fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <asm/uaccess.h>
  40 #include "xattr.h"
  41 #include "acl.h"
  42
  43 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  44 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  45                                int);
  46 static void ext3_commit_super (struct super_block * sb,
  47                                struct ext3_super_block * es,
  48                                int sync);
  49 static void ext3_mark_recovery_complete(struct super_block * sb,
  50                                         struct ext3_super_block * es);
  51 static void ext3_clear_journal_err(struct super_block * sb,
  52                                    struct ext3_super_block * es);
  53 static int ext3_sync_fs(struct super_block *sb, int wait);
  54 static const char *ext3_decode_error(struct super_block * sb, int errno,
  55                                      char nbuf[16]);
  56 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  57 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  58 static void ext3_unlockfs(struct super_block *sb);
  59 static void ext3_write_super (struct super_block * sb);
  60 static void ext3_write_super_lockfs(struct super_block *sb);
  61
  62 /*
  63  * Wrappers for journal_start/end.
  64  *
  65  * The only special thing we need to do here is to make sure that all
  66  * journal_end calls result in the superblock being marked dirty, so
  67  * that sync() will call the filesystem's write_super callback if
  68  * appropriate.
  69  */
  70 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  71 {
  72         journal_t *journal;
  73
  74         if (sb->s_flags & MS_RDONLY)
  75                 return ERR_PTR(-EROFS);
  76
  77         /* Special case here: if the journal has aborted behind our
  78          * backs (eg. EIO in the commit thread), then we still need to
  79          * take the FS itself readonly cleanly. */
  80         journal = EXT3_SB(sb)->s_journal;
  81         if (is_journal_aborted(journal)) {
  82                 ext3_abort(sb, __FUNCTION__,
  83                            "Detected aborted journal");
  84                 return ERR_PTR(-EROFS);
  85         }
  86
  87         return journal_start(journal, nblocks);
  88 }
  89
  90 /*
  91  * The only special thing we need to do here is to make sure that all
  92  * journal_stop calls result in the superblock being marked dirty, so
  93  * that sync() will call the filesystem's write_super callback if
  94  * appropriate.
  95  */
  96 int __ext3_journal_stop(const char *where, handle_t *handle)
  97 {
  98         struct super_block *sb;
  99         int err;
 100         int rc;
 101
 102         sb = handle->h_transaction->t_journal->j_private;
 103         err = handle->h_err;
 104         rc = journal_stop(handle);
 105
 106         if (!err)
 107                 err = rc;
 108         if (err)
 109                 __ext3_std_error(sb, where, err);
 110         return err;
 111 }
 112
 113 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 114                 struct buffer_head *bh, handle_t *handle, int err)
 115 {
 116         char nbuf[16];
 117         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 118
 119         if (bh)
 120                 BUFFER_TRACE(bh, "abort");
 121
 122         if (!handle->h_err)
 123                 handle->h_err = err;
 124
 125         if (is_handle_aborted(handle))
 126                 return;
 127
 128         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 129                caller, errstr, err_fn);
 130
 131         journal_abort_handle(handle);
 132 }
 133
 134 /* Deal with the reporting of failure conditions on a filesystem such as
 135  * inconsistencies detected or read IO failures.
 136  *
 137  * On ext2, we can store the error state of the filesystem in the
 138  * superblock.  That is not possible on ext3, because we may have other
 139  * write ordering constraints on the superblock which prevent us from
 140  * writing it out straight away; and given that the journal is about to
 141  * be aborted, we can't rely on the current, or future, transactions to
 142  * write out the superblock safely.
 143  *
 144  * We'll just use the journal_abort() error code to record an error in
 145  * the journal instead.  On recovery, the journal will compain about
 146  * that error until we've noted it down and cleared it.
 147  */
 148
 149 static void ext3_handle_error(struct super_block *sb)
 150 {
 151         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 152
 153         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 154         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 155
 156         if (sb->s_flags & MS_RDONLY)
 157                 return;
 158
 159         if (test_opt (sb, ERRORS_RO)) {
 160                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 161                 sb->s_flags |= MS_RDONLY;
 162         } else {
 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_PANIC))
 170                 panic("EXT3-fs (device %s): panic forced after error\n",
 171                         sb->s_id);
 172         ext3_commit_super(sb, es, 1);
 173 }
 174
 175 void ext3_error (struct super_block * sb, const char * function,
 176                  const char * fmt, ...)
 177 {
 178         va_list args;
 179
 180         va_start(args, fmt);
 181         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 182         vprintk(fmt, args);
 183         printk("\n");
 184         va_end(args);
 185
 186         ext3_handle_error(sb);
 187 }
 188
 189 static const char *ext3_decode_error(struct super_block * sb, int errno,
 190                                      char nbuf[16])
 191 {
 192         char *errstr = NULL;
 193
 194         switch (errno) {
 195         case -EIO:
 196                 errstr = "IO failure";
 197                 break;
 198         case -ENOMEM:
 199                 errstr = "Out of memory";
 200                 break;
 201         case -EROFS:
 202                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 203                         errstr = "Journal has aborted";
 204                 else
 205                         errstr = "Readonly filesystem";
 206                 break;
 207         default:
 208                 /* If the caller passed in an extra buffer for unknown
 209                  * errors, textualise them now.  Else we just return
 210                  * NULL. */
 211                 if (nbuf) {
 212                         /* Check for truncated error codes... */
 213                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 214                                 errstr = nbuf;
 215                 }
 216                 break;
 217         }
 218
 219         return errstr;
 220 }
 221
 222 /* __ext3_std_error decodes expected errors from journaling functions
 223  * automatically and invokes the appropriate error response.  */
 224
 225 void __ext3_std_error (struct super_block * sb, const char * function,
 226                        int errno)
 227 {
 228         char nbuf[16];
 229         const char *errstr;
 230
 231         /* Special case: if the error is EROFS, and we're not already
 232          * inside a transaction, then there's really no point in logging
 233          * an error. */
 234         if (errno == -EROFS && journal_current_handle() == NULL &&
 235             (sb->s_flags & MS_RDONLY))
 236                 return;
 237
 238         errstr = ext3_decode_error(sb, errno, nbuf);
 239         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 240                 sb->s_id, function, errstr);
 241
 242         ext3_handle_error(sb);
 243 }
 244
 245 /*
 246  * ext3_abort is a much stronger failure handler than ext3_error.  The
 247  * abort function may be used to deal with unrecoverable failures such
 248  * as journal IO errors or ENOMEM at a critical moment in log management.
 249  *
 250  * We unconditionally force the filesystem into an ABORT|READONLY state,
 251  * unless the error response on the fs has been set to panic in which
 252  * case we take the easy way out and panic immediately.
 253  */
 254
 255 void ext3_abort (struct super_block * sb, const char * function,
 256                  const char * fmt, ...)
 257 {
 258         va_list args;
 259
 260         printk (KERN_CRIT "ext3_abort called.\n");
 261
 262         va_start(args, fmt);
 263         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 264         vprintk(fmt, args);
 265         printk("\n");
 266         va_end(args);
 267
 268         if (test_opt(sb, ERRORS_PANIC))
 269                 panic("EXT3-fs panic from previous error\n");
 270
 271         if (sb->s_flags & MS_RDONLY)
 272                 return;
 273
 274         printk(KERN_CRIT "Remounting filesystem read-only\n");
 275         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 276         sb->s_flags |= MS_RDONLY;
 277         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 278         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 279 }
 280
 281 void ext3_warning (struct super_block * sb, const char * function,
 282                    const char * fmt, ...)
 283 {
 284         va_list args;
 285
 286         va_start(args, fmt);
 287         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 288                sb->s_id, function);
 289         vprintk(fmt, args);
 290         printk("\n");
 291         va_end(args);
 292 }
 293
 294 void ext3_update_dynamic_rev(struct super_block *sb)
 295 {
 296         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 297
 298         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 299                 return;
 300
 301         ext3_warning(sb, __FUNCTION__,
 302                      "updating to rev %d because of new feature flag, "
 303                      "running e2fsck is recommended",
 304                      EXT3_DYNAMIC_REV);
 305
 306         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 307         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 308         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 309         /* leave es->s_feature_*compat flags alone */
 310         /* es->s_uuid will be set by e2fsck if empty */
 311
 312         /*
 313          * The rest of the superblock fields should be zero, and if not it
 314          * means they are likely already in use, so leave them alone.  We
 315          * can leave it up to e2fsck to clean up any inconsistencies there.
 316          */
 317 }
 318
 319 /*
 320  * Open the external journal device
 321  */
 322 static struct block_device *ext3_blkdev_get(dev_t dev)
 323 {
 324         struct block_device *bdev;
 325         char b[BDEVNAME_SIZE];
 326
 327         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 328         if (IS_ERR(bdev))
 329                 goto fail;
 330         return bdev;
 331
 332 fail:
 333         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 334                         __bdevname(dev, b), PTR_ERR(bdev));
 335         return NULL;
 336 }
 337
 338 /*
 339  * Release the journal device
 340  */
 341 static int ext3_blkdev_put(struct block_device *bdev)
 342 {
 343         bd_release(bdev);
 344         return blkdev_put(bdev);
 345 }
 346
 347 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 348 {
 349         struct block_device *bdev;
 350         int ret = -ENODEV;
 351
 352         bdev = sbi->journal_bdev;
 353         if (bdev) {
 354                 ret = ext3_blkdev_put(bdev);
 355                 sbi->journal_bdev = NULL;
 356         }
 357         return ret;
 358 }
 359
 360 static inline struct inode *orphan_list_entry(struct list_head *l)
 361 {
 362         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 363 }
 364
 365 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 366 {
 367         struct list_head *l;
 368
 369         printk(KERN_ERR "sb orphan head is %d\n",
 370                le32_to_cpu(sbi->s_es->s_last_orphan));
 371
 372         printk(KERN_ERR "sb_info orphan list:\n");
 373         list_for_each(l, &sbi->s_orphan) {
 374                 struct inode *inode = orphan_list_entry(l);
 375                 printk(KERN_ERR "  "
 376                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 377                        inode->i_sb->s_id, inode->i_ino, inode,
 378                        inode->i_mode, inode->i_nlink,
 379                        NEXT_ORPHAN(inode));
 380         }
 381 }
 382
 383 static void ext3_put_super (struct super_block * sb)
 384 {
 385         struct ext3_sb_info *sbi = EXT3_SB(sb);
 386         struct ext3_super_block *es = sbi->s_es;
 387         int i;
 388
 389         ext3_xattr_put_super(sb);
 390         journal_destroy(sbi->s_journal);
 391         if (!(sb->s_flags & MS_RDONLY)) {
 392                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 393                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 394                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 395                 mark_buffer_dirty(sbi->s_sbh);
 396                 ext3_commit_super(sb, es, 1);
 397         }
 398
 399         for (i = 0; i < sbi->s_gdb_count; i++)
 400                 brelse(sbi->s_group_desc[i]);
 401         kfree(sbi->s_group_desc);
 402         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 403         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 404         percpu_counter_destroy(&sbi->s_dirs_counter);
 405         brelse(sbi->s_sbh);
 406 #ifdef CONFIG_QUOTA
 407         for (i = 0; i < MAXQUOTAS; i++)
 408                 kfree(sbi->s_qf_names[i]);
 409 #endif
 410
 411         /* Debugging code just in case the in-memory inode orphan list
 412          * isn't empty.  The on-disk one can be non-empty if we've
 413          * detected an error and taken the fs readonly, but the
 414          * in-memory list had better be clean by this point. */
 415         if (!list_empty(&sbi->s_orphan))
 416                 dump_orphan_list(sb, sbi);
 417         J_ASSERT(list_empty(&sbi->s_orphan));
 418
 419         invalidate_bdev(sb->s_bdev, 0);
 420         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 421                 /*
 422                  * Invalidate the journal device's buffers.  We don't want them
 423                  * floating about in memory - the physical journal device may
 424                  * hotswapped, and it breaks the `ro-after' testing code.
