fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39
  40 #include <asm/uaccess.h>
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44 #include "namei.h"
  45
  46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
  47                              unsigned long journal_devnum);
  48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  49                                int);
  50 static void ext3_commit_super (struct super_block * sb,
  51                                struct ext3_super_block * es,
  52                                int sync);
  53 static void ext3_mark_recovery_complete(struct super_block * sb,
  54                                         struct ext3_super_block * es);
  55 static void ext3_clear_journal_err(struct super_block * sb,
  56                                    struct ext3_super_block * es);
  57 static int ext3_sync_fs(struct super_block *sb, int wait);
  58 static const char *ext3_decode_error(struct super_block * sb, int errno,
  59                                      char nbuf[16]);
  60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  61 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  62 static void ext3_unlockfs(struct super_block *sb);
  63 static void ext3_write_super (struct super_block * sb);
  64 static void ext3_write_super_lockfs(struct super_block *sb);
  65
  66 /*
  67  * Wrappers for journal_start/end.
  68  *
  69  * The only special thing we need to do here is to make sure that all
  70  * journal_end calls result in the superblock being marked dirty, so
  71  * that sync() will call the filesystem's write_super callback if
  72  * appropriate.
  73  */
  74 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  75 {
  76         journal_t *journal;
  77
  78         if (sb->s_flags & MS_RDONLY)
  79                 return ERR_PTR(-EROFS);
  80
  81         /* Special case here: if the journal has aborted behind our
  82          * backs (eg. EIO in the commit thread), then we still need to
  83          * take the FS itself readonly cleanly. */
  84         journal = EXT3_SB(sb)->s_journal;
  85         if (is_journal_aborted(journal)) {
  86                 ext3_abort(sb, __FUNCTION__,
  87                            "Detected aborted journal");
  88                 return ERR_PTR(-EROFS);
  89         }
  90
  91         return journal_start(journal, nblocks);
  92 }
  93
  94 /*
  95  * The only special thing we need to do here is to make sure that all
  96  * journal_stop calls result in the superblock being marked dirty, so
  97  * that sync() will call the filesystem's write_super callback if
  98  * appropriate.
  99  */
 100 int __ext3_journal_stop(const char *where, handle_t *handle)
 101 {
 102         struct super_block *sb;
 103         int err;
 104         int rc;
 105
 106         sb = handle->h_transaction->t_journal->j_private;
 107         err = handle->h_err;
 108         rc = journal_stop(handle);
 109
 110         if (!err)
 111                 err = rc;
 112         if (err)
 113                 __ext3_std_error(sb, where, err);
 114         return err;
 115 }
 116
 117 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 118                 struct buffer_head *bh, handle_t *handle, int err)
 119 {
 120         char nbuf[16];
 121         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 122
 123         if (bh)
 124                 BUFFER_TRACE(bh, "abort");
 125
 126         if (!handle->h_err)
 127                 handle->h_err = err;
 128
 129         if (is_handle_aborted(handle))
 130                 return;
 131
 132         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 133                caller, errstr, err_fn);
 134
 135         journal_abort_handle(handle);
 136 }
 137
 138 /* Deal with the reporting of failure conditions on a filesystem such as
 139  * inconsistencies detected or read IO failures.
 140  *
 141  * On ext2, we can store the error state of the filesystem in the
 142  * superblock.  That is not possible on ext3, because we may have other
 143  * write ordering constraints on the superblock which prevent us from
 144  * writing it out straight away; and given that the journal is about to
 145  * be aborted, we can't rely on the current, or future, transactions to
 146  * write out the superblock safely.
 147  *
 148  * We'll just use the journal_abort() error code to record an error in
 149  * the journal instead.  On recovery, the journal will compain about
 150  * that error until we've noted it down and cleared it.
 151  */
 152
 153 static void ext3_handle_error(struct super_block *sb)
 154 {
 155         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 156
 157         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 158         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 159
 160         if (sb->s_flags & MS_RDONLY)
 161                 return;
 162
 163         if (test_opt (sb, ERRORS_RO)) {
 164                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 165                 sb->s_flags |= MS_RDONLY;
 166         } else {
 167                 journal_t *journal = EXT3_SB(sb)->s_journal;
 168
 169                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 170                 if (journal)
 171                         journal_abort(journal, -EIO);
 172         }
 173         if (test_opt(sb, ERRORS_PANIC))
 174                 panic("EXT3-fs (device %s): panic forced after error\n",
 175                         sb->s_id);
 176         ext3_commit_super(sb, es, 1);
 177 }
 178
 179 void ext3_error (struct super_block * sb, const char * function,
 180                  const char * fmt, ...)
