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/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39
  40 #include <asm/uaccess.h>
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44 #include "namei.h"
  45
  46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  47                              unsigned long journal_devnum);
  48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  49                                int);
  50 static void ext3_commit_super (struct super_block * sb,
  51                                struct ext3_super_block * es,
  52                                int sync);
  53 static void ext3_mark_recovery_complete(struct super_block * sb,
  54                                         struct ext3_super_block * es);
  55 static void ext3_clear_journal_err(struct super_block * sb,
  56                                    struct ext3_super_block * es);
  57 static int ext3_sync_fs(struct super_block *sb, int wait);
  58 static const char *ext3_decode_error(struct super_block * sb, int errno,
  59                                      char nbuf[16]);
  60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  61 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  62 static void ext3_unlockfs(struct super_block *sb);
  63 static void ext3_write_super (struct super_block * sb);
  64 static void ext3_write_super_lockfs(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  68  *
  69  * The only special thing we need to do here is to make sure that all
  70  * journal_end calls result in the superblock being marked dirty, so
  71  * that sync() will call the filesystem's write_super callback if
  72  * appropriate.
  73  */
  74 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  75 {
  76         journal_t *journal;
  77
  78         if (sb->s_flags & MS_RDONLY)
  79                 return ERR_PTR(-EROFS);
  80
  81         /* Special case here: if the journal has aborted behind our
  82          * backs (eg. EIO in the commit thread), then we still need to
  83          * take the FS itself readonly cleanly. */
  84         journal = EXT3_SB(sb)->s_journal;
  85         if (is_journal_aborted(journal)) {
  86                 ext3_abort(sb, __FUNCTION__,
  87                            "Detected aborted journal");
  88                 return ERR_PTR(-EROFS);
  89         }
  90
  91         return journal_start(journal, nblocks);
  92 }
  93
  94 /*
  95  * The only special thing we need to do here is to make sure that all
  96  * journal_stop calls result in the superblock being marked dirty, so
  97  * that sync() will call the filesystem's write_super callback if
  98  * appropriate.
  99  */
 100 int __ext3_journal_stop(const char *where, handle_t *handle)
 101 {
 102         struct super_block *sb;
 103         int err;
 104         int rc;
 105
 106         sb = handle->h_transaction->t_journal->j_private;
 107         err = handle->h_err;
 108         rc = journal_stop(handle);
 109
 110         if (!err)
 111                 err = rc;
 112         if (err)
 113                 __ext3_std_error(sb, where, err);
 114         return err;
 115 }
 116
 117 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 118                 struct buffer_head *bh, handle_t *handle, int err)
 119 {
 120         char nbuf[16];
 121         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 122
 123         if (bh)
 124                 BUFFER_TRACE(bh, "abort");
 125
 126         if (!handle->h_err)
 127                 handle->h_err = err;
 128
 129         if (is_handle_aborted(handle))
 130                 return;
 131
 132         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 133                caller, errstr, err_fn);
 134
 135         journal_abort_handle(handle);
 136 }
 137
 138 /* Deal with the reporting of failure conditions on a filesystem such as
 139  * inconsistencies detected or read IO failures.
 140  *
 141  * On ext2, we can store the error state of the filesystem in the
 142  * superblock.  That is not possible on ext3, because we may have other
 143  * write ordering constraints on the superblock which prevent us from
 144  * writing it out straight away; and given that the journal is about to
 145  * be aborted, we can't rely on the current, or future, transactions to
 146  * write out the superblock safely.
 147  *
 148  * We'll just use the journal_abort() error code to record an error in
 149  * the journal instead.  On recovery, the journal will compain about
 150  * that error until we've noted it down and cleared it.
 151  */
 152
 153 static void ext3_handle_error(struct super_block *sb)
 154 {
 155         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 156
 157         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 158         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 159
 160         if (sb->s_flags & MS_RDONLY)
 161                 return;
 162
 163         if (test_opt (sb, ERRORS_RO)) {
 164                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 165                 sb->s_flags |= MS_RDONLY;
 166         } else {
 167                 journal_t *journal = EXT3_SB(sb)->s_journal;
 168
 169                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 170                 if (journal)
 171                         journal_abort(journal, -EIO);
 172         }
 173         if (test_opt(sb, ERRORS_PANIC))
 174                 panic("EXT3-fs (device %s): panic forced after error\n",
 175                         sb->s_id);
 176         ext3_commit_super(sb, es, 1);
 177 }
 178
 179 void ext3_error (struct super_block * sb, const char * function,
 180                  const char * fmt, ...)
 181 {
 182         va_list args;
 183
 184         va_start(args, fmt);
 185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 186         vprintk(fmt, args);
 187         printk("\n");
 188         va_end(args);
 189
 190         ext3_handle_error(sb);
 191 }
 192
 193 static const char *ext3_decode_error(struct super_block * sb, int errno,
 194                                      char nbuf[16])
 195 {
 196         char *errstr = NULL;
 197
 198         switch (errno) {
 199         case -EIO:
 200                 errstr = "IO failure";
 201                 break;
 202         case -ENOMEM:
 203                 errstr = "Out of memory";
 204                 break;
 205         case -EROFS:
 206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 207                         errstr = "Journal has aborted";
 208                 else
 209                         errstr = "Readonly filesystem";
 210                 break;
 211         default:
 212                 /* If the caller passed in an extra buffer for unknown
 213                  * errors, textualise them now.  Else we just return
 214                  * NULL. */
 215                 if (nbuf) {
 216                         /* Check for truncated error codes... */
 217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 218                                 errstr = nbuf;
 219                 }
 220                 break;
 221         }
 222
 223         return errstr;
 224 }
 225
 226 /* __ext3_std_error decodes expected errors from journaling functions
 227  * automatically and invokes the appropriate error response.  */
 228
 229 void __ext3_std_error (struct super_block * sb, const char * function,
 230                        int errno)
 231 {
 232         char nbuf[16];
 233         const char *errstr;
 234
 235         /* Special case: if the error is EROFS, and we're not already
 236          * inside a transaction, then there's really no point in logging
 237          * an error. */
 238         if (errno == -EROFS && journal_current_handle() == NULL &&
 239             (sb->s_flags & MS_RDONLY))
 240                 return;
 241
 242         errstr = ext3_decode_error(sb, errno, nbuf);
 243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 244                 sb->s_id, function, errstr);
 245
 246         ext3_handle_error(sb);
 247 }
 248
 249 /*
 250  * ext3_abort is a much stronger failure handler than ext3_error.  The
 251  * abort function may be used to deal with unrecoverable failures such
 252  * as journal IO errors or ENOMEM at a critical moment in log management.
 253  *
 254  * We unconditionally force the filesystem into an ABORT|READONLY state,
 255  * unless the error response on the fs has been set to panic in which
 256  * case we take the easy way out and panic immediately.
 257  */
 258
 259 void ext3_abort (struct super_block * sb, const char * function,
 260                  const char * fmt, ...)
 261 {
 262         va_list args;
 263
 264         printk (KERN_CRIT "ext3_abort called.\n");
 265
 266         va_start(args, fmt);
 267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 268         vprintk(fmt, args);
 269         printk("\n");
 270         va_end(args);
 271
 272         if (test_opt(sb, ERRORS_PANIC))
 273                 panic("EXT3-fs panic from previous error\n");
 274
 275         if (sb->s_flags & MS_RDONLY)
 276                 return;
 277
 278         printk(KERN_CRIT "Remounting filesystem read-only\n");
 279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 280         sb->s_flags |= MS_RDONLY;
 281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 283 }
 284
 285 void ext3_warning (struct super_block * sb, const char * function,
 286                    const char * fmt, ...)
