211                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2212                 iput(journal_inode);
2213                 return NULL;
2214         }
2215
2216         journal = journal_init_inode(journal_inode);
2217         if (!journal) {
2218                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2219                 iput(journal_inode);
2220                 return NULL;
2221         }
2222         journal->j_private = sb;
2223         ext3_init_journal_params(sb, journal);
2224         return journal;
2225 }
2226
2227 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2228                                        dev_t j_dev)
2229 {
2230         struct buffer_head * bh;
2231         journal_t *journal;
2232         ext3_fsblk_t start;
2233         ext3_fsblk_t len;
2234         int hblock, blocksize;
2235         ext3_fsblk_t sb_block;
2236         unsigned long offset;
2237         struct ext3_super_block * es;
2238         struct block_device *bdev;
2239
2240         bdev = ext3_blkdev_get(j_dev, sb);
2241         if (bdev == NULL)
2242                 return NULL;
2243
2244         blocksize = sb->s_blocksize;
2245         hblock = bdev_logical_block_size(bdev);
2246         if (blocksize < hblock) {
2247                 ext3_msg(sb, KERN_ERR,
2248                         "error: blocksize too small for journal device");
2249                 goto out_bdev;
2250         }
2251
2252         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2253         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2254         set_blocksize(bdev, blocksize);
2255         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2256                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2257                         "external journal");
2258                 goto out_bdev;
2259         }
2260
2261         es = (struct ext3_super_block *) (bh->b_data + offset);
2262         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2263             !(le32_to_cpu(es->s_feature_incompat) &
2264               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2265                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2266                         "bad superblock");
2267                 brelse(bh);
2268                 goto out_bdev;
2269         }
2270
2271         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2272                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2273                 brelse(bh);
2274                 goto out_bdev;
2275         }
2276
2277         len = le32_to_cpu(es->s_blocks_count);
2278         start = sb_block + 1;
2279         brelse(bh);     /* we're done with the superblock */
2280
2281         journal = journal_init_dev(bdev, sb->s_bdev,
2282                                         start, len, blocksize);
2283         if (!journal) {
2284                 ext3_msg(sb, KERN_ERR,
2285                         "error: failed to create device journal");
2286                 goto out_bdev;
2287         }
2288         journal->j_private = sb;
2289         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2290                 if (bh_submit_read(journal->j_sb_buffer)) {
2291                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2292                         goto out_journal;
2293                 }
2294         }
2295         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2296                 ext3_msg(sb, KERN_ERR,
2297                         "error: external journal has more than one "
2298                         "user (unsupported) - %d",
2299                         be32_to_cpu(journal->j_superblock->s_nr_users));
2300                 goto out_journal;
2301         }
2302         EXT3_SB(sb)->journal_bdev = bdev;
2303         ext3_init_journal_params(sb, journal);
2304         return journal;
2305 out_journal:
2306         journal_destroy(journal);
2307 out_bdev:
2308         ext3_blkdev_put(bdev);
2309         return NULL;
2310 }
2311
2312 static int ext3_load_journal(struct super_block *sb,
2313                              struct ext3_super_block *es,
2314                              unsigned long journal_devnum)
2315 {
2316         journal_t *journal;
2317         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2318         dev_t journal_dev;
2319         int err = 0;
2320         int really_read_only;
2321
2322         if (journal_devnum &&
2323             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2324                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2325                         "numbers have changed");
2326                 journal_dev = new_decode_dev(journal_devnum);
2327         } else
2328                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2329
2330         really_read_only = bdev_read_only(sb->s_bdev);
2331
2332         /*
2333          * Are we loading a blank journal or performing recovery after a
2334          * crash?  For recovery, we need to check in advance whether we
2335          * can get read-write access to the device.
