SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd.h>
  24 #include <linux/ext3_fs.h>
  25 #include <linux/ext3_jbd.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38
  39 #include <asm/uaccess.h>
  40
  41 #include "xattr.h"
  42 #include "acl.h"
  43 #include "namei.h"
  44
  45 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  46                              unsigned long journal_devnum);
  47 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  48                                unsigned int);
  49 static void ext3_commit_super (struct super_block * sb,
  50                                struct ext3_super_block * es,
  51                                int sync);
  52 static void ext3_mark_recovery_complete(struct super_block * sb,
  53                                         struct ext3_super_block * es);
  54 static void ext3_clear_journal_err(struct super_block * sb,
  55                                    struct ext3_super_block * es);
  56 static int ext3_sync_fs(struct super_block *sb, int wait);
  57 static const char *ext3_decode_error(struct super_block * sb, int errno,
  58                                      char nbuf[16]);
  59 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  60 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
  61 static void ext3_unlockfs(struct super_block *sb);
  62 static void ext3_write_super (struct super_block * sb);
  63 static void ext3_write_super_lockfs(struct super_block *sb);
  64
  65 /*
  66  * Wrappers for journal_start/end.
  67  *
  68  * The only special thing we need to do here is to make sure that all
  69  * journal_end calls result in the superblock being marked dirty, so
  70  * that sync() will call the filesystem's write_super callback if
  71  * appropriate.
  72  */
  73 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  74 {
  75         journal_t *journal;
  76
  77         if (sb->s_flags & MS_RDONLY)
  78                 return ERR_PTR(-EROFS);
  79
  80         /* Special case here: if the journal has aborted behind our
  81          * backs (eg. EIO in the commit thread), then we still need to
  82          * take the FS itself readonly cleanly. */
  83         journal = EXT3_SB(sb)->s_journal;
  84         if (is_journal_aborted(journal)) {
  85                 ext3_abort(sb, __FUNCTION__,
  86                            "Detected aborted journal");
  87                 return ERR_PTR(-EROFS);
  88         }
  89
  90         return journal_start(journal, nblocks);
  91 }
  92
  93 /*
  94  * The only special thing we need to do here is to make sure that all
  95  * journal_stop calls result in the superblock being marked dirty, so
  96  * that sync() will call the filesystem's write_super callback if
  97  * appropriate.
  98  */
  99 int __ext3_journal_stop(const char *where, handle_t *handle)
 100 {
 101         struct super_block *sb;
 102         int err;
 103         int rc;
 104
 105         sb = handle->h_transaction->t_journal->j_private;
 106         err = handle->h_err;
 107         rc = journal_stop(handle);
 108
 109         if (!err)
 110                 err = rc;
 111         if (err)
 112                 __ext3_std_error(sb, where, err);
 113         return err;
 114 }
 115
 116 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 117                 struct buffer_head *bh, handle_t *handle, int err)
 118 {
 119         char nbuf[16];
 120         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 121
 122         if (bh)
 123                 BUFFER_TRACE(bh, "abort");
 124
 125         if (!handle->h_err)
 126                 handle->h_err = err;
 127
 128         if (is_handle_aborted(handle))
 129                 return;
 130
 131         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 132                caller, errstr, err_fn);
 133
 134         journal_abort_handle(handle);
 135 }
 136
 137 /* Deal with the reporting of failure conditions on a filesystem such as
 138  * inconsistencies detected or read IO failures.
 139  *
 140  * On ext2, we can store the error state of the filesystem in the
 141  * superblock.  That is not possible on ext3, because we may have other
 142  * write ordering constraints on the superblock which prevent us from
 143  * writing it out straight away; and given that the journal is about to
 144  * be aborted, we can't rely on the current, or future, transactions to
 145  * write out the superblock safely.
 146  *
 147  * We'll just use the journal_abort() error code to record an error in
 148  * the journal instead.  On recovery, the journal will compain about
 149  * that error until we've noted it down and cleared it.
 150  */
 151
 152 static void ext3_handle_error(struct super_block *sb)
 153 {
 154         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 155
 156         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 157         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 158
 159         if (sb->s_flags & MS_RDONLY)
 160                 return;
 161
 162         if (test_opt (sb, ERRORS_RO)) {
 163                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 164                 sb->s_flags |= MS_RDONLY;
 165         } else {
 166                 journal_t *journal = EXT3_SB(sb)->s_journal;
 167
 168                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 169                 if (journal)
 170                         journal_abort(journal, -EIO);
 171         }
 172         if (test_opt(sb, ERRORS_PANIC))
 173                 panic("EXT3-fs (device %s): panic forced after error\n",
 174                         sb->s_id);
 175         ext3_commit_super(sb, es, 1);
 176 }
 177
 178 void ext3_error (struct super_block * sb, const char * function,
 179                  const char * fmt, ...)
 180 {
 181         va_list args;
 182
 183         va_start(args, fmt);
 184         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 185         vprintk(fmt, args);
 186         printk("\n");
 187         va_end(args);
 188
 189         ext3_handle_error(sb);
 190 }
 191
 192 static const char *ext3_decode_error(struct super_block * sb, int errno,
 193                                      char nbuf[16])
 194 {
 195         char *errstr = NULL;
 196
 197         switch (errno) {
 198         case -EIO:
 199                 errstr = "IO failure";
 200                 break;
 201         case -ENOMEM:
 202                 errstr = "Out of memory";
 203                 break;
 204         case -EROFS:
 205                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 206                         errstr = "Journal has aborted";
 207                 else
 208                         errstr = "Readonly filesystem";
 209                 break;
 210         default:
 211                 /* If the caller passed in an extra buffer for unknown
 212                  * errors, textualise them now.  Else we just return
 213                  * NULL. */
 214                 if (nbuf) {
 215                         /* Check for truncated error codes... */
 216                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 217                                 errstr = nbuf;
 218                 }
 219                 break;
 220         }
 221
 222         return errstr;
 223 }
 224
 225 /* __ext3_std_error decodes expected errors from journaling functions
 226  * automatically and invokes the appropriate error response.  */
 227
 228 void __ext3_std_error (struct super_block * sb, const char * function,
 229                        int errno)
 230 {
 231         char nbuf[16];
 232         const char *errstr;
 233
 234         /* Special case: if the error is EROFS, and we're not already
 235          * inside a transaction, then there's really no point in logging
 236          * an error. */
 237         if (errno == -EROFS && journal_current_handle() == NULL &&
 238             (sb->s_flags & MS_RDONLY))
 239                 return;
 240
 241         errstr = ext3_decode_error(sb, errno, nbuf);
 242         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 243                 sb->s_id, function, errstr);
 244
 245         ext3_handle_error(sb);
 246 }
 247
 248 /*
 249  * ext3_abort is a much stronger failure handler than ext3_error.  The
 250  * abort function may be used to deal with unrecoverable failures such
 251  * as journal IO errors or ENOMEM at a critical moment in log management.
 252  *
 253  * We unconditionally force the filesystem into an ABORT|READONLY state,
 254  * unless the error response on the fs has been set to panic in which
 255  * case we take the easy way out and panic immediately.
 256  */
 257
 258 void ext3_abort (struct super_block * sb, const char * function,
 259                  const char * fmt, ...)
 260 {
 261         va_list args;
 262
 263         printk (KERN_CRIT "ext3_abort called.\n");
 264
 265         va_start(args, fmt);
 266         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 267         vprintk(fmt, args);
 268         printk("\n");
 269         va_end(args);
 270
 271         if (test_opt(sb, ERRORS_PANIC))
 272                 panic("EXT3-fs panic from previous error\n");
 273
 274         if (sb->s_flags & MS_RDONLY)
 275                 return;
 276
 277         printk(KERN_CRIT "Remounting filesystem read-only\n");
 278         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 279         sb->s_flags |= MS_RDONLY;
 280         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 281         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 282 }
 283
 284 void ext3_warning (struct super_block * sb, const char * function,
 285                    const char * fmt, ...)
 286 {
 287         va_list args;
 288
 289         va_start(args, fmt);
 290         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 291                sb->s_id, function);
 292         vprintk(fmt, args);
 293         printk("\n");
 294         va_end(args);
 295 }
 296
 297 void ext3_update_dynamic_rev(struct super_block *sb)
 298 {
 299         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 300
 301         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 302                 return;
 303
 304         ext3_warning(sb, __FUNCTION__,
 305                      "updating to rev %d because of new feature flag, "
 306                      "running e2fsck is recommended",
 307                      EXT3_DYNAMIC_REV);
 308
 309         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 310         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 311         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 312         /* leave es->s_feature_*compat flags alone */
 313         /* es->s_uuid will be set by e2fsck if empty */
 314
 315         /*
 316          * The rest of the superblock fields should be zero, and if not it
 317          * means they are likely already in use, so leave them alone.  We
 318          * can leave it up to e2fsck to clean up any inconsistencies there.
