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/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/log2.h>
  39 #include <linux/cleancache.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 #define CREATE_TRACE_POINTS
  48 #include <trace/events/ext3.h>
  49
  50 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  51   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  52 #else
  53   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  54 #endif
  55
  56 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  57                              unsigned long journal_devnum);
  58 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  59                                unsigned int);
  60 static int ext3_commit_super(struct super_block *sb,
  61                                struct ext3_super_block *es,
  62                                int sync);
  63 static void ext3_mark_recovery_complete(struct super_block * sb,
  64                                         struct ext3_super_block * es);
  65 static void ext3_clear_journal_err(struct super_block * sb,
  66                                    struct ext3_super_block * es);
  67 static int ext3_sync_fs(struct super_block *sb, int wait);
  68 static const char *ext3_decode_error(struct super_block * sb, int errno,
  69                                      char nbuf[16]);
  70 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  71 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  72 static int ext3_unfreeze(struct super_block *sb);
  73 static int ext3_freeze(struct super_block *sb);
  74
  75 /*
  76  * Wrappers for journal_start/end.
  77  *
  78  * The only special thing we need to do here is to make sure that all
  79  * journal_end calls result in the superblock being marked dirty, so
  80  * that sync() will call the filesystem's write_super callback if
  81  * appropriate.
  82  */
  83 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  84 {
  85         journal_t *journal;
  86
  87         if (sb->s_flags & MS_RDONLY)
  88                 return ERR_PTR(-EROFS);
  89
  90         /* Special case here: if the journal has aborted behind our
  91          * backs (eg. EIO in the commit thread), then we still need to
  92          * take the FS itself readonly cleanly. */
  93         journal = EXT3_SB(sb)->s_journal;
  94         if (is_journal_aborted(journal)) {
  95                 ext3_abort(sb, __func__,
  96                            "Detected aborted journal");
  97                 return ERR_PTR(-EROFS);
  98         }
  99
 100         return journal_start(journal, nblocks);
 101 }
 102
 103 /*
 104  * The only special thing we need to do here is to make sure that all
 105  * journal_stop calls result in the superblock being marked dirty, so
 106  * that sync() will call the filesystem's write_super callback if
 107  * appropriate.
 108  */
 109 int __ext3_journal_stop(const char *where, handle_t *handle)
 110 {
 111         struct super_block *sb;
 112         int err;
 113         int rc;
 114
 115         sb = handle->h_transaction->t_journal->j_private;
 116         err = handle->h_err;
 117         rc = journal_stop(handle);
 118
 119         if (!err)
 120                 err = rc;
 121         if (err)
 122                 __ext3_std_error(sb, where, err);
 123         return err;
 124 }
 125
 126 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 127                 struct buffer_head *bh, handle_t *handle, int err)
 128 {
 129         char nbuf[16];
 130         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 131
 132         if (bh)
 133                 BUFFER_TRACE(bh, "abort");
 134
 135         if (!handle->h_err)
 136                 handle->h_err = err;
 137
 138         if (is_handle_aborted(handle))
 139                 return;
 140
 141         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 142                 caller, errstr, err_fn);
 143
 144         journal_abort_handle(handle);
 145 }
 146
 147 void ext3_msg(struct super_block *sb, const char *prefix,
 148                 const char *fmt, ...)
 149 {
 150         struct va_format vaf;
 151         va_list args;
 152
 153         va_start(args, fmt);
 154
 155         vaf.fmt = fmt;
 156         vaf.va = &args;
 157
 158         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 159
 160         va_end(args);
 161 }
 162
 163 /* Deal with the reporting of failure conditions on a filesystem such as
 164  * inconsistencies detected or read IO failures.
 165  *
 166  * On ext2, we can store the error state of the filesystem in the
 167  * superblock.  That is not possible on ext3, because we may have other
 168  * write ordering constraints on the superblock which prevent us from
 169  * writing it out straight away; and given that the journal is about to
 170  * be aborted, we can't rely on the current, or future, transactions to
 171  * write out the superblock safely.
 172  *
 173  * We'll just use the journal_abort() error code to record an error in
 174  * the journal instead.  On recovery, the journal will complain about
 175  * that error until we've noted it down and cleared it.
 176  */
 177
 178 static void ext3_handle_error(struct super_block *sb)
 179 {
 180         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 181
 182         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 183         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 184
 185         if (sb->s_flags & MS_RDONLY)
 186                 return;
 187
 188         if (!test_opt (sb, ERRORS_CONT)) {
 189                 journal_t *journal = EXT3_SB(sb)->s_journal;
 190
 191                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 192                 if (journal)
 193                         journal_abort(journal, -EIO);
 194         }
 195         if (test_opt (sb, ERRORS_RO)) {
 196                 ext3_msg(sb, KERN_CRIT,
 197                         "error: remounting filesystem read-only");
 198                 sb->s_flags |= MS_RDONLY;
 199         }
 200         ext3_commit_super(sb, es, 1);
 201         if (test_opt(sb, ERRORS_PANIC))
 202                 panic("EXT3-fs (%s): panic forced after error\n",
 203                         sb->s_id);
 204 }
 205
 206 void ext3_error(struct super_block *sb, const char *function,
 207                 const char *fmt, ...)
 208 {
 209         struct va_format vaf;
 210         va_list args;
 211
 212         va_start(args, fmt);
 213
 214         vaf.fmt = fmt;
 215         vaf.va = &args;
 216
 217         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 218                sb->s_id, function, &vaf);
 219
 220         va_end(args);
 221
 222         ext3_handle_error(sb);
 223 }
 224
 225 static const char *ext3_decode_error(struct super_block * sb, int errno,
 226                                      char nbuf[16])
 227 {
 228         char *errstr = NULL;
 229
 230         switch (errno) {
 231         case -EIO:
 232                 errstr = "IO failure";
 233                 break;
 234         case -ENOMEM:
 235                 errstr = "Out of memory";
 236                 break;
 237         case -EROFS:
 238                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 239                         errstr = "Journal has aborted";
 240                 else
 241                         errstr = "Readonly filesystem";
 242                 break;
 243         default:
 244                 /* If the caller passed in an extra buffer for unknown
 245                  * errors, textualise them now.  Else we just return
 246                  * NULL. */
 247                 if (nbuf) {
 248                         /* Check for truncated error codes... */
 249                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 250                                 errstr = nbuf;
 251                 }
 252                 break;
 253         }
 254
 255         return errstr;
 256 }
 257
 258 /* __ext3_std_error decodes expected errors from journaling functions
 259  * automatically and invokes the appropriate error response.  */
 260
 261 void __ext3_std_error (struct super_block * sb, const char * function,
 262                        int errno)
 263 {
 264         char nbuf[16];
 265         const char *errstr;
 266
 267         /* Special case: if the error is EROFS, and we're not already
 268          * inside a transaction, then there's really no point in logging
 269          * an error. */
 270         if (errno == -EROFS && journal_current_handle() == NULL &&
 271             (sb->s_flags & MS_RDONLY))
 272                 return;
 273
 274         errstr = ext3_decode_error(sb, errno, nbuf);
 275         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 276
 277         ext3_handle_error(sb);
 278 }
 279
 280 /*
 281  * ext3_abort is a much stronger failure handler than ext3_error.  The
 282  * abort function may be used to deal with unrecoverable failures such
 283  * as journal IO errors or ENOMEM at a critical moment in log management.
 284  *
 285  * We unconditionally force the filesystem into an ABORT|READONLY state,
 286  * unless the error response on the fs has been set to panic in which
 287  * case we take the easy way out and panic immediately.
 288  */
 289
 290 void ext3_abort(struct super_block *sb, const char *function,
 291                  const char *fmt, ...)
 292 {
 293         struct va_format vaf;
 294         va_list args;
 295
 296         va_start(args, fmt);
 297
 298         vaf.fmt = fmt;
 299         vaf.va = &args;
 300
 301         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 302                sb->s_id, function, &vaf);
 303
 304         va_end(args);
 305
 306         if (test_opt(sb, ERRORS_PANIC))
 307                 panic("EXT3-fs: panic from previous error\n");
 308
 309         if (sb->s_flags & MS_RDONLY)
 310                 return;
 311
 312         ext3_msg(sb, KERN_CRIT,
 313                 "error: remounting filesystem read-only");
 314         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 315         sb->s_flags |= MS_RDONLY;
 316         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 317         if (EXT3_SB(sb)->s_journal)
 318                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 319 }
 320
 321 void ext3_warning(struct super_block *sb, const char *function,
 322                   const char *fmt, ...)
 323 {
 324         struct va_format vaf;
 325         va_list args;
 326
 327         va_start(args, fmt);
 328
 329         vaf.fmt = fmt;
 330         vaf.va = &args;
 331
 332         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 333                sb->s_id, function, &vaf);
 334
 335         va_end(args);
 336 }
 337
 338 void ext3_update_dynamic_rev(struct super_block *sb)
 339 {
 340         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 341
 342         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 343                 return;
 344
 345         ext3_msg(sb, KERN_WARNING,
 346                 "warning: updating to rev %d because of "
 347                 "new feature flag, running e2fsck is recommended",
 348                 EXT3_DYNAMIC_REV);
 349
 350         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 351         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 352         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 353         /* leave es->s_feature_*compat flags alone */
 354         /* es->s_uuid will be set by e2fsck if empty */
 355
 356         /*
 357          * The rest of the superblock fields should be zero, and if not it
 358          * means they are likely already in use, so leave them alone.  We
 359          * can leave it up to e2fsck to clean up any inconsistencies there.
 360          */
 361 }
 362
 363 /*
 364  * Open the external journal device
 365  */
 366 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 367 {
 368         struct block_device *bdev;
 369         char b[BDEVNAME_SIZE];
 370
 371         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 372         if (IS_ERR(bdev))
 373                 goto fail;
 374         return bdev;
 375
 376 fail:
 377         ext3_msg(sb, KERN_ERR, "error: failed to open journal device %s: %ld",
 378                 __bdevname(dev, b), PTR_ERR(bdev));
 379
 380         return NULL;
 381 }
 382
 383 /*
 384  * Release the journal device
 385  */
 386 static int ext3_blkdev_put(struct block_device *bdev)
 387 {
 388         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 389 }
 390
 391 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 392 {
 393         struct block_device *bdev;
 394         int ret = -ENODEV;
 395
 396         bdev = sbi->journal_bdev;
 397         if (bdev) {
 398                 ret = ext3_blkdev_put(bdev);
 399                 sbi->journal_bdev = NULL;
 400         }
 401         return ret;
 402 }
 403
 404 static inline struct inode *orphan_list_entry(struct list_head *l)
 405 {
 406         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 407 }
 408
 409 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 410 {
 411         struct list_head *l;
 412
 413         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 414                le32_to_cpu(sbi->s_es->s_last_orphan));
 415
 416         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 417         list_for_each(l, &sbi->s_orphan) {
 418                 struct inode *inode = orphan_list_entry(l);
 419                 ext3_msg(sb, KERN_ERR, "  "
 420                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 421                        inode->i_sb->s_id, inode->i_ino, inode,
 422                        inode->i_mode, inode->i_nlink,
 423                        NEXT_ORPHAN(inode));
 424         }
 425 }
 426
 427 static void ext3_put_super (struct super_block * sb)
 428 {
 429         struct ext3_sb_info *sbi = EXT3_SB(sb);
 430         struct ext3_super_block *es = sbi->s_es;
 431         int i, err;
 432
 433         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 434         ext3_xattr_put_super(sb);
 435         err = journal_destroy(sbi->s_journal);
 436         sbi->s_journal = NULL;
 437         if (err < 0)
 438                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 439
 440         if (!(sb->s_flags & MS_RDONLY)) {
 441                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 442                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 443                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 444                 mark_buffer_dirty(sbi->s_sbh);
 445                 ext3_commit_super(sb, es, 1);
 446         }
 447
 448         for (i = 0; i < sbi->s_gdb_count; i++)
 449                 brelse(sbi->s_group_desc[i]);
 450         kfree(sbi->s_group_desc);
 451         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 452         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 453         percpu_counter_destroy(&sbi->s_dirs_counter);
 454         brelse(sbi->s_sbh);
 455 #ifdef CONFIG_QUOTA
 456         for (i = 0; i < MAXQUOTAS; i++)
 457                 kfree(sbi->s_qf_names[i]);
 458 #endif
 459
 460         /* Debugging code just in case the in-memory inode orphan list
 461          * isn't empty.  The on-disk one can be non-empty if we've
 462          * detected an error and taken the fs readonly, but the
 463          * in-memory list had better be clean by this point. */
 464         if (!list_empty(&sbi->s_orphan))
 465                 dump_orphan_list(sb, sbi);
 466         J_ASSERT(list_empty(&sbi->s_orphan));
 467
 468         invalidate_bdev(sb->s_bdev);
 469         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 470                 /*
 471                  * Invalidate the journal device's buffers.  We don't want them
 472                  * floating about in memory - the physical journal device may
 473                  * hotswapped, and it breaks the `ro-after' testing code.
