fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/parser.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/statfs.h>
  24 #include <linux/random.h>
  25 #include <linux/mount.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/log2.h>
  29 #include <linux/cleancache.h>
  30
  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  * The only special thing we need to do here is to make sure that all
  69  * journal_end calls result in the superblock being marked dirty, so
  70  * that sync() will call the filesystem's write_super callback if
  71  * appropriate.
  72  */
  73 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  74 {
  75         journal_t *journal;
  76
  77         if (sb->s_flags & MS_RDONLY)
  78                 return ERR_PTR(-EROFS);
  79
  80         /* Special case here: if the journal has aborted behind our
  81          * backs (eg. EIO in the commit thread), then we still need to
  82          * take the FS itself readonly cleanly. */
  83         journal = EXT3_SB(sb)->s_journal;
  84         if (is_journal_aborted(journal)) {
  85                 ext3_abort(sb, __func__,
  86                            "Detected aborted journal");
  87                 return ERR_PTR(-EROFS);
  88         }
  89
  90         return journal_start(journal, nblocks);
  91 }
  92
  93 /*
  94  * The only special thing we need to do here is to make sure that all
  95  * journal_stop calls result in the superblock being marked dirty, so
  96  * that sync() will call the filesystem's write_super callback if
  97  * appropriate.
  98  */
  99 int __ext3_journal_stop(const char *where, handle_t *handle)
 100 {
 101         struct super_block *sb;
 102         int err;
 103         int rc;
 104
 105         sb = handle->h_transaction->t_journal->j_private;
 106         err = handle->h_err;
 107         rc = journal_stop(handle);
 108
 109         if (!err)
 110                 err = rc;
 111         if (err)
 112                 __ext3_std_error(sb, where, err);
 113         return err;
 114 }
 115
 116 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 117                 struct buffer_head *bh, handle_t *handle, int err)
 118 {
 119         char nbuf[16];
 120         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 121
 122         if (bh)
 123                 BUFFER_TRACE(bh, "abort");
 124
 125         if (!handle->h_err)
 126                 handle->h_err = err;
 127
 128         if (is_handle_aborted(handle))
 129                 return;
 130
 131         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 132                 caller, errstr, err_fn);
 133
 134         journal_abort_handle(handle);
 135 }
 136
 137 void ext3_msg(struct super_block *sb, const char *prefix,
 138                 const char *fmt, ...)
 139 {
 140         struct va_format vaf;
 141         va_list args;
 142
 143         va_start(args, fmt);
 144
 145         vaf.fmt = fmt;
 146         vaf.va = &args;
 147
 148         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 149
 150         va_end(args);
 151 }
 152
 153 /* Deal with the reporting of failure conditions on a filesystem such as
 154  * inconsistencies detected or read IO failures.
 155  *
 156  * On ext2, we can store the error state of the filesystem in the
 157  * superblock.  That is not possible on ext3, because we may have other
 158  * write ordering constraints on the superblock which prevent us from
 159  * writing it out straight away; and given that the journal is about to
 160  * be aborted, we can't rely on the current, or future, transactions to
 161  * write out the superblock safely.
 162  *
 163  * We'll just use the journal_abort() error code to record an error in
 164  * the journal instead.  On recovery, the journal will complain about
 165  * that error until we've noted it down and cleared it.
 166  */
 167
 168 static void ext3_handle_error(struct super_block *sb)
 169 {
 170         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 171
 172         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 173         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 174
 175         if (sb->s_flags & MS_RDONLY)
 176                 return;
 177
 178         if (!test_opt (sb, ERRORS_CONT)) {
 179                 journal_t *journal = EXT3_SB(sb)->s_journal;
 180
 181                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 182                 if (journal)
 183                         journal_abort(journal, -EIO);
 184         }
 185         if (test_opt (sb, ERRORS_RO)) {
 186                 ext3_msg(sb, KERN_CRIT,
 187                         "error: remounting filesystem read-only");
 188                 sb->s_flags |= MS_RDONLY;
 189         }
 190         ext3_commit_super(sb, es, 1);
 191         if (test_opt(sb, ERRORS_PANIC))
 192                 panic("EXT3-fs (%s): panic forced after error\n",
 193                         sb->s_id);
 194 }
 195
 196 void ext3_error(struct super_block *sb, const char *function,
 197                 const char *fmt, ...)
 198 {
 199         struct va_format vaf;
 200         va_list args;
 201
 202         va_start(args, fmt);
 203
 204         vaf.fmt = fmt;
 205         vaf.va = &args;
 206
 207         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 208                sb->s_id, function, &vaf);
 209
 210         va_end(args);
 211
 212         ext3_handle_error(sb);
 213 }
 214
 215 static const char *ext3_decode_error(struct super_block * sb, int errno,
 216                                      char nbuf[16])
 217 {
 218         char *errstr = NULL;
 219
 220         switch (errno) {
 221         case -EIO:
 222                 errstr = "IO failure";
 223                 break;
 224         case -ENOMEM:
 225                 errstr = "Out of memory";
 226                 break;
 227         case -EROFS:
 228                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 229                         errstr = "Journal has aborted";
 230                 else
 231                         errstr = "Readonly filesystem";
 232                 break;
 233         default:
 234                 /* If the caller passed in an extra buffer for unknown
 235                  * errors, textualise them now.  Else we just return
 236                  * NULL. */
 237                 if (nbuf) {
 238                         /* Check for truncated error codes... */
 239                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 240                                 errstr = nbuf;
 241                 }
 242                 break;
 243         }
 244
 245         return errstr;
 246 }
 247
 248 /* __ext3_std_error decodes expected errors from journaling functions
 249  * automatically and invokes the appropriate error response.  */
 250
 251 void __ext3_std_error (struct super_block * sb, const char * function,
 252                        int errno)
 253 {
 254         char nbuf[16];
 255         const char *errstr;
 256
 257         /* Special case: if the error is EROFS, and we're not already
 258          * inside a transaction, then there's really no point in logging
 259          * an error. */
 260         if (errno == -EROFS && journal_current_handle() == NULL &&
 261             (sb->s_flags & MS_RDONLY))
 262                 return;
 263
 264         errstr = ext3_decode_error(sb, errno, nbuf);
 265         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 266
 267         ext3_handle_error(sb);
 268 }
 269
 270 /*
 271  * ext3_abort is a much stronger failure handler than ext3_error.  The
 272  * abort function may be used to deal with unrecoverable failures such
 273  * as journal IO errors or ENOMEM at a critical moment in log management.
 274  *
 275  * We unconditionally force the filesystem into an ABORT|READONLY state,
 276  * unless the error response on the fs has been set to panic in which
 277  * case we take the easy way out and panic immediately.
 278  */
 279
 280 void ext3_abort(struct super_block *sb, const char *function,
 281                  const char *fmt, ...)
 282 {
 283         struct va_format vaf;
 284         va_list args;
 285
 286         va_start(args, fmt);
 287
 288         vaf.fmt = fmt;
 289         vaf.va = &args;
 290
 291         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 292                sb->s_id, function, &vaf);
 293
 294         va_end(args);
 295
 296         if (test_opt(sb, ERRORS_PANIC))
 297                 panic("EXT3-fs: panic from previous error\n");
 298
 299         if (sb->s_flags & MS_RDONLY)
 300                 return;
 301
 302         ext3_msg(sb, KERN_CRIT,
 303                 "error: remounting filesystem read-only");
 304         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 305         sb->s_flags |= MS_RDONLY;
 306         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 307         if (EXT3_SB(sb)->s_journal)
 308                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 309 }
 310
 311 void ext3_warning(struct super_block *sb, const char *function,
 312                   const char *fmt, ...)
 313 {
 314         struct va_format vaf;
 315         va_list args;
 316
 317         va_start(args, fmt);
 318
 319         vaf.fmt = fmt;
 320         vaf.va = &args;
 321
 322         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 323                sb->s_id, function, &vaf);
 324
 325         va_end(args);
 326 }
 327
 328 void ext3_update_dynamic_rev(struct super_block *sb)
 329 {
 330         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 331
 332         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 333                 return;
 334
 335         ext3_msg(sb, KERN_WARNING,
 336                 "warning: updating to rev %d because of "
 337                 "new feature flag, running e2fsck is recommended",
 338                 EXT3_DYNAMIC_REV);
 339
 340         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 341         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 342         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 343         /* leave es->s_feature_*compat flags alone */
 344         /* es->s_uuid will be set by e2fsck if empty */
 345
 346         /*
 347          * The rest of the superblock fields should be zero, and if not it
 348          * means they are likely already in use, so leave them alone.  We
 349          * can leave it up to e2fsck to clean up any inconsistencies there.
 350          */
 351 }
 352
 353 /*
 354  * Open the external journal device
 355  */
 356 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 357 {
 358         struct block_device *bdev;
 359         char b[BDEVNAME_SIZE];
 360
 361         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 362         if (IS_ERR(bdev))
 363                 goto fail;
 364         return bdev;
 365
 366 fail:
 367         ext3_msg(sb, "error: failed to open journal device %s: %ld",
 368                 __bdevname(dev, b), PTR_ERR(bdev));
 369
 370         return NULL;
 371 }
 372
 373 /*
 374  * Release the journal device
 375  */
 376 static int ext3_blkdev_put(struct block_device *bdev)
 377 {
 378         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 379 }
 380
 381 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 382 {
 383         struct block_device *bdev;
 384         int ret = -ENODEV;
 385
 386         bdev = sbi->journal_bdev;
 387         if (bdev) {
 388                 ret = ext3_blkdev_put(bdev);
 389                 sbi->journal_bdev = NULL;
 390         }
 391         return ret;
 392 }
 393
 394 static inline struct inode *orphan_list_entry(struct list_head *l)
 395 {
 396         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 397 }
 398
 399 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 400 {
 401         struct list_head *l;
 402
 403         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 404                le32_to_cpu(sbi->s_es->s_last_orphan));
 405
 406         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 407         list_for_each(l, &sbi->s_orphan) {
 408                 struct inode *inode = orphan_list_entry(l);
 409                 ext3_msg(sb, KERN_ERR, "  "
 410                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 411                        inode->i_sb->s_id, inode->i_ino, inode,
 412                        inode->i_mode, inode->i_nlink,
 413                        NEXT_ORPHAN(inode));
 414         }
 415 }
 416
 417 static void ext3_put_super (struct super_block * sb)
 418 {
 419         struct ext3_sb_info *sbi = EXT3_SB(sb);
 420         struct ext3_super_block *es = sbi->s_es;
 421         int i, err;
 422
 423         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 424         ext3_xattr_put_super(sb);
 425         err = journal_destroy(sbi->s_journal);
 426         sbi->s_journal = NULL;
 427         if (err < 0)
 428                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 429
 430         if (!(sb->s_flags & MS_RDONLY)) {
 431                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 432                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 433                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 434                 mark_buffer_dirty(sbi->s_sbh);
 435                 ext3_commit_super(sb, es, 1);
 436         }
 437
 438         for (i = 0; i < sbi->s_gdb_count; i++)
 439                 brelse(sbi->s_group_desc[i]);
 440         kfree(sbi->s_group_desc);
 441         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 442         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 443         percpu_counter_destroy(&sbi->s_dirs_counter);
 444         brelse(sbi->s_sbh);
 445 #ifdef CONFIG_QUOTA
 446         for (i = 0; i < MAXQUOTAS; i++)
 447                 kfree(sbi->s_qf_names[i]);
 448 #endif
 449
 450         /* Debugging code just in case the in-memory inode orphan list
 451          * isn't empty.  The on-disk one can be non-empty if we've
 452          * detected an error and taken the fs readonly, but the
 453          * in-memory list had better be clean by this point. */
 454         if (!list_empty(&sbi->s_orphan))
 455                 dump_orphan_list(sb, sbi);
 456         J_ASSERT(list_empty(&sbi->s_orphan));
 457
 458         invalidate_bdev(sb->s_bdev);
 459         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 460                 /*
 461                  * Invalidate the journal device's buffers.  We don't want them
 462                  * floating about in memory - the physical journal device may
 463                  * hotswapped, and it breaks the `ro-after' testing code.
