fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static void ext3_commit_super (struct super_block * sb,
  52                                struct ext3_super_block * es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static void ext3_unlockfs(struct super_block *sb);
  64 static void ext3_write_super (struct super_block * sb);
  65 static void ext3_write_super_lockfs(struct super_block *sb);
  66
  67 /*
  68  * Wrappers for journal_start/end.
  69  *
  70  * The only special thing we need to do here is to make sure that all
  71  * journal_end calls result in the superblock being marked dirty, so
  72  * that sync() will call the filesystem's write_super callback if
  73  * appropriate.
  74  */
  75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  76 {
  77         journal_t *journal;
  78
  79         if (sb->s_flags & MS_RDONLY)
  80                 return ERR_PTR(-EROFS);
  81
  82         /* Special case here: if the journal has aborted behind our
  83          * backs (eg. EIO in the commit thread), then we still need to
  84          * take the FS itself readonly cleanly. */
  85         journal = EXT3_SB(sb)->s_journal;
  86         if (is_journal_aborted(journal)) {
  87                 ext3_abort(sb, __func__,
  88                            "Detected aborted journal");
  89                 return ERR_PTR(-EROFS);
  90         }
  91
  92         return journal_start(journal, nblocks);
  93 }
  94
  95 /*
  96  * The only special thing we need to do here is to make sure that all
  97  * journal_stop calls result in the superblock being marked dirty, so
  98  * that sync() will call the filesystem's write_super callback if
  99  * appropriate.
 100  */
 101 int __ext3_journal_stop(const char *where, handle_t *handle)
 102 {
 103         struct super_block *sb;
 104         int err;
 105         int rc;
 106
 107         sb = handle->h_transaction->t_journal->j_private;
 108         err = handle->h_err;
 109         rc = journal_stop(handle);
 110
 111         if (!err)
 112                 err = rc;
 113         if (err)
 114                 __ext3_std_error(sb, where, err);
 115         return err;
 116 }
 117
 118 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 119                 struct buffer_head *bh, handle_t *handle, int err)
 120 {
 121         char nbuf[16];
 122         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 123
 124         if (bh)
 125                 BUFFER_TRACE(bh, "abort");
 126
 127         if (!handle->h_err)
 128                 handle->h_err = err;
 129
 130         if (is_handle_aborted(handle))
 131                 return;
 132
 133         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 134                caller, errstr, err_fn);
 135
 136         journal_abort_handle(handle);
 137 }
 138
 139 /* Deal with the reporting of failure conditions on a filesystem such as
 140  * inconsistencies detected or read IO failures.
 141  *
 142  * On ext2, we can store the error state of the filesystem in the
 143  * superblock.  That is not possible on ext3, because we may have other
 144  * write ordering constraints on the superblock which prevent us from
 145  * writing it out straight away; and given that the journal is about to
 146  * be aborted, we can't rely on the current, or future, transactions to
 147  * write out the superblock safely.
 148  *
 149  * We'll just use the journal_abort() error code to record an error in
 150  * the journal instead.  On recovery, the journal will compain about
 151  * that error until we've noted it down and cleared it.
 152  */
 153
 154 static void ext3_handle_error(struct super_block *sb)
 155 {
 156         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 157
 158         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 159         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 160
 161         if (sb->s_flags & MS_RDONLY)
 162                 return;
 163
 164         if (!test_opt (sb, ERRORS_CONT)) {
 165                 journal_t *journal = EXT3_SB(sb)->s_journal;
 166
 167                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 168                 if (journal)
 169                         journal_abort(journal, -EIO);
 170         }
 171         if (test_opt (sb, ERRORS_RO)) {
 172                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 173                 sb->s_flags |= MS_RDONLY;
 174         }
 175         ext3_commit_super(sb, es, 1);
 176         if (test_opt(sb, ERRORS_PANIC))
 177                 panic("EXT3-fs (device %s): panic forced after error\n",
 178                         sb->s_id);
 179 }
 180
 181 void ext3_error (struct super_block * sb, const char * function,
 182                  const char * fmt, ...)
 183 {
 184         va_list args;
 185
 186         va_start(args, fmt);
 187         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 188         vprintk(fmt, args);
 189         printk("\n");
 190         va_end(args);
 191
 192         ext3_handle_error(sb);
 193 }
 194
 195 static const char *ext3_decode_error(struct super_block * sb, int errno,
 196                                      char nbuf[16])
 197 {
 198         char *errstr = NULL;
 199
 200         switch (errno) {
 201         case -EIO:
 202                 errstr = "IO failure";
 203                 break;
 204         case -ENOMEM:
 205                 errstr = "Out of memory";
 206                 break;
 207         case -EROFS:
 208                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 209                         errstr = "Journal has aborted";
 210                 else
 211                         errstr = "Readonly filesystem";
 212                 break;
 213         default:
 214                 /* If the caller passed in an extra buffer for unknown
 215                  * errors, textualise them now.  Else we just return
 216                  * NULL. */
 217                 if (nbuf) {
 218                         /* Check for truncated error codes... */
 219                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 220                                 errstr = nbuf;
 221                 }
 222                 break;
 223         }
 224
 225         return errstr;
 226 }
 227
 228 /* __ext3_std_error decodes expected errors from journaling functions
 229  * automatically and invokes the appropriate error response.  */
 230
 231 void __ext3_std_error (struct super_block * sb, const char * function,
 232                        int errno)
 233 {
 234         char nbuf[16];
 235         const char *errstr;
 236
 237         /* Special case: if the error is EROFS, and we're not already
 238          * inside a transaction, then there's really no point in logging
 239          * an error. */
 240         if (errno == -EROFS && journal_current_handle() == NULL &&
 241             (sb->s_flags & MS_RDONLY))
 242                 return;
 243
 244         errstr = ext3_decode_error(sb, errno, nbuf);
 245         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 246                 sb->s_id, function, errstr);
 247
 248         ext3_handle_error(sb);
 249 }
 250
 251 /*
 252  * ext3_abort is a much stronger failure handler than ext3_error.  The
 253  * abort function may be used to deal with unrecoverable failures such
 254  * as journal IO errors or ENOMEM at a critical moment in log management.
 255  *
 256  * We unconditionally force the filesystem into an ABORT|READONLY state,
 257  * unless the error response on the fs has been set to panic in which
 258  * case we take the easy way out and panic immediately.
 259  */
 260
 261 void ext3_abort (struct super_block * sb, const char * function,
 262                  const char * fmt, ...)
 263 {
 264         va_list args;
 265
 266         printk (KERN_CRIT "ext3_abort called.\n");
 267
 268         va_start(args, fmt);
 269         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 270         vprintk(fmt, args);
 271         printk("\n");
 272         va_end(args);
 273
 274         if (test_opt(sb, ERRORS_PANIC))
 275                 panic("EXT3-fs panic from previous error\n");
 276
 277         if (sb->s_flags & MS_RDONLY)
 278                 return;
 279
 280         printk(KERN_CRIT "Remounting filesystem read-only\n");
 281         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 282         sb->s_flags |= MS_RDONLY;
 283         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 284         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 285 }
 286
 287 void ext3_warning (struct super_block * sb, const char * function,
 288                    const char * fmt, ...)
 289 {
 290         va_list args;
 291
 292         va_start(args, fmt);
 293         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 294                sb->s_id, function);
 295         vprintk(fmt, args);
 296         printk("\n");
 297         va_end(args);
 298 }
 299
 300 void ext3_update_dynamic_rev(struct super_block *sb)
 301 {
 302         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 303
 304         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 305                 return;
 306
 307         ext3_warning(sb, __func__,
 308                      "updating to rev %d because of new feature flag, "
 309                      "running e2fsck is recommended",
 310                      EXT3_DYNAMIC_REV);
 311
 312         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 313         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 314         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 315         /* leave es->s_feature_*compat flags alone */
 316         /* es->s_uuid will be set by e2fsck if empty */
 317
 318         /*
 319          * The rest of the superblock fields should be zero, and if not it
 320          * means they are likely already in use, so leave them alone.  We
 321          * can leave it up to e2fsck to clean up any inconsistencies there.
 322          */
 323 }
 324
 325 /*
 326  * Open the external journal device
 327  */
 328 static struct block_device *ext3_blkdev_get(dev_t dev)
 329 {
 330         struct block_device *bdev;
 331         char b[BDEVNAME_SIZE];
 332
 333         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 334         if (IS_ERR(bdev))
 335                 goto fail;
 336         return bdev;
 337
 338 fail:
 339         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 340                         __bdevname(dev, b), PTR_ERR(bdev));
 341         return NULL;
 342 }
 343
 344 /*
 345  * Release the journal device
 346  */
 347 static int ext3_blkdev_put(struct block_device *bdev)
 348 {
 349         bd_release(bdev);
 350         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 351 }
 352
 353 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 354 {
 355         struct block_device *bdev;
 356         int ret = -ENODEV;
 357
 358         bdev = sbi->journal_bdev;
 359         if (bdev) {
 360                 ret = ext3_blkdev_put(bdev);
 361                 sbi->journal_bdev = NULL;
 362         }
 363         return ret;
 364 }
 365
 366 static inline struct inode *orphan_list_entry(struct list_head *l)
 367 {
 368         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 369 }
 370
 371 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 372 {
 373         struct list_head *l;
 374
 375         printk(KERN_ERR "sb orphan head is %d\n",
 376                le32_to_cpu(sbi->s_es->s_last_orphan));
 377
 378         printk(KERN_ERR "sb_info orphan list:\n");
 379         list_for_each(l, &sbi->s_orphan) {
 380                 struct inode *inode = orphan_list_entry(l);
 381                 printk(KERN_ERR "  "
 382                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 383                        inode->i_sb->s_id, inode->i_ino, inode,
 384                        inode->i_mode, inode->i_nlink,
 385                        NEXT_ORPHAN(inode));
 386         }
 387 }
 388
 389 static void ext3_put_super (struct super_block * sb)
 390 {
 391         struct ext3_sb_info *sbi = EXT3_SB(sb);
 392         struct ext3_super_block *es = sbi->s_es;
 393         int i;
 394
 395         ext3_xattr_put_super(sb);
 396         if (journal_destroy(sbi->s_journal) < 0)
 397                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 398         if (!(sb->s_flags & MS_RDONLY)) {
 399                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 400                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 401                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 402                 mark_buffer_dirty(sbi->s_sbh);
 403                 ext3_commit_super(sb, es, 1);
 404         }
 405
 406         for (i = 0; i < sbi->s_gdb_count; i++)
 407                 brelse(sbi->s_group_desc[i]);
 408         kfree(sbi->s_group_desc);
 409         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 410         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 411         percpu_counter_destroy(&sbi->s_dirs_counter);
 412         brelse(sbi->s_sbh);
 413 #ifdef CONFIG_QUOTA
 414         for (i = 0; i < MAXQUOTAS; i++)
 415                 kfree(sbi->s_qf_names[i]);
 416 #endif
 417
 418         /* Debugging code just in case the in-memory inode orphan list
 419          * isn't empty.  The on-disk one can be non-empty if we've
 420          * detected an error and taken the fs readonly, but the
 421          * in-memory list had better be clean by this point. */
 422         if (!list_empty(&sbi->s_orphan))
 423                 dump_orphan_list(sb, sbi);
 424         J_ASSERT(list_empty(&sbi->s_orphan));
 425
 426         invalidate_bdev(sb->s_bdev);
 427         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 428                 /*
 429                  * Invalidate the journal device's buffers.  We don't want them
 430                  * floating about in memory - the physical journal device may
 431                  * hotswapped, and it breaks the `ro-after' testing code.
