fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46
  47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  49 #else
  50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  51 #endif
  52
  53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  54                              unsigned long journal_devnum);
  55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  56                                unsigned int);
  57 static int ext3_commit_super(struct super_block *sb,
  58                                struct ext3_super_block *es,
  59                                int sync);
  60 static void ext3_mark_recovery_complete(struct super_block * sb,
  61                                         struct ext3_super_block * es);
  62 static void ext3_clear_journal_err(struct super_block * sb,
  63                                    struct ext3_super_block * es);
  64 static int ext3_sync_fs(struct super_block *sb, int wait);
  65 static const char *ext3_decode_error(struct super_block * sb, int errno,
  66                                      char nbuf[16]);
  67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  69 static int ext3_unfreeze(struct super_block *sb);
  70 static int ext3_freeze(struct super_block *sb);
  71
  72 /*
  73  * Wrappers for journal_start/end.
  74  *
  75  * The only special thing we need to do here is to make sure that all
  76  * journal_end calls result in the superblock being marked dirty, so
  77  * that sync() will call the filesystem's write_super callback if
  78  * appropriate.
  79  */
  80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  81 {
  82         journal_t *journal;
  83
  84         if (sb->s_flags & MS_RDONLY)
  85                 return ERR_PTR(-EROFS);
  86
  87         /* Special case here: if the journal has aborted behind our
  88          * backs (eg. EIO in the commit thread), then we still need to
  89          * take the FS itself readonly cleanly. */
  90         journal = EXT3_SB(sb)->s_journal;
  91         if (is_journal_aborted(journal)) {
  92                 ext3_abort(sb, __func__,
  93                            "Detected aborted journal");
  94                 return ERR_PTR(-EROFS);
  95         }
  96
  97         return journal_start(journal, nblocks);
  98 }
  99
 100 /*
 101  * The only special thing we need to do here is to make sure that all
 102  * journal_stop calls result in the superblock being marked dirty, so
 103  * that sync() will call the filesystem's write_super callback if
 104  * appropriate.
 105  */
 106 int __ext3_journal_stop(const char *where, handle_t *handle)
 107 {
 108         struct super_block *sb;
 109         int err;
 110         int rc;
 111
 112         sb = handle->h_transaction->t_journal->j_private;
 113         err = handle->h_err;
 114         rc = journal_stop(handle);
 115
 116         if (!err)
 117                 err = rc;
 118         if (err)
 119                 __ext3_std_error(sb, where, err);
 120         return err;
 121 }
 122
 123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 124                 struct buffer_head *bh, handle_t *handle, int err)
 125 {
 126         char nbuf[16];
 127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 128
 129         if (bh)
 130                 BUFFER_TRACE(bh, "abort");
 131
 132         if (!handle->h_err)
 133                 handle->h_err = err;
 134
 135         if (is_handle_aborted(handle))
 136                 return;
 137
 138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 139                 caller, errstr, err_fn);
 140
 141         journal_abort_handle(handle);
 142 }
 143
 144 void ext3_msg(struct super_block *sb, const char *prefix,
 145                 const char *fmt, ...)
 146 {
 147         va_list args;
 148
 149         va_start(args, fmt);
 150         printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
 151         vprintk(fmt, args);
 152         printk("\n");
 153         va_end(args);
 154 }
 155
 156 /* Deal with the reporting of failure conditions on a filesystem such as
 157  * inconsistencies detected or read IO failures.
 158  *
 159  * On ext2, we can store the error state of the filesystem in the
 160  * superblock.  That is not possible on ext3, because we may have other
 161  * write ordering constraints on the superblock which prevent us from
 162  * writing it out straight away; and given that the journal is about to
 163  * be aborted, we can't rely on the current, or future, transactions to
 164  * write out the superblock safely.
 165  *
 166  * We'll just use the journal_abort() error code to record an error in
 167  * the journal instead.  On recovery, the journal will complain about
 168  * that error until we've noted it down and cleared it.
 169  */
 170
 171 static void ext3_handle_error(struct super_block *sb)
 172 {
 173         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 174
 175         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 176         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 177
 178         if (sb->s_flags & MS_RDONLY)
 179                 return;
 180
 181         if (!test_opt (sb, ERRORS_CONT)) {
 182                 journal_t *journal = EXT3_SB(sb)->s_journal;
 183
 184                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 185                 if (journal)
 186                         journal_abort(journal, -EIO);
 187         }
 188         if (test_opt (sb, ERRORS_RO)) {
 189                 ext3_msg(sb, KERN_CRIT,
 190                         "error: remounting filesystem read-only");
 191                 sb->s_flags |= MS_RDONLY;
 192         }
 193         ext3_commit_super(sb, es, 1);
 194         if (test_opt(sb, ERRORS_PANIC))
 195                 panic("EXT3-fs (%s): panic forced after error\n",
 196                         sb->s_id);
 197 }
 198
 199 void ext3_error (struct super_block * sb, const char * function,
 200                  const char * fmt, ...)
 201 {
 202         va_list args;
 203
 204         va_start(args, fmt);
 205         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 206         vprintk(fmt, args);
 207         printk("\n");
 208         va_end(args);
 209
 210         ext3_handle_error(sb);
 211 }
 212
 213 static const char *ext3_decode_error(struct super_block * sb, int errno,
 214                                      char nbuf[16])
 215 {
 216         char *errstr = NULL;
 217
 218         switch (errno) {
 219         case -EIO:
 220                 errstr = "IO failure";
 221                 break;
 222         case -ENOMEM:
 223                 errstr = "Out of memory";
 224                 break;
 225         case -EROFS:
 226                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 227                         errstr = "Journal has aborted";
 228                 else
 229                         errstr = "Readonly filesystem";
 230                 break;
 231         default:
 232                 /* If the caller passed in an extra buffer for unknown
 233                  * errors, textualise them now.  Else we just return
 234                  * NULL. */
 235                 if (nbuf) {
 236                         /* Check for truncated error codes... */
 237                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 238                                 errstr = nbuf;
 239                 }
 240                 break;
 241         }
 242
 243         return errstr;
 244 }
 245
 246 /* __ext3_std_error decodes expected errors from journaling functions
 247  * automatically and invokes the appropriate error response.  */
 248
 249 void __ext3_std_error (struct super_block * sb, const char * function,
 250                        int errno)
 251 {
 252         char nbuf[16];
 253         const char *errstr;
 254
 255         /* Special case: if the error is EROFS, and we're not already
 256          * inside a transaction, then there's really no point in logging
 257          * an error. */
 258         if (errno == -EROFS && journal_current_handle() == NULL &&
 259             (sb->s_flags & MS_RDONLY))
 260                 return;
 261
 262         errstr = ext3_decode_error(sb, errno, nbuf);
 263         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 264
 265         ext3_handle_error(sb);
 266 }
 267
 268 /*
 269  * ext3_abort is a much stronger failure handler than ext3_error.  The
 270  * abort function may be used to deal with unrecoverable failures such
 271  * as journal IO errors or ENOMEM at a critical moment in log management.
 272  *
 273  * We unconditionally force the filesystem into an ABORT|READONLY state,
 274  * unless the error response on the fs has been set to panic in which
 275  * case we take the easy way out and panic immediately.
 276  */
 277
 278 void ext3_abort (struct super_block * sb, const char * function,
 279                  const char * fmt, ...)
 280 {
 281         va_list args;
 282
 283         va_start(args, fmt);
 284         printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
 285         vprintk(fmt, args);
 286         printk("\n");
 287         va_end(args);
 288
 289         if (test_opt(sb, ERRORS_PANIC))
 290                 panic("EXT3-fs: panic from previous error\n");
 291
 292         if (sb->s_flags & MS_RDONLY)
 293                 return;
 294
 295         ext3_msg(sb, KERN_CRIT,
 296                 "error: remounting filesystem read-only");
 297         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 298         sb->s_flags |= MS_RDONLY;
 299         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 300         if (EXT3_SB(sb)->s_journal)
 301                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 302 }
 303
 304 void ext3_warning (struct super_block * sb, const char * function,
 305                    const char * fmt, ...)
 306 {
 307         va_list args;
 308
 309         va_start(args, fmt);
 310         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
 311                sb->s_id, function);
 312         vprintk(fmt, args);
 313         printk("\n");
 314         va_end(args);
 315 }
 316
 317 void ext3_update_dynamic_rev(struct super_block *sb)
 318 {
 319         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 320
 321         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 322                 return;
 323
 324         ext3_msg(sb, KERN_WARNING,
 325                 "warning: updating to rev %d because of "
 326                 "new feature flag, running e2fsck is recommended",
 327                 EXT3_DYNAMIC_REV);
 328
 329         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 330         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 331         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 332         /* leave es->s_feature_*compat flags alone */
 333         /* es->s_uuid will be set by e2fsck if empty */
 334
 335         /*
 336          * The rest of the superblock fields should be zero, and if not it
 337          * means they are likely already in use, so leave them alone.  We
 338          * can leave it up to e2fsck to clean up any inconsistencies there.
 339          */
 340 }
 341
 342 /*
 343  * Open the external journal device
 344  */
 345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 346 {
 347         struct block_device *bdev;
 348         char b[BDEVNAME_SIZE];
 349
 350         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 351         if (IS_ERR(bdev))
 352                 goto fail;
 353         return bdev;
 354
 355 fail:
 356         ext3_msg(sb, "error: failed to open journal device %s: %ld",
 357                 __bdevname(dev, b), PTR_ERR(bdev));
 358
 359         return NULL;
 360 }
 361
 362 /*
 363  * Release the journal device
 364  */
 365 static int ext3_blkdev_put(struct block_device *bdev)
 366 {
 367         bd_release(bdev);
 368         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 369 }
 370
 371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 372 {
 373         struct block_device *bdev;
 374         int ret = -ENODEV;
 375
 376         bdev = sbi->journal_bdev;
 377         if (bdev) {
 378                 ret = ext3_blkdev_put(bdev);
 379                 sbi->journal_bdev = NULL;
 380         }
 381         return ret;
 382 }
 383
 384 static inline struct inode *orphan_list_entry(struct list_head *l)
 385 {
 386         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 387 }
 388
 389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 390 {
 391         struct list_head *l;
 392
 393         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 394                le32_to_cpu(sbi->s_es->s_last_orphan));
 395
 396         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 397         list_for_each(l, &sbi->s_orphan) {
 398                 struct inode *inode = orphan_list_entry(l);
 399                 ext3_msg(sb, KERN_ERR, "  "
 400                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 401                        inode->i_sb->s_id, inode->i_ino, inode,
 402                        inode->i_mode, inode->i_nlink,
 403                        NEXT_ORPHAN(inode));
 404         }
 405 }
 406
 407 static void ext3_put_super (struct super_block * sb)
 408 {
 409         struct ext3_sb_info *sbi = EXT3_SB(sb);
 410         struct ext3_super_block *es = sbi->s_es;
 411         int i, err;
 412
 413         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 414
 415         lock_kernel();
 416
 417         ext3_xattr_put_super(sb);
 418         err = journal_destroy(sbi->s_journal);
 419         sbi->s_journal = NULL;
 420         if (err < 0)
 421                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 422
 423         if (!(sb->s_flags & MS_RDONLY)) {
 424                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 425                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 426                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 427                 mark_buffer_dirty(sbi->s_sbh);
 428                 ext3_commit_super(sb, es, 1);
 429         }
 430
 431         for (i = 0; i < sbi->s_gdb_count; i++)
 432                 brelse(sbi->s_group_desc[i]);
 433         kfree(sbi->s_group_desc);
 434         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 435         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 436         percpu_counter_destroy(&sbi->s_dirs_counter);
 437         brelse(sbi->s_sbh);
 438 #ifdef CONFIG_QUOTA
 439         for (i = 0; i < MAXQUOTAS; i++)
 440                 kfree(sbi->s_qf_names[i]);
 441 #endif
 442
 443         /* Debugging code just in case the in-memory inode orphan list
 444          * isn't empty.  The on-disk one can be non-empty if we've
 445          * detected an error and taken the fs readonly, but the
 446          * in-memory list had better be clean by this point. */
 447         if (!list_empty(&sbi->s_orphan))
 448                 dump_orphan_list(sb, sbi);
 449         J_ASSERT(list_empty(&sbi->s_orphan));
 450
 451         invalidate_bdev(sb->s_bdev);
 452         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 453                 /*
 454                  * Invalidate the journal device's buffers.  We don't want them
 455                  * floating about in memory - the physical journal device may
 456                  * hotswapped, and it breaks the `ro-after' testing code.