 181 {
 182         va_list args;
 183
 184         va_start(args, fmt);
 185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 186         vprintk(fmt, args);
 187         printk("\n");
 188         va_end(args);
 189
 190         ext3_handle_error(sb);
 191 }
 192
 193 static const char *ext3_decode_error(struct super_block * sb, int errno,
 194                                      char nbuf[16])
 195 {
 196         char *errstr = NULL;
 197
 198         switch (errno) {
 199         case -EIO:
 200                 errstr = "IO failure";
 201                 break;
 202         case -ENOMEM:
 203                 errstr = "Out of memory";
 204                 break;
 205         case -EROFS:
 206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 207                         errstr = "Journal has aborted";
 208                 else
 209                         errstr = "Readonly filesystem";
 210                 break;
 211         default:
 212                 /* If the caller passed in an extra buffer for unknown
 213                  * errors, textualise them now.  Else we just return
 214                  * NULL. */
 215                 if (nbuf) {
 216                         /* Check for truncated error codes... */
 217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 218                                 errstr = nbuf;
 219                 }
 220                 break;
 221         }
 222
 223         return errstr;
 224 }
 225
 226 /* __ext3_std_error decodes expected errors from journaling functions
 227  * automatically and invokes the appropriate error response.  */
 228
 229 void __ext3_std_error (struct super_block * sb, const char * function,
 230                        int errno)
 231 {
 232         char nbuf[16];
 233         const char *errstr;
 234
 235         /* Special case: if the error is EROFS, and we're not already
 236          * inside a transaction, then there's really no point in logging
 237          * an error. */
 238         if (errno == -EROFS && journal_current_handle() == NULL &&
 239             (sb->s_flags & MS_RDONLY))
 240                 return;
 241
 242         errstr = ext3_decode_error(sb, errno, nbuf);
 243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 244                 sb->s_id, function, errstr);
 245
 246         ext3_handle_error(sb);
 247 }
 248
 249 /*
 250  * ext3_abort is a much stronger failure handler than ext3_error.  The
 251  * abort function may be used to deal with unrecoverable failures such
 252  * as journal IO errors or ENOMEM at a critical moment in log management.
 253  *
 254  * We unconditionally force the filesystem into an ABORT|READONLY state,
 255  * unless the error response on the fs has been set to panic in which
 256  * case we take the easy way out and panic immediately.
 257  */
 258
 259 void ext3_abort (struct super_block * sb, const char * function,
 260                  const char * fmt, ...)
 261 {
 262         va_list args;
 263
 264         printk (KERN_CRIT "ext3_abort called.\n");
 265
 266         va_start(args, fmt);
 267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 268         vprintk(fmt, args);
 269         printk("\n");
 270         va_end(args);
 271
 272         if (test_opt(sb, ERRORS_PANIC))
 273                 panic("EXT3-fs panic from previous error\n");
 274
 275         if (sb->s_flags & MS_RDONLY)
 276                 return;
 277
 278         printk(KERN_CRIT "Remounting filesystem read-only\n");
 279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 280         sb->s_flags |= MS_RDONLY;
 281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 283 }
 284
 285 void ext3_warning (struct super_block * sb, const char * function,
 286                    const char * fmt, ...)
