fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/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
  31 #include <asm/uaccess.h>
  32
  33 #define CREATE_TRACE_POINTS
  34
  35 #include "ext3.h"
  36 #include "xattr.h"
  37 #include "acl.h"
  38 #include "namei.h"
  39
  40 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  41   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  42 #else
  43   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  44 #endif
  45
  46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  47                              unsigned long journal_devnum);
  48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  49                                unsigned int);
  50 static int ext3_commit_super(struct super_block *sb,
  51                                struct ext3_super_block *es,
  52                                int sync);
  53 static void ext3_mark_recovery_complete(struct super_block * sb,
  54                                         struct ext3_super_block * es);
  55 static void ext3_clear_journal_err(struct super_block * sb,
  56                                    struct ext3_super_block * es);
  57 static int ext3_sync_fs(struct super_block *sb, int wait);
  58 static const char *ext3_decode_error(struct super_block * sb, int errno,
  59                                      char nbuf[16]);
  60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  61 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  62 static int ext3_unfreeze(struct super_block *sb);
  63 static int ext3_freeze(struct super_block *sb);
  64
  65 /*
  66  * Wrappers for journal_start/end.
  67  */
  68 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  69 {
  70         journal_t *journal;
  71
  72         if (sb->s_flags & MS_RDONLY)
  73                 return ERR_PTR(-EROFS);
  74
  75         /* Special case here: if the journal has aborted behind our
  76          * backs (eg. EIO in the commit thread), then we still need to
  77          * take the FS itself readonly cleanly. */
  78         journal = EXT3_SB(sb)->s_journal;
  79         if (is_journal_aborted(journal)) {
  80                 ext3_abort(sb, __func__,
  81                            "Detected aborted journal");
  82                 return ERR_PTR(-EROFS);
  83         }
  84
  85         return journal_start(journal, nblocks);
  86 }
  87
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         if (bh)
 112                 BUFFER_TRACE(bh, "abort");
 113
 114         if (!handle->h_err)
 115                 handle->h_err = err;
 116
 117         if (is_handle_aborted(handle))
 118                 return;
 119
 120         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 121                 caller, errstr, err_fn);
 122
 123         journal_abort_handle(handle);
 124 }
 125
 126 void ext3_msg(struct super_block *sb, const char *prefix,
 127                 const char *fmt, ...)
 128 {
 129         struct va_format vaf;
 130         va_list args;
 131
 132         va_start(args, fmt);
 133
 134         vaf.fmt = fmt;
 135         vaf.va = &args;
 136
 137         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 138
 139         va_end(args);
 140 }
 141
 142 /* Deal with the reporting of failure conditions on a filesystem such as
 143  * inconsistencies detected or read IO failures.
 144  *
 145  * On ext2, we can store the error state of the filesystem in the
 146  * superblock.  That is not possible on ext3, because we may have other
 147  * write ordering constraints on the superblock which prevent us from
 148  * writing it out straight away; and given that the journal is about to
 149  * be aborted, we can't rely on the current, or future, transactions to
 150  * write out the superblock safely.
 151  *
 152  * We'll just use the journal_abort() error code to record an error in
 153  * the journal instead.  On recovery, the journal will complain about
 154  * that error until we've noted it down and cleared it.
 155  */
 156
 157 static void ext3_handle_error(struct super_block *sb)
 158 {
 159         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 160
 161         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 162         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 163
 164         if (sb->s_flags & MS_RDONLY)
 165                 return;
 166
 167         if (!test_opt (sb, ERRORS_CONT)) {
 168                 journal_t *journal = EXT3_SB(sb)->s_journal;
 169
 170                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 171                 if (journal)
 172                         journal_abort(journal, -EIO);
 173         }
 174         if (test_opt (sb, ERRORS_RO)) {
 175                 ext3_msg(sb, KERN_CRIT,
 176                         "error: remounting filesystem read-only");
 177                 /*
 178                  * Make sure updated value of ->s_mount_state will be visible
 179                  * before ->s_flags update.
 180                  */
 181                 smp_wmb();
 182                 sb->s_flags |= MS_RDONLY;
 183         }
 184         ext3_commit_super(sb, es, 1);
 185         if (test_opt(sb, ERRORS_PANIC))
 186                 panic("EXT3-fs (%s): panic forced after error\n",
 187                         sb->s_id);
 188 }
 189
 190 void ext3_error(struct super_block *sb, const char *function,
 191                 const char *fmt, ...)
 192 {
 193         struct va_format vaf;
 194         va_list args;
 195
 196         va_start(args, fmt);
 197
 198         vaf.fmt = fmt;
 199         vaf.va = &args;
 200
 201         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 202                sb->s_id, function, &vaf);
 203
 204         va_end(args);
 205
 206         ext3_handle_error(sb);
 207 }
 208
 209 static const char *ext3_decode_error(struct super_block * sb, int errno,
 210                                      char nbuf[16])
 211 {
 212         char *errstr = NULL;
 213
 214         switch (errno) {
 215         case -EIO:
 216                 errstr = "IO failure";
 217                 break;
 218         case -ENOMEM:
 219                 errstr = "Out of memory";
 220                 break;
 221         case -EROFS:
 222                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 223                         errstr = "Journal has aborted";
 224                 else
 225                         errstr = "Readonly filesystem";
 226                 break;
 227         default:
 228                 /* If the caller passed in an extra buffer for unknown
 229                  * errors, textualise them now.  Else we just return
 230                  * NULL. */
 231                 if (nbuf) {
 232                         /* Check for truncated error codes... */
 233                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 234                                 errstr = nbuf;
 235                 }
 236                 break;
 237         }
 238
 239         return errstr;
 240 }
 241
 242 /* __ext3_std_error decodes expected errors from journaling functions
 243  * automatically and invokes the appropriate error response.  */
 244
 245 void __ext3_std_error (struct super_block * sb, const char * function,
 246                        int errno)
 247 {
 248         char nbuf[16];
 249         const char *errstr;
 250
 251         /* Special case: if the error is EROFS, and we're not already
 252          * inside a transaction, then there's really no point in logging
 253          * an error. */
 254         if (errno == -EROFS && journal_current_handle() == NULL &&
 255             (sb->s_flags & MS_RDONLY))
 256                 return;
 257
 258         errstr = ext3_decode_error(sb, errno, nbuf);
 259         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 260
 261         ext3_handle_error(sb);
 262 }
 263
 264 /*
 265  * ext3_abort is a much stronger failure handler than ext3_error.  The
 266  * abort function may be used to deal with unrecoverable failures such
 267  * as journal IO errors or ENOMEM at a critical moment in log management.
 268  *
 269  * We unconditionally force the filesystem into an ABORT|READONLY state,
 270  * unless the error response on the fs has been set to panic in which
 271  * case we take the easy way out and panic immediately.
 272  */
 273
 274 void ext3_abort(struct super_block *sb, const char *function,
 275                  const char *fmt, ...)
 276 {
 277         struct va_format vaf;
 278         va_list args;
 279
 280         va_start(args, fmt);
 281
 282         vaf.fmt = fmt;
 283         vaf.va = &args;
 284
 285         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 286                sb->s_id, function, &vaf);
 287
 288         va_end(args);
 289
 290         if (test_opt(sb, ERRORS_PANIC))
 291                 panic("EXT3-fs: panic from previous error\n");
 292
 293         if (sb->s_flags & MS_RDONLY)
 294                 return;
 295
 296         ext3_msg(sb, KERN_CRIT,
 297                 "error: remounting filesystem read-only");
 298         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 299         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 300         /*
 301          * Make sure updated value of ->s_mount_state will be visible
 302          * before ->s_flags update.
 303          */
 304         smp_wmb();
 305         sb->s_flags |= MS_RDONLY;
 306
 307         if (EXT3_SB(sb)->s_journal)
 308                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 309 }
 310
 311 void ext3_warning(struct super_block *sb, const char *function,
 312                   const char *fmt, ...)
 313 {
 314         struct va_format vaf;
 315         va_list args;
 316
 317         va_start(args, fmt);
 318
 319         vaf.fmt = fmt;
 320         vaf.va = &args;
 321
 322         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 323                sb->s_id, function, &vaf);
 324
 325         va_end(args);
 326 }
 327
 328 void ext3_update_dynamic_rev(struct super_block *sb)
 329 {
 330         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 331
 332         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 333                 return;
 334
 335         ext3_msg(sb, KERN_WARNING,
 336                 "warning: updating to rev %d because of "
 337                 "new feature flag, running e2fsck is recommended",
 338                 EXT3_DYNAMIC_REV);
 339
 340         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 341         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 342         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 343         /* leave es->s_feature_*compat flags alone */
 344         /* es->s_uuid will be set by e2fsck if empty */
 345
 346         /*
 347          * The rest of the superblock fields should be zero, and if not it
 348          * means they are likely already in use, so leave them alone.  We
 349          * can leave it up to e2fsck to clean up any inconsistencies there.
 350          */
 351 }
 352
 353 /*
 354  * Open the external journal device
 355  */
 356 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 357 {
 358         struct block_device *bdev;
 359         char b[BDEVNAME_SIZE];
 360
 361         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 362         if (IS_ERR(bdev))
 363                 goto fail;
 364         return bdev;
 365
 366 fail:
 367         ext3_msg(sb, KERN_ERR, "error: failed to open journal device %s: %ld",
 368                 __bdevname(dev, b), PTR_ERR(bdev));
 369
 370         return NULL;
 371 }
 372
 373 /*
 374  * Release the journal device
 375  */
 376 static void ext3_blkdev_put(struct block_device *bdev)
 377 {
 378         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 379 }
 380
 381 static void ext3_blkdev_remove(struct ext3_sb_info *sbi)
 382 {
 383         struct block_device *bdev;
 384         bdev = sbi->journal_bdev;
 385         if (bdev) {
 386                 ext3_blkdev_put(bdev);
 387                 sbi->journal_bdev = NULL;
 388         }
 389 }
 390
 391 static inline struct inode *orphan_list_entry(struct list_head *l)
 392 {
 393         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 394 }
 395
 396 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 397 {
 398         struct list_head *l;
 399
 400         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 401                le32_to_cpu(sbi->s_es->s_last_orphan));
 402
 403         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 404         list_for_each(l, &sbi->s_orphan) {
 405                 struct inode *inode = orphan_list_entry(l);
 406                 ext3_msg(sb, KERN_ERR, "  "
 407                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 408                        inode->i_sb->s_id, inode->i_ino, inode,
 409                        inode->i_mode, inode->i_nlink,
 410                        NEXT_ORPHAN(inode));
 411         }
 412 }
 413
 414 static void ext3_put_super (struct super_block * sb)
 415 {
 416         struct ext3_sb_info *sbi = EXT3_SB(sb);
 417         struct ext3_super_block *es = sbi->s_es;
 418         int i, err;
 419
 420         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 421         ext3_xattr_put_super(sb);
 422         err = journal_destroy(sbi->s_journal);
 423         sbi->s_journal = NULL;
 424         if (err < 0)
 425                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 426
 427         if (!(sb->s_flags & MS_RDONLY)) {
 428                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 429                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 430                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 431                 mark_buffer_dirty(sbi->s_sbh);
 432                 ext3_commit_super(sb, es, 1);
 433         }
 434
 435         for (i = 0; i < sbi->s_gdb_count; i++)
 436                 brelse(sbi->s_group_desc[i]);
 437         kfree(sbi->s_group_desc);
 438         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 439         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 440         percpu_counter_destroy(&sbi->s_dirs_counter);
 441         brelse(sbi->s_sbh);
 442 #ifdef CONFIG_QUOTA
 443         for (i = 0; i < MAXQUOTAS; i++)
 444                 kfree(sbi->s_qf_names[i]);
 445 #endif
 446
 447         /* Debugging code just in case the in-memory inode orphan list
 448          * isn't empty.  The on-disk one can be non-empty if we've
 449          * detected an error and taken the fs readonly, but the
 450          * in-memory list had better be clean by this point. */
 451         if (!list_empty(&sbi->s_orphan))
 452                 dump_orphan_list(sb, sbi);
 453         J_ASSERT(list_empty(&sbi->s_orphan));
 454
 455         invalidate_bdev(sb->s_bdev);
 456         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 457                 /*
 458                  * Invalidate the journal device's buffers.  We don't want them
 459                  * floating about in memory - the physical journal device may
 460                  * hotswapped, and it breaks the `ro-after' testing code.