 464                  */
 465                 sync_blockdev(sbi->journal_bdev);
 466                 invalidate_bdev(sbi->journal_bdev);
 467                 ext3_blkdev_remove(sbi);
 468         }
 469         sb->s_fs_info = NULL;
 470         kfree(sbi->s_blockgroup_lock);
 471         kfree(sbi);
 472 }
 473
 474 static struct kmem_cache *ext3_inode_cachep;
 475
 476 /*
 477  * Called inside transaction, so use GFP_NOFS
 478  */
 479 static struct inode *ext3_alloc_inode(struct super_block *sb)
 480 {
 481         struct ext3_inode_info *ei;
 482
 483         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 484         if (!ei)
 485                 return NULL;
 486         ei->i_block_alloc_info = NULL;
 487         ei->vfs_inode.i_version = 1;
 488         atomic_set(&ei->i_datasync_tid, 0);
 489         atomic_set(&ei->i_sync_tid, 0);
 490         return &ei->vfs_inode;
 491 }
 492
 493 static int ext3_drop_inode(struct inode *inode)
 494 {
 495         int drop = generic_drop_inode(inode);
 496
 497         trace_ext3_drop_inode(inode, drop);
 498         return drop;
 499 }
 500
 501 static void ext3_i_callback(struct rcu_head *head)
 502 {
 503         struct inode *inode = container_of(head, struct inode, i_rcu);
 504         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 505 }
 506
 507 static void ext3_destroy_inode(struct inode *inode)
 508 {
 509         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 510                 printk("EXT3 Inode %p: orphan list check failed!\n",
 511                         EXT3_I(inode));
 512                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 513                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 514                                 false);
 515                 dump_stack();
 516         }
 517         call_rcu(&inode->i_rcu, ext3_i_callback);
 518 }
 519
 520 static void init_once(void *foo)
 521 {
 522         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 523
 524         INIT_LIST_HEAD(&ei->i_orphan);
 525 #ifdef CONFIG_EXT3_FS_XATTR
 526         init_rwsem(&ei->xattr_sem);
 527 #endif
 528         mutex_init(&ei->truncate_mutex);
 529         inode_init_once(&ei->vfs_inode);
 530 }
 531
 532 static int init_inodecache(void)
 533 {
 534         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 535                                              sizeof(struct ext3_inode_info),
 536                                              0, (SLAB_RECLAIM_ACCOUNT|
 537                                                 SLAB_MEM_SPREAD),
 538                                              init_once);
 539         if (ext3_inode_cachep == NULL)
 540                 return -ENOMEM;
 541         return 0;
 542 }
 543
 544 static void destroy_inodecache(void)
 545 {
 546         kmem_cache_destroy(ext3_inode_cachep);
 547 }
 548
 549 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 550 {
 551 #if defined(CONFIG_QUOTA)
 552         struct ext3_sb_info *sbi = EXT3_SB(sb);
 553
 554         if (sbi->s_jquota_fmt) {
 555                 char *fmtname = "";
 556
 557                 switch (sbi->s_jquota_fmt) {
 558                 case QFMT_VFS_OLD:
 559                         fmtname = "vfsold";
 560                         break;
 561                 case QFMT_VFS_V0:
 562                         fmtname = "vfsv0";
 563                         break;
 564                 case QFMT_VFS_V1:
 565                         fmtname = "vfsv1";
 566                         break;
 567                 }
 568                 seq_printf(seq, ",jqfmt=%s", fmtname);
 569         }
 570
 571         if (sbi->s_qf_names[USRQUOTA])
 572                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 573
 574         if (sbi->s_qf_names[GRPQUOTA])
 575                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 576
 577         if (test_opt(sb, USRQUOTA))
 578                 seq_puts(seq, ",usrquota");
 579
 580         if (test_opt(sb, GRPQUOTA))
 581                 seq_puts(seq, ",grpquota");
 582 #endif
 583 }
 584
 585 static char *data_mode_string(unsigned long mode)
 586 {
 587         switch (mode) {
 588         case EXT3_MOUNT_JOURNAL_DATA:
 589                 return "journal";
 590         case EXT3_MOUNT_ORDERED_DATA:
 591                 return "ordered";
 592         case EXT3_MOUNT_WRITEBACK_DATA:
 593                 return "writeback";
 594         }
 595         return "unknown";
 596 }
 597
 598 /*
 599  * Show an option if
 600  *  - it's set to a non-default value OR
 601  *  - if the per-sb default is different from the global default
 602  */
 603 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 604 {
 605         struct super_block *sb = root->d_sb;
 606         struct ext3_sb_info *sbi = EXT3_SB(sb);
 607         struct ext3_super_block *es = sbi->s_es;
 608         unsigned long def_mount_opts;
 609
 610         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 611
 612         if (sbi->s_sb_block != 1)
 613                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 614         if (test_opt(sb, MINIX_DF))
 615                 seq_puts(seq, ",minixdf");
 616         if (test_opt(sb, GRPID))
 617                 seq_puts(seq, ",grpid");
 618         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 619                 seq_puts(seq, ",nogrpid");
 620         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
 621             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 622                 seq_printf(seq, ",resuid=%u",
 623                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
 624         }
 625         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
 626             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 627                 seq_printf(seq, ",resgid=%u",
 628                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
 629         }
 630         if (test_opt(sb, ERRORS_RO)) {
 631                 int def_errors = le16_to_cpu(es->s_errors);
 632
 633                 if (def_errors == EXT3_ERRORS_PANIC ||
 634                     def_errors == EXT3_ERRORS_CONTINUE) {
 635                         seq_puts(seq, ",errors=remount-ro");
 636                 }
 637         }
 638         if (test_opt(sb, ERRORS_CONT))
 639                 seq_puts(seq, ",errors=continue");
 640         if (test_opt(sb, ERRORS_PANIC))
 641                 seq_puts(seq, ",errors=panic");
 642         if (test_opt(sb, NO_UID32))
 643                 seq_puts(seq, ",nouid32");
 644         if (test_opt(sb, DEBUG))
 645                 seq_puts(seq, ",debug");
 646 #ifdef CONFIG_EXT3_FS_XATTR
 647         if (test_opt(sb, XATTR_USER))
 648                 seq_puts(seq, ",user_xattr");
 649         if (!test_opt(sb, XATTR_USER) &&
 650             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 651                 seq_puts(seq, ",nouser_xattr");
 652         }
 653 #endif
 654 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 655         if (test_opt(sb, POSIX_ACL))
 656                 seq_puts(seq, ",acl");
 657         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 658                 seq_puts(seq, ",noacl");
 659 #endif
 660         if (!test_opt(sb, RESERVATION))
 661                 seq_puts(seq, ",noreservation");
 662         if (sbi->s_commit_interval) {
 663                 seq_printf(seq, ",commit=%u",
 664                            (unsigned) (sbi->s_commit_interval / HZ));
 665         }
 666
 667         /*
 668          * Always display barrier state so it's clear what the status is.
 669          */
 670         seq_puts(seq, ",barrier=");
 671         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 672         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 673         if (test_opt(sb, DATA_ERR_ABORT))
 674                 seq_puts(seq, ",data_err=abort");
 675
 676         if (test_opt(sb, NOLOAD))
 677                 seq_puts(seq, ",norecovery");
 678
 679         ext3_show_quota_options(seq, sb);
 680
 681         return 0;
 682 }
 683
 684
 685 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 686                 u64 ino, u32 generation)
 687 {
 688         struct inode *inode;
 689
 690         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 691                 return ERR_PTR(-ESTALE);
 692         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 693                 return ERR_PTR(-ESTALE);
 694
 695         /* iget isn't really right if the inode is currently unallocated!!
 696          *
 697          * ext3_read_inode will return a bad_inode if the inode had been
 698          * deleted, so we should be safe.
 699          *
 700          * Currently we don't know the generation for parent directory, so
 701          * a generation of 0 means "accept any"
 702          */
 703         inode = ext3_iget(sb, ino);
 704         if (IS_ERR(inode))
 705                 return ERR_CAST(inode);
 706         if (generation && inode->i_generation != generation) {
 707                 iput(inode);
 708                 return ERR_PTR(-ESTALE);
 709         }
 710
 711         return inode;
 712 }
 713
 714 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 715                 int fh_len, int fh_type)
 716 {
 717         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 718                                     ext3_nfs_get_inode);
 719 }
 720
 721 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 722                 int fh_len, int fh_type)
 723 {
 724         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 725                                     ext3_nfs_get_inode);
 726 }
 727
 728 /*
 729  * Try to release metadata pages (indirect blocks, directories) which are
 730  * mapped via the block device.  Since these pages could have journal heads
 731  * which would prevent try_to_free_buffers() from freeing them, we must use
 732  * jbd layer's try_to_free_buffers() function to release them.