 457                  */
 458                 sync_blockdev(sbi->journal_bdev);
 459                 invalidate_bdev(sbi->journal_bdev);
 460                 ext3_blkdev_remove(sbi);
 461         }
 462         sb->s_fs_info = NULL;
 463         kfree(sbi->s_blockgroup_lock);
 464         kfree(sbi);
 465
 466         unlock_kernel();
 467 }
 468
 469 static struct kmem_cache *ext3_inode_cachep;
 470
 471 /*
 472  * Called inside transaction, so use GFP_NOFS
 473  */
 474 static struct inode *ext3_alloc_inode(struct super_block *sb)
 475 {
 476         struct ext3_inode_info *ei;
 477
 478         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 479         if (!ei)
 480                 return NULL;
 481         ei->i_block_alloc_info = NULL;
 482         ei->vfs_inode.i_version = 1;
 483         atomic_set(&ei->i_datasync_tid, 0);
 484         atomic_set(&ei->i_sync_tid, 0);
 485         return &ei->vfs_inode;
 486 }
 487
 488 static void ext3_destroy_inode(struct inode *inode)
 489 {
 490         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 491                 printk("EXT3 Inode %p: orphan list check failed!\n",
 492                         EXT3_I(inode));
 493                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 494                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 495                                 false);
 496                 dump_stack();
 497         }
 498         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 499 }
 500
 501 static void init_once(void *foo)
 502 {
 503         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 504
 505         INIT_LIST_HEAD(&ei->i_orphan);
 506 #ifdef CONFIG_EXT3_FS_XATTR
 507         init_rwsem(&ei->xattr_sem);
 508 #endif
 509         mutex_init(&ei->truncate_mutex);
 510         inode_init_once(&ei->vfs_inode);
 511 }
 512
 513 static int init_inodecache(void)
 514 {
 515         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 516                                              sizeof(struct ext3_inode_info),
 517                                              0, (SLAB_RECLAIM_ACCOUNT|
 518                                                 SLAB_MEM_SPREAD),
 519                                              init_once);
 520         if (ext3_inode_cachep == NULL)
 521                 return -ENOMEM;
 522         return 0;
 523 }
 524
 525 static void destroy_inodecache(void)
 526 {
 527         kmem_cache_destroy(ext3_inode_cachep);
 528 }
 529
 530 static void ext3_clear_inode(struct inode *inode)
 531 {
 532         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 533
 534         dquot_drop(inode);
 535         ext3_discard_reservation(inode);
 536         EXT3_I(inode)->i_block_alloc_info = NULL;
 537         if (unlikely(rsv))
 538                 kfree(rsv);
 539 }
 540
 541 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 542 {
 543 #if defined(CONFIG_QUOTA)
 544         struct ext3_sb_info *sbi = EXT3_SB(sb);
 545
 546         if (sbi->s_jquota_fmt) {
 547                 char *fmtname = "";
 548
 549                 switch (sbi->s_jquota_fmt) {
 550                 case QFMT_VFS_OLD:
 551                         fmtname = "vfsold";
 552                         break;
 553                 case QFMT_VFS_V0:
 554                         fmtname = "vfsv0";
 555                         break;
 556                 case QFMT_VFS_V1:
 557                         fmtname = "vfsv1";
 558                         break;
 559                 }
 560                 seq_printf(seq, ",jqfmt=%s", fmtname);
 561         }
 562
 563         if (sbi->s_qf_names[USRQUOTA])
 564                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 565
 566         if (sbi->s_qf_names[GRPQUOTA])
 567                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 568
 569         if (test_opt(sb, USRQUOTA))
 570                 seq_puts(seq, ",usrquota");
 571
 572         if (test_opt(sb, GRPQUOTA))
 573                 seq_puts(seq, ",grpquota");
 574 #endif
 575 }
 576
 577 static char *data_mode_string(unsigned long mode)
 578 {
 579         switch (mode) {
 580         case EXT3_MOUNT_JOURNAL_DATA:
 581                 return "journal";
 582         case EXT3_MOUNT_ORDERED_DATA:
 583                 return "ordered";
 584         case EXT3_MOUNT_WRITEBACK_DATA:
 585                 return "writeback";
 586         }
 587         return "unknown";
 588 }
 589
 590 /*
 591  * Show an option if
 592  *  - it's set to a non-default value OR
 593  *  - if the per-sb default is different from the global default
 594  */
 595 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 596 {
 597         struct super_block *sb = vfs->mnt_sb;
 598         struct ext3_sb_info *sbi = EXT3_SB(sb);
 599         struct ext3_super_block *es = sbi->s_es;
 600         unsigned long def_mount_opts;
 601
 602         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 603
 604         if (sbi->s_sb_block != 1)
 605                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 606         if (test_opt(sb, MINIX_DF))
 607                 seq_puts(seq, ",minixdf");
 608         if (test_opt(sb, GRPID))
 609                 seq_puts(seq, ",grpid");
 610         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 611                 seq_puts(seq, ",nogrpid");
 612         if (sbi->s_resuid != EXT3_DEF_RESUID ||
 613             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 614                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 615         }
 616         if (sbi->s_resgid != EXT3_DEF_RESGID ||
 617             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 618                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 619         }
 620         if (test_opt(sb, ERRORS_RO)) {
 621                 int def_errors = le16_to_cpu(es->s_errors);
 622
 623                 if (def_errors == EXT3_ERRORS_PANIC ||
 624                     def_errors == EXT3_ERRORS_CONTINUE) {
 625                         seq_puts(seq, ",errors=remount-ro");
 626                 }
 627         }
 628         if (test_opt(sb, ERRORS_CONT))
 629                 seq_puts(seq, ",errors=continue");
 630         if (test_opt(sb, ERRORS_PANIC))
 631                 seq_puts(seq, ",errors=panic");
 632         if (test_opt(sb, NO_UID32))
 633                 seq_puts(seq, ",nouid32");
 634         if (test_opt(sb, DEBUG))
 635                 seq_puts(seq, ",debug");
 636         if (test_opt(sb, OLDALLOC))
 637                 seq_puts(seq, ",oldalloc");
 638 #ifdef CONFIG_EXT3_FS_XATTR
 639         if (test_opt(sb, XATTR_USER))
 640                 seq_puts(seq, ",user_xattr");
 641         if (!test_opt(sb, XATTR_USER) &&
 642             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 643                 seq_puts(seq, ",nouser_xattr");
 644         }
 645 #endif
 646 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 647         if (test_opt(sb, POSIX_ACL))
 648                 seq_puts(seq, ",acl");
 649         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 650                 seq_puts(seq, ",noacl");
 651 #endif
 652         if (!test_opt(sb, RESERVATION))
 653                 seq_puts(seq, ",noreservation");
 654         if (sbi->s_commit_interval) {
 655                 seq_printf(seq, ",commit=%u",
 656                            (unsigned) (sbi->s_commit_interval / HZ));
 657         }
 658
 659         /*
 660          * Always display barrier state so it's clear what the status is.
 661          */
 662         seq_puts(seq, ",barrier=");
 663         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 664         if (test_opt(sb, NOBH))
 665                 seq_puts(seq, ",nobh");
 666
 667         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 668         if (test_opt(sb, DATA_ERR_ABORT))
 669                 seq_puts(seq, ",data_err=abort");
 670
 671         if (test_opt(sb, NOLOAD))
 672                 seq_puts(seq, ",norecovery");
 673
 674         ext3_show_quota_options(seq, sb);
 675
 676         return 0;
 677 }
 678
 679
 680 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 681                 u64 ino, u32 generation)
 682 {
 683         struct inode *inode;
 684
 685         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 686                 return ERR_PTR(-ESTALE);
 687         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 688                 return ERR_PTR(-ESTALE);
 689
 690         /* iget isn't really right if the inode is currently unallocated!!
 691          *
 692          * ext3_read_inode will return a bad_inode if the inode had been
 693          * deleted, so we should be safe.
 694          *
 695          * Currently we don't know the generation for parent directory, so
 696          * a generation of 0 means "accept any"
 697          */
 698         inode = ext3_iget(sb, ino);
 699         if (IS_ERR(inode))
 700                 return ERR_CAST(inode);
 701         if (generation && inode->i_generation != generation) {
 702                 iput(inode);
 703                 return ERR_PTR(-ESTALE);
 704         }
 705
 706         return inode;
 707 }
 708
 709 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 710                 int fh_len, int fh_type)
 711 {
 712         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 713                                     ext3_nfs_get_inode);
 714 }
 715
 716 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 717                 int fh_len, int fh_type)
 718 {
 719         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 720                                     ext3_nfs_get_inode);
 721 }
 722
 723 /*
 724  * Try to release metadata pages (indirect blocks, directories) which are
 725  * mapped via the block device.  Since these pages could have journal heads
 726  * which would prevent try_to_free_buffers() from freeing them, we must use
 727  * jbd layer's try_to_free_buffers() function to release them.