2336          */
2337
2338         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2339                 if (sb->s_flags & MS_RDONLY) {
2340                         ext3_msg(sb, KERN_INFO,
2341                                 "recovery required on readonly filesystem");
2342                         if (really_read_only) {
2343                                 ext3_msg(sb, KERN_ERR, "error: write access "
2344                                         "unavailable, cannot proceed");
2345                                 return -EROFS;
2346                         }
2347                         ext3_msg(sb, KERN_INFO,
2348                                 "write access will be enabled during recovery");
2349                 }
2350         }
2351
2352         if (journal_inum && journal_dev) {
2353                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2354                        "and inode journals");
2355                 return -EINVAL;
2356         }
2357
2358         if (journal_inum) {
2359                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2360                         return -EINVAL;
2361         } else {
2362                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2363                         return -EINVAL;
2364         }
2365
2366         if (!(journal->j_flags & JFS_BARRIER))
2367                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2368
2369         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2370                 err = journal_update_format(journal);
2371                 if (err)  {
2372                         ext3_msg(sb, KERN_ERR, "error updating journal");
2373                         journal_destroy(journal);
2374                         return err;
2375                 }
2376         }
2377
2378         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2379                 err = journal_wipe(journal, !really_read_only);
2380         if (!err)
2381                 err = journal_load(journal);
2382
2383         if (err) {
2384                 ext3_msg(sb, KERN_ERR, "error loading journal");
2385                 journal_destroy(journal);
2386                 return err;
2387         }
2388
2389         EXT3_SB(sb)->s_journal = journal;
2390         ext3_clear_journal_err(sb, es);
2391
2392         if (!really_read_only && journal_devnum &&
2393             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2394                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2395
2396                 /* Make sure we flush the recovery flag to disk. */
2397                 ext3_commit_super(sb, es, 1);
2398         }
2399
2400         return 0;
2401 }
2402
2403 static int ext3_create_journal(struct super_block *sb,
2404                                struct ext3_super_block *es,
2405                                unsigned int journal_inum)
2406 {
2407         journal_t *journal;
2408         int err;
2409
2410         if (sb->s_flags & MS_RDONLY) {
2411                 ext3_msg(sb, KERN_ERR,
2412                         "error: readonly filesystem when trying to "
2413                         "create journal");
2414                 return -EROFS;
2415         }
2416
2417         journal = ext3_get_journal(sb, journal_inum);
2418         if (!journal)
2419                 return -EINVAL;
2420
2421         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2422                journal_inum);
2423
2424         err = journal_create(journal);
2425         if (err) {
2426                 ext3_msg(sb, KERN_ERR, "error creating journal");
2427                 journal_destroy(journal);
2428                 return -EIO;
2429         }
2430
2431         EXT3_SB(sb)->s_journal = journal;
2432
2433         ext3_update_dynamic_rev(sb);
2434         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2435         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2436
2437         es->s_journal_inum = cpu_to_le32(journal_inum);
2438
2439         /* Make sure we flush the recovery flag to disk. */
2440         ext3_commit_super(sb, es, 1);
2441
2442         return 0;
2443 }
2444
2445 static int ext3_commit_super(struct super_block *sb,
2446                                struct ext3_super_block *es,
2447                                int sync)
2448 {
2449         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2450         int error = 0;
2451
2452         if (!sbh)
2453                 return error;
2454
2455         if (buffer_write_io_error(sbh)) {
2456                 /*
2457                  * Oh, dear.  A previous attempt to write the
2458                  * superblock failed.  This could happen because the
2459                  * USB device was yanked out.  Or it could happen to
2460                  * be a transient write error and maybe the block will
2461                  * be remapped.  Nothing we can do but to retry the
2462                  * write and hope for the best.
2463                  */
2464                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2465                        "superblock detected");
2466                 clear_buffer_write_io_error(sbh);
2467                 set_buffer_uptodate(sbh);
2468         }
2469         /*
2470          * If the file system is mounted read-only, don't update the
2471          * superblock write time.  This avoids updating the superblock
2472          * write time when we are mounting the root file system
2473          * read/only but we need to replay the journal; at that point,
2474          * for people who are east of GMT and who make their clock
2475          * tick in localtime for Windows bug-for-bug compatibility,
2476          * the clock is set in the future, and this will cause e2fsck
2477          * to complain and force a full file system check.
2478          */
2479         if (!(sb->s_flags & MS_RDONLY))
2480                 es->s_wtime = cpu_to_le32(get_seconds());
2481         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2482         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2483         BUFFER_TRACE(sbh, "marking dirty");
2484         mark_buffer_dirty(sbh);
2485         if (sync) {
2486                 error = sync_dirty_buffer(sbh);
2487                 if (buffer_write_io_error(sbh)) {
2488                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2489                                "superblock");
2490                         clear_buffer_write_io_error(sbh);
2491                         set_buffer_uptodate(sbh);
2492                 }
2493         }
2494         return error;
2495 }
2496
2497
2498 /*
2499  * Have we just finished recovery?  If so, and if we are mounting (or
2500  * remounting) the filesystem readonly, then we will end up with a
2501  * consistent fs on disk.  Record that fact.
