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_CONT)) {
 163                 journal_t *journal = EXT3_SB(sb)->s_journal;
 164
 165                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 166                 if (journal)
 167                         journal_abort(journal, -EIO);
 168         }
 169         if (test_opt (sb, ERRORS_RO)) {
 170                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 171                 sb->s_flags |= MS_RDONLY;
 172         }
 173         ext3_commit_super(sb, es, 1);
 174         if (test_opt(sb, ERRORS_PANIC))
 175                 panic("EXT3-fs (device %s): panic forced after error\n",
 176                         sb->s_id);
 177 }
 178
 179 void ext3_error (struct super_block * sb, const char * function,
 180                  const char * fmt, ...)
 181 {
 182         va_list args;
 183
 184         va_start(args, fmt);
 185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 186         vprintk(fmt, args);
 187         printk("\n");
 188         va_end(args);
 189
 190         ext3_handle_error(sb);
 191 }
 192
 193 static const char *ext3_decode_error(struct super_block * sb, int errno,
 194                                      char nbuf[16])
 195 {
 196         char *errstr = NULL;
 197
 198         switch (errno) {
 199         case -EIO:
 200                 errstr = "IO failure";
 201                 break;
 202         case -ENOMEM:
 203                 errstr = "Out of memory";
 204                 break;
 205         case -EROFS:
 206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 207                         errstr = "Journal has aborted";
 208                 else
 209                         errstr = "Readonly filesystem";
 210                 break;
 211         default:
 212                 /* If the caller passed in an extra buffer for unknown
 213                  * errors, textualise them now.  Else we just return
 214                  * NULL. */
 215                 if (nbuf) {
 216                         /* Check for truncated error codes... */
 217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 218                                 errstr = nbuf;
 219                 }
 220                 break;
 221         }
 222
 223         return errstr;
 224 }
 225
 226 /* __ext3_std_error decodes expected errors from journaling functions
 227  * automatically and invokes the appropriate error response.  */
 228
 229 void __ext3_std_error (struct super_block * sb, const char * function,
 230                        int errno)
 231 {
 232         char nbuf[16];
 233         const char *errstr;
 234
 235         /* Special case: if the error is EROFS, and we're not already
 236          * inside a transaction, then there's really no point in logging
 237          * an error. */
 238         if (errno == -EROFS && journal_current_handle() == NULL &&
 239             (sb->s_flags & MS_RDONLY))
 240                 return;
 241
 242         errstr = ext3_decode_error(sb, errno, nbuf);
 243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 244                 sb->s_id, function, errstr);
 245
 246         ext3_handle_error(sb);
 247 }
 248
 249 /*
 250  * ext3_abort is a much stronger failure handler than ext3_error.  The
 251  * abort function may be used to deal with unrecoverable failures such
 252  * as journal IO errors or ENOMEM at a critical moment in log management.
 253  *
 254  * We unconditionally force the filesystem into an ABORT|READONLY state,
 255  * unless the error response on the fs has been set to panic in which
 256  * case we take the easy way out and panic immediately.
 257  */
 258
 259 void ext3_abort (struct super_block * sb, const char * function,
 260                  const char * fmt, ...)
 261 {
 262         va_list args;
 263
 264         printk (KERN_CRIT "ext3_abort called.\n");
 265
 266         va_start(args, fmt);
 267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 268         vprintk(fmt, args);
 269         printk("\n");
 270         va_end(args);
 271
 272         if (test_opt(sb, ERRORS_PANIC))
 273                 panic("EXT3-fs panic from previous error\n");
 274
 275         if (sb->s_flags & MS_RDONLY)
 276                 return;
 277
 278         printk(KERN_CRIT "Remounting filesystem read-only\n");
 279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 280         sb->s_flags |= MS_RDONLY;
 281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 283 }
 284
 285 void ext3_warning (struct super_block * sb, const char * function,
 286                    const char * fmt, ...)
 287 {
 288         va_list args;
 289
 290         va_start(args, fmt);
 291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 292                sb->s_id, function);
 293         vprintk(fmt, args);
 294         printk("\n");
 295         va_end(args);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        NEXT_ORPHAN(inode));
 384         }
 385 }
 386
 387 static void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 408         percpu_counter_destroy(&sbi->s_dirs_counter);
 409         brelse(sbi->s_sbh);
 410 #ifdef CONFIG_QUOTA
 411         for (i = 0; i < MAXQUOTAS; i++)
 412                 kfree(sbi->s_qf_names[i]);
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static struct kmem_cache *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->i_block_alloc_info = NULL;
 456         ei->vfs_inode.i_version = 1;
 457         return &ei->vfs_inode;
 458 }
 459
 460 static void ext3_destroy_inode(struct inode *inode)
 461 {
 462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 463 }
 464
 465 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
 466 {
 467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 468
 469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 470             SLAB_CTOR_CONSTRUCTOR) {
 471                 INIT_LIST_HEAD(&ei->i_orphan);
 472 #ifdef CONFIG_EXT3_FS_XATTR
 473                 init_rwsem(&ei->xattr_sem);
 474 #endif
 475                 mutex_init(&ei->truncate_mutex);
 476                 inode_init_once(&ei->vfs_inode);
 477         }
 478 }
 479
 480 static int init_inodecache(void)
 481 {
 482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 483                                              sizeof(struct ext3_inode_info),
 484                                              0, (SLAB_RECLAIM_ACCOUNT|
 485                                                 SLAB_MEM_SPREAD),
 486                                              init_once, NULL);
 487         if (ext3_inode_cachep == NULL)
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static void destroy_inodecache(void)
 493 {
 494         kmem_cache_destroy(ext3_inode_cachep);
 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, Opt_tag, Opt_notag, Opt_tagid
 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_tag, "tag"},
 731         {Opt_notag, "notag"},
 732         {Opt_tagid, "tagid=%u"},
 733         {Opt_tag, "tagxid"},
 734         {Opt_err, NULL},
 735         {Opt_resize, "resize"},
 736 };
 737
 738 static ext3_fsblk_t get_sb_block(void **data)
 739 {
 740         ext3_fsblk_t    sb_block;
 741         char            *options = (char *) *data;
 742
 743         if (!options || strncmp(options, "sb=", 3) != 0)
 744                 return 1;       /* Default location */
 745         options += 3;
 746         /*todo: use simple_strtoll with >32bit ext3 */
 747         sb_block = simple_strtoul(options, &options, 0);
 748         if (*options && *options != ',') {
 749                 printk("EXT3-fs: Invalid sb specification: %s\n",
 750                        (char *) *data);
 751                 return 1;
 752         }
 753         if (*options == ',')
 754                 options++;
 755         *data = (void *) options;
 756         return sb_block;
 757 }
 758
 759 static int parse_options (char *options, struct super_block *sb,
 760                           unsigned int *inum, unsigned long *journal_devnum,
 761                           ext3_fsblk_t *n_blocks_count, int is_remount)
 762 {
 763         struct ext3_sb_info *sbi = EXT3_SB(sb);
 764         char * p;
 765         substring_t args[MAX_OPT_ARGS];
 766         int data_opt = 0;
 767         int option;
 768 #ifdef CONFIG_QUOTA
 769         int qtype;
 770         char *qname;
 771 #endif
 772
 773         if (!options)
 774                 return 1;
 775
 776         while ((p = strsep (&options, ",")) != NULL) {
 777                 int token;
 778                 if (!*p)
 779                         continue;
 780
 781                 token = match_token(p, tokens, args);
 782                 switch (token) {
 783                 case Opt_bsd_df:
 784                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 785                         break;
 786                 case Opt_minix_df:
 787                         set_opt (sbi->s_mount_opt, MINIX_DF);
 788                         break;
 789                 case Opt_grpid:
 790                         set_opt (sbi->s_mount_opt, GRPID);
 791                         break;
 792                 case Opt_nogrpid:
 793                         clear_opt (sbi->s_mount_opt, GRPID);
 794                         break;
 795                 case Opt_resuid:
 796                         if (match_int(&args[0], &option))
 797                                 return 0;
 798                         sbi->s_resuid = option;
 799                         break;
 800                 case Opt_resgid:
 801                         if (match_int(&args[0], &option))
 802                                 return 0;
 803                         sbi->s_resgid = option;
 804                         break;
 805                 case Opt_sb:
 806                         /* handled by get_sb_block() instead of here */
 807                         /* *sb_block = match_int(&args[0]); */
 808                         break;
 809                 case Opt_err_panic:
 810                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 811                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 812                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 813                         break;
 814                 case Opt_err_ro:
 815                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 816                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 817                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 818                         break;
 819                 case Opt_err_cont:
 820                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 821                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 822                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 823                         break;
 824                 case Opt_nouid32:
 825                         set_opt (sbi->s_mount_opt, NO_UID32);
 826                         break;
 827 #ifndef CONFIG_TAGGING_NONE
 828                 case Opt_tag:
 829                         set_opt (sbi->s_mount_opt, TAGGED);
 830                         break;
 831                 case Opt_notag:
 832                         clear_opt (sbi->s_mount_opt, TAGGED);
 833                         break;
 834 #endif
 835 #ifdef CONFIG_PROPAGATE
 836                 case Opt_tagid:
 837                         /* use args[0] */
 838                         set_opt (sbi->s_mount_opt, TAGGED);
 839                         break;
 840 #endif
 841                 case Opt_nocheck:
 842                         clear_opt (sbi->s_mount_opt, CHECK);
 843                         break;
 844                 case Opt_debug:
 845                         set_opt (sbi->s_mount_opt, DEBUG);
 846                         break;
 847                 case Opt_oldalloc:
 848                         set_opt (sbi->s_mount_opt, OLDALLOC);
 849                         break;
 850                 case Opt_orlov:
 851                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 852                         break;
 853 #ifdef CONFIG_EXT3_FS_XATTR
 854                 case Opt_user_xattr:
 855                         set_opt (sbi->s_mount_opt, XATTR_USER);
 856                         break;
 857                 case Opt_nouser_xattr:
 858                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 859                         break;
 860 #else