 733  */
 734 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 735                                  gfp_t wait)
 736 {
 737         journal_t *journal = EXT3_SB(sb)->s_journal;
 738
 739         WARN_ON(PageChecked(page));
 740         if (!page_has_buffers(page))
 741                 return 0;
 742         if (journal)
 743                 return journal_try_to_free_buffers(journal, page,
 744                                                    wait & ~__GFP_WAIT);
 745         return try_to_free_buffers(page);
 746 }
 747
 748 #ifdef CONFIG_QUOTA
 749 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 750 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 751
 752 static int ext3_write_dquot(struct dquot *dquot);
 753 static int ext3_acquire_dquot(struct dquot *dquot);
 754 static int ext3_release_dquot(struct dquot *dquot);
 755 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 756 static int ext3_write_info(struct super_block *sb, int type);
 757 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 758                          struct path *path);
 759 static int ext3_quota_on_mount(struct super_block *sb, int type);
 760 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 761                                size_t len, loff_t off);
 762 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 763                                 const char *data, size_t len, loff_t off);
 764
 765 static const struct dquot_operations ext3_quota_operations = {
 766         .write_dquot    = ext3_write_dquot,
 767         .acquire_dquot  = ext3_acquire_dquot,
 768         .release_dquot  = ext3_release_dquot,
 769         .mark_dirty     = ext3_mark_dquot_dirty,
 770         .write_info     = ext3_write_info,
 771         .alloc_dquot    = dquot_alloc,
 772         .destroy_dquot  = dquot_destroy,
 773 };
 774
 775 static const struct quotactl_ops ext3_qctl_operations = {
 776         .quota_on       = ext3_quota_on,
 777         .quota_off      = dquot_quota_off,
 778         .quota_sync     = dquot_quota_sync,
 779         .get_info       = dquot_get_dqinfo,
 780         .set_info       = dquot_set_dqinfo,
 781         .get_dqblk      = dquot_get_dqblk,
 782         .set_dqblk      = dquot_set_dqblk
 783 };
 784 #endif
 785
 786 static const struct super_operations ext3_sops = {
 787         .alloc_inode    = ext3_alloc_inode,
 788         .destroy_inode  = ext3_destroy_inode,
 789         .write_inode    = ext3_write_inode,
 790         .dirty_inode    = ext3_dirty_inode,
 791         .drop_inode     = ext3_drop_inode,
 792         .evict_inode    = ext3_evict_inode,
 793         .put_super      = ext3_put_super,
 794         .sync_fs        = ext3_sync_fs,
 795         .freeze_fs      = ext3_freeze,
 796         .unfreeze_fs    = ext3_unfreeze,
 797         .statfs         = ext3_statfs,
 798         .remount_fs     = ext3_remount,
 799         .show_options   = ext3_show_options,
 800 #ifdef CONFIG_QUOTA
 801         .quota_read     = ext3_quota_read,
 802         .quota_write    = ext3_quota_write,
 803 #endif
 804         .bdev_try_to_free_page = bdev_try_to_free_page,
 805 };
 806
 807 static const struct export_operations ext3_export_ops = {
 808         .fh_to_dentry = ext3_fh_to_dentry,
 809         .fh_to_parent = ext3_fh_to_parent,
 810         .get_parent = ext3_get_parent,
 811 };
 812
 813 enum {
 814         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 815         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 816         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 817         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 818         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 819         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 820         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 821         Opt_data_err_abort, Opt_data_err_ignore,
 822         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 823         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 824         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 825         Opt_resize, Opt_usrquota, Opt_grpquota
 826 };
 827
 828 static const match_table_t tokens = {
 829         {Opt_bsd_df, "bsddf"},
 830         {Opt_minix_df, "minixdf"},
 831         {Opt_grpid, "grpid"},
 832         {Opt_grpid, "bsdgroups"},
 833         {Opt_nogrpid, "nogrpid"},
 834         {Opt_nogrpid, "sysvgroups"},
 835         {Opt_resgid, "resgid=%u"},
 836         {Opt_resuid, "resuid=%u"},
 837         {Opt_sb, "sb=%u"},
 838         {Opt_err_cont, "errors=continue"},
 839         {Opt_err_panic, "errors=panic"},
 840         {Opt_err_ro, "errors=remount-ro"},
 841         {Opt_nouid32, "nouid32"},
 842         {Opt_nocheck, "nocheck"},
 843         {Opt_nocheck, "check=none"},
 844         {Opt_debug, "debug"},
 845         {Opt_oldalloc, "oldalloc"},
 846         {Opt_orlov, "orlov"},
 847         {Opt_user_xattr, "user_xattr"},
 848         {Opt_nouser_xattr, "nouser_xattr"},
 849         {Opt_acl, "acl"},
 850         {Opt_noacl, "noacl"},
 851         {Opt_reservation, "reservation"},
 852         {Opt_noreservation, "noreservation"},
 853         {Opt_noload, "noload"},
 854         {Opt_noload, "norecovery"},
 855         {Opt_nobh, "nobh"},
 856         {Opt_bh, "bh"},
 857         {Opt_commit, "commit=%u"},
 858         {Opt_journal_update, "journal=update"},
 859         {Opt_journal_inum, "journal=%u"},
 860         {Opt_journal_dev, "journal_dev=%u"},
 861         {Opt_abort, "abort"},
 862         {Opt_data_journal, "data=journal"},
 863         {Opt_data_ordered, "data=ordered"},
 864         {Opt_data_writeback, "data=writeback"},
 865         {Opt_data_err_abort, "data_err=abort"},
 866         {Opt_data_err_ignore, "data_err=ignore"},
 867         {Opt_offusrjquota, "usrjquota="},
 868         {Opt_usrjquota, "usrjquota=%s"},
 869         {Opt_offgrpjquota, "grpjquota="},
 870         {Opt_grpjquota, "grpjquota=%s"},
 871         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 872         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 873         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 874         {Opt_grpquota, "grpquota"},
 875         {Opt_noquota, "noquota"},
 876         {Opt_quota, "quota"},
 877         {Opt_usrquota, "usrquota"},
 878         {Opt_barrier, "barrier=%u"},
 879         {Opt_barrier, "barrier"},
 880         {Opt_nobarrier, "nobarrier"},
 881         {Opt_resize, "resize"},
 882         {Opt_err, NULL},
 883 };
 884
 885 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 886 {
 887         ext3_fsblk_t    sb_block;
 888         char            *options = (char *) *data;
 889
 890         if (!options || strncmp(options, "sb=", 3) != 0)
 891                 return 1;       /* Default location */
 892         options += 3;
 893         /*todo: use simple_strtoll with >32bit ext3 */
 894         sb_block = simple_strtoul(options, &options, 0);
 895         if (*options && *options != ',') {
 896                 ext3_msg(sb, "error: invalid sb specification: %s",
 897                        (char *) *data);
 898                 return 1;
 899         }
 900         if (*options == ',')
 901                 options++;
 902         *data = (void *) options;
 903         return sb_block;
 904 }
 905
 906 #ifdef CONFIG_QUOTA
 907 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 908 {
 909         struct ext3_sb_info *sbi = EXT3_SB(sb);
 910         char *qname;
 911
 912         if (sb_any_quota_loaded(sb) &&
 913                 !sbi->s_qf_names[qtype]) {
 914                 ext3_msg(sb, KERN_ERR,
 915                         "Cannot change journaled "
 916                         "quota options when quota turned on");
 917                 return 0;
 918         }
 919         qname = match_strdup(args);
 920         if (!qname) {
 921                 ext3_msg(sb, KERN_ERR,
 922                         "Not enough memory for storing quotafile name");
 923                 return 0;
 924         }
 925         if (sbi->s_qf_names[qtype] &&
 926                 strcmp(sbi->s_qf_names[qtype], qname)) {
 927                 ext3_msg(sb, KERN_ERR,
 928                         "%s quota file already specified", QTYPE2NAME(qtype));
 929                 kfree(qname);
 930                 return 0;
 931         }
 932         sbi->s_qf_names[qtype] = qname;
 933         if (strchr(sbi->s_qf_names[qtype], '/')) {
 934                 ext3_msg(sb, KERN_ERR,
 935                         "quotafile must be on filesystem root");
 936                 kfree(sbi->s_qf_names[qtype]);
 937                 sbi->s_qf_names[qtype] = NULL;
 938                 return 0;
 939         }
 940         set_opt(sbi->s_mount_opt, QUOTA);
 941         return 1;
 942 }
 943
 944 static int clear_qf_name(struct super_block *sb, int qtype) {
 945
 946         struct ext3_sb_info *sbi = EXT3_SB(sb);
 947
 948         if (sb_any_quota_loaded(sb) &&
 949                 sbi->s_qf_names[qtype]) {
 950                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 951                         " when quota turned on");
 952                 return 0;
 953         }
 954         /*
 955          * The space will be released later when all options are confirmed
 956          * to be correct
 957          */
 958         sbi->s_qf_names[qtype] = NULL;
 959         return 1;
 960 }
 961 #endif
 962
 963 static int parse_options (char *options, struct super_block *sb,
 964                           unsigned int *inum, unsigned long *journal_devnum,
 965                           ext3_fsblk_t *n_blocks_count, int is_remount)
 966 {
 967         struct ext3_sb_info *sbi = EXT3_SB(sb);
 968         char * p;
 969         substring_t args[MAX_OPT_ARGS];
 970         int data_opt = 0;
 971         int option;
 972         kuid_t uid;
 973         kgid_t gid;
 974 #ifdef CONFIG_QUOTA
 975         int qfmt;
 976 #endif
 977
 978         if (!options)
 979                 return 1;
 980
 981         while ((p = strsep (&options, ",")) != NULL) {
 982                 int token;
 983                 if (!*p)
 984                         continue;
 985                 /*
 986                  * Initialize args struct so we know whether arg was
 987                  * found; some options take optional arguments.
 988                  */
 989                 args[0].to = args[0].from = 0;
 990                 token = match_token(p, tokens, args);
 991                 switch (token) {
 992                 case Opt_bsd_df:
 993                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 994                         break;
 995                 case Opt_minix_df:
 996                         set_opt (sbi->s_mount_opt, MINIX_DF);
 997                         break;
 998                 case Opt_grpid:
 999                         set_opt (sbi->s_mount_opt, GRPID);
1000                         break;
1001                 case Opt_nogrpid:
1002                         clear_opt (sbi->s_mount_opt, GRPID);
1003                         break;
1004                 case Opt_resuid:
1005                         if (match_int(&args[0], &option))
1006                                 return 0;
1007                         uid = make_kuid(current_user_ns(), option);
1008                         if (!uid_valid(uid)) {
1009                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1010                                 return -1;
1011
1012                         }
1013                         sbi->s_resuid = uid;
1014                         break;
1015                 case Opt_resgid:
1016                         if (match_int(&args[0], &option))
1017                                 return 0;
1018                         gid = make_kgid(current_user_ns(), option);
1019                         if (!gid_valid(gid)) {
1020                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1021                                 return -1;
1022                         }
1023                         sbi->s_resgid = gid;
1024                         break;
1025                 case Opt_sb:
1026                         /* handled by get_sb_block() instead of here */
1027                         /* *sb_block = match_int(&args[0]); */
1028                         break;
1029                 case Opt_err_panic:
1030                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1031                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1032                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1033                         break;
1034                 case Opt_err_ro:
1035                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1036                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1037                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1038                         break;
1039                 case Opt_err_cont:
1040                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1041                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1042                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1043                         break;
1044                 case Opt_nouid32:
1045                         set_opt (sbi->s_mount_opt, NO_UID32);
1046                         break;
1047                 case Opt_nocheck:
1048                         clear_opt (sbi->s_mount_opt, CHECK);
1049                         break;
1050                 case Opt_debug:
1051                         set_opt (sbi->s_mount_opt, DEBUG);
1052                         break;
1053                 case Opt_oldalloc:
1054                         ext3_msg(sb, KERN_WARNING,
1055                                 "Ignoring deprecated oldalloc option");
1056                         break;
1057                 case Opt_orlov:
1058                         ext3_msg(sb, KERN_WARNING,
1059                                 "Ignoring deprecated orlov option");
1060                         break;
1061 #ifdef CONFIG_EXT3_FS_XATTR
1062                 case Opt_user_xattr:
1063                         set_opt (sbi->s_mount_opt, XATTR_USER);
1064                         break;
1065                 case Opt_nouser_xattr:
1066                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1067                         break;
1068 #else
1069                 case Opt_user_xattr:
1070                 case Opt_nouser_xattr:
1071                         ext3_msg(sb, KERN_INFO,
1072                                 "(no)user_xattr options not supported");
1073                         break;
1074 #endif
1075 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1076                 case Opt_acl:
1077                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1078                         break;
1079                 case Opt_noacl:
1080                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1081                         break;
1082 #else
1083                 case Opt_acl:
1084                 case Opt_noacl:
1085                         ext3_msg(sb, KERN_INFO,
1086                                 "(no)acl options not supported");
1087                         break;
1088 #endif
1089                 case Opt_reservation:
1090                         set_opt(sbi->s_mount_opt, RESERVATION);
1091                         break;
1092                 case Opt_noreservation:
1093                         clear_opt(sbi->s_mount_opt, RESERVATION);
1094                         break;
1095                 case Opt_journal_update:
1096                         /* @@@ FIXME */
1097                         /* Eventually we will want to be able to create
1098                            a journal file here.  For now, only allow the
1099                            user to specify an existing inode to be the
1100                            journal file. */
1101                         if (is_remount) {
1102                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1103                                         "journal on remount");
1104                                 return 0;
1105                         }
1106                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1107                         break;
1108                 case Opt_journal_inum:
1109                         if (is_remount) {
1110                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1111                                        "journal on remount");
1112                                 return 0;
1113                         }
1114                         if (match_int(&args[0], &option))
1115                                 return 0;
1116                         *inum = option;
1117                         break;
1118                 case Opt_journal_dev:
1119                         if (is_remount) {
1120                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1121                                        "journal on remount");
1122                                 return 0;
1123                         }
1124                         if (match_int(&args[0], &option))
1125                                 return 0;
1126                         *journal_devnum = option;
1127                         break;
1128                 case Opt_noload:
1129                         set_opt (sbi->s_mount_opt, NOLOAD);
1130                         break;
1131                 case Opt_commit:
1132                         if (match_int(&args[0], &option))
1133                                 return 0;
1134                         if (option < 0)
1135                                 return 0;
1136                         if (option == 0)
1137                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1138                         sbi->s_commit_interval = HZ * option;
1139                         break;
1140                 case Opt_data_journal:
1141                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1142                         goto datacheck;
1143                 case Opt_data_ordered:
1144                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1145                         goto datacheck;
1146                 case Opt_data_writeback:
1147                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1148                 datacheck:
1149                         if (is_remount) {
1150                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1151                                         break;
1152                                 ext3_msg(sb, KERN_ERR,
1153                                         "error: cannot change "
1154                                         "data mode on remount. The filesystem "
1155                                         "is mounted in data=%s mode and you "
1156                                         "try to remount it in data=%s mode.",
1157                                         data_mode_string(test_opt(sb,
1158                                                         DATA_FLAGS)),
1159                                         data_mode_string(data_opt));
1160                                 return 0;
1161                         } else {
1162                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1163                                 sbi->s_mount_opt |= data_opt;