2502  */
2503 static void ext3_mark_recovery_complete(struct super_block * sb,
2504                                         struct ext3_super_block * es)
2505 {
2506         journal_t *journal = EXT3_SB(sb)->s_journal;
2507
2508         journal_lock_updates(journal);
2509         if (journal_flush(journal) < 0)
2510                 goto out;
2511
2512         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2513             sb->s_flags & MS_RDONLY) {
2514                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2515                 ext3_commit_super(sb, es, 1);
2516         }
2517
2518 out:
2519         journal_unlock_updates(journal);
2520 }
2521
2522 /*
2523  * If we are mounting (or read-write remounting) a filesystem whose journal
2524  * has recorded an error from a previous lifetime, move that error to the
2525  * main filesystem now.
2526  */
2527 static void ext3_clear_journal_err(struct super_block *sb,
2528                                    struct ext3_super_block *es)
2529 {
2530         journal_t *journal;
2531         int j_errno;
2532         const char *errstr;
2533
2534         journal = EXT3_SB(sb)->s_journal;
2535
2536         /*
2537          * Now check for any error status which may have been recorded in the
2538          * journal by a prior ext3_error() or ext3_abort()
2539          */
2540
2541         j_errno = journal_errno(journal);
2542         if (j_errno) {
2543                 char nbuf[16];
2544
2545                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2546                 ext3_warning(sb, __func__, "Filesystem error recorded "
2547                              "from previous mount: %s", errstr);
2548                 ext3_warning(sb, __func__, "Marking fs in need of "
2549                              "filesystem check.");
2550
2551                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2552                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2553                 ext3_commit_super (sb, es, 1);
2554
2555                 journal_clear_err(journal);
2556         }
2557 }
2558
2559 /*
2560  * Force the running and committing transactions to commit,
2561  * and wait on the commit.
2562  */
2563 int ext3_force_commit(struct super_block *sb)
2564 {
2565         journal_t *journal;
2566         int ret;
2567
2568         if (sb->s_flags & MS_RDONLY)
2569                 return 0;
2570
2571         journal = EXT3_SB(sb)->s_journal;
2572         ret = ext3_journal_force_commit(journal);
2573         return ret;
2574 }
2575
2576 static int ext3_sync_fs(struct super_block *sb, int wait)
2577 {
2578         tid_t target;
2579
2580         trace_ext3_sync_fs(sb, wait);
2581         /*
2582          * Writeback quota in non-journalled quota case - journalled quota has
2583          * no dirty dquots
2584          */
2585         dquot_writeback_dquots(sb, -1);
2586         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2587                 if (wait)
2588                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2589         }
2590         return 0;
2591 }
2592
2593 /*
2594  * LVM calls this function before a (read-only) snapshot is created.  This
2595  * gives us a chance to flush the journal completely and mark the fs clean.
2596  */
2597 static int ext3_freeze(struct super_block *sb)
2598 {
2599         int error = 0;
2600         journal_t *journal;
2601
2602         if (!(sb->s_flags & MS_RDONLY)) {
2603                 journal = EXT3_SB(sb)->s_journal;
2604
2605                 /* Now we set up the journal barrier. */
2606                 journal_lock_updates(journal);
2607
2608                 /*
2609                  * We don't want to clear needs_recovery flag when we failed
2610                  * to flush the journal.
2611                  */
2612                 error = journal_flush(journal);
2613                 if (error < 0)
2614                         goto out;
2615
2616                 /* Journal blocked and flushed, clear needs_recovery flag. */
2617                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2618                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2619                 if (error)
2620                         goto out;
2621         }
2622         return 0;
2623
2624 out:
2625         journal_unlock_updates(journal);
2626         return error;
2627 }
2628
2629 /*
2630  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2631  * flag here, even though the filesystem is not technically dirty yet.