 861                 case Opt_user_xattr:
 862                 case Opt_nouser_xattr:
 863                         printk("EXT3 (no)user_xattr options not supported\n");
 864                         break;
 865 #endif
 866 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 867                 case Opt_acl:
 868                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 869                         break;
 870                 case Opt_noacl:
 871                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 872                         break;
 873 #else
 874                 case Opt_acl:
 875                 case Opt_noacl:
 876                         printk("EXT3 (no)acl options not supported\n");
 877                         break;
 878 #endif
 879                 case Opt_reservation:
 880                         set_opt(sbi->s_mount_opt, RESERVATION);
 881                         break;
 882                 case Opt_noreservation:
 883                         clear_opt(sbi->s_mount_opt, RESERVATION);
 884                         break;
 885                 case Opt_journal_update:
 886                         /* @@@ FIXME */
 887                         /* Eventually we will want to be able to create
 888                            a journal file here.  For now, only allow the
 889                            user to specify an existing inode to be the
 890                            journal file. */
 891                         if (is_remount) {
 892                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 893                                        "journal on remount\n");
 894                                 return 0;
 895                         }
 896                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 897                         break;
 898                 case Opt_journal_inum:
 899                         if (is_remount) {
 900                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 901                                        "journal on remount\n");
 902                                 return 0;
 903                         }
 904                         if (match_int(&args[0], &option))
 905                                 return 0;
 906                         *inum = option;
 907                         break;
 908                 case Opt_journal_dev:
 909                         if (is_remount) {
 910                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 911                                        "journal on remount\n");
 912                                 return 0;
 913                         }
 914                         if (match_int(&args[0], &option))
 915                                 return 0;
 916                         *journal_devnum = option;
 917                         break;
 918                 case Opt_noload:
 919                         set_opt (sbi->s_mount_opt, NOLOAD);
 920                         break;
 921                 case Opt_commit:
 922                         if (match_int(&args[0], &option))
 923                                 return 0;
 924                         if (option < 0)
 925                                 return 0;
 926                         if (option == 0)
 927                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 928                         sbi->s_commit_interval = HZ * option;
 929                         break;
 930                 case Opt_data_journal:
 931                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 932                         goto datacheck;
 933                 case Opt_data_ordered:
 934                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 935                         goto datacheck;
 936                 case Opt_data_writeback:
 937                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 938                 datacheck:
 939                         if (is_remount) {
 940                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 941                                                 != data_opt) {
 942                                         printk(KERN_ERR
 943                                                 "EXT3-fs: cannot change data "
 944                                                 "mode on remount\n");
 945                                         return 0;
 946                                 }
 947                         } else {
 948                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 949                                 sbi->s_mount_opt |= data_opt;
 950                         }
 951                         break;
 952 #ifdef CONFIG_QUOTA
 953                 case Opt_usrjquota:
 954                         qtype = USRQUOTA;
 955                         goto set_qf_name;
 956                 case Opt_grpjquota:
 957                         qtype = GRPQUOTA;
 958 set_qf_name:
 959                         if (sb_any_quota_enabled(sb)) {
 960                                 printk(KERN_ERR
 961                                         "EXT3-fs: Cannot change journalled "
 962                                         "quota options when quota turned on.\n");
 963                                 return 0;
 964                         }
 965                         qname = match_strdup(&args[0]);
 966                         if (!qname) {
 967                                 printk(KERN_ERR
 968                                         "EXT3-fs: not enough memory for "
 969                                         "storing quotafile name.\n");
 970                                 return 0;
 971                         }
 972                         if (sbi->s_qf_names[qtype] &&
 973                             strcmp(sbi->s_qf_names[qtype], qname)) {
 974                                 printk(KERN_ERR
 975                                         "EXT3-fs: %s quota file already "
 976                                         "specified.\n", QTYPE2NAME(qtype));
 977                                 kfree(qname);
 978                                 return 0;
 979                         }
 980                         sbi->s_qf_names[qtype] = qname;
 981                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 982                                 printk(KERN_ERR
 983                                         "EXT3-fs: quotafile must be on "
 984                                         "filesystem root.\n");
 985                                 kfree(sbi->s_qf_names[qtype]);
 986                                 sbi->s_qf_names[qtype] = NULL;
 987                                 return 0;
 988                         }
 989                         set_opt(sbi->s_mount_opt, QUOTA);
 990                         break;
 991                 case Opt_offusrjquota:
 992                         qtype = USRQUOTA;
 993                         goto clear_qf_name;
 994                 case Opt_offgrpjquota:
 995                         qtype = GRPQUOTA;
 996 clear_qf_name:
 997                         if (sb_any_quota_enabled(sb)) {
 998                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 999                                         "journalled quota options when "
1000                                         "quota turned on.\n");
1001                                 return 0;
1002                         }
1003                         /*
1004                          * The space will be released later when all options
1005                          * are confirmed to be correct
1006                          */
1007                         sbi->s_qf_names[qtype] = NULL;
1008                         break;
1009                 case Opt_jqfmt_vfsold:
1010                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1011                         break;
1012                 case Opt_jqfmt_vfsv0:
1013                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1014                         break;
1015                 case Opt_quota:
1016                 case Opt_usrquota:
1017                         set_opt(sbi->s_mount_opt, QUOTA);
1018                         set_opt(sbi->s_mount_opt, USRQUOTA);
1019                         break;
1020                 case Opt_grpquota:
1021                         set_opt(sbi->s_mount_opt, QUOTA);
1022                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1023                         break;
1024                 case Opt_noquota:
1025                         if (sb_any_quota_enabled(sb)) {
1026                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1027                                         "options when quota turned on.\n");
1028                                 return 0;
1029                         }
1030                         clear_opt(sbi->s_mount_opt, QUOTA);
1031                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1032                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1033                         break;
1034 #else
1035                 case Opt_quota:
1036                 case Opt_usrquota:
1037                 case Opt_grpquota:
1038                 case Opt_usrjquota:
1039                 case Opt_grpjquota:
1040                 case Opt_offusrjquota:
1041                 case Opt_offgrpjquota:
1042                 case Opt_jqfmt_vfsold:
1043                 case Opt_jqfmt_vfsv0:
1044                         printk(KERN_ERR
1045                                 "EXT3-fs: journalled quota options not "
1046                                 "supported.\n");
1047                         break;
1048                 case Opt_noquota:
1049                         break;
1050 #endif
1051                 case Opt_abort:
1052                         set_opt(sbi->s_mount_opt, ABORT);
1053                         break;
1054                 case Opt_barrier:
1055                         if (match_int(&args[0], &option))
1056                                 return 0;
1057                         if (option)
1058                                 set_opt(sbi->s_mount_opt, BARRIER);
1059                         else
1060                                 clear_opt(sbi->s_mount_opt, BARRIER);
1061                         break;
1062                 case Opt_ignore:
1063                         break;
1064                 case Opt_resize:
1065                         if (!is_remount) {
1066                                 printk("EXT3-fs: resize option only available "
1067                                         "for remount\n");
1068                                 return 0;
1069                         }
1070                         if (match_int(&args[0], &option) != 0)
1071                                 return 0;
1072                         *n_blocks_count = option;
1073                         break;
1074                 case Opt_nobh:
1075                         set_opt(sbi->s_mount_opt, NOBH);
1076                         break;
1077                 case Opt_bh:
1078                         clear_opt(sbi->s_mount_opt, NOBH);
1079                         break;
1080                 default:
1081                         printk (KERN_ERR
1082                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1083                                 "or missing value\n", p);
1084                         return 0;
1085                 }
1086         }
1087 #ifdef CONFIG_QUOTA
1088         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1089                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1090                      sbi->s_qf_names[USRQUOTA])
1091                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1092
1093                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1094                      sbi->s_qf_names[GRPQUOTA])
1095                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1096
1097                 if ((sbi->s_qf_names[USRQUOTA] &&
1098                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1099                     (sbi->s_qf_names[GRPQUOTA] &&
1100                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1101                         printk(KERN_ERR "EXT3-fs: old and new quota "
1102                                         "format mixing.\n");
1103                         return 0;
1104                 }
1105
1106                 if (!sbi->s_jquota_fmt) {