 425                  */
 426                 sync_blockdev(sbi->journal_bdev);
 427                 invalidate_bdev(sbi->journal_bdev, 0);
 428                 ext3_blkdev_remove(sbi);
 429         }
 430         sb->s_fs_info = NULL;
 431         kfree(sbi);
 432         return;
 433 }
 434
 435 static kmem_cache_t *ext3_inode_cachep;
 436
 437 /*
 438  * Called inside transaction, so use GFP_NOFS
 439  */
 440 static struct inode *ext3_alloc_inode(struct super_block *sb)
 441 {
 442         struct ext3_inode_info *ei;
 443
 444         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 445         if (!ei)
 446                 return NULL;
 447 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 448         ei->i_acl = EXT3_ACL_NOT_CACHED;
 449         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 450 #endif
 451         ei->i_block_alloc_info = NULL;
 452         ei->vfs_inode.i_version = 1;
 453         return &ei->vfs_inode;
 454 }
 455
 456 static void ext3_destroy_inode(struct inode *inode)
 457 {
 458         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 459 }
 460
 461 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 462 {
 463         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 464
 465         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 466             SLAB_CTOR_CONSTRUCTOR) {
 467                 INIT_LIST_HEAD(&ei->i_orphan);
 468 #ifdef CONFIG_EXT3_FS_XATTR
 469                 init_rwsem(&ei->xattr_sem);
 470 #endif
 471                 init_MUTEX(&ei->truncate_sem);
 472                 inode_init_once(&ei->vfs_inode);
 473         }
 474 }
 475
 476 static int init_inodecache(void)
 477 {
 478         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 479                                              sizeof(struct ext3_inode_info),
 480                                              0, SLAB_RECLAIM_ACCOUNT,
 481                                              init_once, NULL);
 482         if (ext3_inode_cachep == NULL)
 483                 return -ENOMEM;
 484         return 0;
 485 }
 486
 487 static void destroy_inodecache(void)
 488 {
 489         if (kmem_cache_destroy(ext3_inode_cachep))
 490                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 491 }
 492
 493 static void ext3_clear_inode(struct inode *inode)
 494 {
 495         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 497        if (EXT3_I(inode)->i_acl &&
 498            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 499                posix_acl_release(EXT3_I(inode)->i_acl);
 500                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 501        }
 502        if (EXT3_I(inode)->i_default_acl &&
 503            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 504                posix_acl_release(EXT3_I(inode)->i_default_acl);
 505                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 506        }
 507 #endif
 508         ext3_discard_reservation(inode);
 509         EXT3_I(inode)->i_block_alloc_info = NULL;
 510         kfree(rsv);
 511 }
 512
 513 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 514 {
 515 #if defined(CONFIG_QUOTA)
 516         struct ext3_sb_info *sbi = EXT3_SB(sb);
 517
 518         if (sbi->s_jquota_fmt)
 519                 seq_printf(seq, ",jqfmt=%s",
 520                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 521
 522         if (sbi->s_qf_names[USRQUOTA])
 523                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 524
 525         if (sbi->s_qf_names[GRPQUOTA])
 526                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 527
 528         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 529                 seq_puts(seq, ",usrquota");
 530
 531         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 532                 seq_puts(seq, ",grpquota");
 533 #endif
 534 }
 535
 536 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 537 {
 538         struct super_block *sb = vfs->mnt_sb;
 539
 540         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 541                 seq_puts(seq, ",data=journal");
 542         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 543                 seq_puts(seq, ",data=ordered");
 544         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 545                 seq_puts(seq, ",data=writeback");
 546
 547         ext3_show_quota_options(seq, sb);
 548
 549         return 0;
 550 }
 551
 552 #ifdef CONFIG_QUOTA
 553 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 554 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 555
 556 static int ext3_dquot_initialize(struct inode *inode, int type);
 557 static int ext3_dquot_drop(struct inode *inode);
 558 static int ext3_write_dquot(struct dquot *dquot);
 559 static int ext3_acquire_dquot(struct dquot *dquot);
 560 static int ext3_release_dquot(struct dquot *dquot);
 561 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 562 static int ext3_write_info(struct super_block *sb, int type);
 563 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 564 static int ext3_quota_on_mount(struct super_block *sb, int type);
 565 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 566                                size_t len, loff_t off);
 567 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 568                                 const char *data, size_t len, loff_t off);
 569
 570 static struct dquot_operations ext3_quota_operations = {
 571         .initialize     = ext3_dquot_initialize,
 572         .drop           = ext3_dquot_drop,
 573         .alloc_space    = dquot_alloc_space,
 574         .alloc_inode    = dquot_alloc_inode,
 575         .free_space     = dquot_free_space,
 576         .free_inode     = dquot_free_inode,
 577         .transfer       = dquot_transfer,
 578         .write_dquot    = ext3_write_dquot,
 579         .acquire_dquot  = ext3_acquire_dquot,
 580         .release_dquot  = ext3_release_dquot,
 581         .mark_dirty     = ext3_mark_dquot_dirty,
 582         .write_info     = ext3_write_info
 583 };
 584
 585 static struct quotactl_ops ext3_qctl_operations = {
 586         .quota_on       = ext3_quota_on,
 587         .quota_off      = vfs_quota_off,
 588         .quota_sync     = vfs_quota_sync,
 589         .get_info       = vfs_get_dqinfo,
 590         .set_info       = vfs_set_dqinfo,
 591         .get_dqblk      = vfs_get_dqblk,
 592         .set_dqblk      = vfs_set_dqblk
 593 };
 594 #endif
 595
 596 static struct super_operations ext3_sops = {
 597         .alloc_inode    = ext3_alloc_inode,
 598         .destroy_inode  = ext3_destroy_inode,
 599         .read_inode     = ext3_read_inode,
 600         .write_inode    = ext3_write_inode,
 601         .dirty_inode    = ext3_dirty_inode,
 602         .delete_inode   = ext3_delete_inode,
 603         .put_super      = ext3_put_super,
 604         .write_super    = ext3_write_super,
 605         .sync_fs        = ext3_sync_fs,
 606         .write_super_lockfs = ext3_write_super_lockfs,
 607         .unlockfs       = ext3_unlockfs,
 608         .statfs         = ext3_statfs,
 609         .remount_fs     = ext3_remount,
 610         .clear_inode    = ext3_clear_inode,
 611         .show_options   = ext3_show_options,
 612 #ifdef CONFIG_QUOTA
 613         .quota_read     = ext3_quota_read,
 614         .quota_write    = ext3_quota_write,
 615 #endif
 616 };
 617
 618 struct dentry *ext3_get_parent(struct dentry *child);
 619 static struct export_operations ext3_export_ops = {
 620         .get_parent = ext3_get_parent,
 621 };
 622
 623 enum {
 624         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 625         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 626         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 627         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 628         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 629         Opt_commit, Opt_journal_update, Opt_journal_inum,
 630         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 631         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 632         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 633         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 634         Opt_grpquota
 635 };
 636
 637 static match_table_t tokens = {
 638         {Opt_bsd_df, "bsddf"},
 639         {Opt_minix_df, "minixdf"},
 640         {Opt_grpid, "grpid"},
 641         {Opt_grpid, "bsdgroups"},
 642         {Opt_nogrpid, "nogrpid"},
 643         {Opt_nogrpid, "sysvgroups"},
 644         {Opt_resgid, "resgid=%u"},
 645         {Opt_resuid, "resuid=%u"},
 646         {Opt_sb, "sb=%u"},
 647         {Opt_err_cont, "errors=continue"},
 648         {Opt_err_panic, "errors=panic"},
 649         {Opt_err_ro, "errors=remount-ro"},
 650         {Opt_nouid32, "nouid32"},
 651         {Opt_nocheck, "nocheck"},
 652         {Opt_nocheck, "check=none"},
 653         {Opt_check, "check"},
 654         {Opt_debug, "debug"},
 655         {Opt_oldalloc, "oldalloc"},
 656         {Opt_orlov, "orlov"},
 657         {Opt_user_xattr, "user_xattr"},
 658         {Opt_nouser_xattr, "nouser_xattr"},
 659         {Opt_acl, "acl"},
 660         {Opt_noacl, "noacl"},
 661         {Opt_reservation, "reservation"},
 662         {Opt_noreservation, "noreservation"},
 663         {Opt_noload, "noload"},
 664         {Opt_nobh, "nobh"},
 665         {Opt_commit, "commit=%u"},
 666         {Opt_journal_update, "journal=update"},
 667         {Opt_journal_inum, "journal=%u"},
 668         {Opt_abort, "abort"},
 669         {Opt_data_journal, "data=journal"},
 670         {Opt_data_ordered, "data=ordered"},
 671         {Opt_data_writeback, "data=writeback"},
 672         {Opt_offusrjquota, "usrjquota="},
 673         {Opt_usrjquota, "usrjquota=%s"},
 674         {Opt_offgrpjquota, "grpjquota="},
 675         {Opt_grpjquota, "grpjquota=%s"},
 676         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 677         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 678         {Opt_grpquota, "grpquota"},
 679         {Opt_noquota, "noquota"},
 680         {Opt_quota, "quota"},
 681         {Opt_usrquota, "usrquota"},
 682         {Opt_barrier, "barrier=%u"},
 683         {Opt_err, NULL},
 684         {Opt_resize, "resize"},
 685 };
 686
 687 static unsigned long get_sb_block(void **data)
 688 {
 689         unsigned long   sb_block;
 690         char            *options = (char *) *data;
 691
 692         if (!options || strncmp(options, "sb=", 3) != 0)
 693                 return 1;       /* Default location */
 694         options += 3;
 695         sb_block = simple_strtoul(options, &options, 0);
 696         if (*options && *options != ',') {
 697                 printk("EXT3-fs: Invalid sb specification: %s\n",
 698                        (char *) *data);
 699                 return 1;
 700         }
 701         if (*options == ',')
 702                 options++;
 703         *data = (void *) options;
 704         return sb_block;
 705 }
 706
 707 static int parse_options (char * options, struct super_block *sb,
 708                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 709 {
 710         struct ext3_sb_info *sbi = EXT3_SB(sb);
 711         char * p;
 712         substring_t args[MAX_OPT_ARGS];
 713         int data_opt = 0;
 714         int option;
 715 #ifdef CONFIG_QUOTA
 716         int qtype;
 717         char *qname;
 718 #endif
 719
 720         if (!options)
 721                 return 1;
 722
 723         while ((p = strsep (&options, ",")) != NULL) {
 724                 int token;
 725                 if (!*p)
 726                         continue;
 727
 728                 token = match_token(p, tokens, args);
 729                 switch (token) {
 730                 case Opt_bsd_df:
 731                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 732                         break;
 733                 case Opt_minix_df:
 734                         set_opt (sbi->s_mount_opt, MINIX_DF);
 735                         break;
 736                 case Opt_grpid:
 737                         set_opt (sbi->s_mount_opt, GRPID);
 738                         break;
 739                 case Opt_nogrpid:
 740                         clear_opt (sbi->s_mount_opt, GRPID);
 741                         break;
 742                 case Opt_resuid:
 743                         if (match_int(&args[0], &option))
 744                                 return 0;
 745                         sbi->s_resuid = option;
 746                         break;
 747                 case Opt_resgid:
 748                         if (match_int(&args[0], &option))
 749                                 return 0;
 750                         sbi->s_resgid = option;
 751                         break;
 752                 case Opt_sb:
 753                         /* handled by get_sb_block() instead of here */
 754                         /* *sb_block = match_int(&args[0]); */
 755                         break;
 756                 case Opt_err_panic:
 757                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 758                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 759                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 760                         break;
 761                 case Opt_err_ro:
 762                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 763                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 764                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 765                         break;
 766                 case Opt_err_cont:
 767                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 768                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 769                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 770                         break;
 771                 case Opt_nouid32:
 772                         set_opt (sbi->s_mount_opt, NO_UID32);
 773                         break;
 774                 case Opt_check:
 775 #ifdef CONFIG_EXT3_CHECK
 776                         set_opt (sbi->s_mount_opt, CHECK);
 777 #else
 778                         printk(KERN_ERR
 779                                "EXT3 Check option not supported\n");
 780 #endif
 781                         break;
 782                 case Opt_nocheck:
 783                         clear_opt (sbi->s_mount_opt, CHECK);
 784                         break;
 785                 case Opt_debug:
 786                         set_opt (sbi->s_mount_opt, DEBUG);
 787                         break;
 788                 case Opt_oldalloc:
 789                         set_opt (sbi->s_mount_opt, OLDALLOC);
 790                         break;
 791                 case Opt_orlov:
 792                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 793                         break;
 794 #ifdef CONFIG_EXT3_FS_XATTR
 795                 case Opt_user_xattr:
 796                         set_opt (sbi->s_mount_opt, XATTR_USER);
 797                         break;
 798                 case Opt_nouser_xattr:
 799                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 800                         break;
 801 #else
 802                 case Opt_user_xattr:
 803                 case Opt_nouser_xattr:
 804                         printk("EXT3 (no)user_xattr options not supported\n");
 805                         break;
 806 #endif
 807 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 808                 case Opt_acl:
 809                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 810                         break;
 811                 case Opt_noacl:
 812                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 813                         break;
 814 #else
 815                 case Opt_acl:
 816                 case Opt_noacl:
 817                         printk("EXT3 (no)acl options not supported\n");
 818                         break;
 819 #endif
 820                 case Opt_reservation:
 821                         set_opt(sbi->s_mount_opt, RESERVATION);
 822                         break;
 823                 case Opt_noreservation:
 824                         clear_opt(sbi->s_mount_opt, RESERVATION);
 825                         break;
 826                 case Opt_journal_update:
 827                         /* @@@ FIXME */
 828                         /* Eventually we will want to be able to create
 829                            a journal file here.  For now, only allow the
 830                            user to specify an existing inode to be the
 831                            journal file. */
 832                         if (is_remount) {
 833                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 834                                        "journal on remount\n");
 835                                 return 0;
 836                         }
 837                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 838                         break;
 839                 case Opt_journal_inum:
 840                         if (is_remount) {
 841                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 842                                        "journal on remount\n");
 843                                 return 0;
 844                         }
 845                         if (match_int(&args[0], &option))
 846                                 return 0;
 847                         *inum = option;
 848                         break;
 849                 case Opt_noload:
 850                         set_opt (sbi->s_mount_opt, NOLOAD);
 851                         break;
 852                 case Opt_commit:
 853                         if (match_int(&args[0], &option))
 854                                 return 0;
 855                         if (option < 0)
 856                                 return 0;
 857                         if (option == 0)
 858                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 859                         sbi->s_commit_interval = HZ * option;
 860                         break;
 861                 case Opt_data_journal:
 862                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 863                         goto datacheck;
 864                 case Opt_data_ordered:
 865                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 866                         goto datacheck;
 867                 case Opt_data_writeback:
 868                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 869                 datacheck:
 870                         if (is_remount) {
 871                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 872                                                 != data_opt) {
 873                                         printk(KERN_ERR
 874                                                 "EXT3-fs: cannot change data "
 875                                                 "mode on remount\n");
 876                                         return 0;
 877                                 }
 878                         } else {
 879                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 880                                 sbi->s_mount_opt |= data_opt;
 881                         }
 882                         break;
 883 #ifdef CONFIG_QUOTA
 884                 case Opt_usrjquota:
 885                         qtype = USRQUOTA;
 886                         goto set_qf_name;
 887                 case Opt_grpjquota:
 888                         qtype = GRPQUOTA;
 889 set_qf_name:
 890                         if (sb_any_quota_enabled(sb)) {
 891                                 printk(KERN_ERR
 892                                         "EXT3-fs: Cannot change journalled "
 893                                         "quota options when quota turned on.\n");
 894                                 return 0;
 895                         }
 896                         qname = match_strdup(&args[0]);
 897                         if (!qname) {
 898                                 printk(KERN_ERR
 899                                         "EXT3-fs: not enough memory for "
 900                                         "storing quotafile name.\n");
 901                                 return 0;
 902                         }
 903                         if (sbi->s_qf_names[qtype] &&
 904                             strcmp(sbi->s_qf_names[qtype], qname)) {
 905                                 printk(KERN_ERR
 906                                         "EXT3-fs: %s quota file already "
 907                                         "specified.\n", QTYPE2NAME(qtype));
 908                                 kfree(qname);
 909                                 return 0;
 910                         }
 911                         sbi->s_qf_names[qtype] = qname;
 912                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 913                                 printk(KERN_ERR
 914                                         "EXT3-fs: quotafile must be on "
 915                                         "filesystem root.\n");
 916                                 kfree(sbi->s_qf_names[qtype]);
 917                                 sbi->s_qf_names[qtype] = NULL;
 918                                 return 0;
 919                         }
 920                         set_opt(sbi->s_mount_opt, QUOTA);
 921                         break;
 922                 case Opt_offusrjquota:
 923                         qtype = USRQUOTA;
 924                         goto clear_qf_name;
 925                 case Opt_offgrpjquota:
 926                         qtype = GRPQUOTA;
 927 clear_qf_name:
 928                         if (sb_any_quota_enabled(sb)) {
 929                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 930                                         "journalled quota options when "
 931                                         "quota turned on.\n");
 932                                 return 0;
 933                         }
 934                         /*
 935                          * The space will be released later when all options
 936                          * are confirmed to be correct
 937                          */
 938                         sbi->s_qf_names[qtype] = NULL;
 939                         break;
 940                 case Opt_jqfmt_vfsold:
 941                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 942                         break;
 943                 case Opt_jqfmt_vfsv0:
 944                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 945                         break;
 946                 case Opt_quota:
 947                 case Opt_usrquota:
 948                         set_opt(sbi->s_mount_opt, QUOTA);
 949                         set_opt(sbi->s_mount_opt, USRQUOTA);
 950                         break;
 951                 case Opt_grpquota:
 952                         set_opt(sbi->s_mount_opt, QUOTA);
 953                         set_opt(sbi->s_mount_opt, GRPQUOTA);
 954                         break;
 955                 case Opt_noquota:
 956                         if (sb_any_quota_enabled(sb)) {
 957                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
 958                                         "options when quota turned on.\n");
 959                                 return 0;
 960                         }
 961                         clear_opt(sbi->s_mount_opt, QUOTA);
 962                         clear_opt(sbi->s_mount_opt, USRQUOTA);
 963                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
 964                         break;
 965 #else
 966                 case Opt_quota:
 967                 case Opt_usrquota:
 968                 case Opt_grpquota:
 969                 case Opt_usrjquota:
 970                 case Opt_grpjquota:
 971                 case Opt_offusrjquota:
 972                 case Opt_offgrpjquota:
 973                 case Opt_jqfmt_vfsold:
 974                 case Opt_jqfmt_vfsv0:
 975                         printk(KERN_ERR
 976                                 "EXT3-fs: journalled quota options not "
 977                                 "supported.\n");
 978                         break;
 979                 case Opt_noquota:
 980                         break;
 981 #endif
 982                 case Opt_abort:
 983                         set_opt(sbi->s_mount_opt, ABORT);
 984                         break;
 985                 case Opt_barrier:
 986                         if (match_int(&args[0], &option))
 987                                 return 0;
 988                         if (option)
 989                                 set_opt(sbi->s_mount_opt, BARRIER);
 990                         else
 991                                 clear_opt(sbi->s_mount_opt, BARRIER);
 992                         break;
 993                 case Opt_ignore:
 994                         break;
 995                 case Opt_resize:
 996                         if (!is_remount) {
 997                                 printk("EXT3-fs: resize option only available "
 998                                         "for remount\n");
 999                                 return 0;
1000                         }
1001                         if (match_int(&args[0], &option) != 0)
1002                                 return 0;
1003                         *n_blocks_count = option;
1004                         break;
1005                 case Opt_nobh:
1006                         set_opt(sbi->s_mount_opt, NOBH);
1007                         break;
1008                 default:
1009                         printk (KERN_ERR
1010                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1011                                 "or missing value\n", p);
1012                         return 0;
1013                 }
1014         }
1015 #ifdef CONFIG_QUOTA
1016         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1017                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1018                      sbi->s_qf_names[USRQUOTA])