 474                  */
 475                 sync_blockdev(sbi->journal_bdev);
 476                 invalidate_bdev(sbi->journal_bdev);
 477                 ext3_blkdev_remove(sbi);
 478         }
 479         sb->s_fs_info = NULL;
 480         kfree(sbi->s_blockgroup_lock);
 481         kfree(sbi);
 482 }
 483
 484 static struct kmem_cache *ext3_inode_cachep;
 485
 486 /*
 487  * Called inside transaction, so use GFP_NOFS
 488  */
 489 static struct inode *ext3_alloc_inode(struct super_block *sb)
 490 {
 491         struct ext3_inode_info *ei;
 492
 493         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 494         if (!ei)
 495                 return NULL;
 496         ei->i_block_alloc_info = NULL;
 497         ei->vfs_inode.i_version = 1;
 498         atomic_set(&ei->i_datasync_tid, 0);
 499         atomic_set(&ei->i_sync_tid, 0);
 500         return &ei->vfs_inode;
 501 }
 502
 503 static int ext3_drop_inode(struct inode *inode)
 504 {
 505         int drop = generic_drop_inode(inode);
 506
 507         trace_ext3_drop_inode(inode, drop);
 508         return drop;
 509 }
 510
 511 static void ext3_i_callback(struct rcu_head *head)
 512 {
 513         struct inode *inode = container_of(head, struct inode, i_rcu);
 514         INIT_LIST_HEAD(&inode->i_dentry);
 515         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 516 }
 517
 518 static void ext3_destroy_inode(struct inode *inode)
 519 {
 520         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 521                 printk("EXT3 Inode %p: orphan list check failed!\n",
 522                         EXT3_I(inode));
 523                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 524                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 525                                 false);
 526                 dump_stack();
 527         }
 528         call_rcu(&inode->i_rcu, ext3_i_callback);
 529 }
 530
 531 static void init_once(void *foo)
 532 {
 533         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 534
 535         INIT_LIST_HEAD(&ei->i_orphan);
 536 #ifdef CONFIG_EXT3_FS_XATTR
 537         init_rwsem(&ei->xattr_sem);
 538 #endif
 539         mutex_init(&ei->truncate_mutex);
 540         inode_init_once(&ei->vfs_inode);
 541 }
 542
 543 static int init_inodecache(void)
 544 {
 545         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 546                                              sizeof(struct ext3_inode_info),
 547                                              0, (SLAB_RECLAIM_ACCOUNT|
 548                                                 SLAB_MEM_SPREAD),
 549                                              init_once);
 550         if (ext3_inode_cachep == NULL)
 551                 return -ENOMEM;
 552         return 0;
 553 }
 554
 555 static void destroy_inodecache(void)
 556 {
 557         kmem_cache_destroy(ext3_inode_cachep);
 558 }
 559
 560 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 561 {
 562 #if defined(CONFIG_QUOTA)
 563         struct ext3_sb_info *sbi = EXT3_SB(sb);
 564
 565         if (sbi->s_jquota_fmt) {
 566                 char *fmtname = "";
 567
 568                 switch (sbi->s_jquota_fmt) {
 569                 case QFMT_VFS_OLD:
 570                         fmtname = "vfsold";
 571                         break;
 572                 case QFMT_VFS_V0:
 573                         fmtname = "vfsv0";
 574                         break;
 575                 case QFMT_VFS_V1:
 576                         fmtname = "vfsv1";
 577                         break;
 578                 }
 579                 seq_printf(seq, ",jqfmt=%s", fmtname);
 580         }
 581
 582         if (sbi->s_qf_names[USRQUOTA])
 583                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 584
 585         if (sbi->s_qf_names[GRPQUOTA])
 586                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 587
 588         if (test_opt(sb, USRQUOTA))
 589                 seq_puts(seq, ",usrquota");
 590
 591         if (test_opt(sb, GRPQUOTA))
 592                 seq_puts(seq, ",grpquota");
 593 #endif
 594 }
 595
 596 static char *data_mode_string(unsigned long mode)
 597 {
 598         switch (mode) {
 599         case EXT3_MOUNT_JOURNAL_DATA:
 600                 return "journal";
 601         case EXT3_MOUNT_ORDERED_DATA:
 602                 return "ordered";
 603         case EXT3_MOUNT_WRITEBACK_DATA:
 604                 return "writeback";
 605         }
 606         return "unknown";
 607 }
 608
 609 /*
 610  * Show an option if
 611  *  - it's set to a non-default value OR
 612  *  - if the per-sb default is different from the global default
 613  */
 614 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 615 {
 616         struct super_block *sb = vfs->mnt_sb;
 617         struct ext3_sb_info *sbi = EXT3_SB(sb);
 618         struct ext3_super_block *es = sbi->s_es;
 619         unsigned long def_mount_opts;
 620
 621         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 622
 623         if (sbi->s_sb_block != 1)
 624                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 625         if (test_opt(sb, MINIX_DF))
 626                 seq_puts(seq, ",minixdf");
 627         if (test_opt(sb, GRPID))
 628                 seq_puts(seq, ",grpid");
 629         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 630                 seq_puts(seq, ",nogrpid");
 631         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 632             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 633                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 634         }
 635         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 636             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 637                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 638         }
 639         if (test_opt(sb, ERRORS_RO)) {
 640                 int def_errors = le16_to_cpu(es->s_errors);
 641
 642                 if (def_errors == EXT3_ERRORS_PANIC ||
 643                     def_errors == EXT3_ERRORS_CONTINUE) {
 644                         seq_puts(seq, ",errors=remount-ro");
 645                 }
 646         }
 647         if (test_opt(sb, ERRORS_CONT))
 648                 seq_puts(seq, ",errors=continue");
 649         if (test_opt(sb, ERRORS_PANIC))
 650                 seq_puts(seq, ",errors=panic");
 651         if (test_opt(sb, NO_UID32))
 652                 seq_puts(seq, ",nouid32");
 653         if (test_opt(sb, DEBUG))
 654                 seq_puts(seq, ",debug");
 655 #ifdef CONFIG_EXT3_FS_XATTR
 656         if (test_opt(sb, XATTR_USER))
 657                 seq_puts(seq, ",user_xattr");
 658         if (!test_opt(sb, XATTR_USER) &&
 659             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 660                 seq_puts(seq, ",nouser_xattr");
 661         }
 662 #endif
 663 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 664         if (test_opt(sb, POSIX_ACL))
 665                 seq_puts(seq, ",acl");
 666         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 667                 seq_puts(seq, ",noacl");
 668 #endif
 669         if (!test_opt(sb, RESERVATION))
 670                 seq_puts(seq, ",noreservation");
 671         if (sbi->s_commit_interval) {
 672                 seq_printf(seq, ",commit=%u",
 673                            (unsigned) (sbi->s_commit_interval / HZ));
 674         }
 675
 676         /*
 677          * Always display barrier state so it's clear what the status is.
 678          */
 679         seq_puts(seq, ",barrier=");
 680         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 681         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 682         if (test_opt(sb, DATA_ERR_ABORT))
 683                 seq_puts(seq, ",data_err=abort");
 684
 685         if (test_opt(sb, NOLOAD))
 686                 seq_puts(seq, ",norecovery");
 687
 688         ext3_show_quota_options(seq, sb);
 689
 690         return 0;
 691 }
 692
 693
 694 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 695                 u64 ino, u32 generation)
 696 {
 697         struct inode *inode;
 698
 699         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 700                 return ERR_PTR(-ESTALE);
 701         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 702                 return ERR_PTR(-ESTALE);
 703
 704         /* iget isn't really right if the inode is currently unallocated!!
 705          *
 706          * ext3_read_inode will return a bad_inode if the inode had been
 707          * deleted, so we should be safe.
 708          *
 709          * Currently we don't know the generation for parent directory, so
 710          * a generation of 0 means "accept any"
 711          */
 712         inode = ext3_iget(sb, ino);
 713         if (IS_ERR(inode))
 714                 return ERR_CAST(inode);
 715         if (generation && inode->i_generation != generation) {
 716                 iput(inode);
 717                 return ERR_PTR(-ESTALE);
 718         }
 719
 720         return inode;
 721 }
 722
 723 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 724                 int fh_len, int fh_type)
 725 {
 726         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 727                                     ext3_nfs_get_inode);
 728 }
 729
 730 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 731                 int fh_len, int fh_type)
 732 {
 733         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 734                                     ext3_nfs_get_inode);
 735 }
 736
 737 /*
 738  * Try to release metadata pages (indirect blocks, directories) which are
 739  * mapped via the block device.  Since these pages could have journal heads
 740  * which would prevent try_to_free_buffers() from freeing them, we must use
 741  * jbd layer's try_to_free_buffers() function to release them.