 287 {
 288         va_list args;
 289
 290         va_start(args, fmt);
 291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 292                sb->s_id, function);
 293         vprintk(fmt, args);
 294         printk("\n");
 295         va_end(args);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        NEXT_ORPHAN(inode));
 384         }
 385 }
 386
 387 static void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 408         percpu_counter_destroy(&sbi->s_dirs_counter);
 409         brelse(sbi->s_sbh);
 410 #ifdef CONFIG_QUOTA
 411         for (i = 0; i < MAXQUOTAS; i++)
 412                 kfree(sbi->s_qf_names[i]);
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->i_block_alloc_info = NULL;
 456         ei->vfs_inode.i_version = 1;
 457         return &ei->vfs_inode;
 458 }
 459
 460 static void ext3_destroy_inode(struct inode *inode)
 461 {
 462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 463 }
 464
 465 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 466 {
 467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 468
 469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 470             SLAB_CTOR_CONSTRUCTOR) {
 471                 INIT_LIST_HEAD(&ei->i_orphan);
 472 #ifdef CONFIG_EXT3_FS_XATTR
 473                 init_rwsem(&ei->xattr_sem);
 474 #endif
 475                 mutex_init(&ei->truncate_mutex);
 476                 inode_init_once(&ei->vfs_inode);
 477         }
 478 }
 479
 480 static int init_inodecache(void)
 481 {
 482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 483                                              sizeof(struct ext3_inode_info),
 484                                              0, (SLAB_RECLAIM_ACCOUNT|
 485                                                 SLAB_MEM_SPREAD),
 486                                              init_once, NULL);
 487         if (ext3_inode_cachep == NULL)
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static void destroy_inodecache(void)
 493 {
 494         if (kmem_cache_destroy(ext3_inode_cachep))
 495                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 496 }
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 501 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 502        if (EXT3_I(inode)->i_acl &&
 503            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 504                posix_acl_release(EXT3_I(inode)->i_acl);
 505                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 506        }
 507        if (EXT3_I(inode)->i_default_acl &&
 508            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 509                posix_acl_release(EXT3_I(inode)->i_default_acl);
 510                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 511        }
 512 #endif
 513         ext3_discard_reservation(inode);
 514         EXT3_I(inode)->i_block_alloc_info = NULL;
 515         kfree(rsv);
 516 }
 517
 518 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 519 {
 520 #if defined(CONFIG_QUOTA)
 521         struct ext3_sb_info *sbi = EXT3_SB(sb);
 522
 523         if (sbi->s_jquota_fmt)
 524                 seq_printf(seq, ",jqfmt=%s",
 525                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 526
 527         if (sbi->s_qf_names[USRQUOTA])
 528                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 529
 530         if (sbi->s_qf_names[GRPQUOTA])
 531                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 532
 533         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 534                 seq_puts(seq, ",usrquota");
 535
 536         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 537                 seq_puts(seq, ",grpquota");
 538 #endif
 539 }
 540
 541 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 542 {
 543         struct super_block *sb = vfs->mnt_sb;
 544
 545         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 546                 seq_puts(seq, ",data=journal");
 547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 548                 seq_puts(seq, ",data=ordered");
 549         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 550                 seq_puts(seq, ",data=writeback");
 551
 552         ext3_show_quota_options(seq, sb);
 553
 554         return 0;
 555 }
 556
 557 #ifdef CONFIG_QUOTA
 558 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 559 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 560
 561 static int ext3_dquot_initialize(struct inode *inode, int type);
 562 static int ext3_dquot_drop(struct inode *inode);
 563 static int ext3_write_dquot(struct dquot *dquot);
 564 static int ext3_acquire_dquot(struct dquot *dquot);
 565 static int ext3_release_dquot(struct dquot *dquot);
 566 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 567 static int ext3_write_info(struct super_block *sb, int type);
 568 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 569 static int ext3_quota_on_mount(struct super_block *sb, int type);
 570 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 571                                size_t len, loff_t off);
 572 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 573                                 const char *data, size_t len, loff_t off);
 574
 575 static struct dquot_operations ext3_quota_operations = {
 576         .initialize     = ext3_dquot_initialize,
 577         .drop           = ext3_dquot_drop,
 578         .alloc_space    = dquot_alloc_space,
 579         .alloc_inode    = dquot_alloc_inode,
 580         .free_space     = dquot_free_space,
 581         .free_inode     = dquot_free_inode,
 582         .transfer       = dquot_transfer,
 583         .write_dquot    = ext3_write_dquot,
 584         .acquire_dquot  = ext3_acquire_dquot,
 585         .release_dquot  = ext3_release_dquot,
 586         .mark_dirty     = ext3_mark_dquot_dirty,
 587         .write_info     = ext3_write_info
 588 };
 589
 590 static struct quotactl_ops ext3_qctl_operations = {
 591         .quota_on       = ext3_quota_on,
 592         .quota_off      = vfs_quota_off,
 593         .quota_sync     = vfs_quota_sync,
 594         .get_info       = vfs_get_dqinfo,
 595         .set_info       = vfs_set_dqinfo,
 596         .get_dqblk      = vfs_get_dqblk,
 597         .set_dqblk      = vfs_set_dqblk
 598 };
 599 #endif
 600
 601 static struct super_operations ext3_sops = {
 602         .alloc_inode    = ext3_alloc_inode,
 603         .destroy_inode  = ext3_destroy_inode,
 604         .read_inode     = ext3_read_inode,
 605         .write_inode    = ext3_write_inode,
 606         .dirty_inode    = ext3_dirty_inode,
 607         .delete_inode   = ext3_delete_inode,
 608         .put_super      = ext3_put_super,
 609         .write_super    = ext3_write_super,
 610         .sync_fs        = ext3_sync_fs,
 611         .write_super_lockfs = ext3_write_super_lockfs,
 612         .unlockfs       = ext3_unlockfs,
 613         .statfs         = ext3_statfs,
 614         .remount_fs     = ext3_remount,
 615         .clear_inode    = ext3_clear_inode,
 616         .show_options   = ext3_show_options,
 617 #ifdef CONFIG_QUOTA
 618         .quota_read     = ext3_quota_read,
 619         .quota_write    = ext3_quota_write,
 620 #endif
 621 };
 622
 623 static struct export_operations ext3_export_ops = {
 624         .get_parent = ext3_get_parent,
 625 };
 626
 627 enum {
 628         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 629         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 630         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 631         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 632         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 633         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 634         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 635         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 636         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 637         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 638         Opt_grpquota, Opt_tagxid
 639 };
 640
 641 static match_table_t tokens = {
 642         {Opt_bsd_df, "bsddf"},
 643         {Opt_minix_df, "minixdf"},
 644         {Opt_grpid, "grpid"},
 645         {Opt_grpid, "bsdgroups"},
 646         {Opt_nogrpid, "nogrpid"},
 647         {Opt_nogrpid, "sysvgroups"},
 648         {Opt_resgid, "resgid=%u"},
 649         {Opt_resuid, "resuid=%u"},
 650         {Opt_sb, "sb=%u"},
 651         {Opt_err_cont, "errors=continue"},
 652         {Opt_err_panic, "errors=panic"},
 653         {Opt_err_ro, "errors=remount-ro"},
 654         {Opt_nouid32, "nouid32"},
 655         {Opt_nocheck, "nocheck"},
 656         {Opt_nocheck, "check=none"},
 657         {Opt_debug, "debug"},
 658         {Opt_oldalloc, "oldalloc"},
 659         {Opt_orlov, "orlov"},
 660         {Opt_user_xattr, "user_xattr"},
 661         {Opt_nouser_xattr, "nouser_xattr"},
 662         {Opt_acl, "acl"},
 663         {Opt_noacl, "noacl"},
 664         {Opt_reservation, "reservation"},
 665         {Opt_noreservation, "noreservation"},
 666         {Opt_noload, "noload"},
 667         {Opt_nobh, "nobh"},
 668         {Opt_commit, "commit=%u"},
 669         {Opt_journal_update, "journal=update"},
 670         {Opt_journal_inum, "journal=%u"},
 671         {Opt_journal_dev, "journal_dev=%u"},
 672         {Opt_abort, "abort"},
 673         {Opt_data_journal, "data=journal"},
 674         {Opt_data_ordered, "data=ordered"},
 675         {Opt_data_writeback, "data=writeback"},
 676         {Opt_offusrjquota, "usrjquota="},
 677         {Opt_usrjquota, "usrjquota=%s"},
 678         {Opt_offgrpjquota, "grpjquota="},
 679         {Opt_grpjquota, "grpjquota=%s"},
 680         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 681         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 682         {Opt_grpquota, "grpquota"},
 683         {Opt_noquota, "noquota"},
 684         {Opt_quota, "quota"},
 685         {Opt_usrquota, "usrquota"},
 686         {Opt_barrier, "barrier=%u"},
 687         {Opt_tagxid, "tagxid"},
 688         {Opt_err, NULL},
 689         {Opt_resize, "resize"},
 690 };
 691
 692 static unsigned long get_sb_block(void **data)
 693 {
 694         unsigned long   sb_block;
 695         char            *options = (char *) *data;
 696
 697         if (!options || strncmp(options, "sb=", 3) != 0)
 698                 return 1;       /* Default location */
 699         options += 3;
 700         sb_block = simple_strtoul(options, &options, 0);
 701         if (*options && *options != ',') {
 702                 printk("EXT3-fs: Invalid sb specification: %s\n",
 703                        (char *) *data);
 704                 return 1;
 705         }
 706         if (*options == ',')
 707                 options++;
 708         *data = (void *) options;
 709         return sb_block;
 710 }
 711
 712 static int parse_options (char *options, struct super_block *sb,
 713                           unsigned long *inum, unsigned long *journal_devnum,
 714                           unsigned long *n_blocks_count, int is_remount)
 715 {
 716         struct ext3_sb_info *sbi = EXT3_SB(sb);
 717         char * p;
 718         substring_t args[MAX_OPT_ARGS];
 719         int data_opt = 0;
 720         int option;
 721 #ifdef CONFIG_QUOTA
 722         int qtype;
 723         char *qname;
 724 #endif
 725
 726         if (!options)
 727                 return 1;
 728
 729         while ((p = strsep (&options, ",")) != NULL) {
 730                 int token;
 731                 if (!*p)
 732                         continue;
 733
 734                 token = match_token(p, tokens, args);
 735                 switch (token) {
 736                 case Opt_bsd_df:
 737                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 738                         break;
 739                 case Opt_minix_df:
 740                         set_opt (sbi->s_mount_opt, MINIX_DF);
 741                         break;
 742                 case Opt_grpid:
 743                         set_opt (sbi->s_mount_opt, GRPID);
 744                         break;
 745                 case Opt_nogrpid:
 746                         clear_opt (sbi->s_mount_opt, GRPID);
 747                         break;
 748                 case Opt_resuid:
 749                         if (match_int(&args[0], &option))
 750                                 return 0;
 751                         sbi->s_resuid = option;
 752                         break;
 753                 case Opt_resgid:
 754                         if (match_int(&args[0], &option))
 755                                 return 0;
 756                         sbi->s_resgid = option;
 757                         break;
 758                 case Opt_sb:
 759                         /* handled by get_sb_block() instead of here */
 760                         /* *sb_block = match_int(&args[0]); */
 761                         break;
 762                 case Opt_err_panic:
 763                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 764                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 765                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 766                         break;
 767                 case Opt_err_ro:
 768                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 769                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 770                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 771                         break;
 772                 case Opt_err_cont:
 773                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 774                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 775                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 776                         break;
 777                 case Opt_nouid32:
 778                         set_opt (sbi->s_mount_opt, NO_UID32);
 779                         break;
 780 #ifndef CONFIG_INOXID_NONE
 781                 case Opt_tagxid:
 782                         set_opt (sbi->s_mount_opt, TAGXID);