1019                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1020
1021                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1022                      sbi->s_qf_names[GRPQUOTA])
1023                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1024
1025                 if ((sbi->s_qf_names[USRQUOTA] &&
1026                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1027                     (sbi->s_qf_names[GRPQUOTA] &&
1028                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1029                         printk(KERN_ERR "EXT3-fs: old and new quota "
1030                                         "format mixing.\n");
1031                         return 0;
1032                 }
1033
1034                 if (!sbi->s_jquota_fmt) {
1035                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1036                                         "not specified.\n");
1037                         return 0;
1038                 }
1039         } else {
1040                 if (sbi->s_jquota_fmt) {
1041                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1042                                         "specified with no journalling "
1043                                         "enabled.\n");
1044                         return 0;
1045                 }
1046         }
1047 #endif
1048         return 1;
1049 }
1050
1051 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1052                             int read_only)
1053 {
1054         struct ext3_sb_info *sbi = EXT3_SB(sb);
1055         int res = 0;
1056
1057         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1058                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1059                         "forcing read-only mode\n");
1060                 res = MS_RDONLY;
1061         }
1062         if (read_only)
1063                 return res;
1064         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1065                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1066                         "running e2fsck is recommended\n");
1067         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1068                 printk (KERN_WARNING
1069                         "EXT3-fs warning: mounting fs with errors, "
1070                         "running e2fsck is recommended\n");
1071         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1072                  le16_to_cpu(es->s_mnt_count) >=
1073                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1074                 printk (KERN_WARNING
1075                         "EXT3-fs warning: maximal mount count reached, "
1076                         "running e2fsck is recommended\n");
1077         else if (le32_to_cpu(es->s_checkinterval) &&
1078                 (le32_to_cpu(es->s_lastcheck) +
1079                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1080                 printk (KERN_WARNING
1081                         "EXT3-fs warning: checktime reached, "
1082                         "running e2fsck is recommended\n");
1083 #if 0
1084                 /* @@@ We _will_ want to clear the valid bit if we find
1085                    inconsistencies, to force a fsck at reboot.  But for
1086                    a plain journaled filesystem we can keep it set as
1087                    valid forever! :) */
1088         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1089 #endif
1090         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1091                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1092         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1093         es->s_mtime = cpu_to_le32(get_seconds());
1094         ext3_update_dynamic_rev(sb);
1095         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1096
1097         ext3_commit_super(sb, es, 1);
1098         if (test_opt(sb, DEBUG))
1099                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1100                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1101                         sb->s_blocksize,
1102                         sbi->s_groups_count,
1103                         EXT3_BLOCKS_PER_GROUP(sb),
1104                         EXT3_INODES_PER_GROUP(sb),
1105                         sbi->s_mount_opt);
1106
1107         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1108         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1109                 char b[BDEVNAME_SIZE];
1110
1111                 printk("external journal on %s\n",
1112                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1113         } else {
1114                 printk("internal journal\n");
1115         }
1116 #ifdef CONFIG_EXT3_CHECK
1117         if (test_opt (sb, CHECK)) {
1118                 ext3_check_blocks_bitmap (sb);
1119                 ext3_check_inodes_bitmap (sb);
1120         }
1121 #endif
1122         return res;
1123 }
1124
1125 /* Called at mount-time, super-block is locked */
1126 static int ext3_check_descriptors (struct super_block * sb)
1127 {
1128         struct ext3_sb_info *sbi = EXT3_SB(sb);
1129         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1130         struct ext3_group_desc * gdp = NULL;
1131         int desc_block = 0;
1132         int i;
1133
1134         ext3_debug ("Checking group descriptors");
1135
1136         for (i = 0; i < sbi->s_groups_count; i++)
1137         {
1138                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1139                         gdp = (struct ext3_group_desc *)
1140                                         sbi->s_group_desc[desc_block++]->b_data;
1141                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1142                     le32_to_cpu(gdp->bg_block_bitmap) >=
1143                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1144                 {
1145                         ext3_error (sb, "ext3_check_descriptors",
1146                                     "Block bitmap for group %d"
1147                                     " not in group (block %lu)!",
1148                                     i, (unsigned long)
1149                                         le32_to_cpu(gdp->bg_block_bitmap));
1150                         return 0;
1151                 }
1152                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1153                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1154                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1155                 {
1156                         ext3_error (sb, "ext3_check_descriptors",
1157                                     "Inode bitmap for group %d"
1158                                     " not in group (block %lu)!",
1159                                     i, (unsigned long)
1160                                         le32_to_cpu(gdp->bg_inode_bitmap));
1161                         return 0;
1162                 }
1163                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1164                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1165                     block + EXT3_BLOCKS_PER_GROUP(sb))
1166                 {
1167                         ext3_error (sb, "ext3_check_descriptors",
1168                                     "Inode table for group %d"
1169                                     " not in group (block %lu)!",
1170                                     i, (unsigned long)
1171                                         le32_to_cpu(gdp->bg_inode_table));
1172                         return 0;
1173                 }
1174                 block += EXT3_BLOCKS_PER_GROUP(sb);
1175                 gdp++;
1176         }
1177
1178         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1179         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1180         return 1;
1181 }
1182
1183
1184 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1185  * the superblock) which were deleted from all directories, but held open by
1186  * a process at the time of a crash.  We walk the list and try to delete these
1187  * inodes at recovery time (only with a read-write filesystem).
1188  *
1189  * In order to keep the orphan inode chain consistent during traversal (in
1190  * case of crash during recovery), we link each inode into the superblock
1191  * orphan list_head and handle it the same way as an inode deletion during
1192  * normal operation (which journals the operations for us).
1193  *
1194  * We only do an iget() and an iput() on each inode, which is very safe if we
1195  * accidentally point at an in-use or already deleted inode.  The worst that
1196  * can happen in this case is that we get a "bit already cleared" message from
1197  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1198  * e2fsck was run on this filesystem, and it must have already done the orphan
1199  * inode cleanup for us, so we can safely abort without any further action.
1200  */
1201 static void ext3_orphan_cleanup (struct super_block * sb,
1202                                  struct ext3_super_block * es)
1203 {
1204         unsigned int s_flags = sb->s_flags;
1205         int nr_orphans = 0, nr_truncates = 0;
1206 #ifdef CONFIG_QUOTA
1207         int i;
1208 #endif
1209         if (!es->s_last_orphan) {
1210                 jbd_debug(4, "no orphan inodes to clean up\n");
1211                 return;
1212         }
1213
1214         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1215                 if (es->s_last_orphan)
1216                         jbd_debug(1, "Errors on filesystem, "
1217                                   "clearing orphan list.\n");
1218                 es->s_last_orphan = 0;
1219                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1220                 return;
1221         }
1222
1223         if (s_flags & MS_RDONLY) {
1224                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1225                        sb->s_id);
1226                 sb->s_flags &= ~MS_RDONLY;
1227         }
1228 #ifdef CONFIG_QUOTA
1229         /* Needed for iput() to work correctly and not trash data */
1230         sb->s_flags |= MS_ACTIVE;
1231         /* Turn on quotas so that they are updated correctly */
1232         for (i = 0; i < MAXQUOTAS; i++) {
1233                 if (EXT3_SB(sb)->s_qf_names[i]) {
1234                         int ret = ext3_quota_on_mount(sb, i);
1235                         if (ret < 0)
1236                                 printk(KERN_ERR
1237                                         "EXT3-fs: Cannot turn on journalled "
1238                                         "quota: error %d\n", ret);
1239                 }
1240         }
1241 #endif
1242
1243         while (es->s_last_orphan) {
1244                 struct inode *inode;
1245
1246                 if (!(inode =
1247                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1248                         es->s_last_orphan = 0;
1249                         break;
1250                 }
1251
1252                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1253                 DQUOT_INIT(inode);
1254                 if (inode->i_nlink) {
1255                         printk(KERN_DEBUG
1256                                 "%s: truncating inode %ld to %Ld bytes\n",
1257                                 __FUNCTION__, inode->i_ino, inode->i_size);
1258                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1259                                   inode->i_ino, inode->i_size);
1260                         ext3_truncate(inode);
1261                         nr_truncates++;
1262                 } else {
1263                         printk(KERN_DEBUG
1264                                 "%s: deleting unreferenced inode %ld\n",
1265                                 __FUNCTION__, inode->i_ino);
1266                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1267                                   inode->i_ino);
1268                         nr_orphans++;
1269                 }
1270                 iput(inode);  /* The delete magic happens here! */
1271         }
1272
1273 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1274
1275         if (nr_orphans)
1276                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1277                        sb->s_id, PLURAL(nr_orphans));
1278         if (nr_truncates)
1279                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1280                        sb->s_id, PLURAL(nr_truncates));
1281 #ifdef CONFIG_QUOTA
1282         /* Turn quotas off */
1283         for (i = 0; i < MAXQUOTAS; i++) {
1284                 if (sb_dqopt(sb)->files[i])
1285                         vfs_quota_off(sb, i);
1286         }
1287 #endif
1288         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1289 }
1290
1291 #define log2(n) ffz(~(n))
1292
1293 /*
1294  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1295  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1296  * We need to be 1 filesystem block less than the 2^32 sector limit.