 461                  */
 462                 sync_blockdev(sbi->journal_bdev);
 463                 invalidate_bdev(sbi->journal_bdev);
 464                 ext3_blkdev_remove(sbi);
 465         }
 466         sb->s_fs_info = NULL;
 467         kfree(sbi->s_blockgroup_lock);
 468         kfree(sbi);
 469 }
 470
 471 static struct kmem_cache *ext3_inode_cachep;
 472
 473 /*
 474  * Called inside transaction, so use GFP_NOFS
 475  */
 476 static struct inode *ext3_alloc_inode(struct super_block *sb)
 477 {
 478         struct ext3_inode_info *ei;
 479
 480         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 481         if (!ei)
 482                 return NULL;
 483         ei->i_block_alloc_info = NULL;
 484         ei->vfs_inode.i_version = 1;
 485         atomic_set(&ei->i_datasync_tid, 0);
 486         atomic_set(&ei->i_sync_tid, 0);
 487         return &ei->vfs_inode;
 488 }
 489
 490 static int ext3_drop_inode(struct inode *inode)
 491 {
 492         int drop = generic_drop_inode(inode);
 493
 494         trace_ext3_drop_inode(inode, drop);
 495         return drop;
 496 }
 497
 498 static void ext3_i_callback(struct rcu_head *head)
 499 {
 500         struct inode *inode = container_of(head, struct inode, i_rcu);
 501         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 502 }
 503
 504 static void ext3_destroy_inode(struct inode *inode)
 505 {
 506         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 507                 printk("EXT3 Inode %p: orphan list check failed!\n",
 508                         EXT3_I(inode));
 509                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 510                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 511                                 false);
 512                 dump_stack();
 513         }
 514         call_rcu(&inode->i_rcu, ext3_i_callback);
 515 }
 516
 517 static void init_once(void *foo)
 518 {
 519         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 520
 521         INIT_LIST_HEAD(&ei->i_orphan);
 522 #ifdef CONFIG_EXT3_FS_XATTR
 523         init_rwsem(&ei->xattr_sem);
 524 #endif
 525         mutex_init(&ei->truncate_mutex);
 526         inode_init_once(&ei->vfs_inode);
 527 }
 528
 529 static int init_inodecache(void)
 530 {
 531         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 532                                              sizeof(struct ext3_inode_info),
 533                                              0, (SLAB_RECLAIM_ACCOUNT|
 534                                                 SLAB_MEM_SPREAD),
 535                                              init_once);
 536         if (ext3_inode_cachep == NULL)
 537                 return -ENOMEM;
 538         return 0;
 539 }
 540
 541 static void destroy_inodecache(void)
 542 {
 543         /*
 544          * Make sure all delayed rcu free inodes are flushed before we
 545          * destroy cache.
 546          */
 547         rcu_barrier();
 548         kmem_cache_destroy(ext3_inode_cachep);
 549 }
 550
 551 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 552 {
 553 #if defined(CONFIG_QUOTA)
 554         struct ext3_sb_info *sbi = EXT3_SB(sb);
 555
 556         if (sbi->s_jquota_fmt) {
 557                 char *fmtname = "";
 558
 559                 switch (sbi->s_jquota_fmt) {
 560                 case QFMT_VFS_OLD:
 561                         fmtname = "vfsold";
 562                         break;
 563                 case QFMT_VFS_V0:
 564                         fmtname = "vfsv0";
 565                         break;
 566                 case QFMT_VFS_V1:
 567                         fmtname = "vfsv1";
 568                         break;
 569                 }
 570                 seq_printf(seq, ",jqfmt=%s", fmtname);
 571         }
 572
 573         if (sbi->s_qf_names[USRQUOTA])
 574                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 575
 576         if (sbi->s_qf_names[GRPQUOTA])
 577                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 578
 579         if (test_opt(sb, USRQUOTA))
 580                 seq_puts(seq, ",usrquota");
 581
 582         if (test_opt(sb, GRPQUOTA))
 583                 seq_puts(seq, ",grpquota");
 584 #endif
 585 }
 586
 587 static char *data_mode_string(unsigned long mode)
 588 {
 589         switch (mode) {
 590         case EXT3_MOUNT_JOURNAL_DATA:
 591                 return "journal";
 592         case EXT3_MOUNT_ORDERED_DATA:
 593                 return "ordered";
 594         case EXT3_MOUNT_WRITEBACK_DATA:
 595                 return "writeback";
 596         }
 597         return "unknown";
 598 }
 599
 600 /*
 601  * Show an option if
 602  *  - it's set to a non-default value OR
 603  *  - if the per-sb default is different from the global default
 604  */
 605 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 606 {
 607         struct super_block *sb = root->d_sb;
 608         struct ext3_sb_info *sbi = EXT3_SB(sb);
 609         struct ext3_super_block *es = sbi->s_es;
 610         unsigned long def_mount_opts;
 611
 612         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 613
 614         if (sbi->s_sb_block != 1)
 615                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 616         if (test_opt(sb, MINIX_DF))
 617                 seq_puts(seq, ",minixdf");
 618         if (test_opt(sb, GRPID))
 619                 seq_puts(seq, ",grpid");
 620         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 621                 seq_puts(seq, ",nogrpid");
 622         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
 623             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 624                 seq_printf(seq, ",resuid=%u",
 625                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
 626         }
 627         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
 628             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 629                 seq_printf(seq, ",resgid=%u",
 630                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
 631         }
 632         if (test_opt(sb, ERRORS_RO)) {
 633                 int def_errors = le16_to_cpu(es->s_errors);
 634
 635                 if (def_errors == EXT3_ERRORS_PANIC ||
 636                     def_errors == EXT3_ERRORS_CONTINUE) {
 637                         seq_puts(seq, ",errors=remount-ro");
 638                 }
 639         }
 640         if (test_opt(sb, ERRORS_CONT))
 641                 seq_puts(seq, ",errors=continue");
 642         if (test_opt(sb, ERRORS_PANIC))
 643                 seq_puts(seq, ",errors=panic");
 644         if (test_opt(sb, NO_UID32))
 645                 seq_puts(seq, ",nouid32");
 646         if (test_opt(sb, DEBUG))
 647                 seq_puts(seq, ",debug");
 648 #ifdef CONFIG_EXT3_FS_XATTR
 649         if (test_opt(sb, XATTR_USER))
 650                 seq_puts(seq, ",user_xattr");
 651         if (!test_opt(sb, XATTR_USER) &&
 652             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 653                 seq_puts(seq, ",nouser_xattr");
 654         }
 655 #endif
 656 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 657         if (test_opt(sb, POSIX_ACL))
 658                 seq_puts(seq, ",acl");
 659         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 660                 seq_puts(seq, ",noacl");
 661 #endif
 662         if (!test_opt(sb, RESERVATION))
 663                 seq_puts(seq, ",noreservation");
 664         if (sbi->s_commit_interval) {
 665                 seq_printf(seq, ",commit=%u",
 666                            (unsigned) (sbi->s_commit_interval / HZ));
 667         }
 668
 669         /*
 670          * Always display barrier state so it's clear what the status is.
 671          */
 672         seq_puts(seq, ",barrier=");
 673         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 674         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 675         if (test_opt(sb, DATA_ERR_ABORT))
 676                 seq_puts(seq, ",data_err=abort");
 677
 678         if (test_opt(sb, NOLOAD))
 679                 seq_puts(seq, ",norecovery");
 680
 681         ext3_show_quota_options(seq, sb);
 682
 683         return 0;
 684 }
 685
 686
 687 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 688                 u64 ino, u32 generation)
 689 {
 690         struct inode *inode;
 691
 692         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 693                 return ERR_PTR(-ESTALE);
 694         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 695                 return ERR_PTR(-ESTALE);
 696
 697         /* iget isn't really right if the inode is currently unallocated!!
 698          *
 699          * ext3_read_inode will return a bad_inode if the inode had been
 700          * deleted, so we should be safe.
 701          *
 702          * Currently we don't know the generation for parent directory, so
 703          * a generation of 0 means "accept any"
 704          */
 705         inode = ext3_iget(sb, ino);
 706         if (IS_ERR(inode))
 707                 return ERR_CAST(inode);
 708         if (generation && inode->i_generation != generation) {
 709                 iput(inode);
 710                 return ERR_PTR(-ESTALE);
 711         }
 712
 713         return inode;
 714 }
 715
 716 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 717                 int fh_len, int fh_type)
 718 {
 719         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 720                                     ext3_nfs_get_inode);
 721 }
 722
 723 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 724                 int fh_len, int fh_type)
 725 {
 726         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 727                                     ext3_nfs_get_inode);
 728 }
 729
 730 /*
 731  * Try to release metadata pages (indirect blocks, directories) which are
 732  * mapped via the block device.  Since these pages could have journal heads
 733  * which would prevent try_to_free_buffers() from freeing them, we must use
 734  * jbd layer's try_to_free_buffers() function to release them.