 783                         break;
 784 #endif
 785                 case Opt_nocheck:
 786                         clear_opt (sbi->s_mount_opt, CHECK);
 787                         break;
 788                 case Opt_debug:
 789                         set_opt (sbi->s_mount_opt, DEBUG);
 790                         break;
 791                 case Opt_oldalloc:
 792                         set_opt (sbi->s_mount_opt, OLDALLOC);
 793                         break;
 794                 case Opt_orlov:
 795                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 796                         break;
 797 #ifdef CONFIG_EXT3_FS_XATTR
 798                 case Opt_user_xattr:
 799                         set_opt (sbi->s_mount_opt, XATTR_USER);
 800                         break;
 801                 case Opt_nouser_xattr:
 802                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 803                         break;
 804 #else
 805                 case Opt_user_xattr:
 806                 case Opt_nouser_xattr:
 807                         printk("EXT3 (no)user_xattr options not supported\n");
 808                         break;
 809 #endif
 810 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 811                 case Opt_acl:
 812                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 813                         break;
 814                 case Opt_noacl:
 815                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 816                         break;
 817 #else
 818                 case Opt_acl:
 819                 case Opt_noacl:
 820                         printk("EXT3 (no)acl options not supported\n");
 821                         break;
 822 #endif
 823                 case Opt_reservation:
 824                         set_opt(sbi->s_mount_opt, RESERVATION);
 825                         break;
 826                 case Opt_noreservation:
 827                         clear_opt(sbi->s_mount_opt, RESERVATION);
 828                         break;
 829                 case Opt_journal_update:
 830                         /* @@@ FIXME */
 831                         /* Eventually we will want to be able to create
 832                            a journal file here.  For now, only allow the
 833                            user to specify an existing inode to be the
 834                            journal file. */
 835                         if (is_remount) {
 836                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 837                                        "journal on remount\n");
 838                                 return 0;
 839                         }
 840                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 841                         break;
 842                 case Opt_journal_inum:
 843                         if (is_remount) {
 844                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 845                                        "journal on remount\n");
 846                                 return 0;
 847                         }
 848                         if (match_int(&args[0], &option))
 849                                 return 0;
 850                         *inum = option;
 851                         break;
 852                 case Opt_journal_dev:
 853                         if (is_remount) {
 854                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 855                                        "journal on remount\n");
 856                                 return 0;
 857                         }
 858                         if (match_int(&args[0], &option))
 859                                 return 0;
 860                         *journal_devnum = option;
 861                         break;
 862                 case Opt_noload:
 863                         set_opt (sbi->s_mount_opt, NOLOAD);
 864                         break;
 865                 case Opt_commit:
 866                         if (match_int(&args[0], &option))
 867                                 return 0;
 868                         if (option < 0)
 869                                 return 0;
 870                         if (option == 0)
 871                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 872                         sbi->s_commit_interval = HZ * option;
 873                         break;
 874                 case Opt_data_journal:
 875                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 876                         goto datacheck;
 877                 case Opt_data_ordered:
 878                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 879                         goto datacheck;
 880                 case Opt_data_writeback:
 881                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 882                 datacheck:
 883                         if (is_remount) {
 884                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 885                                                 != data_opt) {
 886                                         printk(KERN_ERR
 887                                                 "EXT3-fs: cannot change data "
 888                                                 "mode on remount\n");
 889                                         return 0;
 890                                 }
 891                         } else {
 892                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 893                                 sbi->s_mount_opt |= data_opt;
 894                         }
 895                         break;
 896 #ifdef CONFIG_QUOTA
 897                 case Opt_usrjquota:
 898                         qtype = USRQUOTA;
 899                         goto set_qf_name;
 900                 case Opt_grpjquota:
 901                         qtype = GRPQUOTA;
 902 set_qf_name:
 903                         if (sb_any_quota_enabled(sb)) {
 904                                 printk(KERN_ERR
 905                                         "EXT3-fs: Cannot change journalled "
 906                                         "quota options when quota turned on.\n");
 907                                 return 0;
 908                         }
 909                         qname = match_strdup(&args[0]);
 910                         if (!qname) {
 911                                 printk(KERN_ERR
 912                                         "EXT3-fs: not enough memory for "
 913                                         "storing quotafile name.\n");
 914                                 return 0;
 915                         }
 916                         if (sbi->s_qf_names[qtype] &&
 917                             strcmp(sbi->s_qf_names[qtype], qname)) {
 918                                 printk(KERN_ERR
 919                                         "EXT3-fs: %s quota file already "
 920                                         "specified.\n", QTYPE2NAME(qtype));
 921                                 kfree(qname);
 922                                 return 0;
 923                         }
 924                         sbi->s_qf_names[qtype] = qname;
 925                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 926                                 printk(KERN_ERR
 927                                         "EXT3-fs: quotafile must be on "
 928                                         "filesystem root.\n");
 929                                 kfree(sbi->s_qf_names[qtype]);
 930                                 sbi->s_qf_names[qtype] = NULL;
 931                                 return 0;
 932                         }
 933                         set_opt(sbi->s_mount_opt, QUOTA);
 934                         break;
 935                 case Opt_offusrjquota:
 936                         qtype = USRQUOTA;
 937                         goto clear_qf_name;
 938                 case Opt_offgrpjquota:
 939                         qtype = GRPQUOTA;
 940 clear_qf_name:
 941                         if (sb_any_quota_enabled(sb)) {
 942                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 943                                         "journalled quota options when "
 944                                         "quota turned on.\n");
 945                                 return 0;
 946                         }
 947                         /*
 948                          * The space will be released later when all options
 949                          * are confirmed to be correct
 950                          */
 951                         sbi->s_qf_names[qtype] = NULL;
 952                         break;
 953                 case Opt_jqfmt_vfsold:
 954                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 955                         break;
 956                 case Opt_jqfmt_vfsv0:
 957                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 958                         break;
 959                 case Opt_quota:
 960                 case Opt_usrquota:
 961                         set_opt(sbi->s_mount_opt, QUOTA);
 962                         set_opt(sbi->s_mount_opt, USRQUOTA);
 963                         break;
 964                 case Opt_grpquota:
 965                         set_opt(sbi->s_mount_opt, QUOTA);
 966                         set_opt(sbi->s_mount_opt, GRPQUOTA);
 967                         break;
 968                 case Opt_noquota:
 969                         if (sb_any_quota_enabled(sb)) {
 970                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
 971                                         "options when quota turned on.\n");
 972                                 return 0;
 973                         }
 974                         clear_opt(sbi->s_mount_opt, QUOTA);
 975                         clear_opt(sbi->s_mount_opt, USRQUOTA);
 976                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
 977                         break;
 978 #else
 979                 case Opt_quota:
 980                 case Opt_usrquota:
 981                 case Opt_grpquota:
 982                 case Opt_usrjquota:
 983                 case Opt_grpjquota:
 984                 case Opt_offusrjquota:
 985                 case Opt_offgrpjquota:
 986                 case Opt_jqfmt_vfsold:
 987                 case Opt_jqfmt_vfsv0:
 988                         printk(KERN_ERR
 989                                 "EXT3-fs: journalled quota options not "
 990                                 "supported.\n");
 991                         break;
 992                 case Opt_noquota:
 993                         break;
 994 #endif
 995                 case Opt_abort:
 996                         set_opt(sbi->s_mount_opt, ABORT);
 997                         break;
 998                 case Opt_barrier:
 999                         if (match_int(&args[0], &option))
1000                                 return 0;
1001                         if (option)
1002                                 set_opt(sbi->s_mount_opt, BARRIER);
1003                         else
1004                                 clear_opt(sbi->s_mount_opt, BARRIER);
1005                         break;
1006                 case Opt_ignore:
1007                         break;
1008                 case Opt_resize:
1009                         if (!is_remount) {
1010                                 printk("EXT3-fs: resize option only available "
1011                                         "for remount\n");
1012                                 return 0;
1013                         }
1014                         if (match_int(&args[0], &option) != 0)
1015                                 return 0;
1016                         *n_blocks_count = option;
1017                         break;
1018                 case Opt_nobh:
1019                         set_opt(sbi->s_mount_opt, NOBH);
1020                         break;
1021                 default:
1022                         printk (KERN_ERR
1023                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1024                                 "or missing value\n", p);
1025                         return 0;
1026                 }
1027         }
1028 #ifdef CONFIG_QUOTA
1029         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1030                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1031                      sbi->s_qf_names[USRQUOTA])
1032                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1033
1034                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1035                      sbi->s_qf_names[GRPQUOTA])
1036                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1037
1038                 if ((sbi->s_qf_names[USRQUOTA] &&
1039                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1040                     (sbi->s_qf_names[GRPQUOTA] &&
1041                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1042                         printk(KERN_ERR "EXT3-fs: old and new quota "
1043                                         "format mixing.\n");
1044                         return 0;
1045                 }
1046
1047                 if (!sbi->s_jquota_fmt) {
1048                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1049                                         "not specified.\n");
1050                         return 0;
1051                 }
1052         } else {
1053                 if (sbi->s_jquota_fmt) {
1054                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1055                                         "specified with no journalling "
1056                                         "enabled.\n");
1057                         return 0;
1058                 }
1059         }
1060 #endif
1061         return 1;
1062 }
1063
1064 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1065                             int read_only)
1066 {
1067         struct ext3_sb_info *sbi = EXT3_SB(sb);
1068         int res = 0;
1069
1070         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1071                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1072                         "forcing read-only mode\n");