 319          */
 320 }
 321
 322 /*
 323  * Open the external journal device
 324  */
 325 static struct block_device *ext3_blkdev_get(dev_t dev)
 326 {
 327         struct block_device *bdev;
 328         char b[BDEVNAME_SIZE];
 329
 330         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 331         if (IS_ERR(bdev))
 332                 goto fail;
 333         return bdev;
 334
 335 fail:
 336         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 337                         __bdevname(dev, b), PTR_ERR(bdev));
 338         return NULL;
 339 }
 340
 341 /*
 342  * Release the journal device
 343  */
 344 static int ext3_blkdev_put(struct block_device *bdev)
 345 {
 346         bd_release(bdev);
 347         return blkdev_put(bdev);
 348 }
 349
 350 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 351 {
 352         struct block_device *bdev;
 353         int ret = -ENODEV;
 354
 355         bdev = sbi->journal_bdev;
 356         if (bdev) {
 357                 ret = ext3_blkdev_put(bdev);
 358                 sbi->journal_bdev = NULL;
 359         }
 360         return ret;
 361 }
 362
 363 static inline struct inode *orphan_list_entry(struct list_head *l)
 364 {
 365         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 366 }
 367
 368 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 369 {
 370         struct list_head *l;
 371
 372         printk(KERN_ERR "sb orphan head is %d\n",
 373                le32_to_cpu(sbi->s_es->s_last_orphan));
 374
 375         printk(KERN_ERR "sb_info orphan list:\n");
 376         list_for_each(l, &sbi->s_orphan) {
 377                 struct inode *inode = orphan_list_entry(l);
 378                 printk(KERN_ERR "  "
 379                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 380                        inode->i_sb->s_id, inode->i_ino, inode,
 381                        inode->i_mode, inode->i_nlink,
 382                        NEXT_ORPHAN(inode));
 383         }
 384 }
 385
 386 static void ext3_put_super (struct super_block * sb)
 387 {
 388         struct ext3_sb_info *sbi = EXT3_SB(sb);
 389         struct ext3_super_block *es = sbi->s_es;
 390         int i;
 391
 392         ext3_xattr_put_super(sb);
 393         journal_destroy(sbi->s_journal);
 394         if (!(sb->s_flags & MS_RDONLY)) {
 395                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 396                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 397                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 398                 mark_buffer_dirty(sbi->s_sbh);
 399                 ext3_commit_super(sb, es, 1);
 400         }
 401
 402         for (i = 0; i < sbi->s_gdb_count; i++)
 403                 brelse(sbi->s_group_desc[i]);
 404         kfree(sbi->s_group_desc);
 405         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 406         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 407         percpu_counter_destroy(&sbi->s_dirs_counter);
 408         brelse(sbi->s_sbh);
 409 #ifdef CONFIG_QUOTA
 410         for (i = 0; i < MAXQUOTAS; i++)
 411                 kfree(sbi->s_qf_names[i]);
 412 #endif
 413
 414         /* Debugging code just in case the in-memory inode orphan list
 415          * isn't empty.  The on-disk one can be non-empty if we've
 416          * detected an error and taken the fs readonly, but the
 417          * in-memory list had better be clean by this point. */
 418         if (!list_empty(&sbi->s_orphan))
 419                 dump_orphan_list(sb, sbi);
 420         J_ASSERT(list_empty(&sbi->s_orphan));
 421
 422         invalidate_bdev(sb->s_bdev, 0);
 423         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 424                 /*
 425                  * Invalidate the journal device's buffers.  We don't want them
 426                  * floating about in memory - the physical journal device may
 427                  * hotswapped, and it breaks the `ro-after' testing code.
 428                  */
 429                 sync_blockdev(sbi->journal_bdev);
 430                 invalidate_bdev(sbi->journal_bdev, 0);
 431                 ext3_blkdev_remove(sbi);
 432         }
 433         sb->s_fs_info = NULL;
 434         kfree(sbi);
 435         return;
 436 }
 437
 438 static kmem_cache_t *ext3_inode_cachep;
 439
 440 /*
 441  * Called inside transaction, so use GFP_NOFS
 442  */
 443 static struct inode *ext3_alloc_inode(struct super_block *sb)
 444 {
 445         struct ext3_inode_info *ei;
 446
 447         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 448         if (!ei)
 449                 return NULL;
 450 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 451         ei->i_acl = EXT3_ACL_NOT_CACHED;
 452         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 453 #endif
 454         ei->i_block_alloc_info = NULL;
 455         ei->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 mutex_init(&ei->truncate_mutex);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, (SLAB_RECLAIM_ACCOUNT|
 484                                                 SLAB_MEM_SPREAD),
 485                                              init_once, NULL);
 486         if (ext3_inode_cachep == NULL)
 487                 return -ENOMEM;
 488         return 0;
 489 }
 490
 491 static void destroy_inodecache(void)
 492 {
 493         if (kmem_cache_destroy(ext3_inode_cachep))
 494                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 495 }
 496
 497 static void ext3_clear_inode(struct inode *inode)
 498 {
 499         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 500 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 501         if (EXT3_I(inode)->i_acl &&
 502                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 503                 posix_acl_release(EXT3_I(inode)->i_acl);
 504                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 505         }
 506         if (EXT3_I(inode)->i_default_acl &&
 507                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 508                 posix_acl_release(EXT3_I(inode)->i_default_acl);
 509                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 510         }
 511 #endif
 512         ext3_discard_reservation(inode);
 513         EXT3_I(inode)->i_block_alloc_info = NULL;
 514         if (unlikely(rsv))
 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
 558 static struct dentry *ext3_get_dentry(struct super_block *sb, void *vobjp)
 559 {
 560         __u32 *objp = vobjp;
 561         unsigned long ino = objp[0];
 562         __u32 generation = objp[1];
 563         struct inode *inode;
 564         struct dentry *result;
 565
 566         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
 567                 return ERR_PTR(-ESTALE);
 568         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 569                 return ERR_PTR(-ESTALE);
 570
 571         /* iget isn't really right if the inode is currently unallocated!!
 572          *
 573          * ext3_read_inode will return a bad_inode if the inode had been
 574          * deleted, so we should be safe.
 575          *
 576          * Currently we don't know the generation for parent directory, so
 577          * a generation of 0 means "accept any"
 578          */
 579         inode = iget(sb, ino);
 580         if (inode == NULL)
 581                 return ERR_PTR(-ENOMEM);
 582         if (is_bad_inode(inode) ||
 583             (generation && inode->i_generation != generation)) {
 584                 iput(inode);
 585                 return ERR_PTR(-ESTALE);
 586         }
 587         /* now to find a dentry.
 588          * If possible, get a well-connected one
 589          */
 590         result = d_alloc_anon(inode);
 591         if (!result) {
 592                 iput(inode);
 593                 return ERR_PTR(-ENOMEM);
 594         }
 595         return result;
 596 }
 597
 598 #ifdef CONFIG_QUOTA
 599 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 600 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 601
 602 static int ext3_dquot_initialize(struct inode *inode, int type);
 603 static int ext3_dquot_drop(struct inode *inode);
 604 static int ext3_write_dquot(struct dquot *dquot);
 605 static int ext3_acquire_dquot(struct dquot *dquot);
 606 static int ext3_release_dquot(struct dquot *dquot);
 607 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 608 static int ext3_write_info(struct super_block *sb, int type);
 609 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 610 static int ext3_quota_on_mount(struct super_block *sb, int type);
 611 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 612                                size_t len, loff_t off);
 613 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 614                                 const char *data, size_t len, loff_t off);
 615
 616 static struct dquot_operations ext3_quota_operations = {
 617         .initialize     = ext3_dquot_initialize,
 618         .drop           = ext3_dquot_drop,
 619         .alloc_space    = dquot_alloc_space,
 620         .alloc_inode    = dquot_alloc_inode,
 621         .free_space     = dquot_free_space,
 622         .free_inode     = dquot_free_inode,
 623         .transfer       = dquot_transfer,
 624         .write_dquot    = ext3_write_dquot,
 625         .acquire_dquot  = ext3_acquire_dquot,
 626         .release_dquot  = ext3_release_dquot,
 627         .mark_dirty     = ext3_mark_dquot_dirty,
 628         .write_info     = ext3_write_info
 629 };
 630
 631 static struct quotactl_ops ext3_qctl_operations = {
 632         .quota_on       = ext3_quota_on,
 633         .quota_off      = vfs_quota_off,
 634         .quota_sync     = vfs_quota_sync,
 635         .get_info       = vfs_get_dqinfo,
 636         .set_info       = vfs_set_dqinfo,
 637         .get_dqblk      = vfs_get_dqblk,
 638         .set_dqblk      = vfs_set_dqblk
 639 };
 640 #endif
 641
 642 static struct super_operations ext3_sops = {
 643         .alloc_inode    = ext3_alloc_inode,
 644         .destroy_inode  = ext3_destroy_inode,
 645         .read_inode     = ext3_read_inode,
 646         .write_inode    = ext3_write_inode,
 647         .dirty_inode    = ext3_dirty_inode,
 648         .delete_inode   = ext3_delete_inode,
 649         .put_super      = ext3_put_super,
 650         .write_super    = ext3_write_super,
 651         .sync_fs        = ext3_sync_fs,
 652         .write_super_lockfs = ext3_write_super_lockfs,
 653         .unlockfs       = ext3_unlockfs,
 654         .statfs         = ext3_statfs,
 655         .remount_fs     = ext3_remount,
 656         .clear_inode    = ext3_clear_inode,
 657         .show_options   = ext3_show_options,
 658 #ifdef CONFIG_QUOTA
 659         .quota_read     = ext3_quota_read,
 660         .quota_write    = ext3_quota_write,
 661 #endif
 662 };
 663
 664 static struct export_operations ext3_export_ops = {
 665         .get_parent = ext3_get_parent,
 666         .get_dentry = ext3_get_dentry,
 667 };
 668
 669 enum {
 670         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 671         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 672         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 673         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 674         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 675         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 676         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 677         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 678         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 679         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 680         Opt_grpquota
 681 };
 682
 683 static match_table_t tokens = {
 684         {Opt_bsd_df, "bsddf"},
 685         {Opt_minix_df, "minixdf"},
 686         {Opt_grpid, "grpid"},
 687         {Opt_grpid, "bsdgroups"},
 688         {Opt_nogrpid, "nogrpid"},
 689         {Opt_nogrpid, "sysvgroups"},
 690         {Opt_resgid, "resgid=%u"},
 691         {Opt_resuid, "resuid=%u"},
 692         {Opt_sb, "sb=%u"},
 693         {Opt_err_cont, "errors=continue"},
 694         {Opt_err_panic, "errors=panic"},
 695         {Opt_err_ro, "errors=remount-ro"},
 696         {Opt_nouid32, "nouid32"},
 697         {Opt_nocheck, "nocheck"},
 698         {Opt_nocheck, "check=none"},
 699         {Opt_debug, "debug"},
 700         {Opt_oldalloc, "oldalloc"},
 701         {Opt_orlov, "orlov"},
 702         {Opt_user_xattr, "user_xattr"},
 703         {Opt_nouser_xattr, "nouser_xattr"},
 704         {Opt_acl, "acl"},
 705         {Opt_noacl, "noacl"},
 706         {Opt_reservation, "reservation"},
 707         {Opt_noreservation, "noreservation"},
 708         {Opt_noload, "noload"},
 709         {Opt_nobh, "nobh"},
 710         {Opt_bh, "bh"},
 711         {Opt_commit, "commit=%u"},
 712         {Opt_journal_update, "journal=update"},
 713         {Opt_journal_inum, "journal=%u"},
 714         {Opt_journal_dev, "journal_dev=%u"},
 715         {Opt_abort, "abort"},
 716         {Opt_data_journal, "data=journal"},
 717         {Opt_data_ordered, "data=ordered"},
 718         {Opt_data_writeback, "data=writeback"},
 719         {Opt_offusrjquota, "usrjquota="},
 720         {Opt_usrjquota, "usrjquota=%s"},
 721         {Opt_offgrpjquota, "grpjquota="},
 722         {Opt_grpjquota, "grpjquota=%s"},
 723         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 724         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 725         {Opt_grpquota, "grpquota"},
 726         {Opt_noquota, "noquota"},
 727         {Opt_quota, "quota"},
 728         {Opt_usrquota, "usrquota"},
 729         {Opt_barrier, "barrier=%u"},
 730         {Opt_err, NULL},
 731         {Opt_resize, "resize"},
 732 };
 733
 734 static ext3_fsblk_t get_sb_block(void **data)
 735 {
 736         ext3_fsblk_t    sb_block;
 737         char            *options = (char *) *data;
 738
 739         if (!options || strncmp(options, "sb=", 3) != 0)
 740                 return 1;       /* Default location */
 741         options += 3;
 742         /*todo: use simple_strtoll with >32bit ext3 */
 743         sb_block = simple_strtoul(options, &options, 0);
 744         if (*options && *options != ',') {
 745                 printk("EXT3-fs: Invalid sb specification: %s\n",