 735  */
 736 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 737                                  gfp_t wait)
 738 {
 739         journal_t *journal = EXT3_SB(sb)->s_journal;
 740
 741         WARN_ON(PageChecked(page));
 742         if (!page_has_buffers(page))
 743                 return 0;
 744         if (journal)
 745                 return journal_try_to_free_buffers(journal, page,
 746                                                    wait & ~__GFP_WAIT);
 747         return try_to_free_buffers(page);
 748 }
 749
 750 #ifdef CONFIG_QUOTA
 751 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 752 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 753
 754 static int ext3_write_dquot(struct dquot *dquot);
 755 static int ext3_acquire_dquot(struct dquot *dquot);
 756 static int ext3_release_dquot(struct dquot *dquot);
 757 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 758 static int ext3_write_info(struct super_block *sb, int type);
 759 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 760                          struct path *path);
 761 static int ext3_quota_on_mount(struct super_block *sb, int type);
 762 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 763                                size_t len, loff_t off);
 764 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 765                                 const char *data, size_t len, loff_t off);
 766
 767 static const struct dquot_operations ext3_quota_operations = {
 768         .write_dquot    = ext3_write_dquot,
 769         .acquire_dquot  = ext3_acquire_dquot,
 770         .release_dquot  = ext3_release_dquot,
 771         .mark_dirty     = ext3_mark_dquot_dirty,
 772         .write_info     = ext3_write_info,
 773         .alloc_dquot    = dquot_alloc,
 774         .destroy_dquot  = dquot_destroy,
 775 };
 776
 777 static const struct quotactl_ops ext3_qctl_operations = {
 778         .quota_on       = ext3_quota_on,
 779         .quota_off      = dquot_quota_off,
 780         .quota_sync     = dquot_quota_sync,
 781         .get_info       = dquot_get_dqinfo,
 782         .set_info       = dquot_set_dqinfo,
 783         .get_dqblk      = dquot_get_dqblk,
 784         .set_dqblk      = dquot_set_dqblk
 785 };
 786 #endif
 787
 788 static const struct super_operations ext3_sops = {
 789         .alloc_inode    = ext3_alloc_inode,
 790         .destroy_inode  = ext3_destroy_inode,
 791         .write_inode    = ext3_write_inode,
 792         .dirty_inode    = ext3_dirty_inode,
 793         .drop_inode     = ext3_drop_inode,
 794         .evict_inode    = ext3_evict_inode,
 795         .put_super      = ext3_put_super,
 796         .sync_fs        = ext3_sync_fs,
 797         .freeze_fs      = ext3_freeze,
 798         .unfreeze_fs    = ext3_unfreeze,
 799         .statfs         = ext3_statfs,
 800         .remount_fs     = ext3_remount,
 801         .show_options   = ext3_show_options,
 802 #ifdef CONFIG_QUOTA
 803         .quota_read     = ext3_quota_read,
 804         .quota_write    = ext3_quota_write,
 805 #endif
 806         .bdev_try_to_free_page = bdev_try_to_free_page,
 807 };
 808
 809 static const struct export_operations ext3_export_ops = {
 810         .fh_to_dentry = ext3_fh_to_dentry,
 811         .fh_to_parent = ext3_fh_to_parent,
 812         .get_parent = ext3_get_parent,
 813 };
 814
 815 enum {
 816         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 817         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 818         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 819         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 820         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 821         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 822         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 823         Opt_data_err_abort, Opt_data_err_ignore,
 824         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 825         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 826         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 827         Opt_resize, Opt_usrquota, Opt_grpquota
 828 };
 829
 830 static const match_table_t tokens = {
 831         {Opt_bsd_df, "bsddf"},
 832         {Opt_minix_df, "minixdf"},
 833         {Opt_grpid, "grpid"},
 834         {Opt_grpid, "bsdgroups"},
 835         {Opt_nogrpid, "nogrpid"},
 836         {Opt_nogrpid, "sysvgroups"},
 837         {Opt_resgid, "resgid=%u"},
 838         {Opt_resuid, "resuid=%u"},
 839         {Opt_sb, "sb=%u"},
 840         {Opt_err_cont, "errors=continue"},
 841         {Opt_err_panic, "errors=panic"},
 842         {Opt_err_ro, "errors=remount-ro"},
 843         {Opt_nouid32, "nouid32"},
 844         {Opt_nocheck, "nocheck"},
 845         {Opt_nocheck, "check=none"},
 846         {Opt_debug, "debug"},
 847         {Opt_oldalloc, "oldalloc"},
 848         {Opt_orlov, "orlov"},
 849         {Opt_user_xattr, "user_xattr"},
 850         {Opt_nouser_xattr, "nouser_xattr"},
 851         {Opt_acl, "acl"},
 852         {Opt_noacl, "noacl"},
 853         {Opt_reservation, "reservation"},
 854         {Opt_noreservation, "noreservation"},
 855         {Opt_noload, "noload"},
 856         {Opt_noload, "norecovery"},
 857         {Opt_nobh, "nobh"},
 858         {Opt_bh, "bh"},
 859         {Opt_commit, "commit=%u"},
 860         {Opt_journal_update, "journal=update"},
 861         {Opt_journal_inum, "journal=%u"},
 862         {Opt_journal_dev, "journal_dev=%u"},
 863         {Opt_abort, "abort"},
 864         {Opt_data_journal, "data=journal"},
 865         {Opt_data_ordered, "data=ordered"},
 866         {Opt_data_writeback, "data=writeback"},
 867         {Opt_data_err_abort, "data_err=abort"},
 868         {Opt_data_err_ignore, "data_err=ignore"},
 869         {Opt_offusrjquota, "usrjquota="},
 870         {Opt_usrjquota, "usrjquota=%s"},
 871         {Opt_offgrpjquota, "grpjquota="},
 872         {Opt_grpjquota, "grpjquota=%s"},
 873         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 874         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 875         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 876         {Opt_grpquota, "grpquota"},
 877         {Opt_noquota, "noquota"},
 878         {Opt_quota, "quota"},
 879         {Opt_usrquota, "usrquota"},
 880         {Opt_barrier, "barrier=%u"},
 881         {Opt_barrier, "barrier"},
 882         {Opt_nobarrier, "nobarrier"},
 883         {Opt_resize, "resize"},
 884         {Opt_err, NULL},
 885 };
 886
 887 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 888 {
 889         ext3_fsblk_t    sb_block;
 890         char            *options = (char *) *data;
 891
 892         if (!options || strncmp(options, "sb=", 3) != 0)
 893                 return 1;       /* Default location */
 894         options += 3;
 895         /*todo: use simple_strtoll with >32bit ext3 */
 896         sb_block = simple_strtoul(options, &options, 0);
 897         if (*options && *options != ',') {
 898                 ext3_msg(sb, KERN_ERR, "error: invalid sb specification: %s",
 899                        (char *) *data);
 900                 return 1;
 901         }
 902         if (*options == ',')
 903                 options++;
 904         *data = (void *) options;
 905         return sb_block;
 906 }
 907
 908 #ifdef CONFIG_QUOTA
 909 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 910 {
 911         struct ext3_sb_info *sbi = EXT3_SB(sb);
 912         char *qname;
 913
 914         if (sb_any_quota_loaded(sb) &&
 915                 !sbi->s_qf_names[qtype]) {
 916                 ext3_msg(sb, KERN_ERR,
 917                         "Cannot change journaled "
 918                         "quota options when quota turned on");
 919                 return 0;
 920         }
 921         qname = match_strdup(args);
 922         if (!qname) {
 923                 ext3_msg(sb, KERN_ERR,
 924                         "Not enough memory for storing quotafile name");
 925                 return 0;
 926         }
 927         if (sbi->s_qf_names[qtype]) {
 928                 int same = !strcmp(sbi->s_qf_names[qtype], qname);
 929
 930                 kfree(qname);
 931                 if (!same) {
 932                         ext3_msg(sb, KERN_ERR,
 933                                  "%s quota file already specified",
 934                                  QTYPE2NAME(qtype));
 935                 }
 936                 return same;
 937         }
 938         if (strchr(qname, '/')) {
 939                 ext3_msg(sb, KERN_ERR,
 940                         "quotafile must be on filesystem root");
 941                 kfree(qname);
 942                 return 0;
 943         }
 944         sbi->s_qf_names[qtype] = qname;
 945         set_opt(sbi->s_mount_opt, QUOTA);
 946         return 1;
 947 }
 948
 949 static int clear_qf_name(struct super_block *sb, int qtype) {
 950
 951         struct ext3_sb_info *sbi = EXT3_SB(sb);
 952
 953         if (sb_any_quota_loaded(sb) &&
 954                 sbi->s_qf_names[qtype]) {
 955                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 956                         " when quota turned on");
 957                 return 0;
 958         }
 959         if (sbi->s_qf_names[qtype]) {
 960                 kfree(sbi->s_qf_names[qtype]);
 961                 sbi->s_qf_names[qtype] = NULL;
 962         }
 963         return 1;
 964 }
 965 #endif
 966
 967 static int parse_options (char *options, struct super_block *sb,
 968                           unsigned int *inum, unsigned long *journal_devnum,
 969                           ext3_fsblk_t *n_blocks_count, int is_remount)
 970 {
 971         struct ext3_sb_info *sbi = EXT3_SB(sb);
 972         char * p;
 973         substring_t args[MAX_OPT_ARGS];
 974         int data_opt = 0;
 975         int option;
 976         kuid_t uid;
 977         kgid_t gid;
 978 #ifdef CONFIG_QUOTA
 979         int qfmt;
 980 #endif
 981
 982         if (!options)
 983                 return 1;
 984
 985         while ((p = strsep (&options, ",")) != NULL) {
 986                 int token;
 987                 if (!*p)
 988                         continue;
 989                 /*
 990                  * Initialize args struct so we know whether arg was
 991                  * found; some options take optional arguments.
 992                  */
 993                 args[0].to = args[0].from = NULL;
 994                 token = match_token(p, tokens, args);
 995                 switch (token) {
 996                 case Opt_bsd_df:
 997                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 998                         break;
 999                 case Opt_minix_df:
1000                         set_opt (sbi->s_mount_opt, MINIX_DF);
1001                         break;
1002                 case Opt_grpid:
1003                         set_opt (sbi->s_mount_opt, GRPID);
1004                         break;
1005                 case Opt_nogrpid:
1006                         clear_opt (sbi->s_mount_opt, GRPID);
1007                         break;
1008                 case Opt_resuid:
1009                         if (match_int(&args[0], &option))
1010                                 return 0;
1011                         uid = make_kuid(current_user_ns(), option);
1012                         if (!uid_valid(uid)) {
1013                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1014                                 return 0;
1015
1016                         }
1017                         sbi->s_resuid = uid;
1018                         break;
1019                 case Opt_resgid:
1020                         if (match_int(&args[0], &option))
1021                                 return 0;
1022                         gid = make_kgid(current_user_ns(), option);
1023                         if (!gid_valid(gid)) {
1024                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1025                                 return 0;
1026                         }
1027                         sbi->s_resgid = gid;
1028                         break;
1029                 case Opt_sb:
1030                         /* handled by get_sb_block() instead of here */
1031                         /* *sb_block = match_int(&args[0]); */
1032                         break;
1033                 case Opt_err_panic:
1034                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1035                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1036                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1037                         break;
1038                 case Opt_err_ro:
1039                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1040                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1041                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1042                         break;
1043                 case Opt_err_cont:
1044                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1045                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1046                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1047                         break;
1048                 case Opt_nouid32:
1049                         set_opt (sbi->s_mount_opt, NO_UID32);
1050                         break;
1051                 case Opt_nocheck:
1052                         clear_opt (sbi->s_mount_opt, CHECK);
1053                         break;
1054                 case Opt_debug:
1055                         set_opt (sbi->s_mount_opt, DEBUG);
1056                         break;
1057                 case Opt_oldalloc:
1058                         ext3_msg(sb, KERN_WARNING,
1059                                 "Ignoring deprecated oldalloc option");
1060                         break;
1061                 case Opt_orlov:
1062                         ext3_msg(sb, KERN_WARNING,
1063                                 "Ignoring deprecated orlov option");
1064                         break;
1065 #ifdef CONFIG_EXT3_FS_XATTR
1066                 case Opt_user_xattr:
1067                         set_opt (sbi->s_mount_opt, XATTR_USER);
1068                         break;
1069                 case Opt_nouser_xattr:
1070                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1071                         break;
1072 #else
1073                 case Opt_user_xattr:
1074                 case Opt_nouser_xattr:
1075                         ext3_msg(sb, KERN_INFO,
1076                                 "(no)user_xattr options not supported");
1077                         break;
1078 #endif
1079 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1080                 case Opt_acl:
1081                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1082                         break;
1083                 case Opt_noacl:
1084                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1085                         break;
1086 #else
1087                 case Opt_acl:
1088                 case Opt_noacl:
1089                         ext3_msg(sb, KERN_INFO,
1090                                 "(no)acl options not supported");
1091                         break;
1092 #endif
1093                 case Opt_reservation:
1094                         set_opt(sbi->s_mount_opt, RESERVATION);
1095                         break;
1096                 case Opt_noreservation:
1097                         clear_opt(sbi->s_mount_opt, RESERVATION);
1098                         break;
1099                 case Opt_journal_update:
1100                         /* @@@ FIXME */
1101                         /* Eventually we will want to be able to create
1102                            a journal file here.  For now, only allow the
1103                            user to specify an existing inode to be the
1104                            journal file. */
1105                         if (is_remount) {
1106                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1107                                         "journal on remount");
1108                                 return 0;
1109                         }
1110                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1111                         break;
1112                 case Opt_journal_inum:
1113                         if (is_remount) {
1114                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1115                                        "journal on remount");
1116                                 return 0;
1117                         }
1118                         if (match_int(&args[0], &option))
1119                                 return 0;
1120                         *inum = option;
1121                         break;
1122                 case Opt_journal_dev:
1123                         if (is_remount) {
1124                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1125                                        "journal on remount");
1126                                 return 0;
1127                         }
1128                         if (match_int(&args[0], &option))
1129                                 return 0;
1130                         *journal_devnum = option;
1131                         break;
1132                 case Opt_noload:
1133                         set_opt (sbi->s_mount_opt, NOLOAD);
1134                         break;
1135                 case Opt_commit:
1136                         if (match_int(&args[0], &option))
1137                                 return 0;
1138                         if (option < 0)
1139                                 return 0;
1140                         if (option == 0)
1141                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1142                         sbi->s_commit_interval = HZ * option;
1143                         break;
1144                 case Opt_data_journal:
1145                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1146                         goto datacheck;
1147                 case Opt_data_ordered:
1148                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1149                         goto datacheck;
1150                 case Opt_data_writeback:
1151                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1152                 datacheck:
1153                         if (is_remount) {
1154                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1155                                         break;
1156                                 ext3_msg(sb, KERN_ERR,
1157                                         "error: cannot change "
1158                                         "data mode on remount. The filesystem "
1159                                         "is mounted in data=%s mode and you "
1160                                         "try to remount it in data=%s mode.",
1161                                         data_mode_string(test_opt(sb,
1162                                                         DATA_FLAGS)),
1163                                         data_mode_string(data_opt));
1164                                 return 0;
1165                         } else {
1166                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1167                                 sbi->s_mount_opt |= data_opt;
1168                         }
