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/blkdev.h>
  21 #include <linux/parser.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/statfs.h>
  24 #include <linux/random.h>
  25 #include <linux/mount.h>
  26 #include <linux/quotaops.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/log2.h>
  29 #include <linux/cleancache.h>
  30 #include <linux/namei.h>
  31
  32 #include <asm/uaccess.h>
  33
  34 #define CREATE_TRACE_POINTS
  35
  36 #include "ext3.h"
  37 #include "xattr.h"
  38 #include "acl.h"
  39 #include "namei.h"
  40
  41 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
  42   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
  43 #else
  44   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
  45 #endif
  46
  47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  48                              unsigned long journal_devnum);
  49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  50                                unsigned int);
  51 static int ext3_commit_super(struct super_block *sb,
  52                                struct ext3_super_block *es,
  53                                int sync);
  54 static void ext3_mark_recovery_complete(struct super_block * sb,
  55                                         struct ext3_super_block * es);
  56 static void ext3_clear_journal_err(struct super_block * sb,
  57                                    struct ext3_super_block * es);
  58 static int ext3_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext3_decode_error(struct super_block * sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  63 static int ext3_unfreeze(struct super_block *sb);
  64 static int ext3_freeze(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  68  */
  69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  70 {
  71         journal_t *journal;
  72
  73         if (sb->s_flags & MS_RDONLY)
  74                 return ERR_PTR(-EROFS);
  75
  76         /* Special case here: if the journal has aborted behind our
  77          * backs (eg. EIO in the commit thread), then we still need to
  78          * take the FS itself readonly cleanly. */
  79         journal = EXT3_SB(sb)->s_journal;
  80         if (is_journal_aborted(journal)) {
  81                 ext3_abort(sb, __func__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 int __ext3_journal_stop(const char *where, handle_t *handle)
  90 {
  91         struct super_block *sb;
  92         int err;
  93         int rc;
  94
  95         sb = handle->h_transaction->t_journal->j_private;
  96         err = handle->h_err;
  97         rc = journal_stop(handle);
  98
  99         if (!err)
 100                 err = rc;
 101         if (err)
 102                 __ext3_std_error(sb, where, err);
 103         return err;
 104 }
 105
 106 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 107                 struct buffer_head *bh, handle_t *handle, int err)
 108 {
 109         char nbuf[16];
 110         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 111
 112         if (bh)
 113                 BUFFER_TRACE(bh, "abort");
 114
 115         if (!handle->h_err)
 116                 handle->h_err = err;
 117
 118         if (is_handle_aborted(handle))
 119                 return;
 120
 121         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
 122                 caller, errstr, err_fn);
 123
 124         journal_abort_handle(handle);
 125 }
 126
 127 void ext3_msg(struct super_block *sb, const char *prefix,
 128                 const char *fmt, ...)
 129 {
 130         struct va_format vaf;
 131         va_list args;
 132
 133         va_start(args, fmt);
 134
 135         vaf.fmt = fmt;
 136         vaf.va = &args;
 137
 138         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 139
 140         va_end(args);
 141 }
 142
 143 /* Deal with the reporting of failure conditions on a filesystem such as
 144  * inconsistencies detected or read IO failures.
 145  *
 146  * On ext2, we can store the error state of the filesystem in the
 147  * superblock.  That is not possible on ext3, because we may have other
 148  * write ordering constraints on the superblock which prevent us from
 149  * writing it out straight away; and given that the journal is about to
 150  * be aborted, we can't rely on the current, or future, transactions to
 151  * write out the superblock safely.
 152  *
 153  * We'll just use the journal_abort() error code to record an error in
 154  * the journal instead.  On recovery, the journal will complain about
 155  * that error until we've noted it down and cleared it.
 156  */
 157
 158 static void ext3_handle_error(struct super_block *sb)
 159 {
 160         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 161
 162         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 163         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 164
 165         if (sb->s_flags & MS_RDONLY)
 166                 return;
 167
 168         if (!test_opt (sb, ERRORS_CONT)) {
 169                 journal_t *journal = EXT3_SB(sb)->s_journal;
 170
 171                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 172                 if (journal)
 173                         journal_abort(journal, -EIO);
 174         }
 175         if (test_opt (sb, ERRORS_RO)) {
 176                 ext3_msg(sb, KERN_CRIT,
 177                         "error: remounting filesystem read-only");
 178                 /*
 179                  * Make sure updated value of ->s_mount_state will be visible
 180                  * before ->s_flags update.
 181                  */
 182                 smp_wmb();
 183                 sb->s_flags |= MS_RDONLY;
 184         }
 185         ext3_commit_super(sb, es, 1);
 186         if (test_opt(sb, ERRORS_PANIC))
 187                 panic("EXT3-fs (%s): panic forced after error\n",
 188                         sb->s_id);
 189 }
 190
 191 void ext3_error(struct super_block *sb, const char *function,
 192                 const char *fmt, ...)
 193 {
 194         struct va_format vaf;
 195         va_list args;
 196
 197         va_start(args, fmt);
 198
 199         vaf.fmt = fmt;
 200         vaf.va = &args;
 201
 202         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
 203                sb->s_id, function, &vaf);
 204
 205         va_end(args);
 206
 207         ext3_handle_error(sb);
 208 }
 209
 210 static const char *ext3_decode_error(struct super_block * sb, int errno,
 211                                      char nbuf[16])
 212 {
 213         char *errstr = NULL;
 214
 215         switch (errno) {
 216         case -EIO:
 217                 errstr = "IO failure";
 218                 break;
 219         case -ENOMEM:
 220                 errstr = "Out of memory";
 221                 break;
 222         case -EROFS:
 223                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 224                         errstr = "Journal has aborted";
 225                 else
 226                         errstr = "Readonly filesystem";
 227                 break;
 228         default:
 229                 /* If the caller passed in an extra buffer for unknown
 230                  * errors, textualise them now.  Else we just return
 231                  * NULL. */
 232                 if (nbuf) {
 233                         /* Check for truncated error codes... */
 234                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 235                                 errstr = nbuf;
 236                 }
 237                 break;
 238         }
 239
 240         return errstr;
 241 }
 242
 243 /* __ext3_std_error decodes expected errors from journaling functions
 244  * automatically and invokes the appropriate error response.  */
 245
 246 void __ext3_std_error (struct super_block * sb, const char * function,
 247                        int errno)
 248 {
 249         char nbuf[16];
 250         const char *errstr;
 251
 252         /* Special case: if the error is EROFS, and we're not already
 253          * inside a transaction, then there's really no point in logging
 254          * an error. */
 255         if (errno == -EROFS && journal_current_handle() == NULL &&
 256             (sb->s_flags & MS_RDONLY))
 257                 return;
 258
 259         errstr = ext3_decode_error(sb, errno, nbuf);
 260         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
 261
 262         ext3_handle_error(sb);
 263 }
 264
 265 /*
 266  * ext3_abort is a much stronger failure handler than ext3_error.  The
 267  * abort function may be used to deal with unrecoverable failures such
 268  * as journal IO errors or ENOMEM at a critical moment in log management.
 269  *
 270  * We unconditionally force the filesystem into an ABORT|READONLY state,
 271  * unless the error response on the fs has been set to panic in which
 272  * case we take the easy way out and panic immediately.
 273  */
 274
 275 void ext3_abort(struct super_block *sb, const char *function,
 276                  const char *fmt, ...)
 277 {
 278         struct va_format vaf;
 279         va_list args;
 280
 281         va_start(args, fmt);
 282
 283         vaf.fmt = fmt;
 284         vaf.va = &args;
 285
 286         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
 287                sb->s_id, function, &vaf);
 288
 289         va_end(args);
 290
 291         if (test_opt(sb, ERRORS_PANIC))
 292                 panic("EXT3-fs: panic from previous error\n");
 293
 294         if (sb->s_flags & MS_RDONLY)
 295                 return;
 296
 297         ext3_msg(sb, KERN_CRIT,
 298                 "error: remounting filesystem read-only");
 299         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 300         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
 301         /*
 302          * Make sure updated value of ->s_mount_state will be visible
 303          * before ->s_flags update.
 304          */
 305         smp_wmb();
 306         sb->s_flags |= MS_RDONLY;
 307
 308         if (EXT3_SB(sb)->s_journal)
 309                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 310 }
 311
 312 void ext3_warning(struct super_block *sb, const char *function,
 313                   const char *fmt, ...)
 314 {
 315         struct va_format vaf;
 316         va_list args;
 317
 318         va_start(args, fmt);
 319
 320         vaf.fmt = fmt;
 321         vaf.va = &args;
 322
 323         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
 324                sb->s_id, function, &vaf);
 325
 326         va_end(args);
 327 }
 328
 329 void ext3_update_dynamic_rev(struct super_block *sb)
 330 {
 331         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 332
 333         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 334                 return;
 335
 336         ext3_msg(sb, KERN_WARNING,
 337                 "warning: updating to rev %d because of "
 338                 "new feature flag, running e2fsck is recommended",
 339                 EXT3_DYNAMIC_REV);
 340
 341         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 342         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 343         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 344         /* leave es->s_feature_*compat flags alone */
 345         /* es->s_uuid will be set by e2fsck if empty */
 346
 347         /*
 348          * The rest of the superblock fields should be zero, and if not it
 349          * means they are likely already in use, so leave them alone.  We
 350          * can leave it up to e2fsck to clean up any inconsistencies there.
 351          */
 352 }
 353
 354 /*
 355  * Open the external journal device
 356  */
 357 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
 358 {
 359         struct block_device *bdev;
 360         char b[BDEVNAME_SIZE];
 361
 362         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 363         if (IS_ERR(bdev))
 364                 goto fail;
 365         return bdev;
 366
 367 fail:
 368         ext3_msg(sb, KERN_ERR, "error: failed to open journal device %s: %ld",
 369                 __bdevname(dev, b), PTR_ERR(bdev));
 370
 371         return NULL;
 372 }
 373
 374 /*
 375  * Release the journal device
 376  */
 377 static void ext3_blkdev_put(struct block_device *bdev)
 378 {
 379         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 380 }
 381
 382 static void ext3_blkdev_remove(struct ext3_sb_info *sbi)
 383 {
 384         struct block_device *bdev;
 385         bdev = sbi->journal_bdev;
 386         if (bdev) {
 387                 ext3_blkdev_put(bdev);
 388                 sbi->journal_bdev = NULL;
 389         }
 390 }
 391
 392 static inline struct inode *orphan_list_entry(struct list_head *l)
 393 {
 394         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 395 }
 396
 397 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 398 {
 399         struct list_head *l;
 400
 401         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
 402                le32_to_cpu(sbi->s_es->s_last_orphan));
 403
 404         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
 405         list_for_each(l, &sbi->s_orphan) {
 406                 struct inode *inode = orphan_list_entry(l);
 407                 ext3_msg(sb, KERN_ERR, "  "
 408                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 409                        inode->i_sb->s_id, inode->i_ino, inode,
 410                        inode->i_mode, inode->i_nlink,
 411                        NEXT_ORPHAN(inode));
 412         }
 413 }
 414
 415 static void ext3_put_super (struct super_block * sb)
 416 {
 417         struct ext3_sb_info *sbi = EXT3_SB(sb);
 418         struct ext3_super_block *es = sbi->s_es;
 419         int i, err;
 420
 421         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 422         ext3_xattr_put_super(sb);
 423         err = journal_destroy(sbi->s_journal);
 424         sbi->s_journal = NULL;
 425         if (err < 0)
 426                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
 427
 428         if (!(sb->s_flags & MS_RDONLY)) {
 429                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 430                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 431                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 432                 mark_buffer_dirty(sbi->s_sbh);
 433                 ext3_commit_super(sb, es, 1);
 434         }
 435
 436         for (i = 0; i < sbi->s_gdb_count; i++)
 437                 brelse(sbi->s_group_desc[i]);
 438         kfree(sbi->s_group_desc);
 439         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 440         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 441         percpu_counter_destroy(&sbi->s_dirs_counter);
 442         brelse(sbi->s_sbh);
 443 #ifdef CONFIG_QUOTA
 444         for (i = 0; i < MAXQUOTAS; i++)
 445                 kfree(sbi->s_qf_names[i]);
 446 #endif
 447
 448         /* Debugging code just in case the in-memory inode orphan list
 449          * isn't empty.  The on-disk one can be non-empty if we've
 450          * detected an error and taken the fs readonly, but the
 451          * in-memory list had better be clean by this point. */
 452         if (!list_empty(&sbi->s_orphan))
 453                 dump_orphan_list(sb, sbi);
 454         J_ASSERT(list_empty(&sbi->s_orphan));
 455
 456         invalidate_bdev(sb->s_bdev);
 457         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 458                 /*
 459                  * Invalidate the journal device's buffers.  We don't want them
 460                  * floating about in memory - the physical journal device may
 461                  * hotswapped, and it breaks the `ro-after' testing code.