 287 {
 288         va_list args;
 289
 290         va_start(args, fmt);
 291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 292                sb->s_id, function);
 293         vprintk(fmt, args);
 294         printk("\n");
 295         va_end(args);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = NULL;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        NEXT_ORPHAN(inode));
 384         }
 385 }
 386
 387 static void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 408         percpu_counter_destroy(&sbi->s_dirs_counter);
 409         brelse(sbi->s_sbh);
 410 #ifdef CONFIG_QUOTA
 411         for (i = 0; i < MAXQUOTAS; i++)
 412                 kfree(sbi->s_qf_names[i]);
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->i_block_alloc_info = NULL;
 456         ei->vfs_inode.i_version = 1;
 457         return &ei->vfs_inode;
 458 }
 459
 460 static void ext3_destroy_inode(struct inode *inode)
 461 {
 462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 463 }
 464
 465 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 466 {
 467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 468
 469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 470             SLAB_CTOR_CONSTRUCTOR) {
 471                 INIT_LIST_HEAD(&ei->i_orphan);
 472 #ifdef CONFIG_EXT3_FS_XATTR
 473                 init_rwsem(&ei->xattr_sem);
 474 #endif
 475                 mutex_init(&ei->truncate_mutex);
 476                 inode_init_once(&ei->vfs_inode);
 477         }
 478 }
 479
 480 static int init_inodecache(void)
 481 {
 482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 483                                              sizeof(struct ext3_inode_info),
 484                                              0, (SLAB_RECLAIM_ACCOUNT|
 485                                                 SLAB_MEM_SPREAD),
 486                                              init_once, NULL);
 487         if (ext3_inode_cachep == NULL)
 488                 return -ENOMEM;
 489         return 0;
 490 }
 491
 492 static void destroy_inodecache(void)
 493 {
 494         if (kmem_cache_destroy(ext3_inode_cachep))
 495                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 496 }
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
 501 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 502        if (EXT3_I(inode)->i_acl &&
 503            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 504                posix_acl_release(EXT3_I(inode)->i_acl);
 505                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 506        }
 507        if (EXT3_I(inode)->i_default_acl &&
 508            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 509                posix_acl_release(EXT3_I(inode)->i_default_acl);
 510                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 511        }
 512 #endif
 513         ext3_discard_reservation(inode);
 514         EXT3_I(inode)->i_block_alloc_info = NULL;
 515         kfree(rsv);
 516 }
 517
 518 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
 519 {
 520 #if defined(CONFIG_QUOTA)
 521         struct ext3_sb_info *sbi = EXT3_SB(sb);
 522
 523         if (sbi->s_jquota_fmt)
 524                 seq_printf(seq, ",jqfmt=%s",
 525                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 526
 527         if (sbi->s_qf_names[USRQUOTA])
 528                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 529
 530         if (sbi->s_qf_names[GRPQUOTA])
 531                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 532
 533         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
 534                 seq_puts(seq, ",usrquota");
 535
 536         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
 537                 seq_puts(seq, ",grpquota");
 538 #endif
 539 }
 540
 541 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
 542 {
 543         struct super_block *sb = vfs->mnt_sb;
 544
 545         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
 546                 seq_puts(seq, ",data=journal");
 547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
 548                 seq_puts(seq, ",data=ordered");
 549         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
 550                 seq_puts(seq, ",data=writeback");
 551
 552         ext3_show_quota_options(seq, sb);
 553
 554         return 0;
 555 }
 556
 557 #ifdef CONFIG_QUOTA
 558 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 559 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 560
 561 static int ext3_dquot_initialize(struct inode *inode, int type);
 562 static int ext3_dquot_drop(struct inode *inode);
 563 static int ext3_write_dquot(struct dquot *dquot);
 564 static int ext3_acquire_dquot(struct dquot *dquot);
 565 static int ext3_release_dquot(struct dquot *dquot);
 566 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 567 static int ext3_write_info(struct super_block *sb, int type);
 568 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 569 static int ext3_quota_on_mount(struct super_block *sb, int type);
 570 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 571                                size_t len, loff_t off);
 572 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 573                                 const char *data, size_t len, loff_t off);
 574
 575 static struct dquot_operations ext3_quota_operations = {
 576         .initialize     = ext3_dquot_initialize,
 577         .drop           = ext3_dquot_drop,
 578         .alloc_space    = dquot_alloc_space,
 579         .alloc_inode    = dquot_alloc_inode,