1169                         break;
1170                 case Opt_data_err_abort:
1171                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1172                         break;
1173                 case Opt_data_err_ignore:
1174                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1175                         break;
1176 #ifdef CONFIG_QUOTA
1177                 case Opt_usrjquota:
1178                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1179                                 return 0;
1180                         break;
1181                 case Opt_grpjquota:
1182                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1183                                 return 0;
1184                         break;
1185                 case Opt_offusrjquota:
1186                         if (!clear_qf_name(sb, USRQUOTA))
1187                                 return 0;
1188                         break;
1189                 case Opt_offgrpjquota:
1190                         if (!clear_qf_name(sb, GRPQUOTA))
1191                                 return 0;
1192                         break;
1193                 case Opt_jqfmt_vfsold:
1194                         qfmt = QFMT_VFS_OLD;
1195                         goto set_qf_format;
1196                 case Opt_jqfmt_vfsv0:
1197                         qfmt = QFMT_VFS_V0;
1198                         goto set_qf_format;
1199                 case Opt_jqfmt_vfsv1:
1200                         qfmt = QFMT_VFS_V1;
1201 set_qf_format:
1202                         if (sb_any_quota_loaded(sb) &&
1203                             sbi->s_jquota_fmt != qfmt) {
1204                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1205                                         "journaled quota options when "
1206                                         "quota turned on.");
1207                                 return 0;
1208                         }
1209                         sbi->s_jquota_fmt = qfmt;
1210                         break;
1211                 case Opt_quota:
1212                 case Opt_usrquota:
1213                         set_opt(sbi->s_mount_opt, QUOTA);
1214                         set_opt(sbi->s_mount_opt, USRQUOTA);
1215                         break;
1216                 case Opt_grpquota:
1217                         set_opt(sbi->s_mount_opt, QUOTA);
1218                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1219                         break;
1220                 case Opt_noquota:
1221                         if (sb_any_quota_loaded(sb)) {
1222                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1223                                         "quota options when quota turned on.");
1224                                 return 0;
1225                         }
1226                         clear_opt(sbi->s_mount_opt, QUOTA);
1227                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1228                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1229                         break;
1230 #else
1231                 case Opt_quota:
1232                 case Opt_usrquota:
1233                 case Opt_grpquota:
1234                         ext3_msg(sb, KERN_ERR,
1235                                 "error: quota options not supported.");
1236                         break;
1237                 case Opt_usrjquota:
1238                 case Opt_grpjquota:
1239                 case Opt_offusrjquota:
1240                 case Opt_offgrpjquota:
1241                 case Opt_jqfmt_vfsold:
1242                 case Opt_jqfmt_vfsv0:
1243                 case Opt_jqfmt_vfsv1:
1244                         ext3_msg(sb, KERN_ERR,
1245                                 "error: journaled quota options not "
1246                                 "supported.");
1247                         break;
1248                 case Opt_noquota:
1249                         break;
1250 #endif
1251                 case Opt_abort:
1252                         set_opt(sbi->s_mount_opt, ABORT);
1253                         break;
1254                 case Opt_nobarrier:
1255                         clear_opt(sbi->s_mount_opt, BARRIER);
1256                         break;
1257                 case Opt_barrier:
1258                         if (args[0].from) {
1259                                 if (match_int(&args[0], &option))
1260                                         return 0;
1261                         } else
1262                                 option = 1;     /* No argument, default to 1 */
1263                         if (option)
1264                                 set_opt(sbi->s_mount_opt, BARRIER);
1265                         else
1266                                 clear_opt(sbi->s_mount_opt, BARRIER);
1267                         break;
1268                 case Opt_ignore:
1269                         break;
1270                 case Opt_resize:
1271                         if (!is_remount) {
1272                                 ext3_msg(sb, KERN_ERR,
1273                                         "error: resize option only available "
1274                                         "for remount");
1275                                 return 0;
1276                         }
1277                         if (match_int(&args[0], &option) != 0)
1278                                 return 0;
1279                         *n_blocks_count = option;
1280                         break;
1281                 case Opt_nobh:
1282                         ext3_msg(sb, KERN_WARNING,
1283                                 "warning: ignoring deprecated nobh option");
1284                         break;
1285                 case Opt_bh:
1286                         ext3_msg(sb, KERN_WARNING,
1287                                 "warning: ignoring deprecated bh option");
1288                         break;
1289                 default:
1290                         ext3_msg(sb, KERN_ERR,
1291                                 "error: unrecognized mount option \"%s\" "
1292                                 "or missing value", p);
1293                         return 0;
1294                 }
1295         }
1296 #ifdef CONFIG_QUOTA
1297         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1298                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1299                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1300                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1301                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1302
1303                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1304                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1305                                         "format mixing.");
1306                         return 0;
1307                 }
1308
1309                 if (!sbi->s_jquota_fmt) {
1310                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1311                                         "not specified.");
1312                         return 0;
1313                 }
1314         } else {
1315                 if (sbi->s_jquota_fmt) {
1316                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1317                                         "specified with no journaling "
1318                                         "enabled.");
1319                         return 0;
1320                 }
1321         }
1322 #endif
1323         return 1;
1324 }
1325
1326 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1327                             int read_only)
1328 {
1329         struct ext3_sb_info *sbi = EXT3_SB(sb);
1330         int res = 0;
1331
1332         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1333                 ext3_msg(sb, KERN_ERR,
1334                         "error: revision level too high, "
1335                         "forcing read-only mode");
1336                 res = MS_RDONLY;
1337         }
1338         if (read_only)
1339                 return res;
1340         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1341                 ext3_msg(sb, KERN_WARNING,
1342                         "warning: mounting unchecked fs, "
1343                         "running e2fsck is recommended");
1344         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1345                 ext3_msg(sb, KERN_WARNING,
1346                         "warning: mounting fs with errors, "
1347                         "running e2fsck is recommended");
1348         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1349                  le16_to_cpu(es->s_mnt_count) >=
1350                         le16_to_cpu(es->s_max_mnt_count))
1351                 ext3_msg(sb, KERN_WARNING,
1352                         "warning: maximal mount count reached, "
1353                         "running e2fsck is recommended");
1354         else if (le32_to_cpu(es->s_checkinterval) &&
1355                 (le32_to_cpu(es->s_lastcheck) +
1356                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1357                 ext3_msg(sb, KERN_WARNING,
1358                         "warning: checktime reached, "
1359                         "running e2fsck is recommended");
1360 #if 0
1361                 /* @@@ We _will_ want to clear the valid bit if we find
1362                    inconsistencies, to force a fsck at reboot.  But for
1363                    a plain journaled filesystem we can keep it set as
1364                    valid forever! :) */
1365         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1366 #endif
1367         if (!le16_to_cpu(es->s_max_mnt_count))
1368                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1369         le16_add_cpu(&es->s_mnt_count, 1);
1370         es->s_mtime = cpu_to_le32(get_seconds());
1371         ext3_update_dynamic_rev(sb);
1372         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1373
1374         ext3_commit_super(sb, es, 1);
1375         if (test_opt(sb, DEBUG))
1376                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1377                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1378                         sb->s_blocksize,
1379                         sbi->s_groups_count,
1380                         EXT3_BLOCKS_PER_GROUP(sb),
1381                         EXT3_INODES_PER_GROUP(sb),
1382                         sbi->s_mount_opt);
1383
1384         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1385                 char b[BDEVNAME_SIZE];
1386                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1387                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1388         } else {
1389                 ext3_msg(sb, KERN_INFO, "using internal journal");
1390         }
1391         cleancache_init_fs(sb);
1392         return res;
1393 }
1394
1395 /* Called at mount-time, super-block is locked */
1396 static int ext3_check_descriptors(struct super_block *sb)
1397 {
1398         struct ext3_sb_info *sbi = EXT3_SB(sb);
1399         int i;
1400
1401         ext3_debug ("Checking group descriptors");
1402
1403         for (i = 0; i < sbi->s_groups_count; i++) {
1404                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1405                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1406                 ext3_fsblk_t last_block;
1407
1408                 if (i == sbi->s_groups_count - 1)
1409                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1410                 else
1411                         last_block = first_block +
1412                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1413
1414                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1415                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1416                 {
1417                         ext3_error (sb, "ext3_check_descriptors",
1418                                     "Block bitmap for group %d"
1419                                     " not in group (block %lu)!",
1420                                     i, (unsigned long)
1421                                         le32_to_cpu(gdp->bg_block_bitmap));
1422                         return 0;
1423                 }
1424                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1425                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1426                 {
1427                         ext3_error (sb, "ext3_check_descriptors",
1428                                     "Inode bitmap for group %d"
1429                                     " not in group (block %lu)!",
1430                                     i, (unsigned long)
1431                                         le32_to_cpu(gdp->bg_inode_bitmap));
1432                         return 0;
1433                 }
1434                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1435                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1436                     last_block)
1437                 {
1438                         ext3_error (sb, "ext3_check_descriptors",
1439                                     "Inode table for group %d"
1440                                     " not in group (block %lu)!",
1441                                     i, (unsigned long)
1442                                         le32_to_cpu(gdp->bg_inode_table));
1443                         return 0;
1444                 }
1445         }
1446
1447         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1448         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1449         return 1;
1450 }
1451
1452
1453 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1454  * the superblock) which were deleted from all directories, but held open by
1455  * a process at the time of a crash.  We walk the list and try to delete these
1456  * inodes at recovery time (only with a read-write filesystem).
1457  *
1458  * In order to keep the orphan inode chain consistent during traversal (in
1459  * case of crash during recovery), we link each inode into the superblock
1460  * orphan list_head and handle it the same way as an inode deletion during
1461  * normal operation (which journals the operations for us).
1462  *
1463  * We only do an iget() and an iput() on each inode, which is very safe if we
1464  * accidentally point at an in-use or already deleted inode.  The worst that
1465  * can happen in this case is that we get a "bit already cleared" message from
1466  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1467  * e2fsck was run on this filesystem, and it must have already done the orphan
1468  * inode cleanup for us, so we can safely abort without any further action.
1469  */
1470 static void ext3_orphan_cleanup (struct super_block * sb,
1471                                  struct ext3_super_block * es)
1472 {
1473         unsigned int s_flags = sb->s_flags;
1474         int nr_orphans = 0, nr_truncates = 0;
1475 #ifdef CONFIG_QUOTA
1476         int i;
1477 #endif
1478         if (!es->s_last_orphan) {
1479                 jbd_debug(4, "no orphan inodes to clean up\n");
1480                 return;
1481         }
1482
1483         if (bdev_read_only(sb->s_bdev)) {
1484                 ext3_msg(sb, KERN_ERR, "error: write access "
1485                         "unavailable, skipping orphan cleanup.");
1486                 return;
1487         }
1488
1489         /* Check if feature set allows readwrite operations */
1490         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1491                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1492                          "unknown ROCOMPAT features");
1493                 return;
1494         }
1495
1496         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1497                 /* don't clear list on RO mount w/ errors */
1498                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
1499                         jbd_debug(1, "Errors on filesystem, "
1500                                   "clearing orphan list.\n");
1501                         es->s_last_orphan = 0;
1502                 }
1503                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1504                 return;
1505         }
1506
1507         if (s_flags & MS_RDONLY) {
1508                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1509                 sb->s_flags &= ~MS_RDONLY;
1510         }
1511 #ifdef CONFIG_QUOTA
1512         /* Needed for iput() to work correctly and not trash data */
1513         sb->s_flags |= MS_ACTIVE;
1514         /* Turn on quotas so that they are updated correctly */
1515         for (i = 0; i < MAXQUOTAS; i++) {
1516                 if (EXT3_SB(sb)->s_qf_names[i]) {
1517                         int ret = ext3_quota_on_mount(sb, i);
1518                         if (ret < 0)
1519                                 ext3_msg(sb, KERN_ERR,
1520                                         "error: cannot turn on journaled "
1521                                         "quota: %d", ret);
1522                 }
1523         }
1524 #endif
1525
1526         while (es->s_last_orphan) {
1527                 struct inode *inode;
1528
1529                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1530                 if (IS_ERR(inode)) {
1531                         es->s_last_orphan = 0;
1532                         break;
1533                 }
1534
1535                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1536                 dquot_initialize(inode);
1537                 if (inode->i_nlink) {
1538                         printk(KERN_DEBUG
1539                                 "%s: truncating inode %lu to %Ld bytes\n",
1540                                 __func__, inode->i_ino, inode->i_size);
1541                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1542                                   inode->i_ino, inode->i_size);
1543                         ext3_truncate(inode);
1544                         nr_truncates++;
1545                 } else {
1546                         printk(KERN_DEBUG
1547                                 "%s: deleting unreferenced inode %lu\n",
1548                                 __func__, inode->i_ino);
1549                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1550                                   inode->i_ino);
1551                         nr_orphans++;
1552                 }
1553                 iput(inode);  /* The delete magic happens here! */
1554         }
1555
1556 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1557
1558         if (nr_orphans)
1559                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1560                        PLURAL(nr_orphans));
1561         if (nr_truncates)
1562                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1563                        PLURAL(nr_truncates));
1564 #ifdef CONFIG_QUOTA
1565         /* Turn quotas off */
1566         for (i = 0; i < MAXQUOTAS; i++) {
1567                 if (sb_dqopt(sb)->files[i])
1568                         dquot_quota_off(sb, i);
1569         }
1570 #endif
1571         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1572 }
1573
1574 /*
1575  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1576  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1577  * We need to be 1 filesystem block less than the 2^32 sector limit.