 462                  */
 463                 sync_blockdev(sbi->journal_bdev);
 464                 invalidate_bdev(sbi->journal_bdev);
 465                 ext3_blkdev_remove(sbi);
 466         }
 467         sb->s_fs_info = NULL;
 468         kfree(sbi->s_blockgroup_lock);
 469         kfree(sbi);
 470 }
 471
 472 static struct kmem_cache *ext3_inode_cachep;
 473
 474 /*
 475  * Called inside transaction, so use GFP_NOFS
 476  */
 477 static struct inode *ext3_alloc_inode(struct super_block *sb)
 478 {
 479         struct ext3_inode_info *ei;
 480
 481         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 482         if (!ei)
 483                 return NULL;
 484         ei->i_block_alloc_info = NULL;
 485         ei->vfs_inode.i_version = 1;
 486         atomic_set(&ei->i_datasync_tid, 0);
 487         atomic_set(&ei->i_sync_tid, 0);
 488         return &ei->vfs_inode;
 489 }
 490
 491 static int ext3_drop_inode(struct inode *inode)
 492 {
 493         int drop = generic_drop_inode(inode);
 494
 495         trace_ext3_drop_inode(inode, drop);
 496         return drop;
 497 }
 498
 499 static void ext3_i_callback(struct rcu_head *head)
 500 {
 501         struct inode *inode = container_of(head, struct inode, i_rcu);
 502         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 503 }
 504
 505 static void ext3_destroy_inode(struct inode *inode)
 506 {
 507         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
 508                 printk("EXT3 Inode %p: orphan list check failed!\n",
 509                         EXT3_I(inode));
 510                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 511                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
 512                                 false);
 513                 dump_stack();
 514         }
 515         call_rcu(&inode->i_rcu, ext3_i_callback);
 516 }
 517
 518 static void init_once(void *foo)
 519 {
 520         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 521
 522         INIT_LIST_HEAD(&ei->i_orphan);
 523 #ifdef CONFIG_EXT3_FS_XATTR
 524         init_rwsem(&ei->xattr_sem);
 525 #endif
 526         mutex_init(&ei->truncate_mutex);
 527         inode_init_once(&ei->vfs_inode);
 528 }
 529
 530 static int __init init_inodecache(void)
 531 {
 532         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 533                                              sizeof(struct ext3_inode_info),
 534                                              0, (SLAB_RECLAIM_ACCOUNT|
 535                                                 SLAB_MEM_SPREAD),
 536                                              init_once);
 537         if (ext3_inode_cachep == NULL)
 538                 return -ENOMEM;
 539         return 0;
 540 }
 541
 542 static void destroy_inodecache(void)
 543 {
 544         /*
 545          * Make sure all delayed rcu free inodes are flushed before we
 546          * destroy cache.
 547          */
 548         rcu_barrier();
 549         kmem_cache_destroy(ext3_inode_cachep);
 550 }
 551
 552 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 553 {
 554 #if defined(CONFIG_QUOTA)
 555         struct ext3_sb_info *sbi = EXT3_SB(sb);
 556
 557         if (sbi->s_jquota_fmt) {
 558                 char *fmtname = "";
 559
 560                 switch (sbi->s_jquota_fmt) {
 561                 case QFMT_VFS_OLD:
 562                         fmtname = "vfsold";
 563                         break;
 564                 case QFMT_VFS_V0:
 565                         fmtname = "vfsv0";
 566                         break;
 567                 case QFMT_VFS_V1:
 568                         fmtname = "vfsv1";
 569                         break;
 570                 }
 571                 seq_printf(seq, ",jqfmt=%s", fmtname);
 572         }
 573
 574         if (sbi->s_qf_names[USRQUOTA])
 575                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 576
 577         if (sbi->s_qf_names[GRPQUOTA])
 578                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 579
 580         if (test_opt(sb, USRQUOTA))
 581                 seq_puts(seq, ",usrquota");
 582
 583         if (test_opt(sb, GRPQUOTA))
 584                 seq_puts(seq, ",grpquota");
 585 #endif
 586 }
 587
 588 static char *data_mode_string(unsigned long mode)
 589 {
 590         switch (mode) {
 591         case EXT3_MOUNT_JOURNAL_DATA:
 592                 return "journal";
 593         case EXT3_MOUNT_ORDERED_DATA:
 594                 return "ordered";
 595         case EXT3_MOUNT_WRITEBACK_DATA:
 596                 return "writeback";
 597         }
 598         return "unknown";
 599 }
 600
 601 /*
 602  * Show an option if
 603  *  - it's set to a non-default value OR
 604  *  - if the per-sb default is different from the global default
 605  */
 606 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
 607 {
 608         struct super_block *sb = root->d_sb;
 609         struct ext3_sb_info *sbi = EXT3_SB(sb);
 610         struct ext3_super_block *es = sbi->s_es;
 611         unsigned long def_mount_opts;
 612
 613         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 614
 615         if (sbi->s_sb_block != 1)
 616                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
 617         if (test_opt(sb, MINIX_DF))
 618                 seq_puts(seq, ",minixdf");
 619         if (test_opt(sb, GRPID))
 620                 seq_puts(seq, ",grpid");
 621         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
 622                 seq_puts(seq, ",nogrpid");
 623         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
 624             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
 625                 seq_printf(seq, ",resuid=%u",
 626                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
 627         }
 628         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
 629             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
 630                 seq_printf(seq, ",resgid=%u",
 631                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
 632         }
 633         if (test_opt(sb, ERRORS_RO)) {
 634                 int def_errors = le16_to_cpu(es->s_errors);
 635
 636                 if (def_errors == EXT3_ERRORS_PANIC ||
 637                     def_errors == EXT3_ERRORS_CONTINUE) {
 638                         seq_puts(seq, ",errors=remount-ro");
 639                 }
 640         }
 641         if (test_opt(sb, ERRORS_CONT))
 642                 seq_puts(seq, ",errors=continue");
 643         if (test_opt(sb, ERRORS_PANIC))
 644                 seq_puts(seq, ",errors=panic");
 645         if (test_opt(sb, NO_UID32))
 646                 seq_puts(seq, ",nouid32");
 647         if (test_opt(sb, DEBUG))
 648                 seq_puts(seq, ",debug");
 649 #ifdef CONFIG_EXT3_FS_XATTR
 650         if (test_opt(sb, XATTR_USER))
 651                 seq_puts(seq, ",user_xattr");
 652         if (!test_opt(sb, XATTR_USER) &&
 653             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
 654                 seq_puts(seq, ",nouser_xattr");
 655         }
 656 #endif
 657 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 658         if (test_opt(sb, POSIX_ACL))
 659                 seq_puts(seq, ",acl");
 660         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
 661                 seq_puts(seq, ",noacl");
 662 #endif
 663         if (!test_opt(sb, RESERVATION))
 664                 seq_puts(seq, ",noreservation");
 665         if (sbi->s_commit_interval) {
 666                 seq_printf(seq, ",commit=%u",
 667                            (unsigned) (sbi->s_commit_interval / HZ));
 668         }
 669
 670         /*
 671          * Always display barrier state so it's clear what the status is.
 672          */
 673         seq_puts(seq, ",barrier=");
 674         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 675         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
 676         if (test_opt(sb, DATA_ERR_ABORT))
 677                 seq_puts(seq, ",data_err=abort");
 678
 679         if (test_opt(sb, NOLOAD))
 680                 seq_puts(seq, ",norecovery");
 681
 682         ext3_show_quota_options(seq, sb);
 683
 684         return 0;
 685 }
 686
 687
 688 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
 689                 u64 ino, u32 generation)
 690 {
 691         struct inode *inode;
 692
 693         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 694                 return ERR_PTR(-ESTALE);
 695         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 696                 return ERR_PTR(-ESTALE);
 697
 698         /* iget isn't really right if the inode is currently unallocated!!
 699          *
 700          * ext3_read_inode will return a bad_inode if the inode had been
 701          * deleted, so we should be safe.
 702          *
 703          * Currently we don't know the generation for parent directory, so
 704          * a generation of 0 means "accept any"
 705          */
 706         inode = ext3_iget(sb, ino);
 707         if (IS_ERR(inode))
 708                 return ERR_CAST(inode);
 709         if (generation && inode->i_generation != generation) {
 710                 iput(inode);
 711                 return ERR_PTR(-ESTALE);
 712         }
 713
 714         return inode;
 715 }
 716
 717 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
 718                 int fh_len, int fh_type)
 719 {
 720         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 721                                     ext3_nfs_get_inode);
 722 }
 723
 724 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
 725                 int fh_len, int fh_type)
 726 {
 727         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 728                                     ext3_nfs_get_inode);
 729 }
 730
 731 /*
 732  * Try to release metadata pages (indirect blocks, directories) which are
 733  * mapped via the block device.  Since these pages could have journal heads
 734  * which would prevent try_to_free_buffers() from freeing them, we must use
 735  * jbd layer's try_to_free_buffers() function to release them.