 580         .free_space     = dquot_free_space,
 581         .free_inode     = dquot_free_inode,
 582         .transfer       = dquot_transfer,
 583         .write_dquot    = ext3_write_dquot,
 584         .acquire_dquot  = ext3_acquire_dquot,
 585         .release_dquot  = ext3_release_dquot,
 586         .mark_dirty     = ext3_mark_dquot_dirty,
 587         .write_info     = ext3_write_info
 588 };
 589
 590 static struct quotactl_ops ext3_qctl_operations = {
 591         .quota_on       = ext3_quota_on,
 592         .quota_off      = vfs_quota_off,
 593         .quota_sync     = vfs_quota_sync,
 594         .get_info       = vfs_get_dqinfo,
 595         .set_info       = vfs_set_dqinfo,
 596         .get_dqblk      = vfs_get_dqblk,
 597         .set_dqblk      = vfs_set_dqblk
 598 };
 599 #endif
 600
 601 static struct super_operations ext3_sops = {
 602         .alloc_inode    = ext3_alloc_inode,
 603         .destroy_inode  = ext3_destroy_inode,
 604         .read_inode     = ext3_read_inode,
 605         .write_inode    = ext3_write_inode,
 606         .dirty_inode    = ext3_dirty_inode,
 607         .delete_inode   = ext3_delete_inode,
 608         .put_super      = ext3_put_super,
 609         .write_super    = ext3_write_super,
 610         .sync_fs        = ext3_sync_fs,
 611         .write_super_lockfs = ext3_write_super_lockfs,
 612         .unlockfs       = ext3_unlockfs,
 613         .statfs         = ext3_statfs,
 614         .remount_fs     = ext3_remount,
 615         .clear_inode    = ext3_clear_inode,
 616         .show_options   = ext3_show_options,
 617 #ifdef CONFIG_QUOTA
 618         .quota_read     = ext3_quota_read,
 619         .quota_write    = ext3_quota_write,
 620 #endif
 621 };
 622
 623 static struct dentry *ext3_get_dentry(struct super_block *sb, void *vobjp)
 624 {
 625         __u32 *objp = vobjp;
 626         unsigned long ino = objp[0];
 627         __u32 generation = objp[1];
 628         struct inode *inode;
 629         struct dentry *result;
 630
 631         if (ino != EXT3_ROOT_INO && ino < EXT3_FIRST_INO(sb))
 632                 return ERR_PTR(-ESTALE);
 633         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
 634                 return ERR_PTR(-ESTALE);
 635
 636         /* iget isn't really right if the inode is currently unallocated!!
 637          * ext3_read_inode currently does appropriate checks, but
 638          * it might be "neater" to call ext3_get_inode first and check
 639          * if the inode is valid.....
 640          */
 641         inode = iget(sb, ino);
 642         if (inode == NULL)
 643                 return ERR_PTR(-ENOMEM);
 644         if (is_bad_inode(inode)
 645             || (generation && inode->i_generation != generation)
 646                 ) {
 647                 /* we didn't find the right inode.. */
 648                 iput(inode);
 649                 return ERR_PTR(-ESTALE);
 650         }
 651         /* now to find a dentry.
 652          * If possible, get a well-connected one
 653          */
 654         result = d_alloc_anon(inode);
 655         if (!result) {
 656                 iput(inode);
 657                 return ERR_PTR(-ENOMEM);
 658         }
 659         return result;
 660 }
 661
 662 static struct export_operations ext3_export_ops = {
 663         .get_parent = ext3_get_parent,
 664         .get_dentry = ext3_get_dentry,
 665 };
 666
 667 enum {
 668         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 669         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 670         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 671         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 672         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
 673         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 674         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 675         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 676         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 677         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 678         Opt_grpquota, Opt_tagxid
 679 };
 680
 681 static match_table_t tokens = {
 682         {Opt_bsd_df, "bsddf"},
 683         {Opt_minix_df, "minixdf"},
 684         {Opt_grpid, "grpid"},
 685         {Opt_grpid, "bsdgroups"},
 686         {Opt_nogrpid, "nogrpid"},
 687         {Opt_nogrpid, "sysvgroups"},
 688         {Opt_resgid, "resgid=%u"},
 689         {Opt_resuid, "resuid=%u"},
 690         {Opt_sb, "sb=%u"},
 691         {Opt_err_cont, "errors=continue"},
 692         {Opt_err_panic, "errors=panic"},
 693         {Opt_err_ro, "errors=remount-ro"},
 694         {Opt_nouid32, "nouid32"},
 695         {Opt_nocheck, "nocheck"},
 696         {Opt_nocheck, "check=none"},
 697         {Opt_debug, "debug"},
 698         {Opt_oldalloc, "oldalloc"},
 699         {Opt_orlov, "orlov"},
 700         {Opt_user_xattr, "user_xattr"},
 701         {Opt_nouser_xattr, "nouser_xattr"},
 702         {Opt_acl, "acl"},
 703         {Opt_noacl, "noacl"},
 704         {Opt_reservation, "reservation"},
 705         {Opt_noreservation, "noreservation"},
 706         {Opt_noload, "noload"},
 707         {Opt_nobh, "nobh"},
 708         {Opt_commit, "commit=%u"},
 709         {Opt_journal_update, "journal=update"},
 710         {Opt_journal_inum, "journal=%u"},
 711         {Opt_journal_dev, "journal_dev=%u"},
 712         {Opt_abort, "abort"},
 713         {Opt_data_journal, "data=journal"},
 714         {Opt_data_ordered, "data=ordered"},
 715         {Opt_data_writeback, "data=writeback"},