1073                 res = MS_RDONLY;
1074         }
1075         if (read_only)
1076                 return res;
1077         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1078                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1079                         "running e2fsck is recommended\n");
1080         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1081                 printk (KERN_WARNING
1082                         "EXT3-fs warning: mounting fs with errors, "
1083                         "running e2fsck is recommended\n");
1084         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1085                  le16_to_cpu(es->s_mnt_count) >=
1086                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1087                 printk (KERN_WARNING
1088                         "EXT3-fs warning: maximal mount count reached, "
1089                         "running e2fsck is recommended\n");
1090         else if (le32_to_cpu(es->s_checkinterval) &&
1091                 (le32_to_cpu(es->s_lastcheck) +
1092                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1093                 printk (KERN_WARNING
1094                         "EXT3-fs warning: checktime reached, "
1095                         "running e2fsck is recommended\n");
1096 #if 0
1097                 /* @@@ We _will_ want to clear the valid bit if we find
1098                    inconsistencies, to force a fsck at reboot.  But for
1099                    a plain journaled filesystem we can keep it set as
1100                    valid forever! :) */
1101         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1102 #endif
1103         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1104                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1105         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1106         es->s_mtime = cpu_to_le32(get_seconds());
1107         ext3_update_dynamic_rev(sb);
1108         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1109
1110         ext3_commit_super(sb, es, 1);
1111         if (test_opt(sb, DEBUG))
1112                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1113                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1114                         sb->s_blocksize,
1115                         sbi->s_groups_count,
1116                         EXT3_BLOCKS_PER_GROUP(sb),
1117                         EXT3_INODES_PER_GROUP(sb),
1118                         sbi->s_mount_opt);
1119
1120         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1121         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1122                 char b[BDEVNAME_SIZE];
1123
1124                 printk("external journal on %s\n",
1125                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1126         } else {
1127                 printk("internal journal\n");
1128         }
1129         return res;
1130 }
1131
1132 /* Called at mount-time, super-block is locked */
1133 static int ext3_check_descriptors (struct super_block * sb)
1134 {
1135         struct ext3_sb_info *sbi = EXT3_SB(sb);
1136         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1137         struct ext3_group_desc * gdp = NULL;
1138         int desc_block = 0;
1139         int i;
1140
1141         ext3_debug ("Checking group descriptors");
1142
1143         for (i = 0; i < sbi->s_groups_count; i++)
1144         {
1145                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1146                         gdp = (struct ext3_group_desc *)
1147                                         sbi->s_group_desc[desc_block++]->b_data;
1148                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1149                     le32_to_cpu(gdp->bg_block_bitmap) >=
1150                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1151                 {
1152                         ext3_error (sb, "ext3_check_descriptors",
1153                                     "Block bitmap for group %d"
1154                                     " not in group (block %lu)!",
1155                                     i, (unsigned long)
1156                                         le32_to_cpu(gdp->bg_block_bitmap));
1157                         return 0;
1158                 }
1159                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1160                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1161                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1162                 {
1163                         ext3_error (sb, "ext3_check_descriptors",
1164                                     "Inode bitmap for group %d"
1165                                     " not in group (block %lu)!",
1166                                     i, (unsigned long)
1167                                         le32_to_cpu(gdp->bg_inode_bitmap));
1168                         return 0;
1169                 }
1170                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1171                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1172                     block + EXT3_BLOCKS_PER_GROUP(sb))
1173                 {
1174                         ext3_error (sb, "ext3_check_descriptors",
1175                                     "Inode table for group %d"
1176                                     " not in group (block %lu)!",
1177                                     i, (unsigned long)
1178                                         le32_to_cpu(gdp->bg_inode_table));
1179                         return 0;
1180                 }
1181                 block += EXT3_BLOCKS_PER_GROUP(sb);
1182                 gdp++;
1183         }
1184
1185         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1186         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1187         return 1;
1188 }
1189
1190
1191 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1192  * the superblock) which were deleted from all directories, but held open by
1193  * a process at the time of a crash.  We walk the list and try to delete these
1194  * inodes at recovery time (only with a read-write filesystem).
1195  *
1196  * In order to keep the orphan inode chain consistent during traversal (in
1197  * case of crash during recovery), we link each inode into the superblock
1198  * orphan list_head and handle it the same way as an inode deletion during
1199  * normal operation (which journals the operations for us).
1200  *
1201  * We only do an iget() and an iput() on each inode, which is very safe if we
1202  * accidentally point at an in-use or already deleted inode.  The worst that
1203  * can happen in this case is that we get a "bit already cleared" message from
1204  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1205  * e2fsck was run on this filesystem, and it must have already done the orphan
1206  * inode cleanup for us, so we can safely abort without any further action.
1207  */
1208 static void ext3_orphan_cleanup (struct super_block * sb,
1209                                  struct ext3_super_block * es)
1210 {
1211         unsigned int s_flags = sb->s_flags;
1212         int nr_orphans = 0, nr_truncates = 0;
1213 #ifdef CONFIG_QUOTA
1214         int i;
1215 #endif
1216         if (!es->s_last_orphan) {
1217                 jbd_debug(4, "no orphan inodes to clean up\n");
1218                 return;
1219         }
1220
1221         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1222                 if (es->s_last_orphan)
1223                         jbd_debug(1, "Errors on filesystem, "
1224                                   "clearing orphan list.\n");
1225                 es->s_last_orphan = 0;
1226                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1227                 return;
1228         }
1229
1230         if (s_flags & MS_RDONLY) {
1231                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1232                        sb->s_id);
1233                 sb->s_flags &= ~MS_RDONLY;
1234         }
1235 #ifdef CONFIG_QUOTA
1236         /* Needed for iput() to work correctly and not trash data */
1237         sb->s_flags |= MS_ACTIVE;
1238         /* Turn on quotas so that they are updated correctly */
1239         for (i = 0; i < MAXQUOTAS; i++) {
1240                 if (EXT3_SB(sb)->s_qf_names[i]) {
1241                         int ret = ext3_quota_on_mount(sb, i);
1242                         if (ret < 0)
1243                                 printk(KERN_ERR
1244                                         "EXT3-fs: Cannot turn on journalled "
1245                                         "quota: error %d\n", ret);
1246                 }
1247         }
1248 #endif
1249
1250         while (es->s_last_orphan) {
1251                 struct inode *inode;
1252
1253                 if (!(inode =
1254                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1255                         es->s_last_orphan = 0;
1256                         break;
1257                 }
1258
1259                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1260                 DQUOT_INIT(inode);
1261                 if (inode->i_nlink) {
1262                         printk(KERN_DEBUG
1263                                 "%s: truncating inode %ld to %Ld bytes\n",
1264                                 __FUNCTION__, inode->i_ino, inode->i_size);
1265                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1266                                   inode->i_ino, inode->i_size);
1267                         ext3_truncate(inode);
1268                         nr_truncates++;
1269                 } else {
1270                         printk(KERN_DEBUG
1271                                 "%s: deleting unreferenced inode %ld\n",
1272                                 __FUNCTION__, inode->i_ino);
1273                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1274                                   inode->i_ino);
1275                         nr_orphans++;
1276                 }
1277                 iput(inode);  /* The delete magic happens here! */
1278         }
1279
1280 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1281
1282         if (nr_orphans)
1283                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1284                        sb->s_id, PLURAL(nr_orphans));
1285         if (nr_truncates)
1286                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1287                        sb->s_id, PLURAL(nr_truncates));
1288 #ifdef CONFIG_QUOTA
1289         /* Turn quotas off */
1290         for (i = 0; i < MAXQUOTAS; i++) {
1291                 if (sb_dqopt(sb)->files[i])
1292                         vfs_quota_off(sb, i);
1293         }
1294 #endif
1295         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1296 }
1297
1298 #define log2(n) ffz(~(n))
1299
1300 /*
1301  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1302  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1303  * We need to be 1 filesystem block less than the 2^32 sector limit.
1304  */
1305 static loff_t ext3_max_size(int bits)
1306 {
1307         loff_t res = EXT3_NDIR_BLOCKS;
1308         /* This constant is calculated to be the largest file size for a
1309          * dense, 4k-blocksize file such that the total number of
1310          * sectors in the file, including data and all indirect blocks,
1311          * does not exceed 2^32. */
1312         const loff_t upper_limit = 0x1ff7fffd000LL;
1313
1314         res += 1LL << (bits-2);
1315         res += 1LL << (2*(bits-2));
1316         res += 1LL << (3*(bits-2));
1317         res <<= bits;
1318         if (res > upper_limit)
1319                 res = upper_limit;
1320         return res;
1321 }
1322
1323 static unsigned long descriptor_loc(struct super_block *sb,
1324                                     unsigned long logic_sb_block,
1325                                     int nr)
1326 {
1327         struct ext3_sb_info *sbi = EXT3_SB(sb);
1328         unsigned long bg, first_data_block, first_meta_bg;
1329         int has_super = 0;
1330
1331         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1332         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1333
1334         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1335             nr < first_meta_bg)
1336                 return (logic_sb_block + nr + 1);
1337         bg = sbi->s_desc_per_block * nr;
1338         if (ext3_bg_has_super(sb, bg))
1339                 has_super = 1;
1340         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1341 }
1342
1343
1344 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1345 {
1346         struct buffer_head * bh;
1347         struct ext3_super_block *es = NULL;
1348         struct ext3_sb_info *sbi;
1349         unsigned long block;
1350         unsigned long sb_block = get_sb_block(&data);
1351         unsigned long logic_sb_block;
1352         unsigned long offset = 0;
1353         unsigned long journal_inum = 0;
1354         unsigned long journal_devnum = 0;
1355         unsigned long def_mount_opts;
1356         struct inode *root;
1357         int blocksize;
1358         int hblock;
1359         int db_count;
1360         int i;
1361         int needs_recovery;
1362         __le32 features;
1363
1364         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1365         if (!sbi)
1366                 return -ENOMEM;
1367         sb->s_fs_info = sbi;
1368         memset(sbi, 0, sizeof(*sbi));
1369         sbi->s_mount_opt = 0;
1370         sbi->s_resuid = EXT3_DEF_RESUID;
1371         sbi->s_resgid = EXT3_DEF_RESGID;
1372
1373         unlock_kernel();
1374
1375         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1376         if (!blocksize) {
1377                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1378                 goto out_fail;
1379         }
1380
1381         /*
1382          * The ext3 superblock will not be buffer aligned for other than 1kB
1383          * block sizes.  We need to calculate the offset from buffer start.