 742  */
 743 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 744                                  gfp_t wait)
 745 {
 746         journal_t *journal = EXT3_SB(sb)->s_journal;
 747
 748         WARN_ON(PageChecked(page));
 749         if (!page_has_buffers(page))
 750                 return 0;
 751         if (journal)
 752                 return journal_try_to_free_buffers(journal, page,
 753                                                    wait & ~__GFP_WAIT);
 754         return try_to_free_buffers(page);
 755 }
 756
 757 #ifdef CONFIG_QUOTA
 758 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 759 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 760
 761 static int ext3_write_dquot(struct dquot *dquot);
 762 static int ext3_acquire_dquot(struct dquot *dquot);
 763 static int ext3_release_dquot(struct dquot *dquot);
 764 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 765 static int ext3_write_info(struct super_block *sb, int type);
 766 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 767                          struct path *path);
 768 static int ext3_quota_on_mount(struct super_block *sb, int type);
 769 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 770                                size_t len, loff_t off);
 771 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 772                                 const char *data, size_t len, loff_t off);
 773
 774 static const struct dquot_operations ext3_quota_operations = {
 775         .write_dquot    = ext3_write_dquot,
 776         .acquire_dquot  = ext3_acquire_dquot,
 777         .release_dquot  = ext3_release_dquot,
 778         .mark_dirty     = ext3_mark_dquot_dirty,
 779         .write_info     = ext3_write_info,
 780         .alloc_dquot    = dquot_alloc,
 781         .destroy_dquot  = dquot_destroy,
 782 };
 783
 784 static const struct quotactl_ops ext3_qctl_operations = {
 785         .quota_on       = ext3_quota_on,
 786         .quota_off      = dquot_quota_off,
 787         .quota_sync     = dquot_quota_sync,
 788         .get_info       = dquot_get_dqinfo,
 789         .set_info       = dquot_set_dqinfo,
 790         .get_dqblk      = dquot_get_dqblk,
 791         .set_dqblk      = dquot_set_dqblk
 792 };
 793 #endif
 794
 795 static const struct super_operations ext3_sops = {
 796         .alloc_inode    = ext3_alloc_inode,
 797         .destroy_inode  = ext3_destroy_inode,
 798         .write_inode    = ext3_write_inode,
 799         .dirty_inode    = ext3_dirty_inode,
 800         .drop_inode     = ext3_drop_inode,
 801         .evict_inode    = ext3_evict_inode,
 802         .put_super      = ext3_put_super,
 803         .sync_fs        = ext3_sync_fs,
 804         .freeze_fs      = ext3_freeze,
 805         .unfreeze_fs    = ext3_unfreeze,
 806         .statfs         = ext3_statfs,
 807         .remount_fs     = ext3_remount,
 808         .show_options   = ext3_show_options,
 809 #ifdef CONFIG_QUOTA
 810         .quota_read     = ext3_quota_read,
 811         .quota_write    = ext3_quota_write,
 812 #endif
 813         .bdev_try_to_free_page = bdev_try_to_free_page,
 814 };
 815
 816 static const struct export_operations ext3_export_ops = {
 817         .fh_to_dentry = ext3_fh_to_dentry,
 818         .fh_to_parent = ext3_fh_to_parent,
 819         .get_parent = ext3_get_parent,
 820 };
 821
 822 enum {
 823         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 824         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 825         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 826         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 827         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 828         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 829         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 830         Opt_data_err_abort, Opt_data_err_ignore,
 831         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 832         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 833         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 834         Opt_resize, Opt_usrquota, Opt_grpquota
 835 };
 836
 837 static const match_table_t tokens = {
 838         {Opt_bsd_df, "bsddf"},
 839         {Opt_minix_df, "minixdf"},
 840         {Opt_grpid, "grpid"},
 841         {Opt_grpid, "bsdgroups"},
 842         {Opt_nogrpid, "nogrpid"},
 843         {Opt_nogrpid, "sysvgroups"},
 844         {Opt_resgid, "resgid=%u"},
 845         {Opt_resuid, "resuid=%u"},
 846         {Opt_sb, "sb=%u"},
 847         {Opt_err_cont, "errors=continue"},
 848         {Opt_err_panic, "errors=panic"},
 849         {Opt_err_ro, "errors=remount-ro"},
 850         {Opt_nouid32, "nouid32"},
 851         {Opt_nocheck, "nocheck"},
 852         {Opt_nocheck, "check=none"},
 853         {Opt_debug, "debug"},
 854         {Opt_oldalloc, "oldalloc"},
 855         {Opt_orlov, "orlov"},
 856         {Opt_user_xattr, "user_xattr"},
 857         {Opt_nouser_xattr, "nouser_xattr"},
 858         {Opt_acl, "acl"},
 859         {Opt_noacl, "noacl"},
 860         {Opt_reservation, "reservation"},
 861         {Opt_noreservation, "noreservation"},
 862         {Opt_noload, "noload"},
 863         {Opt_noload, "norecovery"},
 864         {Opt_nobh, "nobh"},
 865         {Opt_bh, "bh"},
 866         {Opt_commit, "commit=%u"},
 867         {Opt_journal_update, "journal=update"},
 868         {Opt_journal_inum, "journal=%u"},
 869         {Opt_journal_dev, "journal_dev=%u"},
 870         {Opt_abort, "abort"},
 871         {Opt_data_journal, "data=journal"},
 872         {Opt_data_ordered, "data=ordered"},
 873         {Opt_data_writeback, "data=writeback"},
 874         {Opt_data_err_abort, "data_err=abort"},
 875         {Opt_data_err_ignore, "data_err=ignore"},
 876         {Opt_offusrjquota, "usrjquota="},
 877         {Opt_usrjquota, "usrjquota=%s"},
 878         {Opt_offgrpjquota, "grpjquota="},
 879         {Opt_grpjquota, "grpjquota=%s"},
 880         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 881         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 882         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 883         {Opt_grpquota, "grpquota"},
 884         {Opt_noquota, "noquota"},
 885         {Opt_quota, "quota"},
 886         {Opt_usrquota, "usrquota"},
 887         {Opt_barrier, "barrier=%u"},
 888         {Opt_barrier, "barrier"},
 889         {Opt_nobarrier, "nobarrier"},
 890         {Opt_resize, "resize"},
 891         {Opt_err, NULL},
 892 };
 893
 894 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 895 {
 896         ext3_fsblk_t    sb_block;
 897         char            *options = (char *) *data;
 898
 899         if (!options || strncmp(options, "sb=", 3) != 0)
 900                 return 1;       /* Default location */
 901         options += 3;
 902         /*todo: use simple_strtoll with >32bit ext3 */
 903         sb_block = simple_strtoul(options, &options, 0);
 904         if (*options && *options != ',') {
 905                 ext3_msg(sb, KERN_ERR, "error: invalid sb specification: %s",
 906                        (char *) *data);
 907                 return 1;
 908         }
 909         if (*options == ',')
 910                 options++;
 911         *data = (void *) options;
 912         return sb_block;
 913 }
 914
 915 #ifdef CONFIG_QUOTA
 916 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 917 {
 918         struct ext3_sb_info *sbi = EXT3_SB(sb);
 919         char *qname;
 920
 921         if (sb_any_quota_loaded(sb) &&
 922                 !sbi->s_qf_names[qtype]) {
 923                 ext3_msg(sb, KERN_ERR,
 924                         "Cannot change journaled "
 925                         "quota options when quota turned on");
 926                 return 0;
 927         }
 928         qname = match_strdup(args);
 929         if (!qname) {
 930                 ext3_msg(sb, KERN_ERR,
 931                         "Not enough memory for storing quotafile name");
 932                 return 0;
 933         }
 934         if (sbi->s_qf_names[qtype] &&
 935                 strcmp(sbi->s_qf_names[qtype], qname)) {
 936                 ext3_msg(sb, KERN_ERR,
 937                         "%s quota file already specified", QTYPE2NAME(qtype));
 938                 kfree(qname);
 939                 return 0;
 940         }
 941         sbi->s_qf_names[qtype] = qname;
 942         if (strchr(sbi->s_qf_names[qtype], '/')) {
 943                 ext3_msg(sb, KERN_ERR,
 944                         "quotafile must be on filesystem root");
 945                 kfree(sbi->s_qf_names[qtype]);
 946                 sbi->s_qf_names[qtype] = NULL;
 947                 return 0;
 948         }
 949         set_opt(sbi->s_mount_opt, QUOTA);
 950         return 1;
 951 }
 952
 953 static int clear_qf_name(struct super_block *sb, int qtype) {
 954
 955         struct ext3_sb_info *sbi = EXT3_SB(sb);
 956
 957         if (sb_any_quota_loaded(sb) &&
 958                 sbi->s_qf_names[qtype]) {
 959                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 960                         " when quota turned on");
 961                 return 0;
 962         }
 963         /*
 964          * The space will be released later when all options are confirmed
 965          * to be correct
 966          */
 967         sbi->s_qf_names[qtype] = NULL;
 968         return 1;
 969 }
 970 #endif
 971
 972 static int parse_options (char *options, struct super_block *sb,
 973                           unsigned int *inum, unsigned long *journal_devnum,
 974                           ext3_fsblk_t *n_blocks_count, int is_remount)
 975 {
 976         struct ext3_sb_info *sbi = EXT3_SB(sb);
 977         char * p;
 978         substring_t args[MAX_OPT_ARGS];
 979         int data_opt = 0;
 980         int option;
 981 #ifdef CONFIG_QUOTA
 982         int qfmt;
 983 #endif
 984
 985         if (!options)
 986                 return 1;
 987
 988         while ((p = strsep (&options, ",")) != NULL) {
 989                 int token;
 990                 if (!*p)
 991                         continue;
 992                 /*
 993                  * Initialize args struct so we know whether arg was
 994                  * found; some options take optional arguments.
 995                  */
 996                 args[0].to = args[0].from = 0;
 997                 token = match_token(p, tokens, args);
 998                 switch (token) {
 999                 case Opt_bsd_df:
1000                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1001                         break;
1002                 case Opt_minix_df:
1003                         set_opt (sbi->s_mount_opt, MINIX_DF);
1004                         break;
1005                 case Opt_grpid:
1006                         set_opt (sbi->s_mount_opt, GRPID);
1007                         break;
1008                 case Opt_nogrpid:
1009                         clear_opt (sbi->s_mount_opt, GRPID);
1010                         break;
1011                 case Opt_resuid:
1012                         if (match_int(&args[0], &option))
1013                                 return 0;
1014                         sbi->s_resuid = option;
1015                         break;
1016                 case Opt_resgid:
1017                         if (match_int(&args[0], &option))
1018                                 return 0;
1019                         sbi->s_resgid = option;
1020                         break;
1021                 case Opt_sb:
1022                         /* handled by get_sb_block() instead of here */
1023                         /* *sb_block = match_int(&args[0]); */
1024                         break;
1025                 case Opt_err_panic:
1026                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1027                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1028                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1029                         break;
1030                 case Opt_err_ro:
1031                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1032                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1033                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1034                         break;
1035                 case Opt_err_cont:
1036                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1037                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1038                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1039                         break;
1040                 case Opt_nouid32:
1041                         set_opt (sbi->s_mount_opt, NO_UID32);
1042                         break;
1043                 case Opt_nocheck:
1044                         clear_opt (sbi->s_mount_opt, CHECK);
1045                         break;
1046                 case Opt_debug:
1047                         set_opt (sbi->s_mount_opt, DEBUG);
1048                         break;
1049                 case Opt_oldalloc:
1050                         ext3_msg(sb, KERN_WARNING,
1051                                 "Ignoring deprecated oldalloc option");
1052                         break;
1053                 case Opt_orlov:
1054                         ext3_msg(sb, KERN_WARNING,
1055                                 "Ignoring deprecated orlov option");
1056                         break;
1057 #ifdef CONFIG_EXT3_FS_XATTR
1058                 case Opt_user_xattr:
1059                         set_opt (sbi->s_mount_opt, XATTR_USER);
1060                         break;
1061                 case Opt_nouser_xattr:
1062                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1063                         break;
1064 #else
1065                 case Opt_user_xattr:
1066                 case Opt_nouser_xattr:
1067                         ext3_msg(sb, KERN_INFO,
1068                                 "(no)user_xattr options not supported");
1069                         break;
1070 #endif
1071 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1072                 case Opt_acl:
1073                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1074                         break;
1075                 case Opt_noacl:
1076                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1077                         break;
1078 #else
1079                 case Opt_acl:
1080                 case Opt_noacl:
1081                         ext3_msg(sb, KERN_INFO,
1082                                 "(no)acl options not supported");
1083                         break;
1084 #endif
1085                 case Opt_reservation:
1086                         set_opt(sbi->s_mount_opt, RESERVATION);
1087                         break;
1088                 case Opt_noreservation:
1089                         clear_opt(sbi->s_mount_opt, RESERVATION);
1090                         break;
1091                 case Opt_journal_update:
1092                         /* @@@ FIXME */
1093                         /* Eventually we will want to be able to create
1094                            a journal file here.  For now, only allow the
1095                            user to specify an existing inode to be the
1096                            journal file. */
1097                         if (is_remount) {
1098                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1099                                         "journal on remount");
1100                                 return 0;
1101                         }
1102                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1103                         break;
1104                 case Opt_journal_inum:
1105                         if (is_remount) {
1106                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1107                                        "journal on remount");
1108                                 return 0;
1109                         }
1110                         if (match_int(&args[0], &option))
1111                                 return 0;
1112                         *inum = option;
1113                         break;
1114                 case Opt_journal_dev:
1115                         if (is_remount) {
1116                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1117                                        "journal on remount");
1118                                 return 0;
1119                         }
1120                         if (match_int(&args[0], &option))
1121                                 return 0;
1122                         *journal_devnum = option;
1123                         break;
1124                 case Opt_noload:
1125                         set_opt (sbi->s_mount_opt, NOLOAD);
1126                         break;
1127                 case Opt_commit:
1128                         if (match_int(&args[0], &option))
1129                                 return 0;
1130                         if (option < 0)
1131                                 return 0;
1132                         if (option == 0)
1133                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1134                         sbi->s_commit_interval = HZ * option;
1135                         break;
1136                 case Opt_data_journal:
1137                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1138                         goto datacheck;
1139                 case Opt_data_ordered:
1140                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1141                         goto datacheck;
1142                 case Opt_data_writeback:
1143                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1144                 datacheck:
1145                         if (is_remount) {
1146                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1147                                         break;
1148                                 ext3_msg(sb, KERN_ERR,
1149                                         "error: cannot change "
1150                                         "data mode on remount. The filesystem "
1151                                         "is mounted in data=%s mode and you "
1152                                         "try to remount it in data=%s mode.",
1153                                         data_mode_string(test_opt(sb,
1154                                                         DATA_FLAGS)),
1155                                         data_mode_string(data_opt));
1156                                 return 0;
1157                         } else {
1158                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1159                                 sbi->s_mount_opt |= data_opt;
1160                         }
1161                         break;
1162                 case Opt_data_err_abort:
1163                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1164                         break;
1165                 case Opt_data_err_ignore:
1166                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1167                         break;
1168 #ifdef CONFIG_QUOTA
1169                 case Opt_usrjquota:
1170                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1171                                 return 0;
1172                         break;
1173                 case Opt_grpjquota:
1174                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1175                                 return 0;
1176                         break;
1177                 case Opt_offusrjquota:
1178                         if (!clear_qf_name(sb, USRQUOTA))
1179                                 return 0;
1180                         break;
1181                 case Opt_offgrpjquota:
1182                         if (!clear_qf_name(sb, GRPQUOTA))
1183                                 return 0;
1184                         break;
1185                 case Opt_jqfmt_vfsold:
1186                         qfmt = QFMT_VFS_OLD;
1187                         goto set_qf_format;
1188                 case Opt_jqfmt_vfsv0:
1189                         qfmt = QFMT_VFS_V0;
1190                         goto set_qf_format;
1191                 case Opt_jqfmt_vfsv1:
1192                         qfmt = QFMT_VFS_V1;
1193 set_qf_format:
1194                         if (sb_any_quota_loaded(sb) &&
1195                             sbi->s_jquota_fmt != qfmt) {
1196                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1197                                         "journaled quota options when "
1198                                         "quota turned on.");
1199                                 return 0;
1200                         }
1201                         sbi->s_jquota_fmt = qfmt;
1202                         break;
1203                 case Opt_quota:
1204                 case Opt_usrquota:
1205                         set_opt(sbi->s_mount_opt, QUOTA);
1206                         set_opt(sbi->s_mount_opt, USRQUOTA);
1207                         break;
1208                 case Opt_grpquota:
1209                         set_opt(sbi->s_mount_opt, QUOTA);
1210                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1211                         break;
1212                 case Opt_noquota:
1213                         if (sb_any_quota_loaded(sb)) {
1214                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1215                                         "quota options when quota turned on.");
1216                                 return 0;
1217                         }
1218                         clear_opt(sbi->s_mount_opt, QUOTA);
1219                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1220                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1221                         break;
1222 #else
1223                 case Opt_quota:
1224                 case Opt_usrquota:
1225                 case Opt_grpquota:
1226                         ext3_msg(sb, KERN_ERR,
1227                                 "error: quota options not supported.");
1228                         break;
1229                 case Opt_usrjquota:
1230                 case Opt_grpjquota:
1231                 case Opt_offusrjquota:
1232                 case Opt_offgrpjquota:
1233                 case Opt_jqfmt_vfsold:
1234                 case Opt_jqfmt_vfsv0:
1235                 case Opt_jqfmt_vfsv1:
1236                         ext3_msg(sb, KERN_ERR,
1237                                 "error: journaled quota options not "
1238                                 "supported.");
1239                         break;
1240                 case Opt_noquota:
1241                         break;
1242 #endif
1243                 case Opt_abort:
1244                         set_opt(sbi->s_mount_opt, ABORT);
1245                         break;
1246                 case Opt_nobarrier:
1247                         clear_opt(sbi->s_mount_opt, BARRIER);
1248                         break;
1249                 case Opt_barrier:
1250                         if (args[0].from) {
1251                                 if (match_int(&args[0], &option))
1252                                         return 0;
1253                         } else
1254                                 option = 1;     /* No argument, default to 1 */
1255                         if (option)
1256                                 set_opt(sbi->s_mount_opt, BARRIER);
1257                         else
1258                                 clear_opt(sbi->s_mount_opt, BARRIER);
1259                         break;
1260                 case Opt_ignore:
1261                         break;
1262                 case Opt_resize:
1263                         if (!is_remount) {
1264                                 ext3_msg(sb, KERN_ERR,
1265                                         "error: resize option only available "
1266                                         "for remount");
1267                                 return 0;
1268                         }
1269                         if (match_int(&args[0], &option) != 0)
1270                                 return 0;
1271                         *n_blocks_count = option;
1272                         break;
1273                 case Opt_nobh:
1274                         ext3_msg(sb, KERN_WARNING,
1275                                 "warning: ignoring deprecated nobh option");
1276                         break;
1277                 case Opt_bh:
1278                         ext3_msg(sb, KERN_WARNING,
1279                                 "warning: ignoring deprecated bh option");
1280                         break;
1281                 default:
1282                         ext3_msg(sb, KERN_ERR,
1283                                 "error: unrecognized mount option \"%s\" "
1284                                 "or missing value", p);
1285                         return 0;
1286                 }
1287         }
1288 #ifdef CONFIG_QUOTA
1289         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1290                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1291                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1292                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1293                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1294
1295                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1296                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1297                                         "format mixing.");
1298                         return 0;
1299                 }
1300
1301                 if (!sbi->s_jquota_fmt) {
1302                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1303                                         "not specified.");
1304                         return 0;
1305                 }
1306         }
1307 #endif
1308         return 1;
1309 }
1310
1311 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1312                             int read_only)
1313 {
1314         struct ext3_sb_info *sbi = EXT3_SB(sb);
1315         int res = 0;
1316
1317         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1318                 ext3_msg(sb, KERN_ERR,
1319                         "error: revision level too high, "
1320                         "forcing read-only mode");
1321                 res = MS_RDONLY;
1322         }
1323         if (read_only)
1324                 return res;
1325         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1326                 ext3_msg(sb, KERN_WARNING,
1327                         "warning: mounting unchecked fs, "
1328                         "running e2fsck is recommended");
1329         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1330                 ext3_msg(sb, KERN_WARNING,
1331                         "warning: mounting fs with errors, "
1332                         "running e2fsck is recommended");
1333         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1334                  le16_to_cpu(es->s_mnt_count) >=
1335                         le16_to_cpu(es->s_max_mnt_count))
1336                 ext3_msg(sb, KERN_WARNING,
1337                         "warning: maximal mount count reached, "
1338                         "running e2fsck is recommended");
1339         else if (le32_to_cpu(es->s_checkinterval) &&
1340                 (le32_to_cpu(es->s_lastcheck) +
1341                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1342                 ext3_msg(sb, KERN_WARNING,
1343                         "warning: checktime reached, "
1344                         "running e2fsck is recommended");
1345 #if 0
1346                 /* @@@ We _will_ want to clear the valid bit if we find
1347                    inconsistencies, to force a fsck at reboot.  But for
1348                    a plain journaled filesystem we can keep it set as
1349                    valid forever! :) */
1350         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1351 #endif
1352         if (!le16_to_cpu(es->s_max_mnt_count))
1353                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1354         le16_add_cpu(&es->s_mnt_count, 1);
1355         es->s_mtime = cpu_to_le32(get_seconds());
1356         ext3_update_dynamic_rev(sb);
1357         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1358
1359         ext3_commit_super(sb, es, 1);
1360         if (test_opt(sb, DEBUG))
1361                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1362                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1363                         sb->s_blocksize,
1364                         sbi->s_groups_count,
1365                         EXT3_BLOCKS_PER_GROUP(sb),
1366                         EXT3_INODES_PER_GROUP(sb),
1367                         sbi->s_mount_opt);
1368
1369         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1370                 char b[BDEVNAME_SIZE];
1371                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1372                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1373         } else {
1374                 ext3_msg(sb, KERN_INFO, "using internal journal");
1375         }
1376         cleancache_init_fs(sb);
1377         return res;
1378 }
1379
1380 /* Called at mount-time, super-block is locked */
1381 static int ext3_check_descriptors(struct super_block *sb)
1382 {
1383         struct ext3_sb_info *sbi = EXT3_SB(sb);
1384         int i;
1385
1386         ext3_debug ("Checking group descriptors");
1387
1388         for (i = 0; i < sbi->s_groups_count; i++) {
1389                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1390                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1391                 ext3_fsblk_t last_block;
1392
1393                 if (i == sbi->s_groups_count - 1)
1394                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1395                 else
1396                         last_block = first_block +
1397                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1398
1399                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1400                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1401                 {
1402                         ext3_error (sb, "ext3_check_descriptors",
1403                                     "Block bitmap for group %d"
1404                                     " not in group (block %lu)!",
1405                                     i, (unsigned long)
1406                                         le32_to_cpu(gdp->bg_block_bitmap));
1407                         return 0;
1408                 }
1409                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1410                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1411                 {
1412                         ext3_error (sb, "ext3_check_descriptors",
1413                                     "Inode bitmap for group %d"
1414                                     " not in group (block %lu)!",
1415                                     i, (unsigned long)
1416                                         le32_to_cpu(gdp->bg_inode_bitmap));
1417                         return 0;
1418                 }
1419                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1420                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1421                     last_block)
1422                 {
1423                         ext3_error (sb, "ext3_check_descriptors",
1424                                     "Inode table for group %d"
1425                                     " not in group (block %lu)!",
1426                                     i, (unsigned long)
1427                                         le32_to_cpu(gdp->bg_inode_table));
1428                         return 0;
1429                 }
1430         }
1431
1432         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1433         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1434         return 1;
1435 }
1436
1437
1438 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1439  * the superblock) which were deleted from all directories, but held open by
1440  * a process at the time of a crash.  We walk the list and try to delete these
1441  * inodes at recovery time (only with a read-write filesystem).
1442  *
1443  * In order to keep the orphan inode chain consistent during traversal (in
1444  * case of crash during recovery), we link each inode into the superblock
1445  * orphan list_head and handle it the same way as an inode deletion during
1446  * normal operation (which journals the operations for us).
1447  *
1448  * We only do an iget() and an iput() on each inode, which is very safe if we
1449  * accidentally point at an in-use or already deleted inode.  The worst that
1450  * can happen in this case is that we get a "bit already cleared" message from
1451  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1452  * e2fsck was run on this filesystem, and it must have already done the orphan
1453  * inode cleanup for us, so we can safely abort without any further action.
1454  */
1455 static void ext3_orphan_cleanup (struct super_block * sb,
1456                                  struct ext3_super_block * es)
1457 {
1458         unsigned int s_flags = sb->s_flags;
1459         int nr_orphans = 0, nr_truncates = 0;
1460 #ifdef CONFIG_QUOTA
1461         int i;
1462 #endif
1463         if (!es->s_last_orphan) {
1464                 jbd_debug(4, "no orphan inodes to clean up\n");
1465                 return;
1466         }
1467
1468         if (bdev_read_only(sb->s_bdev)) {
1469                 ext3_msg(sb, KERN_ERR, "error: write access "
1470                         "unavailable, skipping orphan cleanup.");
1471                 return;
1472         }
1473
1474         /* Check if feature set allows readwrite operations */
1475         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1476                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1477                          "unknown ROCOMPAT features");
1478                 return;
1479         }
1480
1481         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1482                 if (es->s_last_orphan)
1483                         jbd_debug(1, "Errors on filesystem, "
1484                                   "clearing orphan list.\n");
1485                 es->s_last_orphan = 0;
1486                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1487                 return;
1488         }
1489
1490         if (s_flags & MS_RDONLY) {
1491                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1492                 sb->s_flags &= ~MS_RDONLY;
1493         }
1494 #ifdef CONFIG_QUOTA
1495         /* Needed for iput() to work correctly and not trash data */
1496         sb->s_flags |= MS_ACTIVE;
1497         /* Turn on quotas so that they are updated correctly */
1498         for (i = 0; i < MAXQUOTAS; i++) {
1499                 if (EXT3_SB(sb)->s_qf_names[i]) {
1500                         int ret = ext3_quota_on_mount(sb, i);
1501                         if (ret < 0)
1502                                 ext3_msg(sb, KERN_ERR,
1503                                         "error: cannot turn on journaled "
1504                                         "quota: %d", ret);
1505                 }
1506         }
1507 #endif
1508
1509         while (es->s_last_orphan) {
1510                 struct inode *inode;
1511
1512                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1513                 if (IS_ERR(inode)) {
1514                         es->s_last_orphan = 0;
1515                         break;
1516                 }
1517
1518                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1519                 dquot_initialize(inode);
1520                 if (inode->i_nlink) {
1521                         printk(KERN_DEBUG
1522                                 "%s: truncating inode %lu to %Ld bytes\n",
1523                                 __func__, inode->i_ino, inode->i_size);
1524                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1525                                   inode->i_ino, inode->i_size);
1526                         ext3_truncate(inode);
1527                         nr_truncates++;
1528                 } else {
1529                         printk(KERN_DEBUG
1530                                 "%s: deleting unreferenced inode %lu\n",
1531                                 __func__, inode->i_ino);
1532                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1533                                   inode->i_ino);
1534                         nr_orphans++;
1535                 }
1536                 iput(inode);  /* The delete magic happens here! */
1537         }
1538
1539 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1540
1541         if (nr_orphans)
1542                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1543                        PLURAL(nr_orphans));
1544         if (nr_truncates)
1545                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1546                        PLURAL(nr_truncates));
1547 #ifdef CONFIG_QUOTA
1548         /* Turn quotas off */
1549         for (i = 0; i < MAXQUOTAS; i++) {
1550                 if (sb_dqopt(sb)->files[i])
1551                         dquot_quota_off(sb, i);
1552         }
1553 #endif
1554         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1555 }
1556
1557 /*
1558  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1559  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1560  * We need to be 1 filesystem block less than the 2^32 sector limit.