 728  */
 729 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 730                                  gfp_t wait)
 731 {
 732         journal_t *journal = EXT3_SB(sb)->s_journal;
 733
 734         WARN_ON(PageChecked(page));
 735         if (!page_has_buffers(page))
 736                 return 0;
 737         if (journal)
 738                 return journal_try_to_free_buffers(journal, page,
 739                                                    wait & ~__GFP_WAIT);
 740         return try_to_free_buffers(page);
 741 }
 742
 743 #ifdef CONFIG_QUOTA
 744 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 745 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 746
 747 static int ext3_write_dquot(struct dquot *dquot);
 748 static int ext3_acquire_dquot(struct dquot *dquot);
 749 static int ext3_release_dquot(struct dquot *dquot);
 750 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 751 static int ext3_write_info(struct super_block *sb, int type);
 752 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 753                                 char *path);
 754 static int ext3_quota_on_mount(struct super_block *sb, int type);
 755 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 756                                size_t len, loff_t off);
 757 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 758                                 const char *data, size_t len, loff_t off);
 759
 760 static const struct dquot_operations ext3_quota_operations = {
 761         .write_dquot    = ext3_write_dquot,
 762         .acquire_dquot  = ext3_acquire_dquot,
 763         .release_dquot  = ext3_release_dquot,
 764         .mark_dirty     = ext3_mark_dquot_dirty,
 765         .write_info     = ext3_write_info,
 766         .alloc_dquot    = dquot_alloc,
 767         .destroy_dquot  = dquot_destroy,
 768 };
 769
 770 static const struct quotactl_ops ext3_qctl_operations = {
 771         .quota_on       = ext3_quota_on,
 772         .quota_off      = dquot_quota_off,
 773         .quota_sync     = dquot_quota_sync,
 774         .get_info       = dquot_get_dqinfo,
 775         .set_info       = dquot_set_dqinfo,
 776         .get_dqblk      = dquot_get_dqblk,
 777         .set_dqblk      = dquot_set_dqblk
 778 };
 779 #endif
 780
 781 static const struct super_operations ext3_sops = {
 782         .alloc_inode    = ext3_alloc_inode,
 783         .destroy_inode  = ext3_destroy_inode,
 784         .write_inode    = ext3_write_inode,
 785         .dirty_inode    = ext3_dirty_inode,
 786         .delete_inode   = ext3_delete_inode,
 787         .put_super      = ext3_put_super,
 788         .sync_fs        = ext3_sync_fs,
 789         .freeze_fs      = ext3_freeze,
 790         .unfreeze_fs    = ext3_unfreeze,
 791         .statfs         = ext3_statfs,
 792         .remount_fs     = ext3_remount,
 793         .clear_inode    = ext3_clear_inode,
 794         .show_options   = ext3_show_options,
 795 #ifdef CONFIG_QUOTA
 796         .quota_read     = ext3_quota_read,
 797         .quota_write    = ext3_quota_write,
 798 #endif
 799         .bdev_try_to_free_page = bdev_try_to_free_page,
 800 };
 801
 802 static const struct export_operations ext3_export_ops = {
 803         .fh_to_dentry = ext3_fh_to_dentry,
 804         .fh_to_parent = ext3_fh_to_parent,
 805         .get_parent = ext3_get_parent,
 806 };
 807
 808 enum {
 809         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 810         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 811         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 812         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 813         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 814         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 815         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 816         Opt_data_err_abort, Opt_data_err_ignore,
 817         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 818         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 819         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 820         Opt_resize, Opt_usrquota, Opt_grpquota
 821 };
 822
 823 static const match_table_t tokens = {
 824         {Opt_bsd_df, "bsddf"},
 825         {Opt_minix_df, "minixdf"},
 826         {Opt_grpid, "grpid"},
 827         {Opt_grpid, "bsdgroups"},
 828         {Opt_nogrpid, "nogrpid"},
 829         {Opt_nogrpid, "sysvgroups"},
 830         {Opt_resgid, "resgid=%u"},
 831         {Opt_resuid, "resuid=%u"},
 832         {Opt_sb, "sb=%u"},
 833         {Opt_err_cont, "errors=continue"},
 834         {Opt_err_panic, "errors=panic"},
 835         {Opt_err_ro, "errors=remount-ro"},
 836         {Opt_nouid32, "nouid32"},
 837         {Opt_nocheck, "nocheck"},
 838         {Opt_nocheck, "check=none"},
 839         {Opt_debug, "debug"},
 840         {Opt_oldalloc, "oldalloc"},
 841         {Opt_orlov, "orlov"},
 842         {Opt_user_xattr, "user_xattr"},
 843         {Opt_nouser_xattr, "nouser_xattr"},
 844         {Opt_acl, "acl"},
 845         {Opt_noacl, "noacl"},
 846         {Opt_reservation, "reservation"},
 847         {Opt_noreservation, "noreservation"},
 848         {Opt_noload, "noload"},
 849         {Opt_noload, "norecovery"},
 850         {Opt_nobh, "nobh"},
 851         {Opt_bh, "bh"},
 852         {Opt_commit, "commit=%u"},
 853         {Opt_journal_update, "journal=update"},
 854         {Opt_journal_inum, "journal=%u"},
 855         {Opt_journal_dev, "journal_dev=%u"},
 856         {Opt_abort, "abort"},
 857         {Opt_data_journal, "data=journal"},
 858         {Opt_data_ordered, "data=ordered"},
 859         {Opt_data_writeback, "data=writeback"},
 860         {Opt_data_err_abort, "data_err=abort"},
 861         {Opt_data_err_ignore, "data_err=ignore"},
 862         {Opt_offusrjquota, "usrjquota="},
 863         {Opt_usrjquota, "usrjquota=%s"},
 864         {Opt_offgrpjquota, "grpjquota="},
 865         {Opt_grpjquota, "grpjquota=%s"},
 866         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 867         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 868         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 869         {Opt_grpquota, "grpquota"},
 870         {Opt_noquota, "noquota"},
 871         {Opt_quota, "quota"},
 872         {Opt_usrquota, "usrquota"},
 873         {Opt_barrier, "barrier=%u"},
 874         {Opt_barrier, "barrier"},
 875         {Opt_nobarrier, "nobarrier"},
 876         {Opt_resize, "resize"},
 877         {Opt_err, NULL},
 878 };
 879
 880 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 881 {
 882         ext3_fsblk_t    sb_block;
 883         char            *options = (char *) *data;
 884
 885         if (!options || strncmp(options, "sb=", 3) != 0)
 886                 return 1;       /* Default location */
 887         options += 3;
 888         /*todo: use simple_strtoll with >32bit ext3 */
 889         sb_block = simple_strtoul(options, &options, 0);
 890         if (*options && *options != ',') {
 891                 ext3_msg(sb, "error: invalid sb specification: %s",
 892                        (char *) *data);
 893                 return 1;
 894         }
 895         if (*options == ',')
 896                 options++;
 897         *data = (void *) options;
 898         return sb_block;
 899 }
 900
 901 #ifdef CONFIG_QUOTA
 902 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 903 {
 904         struct ext3_sb_info *sbi = EXT3_SB(sb);
 905         char *qname;
 906
 907         if (sb_any_quota_loaded(sb) &&
 908                 !sbi->s_qf_names[qtype]) {
 909                 ext3_msg(sb, KERN_ERR,
 910                         "Cannot change journaled "
 911                         "quota options when quota turned on");
 912                 return 0;
 913         }
 914         qname = match_strdup(args);
 915         if (!qname) {
 916                 ext3_msg(sb, KERN_ERR,
 917                         "Not enough memory for storing quotafile name");
 918                 return 0;
 919         }
 920         if (sbi->s_qf_names[qtype] &&
 921                 strcmp(sbi->s_qf_names[qtype], qname)) {
 922                 ext3_msg(sb, KERN_ERR,
 923                         "%s quota file already specified", QTYPE2NAME(qtype));
 924                 kfree(qname);
 925                 return 0;
 926         }
 927         sbi->s_qf_names[qtype] = qname;
 928         if (strchr(sbi->s_qf_names[qtype], '/')) {
 929                 ext3_msg(sb, KERN_ERR,
 930                         "quotafile must be on filesystem root");
 931                 kfree(sbi->s_qf_names[qtype]);
 932                 sbi->s_qf_names[qtype] = NULL;
 933                 return 0;
 934         }
 935         set_opt(sbi->s_mount_opt, QUOTA);
 936         return 1;
 937 }
 938
 939 static int clear_qf_name(struct super_block *sb, int qtype) {
 940
 941         struct ext3_sb_info *sbi = EXT3_SB(sb);
 942
 943         if (sb_any_quota_loaded(sb) &&
 944                 sbi->s_qf_names[qtype]) {
 945                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 946                         " when quota turned on");
 947                 return 0;
 948         }
 949         /*
 950          * The space will be released later when all options are confirmed
 951          * to be correct
 952          */
 953         sbi->s_qf_names[qtype] = NULL;
 954         return 1;
 955 }
 956 #endif
 957
 958 static int parse_options (char *options, struct super_block *sb,
 959                           unsigned int *inum, unsigned long *journal_devnum,
 960                           ext3_fsblk_t *n_blocks_count, int is_remount)
 961 {
 962         struct ext3_sb_info *sbi = EXT3_SB(sb);
 963         char * p;
 964         substring_t args[MAX_OPT_ARGS];
 965         int data_opt = 0;
 966         int option;
 967 #ifdef CONFIG_QUOTA
 968         int qfmt;
 969 #endif
 970
 971         if (!options)
 972                 return 1;
 973
 974         while ((p = strsep (&options, ",")) != NULL) {
 975                 int token;
 976                 if (!*p)
 977                         continue;
 978                 /*
 979                  * Initialize args struct so we know whether arg was
 980                  * found; some options take optional arguments.
 981                  */
 982                 args[0].to = args[0].from = 0;
 983                 token = match_token(p, tokens, args);
 984                 switch (token) {
 985                 case Opt_bsd_df:
 986                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 987                         break;
 988                 case Opt_minix_df:
 989                         set_opt (sbi->s_mount_opt, MINIX_DF);
 990                         break;
 991                 case Opt_grpid:
 992                         set_opt (sbi->s_mount_opt, GRPID);
 993                         break;
 994                 case Opt_nogrpid:
 995                         clear_opt (sbi->s_mount_opt, GRPID);
 996                         break;
 997                 case Opt_resuid:
 998                         if (match_int(&args[0], &option))
 999                                 return 0;
1000                         sbi->s_resuid = option;
1001                         break;
1002                 case Opt_resgid:
1003                         if (match_int(&args[0], &option))
1004                                 return 0;
1005                         sbi->s_resgid = option;
1006                         break;
1007                 case Opt_sb:
1008                         /* handled by get_sb_block() instead of here */
1009                         /* *sb_block = match_int(&args[0]); */
1010                         break;
1011                 case Opt_err_panic:
1012                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1013                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1014                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1015                         break;
1016                 case Opt_err_ro:
1017                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1018                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1019                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1020                         break;
1021                 case Opt_err_cont:
1022                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1023                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1024                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1025                         break;
1026                 case Opt_nouid32:
1027                         set_opt (sbi->s_mount_opt, NO_UID32);
1028                         break;
1029                 case Opt_nocheck:
1030                         clear_opt (sbi->s_mount_opt, CHECK);
1031                         break;
1032                 case Opt_debug:
1033                         set_opt (sbi->s_mount_opt, DEBUG);
1034                         break;
1035                 case Opt_oldalloc:
1036                         set_opt (sbi->s_mount_opt, OLDALLOC);
1037                         break;
1038                 case Opt_orlov:
1039                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1040                         break;
1041 #ifdef CONFIG_EXT3_FS_XATTR
1042                 case Opt_user_xattr:
1043                         set_opt (sbi->s_mount_opt, XATTR_USER);
1044                         break;
1045                 case Opt_nouser_xattr:
1046                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1047                         break;
1048 #else
1049                 case Opt_user_xattr:
1050                 case Opt_nouser_xattr:
1051                         ext3_msg(sb, KERN_INFO,
1052                                 "(no)user_xattr options not supported");
1053                         break;
1054 #endif
1055 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1056                 case Opt_acl:
1057                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1058                         break;
1059                 case Opt_noacl:
1060                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1061                         break;
1062 #else
1063                 case Opt_acl:
1064                 case Opt_noacl:
1065                         ext3_msg(sb, KERN_INFO,
1066                                 "(no)acl options not supported");
1067                         break;
1068 #endif
1069                 case Opt_reservation:
1070                         set_opt(sbi->s_mount_opt, RESERVATION);
1071                         break;
1072                 case Opt_noreservation:
1073                         clear_opt(sbi->s_mount_opt, RESERVATION);
1074                         break;
1075                 case Opt_journal_update:
1076                         /* @@@ FIXME */
1077                         /* Eventually we will want to be able to create
1078                            a journal file here.  For now, only allow the
1079                            user to specify an existing inode to be the
1080                            journal file. */
1081                         if (is_remount) {
1082                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1083                                         "journal on remount");
1084                                 return 0;
1085                         }
1086                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1087                         break;
1088                 case Opt_journal_inum:
1089                         if (is_remount) {
1090                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1091                                        "journal on remount");
1092                                 return 0;
1093                         }
1094                         if (match_int(&args[0], &option))
1095                                 return 0;
1096                         *inum = option;
1097                         break;
1098                 case Opt_journal_dev:
1099                         if (is_remount) {
1100                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1101                                        "journal on remount");
1102                                 return 0;
1103                         }
1104                         if (match_int(&args[0], &option))
1105                                 return 0;
1106                         *journal_devnum = option;
1107                         break;
1108                 case Opt_noload:
1109                         set_opt (sbi->s_mount_opt, NOLOAD);
1110                         break;
1111                 case Opt_commit:
1112                         if (match_int(&args[0], &option))
1113                                 return 0;
1114                         if (option < 0)
1115                                 return 0;
1116                         if (option == 0)
1117                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1118                         sbi->s_commit_interval = HZ * option;
1119                         break;
1120                 case Opt_data_journal:
1121                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1122                         goto datacheck;
1123                 case Opt_data_ordered:
1124                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1125                         goto datacheck;
1126                 case Opt_data_writeback:
1127                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1128                 datacheck:
1129                         if (is_remount) {
1130                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1131                                         break;
1132                                 ext3_msg(sb, KERN_ERR,
1133                                         "error: cannot change "
1134                                         "data mode on remount. The filesystem "
1135                                         "is mounted in data=%s mode and you "
1136                                         "try to remount it in data=%s mode.",
1137                                         data_mode_string(test_opt(sb,
1138                                                         DATA_FLAGS)),
1139                                         data_mode_string(data_opt));
1140                                 return 0;
1141                         } else {
1142                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1143                                 sbi->s_mount_opt |= data_opt;
1144                         }
1145                         break;
1146                 case Opt_data_err_abort:
1147                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1148                         break;
1149                 case Opt_data_err_ignore:
1150                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1151                         break;
1152 #ifdef CONFIG_QUOTA
1153                 case Opt_usrjquota:
1154                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1155                                 return 0;
1156                         break;
1157                 case Opt_grpjquota:
1158                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1159                                 return 0;
1160                         break;
1161                 case Opt_offusrjquota:
1162                         if (!clear_qf_name(sb, USRQUOTA))
1163                                 return 0;
1164                         break;
1165                 case Opt_offgrpjquota:
1166                         if (!clear_qf_name(sb, GRPQUOTA))
1167                                 return 0;
1168                         break;
1169                 case Opt_jqfmt_vfsold:
1170                         qfmt = QFMT_VFS_OLD;
1171                         goto set_qf_format;
1172                 case Opt_jqfmt_vfsv0:
1173                         qfmt = QFMT_VFS_V0;
1174                         goto set_qf_format;
1175                 case Opt_jqfmt_vfsv1:
1176                         qfmt = QFMT_VFS_V1;
1177 set_qf_format:
1178                         if (sb_any_quota_loaded(sb) &&
1179                             sbi->s_jquota_fmt != qfmt) {
1180                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1181                                         "journaled quota options when "
1182                                         "quota turned on.");
1183                                 return 0;
1184                         }
1185                         sbi->s_jquota_fmt = qfmt;
1186                         break;
1187                 case Opt_quota:
1188                 case Opt_usrquota:
1189                         set_opt(sbi->s_mount_opt, QUOTA);
1190                         set_opt(sbi->s_mount_opt, USRQUOTA);
1191                         break;
1192                 case Opt_grpquota:
1193                         set_opt(sbi->s_mount_opt, QUOTA);
1194                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1195                         break;
1196                 case Opt_noquota:
1197                         if (sb_any_quota_loaded(sb)) {
1198                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1199                                         "quota options when quota turned on.");
1200                                 return 0;
1201                         }
1202                         clear_opt(sbi->s_mount_opt, QUOTA);
1203                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1204                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1205                         break;
1206 #else
1207                 case Opt_quota:
1208                 case Opt_usrquota:
1209                 case Opt_grpquota:
1210                         ext3_msg(sb, KERN_ERR,
1211                                 "error: quota options not supported.");
1212                         break;
1213                 case Opt_usrjquota:
1214                 case Opt_grpjquota:
1215                 case Opt_offusrjquota:
1216                 case Opt_offgrpjquota:
1217                 case Opt_jqfmt_vfsold:
1218                 case Opt_jqfmt_vfsv0:
1219                 case Opt_jqfmt_vfsv1:
1220                         ext3_msg(sb, KERN_ERR,
1221                                 "error: journaled quota options not "
1222                                 "supported.");
1223                         break;
1224                 case Opt_noquota:
1225                         break;
1226 #endif
1227                 case Opt_abort:
1228                         set_opt(sbi->s_mount_opt, ABORT);
1229                         break;
1230                 case Opt_nobarrier:
1231                         clear_opt(sbi->s_mount_opt, BARRIER);
1232                         break;
1233                 case Opt_barrier:
1234                         if (args[0].from) {
1235                                 if (match_int(&args[0], &option))
1236                                         return 0;
1237                         } else
1238                                 option = 1;     /* No argument, default to 1 */
1239                         if (option)
1240                                 set_opt(sbi->s_mount_opt, BARRIER);
1241                         else
1242                                 clear_opt(sbi->s_mount_opt, BARRIER);
1243                         break;
1244                 case Opt_ignore:
1245                         break;
1246                 case Opt_resize:
1247                         if (!is_remount) {
1248                                 ext3_msg(sb, KERN_ERR,
1249                                         "error: resize option only available "
1250                                         "for remount");
1251                                 return 0;
1252                         }
1253                         if (match_int(&args[0], &option) != 0)
1254                                 return 0;
1255                         *n_blocks_count = option;
1256                         break;
1257                 case Opt_nobh:
1258                         set_opt(sbi->s_mount_opt, NOBH);
1259                         break;
1260                 case Opt_bh:
1261                         clear_opt(sbi->s_mount_opt, NOBH);
1262                         break;
1263                 default:
1264                         ext3_msg(sb, KERN_ERR,
1265                                 "error: unrecognized mount option \"%s\" "
1266                                 "or missing value", p);
1267                         return 0;
1268                 }
1269         }
1270 #ifdef CONFIG_QUOTA
1271         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1272                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1273                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1274                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1275                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1276
1277                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1278                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1279                                         "format mixing.");
1280                         return 0;
1281                 }
1282
1283                 if (!sbi->s_jquota_fmt) {
1284                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1285                                         "not specified.");
1286                         return 0;
1287                 }
1288         } else {
1289                 if (sbi->s_jquota_fmt) {
1290                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1291                                         "specified with no journaling "
1292                                         "enabled.");
1293                         return 0;
1294                 }
1295         }
1296 #endif
1297         return 1;
1298 }
1299
1300 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1301                             int read_only)
1302 {
1303         struct ext3_sb_info *sbi = EXT3_SB(sb);
1304         int res = 0;
1305
1306         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1307                 ext3_msg(sb, KERN_ERR,
1308                         "error: revision level too high, "
1309                         "forcing read-only mode");
1310                 res = MS_RDONLY;
1311         }
1312         if (read_only)
1313                 return res;
1314         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1315                 ext3_msg(sb, KERN_WARNING,
1316                         "warning: mounting unchecked fs, "
1317                         "running e2fsck is recommended");
1318         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1319                 ext3_msg(sb, KERN_WARNING,
1320                         "warning: mounting fs with errors, "
1321                         "running e2fsck is recommended");
1322         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1323                  le16_to_cpu(es->s_mnt_count) >=
1324                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1325                 ext3_msg(sb, KERN_WARNING,
1326                         "warning: maximal mount count reached, "
1327                         "running e2fsck is recommended");
1328         else if (le32_to_cpu(es->s_checkinterval) &&
1329                 (le32_to_cpu(es->s_lastcheck) +
1330                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1331                 ext3_msg(sb, KERN_WARNING,
1332                         "warning: checktime reached, "
1333                         "running e2fsck is recommended");
1334 #if 0
1335                 /* @@@ We _will_ want to clear the valid bit if we find
1336                    inconsistencies, to force a fsck at reboot.  But for
1337                    a plain journaled filesystem we can keep it set as
1338                    valid forever! :) */
1339         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1340 #endif
1341         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1342                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1343         le16_add_cpu(&es->s_mnt_count, 1);
1344         es->s_mtime = cpu_to_le32(get_seconds());
1345         ext3_update_dynamic_rev(sb);
1346         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1347
1348         ext3_commit_super(sb, es, 1);
1349         if (test_opt(sb, DEBUG))
1350                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1351                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1352                         sb->s_blocksize,
1353                         sbi->s_groups_count,
1354                         EXT3_BLOCKS_PER_GROUP(sb),
1355                         EXT3_INODES_PER_GROUP(sb),
1356                         sbi->s_mount_opt);
1357
1358         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1359                 char b[BDEVNAME_SIZE];
1360                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1361                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1362         } else {
1363                 ext3_msg(sb, KERN_INFO, "using internal journal");
1364         }
1365         return res;
1366 }
1367
1368 /* Called at mount-time, super-block is locked */
1369 static int ext3_check_descriptors(struct super_block *sb)
1370 {
1371         struct ext3_sb_info *sbi = EXT3_SB(sb);
1372         int i;
1373
1374         ext3_debug ("Checking group descriptors");
1375
1376         for (i = 0; i < sbi->s_groups_count; i++) {
1377                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1378                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1379                 ext3_fsblk_t last_block;
1380
1381                 if (i == sbi->s_groups_count - 1)
1382                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1383                 else
1384                         last_block = first_block +
1385                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1386
1387                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1388                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1389                 {
1390                         ext3_error (sb, "ext3_check_descriptors",
1391                                     "Block bitmap for group %d"
1392                                     " not in group (block %lu)!",
1393                                     i, (unsigned long)
1394                                         le32_to_cpu(gdp->bg_block_bitmap));
1395                         return 0;
1396                 }
1397                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1398                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1399                 {
1400                         ext3_error (sb, "ext3_check_descriptors",
1401                                     "Inode bitmap for group %d"
1402                                     " not in group (block %lu)!",
1403                                     i, (unsigned long)
1404                                         le32_to_cpu(gdp->bg_inode_bitmap));
1405                         return 0;
1406                 }
1407                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1408                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1409                     last_block)
1410                 {
1411                         ext3_error (sb, "ext3_check_descriptors",
1412                                     "Inode table for group %d"
1413                                     " not in group (block %lu)!",
1414                                     i, (unsigned long)
1415                                         le32_to_cpu(gdp->bg_inode_table));
1416                         return 0;
1417                 }
1418         }
1419
1420         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1421         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1422         return 1;
1423 }
1424
1425
1426 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1427  * the superblock) which were deleted from all directories, but held open by
1428  * a process at the time of a crash.  We walk the list and try to delete these
1429  * inodes at recovery time (only with a read-write filesystem).
1430  *
1431  * In order to keep the orphan inode chain consistent during traversal (in
1432  * case of crash during recovery), we link each inode into the superblock
1433  * orphan list_head and handle it the same way as an inode deletion during
1434  * normal operation (which journals the operations for us).
1435  *
1436  * We only do an iget() and an iput() on each inode, which is very safe if we
1437  * accidentally point at an in-use or already deleted inode.  The worst that
1438  * can happen in this case is that we get a "bit already cleared" message from
1439  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1440  * e2fsck was run on this filesystem, and it must have already done the orphan
1441  * inode cleanup for us, so we can safely abort without any further action.
1442  */
1443 static void ext3_orphan_cleanup (struct super_block * sb,
1444                                  struct ext3_super_block * es)
1445 {
1446         unsigned int s_flags = sb->s_flags;
1447         int nr_orphans = 0, nr_truncates = 0;
1448 #ifdef CONFIG_QUOTA
1449         int i;
1450 #endif
1451         if (!es->s_last_orphan) {
1452                 jbd_debug(4, "no orphan inodes to clean up\n");
1453                 return;
1454         }
1455
1456         if (bdev_read_only(sb->s_bdev)) {
1457                 ext3_msg(sb, KERN_ERR, "error: write access "
1458                         "unavailable, skipping orphan cleanup.");
1459                 return;
1460         }
1461
1462         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1463                 if (es->s_last_orphan)
1464                         jbd_debug(1, "Errors on filesystem, "
1465                                   "clearing orphan list.\n");
1466                 es->s_last_orphan = 0;
1467                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1468                 return;
1469         }
1470
1471         if (s_flags & MS_RDONLY) {
1472                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1473                 sb->s_flags &= ~MS_RDONLY;
1474         }
1475 #ifdef CONFIG_QUOTA
1476         /* Needed for iput() to work correctly and not trash data */
1477         sb->s_flags |= MS_ACTIVE;
1478         /* Turn on quotas so that they are updated correctly */
1479         for (i = 0; i < MAXQUOTAS; i++) {
1480                 if (EXT3_SB(sb)->s_qf_names[i]) {
1481                         int ret = ext3_quota_on_mount(sb, i);
1482                         if (ret < 0)
1483                                 ext3_msg(sb, KERN_ERR,
1484                                         "error: cannot turn on journaled "
1485                                         "quota: %d", ret);
1486                 }
1487         }
1488 #endif
1489
1490         while (es->s_last_orphan) {
1491                 struct inode *inode;
1492
1493                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1494                 if (IS_ERR(inode)) {
1495                         es->s_last_orphan = 0;
1496                         break;
1497                 }
1498
1499                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1500                 dquot_initialize(inode);
1501                 if (inode->i_nlink) {
1502                         printk(KERN_DEBUG
1503                                 "%s: truncating inode %lu to %Ld bytes\n",
1504                                 __func__, inode->i_ino, inode->i_size);
1505                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1506                                   inode->i_ino, inode->i_size);
1507                         ext3_truncate(inode);
1508                         nr_truncates++;
1509                 } else {
1510                         printk(KERN_DEBUG
1511                                 "%s: deleting unreferenced inode %lu\n",
1512                                 __func__, inode->i_ino);
1513                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1514                                   inode->i_ino);
1515                         nr_orphans++;
1516                 }
1517                 iput(inode);  /* The delete magic happens here! */
1518         }
1519
1520 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1521
1522         if (nr_orphans)
1523                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1524                        PLURAL(nr_orphans));
1525         if (nr_truncates)
1526                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1527                        PLURAL(nr_truncates));
1528 #ifdef CONFIG_QUOTA
1529         /* Turn quotas off */
1530         for (i = 0; i < MAXQUOTAS; i++) {
1531                 if (sb_dqopt(sb)->files[i])
1532                         dquot_quota_off(sb, i);
1533         }
1534 #endif
1535         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1536 }
1537
1538 /*
1539  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1540  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1541  * We need to be 1 filesystem block less than the 2^32 sector limit.