1384          */
1385         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1386                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1387                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1388         } else {
1389                 logic_sb_block = sb_block;
1390         }
1391
1392         if (!(bh = sb_bread(sb, logic_sb_block))) {
1393                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1394                 goto out_fail;
1395         }
1396         /*
1397          * Note: s_es must be initialized as soon as possible because
1398          *       some ext3 macro-instructions depend on its value
1399          */
1400         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1401         sbi->s_es = es;
1402         sb->s_magic = le16_to_cpu(es->s_magic);
1403         if (sb->s_magic != EXT3_SUPER_MAGIC)
1404                 goto cantfind_ext3;
1405
1406         /* Set defaults before we parse the mount options */
1407         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1408         if (def_mount_opts & EXT3_DEFM_DEBUG)
1409                 set_opt(sbi->s_mount_opt, DEBUG);
1410         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1411                 set_opt(sbi->s_mount_opt, GRPID);
1412         if (def_mount_opts & EXT3_DEFM_UID16)
1413                 set_opt(sbi->s_mount_opt, NO_UID32);
1414         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1415                 set_opt(sbi->s_mount_opt, XATTR_USER);
1416         if (def_mount_opts & EXT3_DEFM_ACL)
1417                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1418         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1419                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1420         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1421                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1422         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1423                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1424
1425         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1426                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1427         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1428                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1429
1430         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1431         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1432
1433         set_opt(sbi->s_mount_opt, RESERVATION);
1434
1435         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1436                             NULL, 0))
1437                 goto failed_mount;
1438
1439         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAGXID)
1440                 sb->s_flags |= MS_TAGXID;
1441
1442         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1443                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1444
1445         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1446             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1447              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1448              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1449                 printk(KERN_WARNING
1450                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1451                        "running e2fsck is recommended\n");
1452         /*
1453          * Check feature flags regardless of the revision level, since we
1454          * previously didn't change the revision level when setting the flags,
1455          * so there is a chance incompat flags are set on a rev 0 filesystem.
1456          */
1457         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1458         if (features) {
1459                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1460                        "unsupported optional features (%x).\n",
1461                        sb->s_id, le32_to_cpu(features));
1462                 goto failed_mount;
1463         }
1464         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1465         if (!(sb->s_flags & MS_RDONLY) && features) {
1466                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1467                        "unsupported optional features (%x).\n",
1468                        sb->s_id, le32_to_cpu(features));
1469                 goto failed_mount;
1470         }
1471         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1472
1473         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1474             blocksize > EXT3_MAX_BLOCK_SIZE) {
1475                 printk(KERN_ERR
1476                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1477                        blocksize, sb->s_id);
1478                 goto failed_mount;
1479         }
1480
1481         hblock = bdev_hardsect_size(sb->s_bdev);
1482         if (sb->s_blocksize != blocksize) {
1483                 /*
1484                  * Make sure the blocksize for the filesystem is larger
1485                  * than the hardware sectorsize for the machine.
1486                  */
1487                 if (blocksize < hblock) {
1488                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1489                                "device blocksize %d.\n", blocksize, hblock);
1490                         goto failed_mount;
1491                 }
1492
1493                 brelse (bh);
1494                 sb_set_blocksize(sb, blocksize);
1495                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1496                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1497                 bh = sb_bread(sb, logic_sb_block);
1498                 if (!bh) {
1499                         printk(KERN_ERR
1500                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1501                         goto failed_mount;
1502                 }
1503                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1504                 sbi->s_es = es;
1505                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1506                         printk (KERN_ERR
1507                                 "EXT3-fs: Magic mismatch, very weird !\n");
1508                         goto failed_mount;
1509                 }
1510         }
1511
1512         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1513
1514         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1515                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1516                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1517         } else {
1518                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1519                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1520                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1521                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1522                     (sbi->s_inode_size > blocksize)) {
1523                         printk (KERN_ERR
1524                                 "EXT3-fs: unsupported inode size: %d\n",
1525                                 sbi->s_inode_size);
1526                         goto failed_mount;
1527                 }
1528         }
1529         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1530                                    le32_to_cpu(es->s_log_frag_size);
1531         if (blocksize != sbi->s_frag_size) {
1532                 printk(KERN_ERR
1533                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1534                        sbi->s_frag_size, blocksize);
1535                 goto failed_mount;
1536         }
1537         sbi->s_frags_per_block = 1;
1538         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1539         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1540         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1541         if (EXT3_INODE_SIZE(sb) == 0)
1542                 goto cantfind_ext3;
1543         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1544         if (sbi->s_inodes_per_block == 0)
1545                 goto cantfind_ext3;
1546         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1547                                         sbi->s_inodes_per_block;
1548         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1549         sbi->s_sbh = bh;
1550         sbi->s_mount_state = le16_to_cpu(es->s_state);
1551         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1552         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1553         for (i=0; i < 4; i++)
1554                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1555         sbi->s_def_hash_version = es->s_def_hash_version;
1556
1557         if (sbi->s_blocks_per_group > blocksize * 8) {
1558                 printk (KERN_ERR
1559                         "EXT3-fs: #blocks per group too big: %lu\n",
1560                         sbi->s_blocks_per_group);
1561                 goto failed_mount;
1562         }
1563         if (sbi->s_frags_per_group > blocksize * 8) {
1564                 printk (KERN_ERR
1565                         "EXT3-fs: #fragments per group too big: %lu\n",
1566                         sbi->s_frags_per_group);
1567                 goto failed_mount;
1568         }
1569         if (sbi->s_inodes_per_group > blocksize * 8) {
1570                 printk (KERN_ERR
1571                         "EXT3-fs: #inodes per group too big: %lu\n",
1572                         sbi->s_inodes_per_group);
1573                 goto failed_mount;
1574         }
1575
1576         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1577                 goto cantfind_ext3;
1578         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1579                                le32_to_cpu(es->s_first_data_block) +
1580                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1581                               EXT3_BLOCKS_PER_GROUP(sb);
1582         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1583                    EXT3_DESC_PER_BLOCK(sb);
1584         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1585                                     GFP_KERNEL);
1586         if (sbi->s_group_desc == NULL) {
1587                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1588                 goto failed_mount;
1589         }
1590
1591         percpu_counter_init(&sbi->s_freeblocks_counter);
1592         percpu_counter_init(&sbi->s_freeinodes_counter);
1593         percpu_counter_init(&sbi->s_dirs_counter);
1594         bgl_lock_init(&sbi->s_blockgroup_lock);
1595
1596         for (i = 0; i < db_count; i++) {
1597                 block = descriptor_loc(sb, logic_sb_block, i);
1598                 sbi->s_group_desc[i] = sb_bread(sb, block);
1599                 if (!sbi->s_group_desc[i]) {
1600                         printk (KERN_ERR "EXT3-fs: "
1601                                 "can't read group descriptor %d\n", i);
1602                         db_count = i;
1603                         goto failed_mount2;
1604                 }
1605         }
1606         if (!ext3_check_descriptors (sb)) {
1607                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1608                 goto failed_mount2;
1609         }
1610         sbi->s_gdb_count = db_count;
1611         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1612         spin_lock_init(&sbi->s_next_gen_lock);
1613         /* per fileystem reservation list head & lock */
1614         spin_lock_init(&sbi->s_rsv_window_lock);
1615         sbi->s_rsv_window_root = RB_ROOT;
1616         /* Add a single, static dummy reservation to the start of the
1617          * reservation window list --- it gives us a placeholder for
1618          * append-at-start-of-list which makes the allocation logic
1619          * _much_ simpler. */
1620         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1621         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1622         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1623         sbi->s_rsv_window_head.rsv_goal_size = 0;
1624         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1625
1626         /*
1627          * set up enough so that it can read an inode
1628          */
1629         sb->s_op = &ext3_sops;
1630         sb->s_export_op = &ext3_export_ops;
1631         sb->s_xattr = ext3_xattr_handlers;
1632 #ifdef CONFIG_QUOTA
1633         sb->s_qcop = &ext3_qctl_operations;
1634         sb->dq_op = &ext3_quota_operations;
1635 #endif
1636         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1637
1638         sb->s_root = NULL;
1639
1640         needs_recovery = (es->s_last_orphan != 0 ||
1641                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1642                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1643
1644         /*
1645          * The first inode we look at is the journal inode.  Don't try
1646          * root first: it may be modified in the journal!
1647          */
1648         if (!test_opt(sb, NOLOAD) &&
1649             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1650                 if (ext3_load_journal(sb, es, journal_devnum))
1651                         goto failed_mount2;
1652         } else if (journal_inum) {
1653                 if (ext3_create_journal(sb, es, journal_inum))
1654                         goto failed_mount2;
1655         } else {
1656                 if (!silent)
1657                         printk (KERN_ERR
1658                                 "ext3: No journal on filesystem on %s\n",
1659                                 sb->s_id);
1660                 goto failed_mount2;
1661         }
1662
1663         /* We have now updated the journal if required, so we can
1664          * validate the data journaling mode. */
1665         switch (test_opt(sb, DATA_FLAGS)) {
1666         case 0:
1667                 /* No mode set, assume a default based on the journal
1668                    capabilities: ORDERED_DATA if the journal can
1669                    cope, else JOURNAL_DATA */
1670                 if (journal_check_available_features
1671                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1672                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1673                 else
1674                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1675                 break;
1676
1677         case EXT3_MOUNT_ORDERED_DATA:
1678         case EXT3_MOUNT_WRITEBACK_DATA:
1679                 if (!journal_check_available_features
1680                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1681                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1682                                "requested data journaling mode\n");
1683                         goto failed_mount3;
1684                 }
1685         default:
1686                 break;
1687         }
1688
1689         if (test_opt(sb, NOBH)) {
1690                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1691                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1692                                 "its supported only with writeback mode\n");
1693                         clear_opt(sbi->s_mount_opt, NOBH);
1694                 }
1695         }
1696         /*
1697          * The journal_load will have done any necessary log recovery,
1698          * so we can safely mount the rest of the filesystem now.
1699          */
1700
1701         root = iget(sb, EXT3_ROOT_INO);
1702         sb->s_root = d_alloc_root(root);
1703         if (!sb->s_root) {
1704                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1705                 iput(root);
1706                 goto failed_mount3;
1707         }
1708         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1709                 dput(sb->s_root);
1710                 sb->s_root = NULL;
1711                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1712                 goto failed_mount3;
1713         }
1714
1715         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1716         /*
1717          * akpm: core read_super() calls in here with the superblock locked.