1561  */
1562 static loff_t ext3_max_size(int bits)
1563 {
1564         loff_t res = EXT3_NDIR_BLOCKS;
1565         int meta_blocks;
1566         loff_t upper_limit;
1567
1568         /* This is calculated to be the largest file size for a
1569          * dense, file such that the total number of
1570          * sectors in the file, including data and all indirect blocks,
1571          * does not exceed 2^32 -1
1572          * __u32 i_blocks representing the total number of
1573          * 512 bytes blocks of the file
1574          */
1575         upper_limit = (1LL << 32) - 1;
1576
1577         /* total blocks in file system block size */
1578         upper_limit >>= (bits - 9);
1579
1580
1581         /* indirect blocks */
1582         meta_blocks = 1;
1583         /* double indirect blocks */
1584         meta_blocks += 1 + (1LL << (bits-2));
1585         /* tripple indirect blocks */
1586         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1587
1588         upper_limit -= meta_blocks;
1589         upper_limit <<= bits;
1590
1591         res += 1LL << (bits-2);
1592         res += 1LL << (2*(bits-2));
1593         res += 1LL << (3*(bits-2));
1594         res <<= bits;
1595         if (res > upper_limit)
1596                 res = upper_limit;
1597
1598         if (res > MAX_LFS_FILESIZE)
1599                 res = MAX_LFS_FILESIZE;
1600
1601         return res;
1602 }
1603
1604 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1605                                     ext3_fsblk_t logic_sb_block,
1606                                     int nr)
1607 {
1608         struct ext3_sb_info *sbi = EXT3_SB(sb);
1609         unsigned long bg, first_meta_bg;
1610         int has_super = 0;
1611
1612         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1613
1614         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1615             nr < first_meta_bg)
1616                 return (logic_sb_block + nr + 1);
1617         bg = sbi->s_desc_per_block * nr;
1618         if (ext3_bg_has_super(sb, bg))
1619                 has_super = 1;
1620         return (has_super + ext3_group_first_block_no(sb, bg));
1621 }
1622
1623
1624 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1625 {
1626         struct buffer_head * bh;
1627         struct ext3_super_block *es = NULL;
1628         struct ext3_sb_info *sbi;
1629         ext3_fsblk_t block;
1630         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1631         ext3_fsblk_t logic_sb_block;
1632         unsigned long offset = 0;
1633         unsigned int journal_inum = 0;
1634         unsigned long journal_devnum = 0;
1635         unsigned long def_mount_opts;
1636         struct inode *root;
1637         int blocksize;
1638         int hblock;
1639         int db_count;
1640         int i;
1641         int needs_recovery;
1642         int ret = -EINVAL;
1643         __le32 features;
1644         int err;
1645
1646         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1647         if (!sbi)
1648                 return -ENOMEM;
1649
1650         sbi->s_blockgroup_lock =
1651                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1652         if (!sbi->s_blockgroup_lock) {
1653                 kfree(sbi);
1654                 return -ENOMEM;
1655         }
1656         sb->s_fs_info = sbi;
1657         sbi->s_mount_opt = 0;
1658         sbi->s_resuid = EXT3_DEF_RESUID;
1659         sbi->s_resgid = EXT3_DEF_RESGID;
1660         sbi->s_sb_block = sb_block;
1661
1662         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1663         if (!blocksize) {
1664                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1665                 goto out_fail;
1666         }
1667
1668         /*
1669          * The ext3 superblock will not be buffer aligned for other than 1kB
1670          * block sizes.  We need to calculate the offset from buffer start.
1671          */
1672         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1673                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1674                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1675         } else {
1676                 logic_sb_block = sb_block;
1677         }
1678
1679         if (!(bh = sb_bread(sb, logic_sb_block))) {
1680                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1681                 goto out_fail;
1682         }
1683         /*
1684          * Note: s_es must be initialized as soon as possible because
1685          *       some ext3 macro-instructions depend on its value
1686          */
1687         es = (struct ext3_super_block *) (bh->b_data + offset);
1688         sbi->s_es = es;
1689         sb->s_magic = le16_to_cpu(es->s_magic);
1690         if (sb->s_magic != EXT3_SUPER_MAGIC)
1691                 goto cantfind_ext3;
1692
1693         /* Set defaults before we parse the mount options */
1694         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1695         if (def_mount_opts & EXT3_DEFM_DEBUG)
1696                 set_opt(sbi->s_mount_opt, DEBUG);
1697         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1698                 set_opt(sbi->s_mount_opt, GRPID);
1699         if (def_mount_opts & EXT3_DEFM_UID16)
1700                 set_opt(sbi->s_mount_opt, NO_UID32);
1701 #ifdef CONFIG_EXT3_FS_XATTR
1702         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1703                 set_opt(sbi->s_mount_opt, XATTR_USER);
1704 #endif
1705 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1706         if (def_mount_opts & EXT3_DEFM_ACL)
1707                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1708 #endif
1709         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1710                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1711         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1712                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1713         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1714                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1715
1716         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1717                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1718         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1719                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1720         else
1721                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1722
1723         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1724         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1725
1726         /* enable barriers by default */
1727         set_opt(sbi->s_mount_opt, BARRIER);
1728         set_opt(sbi->s_mount_opt, RESERVATION);
1729
1730         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1731                             NULL, 0))
1732                 goto failed_mount;
1733
1734         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1735                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1736
1737         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1738             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1739              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1740              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1741                 ext3_msg(sb, KERN_WARNING,
1742                         "warning: feature flags set on rev 0 fs, "
1743                         "running e2fsck is recommended");
1744         /*
1745          * Check feature flags regardless of the revision level, since we
1746          * previously didn't change the revision level when setting the flags,
1747          * so there is a chance incompat flags are set on a rev 0 filesystem.
1748          */
1749         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1750         if (features) {
1751                 ext3_msg(sb, KERN_ERR,
1752                         "error: couldn't mount because of unsupported "
1753                         "optional features (%x)", le32_to_cpu(features));
1754                 goto failed_mount;
1755         }
1756         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1757         if (!(sb->s_flags & MS_RDONLY) && features) {
1758                 ext3_msg(sb, KERN_ERR,
1759                         "error: couldn't mount RDWR because of unsupported "
1760                         "optional features (%x)", le32_to_cpu(features));
1761                 goto failed_mount;
1762         }
1763         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1764
1765         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1766             blocksize > EXT3_MAX_BLOCK_SIZE) {
1767                 ext3_msg(sb, KERN_ERR,
1768                         "error: couldn't mount because of unsupported "
1769                         "filesystem blocksize %d", blocksize);
1770                 goto failed_mount;
1771         }
1772
1773         hblock = bdev_logical_block_size(sb->s_bdev);
1774         if (sb->s_blocksize != blocksize) {
1775                 /*
1776                  * Make sure the blocksize for the filesystem is larger
1777                  * than the hardware sectorsize for the machine.
1778                  */
1779                 if (blocksize < hblock) {
1780                         ext3_msg(sb, KERN_ERR,
1781                                 "error: fsblocksize %d too small for "
1782                                 "hardware sectorsize %d", blocksize, hblock);
1783                         goto failed_mount;
1784                 }
1785
1786                 brelse (bh);
1787                 if (!sb_set_blocksize(sb, blocksize)) {
1788                         ext3_msg(sb, KERN_ERR,
1789                                 "error: bad blocksize %d", blocksize);
1790                         goto out_fail;
1791                 }
1792                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1793                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1794                 bh = sb_bread(sb, logic_sb_block);
1795                 if (!bh) {
1796                         ext3_msg(sb, KERN_ERR,
1797                                "error: can't read superblock on 2nd try");
1798                         goto failed_mount;
1799                 }
1800                 es = (struct ext3_super_block *)(bh->b_data + offset);
1801                 sbi->s_es = es;
1802                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1803                         ext3_msg(sb, KERN_ERR,
1804                                 "error: magic mismatch");
1805                         goto failed_mount;
1806                 }
1807         }
1808
1809         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1810
1811         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1812                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1813                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1814         } else {
1815                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1816                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1817                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1818                     (!is_power_of_2(sbi->s_inode_size)) ||
1819                     (sbi->s_inode_size > blocksize)) {
1820                         ext3_msg(sb, KERN_ERR,
1821                                 "error: unsupported inode size: %d",
1822                                 sbi->s_inode_size);
1823                         goto failed_mount;
1824                 }
1825         }
1826         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1827                                    le32_to_cpu(es->s_log_frag_size);
1828         if (blocksize != sbi->s_frag_size) {
1829                 ext3_msg(sb, KERN_ERR,
1830                        "error: fragsize %lu != blocksize %u (unsupported)",
1831                        sbi->s_frag_size, blocksize);
1832                 goto failed_mount;
1833         }
1834         sbi->s_frags_per_block = 1;
1835         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1836         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1837         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1838         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1839                 goto cantfind_ext3;
1840         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1841         if (sbi->s_inodes_per_block == 0)
1842                 goto cantfind_ext3;
1843         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1844                                         sbi->s_inodes_per_block;
1845         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1846         sbi->s_sbh = bh;
1847         sbi->s_mount_state = le16_to_cpu(es->s_state);
1848         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1849         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1850         for (i=0; i < 4; i++)
1851                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1852         sbi->s_def_hash_version = es->s_def_hash_version;
1853         i = le32_to_cpu(es->s_flags);
1854         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1855                 sbi->s_hash_unsigned = 3;
1856         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1857 #ifdef __CHAR_UNSIGNED__
1858                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1859                 sbi->s_hash_unsigned = 3;
1860 #else
1861                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1862 #endif
1863         }
1864
1865         if (sbi->s_blocks_per_group > blocksize * 8) {
1866                 ext3_msg(sb, KERN_ERR,
1867                         "#blocks per group too big: %lu",
1868                         sbi->s_blocks_per_group);
1869                 goto failed_mount;
1870         }
1871         if (sbi->s_frags_per_group > blocksize * 8) {
1872                 ext3_msg(sb, KERN_ERR,
1873                         "error: #fragments per group too big: %lu",
1874                         sbi->s_frags_per_group);
1875                 goto failed_mount;
1876         }
1877         if (sbi->s_inodes_per_group > blocksize * 8) {
1878                 ext3_msg(sb, KERN_ERR,
1879                         "error: #inodes per group too big: %lu",
1880                         sbi->s_inodes_per_group);
1881                 goto failed_mount;
1882         }
1883
1884         err = generic_check_addressable(sb->s_blocksize_bits,
1885                                         le32_to_cpu(es->s_blocks_count));
1886         if (err) {
1887                 ext3_msg(sb, KERN_ERR,
1888                         "error: filesystem is too large to mount safely");
1889                 if (sizeof(sector_t) < 8)
1890                         ext3_msg(sb, KERN_ERR,
1891                                 "error: CONFIG_LBDAF not enabled");
1892                 ret = err;
1893                 goto failed_mount;
1894         }
1895
1896         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1897                 goto cantfind_ext3;
1898         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1899                                le32_to_cpu(es->s_first_data_block) - 1)
1900                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1901         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1902         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1903                                     GFP_KERNEL);
1904         if (sbi->s_group_desc == NULL) {
1905                 ext3_msg(sb, KERN_ERR,
1906                         "error: not enough memory");
1907                 ret = -ENOMEM;
1908                 goto failed_mount;
1909         }
1910
1911         bgl_lock_init(sbi->s_blockgroup_lock);
1912
1913         for (i = 0; i < db_count; i++) {
1914                 block = descriptor_loc(sb, logic_sb_block, i);
1915                 sbi->s_group_desc[i] = sb_bread(sb, block);
1916                 if (!sbi->s_group_desc[i]) {
1917                         ext3_msg(sb, KERN_ERR,
1918                                 "error: can't read group descriptor %d", i);
1919                         db_count = i;
1920                         goto failed_mount2;
1921                 }
1922         }
1923         if (!ext3_check_descriptors (sb)) {
1924                 ext3_msg(sb, KERN_ERR,
1925                         "error: group descriptors corrupted");
1926                 goto failed_mount2;
1927         }
1928         sbi->s_gdb_count = db_count;
1929         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1930         spin_lock_init(&sbi->s_next_gen_lock);
1931
1932         /* per fileystem reservation list head & lock */
1933         spin_lock_init(&sbi->s_rsv_window_lock);
1934         sbi->s_rsv_window_root = RB_ROOT;
1935         /* Add a single, static dummy reservation to the start of the
1936          * reservation window list --- it gives us a placeholder for
1937          * append-at-start-of-list which makes the allocation logic
1938          * _much_ simpler. */
1939         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1940         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1941         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1942         sbi->s_rsv_window_head.rsv_goal_size = 0;
1943         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1944
1945         /*
1946          * set up enough so that it can read an inode
1947          */
1948         sb->s_op = &ext3_sops;
1949         sb->s_export_op = &ext3_export_ops;
1950         sb->s_xattr = ext3_xattr_handlers;
1951 #ifdef CONFIG_QUOTA
1952         sb->s_qcop = &ext3_qctl_operations;
1953         sb->dq_op = &ext3_quota_operations;
1954 #endif
1955         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1956         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1957         mutex_init(&sbi->s_orphan_lock);
1958         mutex_init(&sbi->s_resize_lock);
1959
1960         sb->s_root = NULL;
1961
1962         needs_recovery = (es->s_last_orphan != 0 ||
1963                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1964                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1965
1966         /*
1967          * The first inode we look at is the journal inode.  Don't try
1968          * root first: it may be modified in the journal!