1542  */
1543 static loff_t ext3_max_size(int bits)
1544 {
1545         loff_t res = EXT3_NDIR_BLOCKS;
1546         int meta_blocks;
1547         loff_t upper_limit;
1548
1549         /* This is calculated to be the largest file size for a
1550          * dense, file such that the total number of
1551          * sectors in the file, including data and all indirect blocks,
1552          * does not exceed 2^32 -1
1553          * __u32 i_blocks representing the total number of
1554          * 512 bytes blocks of the file
1555          */
1556         upper_limit = (1LL << 32) - 1;
1557
1558         /* total blocks in file system block size */
1559         upper_limit >>= (bits - 9);
1560
1561
1562         /* indirect blocks */
1563         meta_blocks = 1;
1564         /* double indirect blocks */
1565         meta_blocks += 1 + (1LL << (bits-2));
1566         /* tripple indirect blocks */
1567         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1568
1569         upper_limit -= meta_blocks;
1570         upper_limit <<= bits;
1571
1572         res += 1LL << (bits-2);
1573         res += 1LL << (2*(bits-2));
1574         res += 1LL << (3*(bits-2));
1575         res <<= bits;
1576         if (res > upper_limit)
1577                 res = upper_limit;
1578
1579         if (res > MAX_LFS_FILESIZE)
1580                 res = MAX_LFS_FILESIZE;
1581
1582         return res;
1583 }
1584
1585 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1586                                     ext3_fsblk_t logic_sb_block,
1587                                     int nr)
1588 {
1589         struct ext3_sb_info *sbi = EXT3_SB(sb);
1590         unsigned long bg, first_meta_bg;
1591         int has_super = 0;
1592
1593         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1594
1595         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1596             nr < first_meta_bg)
1597                 return (logic_sb_block + nr + 1);
1598         bg = sbi->s_desc_per_block * nr;
1599         if (ext3_bg_has_super(sb, bg))
1600                 has_super = 1;
1601         return (has_super + ext3_group_first_block_no(sb, bg));
1602 }
1603
1604
1605 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1606 {
1607         struct buffer_head * bh;
1608         struct ext3_super_block *es = NULL;
1609         struct ext3_sb_info *sbi;
1610         ext3_fsblk_t block;
1611         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1612         ext3_fsblk_t logic_sb_block;
1613         unsigned long offset = 0;
1614         unsigned int journal_inum = 0;
1615         unsigned long journal_devnum = 0;
1616         unsigned long def_mount_opts;
1617         struct inode *root;
1618         int blocksize;
1619         int hblock;
1620         int db_count;
1621         int i;
1622         int needs_recovery;
1623         int ret = -EINVAL;
1624         __le32 features;
1625         int err;
1626
1627         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1628         if (!sbi)
1629                 return -ENOMEM;
1630
1631         sbi->s_blockgroup_lock =
1632                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1633         if (!sbi->s_blockgroup_lock) {
1634                 kfree(sbi);
1635                 return -ENOMEM;
1636         }
1637         sb->s_fs_info = sbi;
1638         sbi->s_mount_opt = 0;
1639         sbi->s_resuid = EXT3_DEF_RESUID;
1640         sbi->s_resgid = EXT3_DEF_RESGID;
1641         sbi->s_sb_block = sb_block;
1642
1643         unlock_kernel();
1644
1645         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1646         if (!blocksize) {
1647                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1648                 goto out_fail;
1649         }
1650
1651         /*
1652          * The ext3 superblock will not be buffer aligned for other than 1kB
1653          * block sizes.  We need to calculate the offset from buffer start.
1654          */
1655         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1656                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1657                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1658         } else {
1659                 logic_sb_block = sb_block;
1660         }
1661
1662         if (!(bh = sb_bread(sb, logic_sb_block))) {
1663                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1664                 goto out_fail;
1665         }
1666         /*
1667          * Note: s_es must be initialized as soon as possible because
1668          *       some ext3 macro-instructions depend on its value
1669          */
1670         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1671         sbi->s_es = es;
1672         sb->s_magic = le16_to_cpu(es->s_magic);
1673         if (sb->s_magic != EXT3_SUPER_MAGIC)
1674                 goto cantfind_ext3;
1675
1676         /* Set defaults before we parse the mount options */
1677         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1678         if (def_mount_opts & EXT3_DEFM_DEBUG)
1679                 set_opt(sbi->s_mount_opt, DEBUG);
1680         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1681                 set_opt(sbi->s_mount_opt, GRPID);
1682         if (def_mount_opts & EXT3_DEFM_UID16)
1683                 set_opt(sbi->s_mount_opt, NO_UID32);
1684 #ifdef CONFIG_EXT3_FS_XATTR
1685         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1686                 set_opt(sbi->s_mount_opt, XATTR_USER);
1687 #endif
1688 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1689         if (def_mount_opts & EXT3_DEFM_ACL)
1690                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1691 #endif
1692         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1693                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1694         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1695                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1696         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1697                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1698
1699         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1700                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1701         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1702                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1703         else
1704                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1705
1706         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1707         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1708
1709         set_opt(sbi->s_mount_opt, RESERVATION);
1710
1711         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1712                             NULL, 0))
1713                 goto failed_mount;
1714
1715         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1716                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1717
1718         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1719             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1720              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1721              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1722                 ext3_msg(sb, KERN_WARNING,
1723                         "warning: feature flags set on rev 0 fs, "
1724                         "running e2fsck is recommended");
1725         /*
1726          * Check feature flags regardless of the revision level, since we
1727          * previously didn't change the revision level when setting the flags,
1728          * so there is a chance incompat flags are set on a rev 0 filesystem.
1729          */
1730         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1731         if (features) {
1732                 ext3_msg(sb, KERN_ERR,
1733                         "error: couldn't mount because of unsupported "
1734                         "optional features (%x)", le32_to_cpu(features));
1735                 goto failed_mount;
1736         }
1737         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1738         if (!(sb->s_flags & MS_RDONLY) && features) {
1739                 ext3_msg(sb, KERN_ERR,
1740                         "error: couldn't mount RDWR because of unsupported "
1741                         "optional features (%x)", le32_to_cpu(features));
1742                 goto failed_mount;
1743         }
1744         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1745
1746         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1747             blocksize > EXT3_MAX_BLOCK_SIZE) {
1748                 ext3_msg(sb, KERN_ERR,
1749                         "error: couldn't mount because of unsupported "
1750                         "filesystem blocksize %d", blocksize);
1751                 goto failed_mount;
1752         }
1753
1754         hblock = bdev_logical_block_size(sb->s_bdev);
1755         if (sb->s_blocksize != blocksize) {
1756                 /*
1757                  * Make sure the blocksize for the filesystem is larger
1758                  * than the hardware sectorsize for the machine.
1759                  */
1760                 if (blocksize < hblock) {
1761                         ext3_msg(sb, KERN_ERR,
1762                                 "error: fsblocksize %d too small for "
1763                                 "hardware sectorsize %d", blocksize, hblock);
1764                         goto failed_mount;
1765                 }
1766
1767                 brelse (bh);
1768                 if (!sb_set_blocksize(sb, blocksize)) {
1769                         ext3_msg(sb, KERN_ERR,
1770                                 "error: bad blocksize %d", blocksize);
1771                         goto out_fail;
1772                 }
1773                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1774                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1775                 bh = sb_bread(sb, logic_sb_block);
1776                 if (!bh) {
1777                         ext3_msg(sb, KERN_ERR,
1778                                "error: can't read superblock on 2nd try");
1779                         goto failed_mount;
1780                 }
1781                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1782                 sbi->s_es = es;
1783                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1784                         ext3_msg(sb, KERN_ERR,
1785                                 "error: magic mismatch");
1786                         goto failed_mount;
1787                 }
1788         }
1789
1790         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1791
1792         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1793                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1794                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1795         } else {
1796                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1797                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1798                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1799                     (!is_power_of_2(sbi->s_inode_size)) ||
1800                     (sbi->s_inode_size > blocksize)) {
1801                         ext3_msg(sb, KERN_ERR,
1802                                 "error: unsupported inode size: %d",
1803                                 sbi->s_inode_size);
1804                         goto failed_mount;
1805                 }
1806         }
1807         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1808                                    le32_to_cpu(es->s_log_frag_size);
1809         if (blocksize != sbi->s_frag_size) {
1810                 ext3_msg(sb, KERN_ERR,
1811                        "error: fragsize %lu != blocksize %u (unsupported)",
1812                        sbi->s_frag_size, blocksize);
1813                 goto failed_mount;
1814         }
1815         sbi->s_frags_per_block = 1;
1816         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1817         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1818         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1819         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1820                 goto cantfind_ext3;
1821         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1822         if (sbi->s_inodes_per_block == 0)
1823                 goto cantfind_ext3;
1824         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1825                                         sbi->s_inodes_per_block;
1826         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1827         sbi->s_sbh = bh;
1828         sbi->s_mount_state = le16_to_cpu(es->s_state);
1829         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1830         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1831         for (i=0; i < 4; i++)
1832                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1833         sbi->s_def_hash_version = es->s_def_hash_version;
1834         i = le32_to_cpu(es->s_flags);
1835         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1836                 sbi->s_hash_unsigned = 3;
1837         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1838 #ifdef __CHAR_UNSIGNED__
1839                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1840                 sbi->s_hash_unsigned = 3;
1841 #else
1842                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1843 #endif
1844         }
1845
1846         if (sbi->s_blocks_per_group > blocksize * 8) {
1847                 ext3_msg(sb, KERN_ERR,
1848                         "#blocks per group too big: %lu",
1849                         sbi->s_blocks_per_group);
1850                 goto failed_mount;
1851         }
1852         if (sbi->s_frags_per_group > blocksize * 8) {
1853                 ext3_msg(sb, KERN_ERR,
1854                         "error: #fragments per group too big: %lu",
1855                         sbi->s_frags_per_group);
1856                 goto failed_mount;
1857         }
1858         if (sbi->s_inodes_per_group > blocksize * 8) {
1859                 ext3_msg(sb, KERN_ERR,
1860                         "error: #inodes per group too big: %lu",
1861                         sbi->s_inodes_per_group);
1862                 goto failed_mount;
1863         }
1864
1865         if (le32_to_cpu(es->s_blocks_count) >
1866                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1867                 ext3_msg(sb, KERN_ERR,
1868                         "error: filesystem is too large to mount safely");
1869                 if (sizeof(sector_t) < 8)
1870                         ext3_msg(sb, KERN_ERR,
1871                                 "error: CONFIG_LBDAF not enabled");
1872                 goto failed_mount;
1873         }
1874
1875         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1876                 goto cantfind_ext3;
1877         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1878                                le32_to_cpu(es->s_first_data_block) - 1)
1879                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1880         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1881                    EXT3_DESC_PER_BLOCK(sb);
1882         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1883                                     GFP_KERNEL);
1884         if (sbi->s_group_desc == NULL) {
1885                 ext3_msg(sb, KERN_ERR,
1886                         "error: not enough memory");
1887                 goto failed_mount;
1888         }
1889
1890         bgl_lock_init(sbi->s_blockgroup_lock);
1891
1892         for (i = 0; i < db_count; i++) {
1893                 block = descriptor_loc(sb, logic_sb_block, i);
1894                 sbi->s_group_desc[i] = sb_bread(sb, block);
1895                 if (!sbi->s_group_desc[i]) {
1896                         ext3_msg(sb, KERN_ERR,
1897                                 "error: can't read group descriptor %d", i);
1898                         db_count = i;
1899                         goto failed_mount2;
1900                 }
1901         }
1902         if (!ext3_check_descriptors (sb)) {
1903                 ext3_msg(sb, KERN_ERR,
1904                         "error: group descriptors corrupted");
1905                 goto failed_mount2;
1906         }
1907         sbi->s_gdb_count = db_count;
1908         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1909         spin_lock_init(&sbi->s_next_gen_lock);
1910
1911         /* per fileystem reservation list head & lock */
1912         spin_lock_init(&sbi->s_rsv_window_lock);
1913         sbi->s_rsv_window_root = RB_ROOT;
1914         /* Add a single, static dummy reservation to the start of the
1915          * reservation window list --- it gives us a placeholder for
1916          * append-at-start-of-list which makes the allocation logic
1917          * _much_ simpler. */
1918         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1919         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1920         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1921         sbi->s_rsv_window_head.rsv_goal_size = 0;
1922         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1923
1924         /*
1925          * set up enough so that it can read an inode
1926          */
1927         sb->s_op = &ext3_sops;
1928         sb->s_export_op = &ext3_export_ops;
1929         sb->s_xattr = ext3_xattr_handlers;
1930 #ifdef CONFIG_QUOTA
1931         sb->s_qcop = &ext3_qctl_operations;
1932         sb->dq_op = &ext3_quota_operations;
1933 #endif
1934         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1935         mutex_init(&sbi->s_orphan_lock);
1936         mutex_init(&sbi->s_resize_lock);
1937
1938         sb->s_root = NULL;
1939
1940         needs_recovery = (es->s_last_orphan != 0 ||
1941                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1942                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1943
1944         /*
1945          * The first inode we look at is the journal inode.  Don't try
1946          * root first: it may be modified in the journal!