1164                         }
1165                         break;
1166                 case Opt_data_err_abort:
1167                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1168                         break;
1169                 case Opt_data_err_ignore:
1170                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1171                         break;
1172 #ifdef CONFIG_QUOTA
1173                 case Opt_usrjquota:
1174                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1175                                 return 0;
1176                         break;
1177                 case Opt_grpjquota:
1178                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1179                                 return 0;
1180                         break;
1181                 case Opt_offusrjquota:
1182                         if (!clear_qf_name(sb, USRQUOTA))
1183                                 return 0;
1184                         break;
1185                 case Opt_offgrpjquota:
1186                         if (!clear_qf_name(sb, GRPQUOTA))
1187                                 return 0;
1188                         break;
1189                 case Opt_jqfmt_vfsold:
1190                         qfmt = QFMT_VFS_OLD;
1191                         goto set_qf_format;
1192                 case Opt_jqfmt_vfsv0:
1193                         qfmt = QFMT_VFS_V0;
1194                         goto set_qf_format;
1195                 case Opt_jqfmt_vfsv1:
1196                         qfmt = QFMT_VFS_V1;
1197 set_qf_format:
1198                         if (sb_any_quota_loaded(sb) &&
1199                             sbi->s_jquota_fmt != qfmt) {
1200                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1201                                         "journaled quota options when "
1202                                         "quota turned on.");
1203                                 return 0;
1204                         }
1205                         sbi->s_jquota_fmt = qfmt;
1206                         break;
1207                 case Opt_quota:
1208                 case Opt_usrquota:
1209                         set_opt(sbi->s_mount_opt, QUOTA);
1210                         set_opt(sbi->s_mount_opt, USRQUOTA);
1211                         break;
1212                 case Opt_grpquota:
1213                         set_opt(sbi->s_mount_opt, QUOTA);
1214                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1215                         break;
1216                 case Opt_noquota:
1217                         if (sb_any_quota_loaded(sb)) {
1218                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1219                                         "quota options when quota turned on.");
1220                                 return 0;
1221                         }
1222                         clear_opt(sbi->s_mount_opt, QUOTA);
1223                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1224                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1225                         break;
1226 #else
1227                 case Opt_quota:
1228                 case Opt_usrquota:
1229                 case Opt_grpquota:
1230                         ext3_msg(sb, KERN_ERR,
1231                                 "error: quota options not supported.");
1232                         break;
1233                 case Opt_usrjquota:
1234                 case Opt_grpjquota:
1235                 case Opt_offusrjquota:
1236                 case Opt_offgrpjquota:
1237                 case Opt_jqfmt_vfsold:
1238                 case Opt_jqfmt_vfsv0:
1239                 case Opt_jqfmt_vfsv1:
1240                         ext3_msg(sb, KERN_ERR,
1241                                 "error: journaled quota options not "
1242                                 "supported.");
1243                         break;
1244                 case Opt_noquota:
1245                         break;
1246 #endif
1247                 case Opt_abort:
1248                         set_opt(sbi->s_mount_opt, ABORT);
1249                         break;
1250                 case Opt_nobarrier:
1251                         clear_opt(sbi->s_mount_opt, BARRIER);
1252                         break;
1253                 case Opt_barrier:
1254                         if (args[0].from) {
1255                                 if (match_int(&args[0], &option))
1256                                         return 0;
1257                         } else
1258                                 option = 1;     /* No argument, default to 1 */
1259                         if (option)
1260                                 set_opt(sbi->s_mount_opt, BARRIER);
1261                         else
1262                                 clear_opt(sbi->s_mount_opt, BARRIER);
1263                         break;
1264                 case Opt_ignore:
1265                         break;
1266                 case Opt_resize:
1267                         if (!is_remount) {
1268                                 ext3_msg(sb, KERN_ERR,
1269                                         "error: resize option only available "
1270                                         "for remount");
1271                                 return 0;
1272                         }
1273                         if (match_int(&args[0], &option) != 0)
1274                                 return 0;
1275                         *n_blocks_count = option;
1276                         break;
1277                 case Opt_nobh:
1278                         ext3_msg(sb, KERN_WARNING,
1279                                 "warning: ignoring deprecated nobh option");
1280                         break;
1281                 case Opt_bh:
1282                         ext3_msg(sb, KERN_WARNING,
1283                                 "warning: ignoring deprecated bh option");
1284                         break;
1285                 default:
1286                         ext3_msg(sb, KERN_ERR,
1287                                 "error: unrecognized mount option \"%s\" "
1288                                 "or missing value", p);
1289                         return 0;
1290                 }
1291         }
1292 #ifdef CONFIG_QUOTA
1293         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1294                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1295                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1296                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1297                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1298
1299                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1300                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1301                                         "format mixing.");
1302                         return 0;
1303                 }
1304
1305                 if (!sbi->s_jquota_fmt) {
1306                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1307                                         "not specified.");
1308                         return 0;
1309                 }
1310         } else {
1311                 if (sbi->s_jquota_fmt) {
1312                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1313                                         "specified with no journaling "
1314                                         "enabled.");
1315                         return 0;
1316                 }
1317         }
1318 #endif
1319         return 1;
1320 }
1321
1322 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1323                             int read_only)
1324 {
1325         struct ext3_sb_info *sbi = EXT3_SB(sb);
1326         int res = 0;
1327
1328         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1329                 ext3_msg(sb, KERN_ERR,
1330                         "error: revision level too high, "
1331                         "forcing read-only mode");
1332                 res = MS_RDONLY;
1333         }
1334         if (read_only)
1335                 return res;
1336         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1337                 ext3_msg(sb, KERN_WARNING,
1338                         "warning: mounting unchecked fs, "
1339                         "running e2fsck is recommended");
1340         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1341                 ext3_msg(sb, KERN_WARNING,
1342                         "warning: mounting fs with errors, "
1343                         "running e2fsck is recommended");
1344         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1345                  le16_to_cpu(es->s_mnt_count) >=
1346                         le16_to_cpu(es->s_max_mnt_count))
1347                 ext3_msg(sb, KERN_WARNING,
1348                         "warning: maximal mount count reached, "
1349                         "running e2fsck is recommended");
1350         else if (le32_to_cpu(es->s_checkinterval) &&
1351                 (le32_to_cpu(es->s_lastcheck) +
1352                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1353                 ext3_msg(sb, KERN_WARNING,
1354                         "warning: checktime reached, "
1355                         "running e2fsck is recommended");
1356 #if 0
1357                 /* @@@ We _will_ want to clear the valid bit if we find
1358                    inconsistencies, to force a fsck at reboot.  But for
1359                    a plain journaled filesystem we can keep it set as
1360                    valid forever! :) */
1361         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1362 #endif
1363         if (!le16_to_cpu(es->s_max_mnt_count))
1364                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1365         le16_add_cpu(&es->s_mnt_count, 1);
1366         es->s_mtime = cpu_to_le32(get_seconds());
1367         ext3_update_dynamic_rev(sb);
1368         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1369
1370         ext3_commit_super(sb, es, 1);
1371         if (test_opt(sb, DEBUG))
1372                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1373                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1374                         sb->s_blocksize,
1375                         sbi->s_groups_count,
1376                         EXT3_BLOCKS_PER_GROUP(sb),
1377                         EXT3_INODES_PER_GROUP(sb),
1378                         sbi->s_mount_opt);
1379
1380         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1381                 char b[BDEVNAME_SIZE];
1382                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1383                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1384         } else {
1385                 ext3_msg(sb, KERN_INFO, "using internal journal");
1386         }
1387         cleancache_init_fs(sb);
1388         return res;
1389 }
1390
1391 /* Called at mount-time, super-block is locked */
1392 static int ext3_check_descriptors(struct super_block *sb)
1393 {
1394         struct ext3_sb_info *sbi = EXT3_SB(sb);
1395         int i;
1396
1397         ext3_debug ("Checking group descriptors");
1398
1399         for (i = 0; i < sbi->s_groups_count; i++) {
1400                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1401                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1402                 ext3_fsblk_t last_block;
1403
1404                 if (i == sbi->s_groups_count - 1)
1405                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1406                 else
1407                         last_block = first_block +
1408                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1409
1410                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1411                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1412                 {
1413                         ext3_error (sb, "ext3_check_descriptors",
1414                                     "Block bitmap for group %d"
1415                                     " not in group (block %lu)!",
1416                                     i, (unsigned long)
1417                                         le32_to_cpu(gdp->bg_block_bitmap));
1418                         return 0;
1419                 }
1420                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1421                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1422                 {
1423                         ext3_error (sb, "ext3_check_descriptors",
1424                                     "Inode bitmap for group %d"
1425                                     " not in group (block %lu)!",
1426                                     i, (unsigned long)
1427                                         le32_to_cpu(gdp->bg_inode_bitmap));
1428                         return 0;
1429                 }
1430                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1431                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1432                     last_block)
1433                 {
1434                         ext3_error (sb, "ext3_check_descriptors",
1435                                     "Inode table for group %d"
1436                                     " not in group (block %lu)!",
1437                                     i, (unsigned long)
1438                                         le32_to_cpu(gdp->bg_inode_table));
1439                         return 0;
1440                 }
1441         }
1442
1443         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1444         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1445         return 1;
1446 }
1447
1448
1449 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1450  * the superblock) which were deleted from all directories, but held open by
1451  * a process at the time of a crash.  We walk the list and try to delete these
1452  * inodes at recovery time (only with a read-write filesystem).
1453  *
1454  * In order to keep the orphan inode chain consistent during traversal (in
1455  * case of crash during recovery), we link each inode into the superblock
1456  * orphan list_head and handle it the same way as an inode deletion during
1457  * normal operation (which journals the operations for us).
1458  *
1459  * We only do an iget() and an iput() on each inode, which is very safe if we
1460  * accidentally point at an in-use or already deleted inode.  The worst that
1461  * can happen in this case is that we get a "bit already cleared" message from
1462  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1463  * e2fsck was run on this filesystem, and it must have already done the orphan
1464  * inode cleanup for us, so we can safely abort without any further action.
1465  */
1466 static void ext3_orphan_cleanup (struct super_block * sb,
1467                                  struct ext3_super_block * es)
1468 {
1469         unsigned int s_flags = sb->s_flags;
1470         int nr_orphans = 0, nr_truncates = 0;
1471 #ifdef CONFIG_QUOTA
1472         int i;
1473 #endif
1474         if (!es->s_last_orphan) {
1475                 jbd_debug(4, "no orphan inodes to clean up\n");
1476                 return;
1477         }
1478
1479         if (bdev_read_only(sb->s_bdev)) {
1480                 ext3_msg(sb, KERN_ERR, "error: write access "
1481                         "unavailable, skipping orphan cleanup.");
1482                 return;
1483         }
1484
1485         /* Check if feature set allows readwrite operations */
1486         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1487                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1488                          "unknown ROCOMPAT features");
1489                 return;
1490         }
1491
1492         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1493                 if (es->s_last_orphan)
1494                         jbd_debug(1, "Errors on filesystem, "
1495                                   "clearing orphan list.\n");
1496                 es->s_last_orphan = 0;
1497                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1498                 return;
1499         }
1500
1501         if (s_flags & MS_RDONLY) {
1502                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1503                 sb->s_flags &= ~MS_RDONLY;
1504         }
1505 #ifdef CONFIG_QUOTA
1506         /* Needed for iput() to work correctly and not trash data */
1507         sb->s_flags |= MS_ACTIVE;
1508         /* Turn on quotas so that they are updated correctly */
1509         for (i = 0; i < MAXQUOTAS; i++) {
1510                 if (EXT3_SB(sb)->s_qf_names[i]) {
1511                         int ret = ext3_quota_on_mount(sb, i);
1512                         if (ret < 0)
1513                                 ext3_msg(sb, KERN_ERR,
1514                                         "error: cannot turn on journaled "
1515                                         "quota: %d", ret);
1516                 }
1517         }
1518 #endif
1519
1520         while (es->s_last_orphan) {
1521                 struct inode *inode;
1522
1523                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1524                 if (IS_ERR(inode)) {
1525                         es->s_last_orphan = 0;
1526                         break;
1527                 }
1528
1529                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1530                 dquot_initialize(inode);
1531                 if (inode->i_nlink) {
1532                         printk(KERN_DEBUG
1533                                 "%s: truncating inode %lu to %Ld bytes\n",
1534                                 __func__, inode->i_ino, inode->i_size);
1535                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1536                                   inode->i_ino, inode->i_size);
1537                         ext3_truncate(inode);
1538                         nr_truncates++;
1539                 } else {
1540                         printk(KERN_DEBUG
1541                                 "%s: deleting unreferenced inode %lu\n",
1542                                 __func__, inode->i_ino);
1543                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1544                                   inode->i_ino);
1545                         nr_orphans++;
1546                 }
1547                 iput(inode);  /* The delete magic happens here! */
1548         }
1549
1550 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1551
1552         if (nr_orphans)
1553                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1554                        PLURAL(nr_orphans));
1555         if (nr_truncates)
1556                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1557                        PLURAL(nr_truncates));
1558 #ifdef CONFIG_QUOTA
1559         /* Turn quotas off */
1560         for (i = 0; i < MAXQUOTAS; i++) {
1561                 if (sb_dqopt(sb)->files[i])
1562                         dquot_quota_off(sb, i);
1563         }
1564 #endif
1565         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1566 }
1567
1568 /*
1569  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1570  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1571  * We need to be 1 filesystem block less than the 2^32 sector limit.