 736  */
 737 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 738                                  gfp_t wait)
 739 {
 740         journal_t *journal = EXT3_SB(sb)->s_journal;
 741
 742         WARN_ON(PageChecked(page));
 743         if (!page_has_buffers(page))
 744                 return 0;
 745         if (journal)
 746                 return journal_try_to_free_buffers(journal, page,
 747                                                    wait & ~__GFP_WAIT);
 748         return try_to_free_buffers(page);
 749 }
 750
 751 #ifdef CONFIG_QUOTA
 752 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 753 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 754
 755 static int ext3_write_dquot(struct dquot *dquot);
 756 static int ext3_acquire_dquot(struct dquot *dquot);
 757 static int ext3_release_dquot(struct dquot *dquot);
 758 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 759 static int ext3_write_info(struct super_block *sb, int type);
 760 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
 761                          struct path *path);
 762 static int ext3_quota_on_mount(struct super_block *sb, int type);
 763 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 764                                size_t len, loff_t off);
 765 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 766                                 const char *data, size_t len, loff_t off);
 767
 768 static const struct dquot_operations ext3_quota_operations = {
 769         .write_dquot    = ext3_write_dquot,
 770         .acquire_dquot  = ext3_acquire_dquot,
 771         .release_dquot  = ext3_release_dquot,
 772         .mark_dirty     = ext3_mark_dquot_dirty,
 773         .write_info     = ext3_write_info,
 774         .alloc_dquot    = dquot_alloc,
 775         .destroy_dquot  = dquot_destroy,
 776 };
 777
 778 static const struct quotactl_ops ext3_qctl_operations = {
 779         .quota_on       = ext3_quota_on,
 780         .quota_off      = dquot_quota_off,
 781         .quota_sync     = dquot_quota_sync,
 782         .get_info       = dquot_get_dqinfo,
 783         .set_info       = dquot_set_dqinfo,
 784         .get_dqblk      = dquot_get_dqblk,
 785         .set_dqblk      = dquot_set_dqblk
 786 };
 787 #endif
 788
 789 static const struct super_operations ext3_sops = {
 790         .alloc_inode    = ext3_alloc_inode,
 791         .destroy_inode  = ext3_destroy_inode,
 792         .write_inode    = ext3_write_inode,
 793         .dirty_inode    = ext3_dirty_inode,
 794         .drop_inode     = ext3_drop_inode,
 795         .evict_inode    = ext3_evict_inode,
 796         .put_super      = ext3_put_super,
 797         .sync_fs        = ext3_sync_fs,
 798         .freeze_fs      = ext3_freeze,
 799         .unfreeze_fs    = ext3_unfreeze,
 800         .statfs         = ext3_statfs,
 801         .remount_fs     = ext3_remount,
 802         .show_options   = ext3_show_options,
 803 #ifdef CONFIG_QUOTA
 804         .quota_read     = ext3_quota_read,
 805         .quota_write    = ext3_quota_write,
 806 #endif
 807         .bdev_try_to_free_page = bdev_try_to_free_page,
 808 };
 809
 810 static const struct export_operations ext3_export_ops = {
 811         .fh_to_dentry = ext3_fh_to_dentry,
 812         .fh_to_parent = ext3_fh_to_parent,
 813         .get_parent = ext3_get_parent,
 814 };
 815
 816 enum {
 817         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 818         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 819         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 820         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 821         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 822         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 823         Opt_journal_path,
 824         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 825         Opt_data_err_abort, Opt_data_err_ignore,
 826         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 827         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 828         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
 829         Opt_resize, Opt_usrquota, Opt_grpquota
 830 };
 831
 832 static const match_table_t tokens = {
 833         {Opt_bsd_df, "bsddf"},
 834         {Opt_minix_df, "minixdf"},
 835         {Opt_grpid, "grpid"},
 836         {Opt_grpid, "bsdgroups"},
 837         {Opt_nogrpid, "nogrpid"},
 838         {Opt_nogrpid, "sysvgroups"},
 839         {Opt_resgid, "resgid=%u"},
 840         {Opt_resuid, "resuid=%u"},
 841         {Opt_sb, "sb=%u"},
 842         {Opt_err_cont, "errors=continue"},
 843         {Opt_err_panic, "errors=panic"},
 844         {Opt_err_ro, "errors=remount-ro"},
 845         {Opt_nouid32, "nouid32"},
 846         {Opt_nocheck, "nocheck"},
 847         {Opt_nocheck, "check=none"},
 848         {Opt_debug, "debug"},
 849         {Opt_oldalloc, "oldalloc"},
 850         {Opt_orlov, "orlov"},
 851         {Opt_user_xattr, "user_xattr"},
 852         {Opt_nouser_xattr, "nouser_xattr"},
 853         {Opt_acl, "acl"},
 854         {Opt_noacl, "noacl"},
 855         {Opt_reservation, "reservation"},
 856         {Opt_noreservation, "noreservation"},
 857         {Opt_noload, "noload"},
 858         {Opt_noload, "norecovery"},
 859         {Opt_nobh, "nobh"},
 860         {Opt_bh, "bh"},
 861         {Opt_commit, "commit=%u"},
 862         {Opt_journal_update, "journal=update"},
 863         {Opt_journal_inum, "journal=%u"},
 864         {Opt_journal_dev, "journal_dev=%u"},
 865         {Opt_journal_path, "journal_path=%s"},
 866         {Opt_abort, "abort"},
 867         {Opt_data_journal, "data=journal"},
 868         {Opt_data_ordered, "data=ordered"},
 869         {Opt_data_writeback, "data=writeback"},
 870         {Opt_data_err_abort, "data_err=abort"},
 871         {Opt_data_err_ignore, "data_err=ignore"},
 872         {Opt_offusrjquota, "usrjquota="},
 873         {Opt_usrjquota, "usrjquota=%s"},
 874         {Opt_offgrpjquota, "grpjquota="},
 875         {Opt_grpjquota, "grpjquota=%s"},
 876         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 877         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 878         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
 879         {Opt_grpquota, "grpquota"},
 880         {Opt_noquota, "noquota"},
 881         {Opt_quota, "quota"},
 882         {Opt_usrquota, "usrquota"},
 883         {Opt_barrier, "barrier=%u"},
 884         {Opt_barrier, "barrier"},
 885         {Opt_nobarrier, "nobarrier"},
 886         {Opt_resize, "resize"},
 887         {Opt_err, NULL},
 888 };
 889
 890 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
 891 {
 892         ext3_fsblk_t    sb_block;
 893         char            *options = (char *) *data;
 894
 895         if (!options || strncmp(options, "sb=", 3) != 0)
 896                 return 1;       /* Default location */
 897         options += 3;
 898         /*todo: use simple_strtoll with >32bit ext3 */
 899         sb_block = simple_strtoul(options, &options, 0);
 900         if (*options && *options != ',') {
 901                 ext3_msg(sb, KERN_ERR, "error: invalid sb specification: %s",
 902                        (char *) *data);
 903                 return 1;
 904         }
 905         if (*options == ',')
 906                 options++;
 907         *data = (void *) options;
 908         return sb_block;
 909 }
 910
 911 #ifdef CONFIG_QUOTA
 912 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
 913 {
 914         struct ext3_sb_info *sbi = EXT3_SB(sb);
 915         char *qname;
 916
 917         if (sb_any_quota_loaded(sb) &&
 918                 !sbi->s_qf_names[qtype]) {
 919                 ext3_msg(sb, KERN_ERR,
 920                         "Cannot change journaled "
 921                         "quota options when quota turned on");
 922                 return 0;
 923         }
 924         qname = match_strdup(args);
 925         if (!qname) {
 926                 ext3_msg(sb, KERN_ERR,
 927                         "Not enough memory for storing quotafile name");
 928                 return 0;
 929         }
 930         if (sbi->s_qf_names[qtype]) {
 931                 int same = !strcmp(sbi->s_qf_names[qtype], qname);
 932
 933                 kfree(qname);
 934                 if (!same) {
 935                         ext3_msg(sb, KERN_ERR,
 936                                  "%s quota file already specified",
 937                                  QTYPE2NAME(qtype));
 938                 }
 939                 return same;
 940         }
 941         if (strchr(qname, '/')) {
 942                 ext3_msg(sb, KERN_ERR,
 943                         "quotafile must be on filesystem root");
 944                 kfree(qname);
 945                 return 0;
 946         }
 947         sbi->s_qf_names[qtype] = qname;
 948         set_opt(sbi->s_mount_opt, QUOTA);
 949         return 1;
 950 }
 951
 952 static int clear_qf_name(struct super_block *sb, int qtype) {
 953
 954         struct ext3_sb_info *sbi = EXT3_SB(sb);
 955
 956         if (sb_any_quota_loaded(sb) &&
 957                 sbi->s_qf_names[qtype]) {
 958                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
 959                         " when quota turned on");
 960                 return 0;
 961         }
 962         if (sbi->s_qf_names[qtype]) {
 963                 kfree(sbi->s_qf_names[qtype]);
 964                 sbi->s_qf_names[qtype] = NULL;
 965         }
 966         return 1;
 967 }
 968 #endif
 969
 970 static int parse_options (char *options, struct super_block *sb,
 971                           unsigned int *inum, unsigned long *journal_devnum,
 972                           ext3_fsblk_t *n_blocks_count, int is_remount)
 973 {
 974         struct ext3_sb_info *sbi = EXT3_SB(sb);
 975         char * p;
 976         substring_t args[MAX_OPT_ARGS];
 977         int data_opt = 0;
 978         int option;
 979         kuid_t uid;
 980         kgid_t gid;
 981         char *journal_path;
 982         struct inode *journal_inode;
 983         struct path path;
 984         int error;
 985
 986 #ifdef CONFIG_QUOTA
 987         int qfmt;
 988 #endif
 989
 990         if (!options)
 991                 return 1;
 992
 993         while ((p = strsep (&options, ",")) != NULL) {
 994                 int token;
 995                 if (!*p)
 996                         continue;
 997                 /*
 998                  * Initialize args struct so we know whether arg was
 999                  * found; some options take optional arguments.
1000                  */
1001                 args[0].to = args[0].from = NULL;
1002                 token = match_token(p, tokens, args);
1003                 switch (token) {
1004                 case Opt_bsd_df:
1005                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1006                         break;
1007                 case Opt_minix_df:
1008                         set_opt (sbi->s_mount_opt, MINIX_DF);
1009                         break;
1010                 case Opt_grpid:
1011                         set_opt (sbi->s_mount_opt, GRPID);
1012                         break;
1013                 case Opt_nogrpid:
1014                         clear_opt (sbi->s_mount_opt, GRPID);
1015                         break;
1016                 case Opt_resuid:
1017                         if (match_int(&args[0], &option))
1018                                 return 0;
1019                         uid = make_kuid(current_user_ns(), option);
1020                         if (!uid_valid(uid)) {
1021                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1022                                 return 0;
1023
1024                         }
1025                         sbi->s_resuid = uid;
1026                         break;
1027                 case Opt_resgid:
1028                         if (match_int(&args[0], &option))
1029                                 return 0;
1030                         gid = make_kgid(current_user_ns(), option);
1031                         if (!gid_valid(gid)) {
1032                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1033                                 return 0;
1034                         }
1035                         sbi->s_resgid = gid;
1036                         break;
1037                 case Opt_sb:
1038                         /* handled by get_sb_block() instead of here */
1039                         /* *sb_block = match_int(&args[0]); */
1040                         break;
1041                 case Opt_err_panic:
1042                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1043                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1044                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1045                         break;
1046                 case Opt_err_ro:
1047                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1048                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1049                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1050                         break;
1051                 case Opt_err_cont:
1052                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1053                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1054                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1055                         break;
1056                 case Opt_nouid32:
1057                         set_opt (sbi->s_mount_opt, NO_UID32);
1058                         break;
1059                 case Opt_nocheck:
1060                         clear_opt (sbi->s_mount_opt, CHECK);
1061                         break;
1062                 case Opt_debug:
1063                         set_opt (sbi->s_mount_opt, DEBUG);
1064                         break;
1065                 case Opt_oldalloc:
1066                         ext3_msg(sb, KERN_WARNING,
1067                                 "Ignoring deprecated oldalloc option");
1068                         break;
1069                 case Opt_orlov:
1070                         ext3_msg(sb, KERN_WARNING,
1071                                 "Ignoring deprecated orlov option");
1072                         break;
1073 #ifdef CONFIG_EXT3_FS_XATTR
1074                 case Opt_user_xattr:
1075                         set_opt (sbi->s_mount_opt, XATTR_USER);
1076                         break;
1077                 case Opt_nouser_xattr:
1078                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1079                         break;
1080 #else
1081                 case Opt_user_xattr:
1082                 case Opt_nouser_xattr:
1083                         ext3_msg(sb, KERN_INFO,
1084                                 "(no)user_xattr options not supported");
1085                         break;
1086 #endif
1087 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1088                 case Opt_acl:
1089                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1090                         break;
1091                 case Opt_noacl:
1092                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1093                         break;
1094 #else
1095                 case Opt_acl:
1096                 case Opt_noacl:
1097                         ext3_msg(sb, KERN_INFO,
1098                                 "(no)acl options not supported");
1099                         break;
1100 #endif
1101                 case Opt_reservation:
1102                         set_opt(sbi->s_mount_opt, RESERVATION);
1103                         break;
1104                 case Opt_noreservation:
1105                         clear_opt(sbi->s_mount_opt, RESERVATION);
1106                         break;
1107                 case Opt_journal_update:
1108                         /* @@@ FIXME */
1109                         /* Eventually we will want to be able to create
1110                            a journal file here.  For now, only allow the
1111                            user to specify an existing inode to be the
1112                            journal file. */
1113                         if (is_remount) {
1114                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1115                                         "journal on remount");
1116                                 return 0;
1117                         }
1118                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1119                         break;
1120                 case Opt_journal_inum:
1121                         if (is_remount) {
1122                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1123                                        "journal on remount");
1124                                 return 0;
1125                         }
1126                         if (match_int(&args[0], &option))
1127                                 return 0;
1128                         *inum = option;
1129                         break;
1130                 case Opt_journal_dev:
1131                         if (is_remount) {
1132                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1133                                        "journal on remount");
1134                                 return 0;
1135                         }
1136                         if (match_int(&args[0], &option))
1137                                 return 0;
1138                         *journal_devnum = option;
1139                         break;
1140                 case Opt_journal_path:
1141                         if (is_remount) {
1142                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1143                                        "journal on remount");
1144                                 return 0;
1145                         }
1146
1147                         journal_path = match_strdup(&args[0]);
1148                         if (!journal_path) {
1149                                 ext3_msg(sb, KERN_ERR, "error: could not dup "
1150                                         "journal device string");
1151                                 return 0;
1152                         }
1153
1154                         error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1155                         if (error) {
1156                                 ext3_msg(sb, KERN_ERR, "error: could not find "
1157                                         "journal device path: error %d", error);
1158                                 kfree(journal_path);
1159                                 return 0;
1160                         }
1161
1162                         journal_inode = path.dentry->d_inode;
1163                         if (!S_ISBLK(journal_inode->i_mode)) {
1164                                 ext3_msg(sb, KERN_ERR, "error: journal path %s "
1165                                         "is not a block device", journal_path);
1166                                 path_put(&path);
1167                                 kfree(journal_path);
1168                                 return 0;
1169                         }
1170
1171                         *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1172                         path_put(&path);
1173                         kfree(journal_path);
1174                         break;
1175                 case Opt_noload:
1176                         set_opt (sbi->s_mount_opt, NOLOAD);
1177                         break;
1178                 case Opt_commit:
1179                         if (match_int(&args[0], &option))
1180                                 return 0;
1181                         if (option < 0)
1182                                 return 0;
1183                         if (option == 0)
1184                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1185                         sbi->s_commit_interval = HZ * option;
1186                         break;
1187                 case Opt_data_journal:
1188                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1189                         goto datacheck;
1190                 case Opt_data_ordered:
1191                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1192                         goto datacheck;