1297  */
1298 static loff_t ext3_max_size(int bits)
1299 {
1300         loff_t res = EXT3_NDIR_BLOCKS;
1301         /* This constant is calculated to be the largest file size for a
1302          * dense, 4k-blocksize file such that the total number of
1303          * sectors in the file, including data and all indirect blocks,
1304          * does not exceed 2^32. */
1305         const loff_t upper_limit = 0x1ff7fffd000LL;
1306
1307         res += 1LL << (bits-2);
1308         res += 1LL << (2*(bits-2));
1309         res += 1LL << (3*(bits-2));
1310         res <<= bits;
1311         if (res > upper_limit)
1312                 res = upper_limit;
1313         return res;
1314 }
1315
1316 static unsigned long descriptor_loc(struct super_block *sb,
1317                                     unsigned long logic_sb_block,
1318                                     int nr)
1319 {
1320         struct ext3_sb_info *sbi = EXT3_SB(sb);
1321         unsigned long bg, first_data_block, first_meta_bg;
1322         int has_super = 0;
1323
1324         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1325         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1326
1327         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1328             nr < first_meta_bg)
1329                 return (logic_sb_block + nr + 1);
1330         bg = sbi->s_desc_per_block * nr;
1331         if (ext3_bg_has_super(sb, bg))
1332                 has_super = 1;
1333         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1334 }
1335
1336
1337 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1338 {
1339         struct buffer_head * bh;
1340         struct ext3_super_block *es = NULL;
1341         struct ext3_sb_info *sbi;
1342         unsigned long block;
1343         unsigned long sb_block = get_sb_block(&data);
1344         unsigned long logic_sb_block;
1345         unsigned long offset = 0;
1346         unsigned long journal_inum = 0;
1347         unsigned long def_mount_opts;
1348         struct inode *root;
1349         int blocksize;
1350         int hblock;
1351         int db_count;
1352         int i;
1353         int needs_recovery;
1354         __le32 features;
1355
1356         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1357         if (!sbi)
1358                 return -ENOMEM;
1359         sb->s_fs_info = sbi;
1360         memset(sbi, 0, sizeof(*sbi));
1361         sbi->s_mount_opt = 0;
1362         sbi->s_resuid = EXT3_DEF_RESUID;
1363         sbi->s_resgid = EXT3_DEF_RESGID;
1364
1365         unlock_kernel();
1366
1367         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1368         if (!blocksize) {
1369                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1370                 goto out_fail;
1371         }
1372
1373         /*
1374          * The ext3 superblock will not be buffer aligned for other than 1kB
1375          * block sizes.  We need to calculate the offset from buffer start.
1376          */
1377         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1378                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1379                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1380         } else {
1381                 logic_sb_block = sb_block;
1382         }
1383
1384         if (!(bh = sb_bread(sb, logic_sb_block))) {
1385                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1386                 goto out_fail;
1387         }
1388         /*
1389          * Note: s_es must be initialized as soon as possible because
1390          *       some ext3 macro-instructions depend on its value
1391          */
1392         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1393         sbi->s_es = es;
1394         sb->s_magic = le16_to_cpu(es->s_magic);
1395         if (sb->s_magic != EXT3_SUPER_MAGIC)
1396                 goto cantfind_ext3;
1397
1398         /* Set defaults before we parse the mount options */
1399         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1400         if (def_mount_opts & EXT3_DEFM_DEBUG)
1401                 set_opt(sbi->s_mount_opt, DEBUG);
1402         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1403                 set_opt(sbi->s_mount_opt, GRPID);
1404         if (def_mount_opts & EXT3_DEFM_UID16)
1405                 set_opt(sbi->s_mount_opt, NO_UID32);
1406         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1407                 set_opt(sbi->s_mount_opt, XATTR_USER);
1408         if (def_mount_opts & EXT3_DEFM_ACL)
1409                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1410         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1411                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1412         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1413                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1414         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1415                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1416
1417         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1418                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1419         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1420                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1421
1422         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1423         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1424
1425         set_opt(sbi->s_mount_opt, RESERVATION);
1426
1427         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1428                 goto failed_mount;
1429
1430         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1431                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1432
1433         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1434             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1435              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1436              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1437                 printk(KERN_WARNING
1438                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1439                        "running e2fsck is recommended\n");
1440         /*
1441          * Check feature flags regardless of the revision level, since we
1442          * previously didn't change the revision level when setting the flags,
1443          * so there is a chance incompat flags are set on a rev 0 filesystem.
1444          */
1445         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1446         if (features) {
1447                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1448                        "unsupported optional features (%x).\n",
1449                        sb->s_id, le32_to_cpu(features));
1450                 goto failed_mount;
1451         }
1452         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1453         if (!(sb->s_flags & MS_RDONLY) && features) {
1454                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1455                        "unsupported optional features (%x).\n",
1456                        sb->s_id, le32_to_cpu(features));
1457                 goto failed_mount;
1458         }
1459         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1460
1461         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1462             blocksize > EXT3_MAX_BLOCK_SIZE) {
1463                 printk(KERN_ERR
1464                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1465                        blocksize, sb->s_id);
1466                 goto failed_mount;
1467         }
1468
1469         hblock = bdev_hardsect_size(sb->s_bdev);
1470         if (sb->s_blocksize != blocksize) {
1471                 /*
1472                  * Make sure the blocksize for the filesystem is larger
1473                  * than the hardware sectorsize for the machine.
1474                  */
1475                 if (blocksize < hblock) {
1476                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1477                                "device blocksize %d.\n", blocksize, hblock);
1478                         goto failed_mount;
1479                 }
1480
1481                 brelse (bh);
1482                 sb_set_blocksize(sb, blocksize);
1483                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1484                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1485                 bh = sb_bread(sb, logic_sb_block);
1486                 if (!bh) {
1487                         printk(KERN_ERR
1488                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1489                         goto failed_mount;
1490                 }
1491                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1492                 sbi->s_es = es;
1493                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1494                         printk (KERN_ERR
1495                                 "EXT3-fs: Magic mismatch, very weird !\n");
1496                         goto failed_mount;
1497                 }
1498         }
1499
1500         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1501
1502         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1503                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1504                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1505         } else {
1506                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1507                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1508                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1509                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1510                     (sbi->s_inode_size > blocksize)) {
1511                         printk (KERN_ERR
1512                                 "EXT3-fs: unsupported inode size: %d\n",
1513                                 sbi->s_inode_size);
1514                         goto failed_mount;
1515                 }
1516         }
1517         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1518                                    le32_to_cpu(es->s_log_frag_size);
1519         if (blocksize != sbi->s_frag_size) {
1520                 printk(KERN_ERR
1521                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1522                        sbi->s_frag_size, blocksize);
1523                 goto failed_mount;
1524         }
1525         sbi->s_frags_per_block = 1;
1526         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1527         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1528         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1529         if (EXT3_INODE_SIZE(sb) == 0)
1530                 goto cantfind_ext3;
1531         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1532         if (sbi->s_inodes_per_block == 0)
1533                 goto cantfind_ext3;
1534         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1535                                         sbi->s_inodes_per_block;
1536         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1537         sbi->s_sbh = bh;
1538         sbi->s_mount_state = le16_to_cpu(es->s_state);
1539         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1540         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1541         for (i=0; i < 4; i++)
1542                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1543         sbi->s_def_hash_version = es->s_def_hash_version;
1544
1545         if (sbi->s_blocks_per_group > blocksize * 8) {
1546                 printk (KERN_ERR
1547                         "EXT3-fs: #blocks per group too big: %lu\n",
1548                         sbi->s_blocks_per_group);
1549                 goto failed_mount;
1550         }
1551         if (sbi->s_frags_per_group > blocksize * 8) {
1552                 printk (KERN_ERR
1553                         "EXT3-fs: #fragments per group too big: %lu\n",
1554                         sbi->s_frags_per_group);
1555                 goto failed_mount;
1556         }
1557         if (sbi->s_inodes_per_group > blocksize * 8) {
1558                 printk (KERN_ERR
1559                         "EXT3-fs: #inodes per group too big: %lu\n",
1560                         sbi->s_inodes_per_group);
1561                 goto failed_mount;
1562         }
1563
1564         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1565                 goto cantfind_ext3;
1566         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1567                                le32_to_cpu(es->s_first_data_block) +
1568                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1569                               EXT3_BLOCKS_PER_GROUP(sb);
1570         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1571                    EXT3_DESC_PER_BLOCK(sb);
1572         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1573                                     GFP_KERNEL);
1574         if (sbi->s_group_desc == NULL) {
1575                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1576                 goto failed_mount;
1577         }
1578
1579         percpu_counter_init(&sbi->s_freeblocks_counter);
1580         percpu_counter_init(&sbi->s_freeinodes_counter);
1581         percpu_counter_init(&sbi->s_dirs_counter);
1582         bgl_lock_init(&sbi->s_blockgroup_lock);
1583
1584         for (i = 0; i < db_count; i++) {
1585                 block = descriptor_loc(sb, logic_sb_block, i);
1586                 sbi->s_group_desc[i] = sb_bread(sb, block);
1587                 if (!sbi->s_group_desc[i]) {
1588                         printk (KERN_ERR "EXT3-fs: "
1589                                 "can't read group descriptor %d\n", i);
1590                         db_count = i;
1591                         goto failed_mount2;
1592                 }
1593         }
1594         if (!ext3_check_descriptors (sb)) {
1595                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1596                 goto failed_mount2;
1597         }
1598         sbi->s_gdb_count = db_count;
1599         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1600         spin_lock_init(&sbi->s_next_gen_lock);
1601         /* per fileystem reservation list head & lock */
1602         spin_lock_init(&sbi->s_rsv_window_lock);
1603         sbi->s_rsv_window_root = RB_ROOT;
1604         /* Add a single, static dummy reservation to the start of the
1605          * reservation window list --- it gives us a placeholder for
1606          * append-at-start-of-list which makes the allocation logic
1607          * _much_ simpler. */
1608         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1609         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1610         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1611         sbi->s_rsv_window_head.rsv_goal_size = 0;
1612         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1613
1614         /*
1615          * set up enough so that it can read an inode
1616          */
1617         sb->s_op = &ext3_sops;
1618         sb->s_export_op = &ext3_export_ops;
1619         sb->s_xattr = ext3_xattr_handlers;
1620 #ifdef CONFIG_QUOTA
1621         sb->s_qcop = &ext3_qctl_operations;
1622         sb->dq_op = &ext3_quota_operations;
1623 #endif
1624         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1625
1626         sb->s_root = NULL;
1627
1628         needs_recovery = (es->s_last_orphan != 0 ||
1629                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1630                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1631
1632         /*
1633          * The first inode we look at is the journal inode.  Don't try
1634          * root first: it may be modified in the journal!