1572  */
1573 static loff_t ext3_max_size(int bits)
1574 {
1575         loff_t res = EXT3_NDIR_BLOCKS;
1576         int meta_blocks;
1577         loff_t upper_limit;
1578
1579         /* This is calculated to be the largest file size for a
1580          * dense, file such that the total number of
1581          * sectors in the file, including data and all indirect blocks,
1582          * does not exceed 2^32 -1
1583          * __u32 i_blocks representing the total number of
1584          * 512 bytes blocks of the file
1585          */
1586         upper_limit = (1LL << 32) - 1;
1587
1588         /* total blocks in file system block size */
1589         upper_limit >>= (bits - 9);
1590
1591
1592         /* indirect blocks */
1593         meta_blocks = 1;
1594         /* double indirect blocks */
1595         meta_blocks += 1 + (1LL << (bits-2));
1596         /* tripple indirect blocks */
1597         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1598
1599         upper_limit -= meta_blocks;
1600         upper_limit <<= bits;
1601
1602         res += 1LL << (bits-2);
1603         res += 1LL << (2*(bits-2));
1604         res += 1LL << (3*(bits-2));
1605         res <<= bits;
1606         if (res > upper_limit)
1607                 res = upper_limit;
1608
1609         if (res > MAX_LFS_FILESIZE)
1610                 res = MAX_LFS_FILESIZE;
1611
1612         return res;
1613 }
1614
1615 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1616                                     ext3_fsblk_t logic_sb_block,
1617                                     int nr)
1618 {
1619         struct ext3_sb_info *sbi = EXT3_SB(sb);
1620         unsigned long bg, first_meta_bg;
1621         int has_super = 0;
1622
1623         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1624
1625         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1626             nr < first_meta_bg)
1627                 return (logic_sb_block + nr + 1);
1628         bg = sbi->s_desc_per_block * nr;
1629         if (ext3_bg_has_super(sb, bg))
1630                 has_super = 1;
1631         return (has_super + ext3_group_first_block_no(sb, bg));
1632 }
1633
1634
1635 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1636 {
1637         struct buffer_head * bh;
1638         struct ext3_super_block *es = NULL;
1639         struct ext3_sb_info *sbi;
1640         ext3_fsblk_t block;
1641         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1642         ext3_fsblk_t logic_sb_block;
1643         unsigned long offset = 0;
1644         unsigned int journal_inum = 0;
1645         unsigned long journal_devnum = 0;
1646         unsigned long def_mount_opts;
1647         struct inode *root;
1648         int blocksize;
1649         int hblock;
1650         int db_count;
1651         int i;
1652         int needs_recovery;
1653         int ret = -EINVAL;
1654         __le32 features;
1655         int err;
1656
1657         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1658         if (!sbi)
1659                 return -ENOMEM;
1660
1661         sbi->s_blockgroup_lock =
1662                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1663         if (!sbi->s_blockgroup_lock) {
1664                 kfree(sbi);
1665                 return -ENOMEM;
1666         }
1667         sb->s_fs_info = sbi;
1668         sbi->s_mount_opt = 0;
1669         sbi->s_resuid = make_kuid(&init_user_ns, EXT3_DEF_RESUID);
1670         sbi->s_resgid = make_kgid(&init_user_ns, EXT3_DEF_RESGID);
1671         sbi->s_sb_block = sb_block;
1672
1673         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1674         if (!blocksize) {
1675                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1676                 goto out_fail;
1677         }
1678
1679         /*
1680          * The ext3 superblock will not be buffer aligned for other than 1kB
1681          * block sizes.  We need to calculate the offset from buffer start.
1682          */
1683         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1684                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1685                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1686         } else {
1687                 logic_sb_block = sb_block;
1688         }
1689
1690         if (!(bh = sb_bread(sb, logic_sb_block))) {
1691                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1692                 goto out_fail;
1693         }
1694         /*
1695          * Note: s_es must be initialized as soon as possible because
1696          *       some ext3 macro-instructions depend on its value
1697          */
1698         es = (struct ext3_super_block *) (bh->b_data + offset);
1699         sbi->s_es = es;
1700         sb->s_magic = le16_to_cpu(es->s_magic);
1701         if (sb->s_magic != EXT3_SUPER_MAGIC)
1702                 goto cantfind_ext3;
1703
1704         /* Set defaults before we parse the mount options */
1705         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1706         if (def_mount_opts & EXT3_DEFM_DEBUG)
1707                 set_opt(sbi->s_mount_opt, DEBUG);
1708         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1709                 set_opt(sbi->s_mount_opt, GRPID);
1710         if (def_mount_opts & EXT3_DEFM_UID16)
1711                 set_opt(sbi->s_mount_opt, NO_UID32);
1712 #ifdef CONFIG_EXT3_FS_XATTR
1713         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1714                 set_opt(sbi->s_mount_opt, XATTR_USER);
1715 #endif
1716 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1717         if (def_mount_opts & EXT3_DEFM_ACL)
1718                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1719 #endif
1720         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1721                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1722         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1723                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1724         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1725                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1726
1727         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1728                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1729         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1730                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1731         else
1732                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1733
1734         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1735         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1736
1737         /* enable barriers by default */
1738         set_opt(sbi->s_mount_opt, BARRIER);
1739         set_opt(sbi->s_mount_opt, RESERVATION);
1740
1741         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1742                             NULL, 0))
1743                 goto failed_mount;
1744
1745         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1746                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1747
1748         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1749             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1750              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1751              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1752                 ext3_msg(sb, KERN_WARNING,
1753                         "warning: feature flags set on rev 0 fs, "
1754                         "running e2fsck is recommended");
1755         /*
1756          * Check feature flags regardless of the revision level, since we
1757          * previously didn't change the revision level when setting the flags,
1758          * so there is a chance incompat flags are set on a rev 0 filesystem.
1759          */
1760         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1761         if (features) {
1762                 ext3_msg(sb, KERN_ERR,
1763                         "error: couldn't mount because of unsupported "
1764                         "optional features (%x)", le32_to_cpu(features));
1765                 goto failed_mount;
1766         }
1767         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1768         if (!(sb->s_flags & MS_RDONLY) && features) {
1769                 ext3_msg(sb, KERN_ERR,
1770                         "error: couldn't mount RDWR because of unsupported "
1771                         "optional features (%x)", le32_to_cpu(features));
1772                 goto failed_mount;
1773         }
1774         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1775
1776         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1777             blocksize > EXT3_MAX_BLOCK_SIZE) {
1778                 ext3_msg(sb, KERN_ERR,
1779                         "error: couldn't mount because of unsupported "
1780                         "filesystem blocksize %d", blocksize);
1781                 goto failed_mount;
1782         }
1783
1784         hblock = bdev_logical_block_size(sb->s_bdev);
1785         if (sb->s_blocksize != blocksize) {
1786                 /*
1787                  * Make sure the blocksize for the filesystem is larger
1788                  * than the hardware sectorsize for the machine.
1789                  */
1790                 if (blocksize < hblock) {
1791                         ext3_msg(sb, KERN_ERR,
1792                                 "error: fsblocksize %d too small for "
1793                                 "hardware sectorsize %d", blocksize, hblock);
1794                         goto failed_mount;
1795                 }
1796
1797                 brelse (bh);
1798                 if (!sb_set_blocksize(sb, blocksize)) {
1799                         ext3_msg(sb, KERN_ERR,
1800                                 "error: bad blocksize %d", blocksize);
1801                         goto out_fail;
1802                 }
1803                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1804                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1805                 bh = sb_bread(sb, logic_sb_block);
1806                 if (!bh) {
1807                         ext3_msg(sb, KERN_ERR,
1808                                "error: can't read superblock on 2nd try");
1809                         goto failed_mount;
1810                 }
1811                 es = (struct ext3_super_block *)(bh->b_data + offset);
1812                 sbi->s_es = es;
1813                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1814                         ext3_msg(sb, KERN_ERR,
1815                                 "error: magic mismatch");
1816                         goto failed_mount;
1817                 }
1818         }
1819
1820         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1821
1822         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1823                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1824                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1825         } else {
1826                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1827                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1828                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1829                     (!is_power_of_2(sbi->s_inode_size)) ||
1830                     (sbi->s_inode_size > blocksize)) {
1831                         ext3_msg(sb, KERN_ERR,
1832                                 "error: unsupported inode size: %d",
1833                                 sbi->s_inode_size);
1834                         goto failed_mount;
1835                 }
1836         }
1837         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1838                                    le32_to_cpu(es->s_log_frag_size);
1839         if (blocksize != sbi->s_frag_size) {
1840                 ext3_msg(sb, KERN_ERR,
1841                        "error: fragsize %lu != blocksize %u (unsupported)",
1842                        sbi->s_frag_size, blocksize);
1843                 goto failed_mount;
1844         }
1845         sbi->s_frags_per_block = 1;
1846         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1847         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1848         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1849         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1850                 goto cantfind_ext3;
1851         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1852         if (sbi->s_inodes_per_block == 0)
1853                 goto cantfind_ext3;
1854         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1855                                         sbi->s_inodes_per_block;
1856         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1857         sbi->s_sbh = bh;
1858         sbi->s_mount_state = le16_to_cpu(es->s_state);
1859         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1860         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1861         for (i=0; i < 4; i++)
1862                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1863         sbi->s_def_hash_version = es->s_def_hash_version;
1864         i = le32_to_cpu(es->s_flags);
1865         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1866                 sbi->s_hash_unsigned = 3;
1867         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1868 #ifdef __CHAR_UNSIGNED__
1869                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1870                 sbi->s_hash_unsigned = 3;
1871 #else
1872                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1873 #endif
1874         }
1875
1876         if (sbi->s_blocks_per_group > blocksize * 8) {
1877                 ext3_msg(sb, KERN_ERR,
1878                         "#blocks per group too big: %lu",
1879                         sbi->s_blocks_per_group);
1880                 goto failed_mount;
1881         }
1882         if (sbi->s_frags_per_group > blocksize * 8) {
1883                 ext3_msg(sb, KERN_ERR,
1884                         "error: #fragments per group too big: %lu",
1885                         sbi->s_frags_per_group);
1886                 goto failed_mount;
1887         }
1888         if (sbi->s_inodes_per_group > blocksize * 8) {
1889                 ext3_msg(sb, KERN_ERR,
1890                         "error: #inodes per group too big: %lu",
1891                         sbi->s_inodes_per_group);
1892                 goto failed_mount;
1893         }
1894
1895         err = generic_check_addressable(sb->s_blocksize_bits,
1896                                         le32_to_cpu(es->s_blocks_count));
1897         if (err) {
1898                 ext3_msg(sb, KERN_ERR,
1899                         "error: filesystem is too large to mount safely");
1900                 if (sizeof(sector_t) < 8)
1901                         ext3_msg(sb, KERN_ERR,
1902                                 "error: CONFIG_LBDAF not enabled");
1903                 ret = err;
1904                 goto failed_mount;
1905         }
1906
1907         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1908                 goto cantfind_ext3;
1909         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1910                                le32_to_cpu(es->s_first_data_block) - 1)
1911                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1912         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1913         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1914                                     GFP_KERNEL);
1915         if (sbi->s_group_desc == NULL) {
1916                 ext3_msg(sb, KERN_ERR,
1917                         "error: not enough memory");
1918                 ret = -ENOMEM;
1919                 goto failed_mount;
1920         }
1921
1922         bgl_lock_init(sbi->s_blockgroup_lock);
1923
1924         for (i = 0; i < db_count; i++) {
1925                 block = descriptor_loc(sb, logic_sb_block, i);
1926                 sbi->s_group_desc[i] = sb_bread(sb, block);
1927                 if (!sbi->s_group_desc[i]) {
1928                         ext3_msg(sb, KERN_ERR,
1929                                 "error: can't read group descriptor %d", i);
1930                         db_count = i;
1931                         goto failed_mount2;
1932                 }
1933         }
1934         if (!ext3_check_descriptors (sb)) {
1935                 ext3_msg(sb, KERN_ERR,
1936                         "error: group descriptors corrupted");
1937                 goto failed_mount2;
1938         }
1939         sbi->s_gdb_count = db_count;
1940         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1941         spin_lock_init(&sbi->s_next_gen_lock);
1942
1943         /* per fileystem reservation list head & lock */
1944         spin_lock_init(&sbi->s_rsv_window_lock);
1945         sbi->s_rsv_window_root = RB_ROOT;
1946         /* Add a single, static dummy reservation to the start of the
1947          * reservation window list --- it gives us a placeholder for
1948          * append-at-start-of-list which makes the allocation logic
1949          * _much_ simpler. */
1950         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1951         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1952         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1953         sbi->s_rsv_window_head.rsv_goal_size = 0;
1954         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1955
1956         /*
1957          * set up enough so that it can read an inode
1958          */
1959         sb->s_op = &ext3_sops;
1960         sb->s_export_op = &ext3_export_ops;
1961         sb->s_xattr = ext3_xattr_handlers;
1962 #ifdef CONFIG_QUOTA
1963         sb->s_qcop = &ext3_qctl_operations;
1964         sb->dq_op = &ext3_quota_operations;
1965 #endif
1966         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1967         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1968         mutex_init(&sbi->s_orphan_lock);
1969         mutex_init(&sbi->s_resize_lock);
1970
1971         sb->s_root = NULL;
1972
1973         needs_recovery = (es->s_last_orphan != 0 ||
1974                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1975                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1976
1977         /*
1978          * The first inode we look at is the journal inode.  Don't try
1979          * root first: it may be modified in the journal!