 746                        (char *) *data);
 747                 return 1;
 748         }
 749         if (*options == ',')
 750                 options++;
 751         *data = (void *) options;
 752         return sb_block;
 753 }
 754
 755 static int parse_options (char *options, struct super_block *sb,
 756                           unsigned int *inum, unsigned long *journal_devnum,
 757                           ext3_fsblk_t *n_blocks_count, int is_remount)
 758 {
 759         struct ext3_sb_info *sbi = EXT3_SB(sb);
 760         char * p;
 761         substring_t args[MAX_OPT_ARGS];
 762         int data_opt = 0;
 763         int option;
 764 #ifdef CONFIG_QUOTA
 765         int qtype;
 766         char *qname;
 767 #endif
 768
 769         if (!options)
 770                 return 1;
 771
 772         while ((p = strsep (&options, ",")) != NULL) {
 773                 int token;
 774                 if (!*p)
 775                         continue;
 776
 777                 token = match_token(p, tokens, args);
 778                 switch (token) {
 779                 case Opt_bsd_df:
 780                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 781                         break;
 782                 case Opt_minix_df:
 783                         set_opt (sbi->s_mount_opt, MINIX_DF);
 784                         break;
 785                 case Opt_grpid:
 786                         set_opt (sbi->s_mount_opt, GRPID);
 787                         break;
 788                 case Opt_nogrpid:
 789                         clear_opt (sbi->s_mount_opt, GRPID);
 790                         break;
 791                 case Opt_resuid:
 792                         if (match_int(&args[0], &option))
 793                                 return 0;
 794                         sbi->s_resuid = option;
 795                         break;
 796                 case Opt_resgid:
 797                         if (match_int(&args[0], &option))
 798                                 return 0;
 799                         sbi->s_resgid = option;
 800                         break;
 801                 case Opt_sb:
 802                         /* handled by get_sb_block() instead of here */
 803                         /* *sb_block = match_int(&args[0]); */
 804                         break;
 805                 case Opt_err_panic:
 806                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 807                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 808                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 809                         break;
 810                 case Opt_err_ro:
 811                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 812                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 813                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 814                         break;
 815                 case Opt_err_cont:
 816                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 817                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 818                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 819                         break;
 820                 case Opt_nouid32:
 821                         set_opt (sbi->s_mount_opt, NO_UID32);
 822                         break;
 823                 case Opt_nocheck:
 824                         clear_opt (sbi->s_mount_opt, CHECK);
 825                         break;
 826                 case Opt_debug:
 827                         set_opt (sbi->s_mount_opt, DEBUG);
 828                         break;
 829                 case Opt_oldalloc:
 830                         set_opt (sbi->s_mount_opt, OLDALLOC);
 831                         break;
 832                 case Opt_orlov:
 833                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 834                         break;
 835 #ifdef CONFIG_EXT3_FS_XATTR
 836                 case Opt_user_xattr:
 837                         set_opt (sbi->s_mount_opt, XATTR_USER);
 838                         break;
 839                 case Opt_nouser_xattr:
 840                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 841                         break;
 842 #else
 843                 case Opt_user_xattr:
 844                 case Opt_nouser_xattr:
 845                         printk("EXT3 (no)user_xattr options not supported\n");
 846                         break;
 847 #endif
 848 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 849                 case Opt_acl:
 850                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 851                         break;
 852                 case Opt_noacl:
 853                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 854                         break;
 855 #else
 856                 case Opt_acl:
 857                 case Opt_noacl:
 858                         printk("EXT3 (no)acl options not supported\n");
 859                         break;
 860 #endif
 861                 case Opt_reservation:
 862                         set_opt(sbi->s_mount_opt, RESERVATION);
 863                         break;
 864                 case Opt_noreservation:
 865                         clear_opt(sbi->s_mount_opt, RESERVATION);
 866                         break;
 867                 case Opt_journal_update:
 868                         /* @@@ FIXME */
 869                         /* Eventually we will want to be able to create
 870                            a journal file here.  For now, only allow the
 871                            user to specify an existing inode to be the
 872                            journal file. */
 873                         if (is_remount) {
 874                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 875                                        "journal on remount\n");
 876                                 return 0;
 877                         }
 878                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 879                         break;
 880                 case Opt_journal_inum:
 881                         if (is_remount) {
 882                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 883                                        "journal on remount\n");
 884                                 return 0;
 885                         }
 886                         if (match_int(&args[0], &option))
 887                                 return 0;
 888                         *inum = option;
 889                         break;
 890                 case Opt_journal_dev:
 891                         if (is_remount) {
 892                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 893                                        "journal on remount\n");
 894                                 return 0;
 895                         }
 896                         if (match_int(&args[0], &option))
 897                                 return 0;
 898                         *journal_devnum = option;
 899                         break;
 900                 case Opt_noload:
 901                         set_opt (sbi->s_mount_opt, NOLOAD);
 902                         break;
 903                 case Opt_commit:
 904                         if (match_int(&args[0], &option))
 905                                 return 0;
 906                         if (option < 0)
 907                                 return 0;
 908                         if (option == 0)
 909                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 910                         sbi->s_commit_interval = HZ * option;
 911                         break;
 912                 case Opt_data_journal:
 913                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 914                         goto datacheck;
 915                 case Opt_data_ordered:
 916                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 917                         goto datacheck;
 918                 case Opt_data_writeback:
 919                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 920                 datacheck:
 921                         if (is_remount) {
 922                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 923                                                 != data_opt) {
 924                                         printk(KERN_ERR
 925                                                 "EXT3-fs: cannot change data "
 926                                                 "mode on remount\n");
 927                                         return 0;
 928                                 }
 929                         } else {
 930                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 931                                 sbi->s_mount_opt |= data_opt;
 932                         }
 933                         break;
 934 #ifdef CONFIG_QUOTA
 935                 case Opt_usrjquota:
 936                         qtype = USRQUOTA;
 937                         goto set_qf_name;
 938                 case Opt_grpjquota:
 939                         qtype = GRPQUOTA;
 940 set_qf_name:
 941                         if (sb_any_quota_enabled(sb)) {
 942                                 printk(KERN_ERR
 943                                         "EXT3-fs: Cannot change journalled "
 944                                         "quota options when quota turned on.\n");
 945                                 return 0;
 946                         }
 947                         qname = match_strdup(&args[0]);
 948                         if (!qname) {
 949                                 printk(KERN_ERR
 950                                         "EXT3-fs: not enough memory for "
 951                                         "storing quotafile name.\n");
 952                                 return 0;
 953                         }
 954                         if (sbi->s_qf_names[qtype] &&
 955                             strcmp(sbi->s_qf_names[qtype], qname)) {
 956                                 printk(KERN_ERR
 957                                         "EXT3-fs: %s quota file already "
 958                                         "specified.\n", QTYPE2NAME(qtype));
 959                                 kfree(qname);
 960                                 return 0;
 961                         }
 962                         sbi->s_qf_names[qtype] = qname;
 963                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 964                                 printk(KERN_ERR
 965                                         "EXT3-fs: quotafile must be on "
 966                                         "filesystem root.\n");
 967                                 kfree(sbi->s_qf_names[qtype]);
 968                                 sbi->s_qf_names[qtype] = NULL;
 969                                 return 0;
 970                         }
 971                         set_opt(sbi->s_mount_opt, QUOTA);
 972                         break;
 973                 case Opt_offusrjquota:
 974                         qtype = USRQUOTA;
 975                         goto clear_qf_name;
 976                 case Opt_offgrpjquota:
 977                         qtype = GRPQUOTA;
 978 clear_qf_name:
 979                         if (sb_any_quota_enabled(sb)) {
 980                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 981                                         "journalled quota options when "
 982                                         "quota turned on.\n");
 983                                 return 0;
 984                         }
 985                         /*
 986                          * The space will be released later when all options
 987                          * are confirmed to be correct
 988                          */
 989                         sbi->s_qf_names[qtype] = NULL;
 990                         break;
 991                 case Opt_jqfmt_vfsold:
 992                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 993                         break;
 994                 case Opt_jqfmt_vfsv0:
 995                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 996                         break;
 997                 case Opt_quota:
 998                 case Opt_usrquota:
 999                         set_opt(sbi->s_mount_opt, QUOTA);
1000                         set_opt(sbi->s_mount_opt, USRQUOTA);
1001                         break;
1002                 case Opt_grpquota:
1003                         set_opt(sbi->s_mount_opt, QUOTA);
1004                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1005                         break;
1006                 case Opt_noquota:
1007                         if (sb_any_quota_enabled(sb)) {
1008                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1009                                         "options when quota turned on.\n");
1010                                 return 0;
1011                         }
1012                         clear_opt(sbi->s_mount_opt, QUOTA);
1013                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1014                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1015                         break;
1016 #else
1017                 case Opt_quota:
1018                 case Opt_usrquota:
1019                 case Opt_grpquota:
1020                 case Opt_usrjquota:
1021                 case Opt_grpjquota:
1022                 case Opt_offusrjquota:
1023                 case Opt_offgrpjquota:
1024                 case Opt_jqfmt_vfsold:
1025                 case Opt_jqfmt_vfsv0:
1026                         printk(KERN_ERR
1027                                 "EXT3-fs: journalled quota options not "
1028                                 "supported.\n");
1029                         break;
1030                 case Opt_noquota:
1031                         break;
1032 #endif
1033                 case Opt_abort:
1034                         set_opt(sbi->s_mount_opt, ABORT);
1035                         break;
1036                 case Opt_barrier:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         if (option)
1040                                 set_opt(sbi->s_mount_opt, BARRIER);
1041                         else
1042                                 clear_opt(sbi->s_mount_opt, BARRIER);
1043                         break;