1635          */
1636         if (!test_opt(sb, NOLOAD) &&
1637             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1638                 if (ext3_load_journal(sb, es))
1639                         goto failed_mount2;
1640         } else if (journal_inum) {
1641                 if (ext3_create_journal(sb, es, journal_inum))
1642                         goto failed_mount2;
1643         } else {
1644                 if (!silent)
1645                         printk (KERN_ERR
1646                                 "ext3: No journal on filesystem on %s\n",
1647                                 sb->s_id);
1648                 goto failed_mount2;
1649         }
1650
1651         /* We have now updated the journal if required, so we can
1652          * validate the data journaling mode. */
1653         switch (test_opt(sb, DATA_FLAGS)) {
1654         case 0:
1655                 /* No mode set, assume a default based on the journal
1656                    capabilities: ORDERED_DATA if the journal can
1657                    cope, else JOURNAL_DATA */
1658                 if (journal_check_available_features
1659                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1660                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1661                 else
1662                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1663                 break;
1664
1665         case EXT3_MOUNT_ORDERED_DATA:
1666         case EXT3_MOUNT_WRITEBACK_DATA:
1667                 if (!journal_check_available_features
1668                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1669                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1670                                "requested data journaling mode\n");
1671                         goto failed_mount3;
1672                 }
1673         default:
1674                 break;
1675         }
1676
1677         if (test_opt(sb, NOBH)) {
1678                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1679                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1680                                 "since filesystem blocksize doesn't match "
1681                                 "pagesize\n");
1682                         clear_opt(sbi->s_mount_opt, NOBH);
1683                 }
1684                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1685                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1686                                 "its supported only with writeback mode\n");
1687                         clear_opt(sbi->s_mount_opt, NOBH);
1688                 }
1689         }
1690         /*
1691          * The journal_load will have done any necessary log recovery,
1692          * so we can safely mount the rest of the filesystem now.
1693          */
1694
1695         root = iget(sb, EXT3_ROOT_INO);
1696         sb->s_root = d_alloc_root(root);
1697         if (!sb->s_root) {
1698                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1699                 iput(root);
1700                 goto failed_mount3;
1701         }
1702         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1703                 dput(sb->s_root);
1704                 sb->s_root = NULL;
1705                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1706                 goto failed_mount3;
1707         }
1708
1709         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1710         /*
1711          * akpm: core read_super() calls in here with the superblock locked.
1712          * That deadlocks, because orphan cleanup needs to lock the superblock
1713          * in numerous places.  Here we just pop the lock - it's relatively
1714          * harmless, because we are now ready to accept write_super() requests,
1715          * and aviro says that's the only reason for hanging onto the
1716          * superblock lock.
1717          */
1718         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1719         ext3_orphan_cleanup(sb, es);
1720         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1721         if (needs_recovery)
1722                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1723         ext3_mark_recovery_complete(sb, es);
1724         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1725                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1726                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1727                 "writeback");
1728
1729         percpu_counter_mod(&sbi->s_freeblocks_counter,
1730                 ext3_count_free_blocks(sb));
1731         percpu_counter_mod(&sbi->s_freeinodes_counter,
1732                 ext3_count_free_inodes(sb));
1733         percpu_counter_mod(&sbi->s_dirs_counter,
1734                 ext3_count_dirs(sb));
1735
1736         lock_kernel();
1737         return 0;
1738
1739 cantfind_ext3:
1740         if (!silent)
1741                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1742                        sb->s_id);
1743         goto failed_mount;
1744
1745 failed_mount3:
1746         journal_destroy(sbi->s_journal);
1747 failed_mount2:
1748         for (i = 0; i < db_count; i++)
1749                 brelse(sbi->s_group_desc[i]);
1750         kfree(sbi->s_group_desc);
1751 failed_mount:
1752 #ifdef CONFIG_QUOTA
1753         for (i = 0; i < MAXQUOTAS; i++)
1754                 kfree(sbi->s_qf_names[i]);
1755 #endif
1756         ext3_blkdev_remove(sbi);
1757         brelse(bh);
1758 out_fail:
1759         sb->s_fs_info = NULL;
1760         kfree(sbi);
1761         lock_kernel();
1762         return -EINVAL;
1763 }
1764
1765 /*
1766  * Setup any per-fs journal parameters now.  We'll do this both on
1767  * initial mount, once the journal has been initialised but before we've
1768  * done any recovery; and again on any subsequent remount.
1769  */
1770 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1771 {
1772         struct ext3_sb_info *sbi = EXT3_SB(sb);
1773
1774         if (sbi->s_commit_interval)
1775                 journal->j_commit_interval = sbi->s_commit_interval;
1776         /* We could also set up an ext3-specific default for the commit
1777          * interval here, but for now we'll just fall back to the jbd
1778          * default. */
1779
1780         spin_lock(&journal->j_state_lock);
1781         if (test_opt(sb, BARRIER))
1782                 journal->j_flags |= JFS_BARRIER;
1783         else
1784                 journal->j_flags &= ~JFS_BARRIER;
1785         spin_unlock(&journal->j_state_lock);
1786 }
1787
1788 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1789 {
1790         struct inode *journal_inode;
1791         journal_t *journal;
1792
1793         /* First, test for the existence of a valid inode on disk.  Bad
1794          * things happen if we iget() an unused inode, as the subsequent
1795          * iput() will try to delete it. */
1796
1797         journal_inode = iget(sb, journal_inum);
1798         if (!journal_inode) {
1799                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1800                 return NULL;
1801         }
1802         if (!journal_inode->i_nlink) {
1803                 make_bad_inode(journal_inode);
1804                 iput(journal_inode);
1805                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1806                 return NULL;
1807         }
1808
1809         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1810                   journal_inode, journal_inode->i_size);
1811         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1812                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1813                 iput(journal_inode);
1814                 return NULL;
1815         }
1816
1817         journal = journal_init_inode(journal_inode);
1818         if (!journal) {
1819                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1820                 iput(journal_inode);
1821                 return NULL;
1822         }
1823         journal->j_private = sb;
1824         ext3_init_journal_params(sb, journal);
1825         return journal;
1826 }
1827
1828 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1829                                        dev_t j_dev)
1830 {
1831         struct buffer_head * bh;
1832         journal_t *journal;
1833         int start;
1834         int len;
1835         int hblock, blocksize;
1836         unsigned long sb_block;
1837         unsigned long offset;
1838         struct ext3_super_block * es;
1839         struct block_device *bdev;
1840
1841         bdev = ext3_blkdev_get(j_dev);
1842         if (bdev == NULL)
1843                 return NULL;
1844
1845         if (bd_claim(bdev, sb)) {
1846                 printk(KERN_ERR
1847                         "EXT3: failed to claim external journal device.\n");
1848                 blkdev_put(bdev);
1849                 return NULL;
1850         }
1851
1852         blocksize = sb->s_blocksize;
1853         hblock = bdev_hardsect_size(bdev);
1854         if (blocksize < hblock) {
1855                 printk(KERN_ERR
1856                         "EXT3-fs: blocksize too small for journal device.\n");
1857                 goto out_bdev;
1858         }
1859
1860         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1861         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1862         set_blocksize(bdev, blocksize);
1863         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1864                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1865                        "external journal\n");
1866                 goto out_bdev;
1867         }
1868
1869         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1870         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1871             !(le32_to_cpu(es->s_feature_incompat) &
1872               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1873                 printk(KERN_ERR "EXT3-fs: external journal has "
1874                                         "bad superblock\n");
1875                 brelse(bh);
1876                 goto out_bdev;
1877         }
1878
1879         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1880                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1881                 brelse(bh);
1882                 goto out_bdev;
1883         }
1884
1885         len = le32_to_cpu(es->s_blocks_count);
1886         start = sb_block + 1;
1887         brelse(bh);     /* we're done with the superblock */
1888
1889         journal = journal_init_dev(bdev, sb->s_bdev,
1890                                         start, len, blocksize);
1891         if (!journal) {
1892                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1893                 goto out_bdev;
1894         }
1895         journal->j_private = sb;
1896         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1897         wait_on_buffer(journal->j_sb_buffer);
1898         if (!buffer_uptodate(journal->j_sb_buffer)) {
1899                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1900                 goto out_journal;
1901         }
1902         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1903                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1904                                         "user (unsupported) - %d\n",
1905                         be32_to_cpu(journal->j_superblock->s_nr_users));
1906                 goto out_journal;
1907         }
1908         EXT3_SB(sb)->journal_bdev = bdev;
1909         ext3_init_journal_params(sb, journal);
1910         return journal;
1911 out_journal:
1912         journal_destroy(journal);
1913 out_bdev:
1914         ext3_blkdev_put(bdev);
1915         return NULL;
1916 }
1917
1918 static int ext3_load_journal(struct super_block * sb,
1919                              struct ext3_super_block * es)
1920 {
1921         journal_t *journal;
1922         int journal_inum = le32_to_cpu(es->s_journal_inum);
1923         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1924         int err = 0;
1925         int really_read_only;
1926
1927         really_read_only = bdev_read_only(sb->s_bdev);
1928
1929         /*
1930          * Are we loading a blank journal or performing recovery after a
1931          * crash?  For recovery, we need to check in advance whether we
1932          * can get read-write access to the device.