1107                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1108                                         "not specified.\n");
1109                         return 0;
1110                 }
1111         } else {
1112                 if (sbi->s_jquota_fmt) {
1113                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1114                                         "specified with no journalling "
1115                                         "enabled.\n");
1116                         return 0;
1117                 }
1118         }
1119 #endif
1120         return 1;
1121 }
1122
1123 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1124                             int read_only)
1125 {
1126         struct ext3_sb_info *sbi = EXT3_SB(sb);
1127         int res = 0;
1128
1129         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1130                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1131                         "forcing read-only mode\n");
1132                 res = MS_RDONLY;
1133         }
1134         if (read_only)
1135                 return res;
1136         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1137                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1138                         "running e2fsck is recommended\n");
1139         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1140                 printk (KERN_WARNING
1141                         "EXT3-fs warning: mounting fs with errors, "
1142                         "running e2fsck is recommended\n");
1143         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1144                  le16_to_cpu(es->s_mnt_count) >=
1145                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1146                 printk (KERN_WARNING
1147                         "EXT3-fs warning: maximal mount count reached, "
1148                         "running e2fsck is recommended\n");
1149         else if (le32_to_cpu(es->s_checkinterval) &&
1150                 (le32_to_cpu(es->s_lastcheck) +
1151                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1152                 printk (KERN_WARNING
1153                         "EXT3-fs warning: checktime reached, "
1154                         "running e2fsck is recommended\n");
1155 #if 0
1156                 /* @@@ We _will_ want to clear the valid bit if we find
1157                    inconsistencies, to force a fsck at reboot.  But for
1158                    a plain journaled filesystem we can keep it set as
1159                    valid forever! :) */
1160         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1161 #endif
1162         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1163                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1164         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1165         es->s_mtime = cpu_to_le32(get_seconds());
1166         ext3_update_dynamic_rev(sb);
1167         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1168
1169         ext3_commit_super(sb, es, 1);
1170         if (test_opt(sb, DEBUG))
1171                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1172                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1173                         sb->s_blocksize,
1174                         sbi->s_groups_count,
1175                         EXT3_BLOCKS_PER_GROUP(sb),
1176                         EXT3_INODES_PER_GROUP(sb),
1177                         sbi->s_mount_opt);
1178
1179         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1180         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1181                 char b[BDEVNAME_SIZE];
1182
1183                 printk("external journal on %s\n",
1184                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1185         } else {
1186                 printk("internal journal\n");
1187         }
1188         return res;
1189 }
1190
1191 /* Called at mount-time, super-block is locked */
1192 static int ext3_check_descriptors (struct super_block * sb)
1193 {
1194         struct ext3_sb_info *sbi = EXT3_SB(sb);
1195         ext3_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1196         ext3_fsblk_t last_block;
1197         struct ext3_group_desc * gdp = NULL;
1198         int desc_block = 0;
1199         int i;
1200
1201         ext3_debug ("Checking group descriptors");
1202
1203         for (i = 0; i < sbi->s_groups_count; i++)
1204         {
1205                 if (i == sbi->s_groups_count - 1)
1206                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1207                 else
1208                         last_block = first_block +
1209                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1210
1211                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1212                         gdp = (struct ext3_group_desc *)
1213                                         sbi->s_group_desc[desc_block++]->b_data;
1214                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1215                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1216                 {
1217                         ext3_error (sb, "ext3_check_descriptors",
1218                                     "Block bitmap for group %d"
1219                                     " not in group (block %lu)!",
1220                                     i, (unsigned long)
1221                                         le32_to_cpu(gdp->bg_block_bitmap));
1222                         return 0;
1223                 }
1224                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1225                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1226                 {
1227                         ext3_error (sb, "ext3_check_descriptors",
1228                                     "Inode bitmap for group %d"
1229                                     " not in group (block %lu)!",
1230                                     i, (unsigned long)
1231                                         le32_to_cpu(gdp->bg_inode_bitmap));
1232                         return 0;
1233                 }
1234                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1235                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >
1236                     last_block)
1237                 {
1238                         ext3_error (sb, "ext3_check_descriptors",
1239                                     "Inode table for group %d"
1240                                     " not in group (block %lu)!",
1241                                     i, (unsigned long)
1242                                         le32_to_cpu(gdp->bg_inode_table));
1243                         return 0;
1244                 }
1245                 first_block += EXT3_BLOCKS_PER_GROUP(sb);
1246                 gdp++;
1247         }
1248
1249         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1250         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1251         return 1;
1252 }
1253
1254
1255 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1256  * the superblock) which were deleted from all directories, but held open by
1257  * a process at the time of a crash.  We walk the list and try to delete these
1258  * inodes at recovery time (only with a read-write filesystem).
1259  *
1260  * In order to keep the orphan inode chain consistent during traversal (in
1261  * case of crash during recovery), we link each inode into the superblock
1262  * orphan list_head and handle it the same way as an inode deletion during
1263  * normal operation (which journals the operations for us).
1264  *
1265  * We only do an iget() and an iput() on each inode, which is very safe if we
1266  * accidentally point at an in-use or already deleted inode.  The worst that
1267  * can happen in this case is that we get a "bit already cleared" message from
1268  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1269  * e2fsck was run on this filesystem, and it must have already done the orphan
1270  * inode cleanup for us, so we can safely abort without any further action.
1271  */
1272 static void ext3_orphan_cleanup (struct super_block * sb,
1273                                  struct ext3_super_block * es)
1274 {
1275         unsigned int s_flags = sb->s_flags;
1276         int nr_orphans = 0, nr_truncates = 0;
1277 #ifdef CONFIG_QUOTA
1278         int i;
1279 #endif
1280         if (!es->s_last_orphan) {
1281                 jbd_debug(4, "no orphan inodes to clean up\n");
1282                 return;
1283         }
1284
1285         if (bdev_read_only(sb->s_bdev)) {
1286                 printk(KERN_ERR "EXT3-fs: write access "
1287                         "unavailable, skipping orphan cleanup.\n");
1288                 return;
1289         }
1290
1291         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1292                 if (es->s_last_orphan)
1293                         jbd_debug(1, "Errors on filesystem, "
1294                                   "clearing orphan list.\n");
1295                 es->s_last_orphan = 0;
1296                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1297                 return;
1298         }
1299
1300         if (s_flags & MS_RDONLY) {
1301                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1302                        sb->s_id);
1303                 sb->s_flags &= ~MS_RDONLY;
1304         }
1305 #ifdef CONFIG_QUOTA
1306         /* Needed for iput() to work correctly and not trash data */
1307         sb->s_flags |= MS_ACTIVE;
1308         /* Turn on quotas so that they are updated correctly */
1309         for (i = 0; i < MAXQUOTAS; i++) {
1310                 if (EXT3_SB(sb)->s_qf_names[i]) {
1311                         int ret = ext3_quota_on_mount(sb, i);
1312                         if (ret < 0)
1313                                 printk(KERN_ERR
1314                                         "EXT3-fs: Cannot turn on journalled "
1315                                         "quota: error %d\n", ret);
1316                 }
1317         }
1318 #endif
1319
1320         while (es->s_last_orphan) {
1321                 struct inode *inode;
1322
1323                 if (!(inode =
1324                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1325                         es->s_last_orphan = 0;
1326                         break;
1327                 }
1328
1329                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1330                 DQUOT_INIT(inode);
1331                 if (inode->i_nlink) {
1332                         printk(KERN_DEBUG
1333                                 "%s: truncating inode %lu to %Ld bytes\n",
1334                                 __FUNCTION__, inode->i_ino, inode->i_size);
1335                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1336                                   inode->i_ino, inode->i_size);
1337                         ext3_truncate(inode);
1338                         nr_truncates++;
1339                 } else {
1340                         printk(KERN_DEBUG
1341                                 "%s: deleting unreferenced inode %lu\n",
1342                                 __FUNCTION__, inode->i_ino);
1343                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1344                                   inode->i_ino);
1345                         nr_orphans++;
1346                 }
1347                 iput(inode);  /* The delete magic happens here! */
1348         }
1349
1350 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1351
1352         if (nr_orphans)
1353                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1354                        sb->s_id, PLURAL(nr_orphans));
1355         if (nr_truncates)
1356                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1357                        sb->s_id, PLURAL(nr_truncates));
1358 #ifdef CONFIG_QUOTA
1359         /* Turn quotas off */
1360         for (i = 0; i < MAXQUOTAS; i++) {
1361                 if (sb_dqopt(sb)->files[i])
1362                         vfs_quota_off(sb, i);
1363         }
1364 #endif
1365         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1366 }
1367
1368 /*
1369  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1370  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1371  * We need to be 1 filesystem block less than the 2^32 sector limit.