2632  */
2633 static int ext3_unfreeze(struct super_block *sb)
2634 {
2635         if (!(sb->s_flags & MS_RDONLY)) {
2636                 /* Reser the needs_recovery flag before the fs is unlocked. */
2637                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2638                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2639                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2640         }
2641         return 0;
2642 }
2643
2644 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2645 {
2646         struct ext3_super_block * es;
2647         struct ext3_sb_info *sbi = EXT3_SB(sb);
2648         ext3_fsblk_t n_blocks_count = 0;
2649         unsigned long old_sb_flags;
2650         struct ext3_mount_options old_opts;
2651         int enable_quota = 0;
2652         int err;
2653 #ifdef CONFIG_QUOTA
2654         int i;
2655 #endif
2656
2657         sync_filesystem(sb);
2658
2659         /* Store the original options */
2660         old_sb_flags = sb->s_flags;
2661         old_opts.s_mount_opt = sbi->s_mount_opt;
2662         old_opts.s_resuid = sbi->s_resuid;
2663         old_opts.s_resgid = sbi->s_resgid;
2664         old_opts.s_commit_interval = sbi->s_commit_interval;
2665 #ifdef CONFIG_QUOTA
2666         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2667         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2668                 if (sbi->s_qf_names[i]) {
2669                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
2670                                                          GFP_KERNEL);
2671                         if (!old_opts.s_qf_names[i]) {
2672                                 int j;
2673
2674                                 for (j = 0; j < i; j++)
2675                                         kfree(old_opts.s_qf_names[j]);
2676                                 return -ENOMEM;
2677                         }
2678                 } else
2679                         old_opts.s_qf_names[i] = NULL;
2680 #endif
2681
2682         /*
2683          * Allow the "check" option to be passed as a remount option.
2684          */
2685         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2686                 err = -EINVAL;
2687                 goto restore_opts;
2688         }
2689
2690         if (test_opt(sb, ABORT))
2691                 ext3_abort(sb, __func__, "Abort forced by user");
2692
2693         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2694                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2695
2696         es = sbi->s_es;
2697
2698         ext3_init_journal_params(sb, sbi->s_journal);
2699
2700         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2701                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2702                 if (test_opt(sb, ABORT)) {
2703                         err = -EROFS;
2704                         goto restore_opts;
2705                 }
2706
2707                 if (*flags & MS_RDONLY) {
2708                         err = dquot_suspend(sb, -1);
2709                         if (err < 0)
2710                                 goto restore_opts;
2711
2712                         /*
2713                          * First of all, the unconditional stuff we have to do
2714                          * to disable replay of the journal when we next remount
2715                          */
2716                         sb->s_flags |= MS_RDONLY;
2717
2718                         /*
2719                          * OK, test if we are remounting a valid rw partition
2720                          * readonly, and if so set the rdonly flag and then
2721                          * mark the partition as valid again.
2722                          */
2723                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2724                             (sbi->s_mount_state & EXT3_VALID_FS))
2725                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2726
2727                         ext3_mark_recovery_complete(sb, es);
2728                 } else {
2729                         __le32 ret;
2730                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2731                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2732                                 ext3_msg(sb, KERN_WARNING,
2733                                         "warning: couldn't remount RDWR "
2734                                         "because of unsupported optional "
2735                                         "features (%x)", le32_to_cpu(ret));
2736                                 err = -EROFS;
2737                                 goto restore_opts;
2738                         }
2739
2740                         /*
2741                          * If we have an unprocessed orphan list hanging
2742                          * around from a previously readonly bdev mount,
2743                          * require a full umount & mount for now.
2744                          */
2745                         if (es->s_last_orphan) {
2746                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2747                                        "remount RDWR because of unprocessed "
2748                                        "orphan inode list.  Please "
2749                                        "umount & mount instead.");
2750                                 err = -EINVAL;
2751                                 goto restore_opts;
2752                         }
2753
2754                         /*
2755                          * Mounting a RDONLY partition read-write, so reread
2756                          * and store the current valid flag.  (It may have
2757                          * been changed by e2fsck since we originally mounted
2758                          * the partition.)