 716         {Opt_offusrjquota, "usrjquota="},
 717         {Opt_usrjquota, "usrjquota=%s"},
 718         {Opt_offgrpjquota, "grpjquota="},
 719         {Opt_grpjquota, "grpjquota=%s"},
 720         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 721         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 722         {Opt_grpquota, "grpquota"},
 723         {Opt_noquota, "noquota"},
 724         {Opt_quota, "quota"},
 725         {Opt_usrquota, "usrquota"},
 726         {Opt_barrier, "barrier=%u"},
 727         {Opt_tagxid, "tagxid"},
 728         {Opt_err, NULL},
 729         {Opt_resize, "resize"},
 730 };
 731
 732 static unsigned long get_sb_block(void **data)
 733 {
 734         unsigned long   sb_block;
 735         char            *options = (char *) *data;
 736
 737         if (!options || strncmp(options, "sb=", 3) != 0)
 738                 return 1;       /* Default location */
 739         options += 3;
 740         sb_block = simple_strtoul(options, &options, 0);
 741         if (*options && *options != ',') {
 742                 printk("EXT3-fs: Invalid sb specification: %s\n",
 743                        (char *) *data);
 744                 return 1;
 745         }
 746         if (*options == ',')
 747                 options++;
 748         *data = (void *) options;
 749         return sb_block;
 750 }
 751
 752 static int parse_options (char *options, struct super_block *sb,
 753                           unsigned long *inum, unsigned long *journal_devnum,
 754                           unsigned long *n_blocks_count, int is_remount)
 755 {
 756         struct ext3_sb_info *sbi = EXT3_SB(sb);
 757         char * p;
 758         substring_t args[MAX_OPT_ARGS];
 759         int data_opt = 0;
 760         int option;
 761 #ifdef CONFIG_QUOTA
 762         int qtype;
 763         char *qname;
 764 #endif
 765
 766         if (!options)
 767                 return 1;
 768
 769         while ((p = strsep (&options, ",")) != NULL) {
 770                 int token;
 771                 if (!*p)
 772                         continue;
 773
 774                 token = match_token(p, tokens, args);
 775                 switch (token) {
 776                 case Opt_bsd_df:
 777                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 778                         break;
 779                 case Opt_minix_df:
 780                         set_opt (sbi->s_mount_opt, MINIX_DF);
 781                         break;
 782                 case Opt_grpid:
 783                         set_opt (sbi->s_mount_opt, GRPID);
 784                         break;
 785                 case Opt_nogrpid:
 786                         clear_opt (sbi->s_mount_opt, GRPID);
 787                         break;
 788                 case Opt_resuid:
 789                         if (match_int(&args[0], &option))
 790                                 return 0;
 791                         sbi->s_resuid = option;
 792                         break;
 793                 case Opt_resgid:
 794                         if (match_int(&args[0], &option))
 795                                 return 0;
 796                         sbi->s_resgid = option;
 797                         break;
 798                 case Opt_sb:
 799                         /* handled by get_sb_block() instead of here */
 800                         /* *sb_block = match_int(&args[0]); */
 801                         break;
 802                 case Opt_err_panic:
 803                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 804                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 805                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 806                         break;
 807                 case Opt_err_ro:
 808                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 809                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 810                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 811                         break;
 812                 case Opt_err_cont:
 813                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 814                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 815                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 816                         break;
 817                 case Opt_nouid32:
 818                         set_opt (sbi->s_mount_opt, NO_UID32);
 819                         break;
 820 #ifndef CONFIG_INOXID_NONE
 821                 case Opt_tagxid:
 822                         set_opt (sbi->s_mount_opt, TAGXID);
 823                         break;
 824 #endif
 825                 case Opt_nocheck:
 826                         clear_opt (sbi->s_mount_opt, CHECK);
 827                         break;
 828                 case Opt_debug:
 829                         set_opt (sbi->s_mount_opt, DEBUG);
 830                         break;
 831                 case Opt_oldalloc:
 832                         set_opt (sbi->s_mount_opt, OLDALLOC);
 833                         break;
 834                 case Opt_orlov:
 835                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 836                         break;
 837 #ifdef CONFIG_EXT3_FS_XATTR
 838                 case Opt_user_xattr:
 839                         set_opt (sbi->s_mount_opt, XATTR_USER);
 840                         break;
 841                 case Opt_nouser_xattr:
 842                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 843                         break;
 844 #else
 845                 case Opt_user_xattr:
 846                 case Opt_nouser_xattr:
 847                         printk("EXT3 (no)user_xattr options not supported\n");
 848                         break;
 849 #endif
 850 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 851                 case Opt_acl:
 852                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 853                         break;
 854                 case Opt_noacl:
 855                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 856                         break;
 857 #else
 858                 case Opt_acl:
 859                 case Opt_noacl:
 860                         printk("EXT3 (no)acl options not supported\n");
 861                         break;
 862 #endif
 863                 case Opt_reservation:
 864                         set_opt(sbi->s_mount_opt, RESERVATION);
 865                         break;
 866                 case Opt_noreservation:
 867                         clear_opt(sbi->s_mount_opt, RESERVATION);
 868                         break;
 869                 case Opt_journal_update:
 870                         /* @@@ FIXME */
 871                         /* Eventually we will want to be able to create
 872                            a journal file here.  For now, only allow the
 873                            user to specify an existing inode to be the
 874                            journal file. */
 875                         if (is_remount) {
 876                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 877                                        "journal on remount\n");
 878                                 return 0;
 879                         }
 880                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 881                         break;
 882                 case Opt_journal_inum:
 883                         if (is_remount) {
 884                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 885                                        "journal on remount\n");
 886                                 return 0;
 887                         }
 888                         if (match_int(&args[0], &option))
 889                                 return 0;
 890                         *inum = option;
 891                         break;
 892                 case Opt_journal_dev:
 893                         if (is_remount) {
 894                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 895                                        "journal on remount\n");
 896                                 return 0;
 897                         }
 898                         if (match_int(&args[0], &option))
 899                                 return 0;
 900                         *journal_devnum = option;
 901                         break;
 902                 case Opt_noload:
 903                         set_opt (sbi->s_mount_opt, NOLOAD);
 904                         break;
 905                 case Opt_commit:
 906                         if (match_int(&args[0], &option))
 907                                 return 0;
 908                         if (option < 0)
 909                                 return 0;
 910                         if (option == 0)
 911                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 912                         sbi->s_commit_interval = HZ * option;
 913                         break;
 914                 case Opt_data_journal:
 915                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 916                         goto datacheck;
 917                 case Opt_data_ordered:
 918                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 919                         goto datacheck;
 920                 case Opt_data_writeback:
 921                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 922                 datacheck:
 923                         if (is_remount) {
 924                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 925                                                 != data_opt) {
 926                                         printk(KERN_ERR
 927                                                 "EXT3-fs: cannot change data "
 928                                                 "mode on remount\n");
 929                                         return 0;
 930                                 }
 931                         } else {
 932                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 933                                 sbi->s_mount_opt |= data_opt;
 934                         }
 935                         break;
 936 #ifdef CONFIG_QUOTA
 937                 case Opt_usrjquota:
 938                         qtype = USRQUOTA;
 939                         goto set_qf_name;
 940                 case Opt_grpjquota:
 941                         qtype = GRPQUOTA;
 942 set_qf_name:
 943                         if (sb_any_quota_enabled(sb)) {
 944                                 printk(KERN_ERR
 945                                         "EXT3-fs: Cannot change journalled "
 946                                         "quota options when quota turned on.\n");
 947                                 return 0;
 948                         }
 949                         qname = match_strdup(&args[0]);
 950                         if (!qname) {
 951                                 printk(KERN_ERR
 952                                         "EXT3-fs: not enough memory for "
 953                                         "storing quotafile name.\n");
 954                                 return 0;
 955                         }