 432                  */
 433                 sync_blockdev(sbi->journal_bdev);
 434                 invalidate_bdev(sbi->journal_bdev);
 435                 ext3_blkdev_remove(sbi);
 436         }
 437         sb->s_fs_info = NULL;
 438         kfree(sbi);
 439         return;
 440 }
 441
 442 static struct kmem_cache *ext3_inode_cachep;
 443
 444 /*
 445  * Called inside transaction, so use GFP_NOFS
 446  */
 447 static struct inode *ext3_alloc_inode(struct super_block *sb)
 448 {
 449         struct ext3_inode_info *ei;
 450
 451         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 452         if (!ei)
 453                 return NULL;
 454 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 455         ei->i_acl = EXT3_ACL_NOT_CACHED;
 456         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 457 #endif
 458         ei->i_block_alloc_info = NULL;
 459         ei->vfs_inode.i_version = 1;
 460         return &ei->vfs_inode;
 461 }
 462
 463 static void ext3_destroy_inode(struct inode *inode)
 464 {
 465         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 466                 printk("EXT3 Inode %p: orphan list check failed!\n",
 467                         EXT3_I(inode));
 468                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 469                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 470                                 false);
 471                 dump_stack();
 472         }
 473         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 474 }
 475
 476 static void init_once(void *foo)
 477 {
 478         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 479
 480         INIT_LIST_HEAD(&ei->i_orphan);
 481 #ifdef CONFIG_EXT3_FS_XATTR
 482         init_rwsem(&ei->xattr_sem);
 483 #endif
 484         mutex_init(&ei->truncate_mutex);
 485         inode_init_once(&ei->vfs_inode);
 486 }
 487
 488 static int init_inodecache(void)
 489 {
 490         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 491                                              sizeof(struct ext3_inode_info),
 492                                              0, (SLAB_RECLAIM_ACCOUNT|
 493                                                 SLAB_MEM_SPREAD),
 494                                              init_once);
 495         if (ext3_inode_cachep == NULL)
 496                 return -ENOMEM;
 497         return 0;
 498 }
 499
 500 static void destroy_inodecache(void)
 501 {
 502         kmem_cache_destroy(ext3_inode_cachep);
 503 }
 504
 505 static void ext3_clear_inode(struct inode *inode)
 506 {
 507         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 508 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 509         if (EXT3_I(inode)->i_acl &&
 510                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 511                 posix_acl_release(EXT3_I(inode)->i_acl);
 512                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 513         }
 514         if (EXT3_I(inode)->i_default_acl &&
 515                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 516                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 517                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 518         }
 519 #endif
 520         ext3_discard_reservation(inode);
 521         EXT3_I(inode)->i_block_alloc_info = NULL;
 522         if (unlikely(rsv))
 523                 kfree(rsv);
 524 }
 525
 526 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 527 {
 528 #if defined(CONFIG_QUOTA)
 529         struct ext3_sb_info *sbi = EXT3_SB(sb);
 530
 531         if (sbi->s_jquota_fmt)
 532                 seq_printf(seq, ",jqfmt=%s",
 533                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 534
 535         if (sbi->s_qf_names[USRQUOTA])
 536                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 537
 538         if (sbi->s_qf_names[GRPQUOTA])
 539                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 540
 541         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 542                 seq_puts(seq, ",usrquota");
 543
 544         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 545                 seq_puts(seq, ",grpquota");
 546 #endif
 547 }
 548
 549 /*
 550  * Show an option if
 551  *  - it's set to a non-default value OR
 552  *  - if the per-sb default is different from the global default
 553  */
 554 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 555 {
 556         struct super_block *sb = vfs->mnt_sb;
 557         struct ext3_sb_info *sbi = EXT3_SB(sb);
 558         struct ext3_super_block *es = sbi->s_es;
 559         unsigned long def_mount_opts;
 560
 561         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 562
 563         if (sbi->s_sb_block != 1)
 564                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 565         if (test_opt(sb, MINIX_DF))
 566                 seq_puts(seq, ",minixdf");
 567         if (test_opt(sb, GRPID))
 568                 seq_puts(seq, ",grpid");
 569         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 570                 seq_puts(seq, ",nogrpid");
 571         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 572             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 573                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 574         }
 575         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 576             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 577                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 578         }
 579         if (test_opt(sb, ERRORS_RO)) {
 580                 int def_errors = le16_to_cpu(es->s_errors);
 581
 582                 if (def_errors == EXT3_ERRORS_PANIC ||
 583                     def_errors == EXT3_ERRORS_CONTINUE) {
 584                         seq_puts(seq, ",errors=remount-ro");
 585                 }
 586         }
 587         if (test_opt(sb, ERRORS_CONT))
 588                 seq_puts(seq, ",errors=continue");
 589         if (test_opt(sb, ERRORS_PANIC))
 590                 seq_puts(seq, ",errors=panic");
 591         if (test_opt(sb, NO_UID32))
 592                 seq_puts(seq, ",nouid32");
 593         if (test_opt(sb, DEBUG))
 594                 seq_puts(seq, ",debug");
 595         if (test_opt(sb, OLDALLOC))
 596                 seq_puts(seq, ",oldalloc");
 597 #ifdef CONFIG_EXT3_FS_XATTR
 598         if (test_opt(sb, XATTR_USER))
 599                 seq_puts(seq, ",user_xattr");
 600         if (!test_opt(sb, XATTR_USER) &&
 601             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 602                 seq_puts(seq, ",nouser_xattr");
 603         }
 604 #endif
 605 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 606         if (test_opt(sb, POSIX_ACL))
 607                 seq_puts(seq, ",acl");
 608         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 609                 seq_puts(seq, ",noacl");
 610 #endif
 611         if (!test_opt(sb, RESERVATION))
 612                 seq_puts(seq, ",noreservation");
 613         if (sbi->s_commit_interval) {
 614                 seq_printf(seq, ",commit=%u",
 615                            (unsigned) (sbi->s_commit_interval / HZ));
 616         }
 617         if (test_opt(sb, BARRIER))
 618                 seq_puts(seq, ",barrier=1");
 619         if (test_opt(sb, NOBH))
 620                 seq_puts(seq, ",nobh");
 621
 622         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 623                 seq_puts(seq, ",data=journal");
 624         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 625                 seq_puts(seq, ",data=ordered");
 626         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 627                 seq_puts(seq, ",data=writeback");
 628
 629         if (test_opt(sb, DATA_ERR_ABORT))
 630                 seq_puts(seq, ",data_err=abort");
 631
 632         ext3_show_quota_options(seq, sb);
 633
 634         return 0;
 635 }
 636
 637
 638 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 639                 u64 ino, u32 generation)
 640 {
 641         struct inode *inode;
 642
 643         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 644                 return ERR_PTR(-ESTALE);
 645         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 646                 return ERR_PTR(-ESTALE);
 647
 648         /* iget isn't really right if the inode is currently unallocated!!
 649          *
 650          * ext3_read_inode will return a bad_inode if the inode had been
 651          * deleted, so we should be safe.
 652          *
 653          * Currently we don't know the generation for parent directory, so
 654          * a generation of 0 means "accept any"
 655          */
 656         inode = ext3_iget(sb, ino);
 657         if (IS_ERR(inode))
 658                 return ERR_CAST(inode);
 659         if (generation && inode->i_generation != generation) {
 660                 iput(inode);
 661                 return ERR_PTR(-ESTALE);
 662         }
 663
 664         return inode;
 665 }
 666
 667 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 668                 int fh_len, int fh_type)
 669 {
 670         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 671                                     ext3_nfs_get_inode);
 672 }
 673
 674 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 675                 int fh_len, int fh_type)
 676 {
 677         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 678                                     ext3_nfs_get_inode);
 679 }
 680
 681 #ifdef CONFIG_QUOTA
 682 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 683 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 684
 685 static int ext3_dquot_initialize(struct inode *inode, int type);
 686 static int ext3_dquot_drop(struct inode *inode);
 687 static int ext3_write_dquot(struct dquot *dquot);
 688 static int ext3_acquire_dquot(struct dquot *dquot);
 689 static int ext3_release_dquot(struct dquot *dquot);
 690 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 691 static int ext3_write_info(struct super_block *sb, int type);
 692 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 693                                 char *path, int remount);
 694 static int ext3_quota_on_mount(struct super_block *sb, int type);
 695 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 696                                size_t len, loff_t off);
 697 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 698                                 const char *data, size_t len, loff_t off);
 699
 700 static struct dquot_operations ext3_quota_operations = {
 701         .initialize     = ext3_dquot_initialize,
 702         .drop           = ext3_dquot_drop,
 703         .alloc_space    = dquot_alloc_space,
 704         .alloc_inode    = dquot_alloc_inode,
 705         .free_space     = dquot_free_space,
 706         .free_inode     = dquot_free_inode,
 707         .transfer       = dquot_transfer,
 708         .write_dquot    = ext3_write_dquot,
 709         .acquire_dquot  = ext3_acquire_dquot,
 710         .release_dquot  = ext3_release_dquot,
 711         .mark_dirty     = ext3_mark_dquot_dirty,
 712         .write_info     = ext3_write_info
 713 };
 714
 715 static struct quotactl_ops ext3_qctl_operations = {
 716         .quota_on       = ext3_quota_on,
 717         .quota_off      = vfs_quota_off,
 718         .quota_sync     = vfs_quota_sync,
 719         .get_info       = vfs_get_dqinfo,
 720         .set_info       = vfs_set_dqinfo,
 721         .get_dqblk      = vfs_get_dqblk,
 722         .set_dqblk      = vfs_set_dqblk
 723 };
 724 #endif
 725
 726 static const struct super_operations ext3_sops = {
 727         .alloc_inode    = ext3_alloc_inode,
 728         .destroy_inode  = ext3_destroy_inode,
 729         .write_inode    = ext3_write_inode,
 730         .dirty_inode    = ext3_dirty_inode,
 731         .delete_inode   = ext3_delete_inode,
 732         .put_super      = ext3_put_super,
 733         .write_super    = ext3_write_super,
 734         .sync_fs        = ext3_sync_fs,
 735         .write_super_lockfs = ext3_write_super_lockfs,
 736         .unlockfs       = ext3_unlockfs,
 737         .statfs         = ext3_statfs,
 738         .remount_fs     = ext3_remount,
 739         .clear_inode    = ext3_clear_inode,
 740         .show_options   = ext3_show_options,
 741 #ifdef CONFIG_QUOTA
 742         .quota_read     = ext3_quota_read,
 743         .quota_write    = ext3_quota_write,
 744 #endif
 745 };
 746
 747 static const struct export_operations ext3_export_ops = {
 748         .fh_to_dentry = ext3_fh_to_dentry,
 749         .fh_to_parent = ext3_fh_to_parent,
 750         .get_parent = ext3_get_parent,
 751 };
 752
 753 enum {
 754         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 755         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 756         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 757         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 758         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 759         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 760         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 761         Opt_data_err_abort, Opt_data_err_ignore,
 762         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 763         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 764         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 765         Opt_grpquota
 766 };
 767
 768 static const match_table_t tokens = {
 769         {Opt_bsd_df, "bsddf"},
 770         {Opt_minix_df, "minixdf"},
 771         {Opt_grpid, "grpid"},
 772         {Opt_grpid, "bsdgroups"},
 773         {Opt_nogrpid, "nogrpid"},
 774         {Opt_nogrpid, "sysvgroups"},
 775         {Opt_resgid, "resgid=%u"},
 776         {Opt_resuid, "resuid=%u"},
 777         {Opt_sb, "sb=%u"},
 778         {Opt_err_cont, "errors=continue"},
 779         {Opt_err_panic, "errors=panic"},
 780         {Opt_err_ro, "errors=remount-ro"},
 781         {Opt_nouid32, "nouid32"},
 782         {Opt_nocheck, "nocheck"},
 783         {Opt_nocheck, "check=none"},
 784         {Opt_debug, "debug"},
 785         {Opt_oldalloc, "oldalloc"},
 786         {Opt_orlov, "orlov"},
 787         {Opt_user_xattr, "user_xattr"},
 788         {Opt_nouser_xattr, "nouser_xattr"},
 789         {Opt_acl, "acl"},
 790         {Opt_noacl, "noacl"},
 791         {Opt_reservation, "reservation"},
 792         {Opt_noreservation, "noreservation"},
 793         {Opt_noload, "noload"},
 794         {Opt_nobh, "nobh"},
 795         {Opt_bh, "bh"},
 796         {Opt_commit, "commit=%u"},
 797         {Opt_journal_update, "journal=update"},
 798         {Opt_journal_inum, "journal=%u"},
 799         {Opt_journal_dev, "journal_dev=%u"},
 800         {Opt_abort, "abort"},
 801         {Opt_data_journal, "data=journal"},
 802         {Opt_data_ordered, "data=ordered"},
 803         {Opt_data_writeback, "data=writeback"},
 804         {Opt_data_err_abort, "data_err=abort"},
 805         {Opt_data_err_ignore, "data_err=ignore"},
 806         {Opt_offusrjquota, "usrjquota="},
 807         {Opt_usrjquota, "usrjquota=%s"},
 808         {Opt_offgrpjquota, "grpjquota="},
 809         {Opt_grpjquota, "grpjquota=%s"},
 810         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 811         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 812         {Opt_grpquota, "grpquota"},