1044                 case Opt_ignore:
1045                         break;
1046                 case Opt_resize:
1047                         if (!is_remount) {
1048                                 printk("EXT3-fs: resize option only available "
1049                                         "for remount\n");
1050                                 return 0;
1051                         }
1052                         if (match_int(&args[0], &option) != 0)
1053                                 return 0;
1054                         *n_blocks_count = option;
1055                         break;
1056                 case Opt_nobh:
1057                         set_opt(sbi->s_mount_opt, NOBH);
1058                         break;
1059                 case Opt_bh:
1060                         clear_opt(sbi->s_mount_opt, NOBH);
1061                         break;
1062                 default:
1063                         printk (KERN_ERR
1064                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1065                                 "or missing value\n", p);
1066                         return 0;
1067                 }
1068         }
1069 #ifdef CONFIG_QUOTA
1070         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1071                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1072                      sbi->s_qf_names[USRQUOTA])
1073                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1074
1075                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1076                      sbi->s_qf_names[GRPQUOTA])
1077                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1078
1079                 if ((sbi->s_qf_names[USRQUOTA] &&
1080                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1081                     (sbi->s_qf_names[GRPQUOTA] &&
1082                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1083                         printk(KERN_ERR "EXT3-fs: old and new quota "
1084                                         "format mixing.\n");
1085                         return 0;
1086                 }
1087
1088                 if (!sbi->s_jquota_fmt) {
1089                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1090                                         "not specified.\n");
1091                         return 0;
1092                 }
1093         } else {
1094                 if (sbi->s_jquota_fmt) {
1095                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1096                                         "specified with no journalling "
1097                                         "enabled.\n");
1098                         return 0;
1099                 }
1100         }
1101 #endif
1102         return 1;
1103 }
1104
1105 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1106                             int read_only)
1107 {
1108         struct ext3_sb_info *sbi = EXT3_SB(sb);
1109         int res = 0;
1110
1111         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1112                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1113                         "forcing read-only mode\n");
1114                 res = MS_RDONLY;
1115         }
1116         if (read_only)
1117                 return res;
1118         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1119                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1120                         "running e2fsck is recommended\n");
1121         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1122                 printk (KERN_WARNING
1123                         "EXT3-fs warning: mounting fs with errors, "
1124                         "running e2fsck is recommended\n");
1125         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1126                  le16_to_cpu(es->s_mnt_count) >=
1127                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1128                 printk (KERN_WARNING
1129                         "EXT3-fs warning: maximal mount count reached, "
1130                         "running e2fsck is recommended\n");
1131         else if (le32_to_cpu(es->s_checkinterval) &&
1132                 (le32_to_cpu(es->s_lastcheck) +
1133                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1134                 printk (KERN_WARNING
1135                         "EXT3-fs warning: checktime reached, "
1136                         "running e2fsck is recommended\n");
1137 #if 0
1138                 /* @@@ We _will_ want to clear the valid bit if we find
1139                    inconsistencies, to force a fsck at reboot.  But for
1140                    a plain journaled filesystem we can keep it set as
1141                    valid forever! :) */
1142         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1143 #endif
1144         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1145                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1146         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1147         es->s_mtime = cpu_to_le32(get_seconds());
1148         ext3_update_dynamic_rev(sb);
1149         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1150
1151         ext3_commit_super(sb, es, 1);
1152         if (test_opt(sb, DEBUG))
1153                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1154                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1155                         sb->s_blocksize,
1156                         sbi->s_groups_count,
1157                         EXT3_BLOCKS_PER_GROUP(sb),
1158                         EXT3_INODES_PER_GROUP(sb),
1159                         sbi->s_mount_opt);
1160
1161         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1162         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1163                 char b[BDEVNAME_SIZE];
1164
1165                 printk("external journal on %s\n",
1166                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1167         } else {
1168                 printk("internal journal\n");
1169         }
1170         return res;
1171 }
1172
1173 /* Called at mount-time, super-block is locked */
1174 static int ext3_check_descriptors (struct super_block * sb)
1175 {
1176         struct ext3_sb_info *sbi = EXT3_SB(sb);
1177         ext3_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1178         ext3_fsblk_t last_block;
1179         struct ext3_group_desc * gdp = NULL;
1180         int desc_block = 0;
1181         int i;
1182
1183         ext3_debug ("Checking group descriptors");
1184
1185         for (i = 0; i < sbi->s_groups_count; i++)
1186         {
1187                 if (i == sbi->s_groups_count - 1)
1188                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1189                 else
1190                         last_block = first_block +
1191                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1192
1193                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1194                         gdp = (struct ext3_group_desc *)
1195                                         sbi->s_group_desc[desc_block++]->b_data;
1196                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1197                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1198                 {
1199                         ext3_error (sb, "ext3_check_descriptors",
1200                                     "Block bitmap for group %d"
1201                                     " not in group (block %lu)!",
1202                                     i, (unsigned long)
1203                                         le32_to_cpu(gdp->bg_block_bitmap));
1204                         return 0;
1205                 }
1206                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1207                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1208                 {
1209                         ext3_error (sb, "ext3_check_descriptors",
1210                                     "Inode bitmap for group %d"
1211                                     " not in group (block %lu)!",
1212                                     i, (unsigned long)
1213                                         le32_to_cpu(gdp->bg_inode_bitmap));
1214                         return 0;
1215                 }
1216                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1217                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >
1218                     last_block)
1219                 {
1220                         ext3_error (sb, "ext3_check_descriptors",
1221                                     "Inode table for group %d"
1222                                     " not in group (block %lu)!",
1223                                     i, (unsigned long)
1224                                         le32_to_cpu(gdp->bg_inode_table));
1225                         return 0;
1226                 }
1227                 first_block += EXT3_BLOCKS_PER_GROUP(sb);
1228                 gdp++;
1229         }
1230
1231         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1232         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1233         return 1;
1234 }
1235
1236
1237 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1238  * the superblock) which were deleted from all directories, but held open by
1239  * a process at the time of a crash.  We walk the list and try to delete these
1240  * inodes at recovery time (only with a read-write filesystem).
1241  *
1242  * In order to keep the orphan inode chain consistent during traversal (in
1243  * case of crash during recovery), we link each inode into the superblock
1244  * orphan list_head and handle it the same way as an inode deletion during
1245  * normal operation (which journals the operations for us).
1246  *
1247  * We only do an iget() and an iput() on each inode, which is very safe if we
1248  * accidentally point at an in-use or already deleted inode.  The worst that
1249  * can happen in this case is that we get a "bit already cleared" message from
1250  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1251  * e2fsck was run on this filesystem, and it must have already done the orphan
1252  * inode cleanup for us, so we can safely abort without any further action.
1253  */
1254 static void ext3_orphan_cleanup (struct super_block * sb,
1255                                  struct ext3_super_block * es)
1256 {
1257         unsigned int s_flags = sb->s_flags;
1258         int nr_orphans = 0, nr_truncates = 0;
1259 #ifdef CONFIG_QUOTA
1260         int i;
1261 #endif
1262         if (!es->s_last_orphan) {
1263                 jbd_debug(4, "no orphan inodes to clean up\n");
1264                 return;
1265         }
1266
1267         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1268                 if (es->s_last_orphan)
1269                         jbd_debug(1, "Errors on filesystem, "
1270                                   "clearing orphan list.\n");
1271                 es->s_last_orphan = 0;
1272                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1273                 return;
1274         }
1275
1276         if (s_flags & MS_RDONLY) {
1277                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1278                        sb->s_id);
1279                 sb->s_flags &= ~MS_RDONLY;
1280         }
1281 #ifdef CONFIG_QUOTA
1282         /* Needed for iput() to work correctly and not trash data */
1283         sb->s_flags |= MS_ACTIVE;
1284         /* Turn on quotas so that they are updated correctly */
1285         for (i = 0; i < MAXQUOTAS; i++) {
1286                 if (EXT3_SB(sb)->s_qf_names[i]) {
1287                         int ret = ext3_quota_on_mount(sb, i);
1288                         if (ret < 0)
1289                                 printk(KERN_ERR
1290                                         "EXT3-fs: Cannot turn on journalled "
1291                                         "quota: error %d\n", ret);
1292                 }
1293         }
1294 #endif
1295
1296         while (es->s_last_orphan) {
1297                 struct inode *inode;
1298
1299                 if (!(inode =
1300                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1301                         es->s_last_orphan = 0;
1302                         break;
1303                 }
1304
1305                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1306                 DQUOT_INIT(inode);
1307                 if (inode->i_nlink) {
1308                         printk(KERN_DEBUG
1309                                 "%s: truncating inode %lu to %Ld bytes\n",
1310                                 __FUNCTION__, inode->i_ino, inode->i_size);
1311                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1312                                   inode->i_ino, inode->i_size);
1313                         ext3_truncate(inode);
1314                         nr_truncates++;
1315                 } else {
1316                         printk(KERN_DEBUG
1317                                 "%s: deleting unreferenced inode %lu\n",
1318                                 __FUNCTION__, inode->i_ino);
1319                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1320                                   inode->i_ino);
1321                         nr_orphans++;
1322                 }
1323                 iput(inode);  /* The delete magic happens here! */
1324         }
1325
1326 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1327
1328         if (nr_orphans)
1329                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1330                        sb->s_id, PLURAL(nr_orphans));
1331         if (nr_truncates)
1332                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1333                        sb->s_id, PLURAL(nr_truncates));
1334 #ifdef CONFIG_QUOTA
1335         /* Turn quotas off */
1336         for (i = 0; i < MAXQUOTAS; i++) {
1337                 if (sb_dqopt(sb)->files[i])
1338                         vfs_quota_off(sb, i);
1339         }
1340 #endif
1341         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1342 }
1343
1344 #define log2(n) ffz(~(n))
1345
1346 /*
1347  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1348  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1349  * We need to be 1 filesystem block less than the 2^32 sector limit.