2759                          */
2760                         ext3_clear_journal_err(sb, es);
2761                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2762                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2763                                 goto restore_opts;
2764                         if (!ext3_setup_super (sb, es, 0))
2765                                 sb->s_flags &= ~MS_RDONLY;
2766                         enable_quota = 1;
2767                 }
2768         }
2769 #ifdef CONFIG_QUOTA
2770         /* Release old quota file names */
2771         for (i = 0; i < EXT3_MAXQUOTAS; i++)
2772                 kfree(old_opts.s_qf_names[i]);
2773 #endif
2774         if (enable_quota)
2775                 dquot_resume(sb, -1);
2776         return 0;
2777 restore_opts:
2778         sb->s_flags = old_sb_flags;
2779         sbi->s_mount_opt = old_opts.s_mount_opt;
2780         sbi->s_resuid = old_opts.s_resuid;
2781         sbi->s_resgid = old_opts.s_resgid;
2782         sbi->s_commit_interval = old_opts.s_commit_interval;
2783 #ifdef CONFIG_QUOTA
2784         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2785         for (i = 0; i < EXT3_MAXQUOTAS; i++) {
2786                 kfree(sbi->s_qf_names[i]);
2787                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2788         }
2789 #endif
2790         return err;
2791 }
2792
2793 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2794 {
2795         struct super_block *sb = dentry->d_sb;
2796         struct ext3_sb_info *sbi = EXT3_SB(sb);
2797         struct ext3_super_block *es = sbi->s_es;
2798         u64 fsid;
2799
2800         if (test_opt(sb, MINIX_DF)) {
2801                 sbi->s_overhead_last = 0;
2802         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2803                 unsigned long ngroups = sbi->s_groups_count, i;
2804                 ext3_fsblk_t overhead = 0;
2805                 smp_rmb();
2806
2807                 /*
2808                  * Compute the overhead (FS structures).  This is constant
2809                  * for a given filesystem unless the number of block groups
2810                  * changes so we cache the previous value until it does.
2811                  */
2812
2813                 /*
2814                  * All of the blocks before first_data_block are
2815                  * overhead
2816                  */
2817                 overhead = le32_to_cpu(es->s_first_data_block);
2818
2819                 /*
2820                  * Add the overhead attributed to the superblock and
2821                  * block group descriptors.  If the sparse superblocks
2822                  * feature is turned on, then not all groups have this.
2823                  */
2824                 for (i = 0; i < ngroups; i++) {
2825                         overhead += ext3_bg_has_super(sb, i) +
2826                                 ext3_bg_num_gdb(sb, i);
2827                         cond_resched();
2828                 }
2829
2830                 /*
2831                  * Every block group has an inode bitmap, a block
2832                  * bitmap, and an inode table.
2833                  */
2834                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2835
2836                 /* Add the internal journal blocks as well */
2837                 if (sbi->s_journal && !sbi->journal_bdev)
2838                         overhead += sbi->s_journal->j_maxlen;
2839
2840                 sbi->s_overhead_last = overhead;
2841                 smp_wmb();
2842                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2843         }
2844
2845         buf->f_type = EXT3_SUPER_MAGIC;
2846         buf->f_bsize = sb->s_blocksize;
2847         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2848         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2849         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2850         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2851                 buf->f_bavail = 0;
2852         buf->f_files = le32_to_cpu(es->s_inodes_count);
2853         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2854         buf->f_namelen = EXT3_NAME_LEN;
2855         fsid = le64_to_cpup((void *)es->s_uuid) ^
2856                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2857         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2858         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2859         return 0;
2860 }
2861
2862 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2863  * is locked for write. Otherwise the are possible deadlocks:
2864  * Process 1                         Process 2
2865  * ext3_create()                     quota_sync()
2866  *   journal_start()                   write_dquot()
2867  *   dquot_initialize()                       down(dqio_mutex)
2868  *     down(dqio_mutex)                    journal_start()
2869  *
2870  */
2871
2872 #ifdef CONFIG_QUOTA
2873
2874 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2875 {
2876         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
2877 }
2878
2879 static int ext3_write_dquot(struct dquot *dquot)
2880 {
2881         int ret, err;
2882         handle_t *handle;
2883         struct inode *inode;
2884
2885         inode = dquot_to_inode(dquot);
2886         handle = ext3_journal_start(inode,
2887                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2888         if (IS_ERR(handle))