 813         {Opt_noquota, "noquota"},
 814         {Opt_quota, "quota"},
 815         {Opt_usrquota, "usrquota"},
 816         {Opt_barrier, "barrier=%u"},
 817         {Opt_resize, "resize"},
 818         {Opt_err, NULL},
 819 };
 820
 821 static ext3_fsblk_t get_sb_block(void **data)
 822 {
 823         ext3_fsblk_t    sb_block;
 824         char            *options = (char *) *data;
 825
 826         if (!options || strncmp(options, "sb=", 3) != 0)
 827                 return 1;       /* Default location */
 828         options += 3;
 829         /*todo: use simple_strtoll with >32bit ext3 */
 830         sb_block = simple_strtoul(options, &options, 0);
 831         if (*options && *options != ',') {
 832                 printk("EXT3-fs: Invalid sb specification: %s\n",
 833                        (char *) *data);
 834                 return 1;
 835         }
 836         if (*options == ',')
 837                 options++;
 838         *data = (void *) options;
 839         return sb_block;
 840 }
 841
 842 static int parse_options (char *options, struct super_block *sb,
 843                           unsigned int *inum, unsigned long *journal_devnum,
 844                           ext3_fsblk_t *n_blocks_count, int is_remount)
 845 {
 846         struct ext3_sb_info *sbi = EXT3_SB(sb);
 847         char * p;
 848         substring_t args[MAX_OPT_ARGS];
 849         int data_opt = 0;
 850         int option;
 851 #ifdef CONFIG_QUOTA
 852         int qtype, qfmt;
 853         char *qname;
 854 #endif
 855
 856         if (!options)
 857                 return 1;
 858
 859         while ((p = strsep (&options, ",")) != NULL) {
 860                 int token;
 861                 if (!*p)
 862                         continue;
 863
 864                 token = match_token(p, tokens, args);
 865                 switch (token) {
 866                 case Opt_bsd_df:
 867                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 868                         break;
 869                 case Opt_minix_df:
 870                         set_opt (sbi->s_mount_opt, MINIX_DF);
 871                         break;
 872                 case Opt_grpid:
 873                         set_opt (sbi->s_mount_opt, GRPID);
 874                         break;
 875                 case Opt_nogrpid:
 876                         clear_opt (sbi->s_mount_opt, GRPID);
 877                         break;
 878                 case Opt_resuid:
 879                         if (match_int(&args[0], &option))
 880                                 return 0;
 881                         sbi->s_resuid = option;
 882                         break;
 883                 case Opt_resgid:
 884                         if (match_int(&args[0], &option))
 885                                 return 0;
 886                         sbi->s_resgid = option;
 887                         break;
 888                 case Opt_sb:
 889                         /* handled by get_sb_block() instead of here */
 890                         /* *sb_block = match_int(&args[0]); */
 891                         break;
 892                 case Opt_err_panic:
 893                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 894                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 895                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 896                         break;
 897                 case Opt_err_ro:
 898                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 899                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 900                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 901                         break;
 902                 case Opt_err_cont:
 903                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 904                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 905                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 906                         break;
 907                 case Opt_nouid32:
 908                         set_opt (sbi->s_mount_opt, NO_UID32);
 909                         break;
 910                 case Opt_nocheck:
 911                         clear_opt (sbi->s_mount_opt, CHECK);
 912                         break;
 913                 case Opt_debug:
 914                         set_opt (sbi->s_mount_opt, DEBUG);
 915                         break;
 916                 case Opt_oldalloc:
 917                         set_opt (sbi->s_mount_opt, OLDALLOC);
 918                         break;
 919                 case Opt_orlov:
 920                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 921                         break;
 922 #ifdef CONFIG_EXT3_FS_XATTR
 923                 case Opt_user_xattr:
 924                         set_opt (sbi->s_mount_opt, XATTR_USER);
 925                         break;
 926                 case Opt_nouser_xattr:
 927                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 928                         break;
 929 #else
 930                 case Opt_user_xattr:
 931                 case Opt_nouser_xattr:
 932                         printk("EXT3 (no)user_xattr options not supported\n");
 933                         break;
 934 #endif
 935 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 936                 case Opt_acl:
 937                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 938                         break;
 939                 case Opt_noacl:
 940                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 941                         break;
 942 #else
 943                 case Opt_acl:
 944                 case Opt_noacl:
 945                         printk("EXT3 (no)acl options not supported\n");
 946                         break;
 947 #endif
 948                 case Opt_reservation:
 949                         set_opt(sbi->s_mount_opt, RESERVATION);
 950                         break;
 951                 case Opt_noreservation:
 952                         clear_opt(sbi->s_mount_opt, RESERVATION);
 953                         break;
 954                 case Opt_journal_update:
 955                         /* @@@ FIXME */
 956                         /* Eventually we will want to be able to create
 957                            a journal file here.  For now, only allow the
 958                            user to specify an existing inode to be the
 959                            journal file. */
 960                         if (is_remount) {
 961                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 962                                        "journal on remount\n");
 963                                 return 0;
 964                         }
 965                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 966                         break;
 967                 case Opt_journal_inum:
 968                         if (is_remount) {
 969                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 970                                        "journal on remount\n");
 971                                 return 0;
 972                         }
 973                         if (match_int(&args[0], &option))
 974                                 return 0;
 975                         *inum = option;
 976                         break;
 977                 case Opt_journal_dev:
 978                         if (is_remount) {
 979                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 980                                        "journal on remount\n");
 981                                 return 0;
 982                         }
 983                         if (match_int(&args[0], &option))
 984                                 return 0;
 985                         *journal_devnum = option;
 986                         break;
 987                 case Opt_noload:
 988                         set_opt (sbi->s_mount_opt, NOLOAD);
 989                         break;
 990                 case Opt_commit:
 991                         if (match_int(&args[0], &option))
 992                                 return 0;
 993                         if (option < 0)
 994                                 return 0;
 995                         if (option == 0)
 996                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 997                         sbi->s_commit_interval = HZ * option;
 998                         break;
 999                 case Opt_data_journal:
1000                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1001                         goto datacheck;
1002                 case Opt_data_ordered:
1003                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1004                         goto datacheck;
1005                 case Opt_data_writeback:
1006                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1007                 datacheck:
1008                         if (is_remount) {
1009                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1010                                                 != data_opt) {
1011                                         printk(KERN_ERR
1012                                                 "EXT3-fs: cannot change data "
1013                                                 "mode on remount\n");
1014                                         return 0;
1015                                 }
1016                         } else {
1017                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1018                                 sbi->s_mount_opt |= data_opt;
1019                         }
1020                         break;
1021                 case Opt_data_err_abort:
1022                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1023                         break;
1024                 case Opt_data_err_ignore:
1025                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1026                         break;
1027 #ifdef CONFIG_QUOTA
1028                 case Opt_usrjquota:
1029                         qtype = USRQUOTA;
1030                         goto set_qf_name;
1031                 case Opt_grpjquota:
1032                         qtype = GRPQUOTA;
1033 set_qf_name:
1034                         if ((sb_any_quota_enabled(sb) ||
1035                              sb_any_quota_suspended(sb)) &&
1036                             !sbi->s_qf_names[qtype]) {
1037                                 printk(KERN_ERR
1038                                         "EXT3-fs: Cannot change journaled "
1039                                         "quota options when quota turned on.\n");
1040                                 return 0;
1041                         }
1042                         qname = match_strdup(&args[0]);
1043                         if (!qname) {
1044                                 printk(KERN_ERR
1045                                         "EXT3-fs: not enough memory for "
1046                                         "storing quotafile name.\n");
1047                                 return 0;
1048                         }
1049                         if (sbi->s_qf_names[qtype] &&
1050                             strcmp(sbi->s_qf_names[qtype], qname)) {
1051                                 printk(KERN_ERR
1052                                         "EXT3-fs: %s quota file already "
1053                                         "specified.\n", QTYPE2NAME(qtype));
1054                                 kfree(qname);
1055                                 return 0;
1056                         }
1057                         sbi->s_qf_names[qtype] = qname;
1058                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1059                                 printk(KERN_ERR
1060                                         "EXT3-fs: quotafile must be on "
1061                                         "filesystem root.\n");
1062                                 kfree(sbi->s_qf_names[qtype]);
1063                                 sbi->s_qf_names[qtype] = NULL;
1064                                 return 0;
1065                         }
1066                         set_opt(sbi->s_mount_opt, QUOTA);
1067                         break;
1068                 case Opt_offusrjquota:
1069                         qtype = USRQUOTA;
1070                         goto clear_qf_name;
1071                 case Opt_offgrpjquota:
1072                         qtype = GRPQUOTA;
1073 clear_qf_name:
1074                         if ((sb_any_quota_enabled(sb) ||
1075                              sb_any_quota_suspended(sb)) &&
1076                             sbi->s_qf_names[qtype]) {
1077                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1078                                         "journaled quota options when "
1079                                         "quota turned on.\n");
1080                                 return 0;
1081                         }
1082                         /*
1083                          * The space will be released later when all options
1084                          * are confirmed to be correct
1085                          */
1086                         sbi->s_qf_names[qtype] = NULL;
1087                         break;
1088                 case Opt_jqfmt_vfsold:
1089                         qfmt = QFMT_VFS_OLD;
1090                         goto set_qf_format;
1091                 case Opt_jqfmt_vfsv0:
1092                         qfmt = QFMT_VFS_V0;
1093 set_qf_format:
1094                         if ((sb_any_quota_enabled(sb) ||
1095                              sb_any_quota_suspended(sb)) &&
1096                             sbi->s_jquota_fmt != qfmt) {
1097                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1098                                         "journaled quota options when "
1099                                         "quota turned on.\n");
1100                                 return 0;
1101                         }
1102                         sbi->s_jquota_fmt = qfmt;
1103                         break;
1104                 case Opt_quota:
1105                 case Opt_usrquota:
1106                         set_opt(sbi->s_mount_opt, QUOTA);
1107                         set_opt(sbi->s_mount_opt, USRQUOTA);
1108                         break;
1109                 case Opt_grpquota:
1110                         set_opt(sbi->s_mount_opt, QUOTA);
1111                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1112                         break;
1113                 case Opt_noquota:
1114                         if (sb_any_quota_enabled(sb) ||
1115                             sb_any_quota_suspended(sb)) {
1116                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1117                                         "options when quota turned on.\n");
1118                                 return 0;
1119                         }
1120                         clear_opt(sbi->s_mount_opt, QUOTA);
1121                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1122                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1123                         break;
1124 #else
1125                 case Opt_quota:
1126                 case Opt_usrquota:
1127                 case Opt_grpquota:
1128                         printk(KERN_ERR