1933          */
1934
1935         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1936                 if (sb->s_flags & MS_RDONLY) {
1937                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1938                                         "required on readonly filesystem.\n");
1939                         if (really_read_only) {
1940                                 printk(KERN_ERR "EXT3-fs: write access "
1941                                         "unavailable, cannot proceed.\n");
1942                                 return -EROFS;
1943                         }
1944                         printk (KERN_INFO "EXT3-fs: write access will "
1945                                         "be enabled during recovery.\n");
1946                 }
1947         }
1948
1949         if (journal_inum && journal_dev) {
1950                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1951                        "and inode journals!\n");
1952                 return -EINVAL;
1953         }
1954
1955         if (journal_inum) {
1956                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1957                         return -EINVAL;
1958         } else {
1959                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1960                         return -EINVAL;
1961         }
1962
1963         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1964                 err = journal_update_format(journal);
1965                 if (err)  {
1966                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1967                         journal_destroy(journal);
1968                         return err;
1969                 }
1970         }
1971
1972         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1973                 err = journal_wipe(journal, !really_read_only);
1974         if (!err)
1975                 err = journal_load(journal);
1976
1977         if (err) {
1978                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1979                 journal_destroy(journal);
1980                 return err;
1981         }
1982
1983         EXT3_SB(sb)->s_journal = journal;
1984         ext3_clear_journal_err(sb, es);
1985         return 0;
1986 }
1987
1988 static int ext3_create_journal(struct super_block * sb,
1989                                struct ext3_super_block * es,
1990                                int journal_inum)
1991 {
1992         journal_t *journal;
1993
1994         if (sb->s_flags & MS_RDONLY) {
1995                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1996                                 "create journal.\n");
1997                 return -EROFS;
1998         }
1999
2000         if (!(journal = ext3_get_journal(sb, journal_inum)))
2001                 return -EINVAL;
2002
2003         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2004                journal_inum);
2005
2006         if (journal_create(journal)) {
2007                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2008                 journal_destroy(journal);
2009                 return -EIO;
2010         }
2011
2012         EXT3_SB(sb)->s_journal = journal;
2013
2014         ext3_update_dynamic_rev(sb);
2015         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2016         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2017
2018         es->s_journal_inum = cpu_to_le32(journal_inum);
2019         sb->s_dirt = 1;
2020
2021         /* Make sure we flush the recovery flag to disk. */
2022         ext3_commit_super(sb, es, 1);
2023
2024         return 0;
2025 }
2026
2027 static void ext3_commit_super (struct super_block * sb,
2028                                struct ext3_super_block * es,
2029                                int sync)
2030 {
2031         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2032
2033         if (!sbh)
2034                 return;
2035         es->s_wtime = cpu_to_le32(get_seconds());
2036         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2037         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2038         BUFFER_TRACE(sbh, "marking dirty");
2039         mark_buffer_dirty(sbh);
2040         if (sync)
2041                 sync_dirty_buffer(sbh);
2042 }
2043
2044
2045 /*
2046  * Have we just finished recovery?  If so, and if we are mounting (or
2047  * remounting) the filesystem readonly, then we will end up with a
2048  * consistent fs on disk.  Record that fact.
2049  */
2050 static void ext3_mark_recovery_complete(struct super_block * sb,
2051                                         struct ext3_super_block * es)
2052 {
2053         journal_t *journal = EXT3_SB(sb)->s_journal;
2054
2055         journal_lock_updates(journal);
2056         journal_flush(journal);
2057         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2058             sb->s_flags & MS_RDONLY) {
2059                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2060                 sb->s_dirt = 0;
2061                 ext3_commit_super(sb, es, 1);
2062         }
2063         journal_unlock_updates(journal);
2064 }
2065
2066 /*
2067  * If we are mounting (or read-write remounting) a filesystem whose journal
2068  * has recorded an error from a previous lifetime, move that error to the
2069  * main filesystem now.
2070  */
2071 static void ext3_clear_journal_err(struct super_block * sb,
2072                                    struct ext3_super_block * es)
2073 {
2074         journal_t *journal;
2075         int j_errno;
2076         const char *errstr;
2077
2078         journal = EXT3_SB(sb)->s_journal;
2079
2080         /*
2081          * Now check for any error status which may have been recorded in the
2082          * journal by a prior ext3_error() or ext3_abort()
2083          */
2084
2085         j_errno = journal_errno(journal);
2086         if (j_errno) {
2087                 char nbuf[16];
2088
2089                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2090                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2091                              "from previous mount: %s", errstr);
2092                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2093                              "filesystem check.");
2094
2095                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2096                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2097                 ext3_commit_super (sb, es, 1);
2098
2099                 journal_clear_err(journal);
2100         }
2101 }
2102
2103 /*
2104  * Force the running and committing transactions to commit,
2105  * and wait on the commit.
2106  */
2107 int ext3_force_commit(struct super_block *sb)
2108 {
2109         journal_t *journal;
2110         int ret;
2111
2112         if (sb->s_flags & MS_RDONLY)
2113                 return 0;
2114
2115         journal = EXT3_SB(sb)->s_journal;
2116         sb->s_dirt = 0;
2117         ret = ext3_journal_force_commit(journal);
2118         return ret;
2119 }
2120
2121 /*
2122  * Ext3 always journals updates to the superblock itself, so we don't
2123  * have to propagate any other updates to the superblock on disk at this
2124  * point.  Just start an async writeback to get the buffers on their way
2125  * to the disk.
2126  *
2127  * This implicitly triggers the writebehind on sync().
2128  */
2129
2130 static void ext3_write_super (struct super_block * sb)
2131 {
2132         if (down_trylock(&sb->s_lock) == 0)
2133                 BUG();
2134         sb->s_dirt = 0;
2135 }
2136
2137 static int ext3_sync_fs(struct super_block *sb, int wait)
2138 {
2139         tid_t target;
2140
2141         sb->s_dirt = 0;
2142         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2143                 if (wait)
2144                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2145         }
2146         return 0;
2147 }
2148
2149 /*
2150  * LVM calls this function before a (read-only) snapshot is created.  This
2151  * gives us a chance to flush the journal completely and mark the fs clean.
2152  */
2153 static void ext3_write_super_lockfs(struct super_block *sb)
2154 {
2155         sb->s_dirt = 0;
2156
2157         if (!(sb->s_flags & MS_RDONLY)) {
2158                 journal_t *journal = EXT3_SB(sb)->s_journal;
2159
2160                 /* Now we set up the journal barrier. */
2161                 journal_lock_updates(journal);
2162                 journal_flush(journal);
2163
2164                 /* Journal blocked and flushed, clear needs_recovery flag. */
2165                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2166                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2167         }
2168 }
2169
2170 /*
2171  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2172  * flag here, even though the filesystem is not technically dirty yet.
2173  */
2174 static void ext3_unlockfs(struct super_block *sb)
2175 {
2176         if (!(sb->s_flags & MS_RDONLY)) {
2177                 lock_super(sb);
2178                 /* Reser the needs_recovery flag before the fs is unlocked. */
2179                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2180                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2181                 unlock_super(sb);
2182                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2183         }
2184 }
2185
2186 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2187 {
2188         struct ext3_super_block * es;
2189         struct ext3_sb_info *sbi = EXT3_SB(sb);
2190         unsigned long n_blocks_count = 0;
2191         unsigned long old_sb_flags;
2192         struct ext3_mount_options old_opts;
2193         int err;
2194 #ifdef CONFIG_QUOTA
2195         int i;
2196 #endif
2197
2198         /* Store the original options */
2199         old_sb_flags = sb->s_flags;
2200         old_opts.s_mount_opt = sbi->s_mount_opt;
2201         old_opts.s_resuid = sbi->s_resuid;
2202         old_opts.s_resgid = sbi->s_resgid;
2203         old_opts.s_commit_interval = sbi->s_commit_interval;
2204 #ifdef CONFIG_QUOTA
2205         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2206         for (i = 0; i < MAXQUOTAS; i++)
2207                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2208 #endif
2209
2210         /*
2211          * Allow the "check" option to be passed as a remount option.
2212          */
2213         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2214                 err = -EINVAL;
2215                 goto restore_opts;
2216         }
2217
2218         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2219                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2220
2221         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2222                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2223
2224         es = sbi->s_es;
2225
2226         ext3_init_journal_params(sb, sbi->s_journal);
2227
2228         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2229                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2230                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2231                         err = -EROFS;
2232                         goto restore_opts;
2233                 }
2234
2235                 if (*flags & MS_RDONLY) {
2236                         /*
2237                          * First of all, the unconditional stuff we have to do
2238                          * to disable replay of the journal when we next remount
2239                          */
2240                         sb->s_flags |= MS_RDONLY;
2241
2242                         /*
2243                          * OK, test if we are remounting a valid rw partition
2244                          * readonly, and if so set the rdonly flag and then
2245                          * mark the partition as valid again.
2246                          */
2247                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2248                             (sbi->s_mount_state & EXT3_VALID_FS))
2249                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2250
2251                         ext3_mark_recovery_complete(sb, es);
2252                 } else {
2253                         __le32 ret;
2254                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2255                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2256                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2257                                        "remount RDWR because of unsupported "
2258                                        "optional features (%x).\n",
2259                                        sb->s_id, le32_to_cpu(ret));
2260                                 err = -EROFS;
2261                                 goto restore_opts;
2262                         }
2263                         /*
2264                          * Mounting a RDONLY partition read-write, so reread
2265                          * and store the current valid flag.  (It may have
2266                          * been changed by e2fsck since we originally mounted
2267                          * the partition.)
2268                          */
2269                         ext3_clear_journal_err(sb, es);
2270                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2271                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2272                                 err = ret;
2273                                 goto restore_opts;
2274                         }
2275                         if (!ext3_setup_super (sb, es, 0))
2276                                 sb->s_flags &= ~MS_RDONLY;
2277                 }
2278         }
2279 #ifdef CONFIG_QUOTA
2280         /* Release old quota file names */
2281         for (i = 0; i < MAXQUOTAS; i++)
2282                 if (old_opts.s_qf_names[i] &&
2283                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2284                         kfree(old_opts.s_qf_names[i]);
2285 #endif
2286         return 0;
2287 restore_opts:
2288         sb->s_flags = old_sb_flags;
2289         sbi->s_mount_opt = old_opts.s_mount_opt;
2290         sbi->s_resuid = old_opts.s_resuid;
2291         sbi->s_resgid = old_opts.s_resgid;
2292         sbi->s_commit_interval = old_opts.s_commit_interval;
2293 #ifdef CONFIG_QUOTA
2294         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2295         for (i = 0; i < MAXQUOTAS; i++) {
2296                 if (sbi->s_qf_names[i] &&
2297                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2298                         kfree(sbi->s_qf_names[i]);
2299                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2300         }
2301 #endif
2302         return err;
2303 }
2304
2305 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2306 {
2307         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2308         unsigned long overhead;
2309         int i;
2310
2311         if (test_opt (sb, MINIX_DF))
2312                 overhead = 0;
2313         else {
2314                 unsigned long ngroups;
2315                 ngroups = EXT3_SB(sb)->s_groups_count;
2316                 smp_rmb();
2317
2318                 /*
2319                  * Compute the overhead (FS structures)
2320                  */
2321
2322                 /*
2323                  * All of the blocks before first_data_block are
2324                  * overhead
2325                  */
2326                 overhead = le32_to_cpu(es->s_first_data_block);
2327
2328                 /*
2329                  * Add the overhead attributed to the superblock and
2330                  * block group descriptors.  If the sparse superblocks
2331                  * feature is turned on, then not all groups have this.