1193                 case Opt_data_writeback:
1194                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1195                 datacheck:
1196                         if (is_remount) {
1197                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1198                                         break;
1199                                 ext3_msg(sb, KERN_ERR,
1200                                         "error: cannot change "
1201                                         "data mode on remount. The filesystem "
1202                                         "is mounted in data=%s mode and you "
1203                                         "try to remount it in data=%s mode.",
1204                                         data_mode_string(test_opt(sb,
1205                                                         DATA_FLAGS)),
1206                                         data_mode_string(data_opt));
1207                                 return 0;
1208                         } else {
1209                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1210                                 sbi->s_mount_opt |= data_opt;
1211                         }
1212                         break;
1213                 case Opt_data_err_abort:
1214                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1215                         break;
1216                 case Opt_data_err_ignore:
1217                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1218                         break;
1219 #ifdef CONFIG_QUOTA
1220                 case Opt_usrjquota:
1221                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1222                                 return 0;
1223                         break;
1224                 case Opt_grpjquota:
1225                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1226                                 return 0;
1227                         break;
1228                 case Opt_offusrjquota:
1229                         if (!clear_qf_name(sb, USRQUOTA))
1230                                 return 0;
1231                         break;
1232                 case Opt_offgrpjquota:
1233                         if (!clear_qf_name(sb, GRPQUOTA))
1234                                 return 0;
1235                         break;
1236                 case Opt_jqfmt_vfsold:
1237                         qfmt = QFMT_VFS_OLD;
1238                         goto set_qf_format;
1239                 case Opt_jqfmt_vfsv0:
1240                         qfmt = QFMT_VFS_V0;
1241                         goto set_qf_format;
1242                 case Opt_jqfmt_vfsv1:
1243                         qfmt = QFMT_VFS_V1;
1244 set_qf_format:
1245                         if (sb_any_quota_loaded(sb) &&
1246                             sbi->s_jquota_fmt != qfmt) {
1247                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1248                                         "journaled quota options when "
1249                                         "quota turned on.");
1250                                 return 0;
1251                         }
1252                         sbi->s_jquota_fmt = qfmt;
1253                         break;
1254                 case Opt_quota:
1255                 case Opt_usrquota:
1256                         set_opt(sbi->s_mount_opt, QUOTA);
1257                         set_opt(sbi->s_mount_opt, USRQUOTA);
1258                         break;
1259                 case Opt_grpquota:
1260                         set_opt(sbi->s_mount_opt, QUOTA);
1261                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1262                         break;
1263                 case Opt_noquota:
1264                         if (sb_any_quota_loaded(sb)) {
1265                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1266                                         "quota options when quota turned on.");
1267                                 return 0;
1268                         }
1269                         clear_opt(sbi->s_mount_opt, QUOTA);
1270                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1271                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1272                         break;
1273 #else
1274                 case Opt_quota:
1275                 case Opt_usrquota:
1276                 case Opt_grpquota:
1277                         ext3_msg(sb, KERN_ERR,
1278                                 "error: quota options not supported.");
1279                         break;
1280                 case Opt_usrjquota:
1281                 case Opt_grpjquota:
1282                 case Opt_offusrjquota:
1283                 case Opt_offgrpjquota:
1284                 case Opt_jqfmt_vfsold:
1285                 case Opt_jqfmt_vfsv0:
1286                 case Opt_jqfmt_vfsv1:
1287                         ext3_msg(sb, KERN_ERR,
1288                                 "error: journaled quota options not "
1289                                 "supported.");
1290                         break;
1291                 case Opt_noquota:
1292                         break;
1293 #endif
1294                 case Opt_abort:
1295                         set_opt(sbi->s_mount_opt, ABORT);
1296                         break;
1297                 case Opt_nobarrier:
1298                         clear_opt(sbi->s_mount_opt, BARRIER);
1299                         break;
1300                 case Opt_barrier:
1301                         if (args[0].from) {
1302                                 if (match_int(&args[0], &option))
1303                                         return 0;
1304                         } else
1305                                 option = 1;     /* No argument, default to 1 */
1306                         if (option)
1307                                 set_opt(sbi->s_mount_opt, BARRIER);
1308                         else
1309                                 clear_opt(sbi->s_mount_opt, BARRIER);
1310                         break;
1311                 case Opt_ignore:
1312                         break;
1313                 case Opt_resize:
1314                         if (!is_remount) {
1315                                 ext3_msg(sb, KERN_ERR,
1316                                         "error: resize option only available "
1317                                         "for remount");
1318                                 return 0;
1319                         }
1320                         if (match_int(&args[0], &option) != 0)
1321                                 return 0;
1322                         *n_blocks_count = option;
1323                         break;
1324                 case Opt_nobh:
1325                         ext3_msg(sb, KERN_WARNING,
1326                                 "warning: ignoring deprecated nobh option");
1327                         break;
1328                 case Opt_bh:
1329                         ext3_msg(sb, KERN_WARNING,
1330                                 "warning: ignoring deprecated bh option");
1331                         break;
1332                 default:
1333                         ext3_msg(sb, KERN_ERR,
1334                                 "error: unrecognized mount option \"%s\" "
1335                                 "or missing value", p);
1336                         return 0;
1337                 }
1338         }
1339 #ifdef CONFIG_QUOTA
1340         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1341                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1342                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1343                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1344                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1345
1346                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1347                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1348                                         "format mixing.");
1349                         return 0;
1350                 }
1351
1352                 if (!sbi->s_jquota_fmt) {
1353                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1354                                         "not specified.");
1355                         return 0;
1356                 }
1357         }
1358 #endif
1359         return 1;
1360 }
1361
1362 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1363                             int read_only)
1364 {
1365         struct ext3_sb_info *sbi = EXT3_SB(sb);
1366         int res = 0;
1367
1368         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1369                 ext3_msg(sb, KERN_ERR,
1370                         "error: revision level too high, "
1371                         "forcing read-only mode");
1372                 res = MS_RDONLY;
1373         }
1374         if (read_only)
1375                 return res;
1376         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1377                 ext3_msg(sb, KERN_WARNING,
1378                         "warning: mounting unchecked fs, "
1379                         "running e2fsck is recommended");
1380         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1381                 ext3_msg(sb, KERN_WARNING,
1382                         "warning: mounting fs with errors, "
1383                         "running e2fsck is recommended");
1384         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1385                  le16_to_cpu(es->s_mnt_count) >=
1386                         le16_to_cpu(es->s_max_mnt_count))
1387                 ext3_msg(sb, KERN_WARNING,
1388                         "warning: maximal mount count reached, "
1389                         "running e2fsck is recommended");
1390         else if (le32_to_cpu(es->s_checkinterval) &&
1391                 (le32_to_cpu(es->s_lastcheck) +
1392                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1393                 ext3_msg(sb, KERN_WARNING,
1394                         "warning: checktime reached, "
1395                         "running e2fsck is recommended");
1396 #if 0
1397                 /* @@@ We _will_ want to clear the valid bit if we find
1398                    inconsistencies, to force a fsck at reboot.  But for
1399                    a plain journaled filesystem we can keep it set as
1400                    valid forever! :) */
1401         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1402 #endif
1403         if (!le16_to_cpu(es->s_max_mnt_count))
1404                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1405         le16_add_cpu(&es->s_mnt_count, 1);
1406         es->s_mtime = cpu_to_le32(get_seconds());
1407         ext3_update_dynamic_rev(sb);
1408         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1409
1410         ext3_commit_super(sb, es, 1);
1411         if (test_opt(sb, DEBUG))
1412                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1413                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1414                         sb->s_blocksize,
1415                         sbi->s_groups_count,
1416                         EXT3_BLOCKS_PER_GROUP(sb),
1417                         EXT3_INODES_PER_GROUP(sb),
1418                         sbi->s_mount_opt);
1419
1420         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1421                 char b[BDEVNAME_SIZE];
1422                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1423                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1424         } else {
1425                 ext3_msg(sb, KERN_INFO, "using internal journal");
1426         }
1427         cleancache_init_fs(sb);
1428         return res;
1429 }
1430
1431 /* Called at mount-time, super-block is locked */
1432 static int ext3_check_descriptors(struct super_block *sb)
1433 {
1434         struct ext3_sb_info *sbi = EXT3_SB(sb);
1435         int i;
1436
1437         ext3_debug ("Checking group descriptors");
1438
1439         for (i = 0; i < sbi->s_groups_count; i++) {
1440                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1441                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1442                 ext3_fsblk_t last_block;
1443
1444                 if (i == sbi->s_groups_count - 1)
1445                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1446                 else
1447                         last_block = first_block +
1448                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1449
1450                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1451                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1452                 {
1453                         ext3_error (sb, "ext3_check_descriptors",
1454                                     "Block bitmap for group %d"
1455                                     " not in group (block %lu)!",
1456                                     i, (unsigned long)
1457                                         le32_to_cpu(gdp->bg_block_bitmap));
1458                         return 0;
1459                 }
1460                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1461                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1462                 {
1463                         ext3_error (sb, "ext3_check_descriptors",
1464                                     "Inode bitmap for group %d"
1465                                     " not in group (block %lu)!",
1466                                     i, (unsigned long)
1467                                         le32_to_cpu(gdp->bg_inode_bitmap));
1468                         return 0;
1469                 }
1470                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1471                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1472                     last_block)
1473                 {
1474                         ext3_error (sb, "ext3_check_descriptors",
1475                                     "Inode table for group %d"
1476                                     " not in group (block %lu)!",
1477                                     i, (unsigned long)
1478                                         le32_to_cpu(gdp->bg_inode_table));
1479                         return 0;
1480                 }
1481         }
1482
1483         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1484         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1485         return 1;
1486 }
1487
1488
1489 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1490  * the superblock) which were deleted from all directories, but held open by
1491  * a process at the time of a crash.  We walk the list and try to delete these
1492  * inodes at recovery time (only with a read-write filesystem).
1493  *
1494  * In order to keep the orphan inode chain consistent during traversal (in
1495  * case of crash during recovery), we link each inode into the superblock
1496  * orphan list_head and handle it the same way as an inode deletion during
1497  * normal operation (which journals the operations for us).
1498  *
1499  * We only do an iget() and an iput() on each inode, which is very safe if we
1500  * accidentally point at an in-use or already deleted inode.  The worst that
1501  * can happen in this case is that we get a "bit already cleared" message from
1502  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1503  * e2fsck was run on this filesystem, and it must have already done the orphan
1504  * inode cleanup for us, so we can safely abort without any further action.
1505  */
1506 static void ext3_orphan_cleanup (struct super_block * sb,
1507                                  struct ext3_super_block * es)
1508 {
1509         unsigned int s_flags = sb->s_flags;
1510         int nr_orphans = 0, nr_truncates = 0;
1511 #ifdef CONFIG_QUOTA
1512         int i;
1513 #endif
1514         if (!es->s_last_orphan) {
1515                 jbd_debug(4, "no orphan inodes to clean up\n");
1516                 return;
1517         }
1518
1519         if (bdev_read_only(sb->s_bdev)) {
1520                 ext3_msg(sb, KERN_ERR, "error: write access "
1521                         "unavailable, skipping orphan cleanup.");
1522                 return;
1523         }
1524
1525         /* Check if feature set allows readwrite operations */
1526         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1527                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1528                          "unknown ROCOMPAT features");
1529                 return;
1530         }
1531
1532         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1533                 /* don't clear list on RO mount w/ errors */
1534                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
1535                         jbd_debug(1, "Errors on filesystem, "
1536                                   "clearing orphan list.\n");
1537                         es->s_last_orphan = 0;
1538                 }
1539                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1540                 return;
1541         }
1542
1543         if (s_flags & MS_RDONLY) {
1544                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1545                 sb->s_flags &= ~MS_RDONLY;
1546         }
1547 #ifdef CONFIG_QUOTA
1548         /* Needed for iput() to work correctly and not trash data */
1549         sb->s_flags |= MS_ACTIVE;
1550         /* Turn on quotas so that they are updated correctly */
1551         for (i = 0; i < MAXQUOTAS; i++) {
1552                 if (EXT3_SB(sb)->s_qf_names[i]) {
1553                         int ret = ext3_quota_on_mount(sb, i);
1554                         if (ret < 0)
1555                                 ext3_msg(sb, KERN_ERR,
1556                                         "error: cannot turn on journaled "
1557                                         "quota: %d", ret);
1558                 }
1559         }
1560 #endif
1561
1562         while (es->s_last_orphan) {
1563                 struct inode *inode;
1564
1565                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1566                 if (IS_ERR(inode)) {
1567                         es->s_last_orphan = 0;
1568                         break;
1569                 }
1570
1571                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1572                 dquot_initialize(inode);
1573                 if (inode->i_nlink) {
1574                         printk(KERN_DEBUG
1575                                 "%s: truncating inode %lu to %Ld bytes\n",
1576                                 __func__, inode->i_ino, inode->i_size);
1577                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1578                                   inode->i_ino, inode->i_size);
1579                         ext3_truncate(inode);
1580                         nr_truncates++;
1581                 } else {
1582                         printk(KERN_DEBUG
1583                                 "%s: deleting unreferenced inode %lu\n",
1584                                 __func__, inode->i_ino);
1585                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1586                                   inode->i_ino);
1587                         nr_orphans++;
1588                 }
1589                 iput(inode);  /* The delete magic happens here! */
1590         }
1591
1592 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1593
1594         if (nr_orphans)
1595                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1596                        PLURAL(nr_orphans));
1597         if (nr_truncates)
1598                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1599                        PLURAL(nr_truncates));
1600 #ifdef CONFIG_QUOTA
1601         /* Turn quotas off */
1602         for (i = 0; i < MAXQUOTAS; i++) {
1603                 if (sb_dqopt(sb)->files[i])
1604                         dquot_quota_off(sb, i);
1605         }
1606 #endif
1607         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1608 }
1609
1610 /*
1611  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1612  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1613  * We need to be 1 filesystem block less than the 2^32 sector limit.