1129                                 "EXT3-fs: quota options not supported.\n");
1130                         break;
1131                 case Opt_usrjquota:
1132                 case Opt_grpjquota:
1133                 case Opt_offusrjquota:
1134                 case Opt_offgrpjquota:
1135                 case Opt_jqfmt_vfsold:
1136                 case Opt_jqfmt_vfsv0:
1137                         printk(KERN_ERR
1138                                 "EXT3-fs: journaled quota options not "
1139                                 "supported.\n");
1140                         break;
1141                 case Opt_noquota:
1142                         break;
1143 #endif
1144                 case Opt_abort:
1145                         set_opt(sbi->s_mount_opt, ABORT);
1146                         break;
1147                 case Opt_barrier:
1148                         if (match_int(&args[0], &option))
1149                                 return 0;
1150                         if (option)
1151                                 set_opt(sbi->s_mount_opt, BARRIER);
1152                         else
1153                                 clear_opt(sbi->s_mount_opt, BARRIER);
1154                         break;
1155                 case Opt_ignore:
1156                         break;
1157                 case Opt_resize:
1158                         if (!is_remount) {
1159                                 printk("EXT3-fs: resize option only available "
1160                                         "for remount\n");
1161                                 return 0;
1162                         }
1163                         if (match_int(&args[0], &option) != 0)
1164                                 return 0;
1165                         *n_blocks_count = option;
1166                         break;
1167                 case Opt_nobh:
1168                         set_opt(sbi->s_mount_opt, NOBH);
1169                         break;
1170                 case Opt_bh:
1171                         clear_opt(sbi->s_mount_opt, NOBH);
1172                         break;
1173                 default:
1174                         printk (KERN_ERR
1175                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1176                                 "or missing value\n", p);
1177                         return 0;
1178                 }
1179         }
1180 #ifdef CONFIG_QUOTA
1181         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1182                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1183                      sbi->s_qf_names[USRQUOTA])
1184                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1185
1186                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1187                      sbi->s_qf_names[GRPQUOTA])
1188                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1189
1190                 if ((sbi->s_qf_names[USRQUOTA] &&
1191                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1192                     (sbi->s_qf_names[GRPQUOTA] &&
1193                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1194                         printk(KERN_ERR "EXT3-fs: old and new quota "
1195                                         "format mixing.\n");
1196                         return 0;
1197                 }
1198
1199                 if (!sbi->s_jquota_fmt) {
1200                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1201                                         "not specified.\n");
1202                         return 0;
1203                 }
1204         } else {
1205                 if (sbi->s_jquota_fmt) {
1206                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1207                                         "specified with no journaling "
1208                                         "enabled.\n");
1209                         return 0;
1210                 }
1211         }
1212 #endif
1213         return 1;
1214 }
1215
1216 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1217                             int read_only)
1218 {
1219         struct ext3_sb_info *sbi = EXT3_SB(sb);
1220         int res = 0;
1221
1222         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1223                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1224                         "forcing read-only mode\n");
1225                 res = MS_RDONLY;
1226         }
1227         if (read_only)
1228                 return res;
1229         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1230                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1231                         "running e2fsck is recommended\n");
1232         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1233                 printk (KERN_WARNING
1234                         "EXT3-fs warning: mounting fs with errors, "
1235                         "running e2fsck is recommended\n");
1236         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1237                  le16_to_cpu(es->s_mnt_count) >=
1238                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1239                 printk (KERN_WARNING
1240                         "EXT3-fs warning: maximal mount count reached, "
1241                         "running e2fsck is recommended\n");
1242         else if (le32_to_cpu(es->s_checkinterval) &&
1243                 (le32_to_cpu(es->s_lastcheck) +
1244                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1245                 printk (KERN_WARNING
1246                         "EXT3-fs warning: checktime reached, "
1247                         "running e2fsck is recommended\n");
1248 #if 0
1249                 /* @@@ We _will_ want to clear the valid bit if we find
1250                    inconsistencies, to force a fsck at reboot.  But for
1251                    a plain journaled filesystem we can keep it set as
1252                    valid forever! :) */
1253         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1254 #endif
1255         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1256                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1257         le16_add_cpu(&es->s_mnt_count, 1);
1258         es->s_mtime = cpu_to_le32(get_seconds());
1259         ext3_update_dynamic_rev(sb);
1260         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1261
1262         ext3_commit_super(sb, es, 1);
1263         if (test_opt(sb, DEBUG))
1264                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1265                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1266                         sb->s_blocksize,
1267                         sbi->s_groups_count,
1268                         EXT3_BLOCKS_PER_GROUP(sb),
1269                         EXT3_INODES_PER_GROUP(sb),
1270                         sbi->s_mount_opt);
1271
1272         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1273         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1274                 char b[BDEVNAME_SIZE];
1275
1276                 printk("external journal on %s\n",
1277                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1278         } else {
1279                 printk("internal journal\n");
1280         }
1281         return res;
1282 }
1283
1284 /* Called at mount-time, super-block is locked */
1285 static int ext3_check_descriptors(struct super_block *sb)
1286 {
1287         struct ext3_sb_info *sbi = EXT3_SB(sb);
1288         int i;
1289
1290         ext3_debug ("Checking group descriptors");
1291
1292         for (i = 0; i < sbi->s_groups_count; i++) {
1293                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1294                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1295                 ext3_fsblk_t last_block;
1296
1297                 if (i == sbi->s_groups_count - 1)
1298                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1299                 else
1300                         last_block = first_block +
1301                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1302
1303                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1304                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1305                 {
1306                         ext3_error (sb, "ext3_check_descriptors",
1307                                     "Block bitmap for group %d"
1308                                     " not in group (block %lu)!",
1309                                     i, (unsigned long)
1310                                         le32_to_cpu(gdp->bg_block_bitmap));
1311                         return 0;
1312                 }
1313                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1314                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1315                 {
1316                         ext3_error (sb, "ext3_check_descriptors",
1317                                     "Inode bitmap for group %d"
1318                                     " not in group (block %lu)!",
1319                                     i, (unsigned long)
1320                                         le32_to_cpu(gdp->bg_inode_bitmap));
1321                         return 0;
1322                 }
1323                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1324                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1325                     last_block)
1326                 {
1327                         ext3_error (sb, "ext3_check_descriptors",
1328                                     "Inode table for group %d"
1329                                     " not in group (block %lu)!",
1330                                     i, (unsigned long)
1331                                         le32_to_cpu(gdp->bg_inode_table));
1332                         return 0;
1333                 }
1334         }
1335
1336         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1337         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1338         return 1;
1339 }
1340
1341
1342 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1343  * the superblock) which were deleted from all directories, but held open by
1344  * a process at the time of a crash.  We walk the list and try to delete these
1345  * inodes at recovery time (only with a read-write filesystem).
1346  *
1347  * In order to keep the orphan inode chain consistent during traversal (in
1348  * case of crash during recovery), we link each inode into the superblock
1349  * orphan list_head and handle it the same way as an inode deletion during
1350  * normal operation (which journals the operations for us).
1351  *
1352  * We only do an iget() and an iput() on each inode, which is very safe if we
1353  * accidentally point at an in-use or already deleted inode.  The worst that
1354  * can happen in this case is that we get a "bit already cleared" message from
1355  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1356  * e2fsck was run on this filesystem, and it must have already done the orphan
1357  * inode cleanup for us, so we can safely abort without any further action.
1358  */
1359 static void ext3_orphan_cleanup (struct super_block * sb,
1360                                  struct ext3_super_block * es)
1361 {
1362         unsigned int s_flags = sb->s_flags;
1363         int nr_orphans = 0, nr_truncates = 0;
1364 #ifdef CONFIG_QUOTA
1365         int i;
1366 #endif
1367         if (!es->s_last_orphan) {
1368                 jbd_debug(4, "no orphan inodes to clean up\n");
1369                 return;
1370         }
1371
1372         if (bdev_read_only(sb->s_bdev)) {
1373                 printk(KERN_ERR "EXT3-fs: write access "
1374                         "unavailable, skipping orphan cleanup.\n");
1375                 return;
1376         }
1377
1378         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1379                 if (es->s_last_orphan)
1380                         jbd_debug(1, "Errors on filesystem, "
1381                                   "clearing orphan list.\n");
1382                 es->s_last_orphan = 0;
1383                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1384                 return;
1385         }
1386
1387         if (s_flags & MS_RDONLY) {
1388                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1389                        sb->s_id);
1390                 sb->s_flags &= ~MS_RDONLY;
1391         }
1392 #ifdef CONFIG_QUOTA
1393         /* Needed for iput() to work correctly and not trash data */
1394         sb->s_flags |= MS_ACTIVE;
1395         /* Turn on quotas so that they are updated correctly */
1396         for (i = 0; i < MAXQUOTAS; i++) {
1397                 if (EXT3_SB(sb)->s_qf_names[i]) {
1398                         int ret = ext3_quota_on_mount(sb, i);
1399                         if (ret < 0)
1400                                 printk(KERN_ERR
1401                                         "EXT3-fs: Cannot turn on journaled "
1402                                         "quota: error %d\n", ret);
1403                 }
1404         }
1405 #endif
1406
1407         while (es->s_last_orphan) {
1408                 struct inode *inode;
1409
1410                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1411                 if (IS_ERR(inode)) {
1412                         es->s_last_orphan = 0;
1413                         break;
1414                 }
1415
1416                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1417                 DQUOT_INIT(inode);
1418                 if (inode->i_nlink) {
1419                         printk(KERN_DEBUG
1420                                 "%s: truncating inode %lu to %Ld bytes\n",
1421                                 __func__, inode->i_ino, inode->i_size);
1422                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1423                                   inode->i_ino, inode->i_size);
1424                         ext3_truncate(inode);
1425                         nr_truncates++;
1426                 } else {
1427                         printk(KERN_DEBUG
1428                                 "%s: deleting unreferenced inode %lu\n",
1429                                 __func__, inode->i_ino);
1430                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1431                                   inode->i_ino);
1432                         nr_orphans++;
1433                 }
1434                 iput(inode);  /* The delete magic happens here! */
1435         }
1436
1437 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1438
1439         if (nr_orphans)
1440                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1441                        sb->s_id, PLURAL(nr_orphans));
1442         if (nr_truncates)
1443                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1444                        sb->s_id, PLURAL(nr_truncates));
1445 #ifdef CONFIG_QUOTA
1446         /* Turn quotas off */
1447         for (i = 0; i < MAXQUOTAS; i++) {
1448                 if (sb_dqopt(sb)->files[i])
1449                         vfs_quota_off(sb, i, 0);
1450         }
1451 #endif
1452         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1453 }
1454
1455 /*
1456  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1457  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1458  * We need to be 1 filesystem block less than the 2^32 sector limit.