1980          */
1981         if (!test_opt(sb, NOLOAD) &&
1982             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1983                 if (ext3_load_journal(sb, es, journal_devnum))
1984                         goto failed_mount2;
1985         } else if (journal_inum) {
1986                 if (ext3_create_journal(sb, es, journal_inum))
1987                         goto failed_mount2;
1988         } else {
1989                 if (!silent)
1990                         ext3_msg(sb, KERN_ERR,
1991                                 "error: no journal found. "
1992                                 "mounting ext3 over ext2?");
1993                 goto failed_mount2;
1994         }
1995         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1996                         ext3_count_free_blocks(sb));
1997         if (!err) {
1998                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1999                                 ext3_count_free_inodes(sb));
2000         }
2001         if (!err) {
2002                 err = percpu_counter_init(&sbi->s_dirs_counter,
2003                                 ext3_count_dirs(sb));
2004         }
2005         if (err) {
2006                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2007                 ret = err;
2008                 goto failed_mount3;
2009         }
2010
2011         /* We have now updated the journal if required, so we can
2012          * validate the data journaling mode. */
2013         switch (test_opt(sb, DATA_FLAGS)) {
2014         case 0:
2015                 /* No mode set, assume a default based on the journal
2016                    capabilities: ORDERED_DATA if the journal can
2017                    cope, else JOURNAL_DATA */
2018                 if (journal_check_available_features
2019                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2020                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2021                 else
2022                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2023                 break;
2024
2025         case EXT3_MOUNT_ORDERED_DATA:
2026         case EXT3_MOUNT_WRITEBACK_DATA:
2027                 if (!journal_check_available_features
2028                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2029                         ext3_msg(sb, KERN_ERR,
2030                                 "error: journal does not support "
2031                                 "requested data journaling mode");
2032                         goto failed_mount3;
2033                 }
2034         default:
2035                 break;
2036         }
2037
2038         /*
2039          * The journal_load will have done any necessary log recovery,
2040          * so we can safely mount the rest of the filesystem now.
2041          */
2042
2043         root = ext3_iget(sb, EXT3_ROOT_INO);
2044         if (IS_ERR(root)) {
2045                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2046                 ret = PTR_ERR(root);
2047                 goto failed_mount3;
2048         }
2049         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2050                 iput(root);
2051                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2052                 goto failed_mount3;
2053         }
2054         sb->s_root = d_make_root(root);
2055         if (!sb->s_root) {
2056                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2057                 ret = -ENOMEM;
2058                 goto failed_mount3;
2059         }
2060
2061         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2062
2063         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2064         ext3_orphan_cleanup(sb, es);
2065         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2066         if (needs_recovery) {
2067                 ext3_mark_recovery_complete(sb, es);
2068                 ext3_msg(sb, KERN_INFO, "recovery complete");
2069         }
2070         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2071                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2072                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2073                 "writeback");
2074
2075         return 0;
2076
2077 cantfind_ext3:
2078         if (!silent)
2079                 ext3_msg(sb, KERN_INFO,
2080                         "error: can't find ext3 filesystem on dev %s.",
2081                        sb->s_id);
2082         goto failed_mount;
2083
2084 failed_mount3:
2085         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2086         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2087         percpu_counter_destroy(&sbi->s_dirs_counter);
2088         journal_destroy(sbi->s_journal);
2089 failed_mount2:
2090         for (i = 0; i < db_count; i++)
2091                 brelse(sbi->s_group_desc[i]);
2092         kfree(sbi->s_group_desc);
2093 failed_mount:
2094 #ifdef CONFIG_QUOTA
2095         for (i = 0; i < MAXQUOTAS; i++)
2096                 kfree(sbi->s_qf_names[i]);
2097 #endif
2098         ext3_blkdev_remove(sbi);
2099         brelse(bh);
2100 out_fail:
2101         sb->s_fs_info = NULL;
2102         kfree(sbi->s_blockgroup_lock);
2103         kfree(sbi);
2104         return ret;
2105 }
2106
2107 /*
2108  * Setup any per-fs journal parameters now.  We'll do this both on
2109  * initial mount, once the journal has been initialised but before we've
2110  * done any recovery; and again on any subsequent remount.
2111  */
2112 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2113 {
2114         struct ext3_sb_info *sbi = EXT3_SB(sb);
2115
2116         if (sbi->s_commit_interval)
2117                 journal->j_commit_interval = sbi->s_commit_interval;
2118         /* We could also set up an ext3-specific default for the commit
2119          * interval here, but for now we'll just fall back to the jbd
2120          * default. */
2121
2122         spin_lock(&journal->j_state_lock);
2123         if (test_opt(sb, BARRIER))
2124                 journal->j_flags |= JFS_BARRIER;
2125         else
2126                 journal->j_flags &= ~JFS_BARRIER;
2127         if (test_opt(sb, DATA_ERR_ABORT))
2128                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2129         else
2130                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2131         spin_unlock(&journal->j_state_lock);
2132 }
2133
2134 static journal_t *ext3_get_journal(struct super_block *sb,
2135                                    unsigned int journal_inum)
2136 {
2137         struct inode *journal_inode;
2138         journal_t *journal;
2139
2140         /* First, test for the existence of a valid inode on disk.  Bad
2141          * things happen if we iget() an unused inode, as the subsequent
2142          * iput() will try to delete it. */
2143
2144         journal_inode = ext3_iget(sb, journal_inum);
2145         if (IS_ERR(journal_inode)) {
2146                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2147                 return NULL;
2148         }
2149         if (!journal_inode->i_nlink) {
2150                 make_bad_inode(journal_inode);
2151                 iput(journal_inode);
2152                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2153                 return NULL;
2154         }
2155
2156         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2157                   journal_inode, journal_inode->i_size);
2158         if (!S_ISREG(journal_inode->i_mode)) {
2159                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2160                 iput(journal_inode);
2161                 return NULL;
2162         }
2163
2164         journal = journal_init_inode(journal_inode);
2165         if (!journal) {
2166                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2167                 iput(journal_inode);
2168                 return NULL;
2169         }
2170         journal->j_private = sb;
2171         ext3_init_journal_params(sb, journal);
2172         return journal;
2173 }
2174
2175 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2176                                        dev_t j_dev)
2177 {
2178         struct buffer_head * bh;
2179         journal_t *journal;
2180         ext3_fsblk_t start;
2181         ext3_fsblk_t len;
2182         int hblock, blocksize;
2183         ext3_fsblk_t sb_block;
2184         unsigned long offset;
2185         struct ext3_super_block * es;
2186         struct block_device *bdev;
2187
2188         bdev = ext3_blkdev_get(j_dev, sb);
2189         if (bdev == NULL)
2190                 return NULL;
2191
2192         blocksize = sb->s_blocksize;
2193         hblock = bdev_logical_block_size(bdev);
2194         if (blocksize < hblock) {
2195                 ext3_msg(sb, KERN_ERR,
2196                         "error: blocksize too small for journal device");
2197                 goto out_bdev;
2198         }
2199
2200         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2201         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2202         set_blocksize(bdev, blocksize);
2203         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2204                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2205                         "external journal");
2206                 goto out_bdev;
2207         }
2208
2209         es = (struct ext3_super_block *) (bh->b_data + offset);
2210         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2211             !(le32_to_cpu(es->s_feature_incompat) &
2212               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2213                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2214                         "bad superblock");
2215                 brelse(bh);
2216                 goto out_bdev;
2217         }
2218
2219         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2220                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2221                 brelse(bh);
2222                 goto out_bdev;
2223         }
2224
2225         len = le32_to_cpu(es->s_blocks_count);
2226         start = sb_block + 1;
2227         brelse(bh);     /* we're done with the superblock */
2228
2229         journal = journal_init_dev(bdev, sb->s_bdev,
2230                                         start, len, blocksize);
2231         if (!journal) {
2232                 ext3_msg(sb, KERN_ERR,
2233                         "error: failed to create device journal");
2234                 goto out_bdev;
2235         }
2236         journal->j_private = sb;
2237         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2238                 if (bh_submit_read(journal->j_sb_buffer)) {
2239                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2240                         goto out_journal;
2241                 }
2242         }
2243         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2244                 ext3_msg(sb, KERN_ERR,
2245                         "error: external journal has more than one "
2246                         "user (unsupported) - %d",
2247                         be32_to_cpu(journal->j_superblock->s_nr_users));
2248                 goto out_journal;
2249         }
2250         EXT3_SB(sb)->journal_bdev = bdev;
2251         ext3_init_journal_params(sb, journal);
2252         return journal;
2253 out_journal:
2254         journal_destroy(journal);
2255 out_bdev:
2256         ext3_blkdev_put(bdev);
2257         return NULL;
2258 }
2259
2260 static int ext3_load_journal(struct super_block *sb,
2261                              struct ext3_super_block *es,
2262                              unsigned long journal_devnum)
2263 {
2264         journal_t *journal;
2265         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2266         dev_t journal_dev;
2267         int err = 0;
2268         int really_read_only;
2269
2270         if (journal_devnum &&
2271             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2272                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2273                         "numbers have changed");
2274                 journal_dev = new_decode_dev(journal_devnum);
2275         } else
2276                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2277
2278         really_read_only = bdev_read_only(sb->s_bdev);
2279
2280         /*
2281          * Are we loading a blank journal or performing recovery after a
2282          * crash?  For recovery, we need to check in advance whether we
2283          * can get read-write access to the device.