1614  */
1615 static loff_t ext3_max_size(int bits)
1616 {
1617         loff_t res = EXT3_NDIR_BLOCKS;
1618         int meta_blocks;
1619         loff_t upper_limit;
1620
1621         /* This is calculated to be the largest file size for a
1622          * dense, file such that the total number of
1623          * sectors in the file, including data and all indirect blocks,
1624          * does not exceed 2^32 -1
1625          * __u32 i_blocks representing the total number of
1626          * 512 bytes blocks of the file
1627          */
1628         upper_limit = (1LL << 32) - 1;
1629
1630         /* total blocks in file system block size */
1631         upper_limit >>= (bits - 9);
1632
1633
1634         /* indirect blocks */
1635         meta_blocks = 1;
1636         /* double indirect blocks */
1637         meta_blocks += 1 + (1LL << (bits-2));
1638         /* tripple indirect blocks */
1639         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1640
1641         upper_limit -= meta_blocks;
1642         upper_limit <<= bits;
1643
1644         res += 1LL << (bits-2);
1645         res += 1LL << (2*(bits-2));
1646         res += 1LL << (3*(bits-2));
1647         res <<= bits;
1648         if (res > upper_limit)
1649                 res = upper_limit;
1650
1651         if (res > MAX_LFS_FILESIZE)
1652                 res = MAX_LFS_FILESIZE;
1653
1654         return res;
1655 }
1656
1657 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1658                                     ext3_fsblk_t logic_sb_block,
1659                                     int nr)
1660 {
1661         struct ext3_sb_info *sbi = EXT3_SB(sb);
1662         unsigned long bg, first_meta_bg;
1663         int has_super = 0;
1664
1665         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1666
1667         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1668             nr < first_meta_bg)
1669                 return (logic_sb_block + nr + 1);
1670         bg = sbi->s_desc_per_block * nr;
1671         if (ext3_bg_has_super(sb, bg))
1672                 has_super = 1;
1673         return (has_super + ext3_group_first_block_no(sb, bg));
1674 }
1675
1676
1677 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1678 {
1679         struct buffer_head * bh;
1680         struct ext3_super_block *es = NULL;
1681         struct ext3_sb_info *sbi;
1682         ext3_fsblk_t block;
1683         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1684         ext3_fsblk_t logic_sb_block;
1685         unsigned long offset = 0;
1686         unsigned int journal_inum = 0;
1687         unsigned long journal_devnum = 0;
1688         unsigned long def_mount_opts;
1689         struct inode *root;
1690         int blocksize;
1691         int hblock;
1692         int db_count;
1693         int i;
1694         int needs_recovery;
1695         int ret = -EINVAL;
1696         __le32 features;
1697         int err;
1698
1699         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1700         if (!sbi)
1701                 return -ENOMEM;
1702
1703         sbi->s_blockgroup_lock =
1704                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1705         if (!sbi->s_blockgroup_lock) {
1706                 kfree(sbi);
1707                 return -ENOMEM;
1708         }
1709         sb->s_fs_info = sbi;
1710         sbi->s_sb_block = sb_block;
1711
1712         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1713         if (!blocksize) {
1714                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1715                 goto out_fail;
1716         }
1717
1718         /*
1719          * The ext3 superblock will not be buffer aligned for other than 1kB
1720          * block sizes.  We need to calculate the offset from buffer start.
1721          */
1722         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1723                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1724                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1725         } else {
1726                 logic_sb_block = sb_block;
1727         }
1728
1729         if (!(bh = sb_bread(sb, logic_sb_block))) {
1730                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1731                 goto out_fail;
1732         }
1733         /*
1734          * Note: s_es must be initialized as soon as possible because
1735          *       some ext3 macro-instructions depend on its value
1736          */
1737         es = (struct ext3_super_block *) (bh->b_data + offset);
1738         sbi->s_es = es;
1739         sb->s_magic = le16_to_cpu(es->s_magic);
1740         if (sb->s_magic != EXT3_SUPER_MAGIC)
1741                 goto cantfind_ext3;
1742
1743         /* Set defaults before we parse the mount options */
1744         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1745         if (def_mount_opts & EXT3_DEFM_DEBUG)
1746                 set_opt(sbi->s_mount_opt, DEBUG);
1747         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1748                 set_opt(sbi->s_mount_opt, GRPID);
1749         if (def_mount_opts & EXT3_DEFM_UID16)
1750                 set_opt(sbi->s_mount_opt, NO_UID32);
1751 #ifdef CONFIG_EXT3_FS_XATTR
1752         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1753                 set_opt(sbi->s_mount_opt, XATTR_USER);
1754 #endif
1755 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1756         if (def_mount_opts & EXT3_DEFM_ACL)
1757                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1758 #endif
1759         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1760                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1761         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1762                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1763         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1764                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1765
1766         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1767                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1768         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1769                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1770         else
1771                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1772
1773         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1774         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1775
1776         /* enable barriers by default */
1777         set_opt(sbi->s_mount_opt, BARRIER);
1778         set_opt(sbi->s_mount_opt, RESERVATION);
1779
1780         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1781                             NULL, 0))
1782                 goto failed_mount;
1783
1784         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1785                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1786
1787         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1788             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1789              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1790              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1791                 ext3_msg(sb, KERN_WARNING,
1792                         "warning: feature flags set on rev 0 fs, "
1793                         "running e2fsck is recommended");
1794         /*
1795          * Check feature flags regardless of the revision level, since we
1796          * previously didn't change the revision level when setting the flags,
1797          * so there is a chance incompat flags are set on a rev 0 filesystem.
1798          */
1799         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1800         if (features) {
1801                 ext3_msg(sb, KERN_ERR,
1802                         "error: couldn't mount because of unsupported "
1803                         "optional features (%x)", le32_to_cpu(features));
1804                 goto failed_mount;
1805         }
1806         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1807         if (!(sb->s_flags & MS_RDONLY) && features) {
1808                 ext3_msg(sb, KERN_ERR,
1809                         "error: couldn't mount RDWR because of unsupported "
1810                         "optional features (%x)", le32_to_cpu(features));
1811                 goto failed_mount;
1812         }
1813         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1814
1815         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1816             blocksize > EXT3_MAX_BLOCK_SIZE) {
1817                 ext3_msg(sb, KERN_ERR,
1818                         "error: couldn't mount because of unsupported "
1819                         "filesystem blocksize %d", blocksize);
1820                 goto failed_mount;
1821         }
1822
1823         hblock = bdev_logical_block_size(sb->s_bdev);
1824         if (sb->s_blocksize != blocksize) {
1825                 /*
1826                  * Make sure the blocksize for the filesystem is larger
1827                  * than the hardware sectorsize for the machine.
1828                  */
1829                 if (blocksize < hblock) {
1830                         ext3_msg(sb, KERN_ERR,
1831                                 "error: fsblocksize %d too small for "
1832                                 "hardware sectorsize %d", blocksize, hblock);
1833                         goto failed_mount;
1834                 }
1835
1836                 brelse (bh);
1837                 if (!sb_set_blocksize(sb, blocksize)) {
1838                         ext3_msg(sb, KERN_ERR,
1839                                 "error: bad blocksize %d", blocksize);
1840                         goto out_fail;
1841                 }
1842                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1843                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1844                 bh = sb_bread(sb, logic_sb_block);
1845                 if (!bh) {
1846                         ext3_msg(sb, KERN_ERR,
1847                                "error: can't read superblock on 2nd try");
1848                         goto failed_mount;
1849                 }
1850                 es = (struct ext3_super_block *)(bh->b_data + offset);
1851                 sbi->s_es = es;
1852                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1853                         ext3_msg(sb, KERN_ERR,
1854                                 "error: magic mismatch");
1855                         goto failed_mount;
1856                 }
1857         }
1858
1859         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1860
1861         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1862                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1863                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1864         } else {
1865                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1866                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1867                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1868                     (!is_power_of_2(sbi->s_inode_size)) ||
1869                     (sbi->s_inode_size > blocksize)) {
1870                         ext3_msg(sb, KERN_ERR,
1871                                 "error: unsupported inode size: %d",
1872                                 sbi->s_inode_size);
1873                         goto failed_mount;
1874                 }
1875         }
1876         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1877                                    le32_to_cpu(es->s_log_frag_size);
1878         if (blocksize != sbi->s_frag_size) {
1879                 ext3_msg(sb, KERN_ERR,
1880                        "error: fragsize %lu != blocksize %u (unsupported)",
1881                        sbi->s_frag_size, blocksize);
1882                 goto failed_mount;
1883         }
1884         sbi->s_frags_per_block = 1;
1885         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1886         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1887         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1888         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1889                 goto cantfind_ext3;
1890         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1891         if (sbi->s_inodes_per_block == 0)
1892                 goto cantfind_ext3;
1893         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1894                                         sbi->s_inodes_per_block;
1895         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1896         sbi->s_sbh = bh;
1897         sbi->s_mount_state = le16_to_cpu(es->s_state);
1898         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1899         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1900         for (i=0; i < 4; i++)
1901                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1902         sbi->s_def_hash_version = es->s_def_hash_version;
1903         i = le32_to_cpu(es->s_flags);
1904         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1905                 sbi->s_hash_unsigned = 3;
1906         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1907 #ifdef __CHAR_UNSIGNED__
1908                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1909                 sbi->s_hash_unsigned = 3;
1910 #else
1911                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1912 #endif
1913         }
1914
1915         if (sbi->s_blocks_per_group > blocksize * 8) {
1916                 ext3_msg(sb, KERN_ERR,
1917                         "#blocks per group too big: %lu",
1918                         sbi->s_blocks_per_group);
1919                 goto failed_mount;
1920         }
1921         if (sbi->s_frags_per_group > blocksize * 8) {
1922                 ext3_msg(sb, KERN_ERR,
1923                         "error: #fragments per group too big: %lu",
1924                         sbi->s_frags_per_group);
1925                 goto failed_mount;
1926         }
1927         if (sbi->s_inodes_per_group > blocksize * 8) {
1928                 ext3_msg(sb, KERN_ERR,
1929                         "error: #inodes per group too big: %lu",
1930                         sbi->s_inodes_per_group);
1931                 goto failed_mount;
1932         }
1933
1934         err = generic_check_addressable(sb->s_blocksize_bits,
1935                                         le32_to_cpu(es->s_blocks_count));
1936         if (err) {
1937                 ext3_msg(sb, KERN_ERR,
1938                         "error: filesystem is too large to mount safely");
1939                 if (sizeof(sector_t) < 8)
1940                         ext3_msg(sb, KERN_ERR,
1941                                 "error: CONFIG_LBDAF not enabled");
1942                 ret = err;
1943                 goto failed_mount;
1944         }
1945
1946         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1947                 goto cantfind_ext3;
1948         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1949                                le32_to_cpu(es->s_first_data_block) - 1)
1950                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1951         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1952         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1953                                     GFP_KERNEL);
1954         if (sbi->s_group_desc == NULL) {
1955                 ext3_msg(sb, KERN_ERR,
1956                         "error: not enough memory");
1957                 ret = -ENOMEM;
1958                 goto failed_mount;
1959         }
1960
1961         bgl_lock_init(sbi->s_blockgroup_lock);
1962
1963         for (i = 0; i < db_count; i++) {
1964                 block = descriptor_loc(sb, logic_sb_block, i);
1965                 sbi->s_group_desc[i] = sb_bread(sb, block);
1966                 if (!sbi->s_group_desc[i]) {
1967                         ext3_msg(sb, KERN_ERR,
1968                                 "error: can't read group descriptor %d", i);
1969                         db_count = i;
1970                         goto failed_mount2;
1971                 }
1972         }
1973         if (!ext3_check_descriptors (sb)) {
1974                 ext3_msg(sb, KERN_ERR,
1975                         "error: group descriptors corrupted");
1976                 goto failed_mount2;
1977         }
1978         sbi->s_gdb_count = db_count;
1979         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1980         spin_lock_init(&sbi->s_next_gen_lock);
1981
1982         /* per fileystem reservation list head & lock */
1983         spin_lock_init(&sbi->s_rsv_window_lock);
1984         sbi->s_rsv_window_root = RB_ROOT;
1985         /* Add a single, static dummy reservation to the start of the
1986          * reservation window list --- it gives us a placeholder for
1987          * append-at-start-of-list which makes the allocation logic
1988          * _much_ simpler. */
1989         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1990         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1991         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1992         sbi->s_rsv_window_head.rsv_goal_size = 0;
1993         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1994
1995         /*
1996          * set up enough so that it can read an inode
1997          */
1998         sb->s_op = &ext3_sops;
1999         sb->s_export_op = &ext3_export_ops;
2000         sb->s_xattr = ext3_xattr_handlers;
2001 #ifdef CONFIG_QUOTA
2002         sb->s_qcop = &ext3_qctl_operations;
2003         sb->dq_op = &ext3_quota_operations;
2004 #endif
2005         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
2006         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2007         mutex_init(&sbi->s_orphan_lock);
2008         mutex_init(&sbi->s_resize_lock);
2009
2010         sb->s_root = NULL;
2011
2012         needs_recovery = (es->s_last_orphan != 0 ||
2013                           EXT3_HAS_INCOMPAT_FEATURE(sb,
2014                                     EXT3_FEATURE_INCOMPAT_RECOVER));
2015
2016         /*
2017          * The first inode we look at is the journal inode.  Don't try
2018          * root first: it may be modified in the journal!