1969          */
1970         if (!test_opt(sb, NOLOAD) &&
1971             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1972                 if (ext3_load_journal(sb, es, journal_devnum))
1973                         goto failed_mount2;
1974         } else if (journal_inum) {
1975                 if (ext3_create_journal(sb, es, journal_inum))
1976                         goto failed_mount2;
1977         } else {
1978                 if (!silent)
1979                         ext3_msg(sb, KERN_ERR,
1980                                 "error: no journal found. "
1981                                 "mounting ext3 over ext2?");
1982                 goto failed_mount2;
1983         }
1984         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1985                         ext3_count_free_blocks(sb));
1986         if (!err) {
1987                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1988                                 ext3_count_free_inodes(sb));
1989         }
1990         if (!err) {
1991                 err = percpu_counter_init(&sbi->s_dirs_counter,
1992                                 ext3_count_dirs(sb));
1993         }
1994         if (err) {
1995                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1996                 ret = err;
1997                 goto failed_mount3;
1998         }
1999
2000         /* We have now updated the journal if required, so we can
2001          * validate the data journaling mode. */
2002         switch (test_opt(sb, DATA_FLAGS)) {
2003         case 0:
2004                 /* No mode set, assume a default based on the journal
2005                    capabilities: ORDERED_DATA if the journal can
2006                    cope, else JOURNAL_DATA */
2007                 if (journal_check_available_features
2008                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2009                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2010                 else
2011                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2012                 break;
2013
2014         case EXT3_MOUNT_ORDERED_DATA:
2015         case EXT3_MOUNT_WRITEBACK_DATA:
2016                 if (!journal_check_available_features
2017                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2018                         ext3_msg(sb, KERN_ERR,
2019                                 "error: journal does not support "
2020                                 "requested data journaling mode");
2021                         goto failed_mount3;
2022                 }
2023         default:
2024                 break;
2025         }
2026
2027         /*
2028          * The journal_load will have done any necessary log recovery,
2029          * so we can safely mount the rest of the filesystem now.
2030          */
2031
2032         root = ext3_iget(sb, EXT3_ROOT_INO);
2033         if (IS_ERR(root)) {
2034                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2035                 ret = PTR_ERR(root);
2036                 goto failed_mount3;
2037         }
2038         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2039                 iput(root);
2040                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2041                 goto failed_mount3;
2042         }
2043         sb->s_root = d_alloc_root(root);
2044         if (!sb->s_root) {
2045                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2046                 iput(root);
2047                 ret = -ENOMEM;
2048                 goto failed_mount3;
2049         }
2050
2051         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2052
2053         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2054         ext3_orphan_cleanup(sb, es);
2055         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2056         if (needs_recovery)
2057                 ext3_msg(sb, KERN_INFO, "recovery complete");
2058         ext3_mark_recovery_complete(sb, es);
2059         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2060                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2061                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2062                 "writeback");
2063
2064         return 0;
2065
2066 cantfind_ext3:
2067         if (!silent)
2068                 ext3_msg(sb, KERN_INFO,
2069                         "error: can't find ext3 filesystem on dev %s.",
2070                        sb->s_id);
2071         goto failed_mount;
2072
2073 failed_mount3:
2074         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2075         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2076         percpu_counter_destroy(&sbi->s_dirs_counter);
2077         journal_destroy(sbi->s_journal);
2078 failed_mount2:
2079         for (i = 0; i < db_count; i++)
2080                 brelse(sbi->s_group_desc[i]);
2081         kfree(sbi->s_group_desc);
2082 failed_mount:
2083 #ifdef CONFIG_QUOTA
2084         for (i = 0; i < MAXQUOTAS; i++)
2085                 kfree(sbi->s_qf_names[i]);
2086 #endif
2087         ext3_blkdev_remove(sbi);
2088         brelse(bh);
2089 out_fail:
2090         sb->s_fs_info = NULL;
2091         kfree(sbi->s_blockgroup_lock);
2092         kfree(sbi);
2093         return ret;
2094 }
2095
2096 /*
2097  * Setup any per-fs journal parameters now.  We'll do this both on
2098  * initial mount, once the journal has been initialised but before we've
2099  * done any recovery; and again on any subsequent remount.
2100  */
2101 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2102 {
2103         struct ext3_sb_info *sbi = EXT3_SB(sb);
2104
2105         if (sbi->s_commit_interval)
2106                 journal->j_commit_interval = sbi->s_commit_interval;
2107         /* We could also set up an ext3-specific default for the commit
2108          * interval here, but for now we'll just fall back to the jbd
2109          * default. */
2110
2111         spin_lock(&journal->j_state_lock);
2112         if (test_opt(sb, BARRIER))
2113                 journal->j_flags |= JFS_BARRIER;
2114         else
2115                 journal->j_flags &= ~JFS_BARRIER;
2116         if (test_opt(sb, DATA_ERR_ABORT))
2117                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2118         else
2119                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2120         spin_unlock(&journal->j_state_lock);
2121 }
2122
2123 static journal_t *ext3_get_journal(struct super_block *sb,
2124                                    unsigned int journal_inum)
2125 {
2126         struct inode *journal_inode;
2127         journal_t *journal;
2128
2129         /* First, test for the existence of a valid inode on disk.  Bad
2130          * things happen if we iget() an unused inode, as the subsequent
2131          * iput() will try to delete it. */
2132
2133         journal_inode = ext3_iget(sb, journal_inum);
2134         if (IS_ERR(journal_inode)) {
2135                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2136                 return NULL;
2137         }
2138         if (!journal_inode->i_nlink) {
2139                 make_bad_inode(journal_inode);
2140                 iput(journal_inode);
2141                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2142                 return NULL;
2143         }
2144
2145         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2146                   journal_inode, journal_inode->i_size);
2147         if (!S_ISREG(journal_inode->i_mode)) {
2148                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2149                 iput(journal_inode);
2150                 return NULL;
2151         }
2152
2153         journal = journal_init_inode(journal_inode);
2154         if (!journal) {
2155                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2156                 iput(journal_inode);
2157                 return NULL;
2158         }
2159         journal->j_private = sb;
2160         ext3_init_journal_params(sb, journal);
2161         return journal;
2162 }
2163
2164 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2165                                        dev_t j_dev)
2166 {
2167         struct buffer_head * bh;
2168         journal_t *journal;
2169         ext3_fsblk_t start;
2170         ext3_fsblk_t len;
2171         int hblock, blocksize;
2172         ext3_fsblk_t sb_block;
2173         unsigned long offset;
2174         struct ext3_super_block * es;
2175         struct block_device *bdev;
2176
2177         bdev = ext3_blkdev_get(j_dev, sb);
2178         if (bdev == NULL)
2179                 return NULL;
2180
2181         blocksize = sb->s_blocksize;
2182         hblock = bdev_logical_block_size(bdev);
2183         if (blocksize < hblock) {
2184                 ext3_msg(sb, KERN_ERR,
2185                         "error: blocksize too small for journal device");
2186                 goto out_bdev;
2187         }
2188
2189         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2190         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2191         set_blocksize(bdev, blocksize);
2192         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2193                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2194                         "external journal");
2195                 goto out_bdev;
2196         }
2197
2198         es = (struct ext3_super_block *) (bh->b_data + offset);
2199         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2200             !(le32_to_cpu(es->s_feature_incompat) &
2201               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2202                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2203                         "bad superblock");
2204                 brelse(bh);
2205                 goto out_bdev;
2206         }
2207
2208         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2209                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2210                 brelse(bh);
2211                 goto out_bdev;
2212         }
2213
2214         len = le32_to_cpu(es->s_blocks_count);
2215         start = sb_block + 1;
2216         brelse(bh);     /* we're done with the superblock */
2217
2218         journal = journal_init_dev(bdev, sb->s_bdev,
2219                                         start, len, blocksize);
2220         if (!journal) {
2221                 ext3_msg(sb, KERN_ERR,
2222                         "error: failed to create device journal");
2223                 goto out_bdev;
2224         }
2225         journal->j_private = sb;
2226         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2227         wait_on_buffer(journal->j_sb_buffer);
2228         if (!buffer_uptodate(journal->j_sb_buffer)) {
2229                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2230                 goto out_journal;
2231         }
2232         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2233                 ext3_msg(sb, KERN_ERR,
2234                         "error: external journal has more than one "
2235                         "user (unsupported) - %d",
2236                         be32_to_cpu(journal->j_superblock->s_nr_users));
2237                 goto out_journal;
2238         }
2239         EXT3_SB(sb)->journal_bdev = bdev;
2240         ext3_init_journal_params(sb, journal);
2241         return journal;
2242 out_journal:
2243         journal_destroy(journal);
2244 out_bdev:
2245         ext3_blkdev_put(bdev);
2246         return NULL;
2247 }
2248
2249 static int ext3_load_journal(struct super_block *sb,
2250                              struct ext3_super_block *es,
2251                              unsigned long journal_devnum)
2252 {
2253         journal_t *journal;
2254         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2255         dev_t journal_dev;
2256         int err = 0;
2257         int really_read_only;
2258
2259         if (journal_devnum &&
2260             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2261                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2262                         "numbers have changed");
2263                 journal_dev = new_decode_dev(journal_devnum);
2264         } else
2265                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2266
2267         really_read_only = bdev_read_only(sb->s_bdev);
2268
2269         /*
2270          * Are we loading a blank journal or performing recovery after a
2271          * crash?  For recovery, we need to check in advance whether we
2272          * can get read-write access to the device.