1350  */
1351 static loff_t ext3_max_size(int bits)
1352 {
1353         loff_t res = EXT3_NDIR_BLOCKS;
1354         /* This constant is calculated to be the largest file size for a
1355          * dense, 4k-blocksize file such that the total number of
1356          * sectors in the file, including data and all indirect blocks,
1357          * does not exceed 2^32. */
1358         const loff_t upper_limit = 0x1ff7fffd000LL;
1359
1360         res += 1LL << (bits-2);
1361         res += 1LL << (2*(bits-2));
1362         res += 1LL << (3*(bits-2));
1363         res <<= bits;
1364         if (res > upper_limit)
1365                 res = upper_limit;
1366         return res;
1367 }
1368
1369 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1370                                     ext3_fsblk_t logic_sb_block,
1371                                     int nr)
1372 {
1373         struct ext3_sb_info *sbi = EXT3_SB(sb);
1374         unsigned long bg, first_meta_bg;
1375         int has_super = 0;
1376
1377         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1378
1379         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1380             nr < first_meta_bg)
1381                 return (logic_sb_block + nr + 1);
1382         bg = sbi->s_desc_per_block * nr;
1383         if (ext3_bg_has_super(sb, bg))
1384                 has_super = 1;
1385         return (has_super + ext3_group_first_block_no(sb, bg));
1386 }
1387
1388
1389 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1390 {
1391         struct buffer_head * bh;
1392         struct ext3_super_block *es = NULL;
1393         struct ext3_sb_info *sbi;
1394         ext3_fsblk_t block;
1395         ext3_fsblk_t sb_block = get_sb_block(&data);
1396         ext3_fsblk_t logic_sb_block;
1397         unsigned long offset = 0;
1398         unsigned int journal_inum = 0;
1399         unsigned long journal_devnum = 0;
1400         unsigned long def_mount_opts;
1401         struct inode *root;
1402         int blocksize;
1403         int hblock;
1404         int db_count;
1405         int i;
1406         int needs_recovery;
1407         __le32 features;
1408
1409         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1410         if (!sbi)
1411                 return -ENOMEM;
1412         sb->s_fs_info = sbi;
1413         memset(sbi, 0, sizeof(*sbi));
1414         sbi->s_mount_opt = 0;
1415         sbi->s_resuid = EXT3_DEF_RESUID;
1416         sbi->s_resgid = EXT3_DEF_RESGID;
1417
1418         unlock_kernel();
1419
1420         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1421         if (!blocksize) {
1422                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1423                 goto out_fail;
1424         }
1425
1426         /*
1427          * The ext3 superblock will not be buffer aligned for other than 1kB
1428          * block sizes.  We need to calculate the offset from buffer start.
1429          */
1430         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1431                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1432                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1433         } else {
1434                 logic_sb_block = sb_block;
1435         }
1436
1437         if (!(bh = sb_bread(sb, logic_sb_block))) {
1438                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1439                 goto out_fail;
1440         }
1441         /*
1442          * Note: s_es must be initialized as soon as possible because
1443          *       some ext3 macro-instructions depend on its value
1444          */
1445         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1446         sbi->s_es = es;
1447         sb->s_magic = le16_to_cpu(es->s_magic);
1448         if (sb->s_magic != EXT3_SUPER_MAGIC)
1449                 goto cantfind_ext3;
1450
1451         /* Set defaults before we parse the mount options */
1452         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1453         if (def_mount_opts & EXT3_DEFM_DEBUG)
1454                 set_opt(sbi->s_mount_opt, DEBUG);
1455         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1456                 set_opt(sbi->s_mount_opt, GRPID);
1457         if (def_mount_opts & EXT3_DEFM_UID16)
1458                 set_opt(sbi->s_mount_opt, NO_UID32);
1459         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1460                 set_opt(sbi->s_mount_opt, XATTR_USER);
1461         if (def_mount_opts & EXT3_DEFM_ACL)
1462                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1463         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1464                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1465         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1466                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1467         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1468                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1469
1470         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1471                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1472         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1473                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1474
1475         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1476         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1477
1478         set_opt(sbi->s_mount_opt, RESERVATION);
1479
1480         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1481                             NULL, 0))
1482                 goto failed_mount;
1483
1484         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1485                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1486
1487         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1488             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1489              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1490              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1491                 printk(KERN_WARNING
1492                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1493                        "running e2fsck is recommended\n");
1494         /*
1495          * Check feature flags regardless of the revision level, since we
1496          * previously didn't change the revision level when setting the flags,
1497          * so there is a chance incompat flags are set on a rev 0 filesystem.
1498          */
1499         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1500         if (features) {
1501                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1502                        "unsupported optional features (%x).\n",
1503                        sb->s_id, le32_to_cpu(features));
1504                 goto failed_mount;
1505         }
1506         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1507         if (!(sb->s_flags & MS_RDONLY) && features) {
1508                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1509                        "unsupported optional features (%x).\n",
1510                        sb->s_id, le32_to_cpu(features));
1511                 goto failed_mount;
1512         }
1513         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1514
1515         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1516             blocksize > EXT3_MAX_BLOCK_SIZE) {
1517                 printk(KERN_ERR
1518                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1519                        blocksize, sb->s_id);
1520                 goto failed_mount;
1521         }
1522
1523         hblock = bdev_hardsect_size(sb->s_bdev);
1524         if (sb->s_blocksize != blocksize) {
1525                 /*
1526                  * Make sure the blocksize for the filesystem is larger
1527                  * than the hardware sectorsize for the machine.
1528                  */
1529                 if (blocksize < hblock) {
1530                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1531                                "device blocksize %d.\n", blocksize, hblock);
1532                         goto failed_mount;
1533                 }
1534
1535                 brelse (bh);
1536                 sb_set_blocksize(sb, blocksize);
1537                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1538                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1539                 bh = sb_bread(sb, logic_sb_block);
1540                 if (!bh) {
1541                         printk(KERN_ERR
1542                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1543                         goto failed_mount;
1544                 }
1545                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1546                 sbi->s_es = es;
1547                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1548                         printk (KERN_ERR
1549                                 "EXT3-fs: Magic mismatch, very weird !\n");
1550                         goto failed_mount;
1551                 }
1552         }
1553
1554         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1555
1556         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1557                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1558                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1559         } else {
1560                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1561                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1562                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1563                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1564                     (sbi->s_inode_size > blocksize)) {
1565                         printk (KERN_ERR
1566                                 "EXT3-fs: unsupported inode size: %d\n",
1567                                 sbi->s_inode_size);
1568                         goto failed_mount;
1569                 }
1570         }
1571         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1572                                    le32_to_cpu(es->s_log_frag_size);
1573         if (blocksize != sbi->s_frag_size) {
1574                 printk(KERN_ERR
1575                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1576                        sbi->s_frag_size, blocksize);
1577                 goto failed_mount;
1578         }
1579         sbi->s_frags_per_block = 1;
1580         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1581         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1582         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1583         if (EXT3_INODE_SIZE(sb) == 0)
1584                 goto cantfind_ext3;
1585         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1586         if (sbi->s_inodes_per_block == 0)
1587                 goto cantfind_ext3;
1588         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1589                                         sbi->s_inodes_per_block;
1590         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1591         sbi->s_sbh = bh;
1592         sbi->s_mount_state = le16_to_cpu(es->s_state);
1593         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1594         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1595         for (i=0; i < 4; i++)
1596                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1597         sbi->s_def_hash_version = es->s_def_hash_version;
1598
1599         if (sbi->s_blocks_per_group > blocksize * 8) {
1600                 printk (KERN_ERR
1601                         "EXT3-fs: #blocks per group too big: %lu\n",
1602                         sbi->s_blocks_per_group);
1603                 goto failed_mount;
1604         }
1605         if (sbi->s_frags_per_group > blocksize * 8) {
1606                 printk (KERN_ERR
1607                         "EXT3-fs: #fragments per group too big: %lu\n",
1608                         sbi->s_frags_per_group);
1609                 goto failed_mount;
1610         }
1611         if (sbi->s_inodes_per_group > blocksize * 8) {
1612                 printk (KERN_ERR
1613                         "EXT3-fs: #inodes per group too big: %lu\n",
1614                         sbi->s_inodes_per_group);
1615                 goto failed_mount;
1616         }
1617
1618         if (le32_to_cpu(es->s_blocks_count) >
1619                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1620                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1621                         " too large to mount safely\n", sb->s_id);
1622                 if (sizeof(sector_t) < 8)
1623                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1624                                         "enabled\n");
1625                 goto failed_mount;
1626         }
1627
1628         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1629                 goto cantfind_ext3;
1630         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1631                                le32_to_cpu(es->s_first_data_block) - 1)
1632                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1633         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1634                    EXT3_DESC_PER_BLOCK(sb);
1635         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1636                                     GFP_KERNEL);
1637         if (sbi->s_group_desc == NULL) {
1638                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1639                 goto failed_mount;
1640         }
1641
1642         bgl_lock_init(&sbi->s_blockgroup_lock);
1643
1644         for (i = 0; i < db_count; i++) {
1645                 block = descriptor_loc(sb, logic_sb_block, i);
1646                 sbi->s_group_desc[i] = sb_bread(sb, block);
1647                 if (!sbi->s_group_desc[i]) {
1648                         printk (KERN_ERR "EXT3-fs: "
1649                                 "can't read group descriptor %d\n", i);
1650                         db_count = i;
1651                         goto failed_mount2;
1652                 }
1653         }
1654         if (!ext3_check_descriptors (sb)) {
1655                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1656                 goto failed_mount2;
1657         }
1658         sbi->s_gdb_count = db_count;
1659         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1660         spin_lock_init(&sbi->s_next_gen_lock);
1661
1662         percpu_counter_init(&sbi->s_freeblocks_counter,
1663                 ext3_count_free_blocks(sb));
1664         percpu_counter_init(&sbi->s_freeinodes_counter,
1665                 ext3_count_free_inodes(sb));
1666         percpu_counter_init(&sbi->s_dirs_counter,
1667                 ext3_count_dirs(sb));
1668
1669         /* per fileystem reservation list head & lock */
1670         spin_lock_init(&sbi->s_rsv_window_lock);
1671         sbi->s_rsv_window_root = RB_ROOT;
1672         /* Add a single, static dummy reservation to the start of the
1673          * reservation window list --- it gives us a placeholder for
1674          * append-at-start-of-list which makes the allocation logic
1675          * _much_ simpler. */
1676         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1677         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1678         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1679         sbi->s_rsv_window_head.rsv_goal_size = 0;
1680         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1681
1682         /*
1683          * set up enough so that it can read an inode
1684          */
1685         sb->s_op = &ext3_sops;
1686         sb->s_export_op = &ext3_export_ops;
1687         sb->s_xattr = ext3_xattr_handlers;
1688 #ifdef CONFIG_QUOTA
1689         sb->s_qcop = &ext3_qctl_operations;
1690         sb->dq_op = &ext3_quota_operations;
1691 #endif
1692         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1693
1694         sb->s_root = NULL;
1695
1696         needs_recovery = (es->s_last_orphan != 0 ||
1697                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1698                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1699
1700         /*
1701          * The first inode we look at is the journal inode.  Don't try
1702          * root first: it may be modified in the journal!