2889                 return PTR_ERR(handle);
2890         ret = dquot_commit(dquot);
2891         err = ext3_journal_stop(handle);
2892         if (!ret)
2893                 ret = err;
2894         return ret;
2895 }
2896
2897 static int ext3_acquire_dquot(struct dquot *dquot)
2898 {
2899         int ret, err;
2900         handle_t *handle;
2901
2902         handle = ext3_journal_start(dquot_to_inode(dquot),
2903                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2904         if (IS_ERR(handle))
2905                 return PTR_ERR(handle);
2906         ret = dquot_acquire(dquot);
2907         err = ext3_journal_stop(handle);
2908         if (!ret)
2909                 ret = err;
2910         return ret;
2911 }
2912
2913 static int ext3_release_dquot(struct dquot *dquot)
2914 {
2915         int ret, err;
2916         handle_t *handle;
2917
2918         handle = ext3_journal_start(dquot_to_inode(dquot),
2919                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2920         if (IS_ERR(handle)) {
2921                 /* Release dquot anyway to avoid endless cycle in dqput() */
2922                 dquot_release(dquot);
2923                 return PTR_ERR(handle);
2924         }
2925         ret = dquot_release(dquot);
2926         err = ext3_journal_stop(handle);
2927         if (!ret)
2928                 ret = err;
2929         return ret;
2930 }
2931
2932 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2933 {
2934         /* Are we journaling quotas? */
2935         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2936             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2937                 dquot_mark_dquot_dirty(dquot);
2938                 return ext3_write_dquot(dquot);
2939         } else {
2940                 return dquot_mark_dquot_dirty(dquot);
2941         }
2942 }
2943
2944 static int ext3_write_info(struct super_block *sb, int type)
2945 {
2946         int ret, err;
2947         handle_t *handle;
2948
2949         /* Data block + inode block */
2950         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2951         if (IS_ERR(handle))
2952                 return PTR_ERR(handle);
2953         ret = dquot_commit_info(sb, type);
2954         err = ext3_journal_stop(handle);
2955         if (!ret)
2956                 ret = err;
2957         return ret;
2958 }
2959
2960 /*
2961  * Turn on quotas during mount time - we need to find
2962  * the quota file and such...
2963  */
2964 static int ext3_quota_on_mount(struct super_block *sb, int type)
2965 {
2966         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2967                                         EXT3_SB(sb)->s_jquota_fmt, type);
2968 }
2969
2970 /*
2971  * Standard function to be called on quota_on
2972  */
2973 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2974                          struct path *path)
2975 {
2976         int err;
2977
2978         if (!test_opt(sb, QUOTA))
2979                 return -EINVAL;
2980
2981         /* Quotafile not on the same filesystem? */
2982         if (path->dentry->d_sb != sb)
2983                 return -EXDEV;
2984         /* Journaling quota? */
2985         if (EXT3_SB(sb)->s_qf_names[type]) {
2986                 /* Quotafile not of fs root? */
2987                 if (path->dentry->d_parent != sb->s_root)
2988                         ext3_msg(sb, KERN_WARNING,
2989                                 "warning: Quota file not on filesystem root. "
2990                                 "Journaled quota will not work.");
2991         }
2992
2993         /*
2994          * When we journal data on quota file, we have to flush journal to see
2995          * all updates to the file when we bypass pagecache...
2996          */
2997         if (ext3_should_journal_data(path->dentry->d_inode)) {
2998                 /*
2999                  * We don't need to lock updates but journal_flush() could
3000                  * otherwise be livelocked...
3001                  */
3002                 journal_lock_updates(EXT3_SB(sb)->s_journal);
3003                 err = journal_flush(EXT3_SB(sb)->s_journal);
3004                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
3005                 if (err)
3006                         return err;
3007         }
3008
3009         return dquot_quota_on(sb, type, format_id, path);
3010 }
3011
3012 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3013  * acquiring the locks... As quota files are never truncated and quota code
3014  * itself serializes the operations (and no one else should touch the files)
3015  * we don't have to be afraid of races */
3016 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
3017                                size_t len, loff_t off)
3018 {
3019         struct inode *inode = sb_dqopt(sb)->files[type];
3020         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3021         int err = 0;
3022         int offset = off & (sb->s_blocksize - 1);
3023         int tocopy;
3024         size_t toread;
3025         struct buffer_head *bh;
3026         loff_t i_size = i_size_read(inode);
3027
3028         if (off > i_size)
3029                 return 0;
3030         if (off+len > i_size)
3031                 len = i_size-off;
3032         toread = len;
3033         while (toread > 0) {
3034                 tocopy = sb->s_blocksize - offset < toread ?