1459  */
1460 static loff_t ext3_max_size(int bits)
1461 {
1462         loff_t res = EXT3_NDIR_BLOCKS;
1463         int meta_blocks;
1464         loff_t upper_limit;
1465
1466         /* This is calculated to be the largest file size for a
1467          * dense, file such that the total number of
1468          * sectors in the file, including data and all indirect blocks,
1469          * does not exceed 2^32 -1
1470          * __u32 i_blocks representing the total number of
1471          * 512 bytes blocks of the file
1472          */
1473         upper_limit = (1LL << 32) - 1;
1474
1475         /* total blocks in file system block size */
1476         upper_limit >>= (bits - 9);
1477
1478
1479         /* indirect blocks */
1480         meta_blocks = 1;
1481         /* double indirect blocks */
1482         meta_blocks += 1 + (1LL << (bits-2));
1483         /* tripple indirect blocks */
1484         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1485
1486         upper_limit -= meta_blocks;
1487         upper_limit <<= bits;
1488
1489         res += 1LL << (bits-2);
1490         res += 1LL << (2*(bits-2));
1491         res += 1LL << (3*(bits-2));
1492         res <<= bits;
1493         if (res > upper_limit)
1494                 res = upper_limit;
1495
1496         if (res > MAX_LFS_FILESIZE)
1497                 res = MAX_LFS_FILESIZE;
1498
1499         return res;
1500 }
1501
1502 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1503                                     ext3_fsblk_t logic_sb_block,
1504                                     int nr)
1505 {
1506         struct ext3_sb_info *sbi = EXT3_SB(sb);
1507         unsigned long bg, first_meta_bg;
1508         int has_super = 0;
1509
1510         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1511
1512         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1513             nr < first_meta_bg)
1514                 return (logic_sb_block + nr + 1);
1515         bg = sbi->s_desc_per_block * nr;
1516         if (ext3_bg_has_super(sb, bg))
1517                 has_super = 1;
1518         return (has_super + ext3_group_first_block_no(sb, bg));
1519 }
1520
1521
1522 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1523 {
1524         struct buffer_head * bh;
1525         struct ext3_super_block *es = NULL;
1526         struct ext3_sb_info *sbi;
1527         ext3_fsblk_t block;
1528         ext3_fsblk_t sb_block = get_sb_block(&data);
1529         ext3_fsblk_t logic_sb_block;
1530         unsigned long offset = 0;
1531         unsigned int journal_inum = 0;
1532         unsigned long journal_devnum = 0;
1533         unsigned long def_mount_opts;
1534         struct inode *root;
1535         int blocksize;
1536         int hblock;
1537         int db_count;
1538         int i;
1539         int needs_recovery;
1540         int ret = -EINVAL;
1541         __le32 features;
1542         int err;
1543
1544         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1545         if (!sbi)
1546                 return -ENOMEM;
1547         sb->s_fs_info = sbi;
1548         sbi->s_mount_opt = 0;
1549         sbi->s_resuid = EXT3_DEF_RESUID;
1550         sbi->s_resgid = EXT3_DEF_RESGID;
1551         sbi->s_sb_block = sb_block;
1552
1553         unlock_kernel();
1554
1555         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1556         if (!blocksize) {
1557                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1558                 goto out_fail;
1559         }
1560
1561         /*
1562          * The ext3 superblock will not be buffer aligned for other than 1kB
1563          * block sizes.  We need to calculate the offset from buffer start.
1564          */
1565         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1566                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1567                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1568         } else {
1569                 logic_sb_block = sb_block;
1570         }
1571
1572         if (!(bh = sb_bread(sb, logic_sb_block))) {
1573                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1574                 goto out_fail;
1575         }
1576         /*
1577          * Note: s_es must be initialized as soon as possible because
1578          *       some ext3 macro-instructions depend on its value
1579          */
1580         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1581         sbi->s_es = es;
1582         sb->s_magic = le16_to_cpu(es->s_magic);
1583         if (sb->s_magic != EXT3_SUPER_MAGIC)
1584                 goto cantfind_ext3;
1585
1586         /* Set defaults before we parse the mount options */
1587         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1588         if (def_mount_opts & EXT3_DEFM_DEBUG)
1589                 set_opt(sbi->s_mount_opt, DEBUG);
1590         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1591                 set_opt(sbi->s_mount_opt, GRPID);
1592         if (def_mount_opts & EXT3_DEFM_UID16)
1593                 set_opt(sbi->s_mount_opt, NO_UID32);
1594 #ifdef CONFIG_EXT3_FS_XATTR
1595         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1596                 set_opt(sbi->s_mount_opt, XATTR_USER);
1597 #endif
1598 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1599         if (def_mount_opts & EXT3_DEFM_ACL)
1600                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1601 #endif
1602         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1603                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1604         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1605                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1606         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1607                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1608
1609         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1610                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1611         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1612                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1613         else
1614                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1615
1616         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1617         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1618
1619         set_opt(sbi->s_mount_opt, RESERVATION);
1620
1621         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1622                             NULL, 0))
1623                 goto failed_mount;
1624
1625         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1626                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1627
1628         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1629             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1630              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1631              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1632                 printk(KERN_WARNING
1633                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1634                        "running e2fsck is recommended\n");
1635         /*
1636          * Check feature flags regardless of the revision level, since we
1637          * previously didn't change the revision level when setting the flags,
1638          * so there is a chance incompat flags are set on a rev 0 filesystem.
1639          */
1640         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1641         if (features) {
1642                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1643                        "unsupported optional features (%x).\n",
1644                        sb->s_id, le32_to_cpu(features));
1645                 goto failed_mount;
1646         }
1647         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1648         if (!(sb->s_flags & MS_RDONLY) && features) {
1649                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1650                        "unsupported optional features (%x).\n",
1651                        sb->s_id, le32_to_cpu(features));
1652                 goto failed_mount;
1653         }
1654         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1655
1656         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1657             blocksize > EXT3_MAX_BLOCK_SIZE) {
1658                 printk(KERN_ERR
1659                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1660                        blocksize, sb->s_id);
1661                 goto failed_mount;
1662         }
1663
1664         hblock = bdev_hardsect_size(sb->s_bdev);
1665         if (sb->s_blocksize != blocksize) {
1666                 /*
1667                  * Make sure the blocksize for the filesystem is larger
1668                  * than the hardware sectorsize for the machine.
1669                  */
1670                 if (blocksize < hblock) {
1671                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1672                                "device blocksize %d.\n", blocksize, hblock);
1673                         goto failed_mount;
1674                 }
1675
1676                 brelse (bh);
1677                 if (!sb_set_blocksize(sb, blocksize)) {
1678                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1679                                 blocksize);
1680                         goto out_fail;
1681                 }
1682                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1683                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1684                 bh = sb_bread(sb, logic_sb_block);
1685                 if (!bh) {
1686                         printk(KERN_ERR
1687                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1688                         goto failed_mount;
1689                 }
1690                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1691                 sbi->s_es = es;
1692                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1693                         printk (KERN_ERR
1694                                 "EXT3-fs: Magic mismatch, very weird !\n");
1695                         goto failed_mount;
1696                 }
1697         }
1698
1699         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1700
1701         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1702                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1703                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1704         } else {
1705                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1706                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1707                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1708                     (!is_power_of_2(sbi->s_inode_size)) ||
1709                     (sbi->s_inode_size > blocksize)) {
1710                         printk (KERN_ERR
1711                                 "EXT3-fs: unsupported inode size: %d\n",
1712                                 sbi->s_inode_size);
1713                         goto failed_mount;
1714                 }
1715         }
1716         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1717                                    le32_to_cpu(es->s_log_frag_size);
1718         if (blocksize != sbi->s_frag_size) {
1719                 printk(KERN_ERR
1720                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1721                        sbi->s_frag_size, blocksize);
1722                 goto failed_mount;
1723         }
1724         sbi->s_frags_per_block = 1;
1725         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1726         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1727         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1728         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1729                 goto cantfind_ext3;
1730         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1731         if (sbi->s_inodes_per_block == 0)
1732                 goto cantfind_ext3;
1733         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1734                                         sbi->s_inodes_per_block;
1735         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1736         sbi->s_sbh = bh;
1737         sbi->s_mount_state = le16_to_cpu(es->s_state);
1738         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1739         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1740         for (i=0; i < 4; i++)
1741                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1742         sbi->s_def_hash_version = es->s_def_hash_version;
1743
1744         if (sbi->s_blocks_per_group > blocksize * 8) {
1745                 printk (KERN_ERR
1746                         "EXT3-fs: #blocks per group too big: %lu\n",
1747                         sbi->s_blocks_per_group);
1748                 goto failed_mount;
1749         }
1750         if (sbi->s_frags_per_group > blocksize * 8) {
1751                 printk (KERN_ERR
1752                         "EXT3-fs: #fragments per group too big: %lu\n",
1753                         sbi->s_frags_per_group);
1754                 goto failed_mount;
1755         }
1756         if (sbi->s_inodes_per_group > blocksize * 8) {
1757                 printk (KERN_ERR
1758                         "EXT3-fs: #inodes per group too big: %lu\n",
1759                         sbi->s_inodes_per_group);
1760                 goto failed_mount;
1761         }
1762
1763         if (le32_to_cpu(es->s_blocks_count) >
1764                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1765                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1766                         " too large to mount safely\n", sb->s_id);
1767                 if (sizeof(sector_t) < 8)
1768                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1769                                         "enabled\n");
1770                 goto failed_mount;
1771         }
1772
1773         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1774                 goto cantfind_ext3;
1775         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1776                                le32_to_cpu(es->s_first_data_block) - 1)
1777                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1778         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1779                    EXT3_DESC_PER_BLOCK(sb);
1780         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1781                                     GFP_KERNEL);
1782         if (sbi->s_group_desc == NULL) {
1783                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1784                 goto failed_mount;
1785         }
1786
1787         bgl_lock_init(&sbi->s_blockgroup_lock);
1788
1789         for (i = 0; i < db_count; i++) {
1790                 block = descriptor_loc(sb, logic_sb_block, i);
1791                 sbi->s_group_desc[i] = sb_bread(sb, block);
1792                 if (!sbi->s_group_desc[i]) {
1793                         printk (KERN_ERR "EXT3-fs: "
1794                                 "can't read group descriptor %d\n", i);
1795                         db_count = i;
1796                         goto failed_mount2;
1797                 }
1798         }
1799         if (!ext3_check_descriptors (sb)) {
1800                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1801                 goto failed_mount2;
1802         }
1803         sbi->s_gdb_count = db_count;
1804         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1805         spin_lock_init(&sbi->s_next_gen_lock);
1806
1807         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1808                         ext3_count_free_blocks(sb));
1809         if (!err) {
1810                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1811                                 ext3_count_free_inodes(sb));
1812         }
1813         if (!err) {
1814                 err = percpu_counter_init(&sbi->s_dirs_counter,
1815                                 ext3_count_dirs(sb));
1816         }
1817         if (err) {
1818                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1819                 goto failed_mount3;
1820         }
1821
1822         /* per fileystem reservation list head & lock */
1823         spin_lock_init(&sbi->s_rsv_window_lock);
1824         sbi->s_rsv_window_root = RB_ROOT;
1825         /* Add a single, static dummy reservation to the start of the
1826          * reservation window list --- it gives us a placeholder for
1827          * append-at-start-of-list which makes the allocation logic
1828          * _much_ simpler. */
1829         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1830         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1831         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1832         sbi->s_rsv_window_head.rsv_goal_size = 0;
1833         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1834
1835         /*
1836          * set up enough so that it can read an inode
1837          */
1838         sb->s_op = &ext3_sops;
1839         sb->s_export_op = &ext3_export_ops;
1840         sb->s_xattr = ext3_xattr_handlers;
1841 #ifdef CONFIG_QUOTA
1842         sb->s_qcop = &ext3_qctl_operations;
1843         sb->dq_op = &ext3_quota_operations;
1844 #endif
1845         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1846
1847         sb->s_root = NULL;
1848
1849         needs_recovery = (es->s_last_orphan != 0 ||
1850                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1851                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1852
1853         /*
1854          * The first inode we look at is the journal inode.  Don't try
1855          * root first: it may be modified in the journal!