2284          */
2285
2286         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2287                 if (sb->s_flags & MS_RDONLY) {
2288                         ext3_msg(sb, KERN_INFO,
2289                                 "recovery required on readonly filesystem");
2290                         if (really_read_only) {
2291                                 ext3_msg(sb, KERN_ERR, "error: write access "
2292                                         "unavailable, cannot proceed");
2293                                 return -EROFS;
2294                         }
2295                         ext3_msg(sb, KERN_INFO,
2296                                 "write access will be enabled during recovery");
2297                 }
2298         }
2299
2300         if (journal_inum && journal_dev) {
2301                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2302                        "and inode journals");
2303                 return -EINVAL;
2304         }
2305
2306         if (journal_inum) {
2307                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2308                         return -EINVAL;
2309         } else {
2310                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2311                         return -EINVAL;
2312         }
2313
2314         if (!(journal->j_flags & JFS_BARRIER))
2315                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2316
2317         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2318                 err = journal_update_format(journal);
2319                 if (err)  {
2320                         ext3_msg(sb, KERN_ERR, "error updating journal");
2321                         journal_destroy(journal);
2322                         return err;
2323                 }
2324         }
2325
2326         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2327                 err = journal_wipe(journal, !really_read_only);
2328         if (!err)
2329                 err = journal_load(journal);
2330
2331         if (err) {
2332                 ext3_msg(sb, KERN_ERR, "error loading journal");
2333                 journal_destroy(journal);
2334                 return err;
2335         }
2336
2337         EXT3_SB(sb)->s_journal = journal;
2338         ext3_clear_journal_err(sb, es);
2339
2340         if (!really_read_only && journal_devnum &&
2341             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2342                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2343
2344                 /* Make sure we flush the recovery flag to disk. */
2345                 ext3_commit_super(sb, es, 1);
2346         }
2347
2348         return 0;
2349 }
2350
2351 static int ext3_create_journal(struct super_block *sb,
2352                                struct ext3_super_block *es,
2353                                unsigned int journal_inum)
2354 {
2355         journal_t *journal;
2356         int err;
2357
2358         if (sb->s_flags & MS_RDONLY) {
2359                 ext3_msg(sb, KERN_ERR,
2360                         "error: readonly filesystem when trying to "
2361                         "create journal");
2362                 return -EROFS;
2363         }
2364
2365         journal = ext3_get_journal(sb, journal_inum);
2366         if (!journal)
2367                 return -EINVAL;
2368
2369         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2370                journal_inum);
2371
2372         err = journal_create(journal);
2373         if (err) {
2374                 ext3_msg(sb, KERN_ERR, "error creating journal");
2375                 journal_destroy(journal);
2376                 return -EIO;
2377         }
2378
2379         EXT3_SB(sb)->s_journal = journal;
2380
2381         ext3_update_dynamic_rev(sb);
2382         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2383         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2384
2385         es->s_journal_inum = cpu_to_le32(journal_inum);
2386
2387         /* Make sure we flush the recovery flag to disk. */
2388         ext3_commit_super(sb, es, 1);
2389
2390         return 0;
2391 }
2392
2393 static int ext3_commit_super(struct super_block *sb,
2394                                struct ext3_super_block *es,
2395                                int sync)
2396 {
2397         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2398         int error = 0;
2399
2400         if (!sbh)
2401                 return error;
2402
2403         if (buffer_write_io_error(sbh)) {
2404                 /*
2405                  * Oh, dear.  A previous attempt to write the
2406                  * superblock failed.  This could happen because the
2407                  * USB device was yanked out.  Or it could happen to
2408                  * be a transient write error and maybe the block will
2409                  * be remapped.  Nothing we can do but to retry the
2410                  * write and hope for the best.
2411                  */
2412                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2413                        "superblock detected");
2414                 clear_buffer_write_io_error(sbh);
2415                 set_buffer_uptodate(sbh);
2416         }
2417         /*
2418          * If the file system is mounted read-only, don't update the
2419          * superblock write time.  This avoids updating the superblock
2420          * write time when we are mounting the root file system
2421          * read/only but we need to replay the journal; at that point,
2422          * for people who are east of GMT and who make their clock
2423          * tick in localtime for Windows bug-for-bug compatibility,
2424          * the clock is set in the future, and this will cause e2fsck
2425          * to complain and force a full file system check.
2426          */
2427         if (!(sb->s_flags & MS_RDONLY))
2428                 es->s_wtime = cpu_to_le32(get_seconds());
2429         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2430         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2431         BUFFER_TRACE(sbh, "marking dirty");
2432         mark_buffer_dirty(sbh);
2433         if (sync) {
2434                 error = sync_dirty_buffer(sbh);
2435                 if (buffer_write_io_error(sbh)) {
2436                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2437                                "superblock");
2438                         clear_buffer_write_io_error(sbh);
2439                         set_buffer_uptodate(sbh);
2440                 }
2441         }
2442         return error;
2443 }
2444
2445
2446 /*
2447  * Have we just finished recovery?  If so, and if we are mounting (or
2448  * remounting) the filesystem readonly, then we will end up with a
2449  * consistent fs on disk.  Record that fact.
2450  */
2451 static void ext3_mark_recovery_complete(struct super_block * sb,
2452                                         struct ext3_super_block * es)
2453 {
2454         journal_t *journal = EXT3_SB(sb)->s_journal;
2455
2456         journal_lock_updates(journal);
2457         if (journal_flush(journal) < 0)
2458                 goto out;
2459
2460         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2461             sb->s_flags & MS_RDONLY) {
2462                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2463                 ext3_commit_super(sb, es, 1);
2464         }
2465
2466 out:
2467         journal_unlock_updates(journal);
2468 }
2469
2470 /*
2471  * If we are mounting (or read-write remounting) a filesystem whose journal
2472  * has recorded an error from a previous lifetime, move that error to the
2473  * main filesystem now.
2474  */
2475 static void ext3_clear_journal_err(struct super_block *sb,
2476                                    struct ext3_super_block *es)
2477 {
2478         journal_t *journal;
2479         int j_errno;
2480         const char *errstr;
2481
2482         journal = EXT3_SB(sb)->s_journal;
2483
2484         /*
2485          * Now check for any error status which may have been recorded in the
2486          * journal by a prior ext3_error() or ext3_abort()
2487          */
2488
2489         j_errno = journal_errno(journal);
2490         if (j_errno) {
2491                 char nbuf[16];
2492
2493                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2494                 ext3_warning(sb, __func__, "Filesystem error recorded "
2495                              "from previous mount: %s", errstr);
2496                 ext3_warning(sb, __func__, "Marking fs in need of "
2497                              "filesystem check.");
2498
2499                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2500                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2501                 ext3_commit_super (sb, es, 1);
2502
2503                 journal_clear_err(journal);
2504         }
2505 }
2506
2507 /*
2508  * Force the running and committing transactions to commit,
2509  * and wait on the commit.
2510  */
2511 int ext3_force_commit(struct super_block *sb)
2512 {
2513         journal_t *journal;
2514         int ret;
2515
2516         if (sb->s_flags & MS_RDONLY)
2517                 return 0;
2518
2519         journal = EXT3_SB(sb)->s_journal;
2520         ret = ext3_journal_force_commit(journal);
2521         return ret;
2522 }
2523
2524 static int ext3_sync_fs(struct super_block *sb, int wait)
2525 {
2526         tid_t target;
2527
2528         trace_ext3_sync_fs(sb, wait);
2529         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2530                 if (wait)
2531                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2532         }
2533         return 0;
2534 }
2535
2536 /*
2537  * LVM calls this function before a (read-only) snapshot is created.  This
2538  * gives us a chance to flush the journal completely and mark the fs clean.
2539  */
2540 static int ext3_freeze(struct super_block *sb)
2541 {
2542         int error = 0;
2543         journal_t *journal;
2544
2545         if (!(sb->s_flags & MS_RDONLY)) {
2546                 journal = EXT3_SB(sb)->s_journal;
2547
2548                 /* Now we set up the journal barrier. */
2549                 journal_lock_updates(journal);
2550
2551                 /*
2552                  * We don't want to clear needs_recovery flag when we failed
2553                  * to flush the journal.
2554                  */
2555                 error = journal_flush(journal);
2556                 if (error < 0)
2557                         goto out;
2558
2559                 /* Journal blocked and flushed, clear needs_recovery flag. */
2560                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2561                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2562                 if (error)
2563                         goto out;
2564         }
2565         return 0;
2566
2567 out:
2568         journal_unlock_updates(journal);
2569         return error;
2570 }
2571
2572 /*
2573  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2574  * flag here, even though the filesystem is not technically dirty yet.
2575  */
2576 static int ext3_unfreeze(struct super_block *sb)
2577 {
2578         if (!(sb->s_flags & MS_RDONLY)) {
2579                 lock_super(sb);
2580                 /* Reser the needs_recovery flag before the fs is unlocked. */
2581                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2582                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2583                 unlock_super(sb);
2584                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2585         }
2586         return 0;
2587 }
2588
2589 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2590 {
2591         struct ext3_super_block * es;
2592         struct ext3_sb_info *sbi = EXT3_SB(sb);
2593         ext3_fsblk_t n_blocks_count = 0;
2594         unsigned long old_sb_flags;
2595         struct ext3_mount_options old_opts;
2596         int enable_quota = 0;
2597         int err;
2598 #ifdef CONFIG_QUOTA
2599         int i;
2600 #endif
2601
2602         /* Store the original options */
2603         lock_super(sb);
2604         old_sb_flags = sb->s_flags;
2605         old_opts.s_mount_opt = sbi->s_mount_opt;
2606         old_opts.s_resuid = sbi->s_resuid;
2607         old_opts.s_resgid = sbi->s_resgid;
2608         old_opts.s_commit_interval = sbi->s_commit_interval;
2609 #ifdef CONFIG_QUOTA
2610         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2611         for (i = 0; i < MAXQUOTAS; i++)
2612                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2613 #endif
2614
2615         /*
2616          * Allow the "check" option to be passed as a remount option.
2617          */
2618         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2619                 err = -EINVAL;
2620                 goto restore_opts;
2621         }
2622
2623         if (test_opt(sb, ABORT))
2624                 ext3_abort(sb, __func__, "Abort forced by user");
2625
2626         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2627                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2628
2629         es = sbi->s_es;
2630
2631         ext3_init_journal_params(sb, sbi->s_journal);
2632
2633         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2634                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2635                 if (test_opt(sb, ABORT)) {
2636                         err = -EROFS;
2637                         goto restore_opts;
2638                 }
2639
2640                 if (*flags & MS_RDONLY) {
2641                         err = dquot_suspend(sb, -1);
2642                         if (err < 0)
2643                                 goto restore_opts;
2644
2645                         /*
2646                          * First of all, the unconditional stuff we have to do
2647                          * to disable replay of the journal when we next remount
2648                          */
2649                         sb->s_flags |= MS_RDONLY;
2650
2651                         /*
2652                          * OK, test if we are remounting a valid rw partition
2653                          * readonly, and if so set the rdonly flag and then
2654                          * mark the partition as valid again.
2655                          */
2656                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2657                             (sbi->s_mount_state & EXT3_VALID_FS))
2658                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2659
2660                         ext3_mark_recovery_complete(sb, es);
2661                 } else {
2662                         __le32 ret;
2663                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2664                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2665                                 ext3_msg(sb, KERN_WARNING,
2666                                         "warning: couldn't remount RDWR "
2667                                         "because of unsupported optional "
2668                                         "features (%x)", le32_to_cpu(ret));
2669                                 err = -EROFS;
2670                                 goto restore_opts;
2671                         }
2672
2673                         /*
2674                          * If we have an unprocessed orphan list hanging
2675                          * around from a previously readonly bdev mount,
2676                          * require a full umount & mount for now.
2677                          */
2678                         if (es->s_last_orphan) {
2679                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2680                                        "remount RDWR because of unprocessed "
2681                                        "orphan inode list.  Please "
2682                                        "umount & mount instead.");
2683                                 err = -EINVAL;
2684                                 goto restore_opts;
2685                         }
2686
2687                         /*
2688                          * Mounting a RDONLY partition read-write, so reread
2689                          * and store the current valid flag.  (It may have
2690                          * been changed by e2fsck since we originally mounted
2691                          * the partition.)