1703          */
1704         if (!test_opt(sb, NOLOAD) &&
1705             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1706                 if (ext3_load_journal(sb, es, journal_devnum))
1707                         goto failed_mount3;
1708         } else if (journal_inum) {
1709                 if (ext3_create_journal(sb, es, journal_inum))
1710                         goto failed_mount3;
1711         } else {
1712                 if (!silent)
1713                         printk (KERN_ERR
1714                                 "ext3: No journal on filesystem on %s\n",
1715                                 sb->s_id);
1716                 goto failed_mount3;
1717         }
1718
1719         /* We have now updated the journal if required, so we can
1720          * validate the data journaling mode. */
1721         switch (test_opt(sb, DATA_FLAGS)) {
1722         case 0:
1723                 /* No mode set, assume a default based on the journal
1724                    capabilities: ORDERED_DATA if the journal can
1725                    cope, else JOURNAL_DATA */
1726                 if (journal_check_available_features
1727                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1728                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1729                 else
1730                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1731                 break;
1732
1733         case EXT3_MOUNT_ORDERED_DATA:
1734         case EXT3_MOUNT_WRITEBACK_DATA:
1735                 if (!journal_check_available_features
1736                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1737                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1738                                "requested data journaling mode\n");
1739                         goto failed_mount4;
1740                 }
1741         default:
1742                 break;
1743         }
1744
1745         if (test_opt(sb, NOBH)) {
1746                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1747                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1748                                 "its supported only with writeback mode\n");
1749                         clear_opt(sbi->s_mount_opt, NOBH);
1750                 }
1751         }
1752         /*
1753          * The journal_load will have done any necessary log recovery,
1754          * so we can safely mount the rest of the filesystem now.
1755          */
1756
1757         root = iget(sb, EXT3_ROOT_INO);
1758         sb->s_root = d_alloc_root(root);
1759         if (!sb->s_root) {
1760                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1761                 iput(root);
1762                 goto failed_mount4;
1763         }
1764         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1765                 dput(sb->s_root);
1766                 sb->s_root = NULL;
1767                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1768                 goto failed_mount4;
1769         }
1770
1771         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1772         /*
1773          * akpm: core read_super() calls in here with the superblock locked.
1774          * That deadlocks, because orphan cleanup needs to lock the superblock
1775          * in numerous places.  Here we just pop the lock - it's relatively
1776          * harmless, because we are now ready to accept write_super() requests,
1777          * and aviro says that's the only reason for hanging onto the
1778          * superblock lock.
1779          */
1780         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1781         ext3_orphan_cleanup(sb, es);
1782         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1783         if (needs_recovery)
1784                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1785         ext3_mark_recovery_complete(sb, es);
1786         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1787                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1788                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1789                 "writeback");
1790
1791         lock_kernel();
1792         return 0;
1793
1794 cantfind_ext3:
1795         if (!silent)
1796                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1797                        sb->s_id);
1798         goto failed_mount;
1799
1800 failed_mount4:
1801         journal_destroy(sbi->s_journal);
1802 failed_mount3:
1803         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1804         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1805         percpu_counter_destroy(&sbi->s_dirs_counter);
1806 failed_mount2:
1807         for (i = 0; i < db_count; i++)
1808                 brelse(sbi->s_group_desc[i]);
1809         kfree(sbi->s_group_desc);
1810 failed_mount:
1811 #ifdef CONFIG_QUOTA
1812         for (i = 0; i < MAXQUOTAS; i++)
1813                 kfree(sbi->s_qf_names[i]);
1814 #endif
1815         ext3_blkdev_remove(sbi);
1816         brelse(bh);
1817 out_fail:
1818         sb->s_fs_info = NULL;
1819         kfree(sbi);
1820         lock_kernel();
1821         return -EINVAL;
1822 }
1823
1824 /*
1825  * Setup any per-fs journal parameters now.  We'll do this both on
1826  * initial mount, once the journal has been initialised but before we've
1827  * done any recovery; and again on any subsequent remount.
1828  */
1829 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1830 {
1831         struct ext3_sb_info *sbi = EXT3_SB(sb);
1832
1833         if (sbi->s_commit_interval)
1834                 journal->j_commit_interval = sbi->s_commit_interval;
1835         /* We could also set up an ext3-specific default for the commit
1836          * interval here, but for now we'll just fall back to the jbd
1837          * default. */
1838
1839         spin_lock(&journal->j_state_lock);
1840         if (test_opt(sb, BARRIER))
1841                 journal->j_flags |= JFS_BARRIER;
1842         else
1843                 journal->j_flags &= ~JFS_BARRIER;
1844         spin_unlock(&journal->j_state_lock);
1845 }
1846
1847 static journal_t *ext3_get_journal(struct super_block *sb,
1848                                    unsigned int journal_inum)
1849 {
1850         struct inode *journal_inode;
1851         journal_t *journal;
1852
1853         /* First, test for the existence of a valid inode on disk.  Bad
1854          * things happen if we iget() an unused inode, as the subsequent
1855          * iput() will try to delete it. */
1856
1857         journal_inode = iget(sb, journal_inum);
1858         if (!journal_inode) {
1859                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1860                 return NULL;
1861         }
1862         if (!journal_inode->i_nlink) {
1863                 make_bad_inode(journal_inode);
1864                 iput(journal_inode);
1865                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1866                 return NULL;
1867         }
1868
1869         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1870                   journal_inode, journal_inode->i_size);
1871         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1872                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1873                 iput(journal_inode);
1874                 return NULL;
1875         }
1876
1877         journal = journal_init_inode(journal_inode);
1878         if (!journal) {
1879                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1880                 iput(journal_inode);
1881                 return NULL;
1882         }
1883         journal->j_private = sb;
1884         ext3_init_journal_params(sb, journal);
1885         return journal;
1886 }
1887
1888 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1889                                        dev_t j_dev)
1890 {
1891         struct buffer_head * bh;
1892         journal_t *journal;
1893         ext3_fsblk_t start;
1894         ext3_fsblk_t len;
1895         int hblock, blocksize;
1896         ext3_fsblk_t sb_block;
1897         unsigned long offset;
1898         struct ext3_super_block * es;
1899         struct block_device *bdev;
1900
1901         bdev = ext3_blkdev_get(j_dev);
1902         if (bdev == NULL)
1903                 return NULL;
1904
1905         if (bd_claim(bdev, sb)) {
1906                 printk(KERN_ERR
1907                         "EXT3: failed to claim external journal device.\n");
1908                 blkdev_put(bdev);
1909                 return NULL;
1910         }
1911
1912         blocksize = sb->s_blocksize;
1913         hblock = bdev_hardsect_size(bdev);
1914         if (blocksize < hblock) {
1915                 printk(KERN_ERR
1916                         "EXT3-fs: blocksize too small for journal device.\n");
1917                 goto out_bdev;
1918         }
1919
1920         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1921         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1922         set_blocksize(bdev, blocksize);
1923         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1924                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1925                        "external journal\n");
1926                 goto out_bdev;
1927         }
1928
1929         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1930         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1931             !(le32_to_cpu(es->s_feature_incompat) &
1932               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1933                 printk(KERN_ERR "EXT3-fs: external journal has "
1934                                         "bad superblock\n");
1935                 brelse(bh);
1936                 goto out_bdev;
1937         }
1938
1939         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1940                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1941                 brelse(bh);
1942                 goto out_bdev;
1943         }
1944
1945         len = le32_to_cpu(es->s_blocks_count);
1946         start = sb_block + 1;
1947         brelse(bh);     /* we're done with the superblock */
1948
1949         journal = journal_init_dev(bdev, sb->s_bdev,
1950                                         start, len, blocksize);
1951         if (!journal) {
1952                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1953                 goto out_bdev;
1954         }
1955         journal->j_private = sb;
1956         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1957         wait_on_buffer(journal->j_sb_buffer);
1958         if (!buffer_uptodate(journal->j_sb_buffer)) {
1959                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1960                 goto out_journal;
1961         }
1962         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1963                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1964                                         "user (unsupported) - %d\n",
1965                         be32_to_cpu(journal->j_superblock->s_nr_users));
1966                 goto out_journal;
1967         }
1968         EXT3_SB(sb)->journal_bdev = bdev;
1969         ext3_init_journal_params(sb, journal);
1970         return journal;
1971 out_journal:
1972         journal_destroy(journal);
1973 out_bdev:
1974         ext3_blkdev_put(bdev);
1975         return NULL;
1976 }
1977
1978 static int ext3_load_journal(struct super_block *sb,
1979                              struct ext3_super_block *es,
1980                              unsigned long journal_devnum)
1981 {
1982         journal_t *journal;
1983         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
1984         dev_t journal_dev;
1985         int err = 0;
1986         int really_read_only;
1987
1988         if (journal_devnum &&
1989             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1990                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1991                         "numbers have changed\n");
1992                 journal_dev = new_decode_dev(journal_devnum);
1993         } else
1994                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1995
1996         really_read_only = bdev_read_only(sb->s_bdev);
1997
1998         /*
1999          * Are we loading a blank journal or performing recovery after a
2000          * crash?  For recovery, we need to check in advance whether we
2001          * can get read-write access to the device.