1856          */
1857         if (!test_opt(sb, NOLOAD) &&
1858             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1859                 if (ext3_load_journal(sb, es, journal_devnum))
1860                         goto failed_mount3;
1861         } else if (journal_inum) {
1862                 if (ext3_create_journal(sb, es, journal_inum))
1863                         goto failed_mount3;
1864         } else {
1865                 if (!silent)
1866                         printk (KERN_ERR
1867                                 "ext3: No journal on filesystem on %s\n",
1868                                 sb->s_id);
1869                 goto failed_mount3;
1870         }
1871
1872         /* We have now updated the journal if required, so we can
1873          * validate the data journaling mode. */
1874         switch (test_opt(sb, DATA_FLAGS)) {
1875         case 0:
1876                 /* No mode set, assume a default based on the journal
1877                    capabilities: ORDERED_DATA if the journal can
1878                    cope, else JOURNAL_DATA */
1879                 if (journal_check_available_features
1880                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1881                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1882                 else
1883                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1884                 break;
1885
1886         case EXT3_MOUNT_ORDERED_DATA:
1887         case EXT3_MOUNT_WRITEBACK_DATA:
1888                 if (!journal_check_available_features
1889                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1890                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1891                                "requested data journaling mode\n");
1892                         goto failed_mount4;
1893                 }
1894         default:
1895                 break;
1896         }
1897
1898         if (test_opt(sb, NOBH)) {
1899                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1900                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1901                                 "its supported only with writeback mode\n");
1902                         clear_opt(sbi->s_mount_opt, NOBH);
1903                 }
1904         }
1905         /*
1906          * The journal_load will have done any necessary log recovery,
1907          * so we can safely mount the rest of the filesystem now.
1908          */
1909
1910         root = ext3_iget(sb, EXT3_ROOT_INO);
1911         if (IS_ERR(root)) {
1912                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1913                 ret = PTR_ERR(root);
1914                 goto failed_mount4;
1915         }
1916         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1917                 iput(root);
1918                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1919                 goto failed_mount4;
1920         }
1921         sb->s_root = d_alloc_root(root);
1922         if (!sb->s_root) {
1923                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1924                 iput(root);
1925                 ret = -ENOMEM;
1926                 goto failed_mount4;
1927         }
1928
1929         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1930         /*
1931          * akpm: core read_super() calls in here with the superblock locked.
1932          * That deadlocks, because orphan cleanup needs to lock the superblock
1933          * in numerous places.  Here we just pop the lock - it's relatively
1934          * harmless, because we are now ready to accept write_super() requests,
1935          * and aviro says that's the only reason for hanging onto the
1936          * superblock lock.
1937          */
1938         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1939         ext3_orphan_cleanup(sb, es);
1940         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1941         if (needs_recovery)
1942                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1943         ext3_mark_recovery_complete(sb, es);
1944         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1945                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1946                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1947                 "writeback");
1948
1949         lock_kernel();
1950         return 0;
1951
1952 cantfind_ext3:
1953         if (!silent)
1954                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1955                        sb->s_id);
1956         goto failed_mount;
1957
1958 failed_mount4:
1959         journal_destroy(sbi->s_journal);
1960 failed_mount3:
1961         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1962         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1963         percpu_counter_destroy(&sbi->s_dirs_counter);
1964 failed_mount2:
1965         for (i = 0; i < db_count; i++)
1966                 brelse(sbi->s_group_desc[i]);
1967         kfree(sbi->s_group_desc);
1968 failed_mount:
1969 #ifdef CONFIG_QUOTA
1970         for (i = 0; i < MAXQUOTAS; i++)
1971                 kfree(sbi->s_qf_names[i]);
1972 #endif
1973         ext3_blkdev_remove(sbi);
1974         brelse(bh);
1975 out_fail:
1976         sb->s_fs_info = NULL;
1977         kfree(sbi);
1978         lock_kernel();
1979         return ret;
1980 }
1981
1982 /*
1983  * Setup any per-fs journal parameters now.  We'll do this both on
1984  * initial mount, once the journal has been initialised but before we've
1985  * done any recovery; and again on any subsequent remount.
1986  */
1987 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1988 {
1989         struct ext3_sb_info *sbi = EXT3_SB(sb);
1990
1991         if (sbi->s_commit_interval)
1992                 journal->j_commit_interval = sbi->s_commit_interval;
1993         /* We could also set up an ext3-specific default for the commit
1994          * interval here, but for now we'll just fall back to the jbd
1995          * default. */
1996
1997         spin_lock(&journal->j_state_lock);
1998         if (test_opt(sb, BARRIER))
1999                 journal->j_flags |= JFS_BARRIER;
2000         else
2001                 journal->j_flags &= ~JFS_BARRIER;
2002         if (test_opt(sb, DATA_ERR_ABORT))
2003                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2004         else
2005                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2006         spin_unlock(&journal->j_state_lock);
2007 }
2008
2009 static journal_t *ext3_get_journal(struct super_block *sb,
2010                                    unsigned int journal_inum)
2011 {
2012         struct inode *journal_inode;
2013         journal_t *journal;
2014
2015         /* First, test for the existence of a valid inode on disk.  Bad
2016          * things happen if we iget() an unused inode, as the subsequent
2017          * iput() will try to delete it. */
2018
2019         journal_inode = ext3_iget(sb, journal_inum);
2020         if (IS_ERR(journal_inode)) {
2021                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2022                 return NULL;
2023         }
2024         if (!journal_inode->i_nlink) {
2025                 make_bad_inode(journal_inode);
2026                 iput(journal_inode);
2027                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2028                 return NULL;
2029         }
2030
2031         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2032                   journal_inode, journal_inode->i_size);
2033         if (!S_ISREG(journal_inode->i_mode)) {
2034                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2035                 iput(journal_inode);
2036                 return NULL;
2037         }
2038
2039         journal = journal_init_inode(journal_inode);
2040         if (!journal) {
2041                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2042                 iput(journal_inode);
2043                 return NULL;
2044         }
2045         journal->j_private = sb;
2046         ext3_init_journal_params(sb, journal);
2047         return journal;
2048 }
2049
2050 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2051                                        dev_t j_dev)
2052 {
2053         struct buffer_head * bh;
2054         journal_t *journal;
2055         ext3_fsblk_t start;
2056         ext3_fsblk_t len;
2057         int hblock, blocksize;
2058         ext3_fsblk_t sb_block;
2059         unsigned long offset;
2060         struct ext3_super_block * es;
2061         struct block_device *bdev;
2062
2063         bdev = ext3_blkdev_get(j_dev);
2064         if (bdev == NULL)
2065                 return NULL;
2066
2067         if (bd_claim(bdev, sb)) {
2068                 printk(KERN_ERR
2069                         "EXT3: failed to claim external journal device.\n");
2070                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2071                 return NULL;
2072         }
2073
2074         blocksize = sb->s_blocksize;
2075         hblock = bdev_hardsect_size(bdev);
2076         if (blocksize < hblock) {
2077                 printk(KERN_ERR
2078                         "EXT3-fs: blocksize too small for journal device.\n");
2079                 goto out_bdev;
2080         }
2081
2082         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2083         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2084         set_blocksize(bdev, blocksize);
2085         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2086                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2087                        "external journal\n");
2088                 goto out_bdev;
2089         }
2090
2091         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2092         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2093             !(le32_to_cpu(es->s_feature_incompat) &
2094               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2095                 printk(KERN_ERR "EXT3-fs: external journal has "
2096                                         "bad superblock\n");
2097                 brelse(bh);
2098                 goto out_bdev;
2099         }
2100
2101         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2102                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2103                 brelse(bh);
2104                 goto out_bdev;
2105         }
2106
2107         len = le32_to_cpu(es->s_blocks_count);
2108         start = sb_block + 1;
2109         brelse(bh);     /* we're done with the superblock */
2110
2111         journal = journal_init_dev(bdev, sb->s_bdev,
2112                                         start, len, blocksize);
2113         if (!journal) {
2114                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2115                 goto out_bdev;
2116         }
2117         journal->j_private = sb;
2118         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2119         wait_on_buffer(journal->j_sb_buffer);
2120         if (!buffer_uptodate(journal->j_sb_buffer)) {
2121                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2122                 goto out_journal;
2123         }
2124         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2125                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2126                                         "user (unsupported) - %d\n",
2127                         be32_to_cpu(journal->j_superblock->s_nr_users));
2128                 goto out_journal;
2129         }
2130         EXT3_SB(sb)->journal_bdev = bdev;
2131         ext3_init_journal_params(sb, journal);
2132         return journal;
2133 out_journal:
2134         journal_destroy(journal);
2135 out_bdev:
2136         ext3_blkdev_put(bdev);
2137         return NULL;
2138 }
2139
2140 static int ext3_load_journal(struct super_block *sb,
2141                              struct ext3_super_block *es,
2142                              unsigned long journal_devnum)
2143 {
2144         journal_t *journal;
2145         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2146         dev_t journal_dev;
2147         int err = 0;
2148         int really_read_only;
2149
2150         if (journal_devnum &&
2151             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2152                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2153                         "numbers have changed\n");
2154                 journal_dev = new_decode_dev(journal_devnum);
2155         } else
2156                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2157
2158         really_read_only = bdev_read_only(sb->s_bdev);
2159
2160         /*
2161          * Are we loading a blank journal or performing recovery after a
2162          * crash?  For recovery, we need to check in advance whether we
2163          * can get read-write access to the device.
2164          */
2165
2166         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2167                 if (sb->s_flags & MS_RDONLY) {
2168                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2169                                         "required on readonly filesystem.\n");
2170                         if (really_read_only) {
2171                                 printk(KERN_ERR "EXT3-fs: write access "
2172                                         "unavailable, cannot proceed.\n");
2173                                 return -EROFS;
2174                         }
2175                         printk (KERN_INFO "EXT3-fs: write access will "
2176                                         "be enabled during recovery.\n");
2177                 }
2178         }
2179
2180         if (journal_inum && journal_dev) {
2181                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2182                        "and inode journals!\n");
2183                 return -EINVAL;
2184         }
2185
2186         if (journal_inum) {
2187                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2188                         return -EINVAL;
2189         } else {
2190                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2191                         return -EINVAL;
2192         }
2193
2194         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2195                 err = journal_update_format(journal);
2196                 if (err)  {
2197                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2198                         journal_destroy(journal);
2199                         return err;
2200                 }
2201         }
2202
2203         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2204                 err = journal_wipe(journal, !really_read_only);
2205         if (!err)
2206                 err = journal_load(journal);
2207
2208         if (err) {
2209                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2210                 journal_destroy(journal);
2211                 return err;
2212         }
2213
2214         EXT3_SB(sb)->s_journal = journal;
2215         ext3_clear_journal_err(sb, es);
2216
2217         if (journal_devnum &&
2218             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2219                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2220                 sb->s_dirt = 1;
2221
2222                 /* Make sure we flush the recovery flag to disk. */
2223                 ext3_commit_super(sb, es, 1);
2224         }
2225
2226         return 0;
2227 }
2228
2229 static int ext3_create_journal(struct super_block * sb,
2230                                struct ext3_super_block * es,
2231                                unsigned int journal_inum)
2232 {
2233         journal_t *journal;
2234         int err;
2235
2236         if (sb->s_flags & MS_RDONLY) {
2237                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2238                                 "create journal.\n");
2239                 return -EROFS;
2240         }
2241
2242         journal = ext3_get_journal(sb, journal_inum);
2243         if (!journal)
2244                 return -EINVAL;
2245
2246         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2247                journal_inum);
2248
2249         err = journal_create(journal);
2250         if (err) {
2251                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2252                 journal_destroy(journal);
2253                 return -EIO;
2254         }
2255
2256         EXT3_SB(sb)->s_journal = journal;
2257
2258         ext3_update_dynamic_rev(sb);
2259         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2260         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2261
2262         es->s_journal_inum = cpu_to_le32(journal_inum);
2263         sb->s_dirt = 1;
2264
2265         /* Make sure we flush the recovery flag to disk. */
2266         ext3_commit_super(sb, es, 1);
2267
2268         return 0;
2269 }
2270
2271 static void ext3_commit_super (struct super_block * sb,
2272                                struct ext3_super_block * es,
2273                                int sync)
2274 {
2275         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2276
2277         if (!sbh)
2278                 return;
2279         es->s_wtime = cpu_to_le32(get_seconds());
2280         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2281         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2282         BUFFER_TRACE(sbh, "marking dirty");
2283         mark_buffer_dirty(sbh);
2284         if (sync)
2285                 sync_dirty_buffer(sbh);
2286 }
2287
2288
2289 /*
2290  * Have we just finished recovery?  If so, and if we are mounting (or
2291  * remounting) the filesystem readonly, then we will end up with a
2292  * consistent fs on disk.  Record that fact.