1718          * That deadlocks, because orphan cleanup needs to lock the superblock
1719          * in numerous places.  Here we just pop the lock - it's relatively
1720          * harmless, because we are now ready to accept write_super() requests,
1721          * and aviro says that's the only reason for hanging onto the
1722          * superblock lock.
1723          */
1724         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1725         ext3_orphan_cleanup(sb, es);
1726         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1727         if (needs_recovery)
1728                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1729         ext3_mark_recovery_complete(sb, es);
1730         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1731                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1732                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1733                 "writeback");
1734
1735         percpu_counter_mod(&sbi->s_freeblocks_counter,
1736                 ext3_count_free_blocks(sb));
1737         percpu_counter_mod(&sbi->s_freeinodes_counter,
1738                 ext3_count_free_inodes(sb));
1739         percpu_counter_mod(&sbi->s_dirs_counter,
1740                 ext3_count_dirs(sb));
1741
1742         lock_kernel();
1743         return 0;
1744
1745 cantfind_ext3:
1746         if (!silent)
1747                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1748                        sb->s_id);
1749         goto failed_mount;
1750
1751 failed_mount3:
1752         journal_destroy(sbi->s_journal);
1753 failed_mount2:
1754         for (i = 0; i < db_count; i++)
1755                 brelse(sbi->s_group_desc[i]);
1756         kfree(sbi->s_group_desc);
1757 failed_mount:
1758 #ifdef CONFIG_QUOTA
1759         for (i = 0; i < MAXQUOTAS; i++)
1760                 kfree(sbi->s_qf_names[i]);
1761 #endif
1762         ext3_blkdev_remove(sbi);
1763         brelse(bh);
1764 out_fail:
1765         sb->s_fs_info = NULL;
1766         kfree(sbi);
1767         lock_kernel();
1768         return -EINVAL;
1769 }
1770
1771 /*
1772  * Setup any per-fs journal parameters now.  We'll do this both on
1773  * initial mount, once the journal has been initialised but before we've
1774  * done any recovery; and again on any subsequent remount.
1775  */
1776 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1777 {
1778         struct ext3_sb_info *sbi = EXT3_SB(sb);
1779
1780         if (sbi->s_commit_interval)
1781                 journal->j_commit_interval = sbi->s_commit_interval;
1782         /* We could also set up an ext3-specific default for the commit
1783          * interval here, but for now we'll just fall back to the jbd
1784          * default. */
1785
1786         spin_lock(&journal->j_state_lock);
1787         if (test_opt(sb, BARRIER))
1788                 journal->j_flags |= JFS_BARRIER;
1789         else
1790                 journal->j_flags &= ~JFS_BARRIER;
1791         spin_unlock(&journal->j_state_lock);
1792 }
1793
1794 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1795 {
1796         struct inode *journal_inode;
1797         journal_t *journal;
1798
1799         /* First, test for the existence of a valid inode on disk.  Bad
1800          * things happen if we iget() an unused inode, as the subsequent
1801          * iput() will try to delete it. */
1802
1803         journal_inode = iget(sb, journal_inum);
1804         if (!journal_inode) {
1805                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1806                 return NULL;
1807         }
1808         if (!journal_inode->i_nlink) {
1809                 make_bad_inode(journal_inode);
1810                 iput(journal_inode);
1811                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1812                 return NULL;
1813         }
1814
1815         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1816                   journal_inode, journal_inode->i_size);
1817         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1818                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1819                 iput(journal_inode);
1820                 return NULL;
1821         }
1822
1823         journal = journal_init_inode(journal_inode);
1824         if (!journal) {
1825                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1826                 iput(journal_inode);
1827                 return NULL;
1828         }
1829         journal->j_private = sb;
1830         ext3_init_journal_params(sb, journal);
1831         return journal;
1832 }
1833
1834 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1835                                        dev_t j_dev)
1836 {
1837         struct buffer_head * bh;
1838         journal_t *journal;
1839         int start;
1840         int len;
1841         int hblock, blocksize;
1842         unsigned long sb_block;
1843         unsigned long offset;
1844         struct ext3_super_block * es;
1845         struct block_device *bdev;
1846
1847         bdev = ext3_blkdev_get(j_dev);
1848         if (bdev == NULL)
1849                 return NULL;
1850
1851         if (bd_claim(bdev, sb)) {
1852                 printk(KERN_ERR
1853                         "EXT3: failed to claim external journal device.\n");
1854                 blkdev_put(bdev);
1855                 return NULL;
1856         }
1857
1858         blocksize = sb->s_blocksize;
1859         hblock = bdev_hardsect_size(bdev);
1860         if (blocksize < hblock) {
1861                 printk(KERN_ERR
1862                         "EXT3-fs: blocksize too small for journal device.\n");
1863                 goto out_bdev;
1864         }
1865
1866         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1867         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1868         set_blocksize(bdev, blocksize);
1869         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1870                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1871                        "external journal\n");
1872                 goto out_bdev;
1873         }
1874
1875         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1876         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1877             !(le32_to_cpu(es->s_feature_incompat) &
1878               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1879                 printk(KERN_ERR "EXT3-fs: external journal has "
1880                                         "bad superblock\n");
1881                 brelse(bh);
1882                 goto out_bdev;
1883         }
1884
1885         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1886                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1887                 brelse(bh);
1888                 goto out_bdev;
1889         }
1890
1891         len = le32_to_cpu(es->s_blocks_count);
1892         start = sb_block + 1;
1893         brelse(bh);     /* we're done with the superblock */
1894
1895         journal = journal_init_dev(bdev, sb->s_bdev,
1896                                         start, len, blocksize);
1897         if (!journal) {
1898                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1899                 goto out_bdev;
1900         }
1901         journal->j_private = sb;
1902         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1903         wait_on_buffer(journal->j_sb_buffer);
1904         if (!buffer_uptodate(journal->j_sb_buffer)) {
1905                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1906                 goto out_journal;
1907         }
1908         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1909                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1910                                         "user (unsupported) - %d\n",
1911                         be32_to_cpu(journal->j_superblock->s_nr_users));
1912                 goto out_journal;
1913         }
1914         EXT3_SB(sb)->journal_bdev = bdev;
1915         ext3_init_journal_params(sb, journal);
1916         return journal;
1917 out_journal:
1918         journal_destroy(journal);
1919 out_bdev:
1920         ext3_blkdev_put(bdev);
1921         return NULL;
1922 }
1923
1924 static int ext3_load_journal(struct super_block *sb,
1925                              struct ext3_super_block *es,
1926                              unsigned long journal_devnum)
1927 {
1928         journal_t *journal;
1929         int journal_inum = le32_to_cpu(es->s_journal_inum);
1930         dev_t journal_dev;
1931         int err = 0;
1932         int really_read_only;
1933
1934         if (journal_devnum &&
1935             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1936                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1937                         "numbers have changed\n");
1938                 journal_dev = new_decode_dev(journal_devnum);
1939         } else
1940                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1941
1942         really_read_only = bdev_read_only(sb->s_bdev);
1943
1944         /*
1945          * Are we loading a blank journal or performing recovery after a
1946          * crash?  For recovery, we need to check in advance whether we
1947          * can get read-write access to the device.
1948          */
1949
1950         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1951                 if (sb->s_flags & MS_RDONLY) {
1952                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1953                                         "required on readonly filesystem.\n");
1954                         if (really_read_only) {
1955                                 printk(KERN_ERR "EXT3-fs: write access "
1956                                         "unavailable, cannot proceed.\n");
1957                                 return -EROFS;
1958                         }
1959                         printk (KERN_INFO "EXT3-fs: write access will "
1960                                         "be enabled during recovery.\n");
1961                 }
1962         }
1963
1964         if (journal_inum && journal_dev) {
1965                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1966                        "and inode journals!\n");
1967                 return -EINVAL;
1968         }
1969
1970         if (journal_inum) {
1971                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1972                         return -EINVAL;
1973         } else {
1974                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1975                         return -EINVAL;
1976         }
1977
1978         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1979                 err = journal_update_format(journal);
1980                 if (err)  {
1981                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1982                         journal_destroy(journal);
1983                         return err;
1984                 }
1985         }
1986
1987         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1988                 err = journal_wipe(journal, !really_read_only);
1989         if (!err)
1990                 err = journal_load(journal);
1991
1992         if (err) {
1993                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1994                 journal_destroy(journal);
1995                 return err;
1996         }
1997
1998         EXT3_SB(sb)->s_journal = journal;
1999         ext3_clear_journal_err(sb, es);
2000
2001         if (journal_devnum &&
2002             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2003                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2004                 sb->s_dirt = 1;
2005
2006                 /* Make sure we flush the recovery flag to disk. */
2007                 ext3_commit_super(sb, es, 1);
2008         }
2009
2010         return 0;
2011 }
2012
2013 static int ext3_create_journal(struct super_block * sb,
2014                                struct ext3_super_block * es,
2015                                int journal_inum)
2016 {
2017         journal_t *journal;
2018
2019         if (sb->s_flags & MS_RDONLY) {
2020                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2021                                 "create journal.\n");
2022                 return -EROFS;
2023         }
2024
2025         if (!(journal = ext3_get_journal(sb, journal_inum)))
2026                 return -EINVAL;
2027
2028         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2029                journal_inum);
2030
2031         if (journal_create(journal)) {
2032                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2033                 journal_destroy(journal);
2034                 return -EIO;
2035         }
2036
2037         EXT3_SB(sb)->s_journal = journal;
2038
2039         ext3_update_dynamic_rev(sb);
2040         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2041         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2042
2043         es->s_journal_inum = cpu_to_le32(journal_inum);
2044         sb->s_dirt = 1;
2045
2046         /* Make sure we flush the recovery flag to disk. */
2047         ext3_commit_super(sb, es, 1);
2048
2049         return 0;
2050 }
2051
2052 static void ext3_commit_super (struct super_block * sb,
2053                                struct ext3_super_block * es,
2054                                int sync)
2055 {
2056         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2057
2058         if (!sbh)
2059                 return;
2060         es->s_wtime = cpu_to_le32(get_seconds());
2061         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2062         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2063         BUFFER_TRACE(sbh, "marking dirty");
2064         mark_buffer_dirty(sbh);
2065         if (sync)
2066                 sync_dirty_buffer(sbh);
2067 }
2068
2069
2070 /*
2071  * Have we just finished recovery?  If so, and if we are mounting (or
2072  * remounting) the filesystem readonly, then we will end up with a
2073  * consistent fs on disk.  Record that fact.