2273          */
2274
2275         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2276                 if (sb->s_flags & MS_RDONLY) {
2277                         ext3_msg(sb, KERN_INFO,
2278                                 "recovery required on readonly filesystem");
2279                         if (really_read_only) {
2280                                 ext3_msg(sb, KERN_ERR, "error: write access "
2281                                         "unavailable, cannot proceed");
2282                                 return -EROFS;
2283                         }
2284                         ext3_msg(sb, KERN_INFO,
2285                                 "write access will be enabled during recovery");
2286                 }
2287         }
2288
2289         if (journal_inum && journal_dev) {
2290                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2291                        "and inode journals");
2292                 return -EINVAL;
2293         }
2294
2295         if (journal_inum) {
2296                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2297                         return -EINVAL;
2298         } else {
2299                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2300                         return -EINVAL;
2301         }
2302
2303         if (!(journal->j_flags & JFS_BARRIER))
2304                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2305
2306         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2307                 err = journal_update_format(journal);
2308                 if (err)  {
2309                         ext3_msg(sb, KERN_ERR, "error updating journal");
2310                         journal_destroy(journal);
2311                         return err;
2312                 }
2313         }
2314
2315         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2316                 err = journal_wipe(journal, !really_read_only);
2317         if (!err)
2318                 err = journal_load(journal);
2319
2320         if (err) {
2321                 ext3_msg(sb, KERN_ERR, "error loading journal");
2322                 journal_destroy(journal);
2323                 return err;
2324         }
2325
2326         EXT3_SB(sb)->s_journal = journal;
2327         ext3_clear_journal_err(sb, es);
2328
2329         if (!really_read_only && journal_devnum &&
2330             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2331                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2332
2333                 /* Make sure we flush the recovery flag to disk. */
2334                 ext3_commit_super(sb, es, 1);
2335         }
2336
2337         return 0;
2338 }
2339
2340 static int ext3_create_journal(struct super_block *sb,
2341                                struct ext3_super_block *es,
2342                                unsigned int journal_inum)
2343 {
2344         journal_t *journal;
2345         int err;
2346
2347         if (sb->s_flags & MS_RDONLY) {
2348                 ext3_msg(sb, KERN_ERR,
2349                         "error: readonly filesystem when trying to "
2350                         "create journal");
2351                 return -EROFS;
2352         }
2353
2354         journal = ext3_get_journal(sb, journal_inum);
2355         if (!journal)
2356                 return -EINVAL;
2357
2358         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2359                journal_inum);
2360
2361         err = journal_create(journal);
2362         if (err) {
2363                 ext3_msg(sb, KERN_ERR, "error creating journal");
2364                 journal_destroy(journal);
2365                 return -EIO;
2366         }
2367
2368         EXT3_SB(sb)->s_journal = journal;
2369
2370         ext3_update_dynamic_rev(sb);
2371         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2372         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2373
2374         es->s_journal_inum = cpu_to_le32(journal_inum);
2375
2376         /* Make sure we flush the recovery flag to disk. */
2377         ext3_commit_super(sb, es, 1);
2378
2379         return 0;
2380 }
2381
2382 static int ext3_commit_super(struct super_block *sb,
2383                                struct ext3_super_block *es,
2384                                int sync)
2385 {
2386         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2387         int error = 0;
2388
2389         if (!sbh)
2390                 return error;
2391
2392         if (buffer_write_io_error(sbh)) {
2393                 /*
2394                  * Oh, dear.  A previous attempt to write the
2395                  * superblock failed.  This could happen because the
2396                  * USB device was yanked out.  Or it could happen to
2397                  * be a transient write error and maybe the block will
2398                  * be remapped.  Nothing we can do but to retry the
2399                  * write and hope for the best.
2400                  */
2401                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2402                        "superblock detected");
2403                 clear_buffer_write_io_error(sbh);
2404                 set_buffer_uptodate(sbh);
2405         }
2406         /*
2407          * If the file system is mounted read-only, don't update the
2408          * superblock write time.  This avoids updating the superblock
2409          * write time when we are mounting the root file system
2410          * read/only but we need to replay the journal; at that point,
2411          * for people who are east of GMT and who make their clock
2412          * tick in localtime for Windows bug-for-bug compatibility,
2413          * the clock is set in the future, and this will cause e2fsck
2414          * to complain and force a full file system check.
2415          */
2416         if (!(sb->s_flags & MS_RDONLY))
2417                 es->s_wtime = cpu_to_le32(get_seconds());
2418         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2419         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2420         BUFFER_TRACE(sbh, "marking dirty");
2421         mark_buffer_dirty(sbh);
2422         if (sync) {
2423                 error = sync_dirty_buffer(sbh);
2424                 if (buffer_write_io_error(sbh)) {
2425                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2426                                "superblock");
2427                         clear_buffer_write_io_error(sbh);
2428                         set_buffer_uptodate(sbh);
2429                 }
2430         }
2431         return error;
2432 }
2433
2434
2435 /*
2436  * Have we just finished recovery?  If so, and if we are mounting (or
2437  * remounting) the filesystem readonly, then we will end up with a
2438  * consistent fs on disk.  Record that fact.
2439  */
2440 static void ext3_mark_recovery_complete(struct super_block * sb,
2441                                         struct ext3_super_block * es)
2442 {
2443         journal_t *journal = EXT3_SB(sb)->s_journal;
2444
2445         journal_lock_updates(journal);
2446         if (journal_flush(journal) < 0)
2447                 goto out;
2448
2449         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2450             sb->s_flags & MS_RDONLY) {
2451                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2452                 ext3_commit_super(sb, es, 1);
2453         }
2454
2455 out:
2456         journal_unlock_updates(journal);
2457 }
2458
2459 /*
2460  * If we are mounting (or read-write remounting) a filesystem whose journal
2461  * has recorded an error from a previous lifetime, move that error to the
2462  * main filesystem now.
2463  */
2464 static void ext3_clear_journal_err(struct super_block *sb,
2465                                    struct ext3_super_block *es)
2466 {
2467         journal_t *journal;
2468         int j_errno;
2469         const char *errstr;
2470
2471         journal = EXT3_SB(sb)->s_journal;
2472
2473         /*
2474          * Now check for any error status which may have been recorded in the
2475          * journal by a prior ext3_error() or ext3_abort()
2476          */
2477
2478         j_errno = journal_errno(journal);
2479         if (j_errno) {
2480                 char nbuf[16];
2481
2482                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2483                 ext3_warning(sb, __func__, "Filesystem error recorded "
2484                              "from previous mount: %s", errstr);
2485                 ext3_warning(sb, __func__, "Marking fs in need of "
2486                              "filesystem check.");
2487
2488                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2489                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2490                 ext3_commit_super (sb, es, 1);
2491
2492                 journal_clear_err(journal);
2493         }
2494 }
2495
2496 /*
2497  * Force the running and committing transactions to commit,
2498  * and wait on the commit.
2499  */
2500 int ext3_force_commit(struct super_block *sb)
2501 {
2502         journal_t *journal;
2503         int ret;
2504
2505         if (sb->s_flags & MS_RDONLY)
2506                 return 0;
2507
2508         journal = EXT3_SB(sb)->s_journal;
2509         ret = ext3_journal_force_commit(journal);
2510         return ret;
2511 }
2512
2513 static int ext3_sync_fs(struct super_block *sb, int wait)
2514 {
2515         tid_t target;
2516
2517         trace_ext3_sync_fs(sb, wait);
2518         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2519                 if (wait)
2520                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2521         }
2522         return 0;
2523 }
2524
2525 /*
2526  * LVM calls this function before a (read-only) snapshot is created.  This
2527  * gives us a chance to flush the journal completely and mark the fs clean.
2528  */
2529 static int ext3_freeze(struct super_block *sb)
2530 {
2531         int error = 0;
2532         journal_t *journal;
2533
2534         if (!(sb->s_flags & MS_RDONLY)) {
2535                 journal = EXT3_SB(sb)->s_journal;
2536
2537                 /* Now we set up the journal barrier. */
2538                 journal_lock_updates(journal);
2539
2540                 /*
2541                  * We don't want to clear needs_recovery flag when we failed
2542                  * to flush the journal.
2543                  */
2544                 error = journal_flush(journal);
2545                 if (error < 0)
2546                         goto out;
2547
2548                 /* Journal blocked and flushed, clear needs_recovery flag. */
2549                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2550                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2551                 if (error)
2552                         goto out;
2553         }
2554         return 0;
2555
2556 out:
2557         journal_unlock_updates(journal);
2558         return error;
2559 }
2560
2561 /*
2562  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2563  * flag here, even though the filesystem is not technically dirty yet.
2564  */
2565 static int ext3_unfreeze(struct super_block *sb)
2566 {
2567         if (!(sb->s_flags & MS_RDONLY)) {
2568                 lock_super(sb);
2569                 /* Reser the needs_recovery flag before the fs is unlocked. */
2570                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2571                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2572                 unlock_super(sb);
2573                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2574         }
2575         return 0;
2576 }
2577
2578 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2579 {
2580         struct ext3_super_block * es;
2581         struct ext3_sb_info *sbi = EXT3_SB(sb);
2582         ext3_fsblk_t n_blocks_count = 0;
2583         unsigned long old_sb_flags;
2584         struct ext3_mount_options old_opts;
2585         int enable_quota = 0;
2586         int err;
2587 #ifdef CONFIG_QUOTA
2588         int i;
2589 #endif
2590
2591         /* Store the original options */
2592         lock_super(sb);
2593         old_sb_flags = sb->s_flags;
2594         old_opts.s_mount_opt = sbi->s_mount_opt;
2595         old_opts.s_resuid = sbi->s_resuid;
2596         old_opts.s_resgid = sbi->s_resgid;
2597         old_opts.s_commit_interval = sbi->s_commit_interval;
2598 #ifdef CONFIG_QUOTA
2599         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2600         for (i = 0; i < MAXQUOTAS; i++)
2601                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2602 #endif
2603
2604         /*
2605          * Allow the "check" option to be passed as a remount option.
2606          */
2607         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2608                 err = -EINVAL;
2609                 goto restore_opts;
2610         }
2611
2612         if (test_opt(sb, ABORT))
2613                 ext3_abort(sb, __func__, "Abort forced by user");
2614
2615         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2616                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2617
2618         es = sbi->s_es;
2619
2620         ext3_init_journal_params(sb, sbi->s_journal);
2621
2622         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2623                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2624                 if (test_opt(sb, ABORT)) {
2625                         err = -EROFS;
2626                         goto restore_opts;
2627                 }
2628
2629                 if (*flags & MS_RDONLY) {
2630                         err = dquot_suspend(sb, -1);
2631                         if (err < 0)
2632                                 goto restore_opts;
2633
2634                         /*
2635                          * First of all, the unconditional stuff we have to do
2636                          * to disable replay of the journal when we next remount
2637                          */
2638                         sb->s_flags |= MS_RDONLY;
2639
2640                         /*
2641                          * OK, test if we are remounting a valid rw partition
2642                          * readonly, and if so set the rdonly flag and then
2643                          * mark the partition as valid again.
2644                          */
2645                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2646                             (sbi->s_mount_state & EXT3_VALID_FS))
2647                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2648
2649                         ext3_mark_recovery_complete(sb, es);
2650                 } else {
2651                         __le32 ret;
2652                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2653                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2654                                 ext3_msg(sb, KERN_WARNING,
2655                                         "warning: couldn't remount RDWR "
2656                                         "because of unsupported optional "
2657                                         "features (%x)", le32_to_cpu(ret));
2658                                 err = -EROFS;
2659                                 goto restore_opts;
2660                         }
2661
2662                         /*
2663                          * If we have an unprocessed orphan list hanging
2664                          * around from a previously readonly bdev mount,
2665                          * require a full umount & mount for now.
2666                          */
2667                         if (es->s_last_orphan) {
2668                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2669                                        "remount RDWR because of unprocessed "
2670                                        "orphan inode list.  Please "
2671                                        "umount & mount instead.");
2672                                 err = -EINVAL;
2673                                 goto restore_opts;
2674                         }
2675
2676                         /*
2677                          * Mounting a RDONLY partition read-write, so reread
2678                          * and store the current valid flag.  (It may have
2679                          * been changed by e2fsck since we originally mounted
2680                          * the partition.)