1372  */
1373 static loff_t ext3_max_size(int bits)
1374 {
1375         loff_t res = EXT3_NDIR_BLOCKS;
1376         /* This constant is calculated to be the largest file size for a
1377          * dense, 4k-blocksize file such that the total number of
1378          * sectors in the file, including data and all indirect blocks,
1379          * does not exceed 2^32. */
1380         const loff_t upper_limit = 0x1ff7fffd000LL;
1381
1382         res += 1LL << (bits-2);
1383         res += 1LL << (2*(bits-2));
1384         res += 1LL << (3*(bits-2));
1385         res <<= bits;
1386         if (res > upper_limit)
1387                 res = upper_limit;
1388         return res;
1389 }
1390
1391 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1392                                     ext3_fsblk_t logic_sb_block,
1393                                     int nr)
1394 {
1395         struct ext3_sb_info *sbi = EXT3_SB(sb);
1396         unsigned long bg, first_meta_bg;
1397         int has_super = 0;
1398
1399         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1400
1401         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1402             nr < first_meta_bg)
1403                 return (logic_sb_block + nr + 1);
1404         bg = sbi->s_desc_per_block * nr;
1405         if (ext3_bg_has_super(sb, bg))
1406                 has_super = 1;
1407         return (has_super + ext3_group_first_block_no(sb, bg));
1408 }
1409
1410
1411 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1412 {
1413         struct buffer_head * bh;
1414         struct ext3_super_block *es = NULL;
1415         struct ext3_sb_info *sbi;
1416         ext3_fsblk_t block;
1417         ext3_fsblk_t sb_block = get_sb_block(&data);
1418         ext3_fsblk_t logic_sb_block;
1419         unsigned long offset = 0;
1420         unsigned int journal_inum = 0;
1421         unsigned long journal_devnum = 0;
1422         unsigned long def_mount_opts;
1423         struct inode *root;
1424         int blocksize;
1425         int hblock;
1426         int db_count;
1427         int i;
1428         int needs_recovery;
1429         __le32 features;
1430
1431         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1432         if (!sbi)
1433                 return -ENOMEM;
1434         sb->s_fs_info = sbi;
1435         sbi->s_mount_opt = 0;
1436         sbi->s_resuid = EXT3_DEF_RESUID;
1437         sbi->s_resgid = EXT3_DEF_RESGID;
1438
1439         unlock_kernel();
1440
1441         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1442         if (!blocksize) {
1443                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1444                 goto out_fail;
1445         }
1446
1447         /*
1448          * The ext3 superblock will not be buffer aligned for other than 1kB
1449          * block sizes.  We need to calculate the offset from buffer start.
1450          */
1451         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1452                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1453                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1454         } else {
1455                 logic_sb_block = sb_block;
1456         }
1457
1458         if (!(bh = sb_bread(sb, logic_sb_block))) {
1459                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1460                 goto out_fail;
1461         }
1462         /*
1463          * Note: s_es must be initialized as soon as possible because
1464          *       some ext3 macro-instructions depend on its value
1465          */
1466         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1467         sbi->s_es = es;
1468         sb->s_magic = le16_to_cpu(es->s_magic);
1469         if (sb->s_magic != EXT3_SUPER_MAGIC)
1470                 goto cantfind_ext3;
1471
1472         /* Set defaults before we parse the mount options */
1473         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1474         if (def_mount_opts & EXT3_DEFM_DEBUG)
1475                 set_opt(sbi->s_mount_opt, DEBUG);
1476         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1477                 set_opt(sbi->s_mount_opt, GRPID);
1478         if (def_mount_opts & EXT3_DEFM_UID16)
1479                 set_opt(sbi->s_mount_opt, NO_UID32);
1480 #ifdef CONFIG_EXT3_FS_XATTR
1481         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1482                 set_opt(sbi->s_mount_opt, XATTR_USER);
1483 #endif
1484 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1485         if (def_mount_opts & EXT3_DEFM_ACL)
1486                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1487 #endif
1488         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1489                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1490         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1491                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1492         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1493                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1494
1495         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1496                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1497         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1498                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1499         else
1500                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1501
1502         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1503         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1504
1505         set_opt(sbi->s_mount_opt, RESERVATION);
1506
1507         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1508                             NULL, 0))
1509                 goto failed_mount;
1510
1511         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAGGED)
1512                 sb->s_flags |= MS_TAGGED;
1513
1514         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1515                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1516
1517         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1518             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1519              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1520              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1521                 printk(KERN_WARNING
1522                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1523                        "running e2fsck is recommended\n");
1524         /*
1525          * Check feature flags regardless of the revision level, since we
1526          * previously didn't change the revision level when setting the flags,
1527          * so there is a chance incompat flags are set on a rev 0 filesystem.
1528          */
1529         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1530         if (features) {
1531                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1532                        "unsupported optional features (%x).\n",
1533                        sb->s_id, le32_to_cpu(features));
1534                 goto failed_mount;
1535         }
1536         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1537         if (!(sb->s_flags & MS_RDONLY) && features) {
1538                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1539                        "unsupported optional features (%x).\n",
1540                        sb->s_id, le32_to_cpu(features));
1541                 goto failed_mount;
1542         }
1543         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1544
1545         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1546             blocksize > EXT3_MAX_BLOCK_SIZE) {
1547                 printk(KERN_ERR
1548                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1549                        blocksize, sb->s_id);
1550                 goto failed_mount;
1551         }
1552
1553         hblock = bdev_hardsect_size(sb->s_bdev);
1554         if (sb->s_blocksize != blocksize) {
1555                 /*
1556                  * Make sure the blocksize for the filesystem is larger
1557                  * than the hardware sectorsize for the machine.
1558                  */
1559                 if (blocksize < hblock) {
1560                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1561                                "device blocksize %d.\n", blocksize, hblock);
1562                         goto failed_mount;
1563                 }
1564
1565                 brelse (bh);
1566                 sb_set_blocksize(sb, blocksize);
1567                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1568                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1569                 bh = sb_bread(sb, logic_sb_block);
1570                 if (!bh) {
1571                         printk(KERN_ERR
1572                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1573                         goto failed_mount;
1574                 }
1575                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1576                 sbi->s_es = es;
1577                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1578                         printk (KERN_ERR
1579                                 "EXT3-fs: Magic mismatch, very weird !\n");
1580                         goto failed_mount;
1581                 }
1582         }
1583
1584         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1585
1586         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1587                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1588                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1589         } else {
1590                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1591                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1592                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1593                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1594                     (sbi->s_inode_size > blocksize)) {
1595                         printk (KERN_ERR
1596                                 "EXT3-fs: unsupported inode size: %d\n",
1597                                 sbi->s_inode_size);
1598                         goto failed_mount;
1599                 }
1600         }
1601         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1602                                    le32_to_cpu(es->s_log_frag_size);
1603         if (blocksize != sbi->s_frag_size) {
1604                 printk(KERN_ERR
1605                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1606                        sbi->s_frag_size, blocksize);
1607                 goto failed_mount;
1608         }
1609         sbi->s_frags_per_block = 1;
1610         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1611         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1612         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1613         if (EXT3_INODE_SIZE(sb) == 0)
1614                 goto cantfind_ext3;