1947          */
1948         if (!test_opt(sb, NOLOAD) &&
1949             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1950                 if (ext3_load_journal(sb, es, journal_devnum))
1951                         goto failed_mount2;
1952         } else if (journal_inum) {
1953                 if (ext3_create_journal(sb, es, journal_inum))
1954                         goto failed_mount2;
1955         } else {
1956                 if (!silent)
1957                         ext3_msg(sb, KERN_ERR,
1958                                 "error: no journal found. "
1959                                 "mounting ext3 over ext2?");
1960                 goto failed_mount2;
1961         }
1962         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1963                         ext3_count_free_blocks(sb));
1964         if (!err) {
1965                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1966                                 ext3_count_free_inodes(sb));
1967         }
1968         if (!err) {
1969                 err = percpu_counter_init(&sbi->s_dirs_counter,
1970                                 ext3_count_dirs(sb));
1971         }
1972         if (err) {
1973                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1974                 goto failed_mount3;
1975         }
1976
1977         /* We have now updated the journal if required, so we can
1978          * validate the data journaling mode. */
1979         switch (test_opt(sb, DATA_FLAGS)) {
1980         case 0:
1981                 /* No mode set, assume a default based on the journal
1982                    capabilities: ORDERED_DATA if the journal can
1983                    cope, else JOURNAL_DATA */
1984                 if (journal_check_available_features
1985                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1986                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1987                 else
1988                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1989                 break;
1990
1991         case EXT3_MOUNT_ORDERED_DATA:
1992         case EXT3_MOUNT_WRITEBACK_DATA:
1993                 if (!journal_check_available_features
1994                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1995                         ext3_msg(sb, KERN_ERR,
1996                                 "error: journal does not support "
1997                                 "requested data journaling mode");
1998                         goto failed_mount3;
1999                 }
2000         default:
2001                 break;
2002         }
2003
2004         if (test_opt(sb, NOBH)) {
2005                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
2006                         ext3_msg(sb, KERN_WARNING,
2007                                 "warning: ignoring nobh option - "
2008                                 "it is supported only with writeback mode");
2009                         clear_opt(sbi->s_mount_opt, NOBH);
2010                 }
2011         }
2012         /*
2013          * The journal_load will have done any necessary log recovery,
2014          * so we can safely mount the rest of the filesystem now.
2015          */
2016
2017         root = ext3_iget(sb, EXT3_ROOT_INO);
2018         if (IS_ERR(root)) {
2019                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2020                 ret = PTR_ERR(root);
2021                 goto failed_mount3;
2022         }
2023         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2024                 iput(root);
2025                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2026                 goto failed_mount3;
2027         }
2028         sb->s_root = d_alloc_root(root);
2029         if (!sb->s_root) {
2030                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2031                 iput(root);
2032                 ret = -ENOMEM;
2033                 goto failed_mount3;
2034         }
2035
2036         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2037
2038         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2039         ext3_orphan_cleanup(sb, es);
2040         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2041         if (needs_recovery)
2042                 ext3_msg(sb, KERN_INFO, "recovery complete");
2043         ext3_mark_recovery_complete(sb, es);
2044         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2045                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2046                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2047                 "writeback");
2048
2049         lock_kernel();
2050         return 0;
2051
2052 cantfind_ext3:
2053         if (!silent)
2054                 ext3_msg(sb, KERN_INFO,
2055                         "error: can't find ext3 filesystem on dev %s.",
2056                        sb->s_id);
2057         goto failed_mount;
2058
2059 failed_mount3:
2060         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2061         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2062         percpu_counter_destroy(&sbi->s_dirs_counter);
2063         journal_destroy(sbi->s_journal);
2064 failed_mount2:
2065         for (i = 0; i < db_count; i++)
2066                 brelse(sbi->s_group_desc[i]);
2067         kfree(sbi->s_group_desc);
2068 failed_mount:
2069 #ifdef CONFIG_QUOTA
2070         for (i = 0; i < MAXQUOTAS; i++)
2071                 kfree(sbi->s_qf_names[i]);
2072 #endif
2073         ext3_blkdev_remove(sbi);
2074         brelse(bh);
2075 out_fail:
2076         sb->s_fs_info = NULL;
2077         kfree(sbi->s_blockgroup_lock);
2078         kfree(sbi);
2079         lock_kernel();
2080         return ret;
2081 }
2082
2083 /*
2084  * Setup any per-fs journal parameters now.  We'll do this both on
2085  * initial mount, once the journal has been initialised but before we've
2086  * done any recovery; and again on any subsequent remount.
2087  */
2088 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2089 {
2090         struct ext3_sb_info *sbi = EXT3_SB(sb);
2091
2092         if (sbi->s_commit_interval)
2093                 journal->j_commit_interval = sbi->s_commit_interval;
2094         /* We could also set up an ext3-specific default for the commit
2095          * interval here, but for now we'll just fall back to the jbd
2096          * default. */
2097
2098         spin_lock(&journal->j_state_lock);
2099         if (test_opt(sb, BARRIER))
2100                 journal->j_flags |= JFS_BARRIER;
2101         else
2102                 journal->j_flags &= ~JFS_BARRIER;
2103         if (test_opt(sb, DATA_ERR_ABORT))
2104                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2105         else
2106                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2107         spin_unlock(&journal->j_state_lock);
2108 }
2109
2110 static journal_t *ext3_get_journal(struct super_block *sb,
2111                                    unsigned int journal_inum)
2112 {
2113         struct inode *journal_inode;
2114         journal_t *journal;
2115
2116         /* First, test for the existence of a valid inode on disk.  Bad
2117          * things happen if we iget() an unused inode, as the subsequent
2118          * iput() will try to delete it. */
2119
2120         journal_inode = ext3_iget(sb, journal_inum);
2121         if (IS_ERR(journal_inode)) {
2122                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2123                 return NULL;
2124         }
2125         if (!journal_inode->i_nlink) {
2126                 make_bad_inode(journal_inode);
2127                 iput(journal_inode);
2128                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2129                 return NULL;
2130         }
2131
2132         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2133                   journal_inode, journal_inode->i_size);
2134         if (!S_ISREG(journal_inode->i_mode)) {
2135                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2136                 iput(journal_inode);
2137                 return NULL;
2138         }
2139
2140         journal = journal_init_inode(journal_inode);
2141         if (!journal) {
2142                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2143                 iput(journal_inode);
2144                 return NULL;
2145         }
2146         journal->j_private = sb;
2147         ext3_init_journal_params(sb, journal);
2148         return journal;
2149 }
2150
2151 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2152                                        dev_t j_dev)
2153 {
2154         struct buffer_head * bh;
2155         journal_t *journal;
2156         ext3_fsblk_t start;
2157         ext3_fsblk_t len;
2158         int hblock, blocksize;
2159         ext3_fsblk_t sb_block;
2160         unsigned long offset;
2161         struct ext3_super_block * es;
2162         struct block_device *bdev;
2163
2164         bdev = ext3_blkdev_get(j_dev, sb);
2165         if (bdev == NULL)
2166                 return NULL;
2167
2168         if (bd_claim(bdev, sb)) {
2169                 ext3_msg(sb, KERN_ERR,
2170                         "error: failed to claim external journal device");
2171                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2172                 return NULL;
2173         }
2174
2175         blocksize = sb->s_blocksize;
2176         hblock = bdev_logical_block_size(bdev);
2177         if (blocksize < hblock) {
2178                 ext3_msg(sb, KERN_ERR,
2179                         "error: blocksize too small for journal device");
2180                 goto out_bdev;
2181         }
2182
2183         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2184         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2185         set_blocksize(bdev, blocksize);
2186         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2187                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2188                         "external journal");
2189                 goto out_bdev;
2190         }
2191
2192         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2193         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2194             !(le32_to_cpu(es->s_feature_incompat) &
2195               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2196                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2197                         "bad superblock");
2198                 brelse(bh);
2199                 goto out_bdev;
2200         }
2201
2202         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2203                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2204                 brelse(bh);
2205                 goto out_bdev;
2206         }
2207
2208         len = le32_to_cpu(es->s_blocks_count);
2209         start = sb_block + 1;
2210         brelse(bh);     /* we're done with the superblock */
2211
2212         journal = journal_init_dev(bdev, sb->s_bdev,
2213                                         start, len, blocksize);
2214         if (!journal) {
2215                 ext3_msg(sb, KERN_ERR,
2216                         "error: failed to create device journal");
2217                 goto out_bdev;
2218         }
2219         journal->j_private = sb;
2220         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2221         wait_on_buffer(journal->j_sb_buffer);
2222         if (!buffer_uptodate(journal->j_sb_buffer)) {
2223                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2224                 goto out_journal;
2225         }
2226         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2227                 ext3_msg(sb, KERN_ERR,
2228                         "error: external journal has more than one "
2229                         "user (unsupported) - %d",
2230                         be32_to_cpu(journal->j_superblock->s_nr_users));
2231                 goto out_journal;
2232         }
2233         EXT3_SB(sb)->journal_bdev = bdev;
2234         ext3_init_journal_params(sb, journal);
2235         return journal;
2236 out_journal:
2237         journal_destroy(journal);
2238 out_bdev:
2239         ext3_blkdev_put(bdev);
2240         return NULL;
2241 }
2242
2243 static int ext3_load_journal(struct super_block *sb,
2244                              struct ext3_super_block *es,
2245                              unsigned long journal_devnum)
2246 {
2247         journal_t *journal;
2248         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2249         dev_t journal_dev;
2250         int err = 0;
2251         int really_read_only;
2252
2253         if (journal_devnum &&
2254             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2255                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2256                         "numbers have changed");
2257                 journal_dev = new_decode_dev(journal_devnum);
2258         } else
2259                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2260
2261         really_read_only = bdev_read_only(sb->s_bdev);
2262
2263         /*
2264          * Are we loading a blank journal or performing recovery after a
2265          * crash?  For recovery, we need to check in advance whether we
2266          * can get read-write access to the device.