1578  */
1579 static loff_t ext3_max_size(int bits)
1580 {
1581         loff_t res = EXT3_NDIR_BLOCKS;
1582         int meta_blocks;
1583         loff_t upper_limit;
1584
1585         /* This is calculated to be the largest file size for a
1586          * dense, file such that the total number of
1587          * sectors in the file, including data and all indirect blocks,
1588          * does not exceed 2^32 -1
1589          * __u32 i_blocks representing the total number of
1590          * 512 bytes blocks of the file
1591          */
1592         upper_limit = (1LL << 32) - 1;
1593
1594         /* total blocks in file system block size */
1595         upper_limit >>= (bits - 9);
1596
1597
1598         /* indirect blocks */
1599         meta_blocks = 1;
1600         /* double indirect blocks */
1601         meta_blocks += 1 + (1LL << (bits-2));
1602         /* tripple indirect blocks */
1603         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1604
1605         upper_limit -= meta_blocks;
1606         upper_limit <<= bits;
1607
1608         res += 1LL << (bits-2);
1609         res += 1LL << (2*(bits-2));
1610         res += 1LL << (3*(bits-2));
1611         res <<= bits;
1612         if (res > upper_limit)
1613                 res = upper_limit;
1614
1615         if (res > MAX_LFS_FILESIZE)
1616                 res = MAX_LFS_FILESIZE;
1617
1618         return res;
1619 }
1620
1621 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1622                                     ext3_fsblk_t logic_sb_block,
1623                                     int nr)
1624 {
1625         struct ext3_sb_info *sbi = EXT3_SB(sb);
1626         unsigned long bg, first_meta_bg;
1627         int has_super = 0;
1628
1629         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1630
1631         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1632             nr < first_meta_bg)
1633                 return (logic_sb_block + nr + 1);
1634         bg = sbi->s_desc_per_block * nr;
1635         if (ext3_bg_has_super(sb, bg))
1636                 has_super = 1;
1637         return (has_super + ext3_group_first_block_no(sb, bg));
1638 }
1639
1640
1641 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1642 {
1643         struct buffer_head * bh;
1644         struct ext3_super_block *es = NULL;
1645         struct ext3_sb_info *sbi;
1646         ext3_fsblk_t block;
1647         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1648         ext3_fsblk_t logic_sb_block;
1649         unsigned long offset = 0;
1650         unsigned int journal_inum = 0;
1651         unsigned long journal_devnum = 0;
1652         unsigned long def_mount_opts;
1653         struct inode *root;
1654         int blocksize;
1655         int hblock;
1656         int db_count;
1657         int i;
1658         int needs_recovery;
1659         int ret = -EINVAL;
1660         __le32 features;
1661         int err;
1662
1663         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1664         if (!sbi)
1665                 return -ENOMEM;
1666
1667         sbi->s_blockgroup_lock =
1668                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1669         if (!sbi->s_blockgroup_lock) {
1670                 kfree(sbi);
1671                 return -ENOMEM;
1672         }
1673         sb->s_fs_info = sbi;
1674         sbi->s_sb_block = sb_block;
1675
1676         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1677         if (!blocksize) {
1678                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1679                 goto out_fail;
1680         }
1681
1682         /*
1683          * The ext3 superblock will not be buffer aligned for other than 1kB
1684          * block sizes.  We need to calculate the offset from buffer start.
1685          */
1686         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1687                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1688                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1689         } else {
1690                 logic_sb_block = sb_block;
1691         }
1692
1693         if (!(bh = sb_bread(sb, logic_sb_block))) {
1694                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1695                 goto out_fail;
1696         }
1697         /*
1698          * Note: s_es must be initialized as soon as possible because
1699          *       some ext3 macro-instructions depend on its value
1700          */
1701         es = (struct ext3_super_block *) (bh->b_data + offset);
1702         sbi->s_es = es;
1703         sb->s_magic = le16_to_cpu(es->s_magic);
1704         if (sb->s_magic != EXT3_SUPER_MAGIC)
1705                 goto cantfind_ext3;
1706
1707         /* Set defaults before we parse the mount options */
1708         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1709         if (def_mount_opts & EXT3_DEFM_DEBUG)
1710                 set_opt(sbi->s_mount_opt, DEBUG);
1711         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1712                 set_opt(sbi->s_mount_opt, GRPID);
1713         if (def_mount_opts & EXT3_DEFM_UID16)
1714                 set_opt(sbi->s_mount_opt, NO_UID32);
1715 #ifdef CONFIG_EXT3_FS_XATTR
1716         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1717                 set_opt(sbi->s_mount_opt, XATTR_USER);
1718 #endif
1719 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1720         if (def_mount_opts & EXT3_DEFM_ACL)
1721                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1722 #endif
1723         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1724                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1725         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1726                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1727         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1728                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1729
1730         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1731                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1732         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1733                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1734         else
1735                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1736
1737         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1738         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1739
1740         /* enable barriers by default */
1741         set_opt(sbi->s_mount_opt, BARRIER);
1742         set_opt(sbi->s_mount_opt, RESERVATION);
1743
1744         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1745                             NULL, 0))
1746                 goto failed_mount;
1747
1748         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1749                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1750
1751         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1752             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1753              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1754              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1755                 ext3_msg(sb, KERN_WARNING,
1756                         "warning: feature flags set on rev 0 fs, "
1757                         "running e2fsck is recommended");
1758         /*
1759          * Check feature flags regardless of the revision level, since we
1760          * previously didn't change the revision level when setting the flags,
1761          * so there is a chance incompat flags are set on a rev 0 filesystem.
1762          */
1763         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1764         if (features) {
1765                 ext3_msg(sb, KERN_ERR,
1766                         "error: couldn't mount because of unsupported "
1767                         "optional features (%x)", le32_to_cpu(features));
1768                 goto failed_mount;
1769         }
1770         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1771         if (!(sb->s_flags & MS_RDONLY) && features) {
1772                 ext3_msg(sb, KERN_ERR,
1773                         "error: couldn't mount RDWR because of unsupported "
1774                         "optional features (%x)", le32_to_cpu(features));
1775                 goto failed_mount;
1776         }
1777         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1778
1779         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1780             blocksize > EXT3_MAX_BLOCK_SIZE) {
1781                 ext3_msg(sb, KERN_ERR,
1782                         "error: couldn't mount because of unsupported "
1783                         "filesystem blocksize %d", blocksize);
1784                 goto failed_mount;
1785         }
1786
1787         hblock = bdev_logical_block_size(sb->s_bdev);
1788         if (sb->s_blocksize != blocksize) {
1789                 /*
1790                  * Make sure the blocksize for the filesystem is larger
1791                  * than the hardware sectorsize for the machine.
1792                  */
1793                 if (blocksize < hblock) {
1794                         ext3_msg(sb, KERN_ERR,
1795                                 "error: fsblocksize %d too small for "
1796                                 "hardware sectorsize %d", blocksize, hblock);
1797                         goto failed_mount;
1798                 }
1799
1800                 brelse (bh);
1801                 if (!sb_set_blocksize(sb, blocksize)) {
1802                         ext3_msg(sb, KERN_ERR,
1803                                 "error: bad blocksize %d", blocksize);
1804                         goto out_fail;
1805                 }
1806                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1807                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1808                 bh = sb_bread(sb, logic_sb_block);
1809                 if (!bh) {
1810                         ext3_msg(sb, KERN_ERR,
1811                                "error: can't read superblock on 2nd try");
1812                         goto failed_mount;
1813                 }
1814                 es = (struct ext3_super_block *)(bh->b_data + offset);
1815                 sbi->s_es = es;
1816                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1817                         ext3_msg(sb, KERN_ERR,
1818                                 "error: magic mismatch");
1819                         goto failed_mount;
1820                 }
1821         }
1822
1823         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1824
1825         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1826                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1827                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1828         } else {
1829                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1830                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1831                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1832                     (!is_power_of_2(sbi->s_inode_size)) ||
1833                     (sbi->s_inode_size > blocksize)) {
1834                         ext3_msg(sb, KERN_ERR,
1835                                 "error: unsupported inode size: %d",
1836                                 sbi->s_inode_size);
1837                         goto failed_mount;
1838                 }
1839         }
1840         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1841                                    le32_to_cpu(es->s_log_frag_size);
1842         if (blocksize != sbi->s_frag_size) {
1843                 ext3_msg(sb, KERN_ERR,
1844                        "error: fragsize %lu != blocksize %u (unsupported)",
1845                        sbi->s_frag_size, blocksize);
1846                 goto failed_mount;
1847         }
1848         sbi->s_frags_per_block = 1;
1849         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1850         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1851         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1852         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1853                 goto cantfind_ext3;
1854         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1855         if (sbi->s_inodes_per_block == 0)
1856                 goto cantfind_ext3;
1857         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1858                                         sbi->s_inodes_per_block;
1859         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1860         sbi->s_sbh = bh;
1861         sbi->s_mount_state = le16_to_cpu(es->s_state);
1862         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1863         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1864         for (i=0; i < 4; i++)
1865                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1866         sbi->s_def_hash_version = es->s_def_hash_version;
1867         i = le32_to_cpu(es->s_flags);
1868         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1869                 sbi->s_hash_unsigned = 3;
1870         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1871 #ifdef __CHAR_UNSIGNED__
1872                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1873                 sbi->s_hash_unsigned = 3;
1874 #else
1875                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1876 #endif
1877         }
1878
1879         if (sbi->s_blocks_per_group > blocksize * 8) {
1880                 ext3_msg(sb, KERN_ERR,
1881                         "#blocks per group too big: %lu",
1882                         sbi->s_blocks_per_group);
1883                 goto failed_mount;
1884         }
1885         if (sbi->s_frags_per_group > blocksize * 8) {
1886                 ext3_msg(sb, KERN_ERR,
1887                         "error: #fragments per group too big: %lu",
1888                         sbi->s_frags_per_group);
1889                 goto failed_mount;
1890         }
1891         if (sbi->s_inodes_per_group > blocksize * 8) {
1892                 ext3_msg(sb, KERN_ERR,
1893                         "error: #inodes per group too big: %lu",
1894                         sbi->s_inodes_per_group);
1895                 goto failed_mount;
1896         }
1897
1898         err = generic_check_addressable(sb->s_blocksize_bits,
1899                                         le32_to_cpu(es->s_blocks_count));
1900         if (err) {
1901                 ext3_msg(sb, KERN_ERR,
1902                         "error: filesystem is too large to mount safely");
1903                 if (sizeof(sector_t) < 8)
1904                         ext3_msg(sb, KERN_ERR,
1905                                 "error: CONFIG_LBDAF not enabled");
1906                 ret = err;
1907                 goto failed_mount;
1908         }
1909
1910         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1911                 goto cantfind_ext3;
1912         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1913                                le32_to_cpu(es->s_first_data_block) - 1)
1914                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1915         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1916         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1917                                     GFP_KERNEL);
1918         if (sbi->s_group_desc == NULL) {
1919                 ext3_msg(sb, KERN_ERR,
1920                         "error: not enough memory");
1921                 ret = -ENOMEM;
1922                 goto failed_mount;
1923         }
1924
1925         bgl_lock_init(sbi->s_blockgroup_lock);
1926
1927         for (i = 0; i < db_count; i++) {
1928                 block = descriptor_loc(sb, logic_sb_block, i);
1929                 sbi->s_group_desc[i] = sb_bread(sb, block);
1930                 if (!sbi->s_group_desc[i]) {
1931                         ext3_msg(sb, KERN_ERR,
1932                                 "error: can't read group descriptor %d", i);
1933                         db_count = i;
1934                         goto failed_mount2;
1935                 }
1936         }
1937         if (!ext3_check_descriptors (sb)) {
1938                 ext3_msg(sb, KERN_ERR,
1939                         "error: group descriptors corrupted");
1940                 goto failed_mount2;
1941         }
1942         sbi->s_gdb_count = db_count;
1943         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1944         spin_lock_init(&sbi->s_next_gen_lock);
1945
1946         /* per fileystem reservation list head & lock */
1947         spin_lock_init(&sbi->s_rsv_window_lock);
1948         sbi->s_rsv_window_root = RB_ROOT;
1949         /* Add a single, static dummy reservation to the start of the
1950          * reservation window list --- it gives us a placeholder for
1951          * append-at-start-of-list which makes the allocation logic
1952          * _much_ simpler. */
1953         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1954         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1955         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1956         sbi->s_rsv_window_head.rsv_goal_size = 0;
1957         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1958
1959         /*
1960          * set up enough so that it can read an inode
1961          */
1962         sb->s_op = &ext3_sops;
1963         sb->s_export_op = &ext3_export_ops;
1964         sb->s_xattr = ext3_xattr_handlers;
1965 #ifdef CONFIG_QUOTA
1966         sb->s_qcop = &ext3_qctl_operations;
1967         sb->dq_op = &ext3_quota_operations;
1968 #endif
1969         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1970         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1971         mutex_init(&sbi->s_orphan_lock);
1972         mutex_init(&sbi->s_resize_lock);
1973
1974         sb->s_root = NULL;
1975
1976         needs_recovery = (es->s_last_orphan != 0 ||
1977                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1978                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1979
1980         /*
1981          * The first inode we look at is the journal inode.  Don't try
1982          * root first: it may be modified in the journal!