1615         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1616         if (sbi->s_inodes_per_block == 0)
1617                 goto cantfind_ext3;
1618         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1619                                         sbi->s_inodes_per_block;
1620         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1621         sbi->s_sbh = bh;
1622         sbi->s_mount_state = le16_to_cpu(es->s_state);
1623         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1624         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1625         for (i=0; i < 4; i++)
1626                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1627         sbi->s_def_hash_version = es->s_def_hash_version;
1628
1629         if (sbi->s_blocks_per_group > blocksize * 8) {
1630                 printk (KERN_ERR
1631                         "EXT3-fs: #blocks per group too big: %lu\n",
1632                         sbi->s_blocks_per_group);
1633                 goto failed_mount;
1634         }
1635         if (sbi->s_frags_per_group > blocksize * 8) {
1636                 printk (KERN_ERR
1637                         "EXT3-fs: #fragments per group too big: %lu\n",
1638                         sbi->s_frags_per_group);
1639                 goto failed_mount;
1640         }
1641         if (sbi->s_inodes_per_group > blocksize * 8) {
1642                 printk (KERN_ERR
1643                         "EXT3-fs: #inodes per group too big: %lu\n",
1644                         sbi->s_inodes_per_group);
1645                 goto failed_mount;
1646         }
1647
1648         if (le32_to_cpu(es->s_blocks_count) >
1649                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1650                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1651                         " too large to mount safely\n", sb->s_id);
1652                 if (sizeof(sector_t) < 8)
1653                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1654                                         "enabled\n");
1655                 goto failed_mount;
1656         }
1657
1658         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1659                 goto cantfind_ext3;
1660         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1661                                le32_to_cpu(es->s_first_data_block) - 1)
1662                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1663         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1664                    EXT3_DESC_PER_BLOCK(sb);
1665         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1666                                     GFP_KERNEL);
1667         if (sbi->s_group_desc == NULL) {
1668                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1669                 goto failed_mount;
1670         }
1671
1672         bgl_lock_init(&sbi->s_blockgroup_lock);
1673
1674         for (i = 0; i < db_count; i++) {
1675                 block = descriptor_loc(sb, logic_sb_block, i);
1676                 sbi->s_group_desc[i] = sb_bread(sb, block);
1677                 if (!sbi->s_group_desc[i]) {
1678                         printk (KERN_ERR "EXT3-fs: "
1679                                 "can't read group descriptor %d\n", i);
1680                         db_count = i;
1681                         goto failed_mount2;
1682                 }
1683         }
1684         if (!ext3_check_descriptors (sb)) {
1685                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1686                 goto failed_mount2;
1687         }
1688         sbi->s_gdb_count = db_count;
1689         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1690         spin_lock_init(&sbi->s_next_gen_lock);
1691
1692         percpu_counter_init(&sbi->s_freeblocks_counter,
1693                 ext3_count_free_blocks(sb));
1694         percpu_counter_init(&sbi->s_freeinodes_counter,
1695                 ext3_count_free_inodes(sb));
1696         percpu_counter_init(&sbi->s_dirs_counter,
1697                 ext3_count_dirs(sb));
1698
1699         /* per fileystem reservation list head & lock */
1700         spin_lock_init(&sbi->s_rsv_window_lock);
1701         sbi->s_rsv_window_root = RB_ROOT;
1702         /* Add a single, static dummy reservation to the start of the
1703          * reservation window list --- it gives us a placeholder for
1704          * append-at-start-of-list which makes the allocation logic
1705          * _much_ simpler. */
1706         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1707         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1708         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1709         sbi->s_rsv_window_head.rsv_goal_size = 0;
1710         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1711
1712         /*
1713          * set up enough so that it can read an inode
1714          */
1715         sb->s_op = &ext3_sops;
1716         sb->s_export_op = &ext3_export_ops;
1717         sb->s_xattr = ext3_xattr_handlers;
1718 #ifdef CONFIG_QUOTA
1719         sb->s_qcop = &ext3_qctl_operations;
1720         sb->dq_op = &ext3_quota_operations;
1721 #endif
1722         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1723
1724         sb->s_root = NULL;
1725
1726         needs_recovery = (es->s_last_orphan != 0 ||
1727                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1728                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1729
1730         /*
1731          * The first inode we look at is the journal inode.  Don't try
1732          * root first: it may be modified in the journal!
1733          */
1734         if (!test_opt(sb, NOLOAD) &&
1735             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1736                 if (ext3_load_journal(sb, es, journal_devnum))
1737                         goto failed_mount3;
1738         } else if (journal_inum) {
1739                 if (ext3_create_journal(sb, es, journal_inum))
1740                         goto failed_mount3;
1741         } else {
1742                 if (!silent)
1743                         printk (KERN_ERR
1744                                 "ext3: No journal on filesystem on %s\n",
1745                                 sb->s_id);
1746                 goto failed_mount3;
1747         }
1748
1749         /* We have now updated the journal if required, so we can
1750          * validate the data journaling mode. */
1751         switch (test_opt(sb, DATA_FLAGS)) {
1752         case 0:
1753                 /* No mode set, assume a default based on the journal
1754                    capabilities: ORDERED_DATA if the journal can
1755                    cope, else JOURNAL_DATA */
1756                 if (journal_check_available_features
1757                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1758                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1759                 else
1760                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1761                 break;
1762
1763         case EXT3_MOUNT_ORDERED_DATA:
1764         case EXT3_MOUNT_WRITEBACK_DATA:
1765                 if (!journal_check_available_features
1766                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1767                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1768                                "requested data journaling mode\n");
1769                         goto failed_mount4;
1770                 }
1771         default:
1772                 break;
1773         }
1774
1775         if (test_opt(sb, NOBH)) {
1776                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1777                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1778                                 "its supported only with writeback mode\n");
1779                         clear_opt(sbi->s_mount_opt, NOBH);
1780                 }
1781         }
1782         /*
1783          * The journal_load will have done any necessary log recovery,
1784          * so we can safely mount the rest of the filesystem now.
1785          */
1786
1787         root = iget(sb, EXT3_ROOT_INO);
1788         sb->s_root = d_alloc_root(root);
1789         if (!sb->s_root) {
1790                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1791                 iput(root);
1792                 goto failed_mount4;
1793         }
1794         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1795                 dput(sb->s_root);
1796                 sb->s_root = NULL;
1797                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1798                 goto failed_mount4;
1799         }
1800
1801         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1802         /*
1803          * akpm: core read_super() calls in here with the superblock locked.
1804          * That deadlocks, because orphan cleanup needs to lock the superblock
1805          * in numerous places.  Here we just pop the lock - it's relatively
1806          * harmless, because we are now ready to accept write_super() requests,
1807          * and aviro says that's the only reason for hanging onto the
1808          * superblock lock.
1809          */
1810         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1811         ext3_orphan_cleanup(sb, es);
1812         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1813         if (needs_recovery)
1814                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1815         ext3_mark_recovery_complete(sb, es);
1816         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1817                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1818                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1819                 "writeback");
1820
1821         lock_kernel();
1822         return 0;
1823
1824 cantfind_ext3:
1825         if (!silent)
1826                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1827                        sb->s_id);
1828         goto failed_mount;
1829
1830 failed_mount4:
1831         journal_destroy(sbi->s_journal);
1832 failed_mount3:
1833         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1834         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1835         percpu_counter_destroy(&sbi->s_dirs_counter);
1836 failed_mount2:
1837         for (i = 0; i < db_count; i++)
1838                 brelse(sbi->s_group_desc[i]);
1839         kfree(sbi->s_group_desc);
1840 failed_mount:
1841 #ifdef CONFIG_QUOTA
1842         for (i = 0; i < MAXQUOTAS; i++)
1843                 kfree(sbi->s_qf_names[i]);
1844 #endif
1845         ext3_blkdev_remove(sbi);
1846         brelse(bh);
1847 out_fail:
1848         sb->s_fs_info = NULL;
1849         kfree(sbi);
1850         lock_kernel();
1851         return -EINVAL;
1852 }
1853
1854 /*
1855  * Setup any per-fs journal parameters now.  We'll do this both on
1856  * initial mount, once the journal has been initialised but before we've
1857  * done any recovery; and again on any subsequent remount.