2019          */
2020         if (!test_opt(sb, NOLOAD) &&
2021             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
2022                 if (ext3_load_journal(sb, es, journal_devnum))
2023                         goto failed_mount2;
2024         } else if (journal_inum) {
2025                 if (ext3_create_journal(sb, es, journal_inum))
2026                         goto failed_mount2;
2027         } else {
2028                 if (!silent)
2029                         ext3_msg(sb, KERN_ERR,
2030                                 "error: no journal found. "
2031                                 "mounting ext3 over ext2?");
2032                 goto failed_mount2;
2033         }
2034         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2035                         ext3_count_free_blocks(sb));
2036         if (!err) {
2037                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2038                                 ext3_count_free_inodes(sb));
2039         }
2040         if (!err) {
2041                 err = percpu_counter_init(&sbi->s_dirs_counter,
2042                                 ext3_count_dirs(sb));
2043         }
2044         if (err) {
2045                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2046                 ret = err;
2047                 goto failed_mount3;
2048         }
2049
2050         /* We have now updated the journal if required, so we can
2051          * validate the data journaling mode. */
2052         switch (test_opt(sb, DATA_FLAGS)) {
2053         case 0:
2054                 /* No mode set, assume a default based on the journal
2055                    capabilities: ORDERED_DATA if the journal can
2056                    cope, else JOURNAL_DATA */
2057                 if (journal_check_available_features
2058                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2059                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2060                 else
2061                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2062                 break;
2063
2064         case EXT3_MOUNT_ORDERED_DATA:
2065         case EXT3_MOUNT_WRITEBACK_DATA:
2066                 if (!journal_check_available_features
2067                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2068                         ext3_msg(sb, KERN_ERR,
2069                                 "error: journal does not support "
2070                                 "requested data journaling mode");
2071                         goto failed_mount3;
2072                 }
2073         default:
2074                 break;
2075         }
2076
2077         /*
2078          * The journal_load will have done any necessary log recovery,
2079          * so we can safely mount the rest of the filesystem now.
2080          */
2081
2082         root = ext3_iget(sb, EXT3_ROOT_INO);
2083         if (IS_ERR(root)) {
2084                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2085                 ret = PTR_ERR(root);
2086                 goto failed_mount3;
2087         }
2088         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2089                 iput(root);
2090                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2091                 goto failed_mount3;
2092         }
2093         sb->s_root = d_make_root(root);
2094         if (!sb->s_root) {
2095                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2096                 ret = -ENOMEM;
2097                 goto failed_mount3;
2098         }
2099
2100         if (ext3_setup_super(sb, es, sb->s_flags & MS_RDONLY))
2101                 sb->s_flags |= MS_RDONLY;
2102
2103         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2104         ext3_orphan_cleanup(sb, es);
2105         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2106         if (needs_recovery) {
2107                 ext3_mark_recovery_complete(sb, es);
2108                 ext3_msg(sb, KERN_INFO, "recovery complete");
2109         }
2110         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2111                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2112                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2113                 "writeback");
2114
2115         return 0;
2116
2117 cantfind_ext3:
2118         if (!silent)
2119                 ext3_msg(sb, KERN_INFO,
2120                         "error: can't find ext3 filesystem on dev %s.",
2121                        sb->s_id);
2122         goto failed_mount;
2123
2124 failed_mount3:
2125         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2126         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2127         percpu_counter_destroy(&sbi->s_dirs_counter);
2128         journal_destroy(sbi->s_journal);
2129 failed_mount2:
2130         for (i = 0; i < db_count; i++)
2131                 brelse(sbi->s_group_desc[i]);
2132         kfree(sbi->s_group_desc);
2133 failed_mount:
2134 #ifdef CONFIG_QUOTA
2135         for (i = 0; i < MAXQUOTAS; i++)
2136                 kfree(sbi->s_qf_names[i]);
2137 #endif
2138         ext3_blkdev_remove(sbi);
2139         brelse(bh);
2140 out_fail:
2141         sb->s_fs_info = NULL;
2142         kfree(sbi->s_blockgroup_lock);
2143         kfree(sbi);
2144         return ret;
2145 }
2146
2147 /*
2148  * Setup any per-fs journal parameters now.  We'll do this both on
2149  * initial mount, once the journal has been initialised but before we've
2150  * done any recovery; and again on any subsequent remount.
2151  */
2152 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2153 {
2154         struct ext3_sb_info *sbi = EXT3_SB(sb);
2155
2156         if (sbi->s_commit_interval)
2157                 journal->j_commit_interval = sbi->s_commit_interval;
2158         /* We could also set up an ext3-specific default for the commit
2159          * interval here, but for now we'll just fall back to the jbd
2160          * default. */
2161
2162         spin_lock(&journal->j_state_lock);
2163         if (test_opt(sb, BARRIER))
2164                 journal->j_flags |= JFS_BARRIER;
2165         else
2166                 journal->j_flags &= ~JFS_BARRIER;
2167         if (test_opt(sb, DATA_ERR_ABORT))
2168                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2169         else
2170                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2171         spin_unlock(&journal->j_state_lock);
2172 }
2173
2174 static journal_t *ext3_get_journal(struct super_block *sb,
2175                                    unsigned int journal_inum)
2176 {
2177         struct inode *journal_inode;
2178         journal_t *journal;
2179
2180         /* First, test for the existence of a valid inode on disk.  Bad
2181          * things happen if we iget() an unused inode, as the subsequent
2182          * iput() will try to delete it. */
2183
2184         journal_inode = ext3_iget(sb, journal_inum);
2185         if (IS_ERR(journal_inode)) {
2186                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2187                 return NULL;
2188         }
2189         if (!journal_inode->i_nlink) {
2190                 make_bad_inode(journal_inode);
2191                 iput(journal_inode);
2192                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2193                 return NULL;
2194         }
2195
2196         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2197                   journal_inode, journal_inode->i_size);
2198         if (!S_ISREG(journal_inode->i_mode)) {
2199                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2200                 iput(journal_inode);
2201                 return NULL;
2202         }
2203
2204         journal = journal_init_inode(journal_inode);
2205         if (!journal) {
2206                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2207                 iput(journal_inode);
2208                 return NULL;
2209         }
2210         journal->j_private = sb;
2211         ext3_init_journal_params(sb, journal);
2212         return journal;
2213 }
2214
2215 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2216                                        dev_t j_dev)
2217 {
2218         struct buffer_head * bh;
2219         journal_t *journal;
2220         ext3_fsblk_t start;
2221         ext3_fsblk_t len;
2222         int hblock, blocksize;
2223         ext3_fsblk_t sb_block;
2224         unsigned long offset;
2225         struct ext3_super_block * es;
2226         struct block_device *bdev;
2227
2228         bdev = ext3_blkdev_get(j_dev, sb);
2229         if (bdev == NULL)
2230                 return NULL;
2231
2232         blocksize = sb->s_blocksize;
2233         hblock = bdev_logical_block_size(bdev);
2234         if (blocksize < hblock) {
2235                 ext3_msg(sb, KERN_ERR,
2236                         "error: blocksize too small for journal device");
2237                 goto out_bdev;
2238         }
2239
2240         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2241         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2242         set_blocksize(bdev, blocksize);
2243         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2244                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2245                         "external journal");
2246                 goto out_bdev;
2247         }
2248
2249         es = (struct ext3_super_block *) (bh->b_data + offset);
2250         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2251             !(le32_to_cpu(es->s_feature_incompat) &
2252               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2253                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2254                         "bad superblock");
2255                 brelse(bh);
2256                 goto out_bdev;
2257         }
2258
2259         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2260                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2261                 brelse(bh);
2262                 goto out_bdev;
2263         }
2264
2265         len = le32_to_cpu(es->s_blocks_count);
2266         start = sb_block + 1;
2267         brelse(bh);     /* we're done with the superblock */
2268
2269         journal = journal_init_dev(bdev, sb->s_bdev,
2270                                         start, len, blocksize);
2271         if (!journal) {
2272                 ext3_msg(sb, KERN_ERR,
2273                         "error: failed to create device journal");
2274                 goto out_bdev;
2275         }
2276         journal->j_private = sb;
2277         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2278                 if (bh_submit_read(journal->j_sb_buffer)) {
2279                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2280                         goto out_journal;
2281                 }
2282         }
2283         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2284                 ext3_msg(sb, KERN_ERR,
2285                         "error: external journal has more than one "
2286                         "user (unsupported) - %d",
2287                         be32_to_cpu(journal->j_superblock->s_nr_users));
2288                 goto out_journal;
2289         }
2290         EXT3_SB(sb)->journal_bdev = bdev;
2291         ext3_init_journal_params(sb, journal);
2292         return journal;
2293 out_journal:
2294         journal_destroy(journal);
2295 out_bdev:
2296         ext3_blkdev_put(bdev);
2297         return NULL;
2298 }
2299
2300 static int ext3_load_journal(struct super_block *sb,
2301                              struct ext3_super_block *es,
2302                              unsigned long journal_devnum)
2303 {
2304         journal_t *journal;
2305         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2306         dev_t journal_dev;
2307         int err = 0;
2308         int really_read_only;
2309
2310         if (journal_devnum &&
2311             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2312                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2313                         "numbers have changed");
2314                 journal_dev = new_decode_dev(journal_devnum);
2315         } else
2316                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2317
2318         really_read_only = bdev_read_only(sb->s_bdev);
2319
2320         /*
2321          * Are we loading a blank journal or performing recovery after a
2322          * crash?  For recovery, we need to check in advance whether we
2323          * can get read-write access to the device.
