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