1858  */
1859 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1860 {
1861         struct ext3_sb_info *sbi = EXT3_SB(sb);
1862
1863         if (sbi->s_commit_interval)
1864                 journal->j_commit_interval = sbi->s_commit_interval;
1865         /* We could also set up an ext3-specific default for the commit
1866          * interval here, but for now we'll just fall back to the jbd
1867          * default. */
1868
1869         spin_lock(&journal->j_state_lock);
1870         if (test_opt(sb, BARRIER))
1871                 journal->j_flags |= JFS_BARRIER;
1872         else
1873                 journal->j_flags &= ~JFS_BARRIER;
1874         spin_unlock(&journal->j_state_lock);
1875 }
1876
1877 static journal_t *ext3_get_journal(struct super_block *sb,
1878                                    unsigned int journal_inum)
1879 {
1880         struct inode *journal_inode;
1881         journal_t *journal;
1882
1883         /* First, test for the existence of a valid inode on disk.  Bad
1884          * things happen if we iget() an unused inode, as the subsequent
1885          * iput() will try to delete it. */
1886
1887         journal_inode = iget(sb, journal_inum);
1888         if (!journal_inode) {
1889                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1890                 return NULL;
1891         }
1892         if (!journal_inode->i_nlink) {
1893                 make_bad_inode(journal_inode);
1894                 iput(journal_inode);
1895                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1896                 return NULL;
1897         }
1898
1899         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1900                   journal_inode, journal_inode->i_size);
1901         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1902                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1903                 iput(journal_inode);
1904                 return NULL;
1905         }
1906
1907         journal = journal_init_inode(journal_inode);
1908         if (!journal) {
1909                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1910                 iput(journal_inode);
1911                 return NULL;
1912         }
1913         journal->j_private = sb;
1914         ext3_init_journal_params(sb, journal);
1915         return journal;
1916 }
1917
1918 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1919                                        dev_t j_dev)
1920 {
1921         struct buffer_head * bh;
1922         journal_t *journal;
1923         ext3_fsblk_t start;
1924         ext3_fsblk_t len;
1925         int hblock, blocksize;
1926         ext3_fsblk_t sb_block;
1927         unsigned long offset;
1928         struct ext3_super_block * es;
1929         struct block_device *bdev;
1930
1931         bdev = ext3_blkdev_get(j_dev);
1932         if (bdev == NULL)
1933                 return NULL;
1934
1935         if (bd_claim(bdev, sb)) {
1936                 printk(KERN_ERR
1937                         "EXT3: failed to claim external journal device.\n");
1938                 blkdev_put(bdev);
1939                 return NULL;
1940         }
1941
1942         blocksize = sb->s_blocksize;
1943         hblock = bdev_hardsect_size(bdev);
1944         if (blocksize < hblock) {
1945                 printk(KERN_ERR
1946                         "EXT3-fs: blocksize too small for journal device.\n");
1947                 goto out_bdev;
1948         }
1949
1950         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1951         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1952         set_blocksize(bdev, blocksize);
1953         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1954                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1955                        "external journal\n");
1956                 goto out_bdev;
1957         }
1958
1959         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1960         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1961             !(le32_to_cpu(es->s_feature_incompat) &
1962               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1963                 printk(KERN_ERR "EXT3-fs: external journal has "
1964                                         "bad superblock\n");
1965                 brelse(bh);
1966                 goto out_bdev;
1967         }
1968
1969         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1970                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1971                 brelse(bh);
1972                 goto out_bdev;
1973         }
1974
1975         len = le32_to_cpu(es->s_blocks_count);
1976         start = sb_block + 1;
1977         brelse(bh);     /* we're done with the superblock */
1978
1979         journal = journal_init_dev(bdev, sb->s_bdev,
1980                                         start, len, blocksize);
1981         if (!journal) {
1982                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1983                 goto out_bdev;
1984         }
1985         journal->j_private = sb;
1986         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1987         wait_on_buffer(journal->j_sb_buffer);
1988         if (!buffer_uptodate(journal->j_sb_buffer)) {
1989                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1990                 goto out_journal;
1991         }
1992         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1993                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1994                                         "user (unsupported) - %d\n",
1995                         be32_to_cpu(journal->j_superblock->s_nr_users));
1996                 goto out_journal;
1997         }
1998         EXT3_SB(sb)->journal_bdev = bdev;
1999         ext3_init_journal_params(sb, journal);
2000         return journal;
2001 out_journal:
2002         journal_destroy(journal);
2003 out_bdev:
2004         ext3_blkdev_put(bdev);
2005         return NULL;
2006 }
2007
2008 static int ext3_load_journal(struct super_block *sb,
2009                              struct ext3_super_block *es,
2010                              unsigned long journal_devnum)
2011 {
2012         journal_t *journal;
2013         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2014         dev_t journal_dev;
2015         int err = 0;
2016         int really_read_only;
2017
2018         if (journal_devnum &&
2019             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2020                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2021                         "numbers have changed\n");
2022                 journal_dev = new_decode_dev(journal_devnum);
2023         } else
2024                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2025
2026         really_read_only = bdev_read_only(sb->s_bdev);
2027
2028         /*
2029          * Are we loading a blank journal or performing recovery after a
2030          * crash?  For recovery, we need to check in advance whether we
2031          * can get read-write access to the device.
2032          */
2033
2034         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2035                 if (sb->s_flags & MS_RDONLY) {
2036                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2037                                         "required on readonly filesystem.\n");
2038                         if (really_read_only) {
2039                                 printk(KERN_ERR "EXT3-fs: write access "
2040                                         "unavailable, cannot proceed.\n");
2041                                 return -EROFS;
2042                         }
2043                         printk (KERN_INFO "EXT3-fs: write access will "
2044                                         "be enabled during recovery.\n");
2045                 }
2046         }
2047
2048         if (journal_inum && journal_dev) {
2049                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2050                        "and inode journals!\n");
2051                 return -EINVAL;
2052         }
2053
2054         if (journal_inum) {
2055                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2056                         return -EINVAL;
2057         } else {
2058                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2059                         return -EINVAL;
2060         }
2061
2062         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2063                 err = journal_update_format(journal);
2064                 if (err)  {
2065                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2066                         journal_destroy(journal);
2067                         return err;
2068                 }
2069         }
2070
2071         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2072                 err = journal_wipe(journal, !really_read_only);
2073         if (!err)
2074                 err = journal_load(journal);
2075
2076         if (err) {
2077                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2078                 journal_destroy(journal);
2079                 return err;
2080         }
2081
2082         EXT3_SB(sb)->s_journal = journal;
2083         ext3_clear_journal_err(sb, es);
2084
2085         if (journal_devnum &&
2086             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2087                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2088                 sb->s_dirt = 1;
2089
2090                 /* Make sure we flush the recovery flag to disk. */
2091                 ext3_commit_super(sb, es, 1);
2092         }
2093
2094         return 0;
2095 }
2096
2097 static int ext3_create_journal(struct super_block * sb,
2098                                struct ext3_super_block * es,
2099                                unsigned int journal_inum)
2100 {
2101         journal_t *journal;
2102
2103         if (sb->s_flags & MS_RDONLY) {
2104                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2105                                 "create journal.\n");
2106                 return -EROFS;
2107         }
2108
2109         if (!(journal = ext3_get_journal(sb, journal_inum)))
2110                 return -EINVAL;
2111
2112         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2113                journal_inum);
2114
2115         if (journal_create(journal)) {
2116                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2117                 journal_destroy(journal);
2118                 return -EIO;
2119         }
2120
2121         EXT3_SB(sb)->s_journal = journal;
2122
2123         ext3_update_dynamic_rev(sb);
2124         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2125         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2126
2127         es->s_journal_inum = cpu_to_le32(journal_inum);
2128         sb->s_dirt = 1;
2129
2130         /* Make sure we flush the recovery flag to disk. */
2131         ext3_commit_super(sb, es, 1);
2132
2133         return 0;
2134 }
2135
2136 static void ext3_commit_super (struct super_block * sb,
2137                                struct ext3_super_block * es,
2138                                int sync)
2139 {
2140         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2141
2142         if (!sbh)
2143                 return;
2144         es->s_wtime = cpu_to_le32(get_seconds());
2145         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2146         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2147         BUFFER_TRACE(sbh, "marking dirty");
2148         mark_buffer_dirty(sbh);
2149         if (sync)
2150                 sync_dirty_buffer(sbh);
2151 }
2152
2153
2154 /*
2155  * Have we just finished recovery?  If so, and if we are mounting (or
2156  * remounting) the filesystem readonly, then we will end up with a
2157  * consistent fs on disk.  Record that fact.
2158  */
2159 static void ext3_mark_recovery_complete(struct super_block * sb,
2160                                         struct ext3_super_block * es)
2161 {
2162         journal_t *journal = EXT3_SB(sb)->s_journal;
2163
2164         journal_lock_updates(journal);
2165         journal_flush(journal);
2166         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2167             sb->s_flags & MS_RDONLY) {
2168                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2169                 sb->s_dirt = 0;
2170                 ext3_commit_super(sb, es, 1);
2171         }
2172         journal_unlock_updates(journal);
2173 }
2174
2175 /*
2176  * If we are mounting (or read-write remounting) a filesystem whose journal
2177  * has recorded an error from a previous lifetime, move that error to the
2178  * main filesystem now.
2179  */
2180 static void ext3_clear_journal_err(struct super_block * sb,
2181                                    struct ext3_super_block * es)
2182 {
2183         journal_t *journal;
2184         int j_errno;
2185         const char *errstr;
2186
2187         journal = EXT3_SB(sb)->s_journal;
2188
2189         /*
2190          * Now check for any error status which may have been recorded in the
2191          * journal by a prior ext3_error() or ext3_abort()
2192          */
2193
2194         j_errno = journal_errno(journal);
2195         if (j_errno) {
2196                 char nbuf[16];
2197
2198                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2199                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2200                              "from previous mount: %s", errstr);
2201                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2202                              "filesystem check.");
2203
2204                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2205                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2206                 ext3_commit_super (sb, es, 1);
2207
2208                 journal_clear_err(journal);
2209         }
2210 }
2211
2212 /*
2213  * Force the running and committing transactions to commit,
2214  * and wait on the commit.
2215  */
2216 int ext3_force_commit(struct super_block *sb)
2217 {
2218         journal_t *journal;
2219         int ret;
2220
2221         if (sb->s_flags & MS_RDONLY)
2222                 return 0;
2223
2224         journal = EXT3_SB(sb)->s_journal;
2225         sb->s_dirt = 0;
2226         ret = ext3_journal_force_commit(journal);
2227         return ret;
2228 }
2229
2230 /*
2231  * Ext3 always journals updates to the superblock itself, so we don't
2232  * have to propagate any other updates to the superblock on disk at this
2233  * point.  Just start an async writeback to get the buffers on their way
2234  * to the disk.