2692                          */
2693                         ext3_clear_journal_err(sb, es);
2694                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2695                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2696                                 goto restore_opts;
2697                         if (!ext3_setup_super (sb, es, 0))
2698                                 sb->s_flags &= ~MS_RDONLY;
2699                         enable_quota = 1;
2700                 }
2701         }
2702 #ifdef CONFIG_QUOTA
2703         /* Release old quota file names */
2704         for (i = 0; i < MAXQUOTAS; i++)
2705                 if (old_opts.s_qf_names[i] &&
2706                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2707                         kfree(old_opts.s_qf_names[i]);
2708 #endif
2709         unlock_super(sb);
2710
2711         if (enable_quota)
2712                 dquot_resume(sb, -1);
2713         return 0;
2714 restore_opts:
2715         sb->s_flags = old_sb_flags;
2716         sbi->s_mount_opt = old_opts.s_mount_opt;
2717         sbi->s_resuid = old_opts.s_resuid;
2718         sbi->s_resgid = old_opts.s_resgid;
2719         sbi->s_commit_interval = old_opts.s_commit_interval;
2720 #ifdef CONFIG_QUOTA
2721         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2722         for (i = 0; i < MAXQUOTAS; i++) {
2723                 if (sbi->s_qf_names[i] &&
2724                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2725                         kfree(sbi->s_qf_names[i]);
2726                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2727         }
2728 #endif
2729         unlock_super(sb);
2730         return err;
2731 }
2732
2733 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2734 {
2735         struct super_block *sb = dentry->d_sb;
2736         struct ext3_sb_info *sbi = EXT3_SB(sb);
2737         struct ext3_super_block *es = sbi->s_es;
2738         u64 fsid;
2739
2740         if (test_opt(sb, MINIX_DF)) {
2741                 sbi->s_overhead_last = 0;
2742         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2743                 unsigned long ngroups = sbi->s_groups_count, i;
2744                 ext3_fsblk_t overhead = 0;
2745                 smp_rmb();
2746
2747                 /*
2748                  * Compute the overhead (FS structures).  This is constant
2749                  * for a given filesystem unless the number of block groups
2750                  * changes so we cache the previous value until it does.
2751                  */
2752
2753                 /*
2754                  * All of the blocks before first_data_block are
2755                  * overhead
2756                  */
2757                 overhead = le32_to_cpu(es->s_first_data_block);
2758
2759                 /*
2760                  * Add the overhead attributed to the superblock and
2761                  * block group descriptors.  If the sparse superblocks
2762                  * feature is turned on, then not all groups have this.
2763                  */
2764                 for (i = 0; i < ngroups; i++) {
2765                         overhead += ext3_bg_has_super(sb, i) +
2766                                 ext3_bg_num_gdb(sb, i);
2767                         cond_resched();
2768                 }
2769
2770                 /*
2771                  * Every block group has an inode bitmap, a block
2772                  * bitmap, and an inode table.
2773                  */
2774                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2775                 sbi->s_overhead_last = overhead;
2776                 smp_wmb();
2777                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2778         }
2779
2780         buf->f_type = EXT3_SUPER_MAGIC;
2781         buf->f_bsize = sb->s_blocksize;
2782         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2783         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2784         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2785         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2786                 buf->f_bavail = 0;
2787         buf->f_files = le32_to_cpu(es->s_inodes_count);
2788         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2789         buf->f_namelen = EXT3_NAME_LEN;
2790         fsid = le64_to_cpup((void *)es->s_uuid) ^
2791                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2792         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2793         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2794         return 0;
2795 }
2796
2797 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2798  * is locked for write. Otherwise the are possible deadlocks:
2799  * Process 1                         Process 2
2800  * ext3_create()                     quota_sync()
2801  *   journal_start()                   write_dquot()
2802  *   dquot_initialize()                       down(dqio_mutex)
2803  *     down(dqio_mutex)                    journal_start()
2804  *
2805  */
2806
2807 #ifdef CONFIG_QUOTA
2808
2809 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2810 {
2811         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2812 }
2813
2814 static int ext3_write_dquot(struct dquot *dquot)
2815 {
2816         int ret, err;
2817         handle_t *handle;
2818         struct inode *inode;
2819
2820         inode = dquot_to_inode(dquot);
2821         handle = ext3_journal_start(inode,
2822                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2823         if (IS_ERR(handle))
2824                 return PTR_ERR(handle);
2825         ret = dquot_commit(dquot);
2826         err = ext3_journal_stop(handle);
2827         if (!ret)
2828                 ret = err;
2829         return ret;
2830 }
2831
2832 static int ext3_acquire_dquot(struct dquot *dquot)
2833 {
2834         int ret, err;
2835         handle_t *handle;
2836
2837         handle = ext3_journal_start(dquot_to_inode(dquot),
2838                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2839         if (IS_ERR(handle))
2840                 return PTR_ERR(handle);
2841         ret = dquot_acquire(dquot);
2842         err = ext3_journal_stop(handle);
2843         if (!ret)
2844                 ret = err;
2845         return ret;
2846 }
2847
2848 static int ext3_release_dquot(struct dquot *dquot)
2849 {
2850         int ret, err;
2851         handle_t *handle;
2852
2853         handle = ext3_journal_start(dquot_to_inode(dquot),
2854                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2855         if (IS_ERR(handle)) {
2856                 /* Release dquot anyway to avoid endless cycle in dqput() */
2857                 dquot_release(dquot);
2858                 return PTR_ERR(handle);
2859         }
2860         ret = dquot_release(dquot);
2861         err = ext3_journal_stop(handle);
2862         if (!ret)
2863                 ret = err;
2864         return ret;
2865 }
2866
2867 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2868 {
2869         /* Are we journaling quotas? */
2870         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2871             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2872                 dquot_mark_dquot_dirty(dquot);
2873                 return ext3_write_dquot(dquot);
2874         } else {
2875                 return dquot_mark_dquot_dirty(dquot);
2876         }
2877 }
2878
2879 static int ext3_write_info(struct super_block *sb, int type)
2880 {
2881         int ret, err;
2882         handle_t *handle;
2883
2884         /* Data block + inode block */
2885         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2886         if (IS_ERR(handle))
2887                 return PTR_ERR(handle);
2888         ret = dquot_commit_info(sb, type);
2889         err = ext3_journal_stop(handle);
2890         if (!ret)
2891                 ret = err;
2892         return ret;
2893 }
2894
2895 /*
2896  * Turn on quotas during mount time - we need to find
2897  * the quota file and such...
2898  */
2899 static int ext3_quota_on_mount(struct super_block *sb, int type)
2900 {
2901         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2902                                         EXT3_SB(sb)->s_jquota_fmt, type);
2903 }
2904
2905 /*
2906  * Standard function to be called on quota_on
2907  */
2908 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2909                          struct path *path)
2910 {
2911         int err;
2912
2913         if (!test_opt(sb, QUOTA))
2914                 return -EINVAL;
2915
2916         /* Quotafile not on the same filesystem? */
2917         if (path->dentry->d_sb != sb)
2918                 return -EXDEV;
2919         /* Journaling quota? */
2920         if (EXT3_SB(sb)->s_qf_names[type]) {
2921                 /* Quotafile not of fs root? */
2922                 if (path->dentry->d_parent != sb->s_root)
2923                         ext3_msg(sb, KERN_WARNING,
2924                                 "warning: Quota file not on filesystem root. "
2925                                 "Journaled quota will not work.");
2926         }
2927
2928         /*
2929          * When we journal data on quota file, we have to flush journal to see
2930          * all updates to the file when we bypass pagecache...
2931          */
2932         if (ext3_should_journal_data(path->dentry->d_inode)) {
2933                 /*
2934                  * We don't need to lock updates but journal_flush() could
2935                  * otherwise be livelocked...
2936                  */
2937                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2938                 err = journal_flush(EXT3_SB(sb)->s_journal);
2939                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2940                 if (err)
2941                         return err;
2942         }
2943
2944         return dquot_quota_on(sb, type, format_id, path);
2945 }
2946
2947 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2948  * acquiring the locks... As quota files are never truncated and quota code
2949  * itself serializes the operations (and no one else should touch the files)
2950  * we don't have to be afraid of races */
2951 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2952                                size_t len, loff_t off)
2953 {
2954         struct inode *inode = sb_dqopt(sb)->files[type];
2955         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2956         int err = 0;
2957         int offset = off & (sb->s_blocksize - 1);
2958         int tocopy;
2959         size_t toread;
2960         struct buffer_head *bh;
2961         loff_t i_size = i_size_read(inode);
2962
2963         if (off > i_size)
2964                 return 0;
2965         if (off+len > i_size)
2966                 len = i_size-off;
2967         toread = len;
2968         while (toread > 0) {
2969                 tocopy = sb->s_blocksize - offset < toread ?
2970                                 sb->s_blocksize - offset : toread;
2971                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2972                 if (err)
2973                         return err;
2974                 if (!bh)        /* A hole? */
2975                         memset(data, 0, tocopy);
2976                 else
2977                         memcpy(data, bh->b_data+offset, tocopy);
2978                 brelse(bh);
2979                 offset = 0;
2980                 toread -= tocopy;
2981                 data += tocopy;
2982                 blk++;
2983         }
2984         return len;
2985 }
2986
2987 /* Write to quotafile (we know the transaction is already started and has
2988  * enough credits) */
2989 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2990                                 const char *data, size_t len, loff_t off)
2991 {
2992         struct inode *inode = sb_dqopt(sb)->files[type];
2993         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2994         int err = 0;
2995         int offset = off & (sb->s_blocksize - 1);
2996         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2997         struct buffer_head *bh;
2998         handle_t *handle = journal_current_handle();
2999
3000         if (!handle) {
3001                 ext3_msg(sb, KERN_WARNING,
3002                         "warning: quota write (off=%llu, len=%llu)"
3003                         " cancelled because transaction is not started.",
3004                         (unsigned long long)off, (unsigned long long)len);
3005                 return -EIO;
3006         }
3007
3008         /*
3009          * Since we account only one data block in transaction credits,
3010          * then it is impossible to cross a block boundary.
3011          */
3012         if (sb->s_blocksize - offset < len) {
3013                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3014                         " cancelled because not block aligned",
3015                         (unsigned long long)off, (unsigned long long)len);
3016                 return -EIO;
3017         }
3018         bh = ext3_bread(handle, inode, blk, 1, &err);
3019         if (!bh)
3020                 goto out;
3021         if (journal_quota) {
3022                 err = ext3_journal_get_write_access(handle, bh);
3023                 if (err) {
3024                         brelse(bh);
3025                         goto out;
3026                 }
3027         }
3028         lock_buffer(bh);
3029         memcpy(bh->b_data+offset, data, len);
3030         flush_dcache_page(bh->b_page);
3031         unlock_buffer(bh);
3032         if (journal_quota)
3033                 err = ext3_journal_dirty_metadata(handle, bh);
3034         else {
3035                 /* Always do at least ordered writes for quotas */
3036                 err = ext3_journal_dirty_data(handle, bh);
3037                 mark_buffer_dirty(bh);
3038         }
3039         brelse(bh);
3040 out:
3041         if (err)
3042                 return err;
3043         if (inode->i_size < off + len) {
3044                 i_size_write(inode, off + len);
3045                 EXT3_I(inode)->i_disksize = inode->i_size;
3046         }
3047         inode->i_version++;
3048         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3049         ext3_mark_inode_dirty(handle, inode);
3050         return len;
3051 }
3052
3053 #endif
3054
3055 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3056         int flags, const char *dev_name, void *data)
3057 {
3058         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3059 }
3060
3061 static struct file_system_type ext3_fs_type = {
3062         .owner          = THIS_MODULE,
3063         .name           = "ext3",
3064         .mount          = ext3_mount,
3065         .kill_sb        = kill_block_super,
3066         .fs_flags       = FS_REQUIRES_DEV,
3067 };
3068
3069 static int __init init_ext3_fs(void)
3070 {
3071         int err = init_ext3_xattr();
3072         if (err)
3073                 return err;
3074         err = init_inodecache();
3075         if (err)
3076                 goto out1;
3077         err = register_filesystem(&ext3_fs_type);
3078         if (err)
3079                 goto out;
3080         return 0;
3081 out:
3082         destroy_inodecache();
3083 out1:
3084         exit_ext3_xattr();
3085         return err;
3086 }
3087
3088 static void __exit exit_ext3_fs(void)
3089 {
3090         unregister_filesystem(&ext3_fs_type);
3091         destroy_inodecache();
3092         exit_ext3_xattr();
3093 }
3094
3095 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3096 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3097 MODULE_LICENSE("GPL");
3098 module_init(init_ext3_fs)
3099 module_exit(exit_ext3_fs)