1983          */
1984         if (!test_opt(sb, NOLOAD) &&
1985             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1986                 if (ext3_load_journal(sb, es, journal_devnum))
1987                         goto failed_mount2;
1988         } else if (journal_inum) {
1989                 if (ext3_create_journal(sb, es, journal_inum))
1990                         goto failed_mount2;
1991         } else {
1992                 if (!silent)
1993                         ext3_msg(sb, KERN_ERR,
1994                                 "error: no journal found. "
1995                                 "mounting ext3 over ext2?");
1996                 goto failed_mount2;
1997         }
1998         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1999                         ext3_count_free_blocks(sb));
2000         if (!err) {
2001                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2002                                 ext3_count_free_inodes(sb));
2003         }
2004         if (!err) {
2005                 err = percpu_counter_init(&sbi->s_dirs_counter,
2006                                 ext3_count_dirs(sb));
2007         }
2008         if (err) {
2009                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2010                 ret = err;
2011                 goto failed_mount3;
2012         }
2013
2014         /* We have now updated the journal if required, so we can
2015          * validate the data journaling mode. */
2016         switch (test_opt(sb, DATA_FLAGS)) {
2017         case 0:
2018                 /* No mode set, assume a default based on the journal
2019                    capabilities: ORDERED_DATA if the journal can
2020                    cope, else JOURNAL_DATA */
2021                 if (journal_check_available_features
2022                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2023                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2024                 else
2025                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2026                 break;
2027
2028         case EXT3_MOUNT_ORDERED_DATA:
2029         case EXT3_MOUNT_WRITEBACK_DATA:
2030                 if (!journal_check_available_features
2031                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2032                         ext3_msg(sb, KERN_ERR,
2033                                 "error: journal does not support "
2034                                 "requested data journaling mode");
2035                         goto failed_mount3;
2036                 }
2037         default:
2038                 break;
2039         }
2040
2041         /*
2042          * The journal_load will have done any necessary log recovery,
2043          * so we can safely mount the rest of the filesystem now.
2044          */
2045
2046         root = ext3_iget(sb, EXT3_ROOT_INO);
2047         if (IS_ERR(root)) {
2048                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2049                 ret = PTR_ERR(root);
2050                 goto failed_mount3;
2051         }
2052         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2053                 iput(root);
2054                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2055                 goto failed_mount3;
2056         }
2057         sb->s_root = d_make_root(root);
2058         if (!sb->s_root) {
2059                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2060                 ret = -ENOMEM;
2061                 goto failed_mount3;
2062         }
2063
2064         if (ext3_setup_super(sb, es, sb->s_flags & MS_RDONLY))
2065                 sb->s_flags |= MS_RDONLY;
2066
2067         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2068         ext3_orphan_cleanup(sb, es);
2069         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2070         if (needs_recovery) {
2071                 ext3_mark_recovery_complete(sb, es);
2072                 ext3_msg(sb, KERN_INFO, "recovery complete");
2073         }
2074         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2075                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2076                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2077                 "writeback");
2078
2079         return 0;
2080
2081 cantfind_ext3:
2082         if (!silent)
2083                 ext3_msg(sb, KERN_INFO,
2084                         "error: can't find ext3 filesystem on dev %s.",
2085                        sb->s_id);
2086         goto failed_mount;
2087
2088 failed_mount3:
2089         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2090         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2091         percpu_counter_destroy(&sbi->s_dirs_counter);
2092         journal_destroy(sbi->s_journal);
2093 failed_mount2:
2094         for (i = 0; i < db_count; i++)
2095                 brelse(sbi->s_group_desc[i]);
2096         kfree(sbi->s_group_desc);
2097 failed_mount:
2098 #ifdef CONFIG_QUOTA
2099         for (i = 0; i < MAXQUOTAS; i++)
2100                 kfree(sbi->s_qf_names[i]);
2101 #endif
2102         ext3_blkdev_remove(sbi);
2103         brelse(bh);
2104 out_fail:
2105         sb->s_fs_info = NULL;
2106         kfree(sbi->s_blockgroup_lock);
2107         kfree(sbi);
2108         return ret;
2109 }
2110
2111 /*
2112  * Setup any per-fs journal parameters now.  We'll do this both on
2113  * initial mount, once the journal has been initialised but before we've
2114  * done any recovery; and again on any subsequent remount.
2115  */
2116 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2117 {
2118         struct ext3_sb_info *sbi = EXT3_SB(sb);
2119
2120         if (sbi->s_commit_interval)
2121                 journal->j_commit_interval = sbi->s_commit_interval;
2122         /* We could also set up an ext3-specific default for the commit
2123          * interval here, but for now we'll just fall back to the jbd
2124          * default. */
2125
2126         spin_lock(&journal->j_state_lock);
2127         if (test_opt(sb, BARRIER))
2128                 journal->j_flags |= JFS_BARRIER;
2129         else
2130                 journal->j_flags &= ~JFS_BARRIER;
2131         if (test_opt(sb, DATA_ERR_ABORT))
2132                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2133         else
2134                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2135         spin_unlock(&journal->j_state_lock);
2136 }
2137
2138 static journal_t *ext3_get_journal(struct super_block *sb,
2139                                    unsigned int journal_inum)
2140 {
2141         struct inode *journal_inode;
2142         journal_t *journal;
2143
2144         /* First, test for the existence of a valid inode on disk.  Bad
2145          * things happen if we iget() an unused inode, as the subsequent
2146          * iput() will try to delete it. */
2147
2148         journal_inode = ext3_iget(sb, journal_inum);
2149         if (IS_ERR(journal_inode)) {
2150                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2151                 return NULL;
2152         }
2153         if (!journal_inode->i_nlink) {
2154                 make_bad_inode(journal_inode);
2155                 iput(journal_inode);
2156                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2157                 return NULL;
2158         }
2159
2160         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2161                   journal_inode, journal_inode->i_size);
2162         if (!S_ISREG(journal_inode->i_mode)) {
2163                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2164                 iput(journal_inode);
2165                 return NULL;
2166         }
2167
2168         journal = journal_init_inode(journal_inode);
2169         if (!journal) {
2170                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2171                 iput(journal_inode);
2172                 return NULL;
2173         }
2174         journal->j_private = sb;
2175         ext3_init_journal_params(sb, journal);
2176         return journal;
2177 }
2178
2179 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2180                                        dev_t j_dev)
2181 {
2182         struct buffer_head * bh;
2183         journal_t *journal;
2184         ext3_fsblk_t start;
2185         ext3_fsblk_t len;
2186         int hblock, blocksize;
2187         ext3_fsblk_t sb_block;
2188         unsigned long offset;
2189         struct ext3_super_block * es;
2190         struct block_device *bdev;
2191
2192         bdev = ext3_blkdev_get(j_dev, sb);
2193         if (bdev == NULL)
2194                 return NULL;
2195
2196         blocksize = sb->s_blocksize;
2197         hblock = bdev_logical_block_size(bdev);
2198         if (blocksize < hblock) {
2199                 ext3_msg(sb, KERN_ERR,
2200                         "error: blocksize too small for journal device");
2201                 goto out_bdev;
2202         }
2203
2204         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2205         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2206         set_blocksize(bdev, blocksize);
2207         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2208                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2209                         "external journal");
2210                 goto out_bdev;
2211         }
2212
2213         es = (struct ext3_super_block *) (bh->b_data + offset);
2214         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2215             !(le32_to_cpu(es->s_feature_incompat) &
2216               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2217                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2218                         "bad superblock");
2219                 brelse(bh);
2220                 goto out_bdev;
2221         }
2222
2223         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2224                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2225                 brelse(bh);
2226                 goto out_bdev;
2227         }
2228
2229         len = le32_to_cpu(es->s_blocks_count);
2230         start = sb_block + 1;
2231         brelse(bh);     /* we're done with the superblock */
2232
2233         journal = journal_init_dev(bdev, sb->s_bdev,
2234                                         start, len, blocksize);
2235         if (!journal) {
2236                 ext3_msg(sb, KERN_ERR,
2237                         "error: failed to create device journal");
2238                 goto out_bdev;
2239         }
2240         journal->j_private = sb;
2241         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2242                 if (bh_submit_read(journal->j_sb_buffer)) {
2243                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2244                         goto out_journal;
2245                 }
2246         }
2247         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2248                 ext3_msg(sb, KERN_ERR,
2249                         "error: external journal has more than one "
2250                         "user (unsupported) - %d",
2251                         be32_to_cpu(journal->j_superblock->s_nr_users));
2252                 goto out_journal;
2253         }
2254         EXT3_SB(sb)->journal_bdev = bdev;
2255         ext3_init_journal_params(sb, journal);
2256         return journal;
2257 out_journal:
2258         journal_destroy(journal);
2259 out_bdev:
2260         ext3_blkdev_put(bdev);
2261         return NULL;
2262 }
2263
2264 static int ext3_load_journal(struct super_block *sb,
2265                              struct ext3_super_block *es,
2266                              unsigned long journal_devnum)
2267 {
2268         journal_t *journal;
2269         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2270         dev_t journal_dev;
2271         int err = 0;
2272         int really_read_only;
2273
2274         if (journal_devnum &&
2275             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2276                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2277                         "numbers have changed");
2278                 journal_dev = new_decode_dev(journal_devnum);
2279         } else
2280                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2281
2282         really_read_only = bdev_read_only(sb->s_bdev);
2283
2284         /*
2285          * Are we loading a blank journal or performing recovery after a
2286          * crash?  For recovery, we need to check in advance whether we
2287          * can get read-write access to the device.