2235  *
2236  * This implicitly triggers the writebehind on sync().
2237  */
2238
2239 static void ext3_write_super (struct super_block * sb)
2240 {
2241         if (mutex_trylock(&sb->s_lock) != 0)
2242                 BUG();
2243         sb->s_dirt = 0;
2244 }
2245
2246 static int ext3_sync_fs(struct super_block *sb, int wait)
2247 {
2248         tid_t target;
2249
2250         sb->s_dirt = 0;
2251         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2252                 if (wait)
2253                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2254         }
2255         return 0;
2256 }
2257
2258 /*
2259  * LVM calls this function before a (read-only) snapshot is created.  This
2260  * gives us a chance to flush the journal completely and mark the fs clean.
2261  */
2262 static void ext3_write_super_lockfs(struct super_block *sb)
2263 {
2264         sb->s_dirt = 0;
2265
2266         if (!(sb->s_flags & MS_RDONLY)) {
2267                 journal_t *journal = EXT3_SB(sb)->s_journal;
2268
2269                 /* Now we set up the journal barrier. */
2270                 journal_lock_updates(journal);
2271                 journal_flush(journal);
2272
2273                 /* Journal blocked and flushed, clear needs_recovery flag. */
2274                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2275                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2276         }
2277 }
2278
2279 /*
2280  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2281  * flag here, even though the filesystem is not technically dirty yet.
2282  */
2283 static void ext3_unlockfs(struct super_block *sb)
2284 {
2285         if (!(sb->s_flags & MS_RDONLY)) {
2286                 lock_super(sb);
2287                 /* Reser the needs_recovery flag before the fs is unlocked. */
2288                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2289                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2290                 unlock_super(sb);
2291                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2292         }
2293 }
2294
2295 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2296 {
2297         struct ext3_super_block * es;
2298         struct ext3_sb_info *sbi = EXT3_SB(sb);
2299         ext3_fsblk_t n_blocks_count = 0;
2300         unsigned long old_sb_flags;
2301         struct ext3_mount_options old_opts;
2302         int err;
2303 #ifdef CONFIG_QUOTA
2304         int i;
2305 #endif
2306
2307         /* Store the original options */
2308         old_sb_flags = sb->s_flags;
2309         old_opts.s_mount_opt = sbi->s_mount_opt;
2310         old_opts.s_resuid = sbi->s_resuid;
2311         old_opts.s_resgid = sbi->s_resgid;
2312         old_opts.s_commit_interval = sbi->s_commit_interval;
2313 #ifdef CONFIG_QUOTA
2314         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2315         for (i = 0; i < MAXQUOTAS; i++)
2316                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2317 #endif
2318
2319         /*
2320          * Allow the "check" option to be passed as a remount option.
2321          */
2322         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2323                 err = -EINVAL;
2324                 goto restore_opts;
2325         }
2326
2327         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2328                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2329         if ((sbi->s_mount_opt & EXT3_MOUNT_TAGGED) &&
2330                 !(sb->s_flags & MS_TAGGED)) {
2331                 printk("EXT3-fs: %s: tagging not permitted on remount.\n",
2332                         sb->s_id);
2333                 return -EINVAL;
2334         }
2335
2336         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2337                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2338
2339         es = sbi->s_es;
2340
2341         ext3_init_journal_params(sb, sbi->s_journal);
2342
2343         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2344                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2345                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2346                         err = -EROFS;
2347                         goto restore_opts;
2348                 }
2349
2350                 if (*flags & MS_RDONLY) {
2351                         /*
2352                          * First of all, the unconditional stuff we have to do
2353                          * to disable replay of the journal when we next remount
2354                          */
2355                         sb->s_flags |= MS_RDONLY;
2356
2357                         /*
2358                          * OK, test if we are remounting a valid rw partition
2359                          * readonly, and if so set the rdonly flag and then
2360                          * mark the partition as valid again.
2361                          */
2362                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2363                             (sbi->s_mount_state & EXT3_VALID_FS))
2364                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2365
2366                         ext3_mark_recovery_complete(sb, es);
2367                 } else {
2368                         __le32 ret;
2369                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2370                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2371                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2372                                        "remount RDWR because of unsupported "
2373                                        "optional features (%x).\n",
2374                                        sb->s_id, le32_to_cpu(ret));
2375                                 err = -EROFS;
2376                                 goto restore_opts;
2377                         }
2378                         /*
2379                          * Mounting a RDONLY partition read-write, so reread
2380                          * and store the current valid flag.  (It may have
2381                          * been changed by e2fsck since we originally mounted
2382                          * the partition.)
2383                          */
2384                         ext3_clear_journal_err(sb, es);
2385                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2386                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2387                                 goto restore_opts;
2388                         if (!ext3_setup_super (sb, es, 0))
2389                                 sb->s_flags &= ~MS_RDONLY;
2390                 }
2391         }
2392 #ifdef CONFIG_QUOTA
2393         /* Release old quota file names */
2394         for (i = 0; i < MAXQUOTAS; i++)
2395                 if (old_opts.s_qf_names[i] &&
2396                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2397                         kfree(old_opts.s_qf_names[i]);
2398 #endif
2399         return 0;
2400 restore_opts:
2401         sb->s_flags = old_sb_flags;
2402         sbi->s_mount_opt = old_opts.s_mount_opt;
2403         sbi->s_resuid = old_opts.s_resuid;
2404         sbi->s_resgid = old_opts.s_resgid;
2405         sbi->s_commit_interval = old_opts.s_commit_interval;
2406 #ifdef CONFIG_QUOTA
2407         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2408         for (i = 0; i < MAXQUOTAS; i++) {
2409                 if (sbi->s_qf_names[i] &&
2410                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2411                         kfree(sbi->s_qf_names[i]);
2412                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2413         }
2414 #endif
2415         return err;
2416 }
2417
2418 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2419 {
2420         struct super_block *sb = dentry->d_sb;
2421         struct ext3_sb_info *sbi = EXT3_SB(sb);
2422         struct ext3_super_block *es = sbi->s_es;
2423         ext3_fsblk_t overhead;
2424         int i;
2425         u64 fsid;
2426
2427         if (test_opt (sb, MINIX_DF))
2428                 overhead = 0;
2429         else {
2430                 unsigned long ngroups;
2431                 ngroups = EXT3_SB(sb)->s_groups_count;
2432                 smp_rmb();
2433
2434                 /*
2435                  * Compute the overhead (FS structures)
2436                  */
2437
2438                 /*
2439                  * All of the blocks before first_data_block are
2440                  * overhead
2441                  */
2442                 overhead = le32_to_cpu(es->s_first_data_block);
2443
2444                 /*
2445                  * Add the overhead attributed to the superblock and
2446                  * block group descriptors.  If the sparse superblocks
2447                  * feature is turned on, then not all groups have this.
2448                  */
2449                 for (i = 0; i < ngroups; i++) {
2450                         overhead += ext3_bg_has_super(sb, i) +
2451                                 ext3_bg_num_gdb(sb, i);
2452                         cond_resched();
2453                 }
2454
2455                 /*
2456                  * Every block group has an inode bitmap, a block
2457                  * bitmap, and an inode table.