2681                          */
2682                         ext3_clear_journal_err(sb, es);
2683                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2684                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2685                                 goto restore_opts;
2686                         if (!ext3_setup_super (sb, es, 0))
2687                                 sb->s_flags &= ~MS_RDONLY;
2688                         enable_quota = 1;
2689                 }
2690         }
2691 #ifdef CONFIG_QUOTA
2692         /* Release old quota file names */
2693         for (i = 0; i < MAXQUOTAS; i++)
2694                 if (old_opts.s_qf_names[i] &&
2695                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2696                         kfree(old_opts.s_qf_names[i]);
2697 #endif
2698         unlock_super(sb);
2699
2700         if (enable_quota)
2701                 dquot_resume(sb, -1);
2702         return 0;
2703 restore_opts:
2704         sb->s_flags = old_sb_flags;
2705         sbi->s_mount_opt = old_opts.s_mount_opt;
2706         sbi->s_resuid = old_opts.s_resuid;
2707         sbi->s_resgid = old_opts.s_resgid;
2708         sbi->s_commit_interval = old_opts.s_commit_interval;
2709 #ifdef CONFIG_QUOTA
2710         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2711         for (i = 0; i < MAXQUOTAS; i++) {
2712                 if (sbi->s_qf_names[i] &&
2713                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2714                         kfree(sbi->s_qf_names[i]);
2715                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2716         }
2717 #endif
2718         unlock_super(sb);
2719         return err;
2720 }
2721
2722 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2723 {
2724         struct super_block *sb = dentry->d_sb;
2725         struct ext3_sb_info *sbi = EXT3_SB(sb);
2726         struct ext3_super_block *es = sbi->s_es;
2727         u64 fsid;
2728
2729         if (test_opt(sb, MINIX_DF)) {
2730                 sbi->s_overhead_last = 0;
2731         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2732                 unsigned long ngroups = sbi->s_groups_count, i;
2733                 ext3_fsblk_t overhead = 0;
2734                 smp_rmb();
2735
2736                 /*
2737                  * Compute the overhead (FS structures).  This is constant
2738                  * for a given filesystem unless the number of block groups
2739                  * changes so we cache the previous value until it does.
2740                  */
2741
2742                 /*
2743                  * All of the blocks before first_data_block are
2744                  * overhead
2745                  */
2746                 overhead = le32_to_cpu(es->s_first_data_block);
2747
2748                 /*
2749                  * Add the overhead attributed to the superblock and
2750                  * block group descriptors.  If the sparse superblocks
2751                  * feature is turned on, then not all groups have this.
2752                  */
2753                 for (i = 0; i < ngroups; i++) {
2754                         overhead += ext3_bg_has_super(sb, i) +
2755                                 ext3_bg_num_gdb(sb, i);
2756                         cond_resched();
2757                 }
2758
2759                 /*
2760                  * Every block group has an inode bitmap, a block
2761                  * bitmap, and an inode table.
2762                  */
2763                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2764                 sbi->s_overhead_last = overhead;
2765                 smp_wmb();
2766                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2767         }
2768
2769         buf->f_type = EXT3_SUPER_MAGIC;
2770         buf->f_bsize = sb->s_blocksize;
2771         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2772         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2773         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2774         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2775                 buf->f_bavail = 0;
2776         buf->f_files = le32_to_cpu(es->s_inodes_count);
2777         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2778         buf->f_namelen = EXT3_NAME_LEN;
2779         fsid = le64_to_cpup((void *)es->s_uuid) ^
2780                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2781         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2782         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2783         return 0;
2784 }
2785
2786 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2787  * is locked for write. Otherwise the are possible deadlocks:
2788  * Process 1                         Process 2
2789  * ext3_create()                     quota_sync()
2790  *   journal_start()                   write_dquot()
2791  *   dquot_initialize()                       down(dqio_mutex)
2792  *     down(dqio_mutex)                    journal_start()
2793  *
2794  */
2795
2796 #ifdef CONFIG_QUOTA
2797
2798 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2799 {
2800         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2801 }
2802
2803 static int ext3_write_dquot(struct dquot *dquot)
2804 {
2805         int ret, err;
2806         handle_t *handle;
2807         struct inode *inode;
2808
2809         inode = dquot_to_inode(dquot);
2810         handle = ext3_journal_start(inode,
2811                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2812         if (IS_ERR(handle))
2813                 return PTR_ERR(handle);
2814         ret = dquot_commit(dquot);
2815         err = ext3_journal_stop(handle);
2816         if (!ret)
2817                 ret = err;
2818         return ret;
2819 }
2820
2821 static int ext3_acquire_dquot(struct dquot *dquot)
2822 {
2823         int ret, err;
2824         handle_t *handle;
2825
2826         handle = ext3_journal_start(dquot_to_inode(dquot),
2827                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2828         if (IS_ERR(handle))
2829                 return PTR_ERR(handle);
2830         ret = dquot_acquire(dquot);
2831         err = ext3_journal_stop(handle);
2832         if (!ret)
2833                 ret = err;
2834         return ret;
2835 }
2836
2837 static int ext3_release_dquot(struct dquot *dquot)
2838 {
2839         int ret, err;
2840         handle_t *handle;
2841
2842         handle = ext3_journal_start(dquot_to_inode(dquot),
2843                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2844         if (IS_ERR(handle)) {
2845                 /* Release dquot anyway to avoid endless cycle in dqput() */
2846                 dquot_release(dquot);
2847                 return PTR_ERR(handle);
2848         }
2849         ret = dquot_release(dquot);
2850         err = ext3_journal_stop(handle);
2851         if (!ret)
2852                 ret = err;
2853         return ret;
2854 }
2855
2856 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2857 {
2858         /* Are we journaling quotas? */
2859         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2860             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2861                 dquot_mark_dquot_dirty(dquot);
2862                 return ext3_write_dquot(dquot);
2863         } else {
2864                 return dquot_mark_dquot_dirty(dquot);
2865         }
2866 }
2867
2868 static int ext3_write_info(struct super_block *sb, int type)
2869 {
2870         int ret, err;
2871         handle_t *handle;
2872
2873         /* Data block + inode block */
2874         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2875         if (IS_ERR(handle))
2876                 return PTR_ERR(handle);
2877         ret = dquot_commit_info(sb, type);
2878         err = ext3_journal_stop(handle);
2879         if (!ret)
2880                 ret = err;
2881         return ret;
2882 }
2883
2884 /*
2885  * Turn on quotas during mount time - we need to find
2886  * the quota file and such...
2887  */
2888 static int ext3_quota_on_mount(struct super_block *sb, int type)
2889 {
2890         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2891                                         EXT3_SB(sb)->s_jquota_fmt, type);
2892 }
2893
2894 /*
2895  * Standard function to be called on quota_on
2896  */
2897 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2898                          struct path *path)
2899 {
2900         int err;
2901
2902         if (!test_opt(sb, QUOTA))
2903                 return -EINVAL;
2904
2905         /* Quotafile not on the same filesystem? */
2906         if (path->mnt->mnt_sb != sb)
2907                 return -EXDEV;
2908         /* Journaling quota? */
2909         if (EXT3_SB(sb)->s_qf_names[type]) {
2910                 /* Quotafile not of fs root? */
2911                 if (path->dentry->d_parent != sb->s_root)
2912                         ext3_msg(sb, KERN_WARNING,
2913                                 "warning: Quota file not on filesystem root. "
2914                                 "Journaled quota will not work.");
2915         }
2916
2917         /*
2918          * When we journal data on quota file, we have to flush journal to see
2919          * all updates to the file when we bypass pagecache...
2920          */
2921         if (ext3_should_journal_data(path->dentry->d_inode)) {
2922                 /*
2923                  * We don't need to lock updates but journal_flush() could
2924                  * otherwise be livelocked...
2925                  */
2926                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2927                 err = journal_flush(EXT3_SB(sb)->s_journal);
2928                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2929                 if (err)
2930                         return err;
2931         }
2932
2933         return dquot_quota_on(sb, type, format_id, path);
2934 }
2935
2936 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2937  * acquiring the locks... As quota files are never truncated and quota code
2938  * itself serializes the operations (and no one else should touch the files)
2939  * we don't have to be afraid of races */
2940 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2941                                size_t len, loff_t off)
2942 {
2943         struct inode *inode = sb_dqopt(sb)->files[type];
2944         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2945         int err = 0;
2946         int offset = off & (sb->s_blocksize - 1);
2947         int tocopy;
2948         size_t toread;
2949         struct buffer_head *bh;
2950         loff_t i_size = i_size_read(inode);
2951
2952         if (off > i_size)
2953                 return 0;
2954         if (off+len > i_size)
2955                 len = i_size-off;
2956         toread = len;
2957         while (toread > 0) {
2958                 tocopy = sb->s_blocksize - offset < toread ?
2959                                 sb->s_blocksize - offset : toread;
2960                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2961                 if (err)
2962                         return err;
2963                 if (!bh)        /* A hole? */
2964                         memset(data, 0, tocopy);
2965                 else
2966                         memcpy(data, bh->b_data+offset, tocopy);
2967                 brelse(bh);
2968                 offset = 0;
2969                 toread -= tocopy;
2970                 data += tocopy;
2971                 blk++;
2972         }
2973         return len;
2974 }
2975
2976 /* Write to quotafile (we know the transaction is already started and has
2977  * enough credits) */
2978 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2979                                 const char *data, size_t len, loff_t off)
2980 {
2981         struct inode *inode = sb_dqopt(sb)->files[type];
2982         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2983         int err = 0;
2984         int offset = off & (sb->s_blocksize - 1);
2985         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2986         struct buffer_head *bh;
2987         handle_t *handle = journal_current_handle();
2988
2989         if (!handle) {
2990                 ext3_msg(sb, KERN_WARNING,
2991                         "warning: quota write (off=%llu, len=%llu)"
2992                         " cancelled because transaction is not started.",
2993                         (unsigned long long)off, (unsigned long long)len);
2994                 return -EIO;
2995         }
2996
2997         /*
2998          * Since we account only one data block in transaction credits,
2999          * then it is impossible to cross a block boundary.
3000          */
3001         if (sb->s_blocksize - offset < len) {
3002                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3003                         " cancelled because not block aligned",
3004                         (unsigned long long)off, (unsigned long long)len);
3005                 return -EIO;
3006         }
3007         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3008         bh = ext3_bread(handle, inode, blk, 1, &err);
3009         if (!bh)
3010                 goto out;
3011         if (journal_quota) {
3012                 err = ext3_journal_get_write_access(handle, bh);
3013                 if (err) {
3014                         brelse(bh);
3015                         goto out;
3016                 }
3017         }
3018         lock_buffer(bh);
3019         memcpy(bh->b_data+offset, data, len);
3020         flush_dcache_page(bh->b_page);
3021         unlock_buffer(bh);
3022         if (journal_quota)
3023                 err = ext3_journal_dirty_metadata(handle, bh);
3024         else {
3025                 /* Always do at least ordered writes for quotas */
3026                 err = ext3_journal_dirty_data(handle, bh);
3027                 mark_buffer_dirty(bh);
3028         }
3029         brelse(bh);
3030 out:
3031         if (err) {
3032                 mutex_unlock(&inode->i_mutex);
3033                 return err;
3034         }
3035         if (inode->i_size < off + len) {
3036                 i_size_write(inode, off + len);
3037                 EXT3_I(inode)->i_disksize = inode->i_size;
3038         }
3039         inode->i_version++;
3040         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3041         ext3_mark_inode_dirty(handle, inode);
3042         mutex_unlock(&inode->i_mutex);
3043         return len;
3044 }
3045
3046 #endif
3047
3048 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3049         int flags, const char *dev_name, void *data)
3050 {
3051         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3052 }
3053
3054 static struct file_system_type ext3_fs_type = {
3055         .owner          = THIS_MODULE,
3056         .name           = "ext3",
3057         .mount          = ext3_mount,
3058         .kill_sb        = kill_block_super,
3059         .fs_flags       = FS_REQUIRES_DEV,
3060 };
3061
3062 static int __init init_ext3_fs(void)
3063 {
3064         int err = init_ext3_xattr();
3065         if (err)
3066                 return err;
3067         err = init_inodecache();
3068         if (err)
3069                 goto out1;
3070         err = register_filesystem(&ext3_fs_type);
3071         if (err)
3072                 goto out;
3073         return 0;
3074 out:
3075         destroy_inodecache();
3076 out1:
3077         exit_ext3_xattr();
3078         return err;
3079 }
3080
3081 static void __exit exit_ext3_fs(void)
3082 {
3083         unregister_filesystem(&ext3_fs_type);
3084         destroy_inodecache();
3085         exit_ext3_xattr();
3086 }
3087
3088 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3089 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3090 MODULE_LICENSE("GPL");
3091 module_init(init_ext3_fs)
3092 module_exit(exit_ext3_fs)