2267          */
2268
2269         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2270                 if (sb->s_flags & MS_RDONLY) {
2271                         ext3_msg(sb, KERN_INFO,
2272                                 "recovery required on readonly filesystem");
2273                         if (really_read_only) {
2274                                 ext3_msg(sb, KERN_ERR, "error: write access "
2275                                         "unavailable, cannot proceed");
2276                                 return -EROFS;
2277                         }
2278                         ext3_msg(sb, KERN_INFO,
2279                                 "write access will be enabled during recovery");
2280                 }
2281         }
2282
2283         if (journal_inum && journal_dev) {
2284                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2285                        "and inode journals");
2286                 return -EINVAL;
2287         }
2288
2289         if (journal_inum) {
2290                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2291                         return -EINVAL;
2292         } else {
2293                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2294                         return -EINVAL;
2295         }
2296
2297         if (!(journal->j_flags & JFS_BARRIER))
2298                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2299
2300         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2301                 err = journal_update_format(journal);
2302                 if (err)  {
2303                         ext3_msg(sb, KERN_ERR, "error updating journal");
2304                         journal_destroy(journal);
2305                         return err;
2306                 }
2307         }
2308
2309         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2310                 err = journal_wipe(journal, !really_read_only);
2311         if (!err)
2312                 err = journal_load(journal);
2313
2314         if (err) {
2315                 ext3_msg(sb, KERN_ERR, "error loading journal");
2316                 journal_destroy(journal);
2317                 return err;
2318         }
2319
2320         EXT3_SB(sb)->s_journal = journal;
2321         ext3_clear_journal_err(sb, es);
2322
2323         if (journal_devnum &&
2324             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2325                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2326
2327                 /* Make sure we flush the recovery flag to disk. */
2328                 ext3_commit_super(sb, es, 1);
2329         }
2330
2331         return 0;
2332 }
2333
2334 static int ext3_create_journal(struct super_block *sb,
2335                                struct ext3_super_block *es,
2336                                unsigned int journal_inum)
2337 {
2338         journal_t *journal;
2339         int err;
2340
2341         if (sb->s_flags & MS_RDONLY) {
2342                 ext3_msg(sb, KERN_ERR,
2343                         "error: readonly filesystem when trying to "
2344                         "create journal");
2345                 return -EROFS;
2346         }
2347
2348         journal = ext3_get_journal(sb, journal_inum);
2349         if (!journal)
2350                 return -EINVAL;
2351
2352         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2353                journal_inum);
2354
2355         err = journal_create(journal);
2356         if (err) {
2357                 ext3_msg(sb, KERN_ERR, "error creating journal");
2358                 journal_destroy(journal);
2359                 return -EIO;
2360         }
2361
2362         EXT3_SB(sb)->s_journal = journal;
2363
2364         ext3_update_dynamic_rev(sb);
2365         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2366         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2367
2368         es->s_journal_inum = cpu_to_le32(journal_inum);
2369
2370         /* Make sure we flush the recovery flag to disk. */
2371         ext3_commit_super(sb, es, 1);
2372
2373         return 0;
2374 }
2375
2376 static int ext3_commit_super(struct super_block *sb,
2377                                struct ext3_super_block *es,
2378                                int sync)
2379 {
2380         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2381         int error = 0;
2382
2383         if (!sbh)
2384                 return error;
2385         /*
2386          * If the file system is mounted read-only, don't update the
2387          * superblock write time.  This avoids updating the superblock
2388          * write time when we are mounting the root file system
2389          * read/only but we need to replay the journal; at that point,
2390          * for people who are east of GMT and who make their clock
2391          * tick in localtime for Windows bug-for-bug compatibility,
2392          * the clock is set in the future, and this will cause e2fsck
2393          * to complain and force a full file system check.
2394          */
2395         if (!(sb->s_flags & MS_RDONLY))
2396                 es->s_wtime = cpu_to_le32(get_seconds());
2397         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2398         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2399         BUFFER_TRACE(sbh, "marking dirty");
2400         mark_buffer_dirty(sbh);
2401         if (sync)
2402                 error = sync_dirty_buffer(sbh);
2403         return error;
2404 }
2405
2406
2407 /*
2408  * Have we just finished recovery?  If so, and if we are mounting (or
2409  * remounting) the filesystem readonly, then we will end up with a
2410  * consistent fs on disk.  Record that fact.
2411  */
2412 static void ext3_mark_recovery_complete(struct super_block * sb,
2413                                         struct ext3_super_block * es)
2414 {
2415         journal_t *journal = EXT3_SB(sb)->s_journal;
2416
2417         journal_lock_updates(journal);
2418         if (journal_flush(journal) < 0)
2419                 goto out;
2420
2421         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2422             sb->s_flags & MS_RDONLY) {
2423                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2424                 ext3_commit_super(sb, es, 1);
2425         }
2426
2427 out:
2428         journal_unlock_updates(journal);
2429 }
2430
2431 /*
2432  * If we are mounting (or read-write remounting) a filesystem whose journal
2433  * has recorded an error from a previous lifetime, move that error to the
2434  * main filesystem now.
2435  */
2436 static void ext3_clear_journal_err(struct super_block *sb,
2437                                    struct ext3_super_block *es)
2438 {
2439         journal_t *journal;
2440         int j_errno;
2441         const char *errstr;
2442
2443         journal = EXT3_SB(sb)->s_journal;
2444
2445         /*
2446          * Now check for any error status which may have been recorded in the
2447          * journal by a prior ext3_error() or ext3_abort()
2448          */
2449
2450         j_errno = journal_errno(journal);
2451         if (j_errno) {
2452                 char nbuf[16];
2453
2454                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2455                 ext3_warning(sb, __func__, "Filesystem error recorded "
2456                              "from previous mount: %s", errstr);
2457                 ext3_warning(sb, __func__, "Marking fs in need of "
2458                              "filesystem check.");
2459
2460                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2461                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2462                 ext3_commit_super (sb, es, 1);
2463
2464                 journal_clear_err(journal);
2465         }
2466 }
2467
2468 /*
2469  * Force the running and committing transactions to commit,
2470  * and wait on the commit.
2471  */
2472 int ext3_force_commit(struct super_block *sb)
2473 {
2474         journal_t *journal;
2475         int ret;
2476
2477         if (sb->s_flags & MS_RDONLY)
2478                 return 0;
2479
2480         journal = EXT3_SB(sb)->s_journal;
2481         ret = ext3_journal_force_commit(journal);
2482         return ret;
2483 }
2484
2485 static int ext3_sync_fs(struct super_block *sb, int wait)
2486 {
2487         tid_t target;
2488
2489         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2490                 if (wait)
2491                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2492         }
2493         return 0;
2494 }
2495
2496 /*
2497  * LVM calls this function before a (read-only) snapshot is created.  This
2498  * gives us a chance to flush the journal completely and mark the fs clean.
2499  */
2500 static int ext3_freeze(struct super_block *sb)
2501 {
2502         int error = 0;
2503         journal_t *journal;
2504
2505         if (!(sb->s_flags & MS_RDONLY)) {
2506                 journal = EXT3_SB(sb)->s_journal;
2507
2508                 /* Now we set up the journal barrier. */
2509                 journal_lock_updates(journal);
2510
2511                 /*
2512                  * We don't want to clear needs_recovery flag when we failed
2513                  * to flush the journal.
2514                  */
2515                 error = journal_flush(journal);
2516                 if (error < 0)
2517                         goto out;
2518
2519                 /* Journal blocked and flushed, clear needs_recovery flag. */
2520                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2521                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2522                 if (error)
2523                         goto out;
2524         }
2525         return 0;
2526
2527 out:
2528         journal_unlock_updates(journal);
2529         return error;
2530 }
2531
2532 /*
2533  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2534  * flag here, even though the filesystem is not technically dirty yet.
2535  */
2536 static int ext3_unfreeze(struct super_block *sb)
2537 {
2538         if (!(sb->s_flags & MS_RDONLY)) {
2539                 lock_super(sb);
2540                 /* Reser the needs_recovery flag before the fs is unlocked. */
2541                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2542                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2543                 unlock_super(sb);
2544                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2545         }
2546         return 0;
2547 }
2548
2549 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2550 {
2551         struct ext3_super_block * es;
2552         struct ext3_sb_info *sbi = EXT3_SB(sb);
2553         ext3_fsblk_t n_blocks_count = 0;
2554         unsigned long old_sb_flags;
2555         struct ext3_mount_options old_opts;
2556         int enable_quota = 0;
2557         int err;
2558 #ifdef CONFIG_QUOTA
2559         int i;
2560 #endif
2561
2562         lock_kernel();
2563
2564         /* Store the original options */
2565         lock_super(sb);
2566         old_sb_flags = sb->s_flags;
2567         old_opts.s_mount_opt = sbi->s_mount_opt;
2568         old_opts.s_resuid = sbi->s_resuid;
2569         old_opts.s_resgid = sbi->s_resgid;
2570         old_opts.s_commit_interval = sbi->s_commit_interval;
2571 #ifdef CONFIG_QUOTA
2572         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2573         for (i = 0; i < MAXQUOTAS; i++)
2574                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2575 #endif
2576
2577         /*
2578          * Allow the "check" option to be passed as a remount option.
2579          */
2580         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2581                 err = -EINVAL;
2582                 goto restore_opts;
2583         }
2584
2585         if (test_opt(sb, ABORT))
2586                 ext3_abort(sb, __func__, "Abort forced by user");
2587
2588         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2589                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2590
2591         es = sbi->s_es;
2592
2593         ext3_init_journal_params(sb, sbi->s_journal);
2594
2595         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2596                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2597                 if (test_opt(sb, ABORT)) {
2598                         err = -EROFS;
2599                         goto restore_opts;
2600                 }
2601
2602                 if (*flags & MS_RDONLY) {
2603                         err = dquot_suspend(sb, -1);
2604                         if (err < 0)
2605                                 goto restore_opts;
2606
2607                         /*
2608                          * First of all, the unconditional stuff we have to do
2609                          * to disable replay of the journal when we next remount
2610                          */
2611                         sb->s_flags |= MS_RDONLY;
2612
2613                         /*
2614                          * OK, test if we are remounting a valid rw partition
2615                          * readonly, and if so set the rdonly flag and then
2616                          * mark the partition as valid again.
2617                          */
2618                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2619                             (sbi->s_mount_state & EXT3_VALID_FS))
2620                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2621
2622                         ext3_mark_recovery_complete(sb, es);
2623                 } else {
2624                         __le32 ret;
2625                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2626                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2627                                 ext3_msg(sb, KERN_WARNING,
2628                                         "warning: couldn't remount RDWR "
2629                                         "because of unsupported optional "
2630                                         "features (%x)", le32_to_cpu(ret));
2631                                 err = -EROFS;
2632                                 goto restore_opts;
2633                         }
2634
2635                         /*
2636                          * If we have an unprocessed orphan list hanging
2637                          * around from a previously readonly bdev mount,
2638                          * require a full umount/remount for now.
2639                          */
2640                         if (es->s_last_orphan) {
2641                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2642                                        "remount RDWR because of unprocessed "
2643                                        "orphan inode list.  Please "
2644                                        "umount/remount instead.");
2645                                 err = -EINVAL;
2646                                 goto restore_opts;
2647                         }
2648
2649                         /*
2650                          * Mounting a RDONLY partition read-write, so reread
2651                          * and store the current valid flag.  (It may have
2652                          * been changed by e2fsck since we originally mounted
2653                          * the partition.)