2458                  */
2459                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2460         }
2461
2462         buf->f_type = EXT3_SUPER_MAGIC;
2463         buf->f_bsize = sb->s_blocksize;
2464         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2465         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2466         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2467         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2468                 buf->f_bavail = 0;
2469         buf->f_files = le32_to_cpu(es->s_inodes_count);
2470         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2471         buf->f_namelen = EXT3_NAME_LEN;
2472         fsid = le64_to_cpup((void *)es->s_uuid) ^
2473                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2474         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2475         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2476         return 0;
2477 }
2478
2479 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2480  * is locked for write. Otherwise the are possible deadlocks:
2481  * Process 1                         Process 2
2482  * ext3_create()                     quota_sync()
2483  *   journal_start()                   write_dquot()
2484  *   DQUOT_INIT()                        down(dqio_mutex)
2485  *     down(dqio_mutex)                    journal_start()
2486  *
2487  */
2488
2489 #ifdef CONFIG_QUOTA
2490
2491 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2492 {
2493         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2494 }
2495
2496 static int ext3_dquot_initialize(struct inode *inode, int type)
2497 {
2498         handle_t *handle;
2499         int ret, err;
2500
2501         /* We may create quota structure so we need to reserve enough blocks */
2502         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2503         if (IS_ERR(handle))
2504                 return PTR_ERR(handle);
2505         ret = dquot_initialize(inode, type);
2506         err = ext3_journal_stop(handle);
2507         if (!ret)
2508                 ret = err;
2509         return ret;
2510 }
2511
2512 static int ext3_dquot_drop(struct inode *inode)
2513 {
2514         handle_t *handle;
2515         int ret, err;
2516
2517         /* We may delete quota structure so we need to reserve enough blocks */
2518         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2519         if (IS_ERR(handle))
2520                 return PTR_ERR(handle);
2521         ret = dquot_drop(inode);
2522         err = ext3_journal_stop(handle);
2523         if (!ret)
2524                 ret = err;
2525         return ret;
2526 }
2527
2528 static int ext3_write_dquot(struct dquot *dquot)
2529 {
2530         int ret, err;
2531         handle_t *handle;
2532         struct inode *inode;
2533
2534         inode = dquot_to_inode(dquot);
2535         handle = ext3_journal_start(inode,
2536                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2537         if (IS_ERR(handle))
2538                 return PTR_ERR(handle);
2539         ret = dquot_commit(dquot);
2540         err = ext3_journal_stop(handle);
2541         if (!ret)
2542                 ret = err;
2543         return ret;
2544 }
2545
2546 static int ext3_acquire_dquot(struct dquot *dquot)
2547 {
2548         int ret, err;
2549         handle_t *handle;
2550
2551         handle = ext3_journal_start(dquot_to_inode(dquot),
2552                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2553         if (IS_ERR(handle))
2554                 return PTR_ERR(handle);
2555         ret = dquot_acquire(dquot);
2556         err = ext3_journal_stop(handle);
2557         if (!ret)
2558                 ret = err;
2559         return ret;
2560 }
2561
2562 static int ext3_release_dquot(struct dquot *dquot)
2563 {
2564         int ret, err;
2565         handle_t *handle;
2566
2567         handle = ext3_journal_start(dquot_to_inode(dquot),
2568                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2569         if (IS_ERR(handle))
2570                 return PTR_ERR(handle);
2571         ret = dquot_release(dquot);
2572         err = ext3_journal_stop(handle);
2573         if (!ret)
2574                 ret = err;
2575         return ret;
2576 }
2577
2578 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2579 {
2580         /* Are we journalling quotas? */
2581         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2582             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2583                 dquot_mark_dquot_dirty(dquot);
2584                 return ext3_write_dquot(dquot);
2585         } else {
2586                 return dquot_mark_dquot_dirty(dquot);
2587         }
2588 }
2589
2590 static int ext3_write_info(struct super_block *sb, int type)
2591 {
2592         int ret, err;
2593         handle_t *handle;
2594
2595         /* Data block + inode block */
2596         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2597         if (IS_ERR(handle))
2598                 return PTR_ERR(handle);
2599         ret = dquot_commit_info(sb, type);
2600         err = ext3_journal_stop(handle);
2601         if (!ret)
2602                 ret = err;
2603         return ret;
2604 }
2605
2606 /*
2607  * Turn on quotas during mount time - we need to find
2608  * the quota file and such...
2609  */
2610 static int ext3_quota_on_mount(struct super_block *sb, int type)
2611 {
2612         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2613                         EXT3_SB(sb)->s_jquota_fmt, type);
2614 }
2615
2616 /*
2617  * Standard function to be called on quota_on
2618  */
2619 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2620                          char *path)
2621 {
2622         int err;
2623         struct nameidata nd;
2624
2625         if (!test_opt(sb, QUOTA))
2626                 return -EINVAL;
2627         /* Not journalling quota? */
2628         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2629             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2630                 return vfs_quota_on(sb, type, format_id, path);
2631         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2632         if (err)
2633                 return err;
2634         /* Quotafile not on the same filesystem? */
2635         if (nd.mnt->mnt_sb != sb) {
2636                 path_release(&nd);
2637                 return -EXDEV;
2638         }
2639         /* Quotafile not of fs root? */
2640         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2641                 printk(KERN_WARNING
2642                         "EXT3-fs: Quota file not on filesystem root. "
2643                         "Journalled quota will not work.\n");
2644         path_release(&nd);
2645         return vfs_quota_on(sb, type, format_id, path);
2646 }
2647
2648 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2649  * acquiring the locks... As quota files are never truncated and quota code
2650  * itself serializes the operations (and noone else should touch the files)
2651  * we don't have to be afraid of races */
2652 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2653                                size_t len, loff_t off)
2654 {
2655         struct inode *inode = sb_dqopt(sb)->files[type];
2656         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2657         int err = 0;
2658         int offset = off & (sb->s_blocksize - 1);
2659         int tocopy;
2660         size_t toread;
2661         struct buffer_head *bh;
2662         loff_t i_size = i_size_read(inode);
2663
2664         if (off > i_size)
2665                 return 0;
2666         if (off+len > i_size)
2667                 len = i_size-off;
2668         toread = len;
2669         while (toread > 0) {
2670                 tocopy = sb->s_blocksize - offset < toread ?
2671                                 sb->s_blocksize - offset : toread;
2672                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2673                 if (err)
2674                         return err;
2675                 if (!bh)        /* A hole? */
2676                         memset(data, 0, tocopy);
2677                 else
2678                         memcpy(data, bh->b_data+offset, tocopy);
2679                 brelse(bh);
2680                 offset = 0;
2681                 toread -= tocopy;
2682                 data += tocopy;
2683                 blk++;
2684         }
2685         return len;
2686 }
2687
2688 /* Write to quotafile (we know the transaction is already started and has
2689  * enough credits) */
2690 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2691                                 const char *data, size_t len, loff_t off)
2692 {
2693         struct inode *inode = sb_dqopt(sb)->files[type];
2694         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2695         int err = 0;
2696         int offset = off & (sb->s_blocksize - 1);
2697         int tocopy;
2698         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2699         size_t towrite = len;
2700         struct buffer_head *bh;
2701         handle_t *handle = journal_current_handle();
2702
2703         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2704         while (towrite > 0) {
2705                 tocopy = sb->s_blocksize - offset < towrite ?
2706                                 sb->s_blocksize - offset : towrite;
2707                 bh = ext3_bread(handle, inode, blk, 1, &err);
2708                 if (!bh)
2709                         goto out;
2710                 if (journal_quota) {
2711                         err = ext3_journal_get_write_access(handle, bh);
2712                         if (err) {
2713                                 brelse(bh);
2714                                 goto out;
2715                         }
2716                 }
2717                 lock_buffer(bh);
2718                 memcpy(bh->b_data+offset, data, tocopy);
2719                 flush_dcache_page(bh->b_page);
2720                 unlock_buffer(bh);
2721                 if (journal_quota)
2722                         err = ext3_journal_dirty_metadata(handle, bh);
2723                 else {
2724                         /* Always do at least ordered writes for quotas */
2725                         err = ext3_journal_dirty_data(handle, bh);
2726                         mark_buffer_dirty(bh);
2727                 }
2728                 brelse(bh);
2729                 if (err)
2730                         goto out;
2731                 offset = 0;
2732                 towrite -= tocopy;
2733                 data += tocopy;
2734                 blk++;
2735         }
2736 out:
2737         if (len == towrite)
2738                 return err;
2739         if (inode->i_size < off+len-towrite) {
2740                 i_size_write(inode, off+len-towrite);
2741                 EXT3_I(inode)->i_disksize = inode->i_size;
2742         }
2743         inode->i_version++;
2744         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2745         ext3_mark_inode_dirty(handle, inode);
2746         mutex_unlock(&inode->i_mutex);
2747         return len - towrite;
2748 }
2749
2750 #endif
2751
2752 static int ext3_get_sb(struct file_system_type *fs_type,
2753         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2754 {
2755         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2756 }
2757
2758 static struct file_system_type ext3_fs_type = {
2759         .owner          = THIS_MODULE,
2760         .name           = "ext3",
2761         .get_sb         = ext3_get_sb,
2762         .kill_sb        = kill_block_super,
2763         .fs_flags       = FS_REQUIRES_DEV,
2764 };
2765
2766 static int __init init_ext3_fs(void)
2767 {
2768         int err = init_ext3_xattr();
2769         if (err)
2770                 return err;
2771         err = init_inodecache();
2772         if (err)
2773                 goto out1;
2774         err = register_filesystem(&ext3_fs_type);
2775         if (err)
2776                 goto out;
2777         return 0;
2778 out:
2779         destroy_inodecache();
2780 out1:
2781         exit_ext3_xattr();
2782         return err;
2783 }
2784
2785 static void __exit exit_ext3_fs(void)
2786 {
2787         unregister_filesystem(&ext3_fs_type);
2788         destroy_inodecache();
2789         exit_ext3_xattr();
2790 }
2791
2792 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2793 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2794 MODULE_LICENSE("GPL");
2795 module_init(init_ext3_fs)
2796 module_exit(exit_ext3_fs)