2074  */
2075 static void ext3_mark_recovery_complete(struct super_block * sb,
2076                                         struct ext3_super_block * es)
2077 {
2078         journal_t *journal = EXT3_SB(sb)->s_journal;
2079
2080         journal_lock_updates(journal);
2081         journal_flush(journal);
2082         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2083             sb->s_flags & MS_RDONLY) {
2084                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2085                 sb->s_dirt = 0;
2086                 ext3_commit_super(sb, es, 1);
2087         }
2088         journal_unlock_updates(journal);
2089 }
2090
2091 /*
2092  * If we are mounting (or read-write remounting) a filesystem whose journal
2093  * has recorded an error from a previous lifetime, move that error to the
2094  * main filesystem now.
2095  */
2096 static void ext3_clear_journal_err(struct super_block * sb,
2097                                    struct ext3_super_block * es)
2098 {
2099         journal_t *journal;
2100         int j_errno;
2101         const char *errstr;
2102
2103         journal = EXT3_SB(sb)->s_journal;
2104
2105         /*
2106          * Now check for any error status which may have been recorded in the
2107          * journal by a prior ext3_error() or ext3_abort()
2108          */
2109
2110         j_errno = journal_errno(journal);
2111         if (j_errno) {
2112                 char nbuf[16];
2113
2114                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2115                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2116                              "from previous mount: %s", errstr);
2117                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2118                              "filesystem check.");
2119
2120                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2121                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2122                 ext3_commit_super (sb, es, 1);
2123
2124                 journal_clear_err(journal);
2125         }
2126 }
2127
2128 /*
2129  * Force the running and committing transactions to commit,
2130  * and wait on the commit.
2131  */
2132 int ext3_force_commit(struct super_block *sb)
2133 {
2134         journal_t *journal;
2135         int ret;
2136
2137         if (sb->s_flags & MS_RDONLY)
2138                 return 0;
2139
2140         journal = EXT3_SB(sb)->s_journal;
2141         sb->s_dirt = 0;
2142         ret = ext3_journal_force_commit(journal);
2143         return ret;
2144 }
2145
2146 /*
2147  * Ext3 always journals updates to the superblock itself, so we don't
2148  * have to propagate any other updates to the superblock on disk at this
2149  * point.  Just start an async writeback to get the buffers on their way
2150  * to the disk.
2151  *
2152  * This implicitly triggers the writebehind on sync().
2153  */
2154
2155 static void ext3_write_super (struct super_block * sb)
2156 {
2157         if (mutex_trylock(&sb->s_lock) != 0)
2158                 BUG();
2159         sb->s_dirt = 0;
2160 }
2161
2162 static int ext3_sync_fs(struct super_block *sb, int wait)
2163 {
2164         tid_t target;
2165
2166         sb->s_dirt = 0;
2167         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2168                 if (wait)
2169                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2170         }
2171         return 0;
2172 }
2173
2174 /*
2175  * LVM calls this function before a (read-only) snapshot is created.  This
2176  * gives us a chance to flush the journal completely and mark the fs clean.
2177  */
2178 static void ext3_write_super_lockfs(struct super_block *sb)
2179 {
2180         sb->s_dirt = 0;
2181
2182         if (!(sb->s_flags & MS_RDONLY)) {
2183                 journal_t *journal = EXT3_SB(sb)->s_journal;
2184
2185                 /* Now we set up the journal barrier. */
2186                 journal_lock_updates(journal);
2187                 journal_flush(journal);
2188
2189                 /* Journal blocked and flushed, clear needs_recovery flag. */
2190                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2191                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2192         }
2193 }
2194
2195 /*
2196  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2197  * flag here, even though the filesystem is not technically dirty yet.
2198  */
2199 static void ext3_unlockfs(struct super_block *sb)
2200 {
2201         if (!(sb->s_flags & MS_RDONLY)) {
2202                 lock_super(sb);
2203                 /* Reser the needs_recovery flag before the fs is unlocked. */
2204                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2205                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2206                 unlock_super(sb);
2207                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2208         }
2209 }
2210
2211 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2212 {
2213         struct ext3_super_block * es;
2214         struct ext3_sb_info *sbi = EXT3_SB(sb);
2215         unsigned long n_blocks_count = 0;
2216         unsigned long old_sb_flags;
2217         struct ext3_mount_options old_opts;
2218         int err;
2219 #ifdef CONFIG_QUOTA
2220         int i;
2221 #endif
2222
2223         /* Store the original options */
2224         old_sb_flags = sb->s_flags;
2225         old_opts.s_mount_opt = sbi->s_mount_opt;
2226         old_opts.s_resuid = sbi->s_resuid;
2227         old_opts.s_resgid = sbi->s_resgid;
2228         old_opts.s_commit_interval = sbi->s_commit_interval;
2229 #ifdef CONFIG_QUOTA
2230         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2231         for (i = 0; i < MAXQUOTAS; i++)
2232                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2233 #endif
2234
2235         /*
2236          * Allow the "check" option to be passed as a remount option.
2237          */
2238         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2239                 err = -EINVAL;
2240                 goto restore_opts;
2241         }
2242
2243         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2244                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2245         if ((sbi->s_mount_opt & EXT3_MOUNT_TAGXID) &&
2246                 !(sb->s_flags & MS_TAGXID)) {
2247                 printk("EXT3-fs: %s: tagxid not permitted on remount.\n",
2248                         sb->s_id);
2249                 return -EINVAL;
2250         }
2251
2252         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2253                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2254
2255         es = sbi->s_es;
2256
2257         ext3_init_journal_params(sb, sbi->s_journal);
2258
2259         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2260                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2261                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2262                         err = -EROFS;
2263                         goto restore_opts;
2264                 }
2265
2266                 if (*flags & MS_RDONLY) {
2267                         /*
2268                          * First of all, the unconditional stuff we have to do
2269                          * to disable replay of the journal when we next remount
2270                          */
2271                         sb->s_flags |= MS_RDONLY;
2272
2273                         /*
2274                          * OK, test if we are remounting a valid rw partition
2275                          * readonly, and if so set the rdonly flag and then
2276                          * mark the partition as valid again.
2277                          */
2278                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2279                             (sbi->s_mount_state & EXT3_VALID_FS))
2280                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2281
2282                         ext3_mark_recovery_complete(sb, es);
2283                 } else {
2284                         __le32 ret;
2285                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2286                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2287                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2288                                        "remount RDWR because of unsupported "
2289                                        "optional features (%x).\n",
2290                                        sb->s_id, le32_to_cpu(ret));
2291                                 err = -EROFS;
2292                                 goto restore_opts;
2293                         }
2294                         /*
2295                          * Mounting a RDONLY partition read-write, so reread
2296                          * and store the current valid flag.  (It may have
2297                          * been changed by e2fsck since we originally mounted
2298                          * the partition.)
2299                          */
2300                         ext3_clear_journal_err(sb, es);
2301                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2302                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2303                                 err = ret;
2304                                 goto restore_opts;
2305                         }
2306                         if (!ext3_setup_super (sb, es, 0))
2307                                 sb->s_flags &= ~MS_RDONLY;
2308                 }
2309         }
2310 #ifdef CONFIG_QUOTA
2311         /* Release old quota file names */
2312         for (i = 0; i < MAXQUOTAS; i++)
2313                 if (old_opts.s_qf_names[i] &&
2314                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2315                         kfree(old_opts.s_qf_names[i]);
2316 #endif
2317         return 0;
2318 restore_opts:
2319         sb->s_flags = old_sb_flags;
2320         sbi->s_mount_opt = old_opts.s_mount_opt;
2321         sbi->s_resuid = old_opts.s_resuid;
2322         sbi->s_resgid = old_opts.s_resgid;
2323         sbi->s_commit_interval = old_opts.s_commit_interval;
2324 #ifdef CONFIG_QUOTA
2325         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2326         for (i = 0; i < MAXQUOTAS; i++) {
2327                 if (sbi->s_qf_names[i] &&
2328                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2329                         kfree(sbi->s_qf_names[i]);
2330                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2331         }
2332 #endif
2333         return err;
2334 }
2335
2336 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2337 {
2338         struct ext3_sb_info *sbi = EXT3_SB(sb);
2339         struct ext3_super_block *es = sbi->s_es;
2340         unsigned long overhead;
2341         int i;
2342
2343         if (test_opt (sb, MINIX_DF))
2344                 overhead = 0;
2345         else {
2346                 unsigned long ngroups;
2347                 ngroups = EXT3_SB(sb)->s_groups_count;
2348                 smp_rmb();
2349
2350                 /*
2351                  * Compute the overhead (FS structures)
2352                  */
2353
2354                 /*
2355                  * All of the blocks before first_data_block are
2356                  * overhead
2357                  */
2358                 overhead = le32_to_cpu(es->s_first_data_block);
2359
2360                 /*
2361                  * Add the overhead attributed to the superblock and
2362                  * block group descriptors.  If the sparse superblocks
2363                  * feature is turned on, then not all groups have this.
2364                  */
2365                 for (i = 0; i < ngroups; i++) {
2366                         overhead += ext3_bg_has_super(sb, i) +
2367                                 ext3_bg_num_gdb(sb, i);
2368                         cond_resched();
2369                 }
2370
2371                 /*
2372                  * Every block group has an inode bitmap, a block
2373                  * bitmap, and an inode table.
