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