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