3035                                 sb->s_blocksize - offset : toread;
3036                 bh = ext3_bread(NULL, inode, blk, 0, &err);
3037                 if (err)
3038                         return err;
3039                 if (!bh)        /* A hole? */
3040                         memset(data, 0, tocopy);
3041                 else
3042                         memcpy(data, bh->b_data+offset, tocopy);
3043                 brelse(bh);
3044                 offset = 0;
3045                 toread -= tocopy;
3046                 data += tocopy;
3047                 blk++;
3048         }
3049         return len;
3050 }
3051
3052 /* Write to quotafile (we know the transaction is already started and has
3053  * enough credits) */
3054 static ssize_t ext3_quota_write(struct super_block *sb, int type,
3055                                 const char *data, size_t len, loff_t off)
3056 {
3057         struct inode *inode = sb_dqopt(sb)->files[type];
3058         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3059         int err = 0;
3060         int offset = off & (sb->s_blocksize - 1);
3061         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
3062         struct buffer_head *bh;
3063         handle_t *handle = journal_current_handle();
3064
3065         if (!handle) {
3066                 ext3_msg(sb, KERN_WARNING,
3067                         "warning: quota write (off=%llu, len=%llu)"
3068                         " cancelled because transaction is not started.",
3069                         (unsigned long long)off, (unsigned long long)len);
3070                 return -EIO;
3071         }
3072
3073         /*
3074          * Since we account only one data block in transaction credits,
3075          * then it is impossible to cross a block boundary.
3076          */
3077         if (sb->s_blocksize - offset < len) {
3078                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3079                         " cancelled because not block aligned",
3080                         (unsigned long long)off, (unsigned long long)len);
3081                 return -EIO;
3082         }
3083         bh = ext3_bread(handle, inode, blk, 1, &err);
3084         if (!bh)
3085                 goto out;
3086         if (journal_quota) {
3087                 err = ext3_journal_get_write_access(handle, bh);
3088                 if (err) {
3089                         brelse(bh);
3090                         goto out;
3091                 }
3092         }
3093         lock_buffer(bh);
3094         memcpy(bh->b_data+offset, data, len);
3095         flush_dcache_page(bh->b_page);
3096         unlock_buffer(bh);
3097         if (journal_quota)
3098                 err = ext3_journal_dirty_metadata(handle, bh);
3099         else {
3100                 /* Always do at least ordered writes for quotas */
3101                 err = ext3_journal_dirty_data(handle, bh);
3102                 mark_buffer_dirty(bh);
3103         }
3104         brelse(bh);
3105 out:
3106         if (err)
3107                 return err;
3108         if (inode->i_size < off + len) {
3109                 i_size_write(inode, off + len);
3110                 EXT3_I(inode)->i_disksize = inode->i_size;
3111         }
3112         inode->i_version++;
3113         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3114         ext3_mark_inode_dirty(handle, inode);
3115         return len;
3116 }
3117
3118 #endif
3119
3120 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3121         int flags, const char *dev_name, void *data)
3122 {
3123         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3124 }
3125
3126 static struct file_system_type ext3_fs_type = {
3127         .owner          = THIS_MODULE,
3128         .name           = "ext3",
3129         .mount          = ext3_mount,
3130         .kill_sb        = kill_block_super,
3131         .fs_flags       = FS_REQUIRES_DEV,
3132 };
3133 MODULE_ALIAS_FS("ext3");
3134
3135 static int __init init_ext3_fs(void)
3136 {
3137         int err = init_ext3_xattr();
3138         if (err)
3139                 return err;
3140         err = init_inodecache();
3141         if (err)
3142                 goto out1;
3143         err = register_filesystem(&ext3_fs_type);
3144         if (err)
3145                 goto out;
3146         return 0;
3147 out:
3148         destroy_inodecache();
3149 out1:
3150         exit_ext3_xattr();
3151         return err;
3152 }
3153
3154 static void __exit exit_ext3_fs(void)
3155 {
3156         unregister_filesystem(&ext3_fs_type);
3157         destroy_inodecache();
3158         exit_ext3_xattr();
3159 }
3160
3161 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3162 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3163 MODULE_LICENSE("GPL");
3164 module_init(init_ext3_fs)
3165 module_exit(exit_ext3_fs)