2374                  */
2375                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2376         }
2377
2378         buf->f_type = EXT3_SUPER_MAGIC;
2379         buf->f_bsize = sb->s_blocksize;
2380         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2381         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2382         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2383         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2384                 buf->f_bavail = 0;
2385         buf->f_files = le32_to_cpu(es->s_inodes_count);
2386         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2387         buf->f_namelen = EXT3_NAME_LEN;
2388         return 0;
2389 }
2390
2391 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2392  * is locked for write. Otherwise the are possible deadlocks:
2393  * Process 1                         Process 2
2394  * ext3_create()                     quota_sync()
2395  *   journal_start()                   write_dquot()
2396  *   DQUOT_INIT()                        down(dqio_mutex)
2397  *     down(dqio_mutex)                    journal_start()
2398  *
2399  */
2400
2401 #ifdef CONFIG_QUOTA
2402
2403 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2404 {
2405         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2406 }
2407
2408 static int ext3_dquot_initialize(struct inode *inode, int type)
2409 {
2410         handle_t *handle;
2411         int ret, err;
2412
2413         /* We may create quota structure so we need to reserve enough blocks */
2414         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2415         if (IS_ERR(handle))
2416                 return PTR_ERR(handle);
2417         ret = dquot_initialize(inode, type);
2418         err = ext3_journal_stop(handle);
2419         if (!ret)
2420                 ret = err;
2421         return ret;
2422 }
2423
2424 static int ext3_dquot_drop(struct inode *inode)
2425 {
2426         handle_t *handle;
2427         int ret, err;
2428
2429         /* We may delete quota structure so we need to reserve enough blocks */
2430         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2431         if (IS_ERR(handle))
2432                 return PTR_ERR(handle);
2433         ret = dquot_drop(inode);
2434         err = ext3_journal_stop(handle);
2435         if (!ret)
2436                 ret = err;
2437         return ret;
2438 }
2439
2440 static int ext3_write_dquot(struct dquot *dquot)
2441 {
2442         int ret, err;
2443         handle_t *handle;
2444         struct inode *inode;
2445
2446         inode = dquot_to_inode(dquot);
2447         handle = ext3_journal_start(inode,
2448                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2449         if (IS_ERR(handle))
2450                 return PTR_ERR(handle);
2451         ret = dquot_commit(dquot);
2452         err = ext3_journal_stop(handle);
2453         if (!ret)
2454                 ret = err;
2455         return ret;
2456 }
2457
2458 static int ext3_acquire_dquot(struct dquot *dquot)
2459 {
2460         int ret, err;
2461         handle_t *handle;
2462
2463         handle = ext3_journal_start(dquot_to_inode(dquot),
2464                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2465         if (IS_ERR(handle))
2466                 return PTR_ERR(handle);
2467         ret = dquot_acquire(dquot);
2468         err = ext3_journal_stop(handle);
2469         if (!ret)
2470                 ret = err;
2471         return ret;
2472 }
2473
2474 static int ext3_release_dquot(struct dquot *dquot)
2475 {
2476         int ret, err;
2477         handle_t *handle;
2478
2479         handle = ext3_journal_start(dquot_to_inode(dquot),
2480                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2481         if (IS_ERR(handle))
2482                 return PTR_ERR(handle);
2483         ret = dquot_release(dquot);
2484         err = ext3_journal_stop(handle);
2485         if (!ret)
2486                 ret = err;
2487         return ret;
2488 }
2489
2490 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2491 {
2492         /* Are we journalling quotas? */
2493         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2494             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2495                 dquot_mark_dquot_dirty(dquot);
2496                 return ext3_write_dquot(dquot);
2497         } else {
2498                 return dquot_mark_dquot_dirty(dquot);
2499         }
2500 }
2501
2502 static int ext3_write_info(struct super_block *sb, int type)
2503 {
2504         int ret, err;
2505         handle_t *handle;
2506
2507         /* Data block + inode block */
2508         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2509         if (IS_ERR(handle))
2510                 return PTR_ERR(handle);
2511         ret = dquot_commit_info(sb, type);
2512         err = ext3_journal_stop(handle);
2513         if (!ret)
2514                 ret = err;
2515         return ret;
2516 }
2517
2518 /*
2519  * Turn on quotas during mount time - we need to find
2520  * the quota file and such...
2521  */
2522 static int ext3_quota_on_mount(struct super_block *sb, int type)
2523 {
2524         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2525                         EXT3_SB(sb)->s_jquota_fmt, type);
2526 }
2527
2528 /*
2529  * Standard function to be called on quota_on
2530  */
2531 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2532                          char *path)
2533 {
2534         int err;
2535         struct nameidata nd;
2536
2537         if (!test_opt(sb, QUOTA))
2538                 return -EINVAL;
2539         /* Not journalling quota? */
2540         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2541             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2542                 return vfs_quota_on(sb, type, format_id, path);
2543         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2544         if (err)
2545                 return err;
2546         /* Quotafile not on the same filesystem? */
2547         if (nd.mnt->mnt_sb != sb) {
2548                 path_release(&nd);
2549                 return -EXDEV;
2550         }
2551         /* Quotafile not of fs root? */
2552         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2553                 printk(KERN_WARNING
2554                         "EXT3-fs: Quota file not on filesystem root. "
2555                         "Journalled quota will not work.\n");
2556         path_release(&nd);
2557         return vfs_quota_on(sb, type, format_id, path);
2558 }
2559
2560 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2561  * acquiring the locks... As quota files are never truncated and quota code
2562  * itself serializes the operations (and noone else should touch the files)
2563  * we don't have to be afraid of races */
2564 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2565                                size_t len, loff_t off)
2566 {
2567         struct inode *inode = sb_dqopt(sb)->files[type];
2568         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2569         int err = 0;
2570         int offset = off & (sb->s_blocksize - 1);
2571         int tocopy;
2572         size_t toread;
2573         struct buffer_head *bh;
2574         loff_t i_size = i_size_read(inode);
2575
2576         if (off > i_size)
2577                 return 0;
2578         if (off+len > i_size)
2579                 len = i_size-off;
2580         toread = len;
2581         while (toread > 0) {
2582                 tocopy = sb->s_blocksize - offset < toread ?
2583                                 sb->s_blocksize - offset : toread;
2584                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2585                 if (err)
2586                         return err;
2587                 if (!bh)        /* A hole? */
2588                         memset(data, 0, tocopy);
2589                 else
2590                         memcpy(data, bh->b_data+offset, tocopy);
2591                 brelse(bh);
2592                 offset = 0;
2593                 toread -= tocopy;
2594                 data += tocopy;
2595                 blk++;
2596         }
2597         return len;
2598 }
2599
2600 /* Write to quotafile (we know the transaction is already started and has
2601  * enough credits) */
2602 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2603                                 const char *data, size_t len, loff_t off)
2604 {
2605         struct inode *inode = sb_dqopt(sb)->files[type];
2606         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2607         int err = 0;
2608         int offset = off & (sb->s_blocksize - 1);
2609         int tocopy;
2610         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2611         size_t towrite = len;
2612         struct buffer_head *bh;
2613         handle_t *handle = journal_current_handle();
2614
2615         mutex_lock(&inode->i_mutex);
2616         while (towrite > 0) {
2617                 tocopy = sb->s_blocksize - offset < towrite ?
2618                                 sb->s_blocksize - offset : towrite;
2619                 bh = ext3_bread(handle, inode, blk, 1, &err);
2620                 if (!bh)
2621                         goto out;
2622                 if (journal_quota) {
2623                         err = ext3_journal_get_write_access(handle, bh);
2624                         if (err) {
2625                                 brelse(bh);
2626                                 goto out;
2627                         }
2628                 }
2629                 lock_buffer(bh);
2630                 memcpy(bh->b_data+offset, data, tocopy);
2631                 flush_dcache_page(bh->b_page);
2632                 unlock_buffer(bh);
2633                 if (journal_quota)
2634                         err = ext3_journal_dirty_metadata(handle, bh);
2635                 else {
2636                         /* Always do at least ordered writes for quotas */
2637                         err = ext3_journal_dirty_data(handle, bh);
2638                         mark_buffer_dirty(bh);
2639                 }
2640                 brelse(bh);
2641                 if (err)
2642                         goto out;
2643                 offset = 0;
2644                 towrite -= tocopy;
2645                 data += tocopy;
2646                 blk++;
2647         }
2648 out:
2649         if (len == towrite)
2650                 return err;
2651         if (inode->i_size < off+len-towrite) {
2652                 i_size_write(inode, off+len-towrite);
2653                 EXT3_I(inode)->i_disksize = inode->i_size;
2654         }
2655         inode->i_version++;
2656         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2657         ext3_mark_inode_dirty(handle, inode);
2658         mutex_unlock(&inode->i_mutex);
2659         return len - towrite;
2660 }
2661
2662 #endif
2663
2664 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2665         int flags, const char *dev_name, void *data)
2666 {
2667         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2668 }
2669
2670 static struct file_system_type ext3_fs_type = {
2671         .owner          = THIS_MODULE,
2672         .name           = "ext3",
2673         .get_sb         = ext3_get_sb,
2674         .kill_sb        = kill_block_super,
2675         .fs_flags       = FS_REQUIRES_DEV,
2676 };
2677
2678 static int __init init_ext3_fs(void)
2679 {
2680         int err = init_ext3_xattr();
2681         if (err)
2682                 return err;
2683         err = init_inodecache();
2684         if (err)
2685                 goto out1;
2686         err = register_filesystem(&ext3_fs_type);
2687         if (err)
2688                 goto out;
2689         return 0;
2690 out:
2691         destroy_inodecache();
2692 out1:
2693         exit_ext3_xattr();
2694         return err;
2695 }
2696
2697 static void __exit exit_ext3_fs(void)
2698 {
2699         unregister_filesystem(&ext3_fs_type);
2700         destroy_inodecache();
2701         exit_ext3_xattr();
2702 }
2703
2704 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2705 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2706 MODULE_LICENSE("GPL");
2707 module_init(init_ext3_fs)
2708 module_exit(exit_ext3_fs)