2288          */
2289
2290         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2291                 if (sb->s_flags & MS_RDONLY) {
2292                         ext3_msg(sb, KERN_INFO,
2293                                 "recovery required on readonly filesystem");
2294                         if (really_read_only) {
2295                                 ext3_msg(sb, KERN_ERR, "error: write access "
2296                                         "unavailable, cannot proceed");
2297                                 return -EROFS;
2298                         }
2299                         ext3_msg(sb, KERN_INFO,
2300                                 "write access will be enabled during recovery");
2301                 }
2302         }
2303
2304         if (journal_inum && journal_dev) {
2305                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2306                        "and inode journals");
2307                 return -EINVAL;
2308         }
2309
2310         if (journal_inum) {
2311                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2312                         return -EINVAL;
2313         } else {
2314                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2315                         return -EINVAL;
2316         }
2317
2318         if (!(journal->j_flags & JFS_BARRIER))
2319                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2320
2321         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2322                 err = journal_update_format(journal);
2323                 if (err)  {
2324                         ext3_msg(sb, KERN_ERR, "error updating journal");
2325                         journal_destroy(journal);
2326                         return err;
2327                 }
2328         }
2329
2330         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2331                 err = journal_wipe(journal, !really_read_only);
2332         if (!err)
2333                 err = journal_load(journal);
2334
2335         if (err) {
2336                 ext3_msg(sb, KERN_ERR, "error loading journal");
2337                 journal_destroy(journal);
2338                 return err;
2339         }
2340
2341         EXT3_SB(sb)->s_journal = journal;
2342         ext3_clear_journal_err(sb, es);
2343
2344         if (!really_read_only && journal_devnum &&
2345             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2346                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2347
2348                 /* Make sure we flush the recovery flag to disk. */
2349                 ext3_commit_super(sb, es, 1);
2350         }
2351
2352         return 0;
2353 }
2354
2355 static int ext3_create_journal(struct super_block *sb,
2356                                struct ext3_super_block *es,
2357                                unsigned int journal_inum)
2358 {
2359         journal_t *journal;
2360         int err;
2361
2362         if (sb->s_flags & MS_RDONLY) {
2363                 ext3_msg(sb, KERN_ERR,
2364                         "error: readonly filesystem when trying to "
2365                         "create journal");
2366                 return -EROFS;
2367         }
2368
2369         journal = ext3_get_journal(sb, journal_inum);
2370         if (!journal)
2371                 return -EINVAL;
2372
2373         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2374                journal_inum);
2375
2376         err = journal_create(journal);
2377         if (err) {
2378                 ext3_msg(sb, KERN_ERR, "error creating journal");
2379                 journal_destroy(journal);
2380                 return -EIO;
2381         }
2382
2383         EXT3_SB(sb)->s_journal = journal;
2384
2385         ext3_update_dynamic_rev(sb);
2386         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2387         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2388
2389         es->s_journal_inum = cpu_to_le32(journal_inum);
2390
2391         /* Make sure we flush the recovery flag to disk. */
2392         ext3_commit_super(sb, es, 1);
2393
2394         return 0;
2395 }
2396
2397 static int ext3_commit_super(struct super_block *sb,
2398                                struct ext3_super_block *es,
2399                                int sync)
2400 {
2401         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2402         int error = 0;
2403
2404         if (!sbh)
2405                 return error;
2406
2407         if (buffer_write_io_error(sbh)) {
2408                 /*
2409                  * Oh, dear.  A previous attempt to write the
2410                  * superblock failed.  This could happen because the
2411                  * USB device was yanked out.  Or it could happen to
2412                  * be a transient write error and maybe the block will
2413                  * be remapped.  Nothing we can do but to retry the
2414                  * write and hope for the best.
2415                  */
2416                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2417                        "superblock detected");
2418                 clear_buffer_write_io_error(sbh);
2419                 set_buffer_uptodate(sbh);
2420         }
2421         /*
2422          * If the file system is mounted read-only, don't update the
2423          * superblock write time.  This avoids updating the superblock
2424          * write time when we are mounting the root file system
2425          * read/only but we need to replay the journal; at that point,
2426          * for people who are east of GMT and who make their clock
2427          * tick in localtime for Windows bug-for-bug compatibility,
2428          * the clock is set in the future, and this will cause e2fsck
2429          * to complain and force a full file system check.
2430          */
2431         if (!(sb->s_flags & MS_RDONLY))
2432                 es->s_wtime = cpu_to_le32(get_seconds());
2433         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2434         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2435         BUFFER_TRACE(sbh, "marking dirty");
2436         mark_buffer_dirty(sbh);
2437         if (sync) {
2438                 error = sync_dirty_buffer(sbh);
2439                 if (buffer_write_io_error(sbh)) {
2440                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2441                                "superblock");
2442                         clear_buffer_write_io_error(sbh);
2443                         set_buffer_uptodate(sbh);
2444                 }
2445         }
2446         return error;
2447 }
2448
2449
2450 /*
2451  * Have we just finished recovery?  If so, and if we are mounting (or
2452  * remounting) the filesystem readonly, then we will end up with a
2453  * consistent fs on disk.  Record that fact.
2454  */
2455 static void ext3_mark_recovery_complete(struct super_block * sb,
2456                                         struct ext3_super_block * es)
2457 {
2458         journal_t *journal = EXT3_SB(sb)->s_journal;
2459
2460         journal_lock_updates(journal);
2461         if (journal_flush(journal) < 0)
2462                 goto out;
2463
2464         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2465             sb->s_flags & MS_RDONLY) {
2466                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2467                 ext3_commit_super(sb, es, 1);
2468         }
2469
2470 out:
2471         journal_unlock_updates(journal);
2472 }
2473
2474 /*
2475  * If we are mounting (or read-write remounting) a filesystem whose journal
2476  * has recorded an error from a previous lifetime, move that error to the
2477  * main filesystem now.
2478  */
2479 static void ext3_clear_journal_err(struct super_block *sb,
2480                                    struct ext3_super_block *es)
2481 {
2482         journal_t *journal;
2483         int j_errno;
2484         const char *errstr;
2485
2486         journal = EXT3_SB(sb)->s_journal;
2487
2488         /*
2489          * Now check for any error status which may have been recorded in the
2490          * journal by a prior ext3_error() or ext3_abort()
2491          */
2492
2493         j_errno = journal_errno(journal);
2494         if (j_errno) {
2495                 char nbuf[16];
2496
2497                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2498                 ext3_warning(sb, __func__, "Filesystem error recorded "
2499                              "from previous mount: %s", errstr);
2500                 ext3_warning(sb, __func__, "Marking fs in need of "
2501                              "filesystem check.");
2502
2503                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2504                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2505                 ext3_commit_super (sb, es, 1);
2506
2507                 journal_clear_err(journal);
2508         }
2509 }
2510
2511 /*
2512  * Force the running and committing transactions to commit,
2513  * and wait on the commit.
2514  */
2515 int ext3_force_commit(struct super_block *sb)
2516 {
2517         journal_t *journal;
2518         int ret;
2519
2520         if (sb->s_flags & MS_RDONLY)
2521                 return 0;
2522
2523         journal = EXT3_SB(sb)->s_journal;
2524         ret = ext3_journal_force_commit(journal);
2525         return ret;
2526 }
2527
2528 static int ext3_sync_fs(struct super_block *sb, int wait)
2529 {
2530         tid_t target;
2531
2532         trace_ext3_sync_fs(sb, wait);
2533         /*
2534          * Writeback quota in non-journalled quota case - journalled quota has
2535          * no dirty dquots
2536          */
2537         dquot_writeback_dquots(sb, -1);
2538         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2539                 if (wait)
2540                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2541         }
2542         return 0;
2543 }
2544
2545 /*
2546  * LVM calls this function before a (read-only) snapshot is created.  This
2547  * gives us a chance to flush the journal completely and mark the fs clean.
2548  */
2549 static int ext3_freeze(struct super_block *sb)
2550 {
2551         int error = 0;
2552         journal_t *journal;
2553
2554         if (!(sb->s_flags & MS_RDONLY)) {
2555                 journal = EXT3_SB(sb)->s_journal;
2556
2557                 /* Now we set up the journal barrier. */
2558                 journal_lock_updates(journal);
2559
2560                 /*
2561                  * We don't want to clear needs_recovery flag when we failed
2562                  * to flush the journal.
2563                  */
2564                 error = journal_flush(journal);
2565                 if (error < 0)
2566                         goto out;
2567
2568                 /* Journal blocked and flushed, clear needs_recovery flag. */
2569                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2570                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2571                 if (error)
2572                         goto out;
2573         }
2574         return 0;
2575
2576 out:
2577         journal_unlock_updates(journal);
2578         return error;
2579 }
2580
2581 /*
2582  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2583  * flag here, even though the filesystem is not technically dirty yet.
2584  */
2585 static int ext3_unfreeze(struct super_block *sb)
2586 {
2587         if (!(sb->s_flags & MS_RDONLY)) {
2588                 /* Reser the needs_recovery flag before the fs is unlocked. */
2589                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2590                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2591                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2592         }
2593         return 0;
2594 }
2595
2596 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2597 {
2598         struct ext3_super_block * es;
2599         struct ext3_sb_info *sbi = EXT3_SB(sb);
2600         ext3_fsblk_t n_blocks_count = 0;
2601         unsigned long old_sb_flags;
2602         struct ext3_mount_options old_opts;
2603         int enable_quota = 0;
2604         int err;
2605 #ifdef CONFIG_QUOTA
2606         int i;
2607 #endif
2608
2609         /* Store the original options */
2610         old_sb_flags = sb->s_flags;
2611         old_opts.s_mount_opt = sbi->s_mount_opt;
2612         old_opts.s_resuid = sbi->s_resuid;
2613         old_opts.s_resgid = sbi->s_resgid;
2614         old_opts.s_commit_interval = sbi->s_commit_interval;
2615 #ifdef CONFIG_QUOTA
2616         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2617         for (i = 0; i < MAXQUOTAS; i++)
2618                 if (sbi->s_qf_names[i]) {
2619                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
2620                                                          GFP_KERNEL);
2621                         if (!old_opts.s_qf_names[i]) {
2622                                 int j;
2623
2624                                 for (j = 0; j < i; j++)
2625                                         kfree(old_opts.s_qf_names[j]);
2626                                 return -ENOMEM;
2627                         }
2628                 } else
2629                         old_opts.s_qf_names[i] = NULL;
2630 #endif
2631
2632         /*
2633          * Allow the "check" option to be passed as a remount option.
2634          */
2635         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2636                 err = -EINVAL;
2637                 goto restore_opts;
2638         }
2639
2640         if (test_opt(sb, ABORT))
2641                 ext3_abort(sb, __func__, "Abort forced by user");
2642
2643         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2644                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2645
2646         es = sbi->s_es;
2647
2648         ext3_init_journal_params(sb, sbi->s_journal);
2649
2650         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2651                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2652                 if (test_opt(sb, ABORT)) {
2653                         err = -EROFS;
2654                         goto restore_opts;
2655                 }
2656
2657                 if (*flags & MS_RDONLY) {
2658                         err = dquot_suspend(sb, -1);
2659                         if (err < 0)
2660                                 goto restore_opts;
2661
2662                         /*
2663                          * First of all, the unconditional stuff we have to do
2664                          * to disable replay of the journal when we next remount
2665                          */
2666                         sb->s_flags |= MS_RDONLY;
2667
2668                         /*
2669                          * OK, test if we are remounting a valid rw partition
2670                          * readonly, and if so set the rdonly flag and then
2671                          * mark the partition as valid again.
2672                          */
2673                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2674                             (sbi->s_mount_state & EXT3_VALID_FS))
2675                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2676
2677                         ext3_mark_recovery_complete(sb, es);
2678                 } else {
2679                         __le32 ret;
2680                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2681                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2682                                 ext3_msg(sb, KERN_WARNING,
2683                                         "warning: couldn't remount RDWR "
2684                                         "because of unsupported optional "
2685                                         "features (%x)", le32_to_cpu(ret));
2686                                 err = -EROFS;
2687                                 goto restore_opts;
2688                         }
2689
2690                         /*
2691                          * If we have an unprocessed orphan list hanging
2692                          * around from a previously readonly bdev mount,
2693                          * require a full umount & mount for now.
2694                          */
2695                         if (es->s_last_orphan) {
2696                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2697                                        "remount RDWR because of unprocessed "
2698                                        "orphan inode list.  Please "
2699                                        "umount & mount instead.");
2700                                 err = -EINVAL;
2701                                 goto restore_opts;
2702                         }
2703
2704                         /*
2705                          * Mounting a RDONLY partition read-write, so reread
2706                          * and store the current valid flag.  (It may have
2707                          * been changed by e2fsck since we originally mounted
2708                          * the partition.)