2324          */
2325
2326         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2327                 if (sb->s_flags & MS_RDONLY) {
2328                         ext3_msg(sb, KERN_INFO,
2329                                 "recovery required on readonly filesystem");
2330                         if (really_read_only) {
2331                                 ext3_msg(sb, KERN_ERR, "error: write access "
2332                                         "unavailable, cannot proceed");
2333                                 return -EROFS;
2334                         }
2335                         ext3_msg(sb, KERN_INFO,
2336                                 "write access will be enabled during recovery");
2337                 }
2338         }
2339
2340         if (journal_inum && journal_dev) {
2341                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2342                        "and inode journals");
2343                 return -EINVAL;
2344         }
2345
2346         if (journal_inum) {
2347                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2348                         return -EINVAL;
2349         } else {
2350                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2351                         return -EINVAL;
2352         }
2353
2354         if (!(journal->j_flags & JFS_BARRIER))
2355                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2356
2357         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2358                 err = journal_update_format(journal);
2359                 if (err)  {
2360                         ext3_msg(sb, KERN_ERR, "error updating journal");
2361                         journal_destroy(journal);
2362                         return err;
2363                 }
2364         }
2365
2366         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2367                 err = journal_wipe(journal, !really_read_only);
2368         if (!err)
2369                 err = journal_load(journal);
2370
2371         if (err) {
2372                 ext3_msg(sb, KERN_ERR, "error loading journal");
2373                 journal_destroy(journal);
2374                 return err;
2375         }
2376
2377         EXT3_SB(sb)->s_journal = journal;
2378         ext3_clear_journal_err(sb, es);
2379
2380         if (!really_read_only && journal_devnum &&
2381             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2382                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2383
2384                 /* Make sure we flush the recovery flag to disk. */
2385                 ext3_commit_super(sb, es, 1);
2386         }
2387
2388         return 0;
2389 }
2390
2391 static int ext3_create_journal(struct super_block *sb,
2392                                struct ext3_super_block *es,
2393                                unsigned int journal_inum)
2394 {
2395         journal_t *journal;
2396         int err;
2397
2398         if (sb->s_flags & MS_RDONLY) {
2399                 ext3_msg(sb, KERN_ERR,
2400                         "error: readonly filesystem when trying to "
2401                         "create journal");
2402                 return -EROFS;
2403         }
2404
2405         journal = ext3_get_journal(sb, journal_inum);
2406         if (!journal)
2407                 return -EINVAL;
2408
2409         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2410                journal_inum);
2411
2412         err = journal_create(journal);
2413         if (err) {
2414                 ext3_msg(sb, KERN_ERR, "error creating journal");
2415                 journal_destroy(journal);
2416                 return -EIO;
2417         }
2418
2419         EXT3_SB(sb)->s_journal = journal;
2420
2421         ext3_update_dynamic_rev(sb);
2422         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2423         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2424
2425         es->s_journal_inum = cpu_to_le32(journal_inum);
2426
2427         /* Make sure we flush the recovery flag to disk. */
2428         ext3_commit_super(sb, es, 1);
2429
2430         return 0;
2431 }
2432
2433 static int ext3_commit_super(struct super_block *sb,
2434                                struct ext3_super_block *es,
2435                                int sync)
2436 {
2437         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2438         int error = 0;
2439
2440         if (!sbh)
2441                 return error;
2442
2443         if (buffer_write_io_error(sbh)) {
2444                 /*
2445                  * Oh, dear.  A previous attempt to write the
2446                  * superblock failed.  This could happen because the
2447                  * USB device was yanked out.  Or it could happen to
2448                  * be a transient write error and maybe the block will
2449                  * be remapped.  Nothing we can do but to retry the
2450                  * write and hope for the best.
2451                  */
2452                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2453                        "superblock detected");
2454                 clear_buffer_write_io_error(sbh);
2455                 set_buffer_uptodate(sbh);
2456         }
2457         /*
2458          * If the file system is mounted read-only, don't update the
2459          * superblock write time.  This avoids updating the superblock
2460          * write time when we are mounting the root file system
2461          * read/only but we need to replay the journal; at that point,
2462          * for people who are east of GMT and who make their clock
2463          * tick in localtime for Windows bug-for-bug compatibility,
2464          * the clock is set in the future, and this will cause e2fsck
2465          * to complain and force a full file system check.
2466          */
2467         if (!(sb->s_flags & MS_RDONLY))
2468                 es->s_wtime = cpu_to_le32(get_seconds());
2469         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2470         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2471         BUFFER_TRACE(sbh, "marking dirty");
2472         mark_buffer_dirty(sbh);
2473         if (sync) {
2474                 error = sync_dirty_buffer(sbh);
2475                 if (buffer_write_io_error(sbh)) {
2476                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2477                                "superblock");
2478                         clear_buffer_write_io_error(sbh);
2479                         set_buffer_uptodate(sbh);
2480                 }
2481         }
2482         return error;
2483 }
2484
2485
2486 /*
2487  * Have we just finished recovery?  If so, and if we are mounting (or
2488  * remounting) the filesystem readonly, then we will end up with a
2489  * consistent fs on disk.  Record that fact.
2490  */
2491 static void ext3_mark_recovery_complete(struct super_block * sb,
2492                                         struct ext3_super_block * es)
2493 {
2494         journal_t *journal = EXT3_SB(sb)->s_journal;
2495
2496         journal_lock_updates(journal);
2497         if (journal_flush(journal) < 0)
2498                 goto out;
2499
2500         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2501             sb->s_flags & MS_RDONLY) {
2502                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2503                 ext3_commit_super(sb, es, 1);
2504         }
2505
2506 out:
2507         journal_unlock_updates(journal);
2508 }
2509
2510 /*
2511  * If we are mounting (or read-write remounting) a filesystem whose journal
2512  * has recorded an error from a previous lifetime, move that error to the
2513  * main filesystem now.
2514  */
2515 static void ext3_clear_journal_err(struct super_block *sb,
2516                                    struct ext3_super_block *es)
2517 {
2518         journal_t *journal;
2519         int j_errno;
2520         const char *errstr;
2521
2522         journal = EXT3_SB(sb)->s_journal;
2523
2524         /*
2525          * Now check for any error status which may have been recorded in the
2526          * journal by a prior ext3_error() or ext3_abort()
2527          */
2528
2529         j_errno = journal_errno(journal);
2530         if (j_errno) {
2531                 char nbuf[16];
2532
2533                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2534                 ext3_warning(sb, __func__, "Filesystem error recorded "
2535                              "from previous mount: %s", errstr);
2536                 ext3_warning(sb, __func__, "Marking fs in need of "
2537                              "filesystem check.");
2538
2539                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2540                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2541                 ext3_commit_super (sb, es, 1);
2542
2543                 journal_clear_err(journal);
2544         }
2545 }
2546
2547 /*
2548  * Force the running and committing transactions to commit,
2549  * and wait on the commit.
2550  */
2551 int ext3_force_commit(struct super_block *sb)
2552 {
2553         journal_t *journal;
2554         int ret;
2555
2556         if (sb->s_flags & MS_RDONLY)
2557                 return 0;
2558
2559         journal = EXT3_SB(sb)->s_journal;
2560         ret = ext3_journal_force_commit(journal);
2561         return ret;
2562 }
2563
2564 static int ext3_sync_fs(struct super_block *sb, int wait)
2565 {
2566         tid_t target;
2567
2568         trace_ext3_sync_fs(sb, wait);
2569         /*
2570          * Writeback quota in non-journalled quota case - journalled quota has
2571          * no dirty dquots
2572          */
2573         dquot_writeback_dquots(sb, -1);
2574         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2575                 if (wait)
2576                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2577         }
2578         return 0;
2579 }
2580
2581 /*
2582  * LVM calls this function before a (read-only) snapshot is created.  This
2583  * gives us a chance to flush the journal completely and mark the fs clean.
2584  */
2585 static int ext3_freeze(struct super_block *sb)
2586 {
2587         int error = 0;
2588         journal_t *journal;
2589
2590         if (!(sb->s_flags & MS_RDONLY)) {
2591                 journal = EXT3_SB(sb)->s_journal;
2592
2593                 /* Now we set up the journal barrier. */
2594                 journal_lock_updates(journal);
2595
2596                 /*
2597                  * We don't want to clear needs_recovery flag when we failed
2598                  * to flush the journal.
2599                  */
2600                 error = journal_flush(journal);
2601                 if (error < 0)
2602                         goto out;
2603
2604                 /* Journal blocked and flushed, clear needs_recovery flag. */
2605                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2606                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2607                 if (error)
2608                         goto out;
2609         }
2610         return 0;
2611
2612 out:
2613         journal_unlock_updates(journal);
2614         return error;
2615 }
2616
2617 /*
2618  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2619  * flag here, even though the filesystem is not technically dirty yet.
2620  */
2621 static int ext3_unfreeze(struct super_block *sb)
2622 {
2623         if (!(sb->s_flags & MS_RDONLY)) {
2624                 /* Reser the needs_recovery flag before the fs is unlocked. */
2625                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2626                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2627                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2628         }
2629         return 0;
2630 }
2631
2632 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2633 {
2634         struct ext3_super_block * es;
2635         struct ext3_sb_info *sbi = EXT3_SB(sb);
2636         ext3_fsblk_t n_blocks_count = 0;
2637         unsigned long old_sb_flags;
2638         struct ext3_mount_options old_opts;
2639         int enable_quota = 0;
2640         int err;
2641 #ifdef CONFIG_QUOTA
2642         int i;
2643 #endif
2644
2645         sync_filesystem(sb);
2646
2647         /* Store the original options */
2648         old_sb_flags = sb->s_flags;
2649         old_opts.s_mount_opt = sbi->s_mount_opt;
2650         old_opts.s_resuid = sbi->s_resuid;
2651         old_opts.s_resgid = sbi->s_resgid;
2652         old_opts.s_commit_interval = sbi->s_commit_interval;
2653 #ifdef CONFIG_QUOTA
2654         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2655         for (i = 0; i < MAXQUOTAS; i++)
2656                 if (sbi->s_qf_names[i]) {
2657                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
2658                                                          GFP_KERNEL);
2659                         if (!old_opts.s_qf_names[i]) {
2660                                 int j;
2661
2662                                 for (j = 0; j < i; j++)
2663                                         kfree(old_opts.s_qf_names[j]);
2664                                 return -ENOMEM;
2665                         }
2666                 } else
2667                         old_opts.s_qf_names[i] = NULL;
2668 #endif
2669
2670         /*
2671          * Allow the "check" option to be passed as a remount option.
2672          */
2673         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2674                 err = -EINVAL;
2675                 goto restore_opts;
2676         }
2677
2678         if (test_opt(sb, ABORT))
2679                 ext3_abort(sb, __func__, "Abort forced by user");
2680
2681         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2682                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2683
2684         es = sbi->s_es;
2685
2686         ext3_init_journal_params(sb, sbi->s_journal);
2687
2688         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2689                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2690                 if (test_opt(sb, ABORT)) {
2691                         err = -EROFS;
2692                         goto restore_opts;
2693                 }
2694
2695                 if (*flags & MS_RDONLY) {
2696                         err = dquot_suspend(sb, -1);
2697                         if (err < 0)
2698                                 goto restore_opts;
2699
2700                         /*
2701                          * First of all, the unconditional stuff we have to do
2702                          * to disable replay of the journal when we next remount
2703                          */
2704                         sb->s_flags |= MS_RDONLY;
2705
2706                         /*
2707                          * OK, test if we are remounting a valid rw partition
2708                          * readonly, and if so set the rdonly flag and then
2709                          * mark the partition as valid again.
2710                          */
2711                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2712                             (sbi->s_mount_state & EXT3_VALID_FS))
2713                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2714
2715                         ext3_mark_recovery_complete(sb, es);
2716                 } else {
2717                         __le32 ret;
2718                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2719                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2720                                 ext3_msg(sb, KERN_WARNING,
2721                                         "warning: couldn't remount RDWR "
2722                                         "because of unsupported optional "
2723                                         "features (%x)", le32_to_cpu(ret));
2724                                 err = -EROFS;
2725                                 goto restore_opts;
2726                         }
2727
2728                         /*
2729                          * If we have an unprocessed orphan list hanging
2730                          * around from a previously readonly bdev mount,
2731                          * require a full umount & mount for now.
2732                          */
2733                         if (es->s_last_orphan) {
2734                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2735                                        "remount RDWR because of unprocessed "
2736                                        "orphan inode list.  Please "
2737                                        "umount & mount instead.");
2738                                 err = -EINVAL;
2739                                 goto restore_opts;
2740                         }
2741
2742                         /*
2743                          * Mounting a RDONLY partition read-write, so reread
2744                          * and store the current valid flag.  (It may have
2745                          * been changed by e2fsck since we originally mounted
2746                          * the partition.)