2002          */
2003
2004         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2005                 if (sb->s_flags & MS_RDONLY) {
2006                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2007                                         "required on readonly filesystem.\n");
2008                         if (really_read_only) {
2009                                 printk(KERN_ERR "EXT3-fs: write access "
2010                                         "unavailable, cannot proceed.\n");
2011                                 return -EROFS;
2012                         }
2013                         printk (KERN_INFO "EXT3-fs: write access will "
2014                                         "be enabled during recovery.\n");
2015                 }
2016         }
2017
2018         if (journal_inum && journal_dev) {
2019                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2020                        "and inode journals!\n");
2021                 return -EINVAL;
2022         }
2023
2024         if (journal_inum) {
2025                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2026                         return -EINVAL;
2027         } else {
2028                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2029                         return -EINVAL;
2030         }
2031
2032         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2033                 err = journal_update_format(journal);
2034                 if (err)  {
2035                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2036                         journal_destroy(journal);
2037                         return err;
2038                 }
2039         }
2040
2041         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2042                 err = journal_wipe(journal, !really_read_only);
2043         if (!err)
2044                 err = journal_load(journal);
2045
2046         if (err) {
2047                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2048                 journal_destroy(journal);
2049                 return err;
2050         }
2051
2052         EXT3_SB(sb)->s_journal = journal;
2053         ext3_clear_journal_err(sb, es);
2054
2055         if (journal_devnum &&
2056             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2057                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2058                 sb->s_dirt = 1;
2059
2060                 /* Make sure we flush the recovery flag to disk. */
2061                 ext3_commit_super(sb, es, 1);
2062         }
2063
2064         return 0;
2065 }
2066
2067 static int ext3_create_journal(struct super_block * sb,
2068                                struct ext3_super_block * es,
2069                                unsigned int journal_inum)
2070 {
2071         journal_t *journal;
2072
2073         if (sb->s_flags & MS_RDONLY) {
2074                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2075                                 "create journal.\n");
2076                 return -EROFS;
2077         }
2078
2079         if (!(journal = ext3_get_journal(sb, journal_inum)))
2080                 return -EINVAL;
2081
2082         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2083                journal_inum);
2084
2085         if (journal_create(journal)) {
2086                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2087                 journal_destroy(journal);
2088                 return -EIO;
2089         }
2090
2091         EXT3_SB(sb)->s_journal = journal;
2092
2093         ext3_update_dynamic_rev(sb);
2094         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2095         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2096
2097         es->s_journal_inum = cpu_to_le32(journal_inum);
2098         sb->s_dirt = 1;
2099
2100         /* Make sure we flush the recovery flag to disk. */
2101         ext3_commit_super(sb, es, 1);
2102
2103         return 0;
2104 }
2105
2106 static void ext3_commit_super (struct super_block * sb,
2107                                struct ext3_super_block * es,
2108                                int sync)
2109 {
2110         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2111
2112         if (!sbh)
2113                 return;
2114         es->s_wtime = cpu_to_le32(get_seconds());
2115         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2116         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2117         BUFFER_TRACE(sbh, "marking dirty");
2118         mark_buffer_dirty(sbh);
2119         if (sync)
2120                 sync_dirty_buffer(sbh);
2121 }
2122
2123
2124 /*
2125  * Have we just finished recovery?  If so, and if we are mounting (or
2126  * remounting) the filesystem readonly, then we will end up with a
2127  * consistent fs on disk.  Record that fact.
2128  */
2129 static void ext3_mark_recovery_complete(struct super_block * sb,
2130                                         struct ext3_super_block * es)
2131 {
2132         journal_t *journal = EXT3_SB(sb)->s_journal;
2133
2134         journal_lock_updates(journal);
2135         journal_flush(journal);
2136         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2137             sb->s_flags & MS_RDONLY) {
2138                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2139                 sb->s_dirt = 0;
2140                 ext3_commit_super(sb, es, 1);
2141         }
2142         journal_unlock_updates(journal);
2143 }
2144
2145 /*
2146  * If we are mounting (or read-write remounting) a filesystem whose journal
2147  * has recorded an error from a previous lifetime, move that error to the
2148  * main filesystem now.
2149  */
2150 static void ext3_clear_journal_err(struct super_block * sb,
2151                                    struct ext3_super_block * es)
2152 {
2153         journal_t *journal;
2154         int j_errno;
2155         const char *errstr;
2156
2157         journal = EXT3_SB(sb)->s_journal;
2158
2159         /*
2160          * Now check for any error status which may have been recorded in the
2161          * journal by a prior ext3_error() or ext3_abort()
2162          */
2163
2164         j_errno = journal_errno(journal);
2165         if (j_errno) {
2166                 char nbuf[16];
2167
2168                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2169                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2170                              "from previous mount: %s", errstr);
2171                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2172                              "filesystem check.");
2173
2174                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2175                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2176                 ext3_commit_super (sb, es, 1);
2177
2178                 journal_clear_err(journal);
2179         }
2180 }
2181
2182 /*
2183  * Force the running and committing transactions to commit,
2184  * and wait on the commit.
2185  */
2186 int ext3_force_commit(struct super_block *sb)
2187 {
2188         journal_t *journal;
2189         int ret;
2190
2191         if (sb->s_flags & MS_RDONLY)
2192                 return 0;
2193
2194         journal = EXT3_SB(sb)->s_journal;
2195         sb->s_dirt = 0;
2196         ret = ext3_journal_force_commit(journal);
2197         return ret;
2198 }
2199
2200 /*
2201  * Ext3 always journals updates to the superblock itself, so we don't
2202  * have to propagate any other updates to the superblock on disk at this
2203  * point.  Just start an async writeback to get the buffers on their way
2204  * to the disk.
2205  *
2206  * This implicitly triggers the writebehind on sync().
2207  */
2208
2209 static void ext3_write_super (struct super_block * sb)
2210 {
2211         if (mutex_trylock(&sb->s_lock) != 0)
2212                 BUG();
2213         sb->s_dirt = 0;
2214 }
2215
2216 static int ext3_sync_fs(struct super_block *sb, int wait)
2217 {
2218         tid_t target;
2219
2220         sb->s_dirt = 0;
2221         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2222                 if (wait)
2223                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2224         }
2225         return 0;
2226 }
2227
2228 /*
2229  * LVM calls this function before a (read-only) snapshot is created.  This
2230  * gives us a chance to flush the journal completely and mark the fs clean.
2231  */
2232 static void ext3_write_super_lockfs(struct super_block *sb)
2233 {
2234         sb->s_dirt = 0;
2235
2236         if (!(sb->s_flags & MS_RDONLY)) {
2237                 journal_t *journal = EXT3_SB(sb)->s_journal;
2238
2239                 /* Now we set up the journal barrier. */
2240                 journal_lock_updates(journal);
2241                 journal_flush(journal);
2242
2243                 /* Journal blocked and flushed, clear needs_recovery flag. */
2244                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2245                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2246         }
2247 }
2248
2249 /*
2250  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2251  * flag here, even though the filesystem is not technically dirty yet.
2252  */
2253 static void ext3_unlockfs(struct super_block *sb)
2254 {
2255         if (!(sb->s_flags & MS_RDONLY)) {
2256                 lock_super(sb);
2257                 /* Reser the needs_recovery flag before the fs is unlocked. */
2258                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2259                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2260                 unlock_super(sb);
2261                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2262         }
2263 }
2264
2265 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2266 {
2267         struct ext3_super_block * es;
2268         struct ext3_sb_info *sbi = EXT3_SB(sb);
2269         ext3_fsblk_t n_blocks_count = 0;
2270         unsigned long old_sb_flags;
2271         struct ext3_mount_options old_opts;
2272         int err;
2273 #ifdef CONFIG_QUOTA
2274         int i;
2275 #endif
2276
2277         /* Store the original options */
2278         old_sb_flags = sb->s_flags;
2279         old_opts.s_mount_opt = sbi->s_mount_opt;
2280         old_opts.s_resuid = sbi->s_resuid;
2281         old_opts.s_resgid = sbi->s_resgid;
2282         old_opts.s_commit_interval = sbi->s_commit_interval;
2283 #ifdef CONFIG_QUOTA
2284         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2285         for (i = 0; i < MAXQUOTAS; i++)
2286                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2287 #endif
2288
2289         /*
2290          * Allow the "check" option to be passed as a remount option.
2291          */
2292         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2293                 err = -EINVAL;
2294                 goto restore_opts;
2295         }
2296
2297         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2298                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2299
2300         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2301                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2302
2303         es = sbi->s_es;
2304
2305         ext3_init_journal_params(sb, sbi->s_journal);
2306
2307         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2308                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2309                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2310                         err = -EROFS;
2311                         goto restore_opts;
2312                 }
2313
2314                 if (*flags & MS_RDONLY) {
2315                         /*
2316                          * First of all, the unconditional stuff we have to do
2317                          * to disable replay of the journal when we next remount
2318                          */
2319                         sb->s_flags |= MS_RDONLY;
2320
2321                         /*
2322                          * OK, test if we are remounting a valid rw partition
2323                          * readonly, and if so set the rdonly flag and then
2324                          * mark the partition as valid again.
2325                          */
2326                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2327                             (sbi->s_mount_state & EXT3_VALID_FS))
2328                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2329
2330                         ext3_mark_recovery_complete(sb, es);
2331                 } else {
2332                         __le32 ret;
2333                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2334                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2335                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2336                                        "remount RDWR because of unsupported "
2337                                        "optional features (%x).\n",
2338                                        sb->s_id, le32_to_cpu(ret));
2339                                 err = -EROFS;
2340                                 goto restore_opts;
2341                         }
2342                         /*
2343                          * Mounting a RDONLY partition read-write, so reread
2344                          * and store the current valid flag.  (It may have
2345                          * been changed by e2fsck since we originally mounted
2346                          * the partition.)
2347                          */
2348                         ext3_clear_journal_err(sb, es);
2349                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2350                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2351                                 err = ret;
2352                                 goto restore_opts;
2353                         }
2354                         if (!ext3_setup_super (sb, es, 0))
2355                                 sb->s_flags &= ~MS_RDONLY;
2356                 }
2357         }
2358 #ifdef CONFIG_QUOTA
2359         /* Release old quota file names */
2360         for (i = 0; i < MAXQUOTAS; i++)
2361                 if (old_opts.s_qf_names[i] &&
2362                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2363                         kfree(old_opts.s_qf_names[i]);
2364 #endif
2365         return 0;
2366 restore_opts:
2367         sb->s_flags = old_sb_flags;
2368         sbi->s_mount_opt = old_opts.s_mount_opt;
2369         sbi->s_resuid = old_opts.s_resuid;
2370         sbi->s_resgid = old_opts.s_resgid;
2371         sbi->s_commit_interval = old_opts.s_commit_interval;
2372 #ifdef CONFIG_QUOTA
2373         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2374         for (i = 0; i < MAXQUOTAS; i++) {
2375                 if (sbi->s_qf_names[i] &&
2376                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2377                         kfree(sbi->s_qf_names[i]);
2378                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2379         }
2380 #endif
2381         return err;
2382 }
2383
2384 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2385 {
2386         struct super_block *sb = dentry->d_sb;
2387         struct ext3_sb_info *sbi = EXT3_SB(sb);
2388         struct ext3_super_block *es = sbi->s_es;
2389         ext3_fsblk_t overhead;
2390         int i;
2391
2392         if (test_opt (sb, MINIX_DF))
2393                 overhead = 0;
2394         else {
2395                 unsigned long ngroups;
2396                 ngroups = EXT3_SB(sb)->s_groups_count;
2397                 smp_rmb();
2398
2399                 /*
2400                  * Compute the overhead (FS structures)
2401                  */
2402
2403                 /*
2404                  * All of the blocks before first_data_block are
2405                  * overhead
2406                  */
2407                 overhead = le32_to_cpu(es->s_first_data_block);
2408
2409                 /*
2410                  * Add the overhead attributed to the superblock and
2411                  * block group descriptors.  If the sparse superblocks
2412                  * feature is turned on, then not all groups have this.