2293  */
2294 static void ext3_mark_recovery_complete(struct super_block * sb,
2295                                         struct ext3_super_block * es)
2296 {
2297         journal_t *journal = EXT3_SB(sb)->s_journal;
2298
2299         journal_lock_updates(journal);
2300         if (journal_flush(journal) < 0)
2301                 goto out;
2302
2303         lock_super(sb);
2304         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2305             sb->s_flags & MS_RDONLY) {
2306                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2307                 sb->s_dirt = 0;
2308                 ext3_commit_super(sb, es, 1);
2309         }
2310         unlock_super(sb);
2311
2312 out:
2313         journal_unlock_updates(journal);
2314 }
2315
2316 /*
2317  * If we are mounting (or read-write remounting) a filesystem whose journal
2318  * has recorded an error from a previous lifetime, move that error to the
2319  * main filesystem now.
2320  */
2321 static void ext3_clear_journal_err(struct super_block * sb,
2322                                    struct ext3_super_block * es)
2323 {
2324         journal_t *journal;
2325         int j_errno;
2326         const char *errstr;
2327
2328         journal = EXT3_SB(sb)->s_journal;
2329
2330         /*
2331          * Now check for any error status which may have been recorded in the
2332          * journal by a prior ext3_error() or ext3_abort()
2333          */
2334
2335         j_errno = journal_errno(journal);
2336         if (j_errno) {
2337                 char nbuf[16];
2338
2339                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2340                 ext3_warning(sb, __func__, "Filesystem error recorded "
2341                              "from previous mount: %s", errstr);
2342                 ext3_warning(sb, __func__, "Marking fs in need of "
2343                              "filesystem check.");
2344
2345                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2346                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2347                 ext3_commit_super (sb, es, 1);
2348
2349                 journal_clear_err(journal);
2350         }
2351 }
2352
2353 /*
2354  * Force the running and committing transactions to commit,
2355  * and wait on the commit.
2356  */
2357 int ext3_force_commit(struct super_block *sb)
2358 {
2359         journal_t *journal;
2360         int ret;
2361
2362         if (sb->s_flags & MS_RDONLY)
2363                 return 0;
2364
2365         journal = EXT3_SB(sb)->s_journal;
2366         sb->s_dirt = 0;
2367         ret = ext3_journal_force_commit(journal);
2368         return ret;
2369 }
2370
2371 /*
2372  * Ext3 always journals updates to the superblock itself, so we don't
2373  * have to propagate any other updates to the superblock on disk at this
2374  * point.  Just start an async writeback to get the buffers on their way
2375  * to the disk.
2376  *
2377  * This implicitly triggers the writebehind on sync().
2378  */
2379
2380 static void ext3_write_super (struct super_block * sb)
2381 {
2382         if (mutex_trylock(&sb->s_lock) != 0)
2383                 BUG();
2384         sb->s_dirt = 0;
2385 }
2386
2387 static int ext3_sync_fs(struct super_block *sb, int wait)
2388 {
2389         tid_t target;
2390
2391         sb->s_dirt = 0;
2392         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2393                 if (wait)
2394                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2395         }
2396         return 0;
2397 }
2398
2399 /*
2400  * LVM calls this function before a (read-only) snapshot is created.  This
2401  * gives us a chance to flush the journal completely and mark the fs clean.
2402  */
2403 static void ext3_write_super_lockfs(struct super_block *sb)
2404 {
2405         sb->s_dirt = 0;
2406
2407         if (!(sb->s_flags & MS_RDONLY)) {
2408                 journal_t *journal = EXT3_SB(sb)->s_journal;
2409
2410                 /* Now we set up the journal barrier. */
2411                 journal_lock_updates(journal);
2412
2413                 /*
2414                  * We don't want to clear needs_recovery flag when we failed
2415                  * to flush the journal.
2416                  */
2417                 if (journal_flush(journal) < 0)
2418                         return;
2419
2420                 /* Journal blocked and flushed, clear needs_recovery flag. */
2421                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2422                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2423         }
2424 }
2425
2426 /*
2427  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2428  * flag here, even though the filesystem is not technically dirty yet.
2429  */
2430 static void ext3_unlockfs(struct super_block *sb)
2431 {
2432         if (!(sb->s_flags & MS_RDONLY)) {
2433                 lock_super(sb);
2434                 /* Reser the needs_recovery flag before the fs is unlocked. */
2435                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2436                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2437                 unlock_super(sb);
2438                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2439         }
2440 }
2441
2442 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2443 {
2444         struct ext3_super_block * es;
2445         struct ext3_sb_info *sbi = EXT3_SB(sb);
2446         ext3_fsblk_t n_blocks_count = 0;
2447         unsigned long old_sb_flags;
2448         struct ext3_mount_options old_opts;
2449         int err;
2450 #ifdef CONFIG_QUOTA
2451         int i;
2452 #endif
2453
2454         /* Store the original options */
2455         old_sb_flags = sb->s_flags;
2456         old_opts.s_mount_opt = sbi->s_mount_opt;
2457         old_opts.s_resuid = sbi->s_resuid;
2458         old_opts.s_resgid = sbi->s_resgid;
2459         old_opts.s_commit_interval = sbi->s_commit_interval;
2460 #ifdef CONFIG_QUOTA
2461         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2462         for (i = 0; i < MAXQUOTAS; i++)
2463                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2464 #endif
2465
2466         /*
2467          * Allow the "check" option to be passed as a remount option.
2468          */
2469         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2470                 err = -EINVAL;
2471                 goto restore_opts;
2472         }
2473
2474         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2475                 ext3_abort(sb, __func__, "Abort forced by user");
2476
2477         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2478                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2479
2480         es = sbi->s_es;
2481
2482         ext3_init_journal_params(sb, sbi->s_journal);
2483
2484         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2485                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2486                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2487                         err = -EROFS;
2488                         goto restore_opts;
2489                 }
2490
2491                 if (*flags & MS_RDONLY) {
2492                         /*
2493                          * First of all, the unconditional stuff we have to do
2494                          * to disable replay of the journal when we next remount
2495                          */
2496                         sb->s_flags |= MS_RDONLY;
2497
2498                         /*
2499                          * OK, test if we are remounting a valid rw partition
2500                          * readonly, and if so set the rdonly flag and then
2501                          * mark the partition as valid again.
2502                          */
2503                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2504                             (sbi->s_mount_state & EXT3_VALID_FS))
2505                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2506
2507                         /*
2508                          * We have to unlock super so that we can wait for
2509                          * transactions.
2510                          */
2511                         unlock_super(sb);
2512                         ext3_mark_recovery_complete(sb, es);
2513                         lock_super(sb);
2514                 } else {
2515                         __le32 ret;
2516                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2517                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2518                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2519                                        "remount RDWR because of unsupported "
2520                                        "optional features (%x).\n",
2521                                        sb->s_id, le32_to_cpu(ret));
2522                                 err = -EROFS;
2523                                 goto restore_opts;
2524                         }
2525
2526                         /*
2527                          * If we have an unprocessed orphan list hanging
2528                          * around from a previously readonly bdev mount,
2529                          * require a full umount/remount for now.
2530                          */
2531                         if (es->s_last_orphan) {
2532                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2533                                        "remount RDWR because of unprocessed "
2534                                        "orphan inode list.  Please "
2535                                        "umount/remount instead.\n",
2536                                        sb->s_id);
2537                                 err = -EINVAL;
2538                                 goto restore_opts;
2539                         }
2540
2541                         /*
2542                          * Mounting a RDONLY partition read-write, so reread
2543                          * and store the current valid flag.  (It may have
2544                          * been changed by e2fsck since we originally mounted
2545                          * the partition.)
2546                          */
2547                         ext3_clear_journal_err(sb, es);
2548                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2549                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2550                                 goto restore_opts;
2551                         if (!ext3_setup_super (sb, es, 0))
2552                                 sb->s_flags &= ~MS_RDONLY;
2553                 }
2554         }
2555 #ifdef CONFIG_QUOTA
2556         /* Release old quota file names */
2557         for (i = 0; i < MAXQUOTAS; i++)
2558                 if (old_opts.s_qf_names[i] &&
2559                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2560                         kfree(old_opts.s_qf_names[i]);
2561 #endif
2562         return 0;
2563 restore_opts:
2564         sb->s_flags = old_sb_flags;
2565         sbi->s_mount_opt = old_opts.s_mount_opt;
2566         sbi->s_resuid = old_opts.s_resuid;
2567         sbi->s_resgid = old_opts.s_resgid;
2568         sbi->s_commit_interval = old_opts.s_commit_interval;
2569 #ifdef CONFIG_QUOTA
2570         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2571         for (i = 0; i < MAXQUOTAS; i++) {
2572                 if (sbi->s_qf_names[i] &&
2573                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2574                         kfree(sbi->s_qf_names[i]);
2575                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2576         }
2577 #endif
2578         return err;
2579 }
2580
2581 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2582 {
2583         struct super_block *sb = dentry->d_sb;
2584         struct ext3_sb_info *sbi = EXT3_SB(sb);
2585         struct ext3_super_block *es = sbi->s_es;
2586         u64 fsid;
2587
2588         if (test_opt(sb, MINIX_DF)) {
2589                 sbi->s_overhead_last = 0;
2590         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2591                 unsigned long ngroups = sbi->s_groups_count, i;
2592                 ext3_fsblk_t overhead = 0;
2593                 smp_rmb();
2594
2595                 /*
2596                  * Compute the overhead (FS structures).  This is constant
2597                  * for a given filesystem unless the number of block groups
2598                  * changes so we cache the previous value until it does.
2599                  */
2600
2601                 /*
2602                  * All of the blocks before first_data_block are
2603                  * overhead
2604                  */
2605                 overhead = le32_to_cpu(es->s_first_data_block);
2606
2607                 /*
2608                  * Add the overhead attributed to the superblock and
2609                  * block group descriptors.  If the sparse superblocks
2610                  * feature is turned on, then not all groups have this.
2611                  */
2612                 for (i = 0; i < ngroups; i++) {
2613                         overhead += ext3_bg_has_super(sb, i) +
2614                                 ext3_bg_num_gdb(sb, i);
2615                         cond_resched();
2616                 }
2617
2618                 /*
2619                  * Every block group has an inode bitmap, a block
2620                  * bitmap, and an inode table.