2654                          */
2655                         ext3_clear_journal_err(sb, es);
2656                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2657                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2658                                 goto restore_opts;
2659                         if (!ext3_setup_super (sb, es, 0))
2660                                 sb->s_flags &= ~MS_RDONLY;
2661                         enable_quota = 1;
2662                 }
2663         }
2664 #ifdef CONFIG_QUOTA
2665         /* Release old quota file names */
2666         for (i = 0; i < MAXQUOTAS; i++)
2667                 if (old_opts.s_qf_names[i] &&
2668                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2669                         kfree(old_opts.s_qf_names[i]);
2670 #endif
2671         unlock_super(sb);
2672         unlock_kernel();
2673
2674         if (enable_quota)
2675                 dquot_resume(sb, -1);
2676         return 0;
2677 restore_opts:
2678         sb->s_flags = old_sb_flags;
2679         sbi->s_mount_opt = old_opts.s_mount_opt;
2680         sbi->s_resuid = old_opts.s_resuid;
2681         sbi->s_resgid = old_opts.s_resgid;
2682         sbi->s_commit_interval = old_opts.s_commit_interval;
2683 #ifdef CONFIG_QUOTA
2684         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2685         for (i = 0; i < MAXQUOTAS; i++) {
2686                 if (sbi->s_qf_names[i] &&
2687                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2688                         kfree(sbi->s_qf_names[i]);
2689                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2690         }
2691 #endif
2692         unlock_super(sb);
2693         unlock_kernel();
2694         return err;
2695 }
2696
2697 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2698 {
2699         struct super_block *sb = dentry->d_sb;
2700         struct ext3_sb_info *sbi = EXT3_SB(sb);
2701         struct ext3_super_block *es = sbi->s_es;
2702         u64 fsid;
2703
2704         if (test_opt(sb, MINIX_DF)) {
2705                 sbi->s_overhead_last = 0;
2706         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2707                 unsigned long ngroups = sbi->s_groups_count, i;
2708                 ext3_fsblk_t overhead = 0;
2709                 smp_rmb();
2710
2711                 /*
2712                  * Compute the overhead (FS structures).  This is constant
2713                  * for a given filesystem unless the number of block groups
2714                  * changes so we cache the previous value until it does.
2715                  */
2716
2717                 /*
2718                  * All of the blocks before first_data_block are
2719                  * overhead
2720                  */
2721                 overhead = le32_to_cpu(es->s_first_data_block);
2722
2723                 /*
2724                  * Add the overhead attributed to the superblock and
2725                  * block group descriptors.  If the sparse superblocks
2726                  * feature is turned on, then not all groups have this.
2727                  */
2728                 for (i = 0; i < ngroups; i++) {
2729                         overhead += ext3_bg_has_super(sb, i) +
2730                                 ext3_bg_num_gdb(sb, i);
2731                         cond_resched();
2732                 }
2733
2734                 /*
2735                  * Every block group has an inode bitmap, a block
2736                  * bitmap, and an inode table.
2737                  */
2738                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2739                 sbi->s_overhead_last = overhead;
2740                 smp_wmb();
2741                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2742         }
2743
2744         buf->f_type = EXT3_SUPER_MAGIC;
2745         buf->f_bsize = sb->s_blocksize;
2746         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2747         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2748         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2749         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2750                 buf->f_bavail = 0;
2751         buf->f_files = le32_to_cpu(es->s_inodes_count);
2752         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2753         buf->f_namelen = EXT3_NAME_LEN;
2754         fsid = le64_to_cpup((void *)es->s_uuid) ^
2755                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2756         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2757         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2758         return 0;
2759 }
2760
2761 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2762  * is locked for write. Otherwise the are possible deadlocks:
2763  * Process 1                         Process 2
2764  * ext3_create()                     quota_sync()
2765  *   journal_start()                   write_dquot()
2766  *   dquot_initialize()                       down(dqio_mutex)
2767  *     down(dqio_mutex)                    journal_start()
2768  *
2769  */
2770
2771 #ifdef CONFIG_QUOTA
2772
2773 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2774 {
2775         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2776 }
2777
2778 static int ext3_write_dquot(struct dquot *dquot)
2779 {
2780         int ret, err;
2781         handle_t *handle;
2782         struct inode *inode;
2783
2784         inode = dquot_to_inode(dquot);
2785         handle = ext3_journal_start(inode,
2786                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2787         if (IS_ERR(handle))
2788                 return PTR_ERR(handle);
2789         ret = dquot_commit(dquot);
2790         err = ext3_journal_stop(handle);
2791         if (!ret)
2792                 ret = err;
2793         return ret;
2794 }
2795
2796 static int ext3_acquire_dquot(struct dquot *dquot)
2797 {
2798         int ret, err;
2799         handle_t *handle;
2800
2801         handle = ext3_journal_start(dquot_to_inode(dquot),
2802                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2803         if (IS_ERR(handle))
2804                 return PTR_ERR(handle);
2805         ret = dquot_acquire(dquot);
2806         err = ext3_journal_stop(handle);
2807         if (!ret)
2808                 ret = err;
2809         return ret;
2810 }
2811
2812 static int ext3_release_dquot(struct dquot *dquot)
2813 {
2814         int ret, err;
2815         handle_t *handle;
2816
2817         handle = ext3_journal_start(dquot_to_inode(dquot),
2818                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2819         if (IS_ERR(handle)) {
2820                 /* Release dquot anyway to avoid endless cycle in dqput() */
2821                 dquot_release(dquot);
2822                 return PTR_ERR(handle);
2823         }
2824         ret = dquot_release(dquot);
2825         err = ext3_journal_stop(handle);
2826         if (!ret)
2827                 ret = err;
2828         return ret;
2829 }
2830
2831 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2832 {
2833         /* Are we journaling quotas? */
2834         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2835             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2836                 dquot_mark_dquot_dirty(dquot);
2837                 return ext3_write_dquot(dquot);
2838         } else {
2839                 return dquot_mark_dquot_dirty(dquot);
2840         }
2841 }
2842
2843 static int ext3_write_info(struct super_block *sb, int type)
2844 {
2845         int ret, err;
2846         handle_t *handle;
2847
2848         /* Data block + inode block */
2849         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2850         if (IS_ERR(handle))
2851                 return PTR_ERR(handle);
2852         ret = dquot_commit_info(sb, type);
2853         err = ext3_journal_stop(handle);
2854         if (!ret)
2855                 ret = err;
2856         return ret;
2857 }
2858
2859 /*
2860  * Turn on quotas during mount time - we need to find
2861  * the quota file and such...
2862  */
2863 static int ext3_quota_on_mount(struct super_block *sb, int type)
2864 {
2865         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2866                                         EXT3_SB(sb)->s_jquota_fmt, type);
2867 }
2868
2869 /*
2870  * Standard function to be called on quota_on
2871  */
2872 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2873                          char *name)
2874 {
2875         int err;
2876         struct path path;
2877
2878         if (!test_opt(sb, QUOTA))
2879                 return -EINVAL;
2880
2881         err = kern_path(name, LOOKUP_FOLLOW, &path);
2882         if (err)
2883                 return err;
2884
2885         /* Quotafile not on the same filesystem? */
2886         if (path.mnt->mnt_sb != sb) {
2887                 path_put(&path);
2888                 return -EXDEV;
2889         }
2890         /* Journaling quota? */
2891         if (EXT3_SB(sb)->s_qf_names[type]) {
2892                 /* Quotafile not of fs root? */
2893                 if (path.dentry->d_parent != sb->s_root)
2894                         ext3_msg(sb, KERN_WARNING,
2895                                 "warning: Quota file not on filesystem root. "
2896                                 "Journaled quota will not work.");
2897         }
2898
2899         /*
2900          * When we journal data on quota file, we have to flush journal to see
2901          * all updates to the file when we bypass pagecache...
2902          */
2903         if (ext3_should_journal_data(path.dentry->d_inode)) {
2904                 /*
2905                  * We don't need to lock updates but journal_flush() could
2906                  * otherwise be livelocked...
2907                  */
2908                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2909                 err = journal_flush(EXT3_SB(sb)->s_journal);
2910                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2911                 if (err) {
2912                         path_put(&path);
2913                         return err;
2914                 }
2915         }
2916
2917         err = dquot_quota_on_path(sb, type, format_id, &path);
2918         path_put(&path);
2919         return err;
2920 }
2921
2922 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2923  * acquiring the locks... As quota files are never truncated and quota code
2924  * itself serializes the operations (and noone else should touch the files)
2925  * we don't have to be afraid of races */
2926 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2927                                size_t len, loff_t off)
2928 {
2929         struct inode *inode = sb_dqopt(sb)->files[type];
2930         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2931         int err = 0;
2932         int offset = off & (sb->s_blocksize - 1);
2933         int tocopy;
2934         size_t toread;
2935         struct buffer_head *bh;
2936         loff_t i_size = i_size_read(inode);
2937
2938         if (off > i_size)
2939                 return 0;
2940         if (off+len > i_size)
2941                 len = i_size-off;
2942         toread = len;
2943         while (toread > 0) {
2944                 tocopy = sb->s_blocksize - offset < toread ?
2945                                 sb->s_blocksize - offset : toread;
2946                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2947                 if (err)
2948                         return err;
2949                 if (!bh)        /* A hole? */
2950                         memset(data, 0, tocopy);
2951                 else
2952                         memcpy(data, bh->b_data+offset, tocopy);
2953                 brelse(bh);
2954                 offset = 0;
2955                 toread -= tocopy;
2956                 data += tocopy;
2957                 blk++;
2958         }
2959         return len;
2960 }
2961
2962 /* Write to quotafile (we know the transaction is already started and has
2963  * enough credits) */
2964 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2965                                 const char *data, size_t len, loff_t off)
2966 {
2967         struct inode *inode = sb_dqopt(sb)->files[type];
2968         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2969         int err = 0;
2970         int offset = off & (sb->s_blocksize - 1);
2971         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2972         struct buffer_head *bh;
2973         handle_t *handle = journal_current_handle();
2974
2975         if (!handle) {
2976                 ext3_msg(sb, KERN_WARNING,
2977                         "warning: quota write (off=%llu, len=%llu)"
2978                         " cancelled because transaction is not started.",
2979                         (unsigned long long)off, (unsigned long long)len);
2980                 return -EIO;
2981         }
2982
2983         /*
2984          * Since we account only one data block in transaction credits,
2985          * then it is impossible to cross a block boundary.
2986          */
2987         if (sb->s_blocksize - offset < len) {
2988                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2989                         " cancelled because not block aligned",
2990                         (unsigned long long)off, (unsigned long long)len);
2991                 return -EIO;
2992         }
2993         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2994         bh = ext3_bread(handle, inode, blk, 1, &err);
2995         if (!bh)
2996                 goto out;
2997         if (journal_quota) {
2998                 err = ext3_journal_get_write_access(handle, bh);
2999                 if (err) {
3000                         brelse(bh);
3001                         goto out;
3002                 }
3003         }
3004         lock_buffer(bh);
3005         memcpy(bh->b_data+offset, data, len);
3006         flush_dcache_page(bh->b_page);
3007         unlock_buffer(bh);
3008         if (journal_quota)
3009                 err = ext3_journal_dirty_metadata(handle, bh);
3010         else {
3011                 /* Always do at least ordered writes for quotas */
3012                 err = ext3_journal_dirty_data(handle, bh);
3013                 mark_buffer_dirty(bh);
3014         }
3015         brelse(bh);
3016 out:
3017         if (err) {
3018                 mutex_unlock(&inode->i_mutex);
3019                 return err;
3020         }
3021         if (inode->i_size < off + len) {
3022                 i_size_write(inode, off + len);
3023                 EXT3_I(inode)->i_disksize = inode->i_size;
3024         }
3025         inode->i_version++;
3026         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3027         ext3_mark_inode_dirty(handle, inode);
3028         mutex_unlock(&inode->i_mutex);
3029         return len;
3030 }
3031
3032 #endif
3033
3034 static int ext3_get_sb(struct file_system_type *fs_type,
3035         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3036 {
3037         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3038 }
3039
3040 static struct file_system_type ext3_fs_type = {
3041         .owner          = THIS_MODULE,
3042         .name           = "ext3",
3043         .get_sb         = ext3_get_sb,
3044         .kill_sb        = kill_block_super,
3045         .fs_flags       = FS_REQUIRES_DEV,
3046 };
3047
3048 static int __init init_ext3_fs(void)
3049 {
3050         int err = init_ext3_xattr();
3051         if (err)
3052                 return err;
3053         err = init_inodecache();
3054         if (err)
3055                 goto out1;
3056         err = register_filesystem(&ext3_fs_type);
3057         if (err)
3058                 goto out;
3059         return 0;
3060 out:
3061         destroy_inodecache();
3062 out1:
3063         exit_ext3_xattr();
3064         return err;
3065 }
3066
3067 static void __exit exit_ext3_fs(void)
3068 {
3069         unregister_filesystem(&ext3_fs_type);
3070         destroy_inodecache();
3071         exit_ext3_xattr();
3072 }
3073
3074 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3075 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3076 MODULE_LICENSE("GPL");
3077 module_init(init_ext3_fs)
3078 module_exit(exit_ext3_fs)