2709                          */
2710                         ext3_clear_journal_err(sb, es);
2711                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2712                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2713                                 goto restore_opts;
2714                         if (!ext3_setup_super (sb, es, 0))
2715                                 sb->s_flags &= ~MS_RDONLY;
2716                         enable_quota = 1;
2717                 }
2718         }
2719 #ifdef CONFIG_QUOTA
2720         /* Release old quota file names */
2721         for (i = 0; i < MAXQUOTAS; i++)
2722                 kfree(old_opts.s_qf_names[i]);
2723 #endif
2724         if (enable_quota)
2725                 dquot_resume(sb, -1);
2726         return 0;
2727 restore_opts:
2728         sb->s_flags = old_sb_flags;
2729         sbi->s_mount_opt = old_opts.s_mount_opt;
2730         sbi->s_resuid = old_opts.s_resuid;
2731         sbi->s_resgid = old_opts.s_resgid;
2732         sbi->s_commit_interval = old_opts.s_commit_interval;
2733 #ifdef CONFIG_QUOTA
2734         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2735         for (i = 0; i < MAXQUOTAS; i++) {
2736                 kfree(sbi->s_qf_names[i]);
2737                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2738         }
2739 #endif
2740         return err;
2741 }
2742
2743 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2744 {
2745         struct super_block *sb = dentry->d_sb;
2746         struct ext3_sb_info *sbi = EXT3_SB(sb);
2747         struct ext3_super_block *es = sbi->s_es;
2748         u64 fsid;
2749
2750         if (test_opt(sb, MINIX_DF)) {
2751                 sbi->s_overhead_last = 0;
2752         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2753                 unsigned long ngroups = sbi->s_groups_count, i;
2754                 ext3_fsblk_t overhead = 0;
2755                 smp_rmb();
2756
2757                 /*
2758                  * Compute the overhead (FS structures).  This is constant
2759                  * for a given filesystem unless the number of block groups
2760                  * changes so we cache the previous value until it does.
2761                  */
2762
2763                 /*
2764                  * All of the blocks before first_data_block are
2765                  * overhead
2766                  */
2767                 overhead = le32_to_cpu(es->s_first_data_block);
2768
2769                 /*
2770                  * Add the overhead attributed to the superblock and
2771                  * block group descriptors.  If the sparse superblocks
2772                  * feature is turned on, then not all groups have this.
2773                  */
2774                 for (i = 0; i < ngroups; i++) {
2775                         overhead += ext3_bg_has_super(sb, i) +
2776                                 ext3_bg_num_gdb(sb, i);
2777                         cond_resched();
2778                 }
2779
2780                 /*
2781                  * Every block group has an inode bitmap, a block
2782                  * bitmap, and an inode table.
2783                  */
2784                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2785                 sbi->s_overhead_last = overhead;
2786                 smp_wmb();
2787                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2788         }
2789
2790         buf->f_type = EXT3_SUPER_MAGIC;
2791         buf->f_bsize = sb->s_blocksize;
2792         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2793         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2794         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2795         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2796                 buf->f_bavail = 0;
2797         buf->f_files = le32_to_cpu(es->s_inodes_count);
2798         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2799         buf->f_namelen = EXT3_NAME_LEN;
2800         fsid = le64_to_cpup((void *)es->s_uuid) ^
2801                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2802         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2803         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2804         return 0;
2805 }
2806
2807 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2808  * is locked for write. Otherwise the are possible deadlocks:
2809  * Process 1                         Process 2
2810  * ext3_create()                     quota_sync()
2811  *   journal_start()                   write_dquot()
2812  *   dquot_initialize()                       down(dqio_mutex)
2813  *     down(dqio_mutex)                    journal_start()
2814  *
2815  */
2816
2817 #ifdef CONFIG_QUOTA
2818
2819 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2820 {
2821         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
2822 }
2823
2824 static int ext3_write_dquot(struct dquot *dquot)
2825 {
2826         int ret, err;
2827         handle_t *handle;
2828         struct inode *inode;
2829
2830         inode = dquot_to_inode(dquot);
2831         handle = ext3_journal_start(inode,
2832                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2833         if (IS_ERR(handle))
2834                 return PTR_ERR(handle);
2835         ret = dquot_commit(dquot);
2836         err = ext3_journal_stop(handle);
2837         if (!ret)
2838                 ret = err;
2839         return ret;
2840 }
2841
2842 static int ext3_acquire_dquot(struct dquot *dquot)
2843 {
2844         int ret, err;
2845         handle_t *handle;
2846
2847         handle = ext3_journal_start(dquot_to_inode(dquot),
2848                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2849         if (IS_ERR(handle))
2850                 return PTR_ERR(handle);
2851         ret = dquot_acquire(dquot);
2852         err = ext3_journal_stop(handle);
2853         if (!ret)
2854                 ret = err;
2855         return ret;
2856 }
2857
2858 static int ext3_release_dquot(struct dquot *dquot)
2859 {
2860         int ret, err;
2861         handle_t *handle;
2862
2863         handle = ext3_journal_start(dquot_to_inode(dquot),
2864                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2865         if (IS_ERR(handle)) {
2866                 /* Release dquot anyway to avoid endless cycle in dqput() */
2867                 dquot_release(dquot);
2868                 return PTR_ERR(handle);
2869         }
2870         ret = dquot_release(dquot);
2871         err = ext3_journal_stop(handle);
2872         if (!ret)
2873                 ret = err;
2874         return ret;
2875 }
2876
2877 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2878 {
2879         /* Are we journaling quotas? */
2880         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2881             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2882                 dquot_mark_dquot_dirty(dquot);
2883                 return ext3_write_dquot(dquot);
2884         } else {
2885                 return dquot_mark_dquot_dirty(dquot);
2886         }
2887 }
2888
2889 static int ext3_write_info(struct super_block *sb, int type)
2890 {
2891         int ret, err;
2892         handle_t *handle;
2893
2894         /* Data block + inode block */
2895         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2896         if (IS_ERR(handle))
2897                 return PTR_ERR(handle);
2898         ret = dquot_commit_info(sb, type);
2899         err = ext3_journal_stop(handle);
2900         if (!ret)
2901                 ret = err;
2902         return ret;
2903 }
2904
2905 /*
2906  * Turn on quotas during mount time - we need to find
2907  * the quota file and such...
2908  */
2909 static int ext3_quota_on_mount(struct super_block *sb, int type)
2910 {
2911         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2912                                         EXT3_SB(sb)->s_jquota_fmt, type);
2913 }
2914
2915 /*
2916  * Standard function to be called on quota_on
2917  */
2918 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2919                          struct path *path)
2920 {
2921         int err;
2922
2923         if (!test_opt(sb, QUOTA))
2924                 return -EINVAL;
2925
2926         /* Quotafile not on the same filesystem? */
2927         if (path->dentry->d_sb != sb)
2928                 return -EXDEV;
2929         /* Journaling quota? */
2930         if (EXT3_SB(sb)->s_qf_names[type]) {
2931                 /* Quotafile not of fs root? */
2932                 if (path->dentry->d_parent != sb->s_root)
2933                         ext3_msg(sb, KERN_WARNING,
2934                                 "warning: Quota file not on filesystem root. "
2935                                 "Journaled quota will not work.");
2936         }
2937
2938         /*
2939          * When we journal data on quota file, we have to flush journal to see
2940          * all updates to the file when we bypass pagecache...
2941          */
2942         if (ext3_should_journal_data(path->dentry->d_inode)) {
2943                 /*
2944                  * We don't need to lock updates but journal_flush() could
2945                  * otherwise be livelocked...
2946                  */
2947                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2948                 err = journal_flush(EXT3_SB(sb)->s_journal);
2949                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2950                 if (err)
2951                         return err;
2952         }
2953
2954         return dquot_quota_on(sb, type, format_id, path);
2955 }
2956
2957 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2958  * acquiring the locks... As quota files are never truncated and quota code
2959  * itself serializes the operations (and no one else should touch the files)
2960  * we don't have to be afraid of races */
2961 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2962                                size_t len, loff_t off)
2963 {
2964         struct inode *inode = sb_dqopt(sb)->files[type];
2965         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2966         int err = 0;
2967         int offset = off & (sb->s_blocksize - 1);
2968         int tocopy;
2969         size_t toread;
2970         struct buffer_head *bh;
2971         loff_t i_size = i_size_read(inode);
2972
2973         if (off > i_size)
2974                 return 0;
2975         if (off+len > i_size)
2976                 len = i_size-off;
2977         toread = len;
2978         while (toread > 0) {
2979                 tocopy = sb->s_blocksize - offset < toread ?
2980                                 sb->s_blocksize - offset : toread;
2981                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2982                 if (err)
2983                         return err;
2984                 if (!bh)        /* A hole? */
2985                         memset(data, 0, tocopy);
2986                 else
2987                         memcpy(data, bh->b_data+offset, tocopy);
2988                 brelse(bh);
2989                 offset = 0;
2990                 toread -= tocopy;
2991                 data += tocopy;
2992                 blk++;
2993         }
2994         return len;
2995 }
2996
2997 /* Write to quotafile (we know the transaction is already started and has
2998  * enough credits) */
2999 static ssize_t ext3_quota_write(struct super_block *sb, int type,
3000                                 const char *data, size_t len, loff_t off)
3001 {
3002         struct inode *inode = sb_dqopt(sb)->files[type];
3003         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3004         int err = 0;
3005         int offset = off & (sb->s_blocksize - 1);
3006         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
3007         struct buffer_head *bh;
3008         handle_t *handle = journal_current_handle();
3009
3010         if (!handle) {
3011                 ext3_msg(sb, KERN_WARNING,
3012                         "warning: quota write (off=%llu, len=%llu)"
3013                         " cancelled because transaction is not started.",
3014                         (unsigned long long)off, (unsigned long long)len);
3015                 return -EIO;
3016         }
3017
3018         /*
3019          * Since we account only one data block in transaction credits,
3020          * then it is impossible to cross a block boundary.
3021          */
3022         if (sb->s_blocksize - offset < len) {
3023                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3024                         " cancelled because not block aligned",
3025                         (unsigned long long)off, (unsigned long long)len);
3026                 return -EIO;
3027         }
3028         bh = ext3_bread(handle, inode, blk, 1, &err);
3029         if (!bh)
3030                 goto out;
3031         if (journal_quota) {
3032                 err = ext3_journal_get_write_access(handle, bh);
3033                 if (err) {
3034                         brelse(bh);
3035                         goto out;
3036                 }
3037         }
3038         lock_buffer(bh);
3039         memcpy(bh->b_data+offset, data, len);
3040         flush_dcache_page(bh->b_page);
3041         unlock_buffer(bh);
3042         if (journal_quota)
3043                 err = ext3_journal_dirty_metadata(handle, bh);
3044         else {
3045                 /* Always do at least ordered writes for quotas */
3046                 err = ext3_journal_dirty_data(handle, bh);
3047                 mark_buffer_dirty(bh);
3048         }
3049         brelse(bh);
3050 out:
3051         if (err)
3052                 return err;
3053         if (inode->i_size < off + len) {
3054                 i_size_write(inode, off + len);
3055                 EXT3_I(inode)->i_disksize = inode->i_size;
3056         }
3057         inode->i_version++;
3058         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3059         ext3_mark_inode_dirty(handle, inode);
3060         return len;
3061 }
3062
3063 #endif
3064
3065 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3066         int flags, const char *dev_name, void *data)
3067 {
3068         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3069 }
3070
3071 static struct file_system_type ext3_fs_type = {
3072         .owner          = THIS_MODULE,
3073         .name           = "ext3",
3074         .mount          = ext3_mount,
3075         .kill_sb        = kill_block_super,
3076         .fs_flags       = FS_REQUIRES_DEV,
3077 };
3078 MODULE_ALIAS_FS("ext3");
3079
3080 static int __init init_ext3_fs(void)
3081 {
3082         int err = init_ext3_xattr();
3083         if (err)
3084                 return err;
3085         err = init_inodecache();
3086         if (err)
3087                 goto out1;
3088         err = register_filesystem(&ext3_fs_type);
3089         if (err)
3090                 goto out;
3091         return 0;
3092 out:
3093         destroy_inodecache();
3094 out1:
3095         exit_ext3_xattr();
3096         return err;
3097 }
3098
3099 static void __exit exit_ext3_fs(void)
3100 {
3101         unregister_filesystem(&ext3_fs_type);
3102         destroy_inodecache();
3103         exit_ext3_xattr();
3104 }
3105
3106 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3107 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3108 MODULE_LICENSE("GPL");
3109 module_init(init_ext3_fs)
3110 module_exit(exit_ext3_fs)