2332                  */
2333                 for (i = 0; i < ngroups; i++) {
2334                         overhead += ext3_bg_has_super(sb, i) +
2335                                 ext3_bg_num_gdb(sb, i);
2336                         cond_resched();
2337                 }
2338
2339                 /*
2340                  * Every block group has an inode bitmap, a block
2341                  * bitmap, and an inode table.
2342                  */
2343                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2344         }
2345
2346         buf->f_type = EXT3_SUPER_MAGIC;
2347         buf->f_bsize = sb->s_blocksize;
2348         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2349         buf->f_bfree = ext3_count_free_blocks (sb);
2350         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2351         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2352                 buf->f_bavail = 0;
2353         buf->f_files = le32_to_cpu(es->s_inodes_count);
2354         buf->f_ffree = ext3_count_free_inodes (sb);
2355         buf->f_namelen = EXT3_NAME_LEN;
2356         return 0;
2357 }
2358
2359 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2360  * is locked for write. Otherwise the are possible deadlocks:
2361  * Process 1                         Process 2
2362  * ext3_create()                     quota_sync()
2363  *   journal_start()                   write_dquot()
2364  *   DQUOT_INIT()                        down(dqio_sem)
2365  *     down(dqio_sem)                    journal_start()
2366  *
2367  */
2368
2369 #ifdef CONFIG_QUOTA
2370
2371 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2372 {
2373         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2374 }
2375
2376 static int ext3_dquot_initialize(struct inode *inode, int type)
2377 {
2378         handle_t *handle;
2379         int ret, err;
2380
2381         /* We may create quota structure so we need to reserve enough blocks */
2382         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2383         if (IS_ERR(handle))
2384                 return PTR_ERR(handle);
2385         ret = dquot_initialize(inode, type);
2386         err = ext3_journal_stop(handle);
2387         if (!ret)
2388                 ret = err;
2389         return ret;
2390 }
2391
2392 static int ext3_dquot_drop(struct inode *inode)
2393 {
2394         handle_t *handle;
2395         int ret, err;
2396
2397         /* We may delete quota structure so we need to reserve enough blocks */
2398         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2399         if (IS_ERR(handle))
2400                 return PTR_ERR(handle);
2401         ret = dquot_drop(inode);
2402         err = ext3_journal_stop(handle);
2403         if (!ret)
2404                 ret = err;
2405         return ret;
2406 }
2407
2408 static int ext3_write_dquot(struct dquot *dquot)
2409 {
2410         int ret, err;
2411         handle_t *handle;
2412         struct inode *inode;
2413
2414         inode = dquot_to_inode(dquot);
2415         handle = ext3_journal_start(inode,
2416                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2417         if (IS_ERR(handle))
2418                 return PTR_ERR(handle);
2419         ret = dquot_commit(dquot);
2420         err = ext3_journal_stop(handle);
2421         if (!ret)
2422                 ret = err;
2423         return ret;
2424 }
2425
2426 static int ext3_acquire_dquot(struct dquot *dquot)
2427 {
2428         int ret, err;
2429         handle_t *handle;
2430
2431         handle = ext3_journal_start(dquot_to_inode(dquot),
2432                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2433         if (IS_ERR(handle))
2434                 return PTR_ERR(handle);
2435         ret = dquot_acquire(dquot);
2436         err = ext3_journal_stop(handle);
2437         if (!ret)
2438                 ret = err;
2439         return ret;
2440 }
2441
2442 static int ext3_release_dquot(struct dquot *dquot)
2443 {
2444         int ret, err;
2445         handle_t *handle;
2446
2447         handle = ext3_journal_start(dquot_to_inode(dquot),
2448                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2449         if (IS_ERR(handle))
2450                 return PTR_ERR(handle);
2451         ret = dquot_release(dquot);
2452         err = ext3_journal_stop(handle);
2453         if (!ret)
2454                 ret = err;
2455         return ret;
2456 }
2457
2458 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2459 {
2460         /* Are we journalling quotas? */
2461         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2462             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2463                 dquot_mark_dquot_dirty(dquot);
2464                 return ext3_write_dquot(dquot);
2465         } else {
2466                 return dquot_mark_dquot_dirty(dquot);
2467         }
2468 }
2469
2470 static int ext3_write_info(struct super_block *sb, int type)
2471 {
2472         int ret, err;
2473         handle_t *handle;
2474
2475         /* Data block + inode block */
2476         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2477         if (IS_ERR(handle))
2478                 return PTR_ERR(handle);
2479         ret = dquot_commit_info(sb, type);
2480         err = ext3_journal_stop(handle);
2481         if (!ret)
2482                 ret = err;
2483         return ret;
2484 }
2485
2486 /*
2487  * Turn on quotas during mount time - we need to find
2488  * the quota file and such...
2489  */
2490 static int ext3_quota_on_mount(struct super_block *sb, int type)
2491 {
2492         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2493                         EXT3_SB(sb)->s_jquota_fmt, type);
2494 }
2495
2496 /*
2497  * Standard function to be called on quota_on
2498  */
2499 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2500                          char *path)
2501 {
2502         int err;
2503         struct nameidata nd;
2504
2505         if (!test_opt(sb, QUOTA))
2506                 return -EINVAL;
2507         /* Not journalling quota? */
2508         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2509             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2510                 return vfs_quota_on(sb, type, format_id, path);
2511         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2512         if (err)
2513                 return err;
2514         /* Quotafile not on the same filesystem? */
2515         if (nd.mnt->mnt_sb != sb) {
2516                 path_release(&nd);
2517                 return -EXDEV;
2518         }
2519         /* Quotafile not of fs root? */
2520         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2521                 printk(KERN_WARNING
2522                         "EXT3-fs: Quota file not on filesystem root. "
2523                         "Journalled quota will not work.\n");
2524         path_release(&nd);
2525         return vfs_quota_on(sb, type, format_id, path);
2526 }
2527
2528 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2529  * acquiring the locks... As quota files are never truncated and quota code
2530  * itself serializes the operations (and noone else should touch the files)
2531  * we don't have to be afraid of races */
2532 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2533                                size_t len, loff_t off)
2534 {
2535         struct inode *inode = sb_dqopt(sb)->files[type];
2536         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2537         int err = 0;
2538         int offset = off & (sb->s_blocksize - 1);
2539         int tocopy;
2540         size_t toread;
2541         struct buffer_head *bh;
2542         loff_t i_size = i_size_read(inode);
2543
2544         if (off > i_size)
2545                 return 0;
2546         if (off+len > i_size)
2547                 len = i_size-off;
2548         toread = len;
2549         while (toread > 0) {
2550                 tocopy = sb->s_blocksize - offset < toread ?
2551                                 sb->s_blocksize - offset : toread;
2552                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2553                 if (err)
2554                         return err;
2555                 if (!bh)        /* A hole? */
2556                         memset(data, 0, tocopy);
2557                 else
2558                         memcpy(data, bh->b_data+offset, tocopy);
2559                 brelse(bh);
2560                 offset = 0;
2561                 toread -= tocopy;
2562                 data += tocopy;
2563                 blk++;
2564         }
2565         return len;
2566 }
2567
2568 /* Write to quotafile (we know the transaction is already started and has
2569  * enough credits) */
2570 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2571                                 const char *data, size_t len, loff_t off)
2572 {
2573         struct inode *inode = sb_dqopt(sb)->files[type];
2574         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2575         int err = 0;
2576         int offset = off & (sb->s_blocksize - 1);
2577         int tocopy;
2578         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2579         size_t towrite = len;
2580         struct buffer_head *bh;
2581         handle_t *handle = journal_current_handle();
2582
2583         down(&inode->i_sem);
2584         while (towrite > 0) {
2585                 tocopy = sb->s_blocksize - offset < towrite ?
2586                                 sb->s_blocksize - offset : towrite;
2587                 bh = ext3_bread(handle, inode, blk, 1, &err);
2588                 if (!bh)
2589                         goto out;
2590                 if (journal_quota) {
2591                         err = ext3_journal_get_write_access(handle, bh);
2592                         if (err) {
2593                                 brelse(bh);
2594                                 goto out;
2595                         }
2596                 }
2597                 lock_buffer(bh);
2598                 memcpy(bh->b_data+offset, data, tocopy);
2599                 flush_dcache_page(bh->b_page);
2600                 unlock_buffer(bh);
2601                 if (journal_quota)
2602                         err = ext3_journal_dirty_metadata(handle, bh);
2603                 else {
2604                         /* Always do at least ordered writes for quotas */
2605                         err = ext3_journal_dirty_data(handle, bh);
2606                         mark_buffer_dirty(bh);
2607                 }
2608                 brelse(bh);
2609                 if (err)
2610                         goto out;
2611                 offset = 0;
2612                 towrite -= tocopy;
2613                 data += tocopy;
2614                 blk++;
2615         }
2616 out:
2617         if (len == towrite)
2618                 return err;
2619         if (inode->i_size < off+len-towrite) {
2620                 i_size_write(inode, off+len-towrite);
2621                 EXT3_I(inode)->i_disksize = inode->i_size;
2622         }
2623         inode->i_version++;
2624         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2625         ext3_mark_inode_dirty(handle, inode);
2626         up(&inode->i_sem);
2627         return len - towrite;
2628 }
2629
2630 #endif
2631
2632 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2633         int flags, const char *dev_name, void *data)
2634 {
2635         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2636 }
2637
2638 static struct file_system_type ext3_fs_type = {
2639         .owner          = THIS_MODULE,
2640         .name           = "ext3",
2641         .get_sb         = ext3_get_sb,
2642         .kill_sb        = kill_block_super,
2643         .fs_flags       = FS_REQUIRES_DEV,
2644 };
2645
2646 static int __init init_ext3_fs(void)
2647 {
2648         int err = init_ext3_xattr();
2649         if (err)
2650                 return err;
2651         err = init_inodecache();
2652         if (err)
2653                 goto out1;
2654         err = register_filesystem(&ext3_fs_type);
2655         if (err)
2656                 goto out;
2657         return 0;
2658 out:
2659         destroy_inodecache();
2660 out1:
2661         exit_ext3_xattr();
2662         return err;
2663 }
2664
2665 static void __exit exit_ext3_fs(void)
2666 {
2667         unregister_filesystem(&ext3_fs_type);
2668         destroy_inodecache();
2669         exit_ext3_xattr();
2670 }
2671
2672 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2673 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2674 MODULE_LICENSE("GPL");
2675 module_init(init_ext3_fs)
2676 module_exit(exit_ext3_fs)