2413                  */
2414                 for (i = 0; i < ngroups; i++) {
2415                         overhead += ext3_bg_has_super(sb, i) +
2416                                 ext3_bg_num_gdb(sb, i);
2417                         cond_resched();
2418                 }
2419
2420                 /*
2421                  * Every block group has an inode bitmap, a block
2422                  * bitmap, and an inode table.
2423                  */
2424                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2425         }
2426
2427         buf->f_type = EXT3_SUPER_MAGIC;
2428         buf->f_bsize = sb->s_blocksize;
2429         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2430         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2431         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2432         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2433                 buf->f_bavail = 0;
2434         buf->f_files = le32_to_cpu(es->s_inodes_count);
2435         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2436         buf->f_namelen = EXT3_NAME_LEN;
2437         return 0;
2438 }
2439
2440 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2441  * is locked for write. Otherwise the are possible deadlocks:
2442  * Process 1                         Process 2
2443  * ext3_create()                     quota_sync()
2444  *   journal_start()                   write_dquot()
2445  *   DQUOT_INIT()                        down(dqio_mutex)
2446  *     down(dqio_mutex)                    journal_start()
2447  *
2448  */
2449
2450 #ifdef CONFIG_QUOTA
2451
2452 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2453 {
2454         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2455 }
2456
2457 static int ext3_dquot_initialize(struct inode *inode, int type)
2458 {
2459         handle_t *handle;
2460         int ret, err;
2461
2462         /* We may create quota structure so we need to reserve enough blocks */
2463         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2464         if (IS_ERR(handle))
2465                 return PTR_ERR(handle);
2466         ret = dquot_initialize(inode, type);
2467         err = ext3_journal_stop(handle);
2468         if (!ret)
2469                 ret = err;
2470         return ret;
2471 }
2472
2473 static int ext3_dquot_drop(struct inode *inode)
2474 {
2475         handle_t *handle;
2476         int ret, err;
2477
2478         /* We may delete quota structure so we need to reserve enough blocks */
2479         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2480         if (IS_ERR(handle))
2481                 return PTR_ERR(handle);
2482         ret = dquot_drop(inode);
2483         err = ext3_journal_stop(handle);
2484         if (!ret)
2485                 ret = err;
2486         return ret;
2487 }
2488
2489 static int ext3_write_dquot(struct dquot *dquot)
2490 {
2491         int ret, err;
2492         handle_t *handle;
2493         struct inode *inode;
2494
2495         inode = dquot_to_inode(dquot);
2496         handle = ext3_journal_start(inode,
2497                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2498         if (IS_ERR(handle))
2499                 return PTR_ERR(handle);
2500         ret = dquot_commit(dquot);
2501         err = ext3_journal_stop(handle);
2502         if (!ret)
2503                 ret = err;
2504         return ret;
2505 }
2506
2507 static int ext3_acquire_dquot(struct dquot *dquot)
2508 {
2509         int ret, err;
2510         handle_t *handle;
2511
2512         handle = ext3_journal_start(dquot_to_inode(dquot),
2513                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2514         if (IS_ERR(handle))
2515                 return PTR_ERR(handle);
2516         ret = dquot_acquire(dquot);
2517         err = ext3_journal_stop(handle);
2518         if (!ret)
2519                 ret = err;
2520         return ret;
2521 }
2522
2523 static int ext3_release_dquot(struct dquot *dquot)
2524 {
2525         int ret, err;
2526         handle_t *handle;
2527
2528         handle = ext3_journal_start(dquot_to_inode(dquot),
2529                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2530         if (IS_ERR(handle))
2531                 return PTR_ERR(handle);
2532         ret = dquot_release(dquot);
2533         err = ext3_journal_stop(handle);
2534         if (!ret)
2535                 ret = err;
2536         return ret;
2537 }
2538
2539 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2540 {
2541         /* Are we journalling quotas? */
2542         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2543             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2544                 dquot_mark_dquot_dirty(dquot);
2545                 return ext3_write_dquot(dquot);
2546         } else {
2547                 return dquot_mark_dquot_dirty(dquot);
2548         }
2549 }
2550
2551 static int ext3_write_info(struct super_block *sb, int type)
2552 {
2553         int ret, err;
2554         handle_t *handle;
2555
2556         /* Data block + inode block */
2557         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2558         if (IS_ERR(handle))
2559                 return PTR_ERR(handle);
2560         ret = dquot_commit_info(sb, type);
2561         err = ext3_journal_stop(handle);
2562         if (!ret)
2563                 ret = err;
2564         return ret;
2565 }
2566
2567 /*
2568  * Turn on quotas during mount time - we need to find
2569  * the quota file and such...
2570  */
2571 static int ext3_quota_on_mount(struct super_block *sb, int type)
2572 {
2573         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2574                         EXT3_SB(sb)->s_jquota_fmt, type);
2575 }
2576
2577 /*
2578  * Standard function to be called on quota_on
2579  */
2580 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2581                          char *path)
2582 {
2583         int err;
2584         struct nameidata nd;
2585
2586         if (!test_opt(sb, QUOTA))
2587                 return -EINVAL;
2588         /* Not journalling quota? */
2589         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2590             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2591                 return vfs_quota_on(sb, type, format_id, path);
2592         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2593         if (err)
2594                 return err;
2595         /* Quotafile not on the same filesystem? */
2596         if (nd.mnt->mnt_sb != sb) {
2597                 path_release(&nd);
2598                 return -EXDEV;
2599         }
2600         /* Quotafile not of fs root? */
2601         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2602                 printk(KERN_WARNING
2603                         "EXT3-fs: Quota file not on filesystem root. "
2604                         "Journalled quota will not work.\n");
2605         path_release(&nd);
2606         return vfs_quota_on(sb, type, format_id, path);
2607 }
2608
2609 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2610  * acquiring the locks... As quota files are never truncated and quota code
2611  * itself serializes the operations (and noone else should touch the files)
2612  * we don't have to be afraid of races */
2613 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2614                                size_t len, loff_t off)
2615 {
2616         struct inode *inode = sb_dqopt(sb)->files[type];
2617         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2618         int err = 0;
2619         int offset = off & (sb->s_blocksize - 1);
2620         int tocopy;
2621         size_t toread;
2622         struct buffer_head *bh;
2623         loff_t i_size = i_size_read(inode);
2624
2625         if (off > i_size)
2626                 return 0;
2627         if (off+len > i_size)
2628                 len = i_size-off;
2629         toread = len;
2630         while (toread > 0) {
2631                 tocopy = sb->s_blocksize - offset < toread ?
2632                                 sb->s_blocksize - offset : toread;
2633                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2634                 if (err)
2635                         return err;
2636                 if (!bh)        /* A hole? */
2637                         memset(data, 0, tocopy);
2638                 else
2639                         memcpy(data, bh->b_data+offset, tocopy);
2640                 brelse(bh);
2641                 offset = 0;
2642                 toread -= tocopy;
2643                 data += tocopy;
2644                 blk++;
2645         }
2646         return len;
2647 }
2648
2649 /* Write to quotafile (we know the transaction is already started and has
2650  * enough credits) */
2651 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2652                                 const char *data, size_t len, loff_t off)
2653 {
2654         struct inode *inode = sb_dqopt(sb)->files[type];
2655         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2656         int err = 0;
2657         int offset = off & (sb->s_blocksize - 1);
2658         int tocopy;
2659         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2660         size_t towrite = len;
2661         struct buffer_head *bh;
2662         handle_t *handle = journal_current_handle();
2663
2664         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2665         while (towrite > 0) {
2666                 tocopy = sb->s_blocksize - offset < towrite ?
2667                                 sb->s_blocksize - offset : towrite;
2668                 bh = ext3_bread(handle, inode, blk, 1, &err);
2669                 if (!bh)
2670                         goto out;
2671                 if (journal_quota) {
2672                         err = ext3_journal_get_write_access(handle, bh);
2673                         if (err) {
2674                                 brelse(bh);
2675                                 goto out;
2676                         }
2677                 }
2678                 lock_buffer(bh);
2679                 memcpy(bh->b_data+offset, data, tocopy);
2680                 flush_dcache_page(bh->b_page);
2681                 unlock_buffer(bh);
2682                 if (journal_quota)
2683                         err = ext3_journal_dirty_metadata(handle, bh);
2684                 else {
2685                         /* Always do at least ordered writes for quotas */
2686                         err = ext3_journal_dirty_data(handle, bh);
2687                         mark_buffer_dirty(bh);
2688                 }
2689                 brelse(bh);
2690                 if (err)
2691                         goto out;
2692                 offset = 0;
2693                 towrite -= tocopy;
2694                 data += tocopy;
2695                 blk++;
2696         }
2697 out:
2698         if (len == towrite)
2699                 return err;
2700         if (inode->i_size < off+len-towrite) {
2701                 i_size_write(inode, off+len-towrite);
2702                 EXT3_I(inode)->i_disksize = inode->i_size;
2703         }
2704         inode->i_version++;
2705         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2706         ext3_mark_inode_dirty(handle, inode);
2707         mutex_unlock(&inode->i_mutex);
2708         return len - towrite;
2709 }
2710
2711 #endif
2712
2713 static int ext3_get_sb(struct file_system_type *fs_type,
2714         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2715 {
2716         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2717 }
2718
2719 static struct file_system_type ext3_fs_type = {
2720         .owner          = THIS_MODULE,
2721         .name           = "ext3",
2722         .get_sb         = ext3_get_sb,
2723         .kill_sb        = kill_block_super,
2724         .fs_flags       = FS_REQUIRES_DEV,
2725 };
2726
2727 static int __init init_ext3_fs(void)
2728 {
2729         int err = init_ext3_xattr();
2730         if (err)
2731                 return err;
2732         err = init_inodecache();
2733         if (err)
2734                 goto out1;
2735         err = register_filesystem(&ext3_fs_type);
2736         if (err)
2737                 goto out;
2738         return 0;
2739 out:
2740         destroy_inodecache();
2741 out1:
2742         exit_ext3_xattr();
2743         return err;
2744 }
2745
2746 static void __exit exit_ext3_fs(void)
2747 {
2748         unregister_filesystem(&ext3_fs_type);
2749         destroy_inodecache();
2750         exit_ext3_xattr();
2751 }
2752
2753 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2754 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2755 MODULE_LICENSE("GPL");
2756 module_init(init_ext3_fs)
2757 module_exit(exit_ext3_fs)