2747                          */
2748                         ext3_clear_journal_err(sb, es);
2749                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2750                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2751                                 goto restore_opts;
2752                         if (!ext3_setup_super (sb, es, 0))
2753                                 sb->s_flags &= ~MS_RDONLY;
2754                         enable_quota = 1;
2755                 }
2756         }
2757 #ifdef CONFIG_QUOTA
2758         /* Release old quota file names */
2759         for (i = 0; i < MAXQUOTAS; i++)
2760                 kfree(old_opts.s_qf_names[i]);
2761 #endif
2762         if (enable_quota)
2763                 dquot_resume(sb, -1);
2764         return 0;
2765 restore_opts:
2766         sb->s_flags = old_sb_flags;
2767         sbi->s_mount_opt = old_opts.s_mount_opt;
2768         sbi->s_resuid = old_opts.s_resuid;
2769         sbi->s_resgid = old_opts.s_resgid;
2770         sbi->s_commit_interval = old_opts.s_commit_interval;
2771 #ifdef CONFIG_QUOTA
2772         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2773         for (i = 0; i < MAXQUOTAS; i++) {
2774                 kfree(sbi->s_qf_names[i]);
2775                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2776         }
2777 #endif
2778         return err;
2779 }
2780
2781 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2782 {
2783         struct super_block *sb = dentry->d_sb;
2784         struct ext3_sb_info *sbi = EXT3_SB(sb);
2785         struct ext3_super_block *es = sbi->s_es;
2786         u64 fsid;
2787
2788         if (test_opt(sb, MINIX_DF)) {
2789                 sbi->s_overhead_last = 0;
2790         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2791                 unsigned long ngroups = sbi->s_groups_count, i;
2792                 ext3_fsblk_t overhead = 0;
2793                 smp_rmb();
2794
2795                 /*
2796                  * Compute the overhead (FS structures).  This is constant
2797                  * for a given filesystem unless the number of block groups
2798                  * changes so we cache the previous value until it does.
2799                  */
2800
2801                 /*
2802                  * All of the blocks before first_data_block are
2803                  * overhead
2804                  */
2805                 overhead = le32_to_cpu(es->s_first_data_block);
2806
2807                 /*
2808                  * Add the overhead attributed to the superblock and
2809                  * block group descriptors.  If the sparse superblocks
2810                  * feature is turned on, then not all groups have this.
2811                  */
2812                 for (i = 0; i < ngroups; i++) {
2813                         overhead += ext3_bg_has_super(sb, i) +
2814                                 ext3_bg_num_gdb(sb, i);
2815                         cond_resched();
2816                 }
2817
2818                 /*
2819                  * Every block group has an inode bitmap, a block
2820                  * bitmap, and an inode table.
2821                  */
2822                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2823
2824                 /* Add the journal blocks as well */
2825                 overhead += sbi->s_journal->j_maxlen;
2826
2827                 sbi->s_overhead_last = overhead;
2828                 smp_wmb();
2829                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2830         }
2831
2832         buf->f_type = EXT3_SUPER_MAGIC;
2833         buf->f_bsize = sb->s_blocksize;
2834         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2835         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2836         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2837         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2838                 buf->f_bavail = 0;
2839         buf->f_files = le32_to_cpu(es->s_inodes_count);
2840         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2841         buf->f_namelen = EXT3_NAME_LEN;
2842         fsid = le64_to_cpup((void *)es->s_uuid) ^
2843                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2844         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2845         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2846         return 0;
2847 }
2848
2849 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2850  * is locked for write. Otherwise the are possible deadlocks:
2851  * Process 1                         Process 2
2852  * ext3_create()                     quota_sync()
2853  *   journal_start()                   write_dquot()
2854  *   dquot_initialize()                       down(dqio_mutex)
2855  *     down(dqio_mutex)                    journal_start()
2856  *
2857  */
2858
2859 #ifdef CONFIG_QUOTA
2860
2861 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2862 {
2863         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
2864 }
2865
2866 static int ext3_write_dquot(struct dquot *dquot)
2867 {
2868         int ret, err;
2869         handle_t *handle;
2870         struct inode *inode;
2871
2872         inode = dquot_to_inode(dquot);
2873         handle = ext3_journal_start(inode,
2874                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2875         if (IS_ERR(handle))
2876                 return PTR_ERR(handle);
2877         ret = dquot_commit(dquot);
2878         err = ext3_journal_stop(handle);
2879         if (!ret)
2880                 ret = err;
2881         return ret;
2882 }
2883
2884 static int ext3_acquire_dquot(struct dquot *dquot)
2885 {
2886         int ret, err;
2887         handle_t *handle;
2888
2889         handle = ext3_journal_start(dquot_to_inode(dquot),
2890                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2891         if (IS_ERR(handle))
2892                 return PTR_ERR(handle);
2893         ret = dquot_acquire(dquot);
2894         err = ext3_journal_stop(handle);
2895         if (!ret)
2896                 ret = err;
2897         return ret;
2898 }
2899
2900 static int ext3_release_dquot(struct dquot *dquot)
2901 {
2902         int ret, err;
2903         handle_t *handle;
2904
2905         handle = ext3_journal_start(dquot_to_inode(dquot),
2906                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2907         if (IS_ERR(handle)) {
2908                 /* Release dquot anyway to avoid endless cycle in dqput() */
2909                 dquot_release(dquot);
2910                 return PTR_ERR(handle);
2911         }
2912         ret = dquot_release(dquot);
2913         err = ext3_journal_stop(handle);
2914         if (!ret)
2915                 ret = err;
2916         return ret;
2917 }
2918
2919 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2920 {
2921         /* Are we journaling quotas? */
2922         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2923             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2924                 dquot_mark_dquot_dirty(dquot);
2925                 return ext3_write_dquot(dquot);
2926         } else {
2927                 return dquot_mark_dquot_dirty(dquot);
2928         }
2929 }
2930
2931 static int ext3_write_info(struct super_block *sb, int type)
2932 {
2933         int ret, err;
2934         handle_t *handle;
2935
2936         /* Data block + inode block */
2937         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2938         if (IS_ERR(handle))
2939                 return PTR_ERR(handle);
2940         ret = dquot_commit_info(sb, type);
2941         err = ext3_journal_stop(handle);
2942         if (!ret)
2943                 ret = err;
2944         return ret;
2945 }
2946
2947 /*
2948  * Turn on quotas during mount time - we need to find
2949  * the quota file and such...
2950  */
2951 static int ext3_quota_on_mount(struct super_block *sb, int type)
2952 {
2953         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2954                                         EXT3_SB(sb)->s_jquota_fmt, type);
2955 }
2956
2957 /*
2958  * Standard function to be called on quota_on
2959  */
2960 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2961                          struct path *path)
2962 {
2963         int err;
2964
2965         if (!test_opt(sb, QUOTA))
2966                 return -EINVAL;
2967
2968         /* Quotafile not on the same filesystem? */
2969         if (path->dentry->d_sb != sb)
2970                 return -EXDEV;
2971         /* Journaling quota? */
2972         if (EXT3_SB(sb)->s_qf_names[type]) {
2973                 /* Quotafile not of fs root? */
2974                 if (path->dentry->d_parent != sb->s_root)
2975                         ext3_msg(sb, KERN_WARNING,
2976                                 "warning: Quota file not on filesystem root. "
2977                                 "Journaled quota will not work.");
2978         }
2979
2980         /*
2981          * When we journal data on quota file, we have to flush journal to see
2982          * all updates to the file when we bypass pagecache...
2983          */
2984         if (ext3_should_journal_data(path->dentry->d_inode)) {
2985                 /*
2986                  * We don't need to lock updates but journal_flush() could
2987                  * otherwise be livelocked...
2988                  */
2989                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2990                 err = journal_flush(EXT3_SB(sb)->s_journal);
2991                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2992                 if (err)
2993                         return err;
2994         }
2995
2996         return dquot_quota_on(sb, type, format_id, path);
2997 }
2998
2999 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3000  * acquiring the locks... As quota files are never truncated and quota code
3001  * itself serializes the operations (and no one else should touch the files)
3002  * we don't have to be afraid of races */
3003 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
3004                                size_t len, loff_t off)
3005 {
3006         struct inode *inode = sb_dqopt(sb)->files[type];
3007         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3008         int err = 0;
3009         int offset = off & (sb->s_blocksize - 1);
3010         int tocopy;
3011         size_t toread;
3012         struct buffer_head *bh;
3013         loff_t i_size = i_size_read(inode);
3014
3015         if (off > i_size)
3016                 return 0;
3017         if (off+len > i_size)
3018                 len = i_size-off;
3019         toread = len;
3020         while (toread > 0) {
3021                 tocopy = sb->s_blocksize - offset < toread ?
3022                                 sb->s_blocksize - offset : toread;
3023                 bh = ext3_bread(NULL, inode, blk, 0, &err);
3024                 if (err)
3025                         return err;
3026                 if (!bh)        /* A hole? */
3027                         memset(data, 0, tocopy);
3028                 else
3029                         memcpy(data, bh->b_data+offset, tocopy);
3030                 brelse(bh);
3031                 offset = 0;
3032                 toread -= tocopy;
3033                 data += tocopy;
3034                 blk++;
3035         }
3036         return len;
3037 }
3038
3039 /* Write to quotafile (we know the transaction is already started and has
3040  * enough credits) */
3041 static ssize_t ext3_quota_write(struct super_block *sb, int type,
3042                                 const char *data, size_t len, loff_t off)
3043 {
3044         struct inode *inode = sb_dqopt(sb)->files[type];
3045         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
3046         int err = 0;
3047         int offset = off & (sb->s_blocksize - 1);
3048         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
3049         struct buffer_head *bh;
3050         handle_t *handle = journal_current_handle();
3051
3052         if (!handle) {
3053                 ext3_msg(sb, KERN_WARNING,
3054                         "warning: quota write (off=%llu, len=%llu)"
3055                         " cancelled because transaction is not started.",
3056                         (unsigned long long)off, (unsigned long long)len);
3057                 return -EIO;
3058         }
3059
3060         /*
3061          * Since we account only one data block in transaction credits,
3062          * then it is impossible to cross a block boundary.
3063          */
3064         if (sb->s_blocksize - offset < len) {
3065                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3066                         " cancelled because not block aligned",
3067                         (unsigned long long)off, (unsigned long long)len);
3068                 return -EIO;
3069         }
3070         bh = ext3_bread(handle, inode, blk, 1, &err);
3071         if (!bh)
3072                 goto out;
3073         if (journal_quota) {
3074                 err = ext3_journal_get_write_access(handle, bh);
3075                 if (err) {
3076                         brelse(bh);
3077                         goto out;
3078                 }
3079         }
3080         lock_buffer(bh);
3081         memcpy(bh->b_data+offset, data, len);
3082         flush_dcache_page(bh->b_page);
3083         unlock_buffer(bh);
3084         if (journal_quota)
3085                 err = ext3_journal_dirty_metadata(handle, bh);
3086         else {
3087                 /* Always do at least ordered writes for quotas */
3088                 err = ext3_journal_dirty_data(handle, bh);
3089                 mark_buffer_dirty(bh);
3090         }
3091         brelse(bh);
3092 out:
3093         if (err)
3094                 return err;
3095         if (inode->i_size < off + len) {
3096                 i_size_write(inode, off + len);
3097                 EXT3_I(inode)->i_disksize = inode->i_size;
3098         }
3099         inode->i_version++;
3100         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3101         ext3_mark_inode_dirty(handle, inode);
3102         return len;
3103 }
3104
3105 #endif
3106
3107 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3108         int flags, const char *dev_name, void *data)
3109 {
3110         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3111 }
3112
3113 static struct file_system_type ext3_fs_type = {
3114         .owner          = THIS_MODULE,
3115         .name           = "ext3",
3116         .mount          = ext3_mount,
3117         .kill_sb        = kill_block_super,
3118         .fs_flags       = FS_REQUIRES_DEV,
3119 };
3120 MODULE_ALIAS_FS("ext3");
3121
3122 static int __init init_ext3_fs(void)
3123 {
3124         int err = init_ext3_xattr();
3125         if (err)
3126                 return err;
3127         err = init_inodecache();
3128         if (err)
3129                 goto out1;
3130         err = register_filesystem(&ext3_fs_type);
3131         if (err)
3132                 goto out;
3133         return 0;
3134 out:
3135         destroy_inodecache();
3136 out1:
3137         exit_ext3_xattr();
3138         return err;
3139 }
3140
3141 static void __exit exit_ext3_fs(void)
3142 {
3143         unregister_filesystem(&ext3_fs_type);
3144         destroy_inodecache();
3145         exit_ext3_xattr();
3146 }
3147
3148 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3149 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3150 MODULE_LICENSE("GPL");
3151 module_init(init_ext3_fs)
3152 module_exit(exit_ext3_fs)