2621                  */
2622                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2623                 sbi->s_overhead_last = overhead;
2624                 smp_wmb();
2625                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2626         }
2627
2628         buf->f_type = EXT3_SUPER_MAGIC;
2629         buf->f_bsize = sb->s_blocksize;
2630         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2631         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2632         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2633         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2634         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2635                 buf->f_bavail = 0;
2636         buf->f_files = le32_to_cpu(es->s_inodes_count);
2637         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2638         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2639         buf->f_namelen = EXT3_NAME_LEN;
2640         fsid = le64_to_cpup((void *)es->s_uuid) ^
2641                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2642         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2643         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2644         return 0;
2645 }
2646
2647 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2648  * is locked for write. Otherwise the are possible deadlocks:
2649  * Process 1                         Process 2
2650  * ext3_create()                     quota_sync()
2651  *   journal_start()                   write_dquot()
2652  *   DQUOT_INIT()                        down(dqio_mutex)
2653  *     down(dqio_mutex)                    journal_start()
2654  *
2655  */
2656
2657 #ifdef CONFIG_QUOTA
2658
2659 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2660 {
2661         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2662 }
2663
2664 static int ext3_dquot_initialize(struct inode *inode, int type)
2665 {
2666         handle_t *handle;
2667         int ret, err;
2668
2669         /* We may create quota structure so we need to reserve enough blocks */
2670         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2671         if (IS_ERR(handle))
2672                 return PTR_ERR(handle);
2673         ret = dquot_initialize(inode, type);
2674         err = ext3_journal_stop(handle);
2675         if (!ret)
2676                 ret = err;
2677         return ret;
2678 }
2679
2680 static int ext3_dquot_drop(struct inode *inode)
2681 {
2682         handle_t *handle;
2683         int ret, err;
2684
2685         /* We may delete quota structure so we need to reserve enough blocks */
2686         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2687         if (IS_ERR(handle)) {
2688                 /*
2689                  * We call dquot_drop() anyway to at least release references
2690                  * to quota structures so that umount does not hang.
2691                  */
2692                 dquot_drop(inode);
2693                 return PTR_ERR(handle);
2694         }
2695         ret = dquot_drop(inode);
2696         err = ext3_journal_stop(handle);
2697         if (!ret)
2698                 ret = err;
2699         return ret;
2700 }
2701
2702 static int ext3_write_dquot(struct dquot *dquot)
2703 {
2704         int ret, err;
2705         handle_t *handle;
2706         struct inode *inode;
2707
2708         inode = dquot_to_inode(dquot);
2709         handle = ext3_journal_start(inode,
2710                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2711         if (IS_ERR(handle))
2712                 return PTR_ERR(handle);
2713         ret = dquot_commit(dquot);
2714         err = ext3_journal_stop(handle);
2715         if (!ret)
2716                 ret = err;
2717         return ret;
2718 }
2719
2720 static int ext3_acquire_dquot(struct dquot *dquot)
2721 {
2722         int ret, err;
2723         handle_t *handle;
2724
2725         handle = ext3_journal_start(dquot_to_inode(dquot),
2726                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2727         if (IS_ERR(handle))
2728                 return PTR_ERR(handle);
2729         ret = dquot_acquire(dquot);
2730         err = ext3_journal_stop(handle);
2731         if (!ret)
2732                 ret = err;
2733         return ret;
2734 }
2735
2736 static int ext3_release_dquot(struct dquot *dquot)
2737 {
2738         int ret, err;
2739         handle_t *handle;
2740
2741         handle = ext3_journal_start(dquot_to_inode(dquot),
2742                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2743         if (IS_ERR(handle)) {
2744                 /* Release dquot anyway to avoid endless cycle in dqput() */
2745                 dquot_release(dquot);
2746                 return PTR_ERR(handle);
2747         }
2748         ret = dquot_release(dquot);
2749         err = ext3_journal_stop(handle);
2750         if (!ret)
2751                 ret = err;
2752         return ret;
2753 }
2754
2755 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2756 {
2757         /* Are we journaling quotas? */
2758         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2759             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2760                 dquot_mark_dquot_dirty(dquot);
2761                 return ext3_write_dquot(dquot);
2762         } else {
2763                 return dquot_mark_dquot_dirty(dquot);
2764         }
2765 }
2766
2767 static int ext3_write_info(struct super_block *sb, int type)
2768 {
2769         int ret, err;
2770         handle_t *handle;
2771
2772         /* Data block + inode block */
2773         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2774         if (IS_ERR(handle))
2775                 return PTR_ERR(handle);
2776         ret = dquot_commit_info(sb, type);
2777         err = ext3_journal_stop(handle);
2778         if (!ret)
2779                 ret = err;
2780         return ret;
2781 }
2782
2783 /*
2784  * Turn on quotas during mount time - we need to find
2785  * the quota file and such...
2786  */
2787 static int ext3_quota_on_mount(struct super_block *sb, int type)
2788 {
2789         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2790                         EXT3_SB(sb)->s_jquota_fmt, type);
2791 }
2792
2793 /*
2794  * Standard function to be called on quota_on
2795  */
2796 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2797                          char *name, int remount)
2798 {
2799         int err;
2800         struct path path;
2801
2802         if (!test_opt(sb, QUOTA))
2803                 return -EINVAL;
2804         /* When remounting, no checks are needed and in fact, name is NULL */
2805         if (remount)
2806                 return vfs_quota_on(sb, type, format_id, name, remount);
2807
2808         err = kern_path(name, LOOKUP_FOLLOW, &path);
2809         if (err)
2810                 return err;
2811
2812         /* Quotafile not on the same filesystem? */
2813         if (path.mnt->mnt_sb != sb) {
2814                 path_put(&path);
2815                 return -EXDEV;
2816         }
2817         /* Journaling quota? */
2818         if (EXT3_SB(sb)->s_qf_names[type]) {
2819                 /* Quotafile not of fs root? */
2820                 if (path.dentry->d_parent != sb->s_root)
2821                         printk(KERN_WARNING
2822                                 "EXT3-fs: Quota file not on filesystem root. "
2823                                 "Journaled quota will not work.\n");
2824         }
2825
2826         /*
2827          * When we journal data on quota file, we have to flush journal to see
2828          * all updates to the file when we bypass pagecache...
2829          */
2830         if (ext3_should_journal_data(path.dentry->d_inode)) {
2831                 /*
2832                  * We don't need to lock updates but journal_flush() could
2833                  * otherwise be livelocked...
2834                  */
2835                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2836                 err = journal_flush(EXT3_SB(sb)->s_journal);
2837                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2838                 if (err) {
2839                         path_put(&nd.path);
2840                         return err;
2841                 }
2842         }
2843
2844         err = vfs_quota_on_path(sb, type, format_id, &path);
2845         path_put(&path);
2846         return err;
2847 }
2848
2849 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2850  * acquiring the locks... As quota files are never truncated and quota code
2851  * itself serializes the operations (and noone else should touch the files)
2852  * we don't have to be afraid of races */
2853 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2854                                size_t len, loff_t off)
2855 {
2856         struct inode *inode = sb_dqopt(sb)->files[type];
2857         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2858         int err = 0;
2859         int offset = off & (sb->s_blocksize - 1);
2860         int tocopy;
2861         size_t toread;
2862         struct buffer_head *bh;
2863         loff_t i_size = i_size_read(inode);
2864
2865         if (off > i_size)
2866                 return 0;
2867         if (off+len > i_size)
2868                 len = i_size-off;
2869         toread = len;
2870         while (toread > 0) {
2871                 tocopy = sb->s_blocksize - offset < toread ?
2872                                 sb->s_blocksize - offset : toread;
2873                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2874                 if (err)
2875                         return err;
2876                 if (!bh)        /* A hole? */
2877                         memset(data, 0, tocopy);
2878                 else
2879                         memcpy(data, bh->b_data+offset, tocopy);
2880                 brelse(bh);
2881                 offset = 0;
2882                 toread -= tocopy;
2883                 data += tocopy;
2884                 blk++;
2885         }
2886         return len;
2887 }
2888
2889 /* Write to quotafile (we know the transaction is already started and has
2890  * enough credits) */
2891 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2892                                 const char *data, size_t len, loff_t off)
2893 {
2894         struct inode *inode = sb_dqopt(sb)->files[type];
2895         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2896         int err = 0;
2897         int offset = off & (sb->s_blocksize - 1);
2898         int tocopy;
2899         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2900         size_t towrite = len;
2901         struct buffer_head *bh;
2902         handle_t *handle = journal_current_handle();
2903
2904         if (!handle) {
2905                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2906                         " cancelled because transaction is not started.\n",
2907                         (unsigned long long)off, (unsigned long long)len);
2908                 return -EIO;
2909         }
2910         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2911         while (towrite > 0) {
2912                 tocopy = sb->s_blocksize - offset < towrite ?
2913                                 sb->s_blocksize - offset : towrite;
2914                 bh = ext3_bread(handle, inode, blk, 1, &err);
2915                 if (!bh)
2916                         goto out;
2917                 if (journal_quota) {
2918                         err = ext3_journal_get_write_access(handle, bh);
2919                         if (err) {
2920                                 brelse(bh);
2921                                 goto out;
2922                         }
2923                 }
2924                 lock_buffer(bh);
2925                 memcpy(bh->b_data+offset, data, tocopy);
2926                 flush_dcache_page(bh->b_page);
2927                 unlock_buffer(bh);
2928                 if (journal_quota)
2929                         err = ext3_journal_dirty_metadata(handle, bh);
2930                 else {
2931                         /* Always do at least ordered writes for quotas */
2932                         err = ext3_journal_dirty_data(handle, bh);
2933                         mark_buffer_dirty(bh);
2934                 }
2935                 brelse(bh);
2936                 if (err)
2937                         goto out;
2938                 offset = 0;
2939                 towrite -= tocopy;
2940                 data += tocopy;
2941                 blk++;
2942         }
2943 out:
2944         if (len == towrite) {
2945                 mutex_unlock(&inode->i_mutex);
2946                 return err;
2947         }
2948         if (inode->i_size < off+len-towrite) {
2949                 i_size_write(inode, off+len-towrite);
2950                 EXT3_I(inode)->i_disksize = inode->i_size;
2951         }
2952         inode->i_version++;
2953         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2954         ext3_mark_inode_dirty(handle, inode);
2955         mutex_unlock(&inode->i_mutex);
2956         return len - towrite;
2957 }
2958
2959 #endif
2960
2961 static int ext3_get_sb(struct file_system_type *fs_type,
2962         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2963 {
2964         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2965 }
2966
2967 static struct file_system_type ext3_fs_type = {
2968         .owner          = THIS_MODULE,
2969         .name           = "ext3",
2970         .get_sb         = ext3_get_sb,
2971         .kill_sb        = kill_block_super,
2972         .fs_flags       = FS_REQUIRES_DEV,
2973 };
2974
2975 static int __init init_ext3_fs(void)
2976 {
2977         int err = init_ext3_xattr();
2978         if (err)
2979                 return err;
2980         err = init_inodecache();
2981         if (err)
2982                 goto out1;
2983         err = register_filesystem(&ext3_fs_type);
2984         if (err)
2985                 goto out;
2986         return 0;
2987 out:
2988         destroy_inodecache();
2989 out1:
2990         exit_ext3_xattr();
2991         return err;
2992 }
2993
2994 static void __exit exit_ext3_fs(void)
2995 {
2996         unregister_filesystem(&ext3_fs_type);
2997         destroy_inodecache();
2998         exit_ext3_xattr();
2999 }
3000
3001 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3002 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3003 MODULE_LICENSE("GPL");
3004 module_init(init_ext3_fs)
3005 module_exit(exit_ext3_fs)