 956                         if (sbi->s_qf_names[qtype] &&
 957                             strcmp(sbi->s_qf_names[qtype], qname)) {
 958                                 printk(KERN_ERR
 959                                         "EXT3-fs: %s quota file already "
 960                                         "specified.\n", QTYPE2NAME(qtype));
 961                                 kfree(qname);
 962                                 return 0;
 963                         }
 964                         sbi->s_qf_names[qtype] = qname;
 965                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 966                                 printk(KERN_ERR
 967                                         "EXT3-fs: quotafile must be on "
 968                                         "filesystem root.\n");
 969                                 kfree(sbi->s_qf_names[qtype]);
 970                                 sbi->s_qf_names[qtype] = NULL;
 971                                 return 0;
 972                         }
 973                         set_opt(sbi->s_mount_opt, QUOTA);
 974                         break;
 975                 case Opt_offusrjquota:
 976                         qtype = USRQUOTA;
 977                         goto clear_qf_name;
 978                 case Opt_offgrpjquota:
 979                         qtype = GRPQUOTA;
 980 clear_qf_name:
 981                         if (sb_any_quota_enabled(sb)) {
 982                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 983                                         "journalled quota options when "
 984                                         "quota turned on.\n");
 985                                 return 0;
 986                         }
 987                         /*
 988                          * The space will be released later when all options
 989                          * are confirmed to be correct
 990                          */
 991                         sbi->s_qf_names[qtype] = NULL;
 992                         break;
 993                 case Opt_jqfmt_vfsold:
 994                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 995                         break;
 996                 case Opt_jqfmt_vfsv0:
 997                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 998                         break;
 999                 case Opt_quota:
1000                 case Opt_usrquota:
1001                         set_opt(sbi->s_mount_opt, QUOTA);
1002                         set_opt(sbi->s_mount_opt, USRQUOTA);
1003                         break;
1004                 case Opt_grpquota:
1005                         set_opt(sbi->s_mount_opt, QUOTA);
1006                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1007                         break;
1008                 case Opt_noquota:
1009                         if (sb_any_quota_enabled(sb)) {
1010                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1011                                         "options when quota turned on.\n");
1012                                 return 0;
1013                         }
1014                         clear_opt(sbi->s_mount_opt, QUOTA);
1015                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1016                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1017                         break;
1018 #else
1019                 case Opt_quota:
1020                 case Opt_usrquota:
1021                 case Opt_grpquota:
1022                 case Opt_usrjquota:
1023                 case Opt_grpjquota:
1024                 case Opt_offusrjquota:
1025                 case Opt_offgrpjquota:
1026                 case Opt_jqfmt_vfsold:
1027                 case Opt_jqfmt_vfsv0:
1028                         printk(KERN_ERR
1029                                 "EXT3-fs: journalled quota options not "
1030                                 "supported.\n");
1031                         break;
1032                 case Opt_noquota:
1033                         break;
1034 #endif
1035                 case Opt_abort:
1036                         set_opt(sbi->s_mount_opt, ABORT);
1037                         break;
1038                 case Opt_barrier:
1039                         if (match_int(&args[0], &option))
1040                                 return 0;
1041                         if (option)
1042                                 set_opt(sbi->s_mount_opt, BARRIER);
1043                         else
1044                                 clear_opt(sbi->s_mount_opt, BARRIER);
1045                         break;
1046                 case Opt_ignore:
1047                         break;
1048                 case Opt_resize:
1049                         if (!is_remount) {
1050                                 printk("EXT3-fs: resize option only available "
1051                                         "for remount\n");
1052                                 return 0;
1053                         }
1054                         if (match_int(&args[0], &option) != 0)
1055                                 return 0;
1056                         *n_blocks_count = option;
1057                         break;
1058                 case Opt_nobh:
1059                         set_opt(sbi->s_mount_opt, NOBH);
1060                         break;
1061                 default:
1062                         printk (KERN_ERR
1063                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1064                                 "or missing value\n", p);
1065                         return 0;
1066                 }
1067         }
1068 #ifdef CONFIG_QUOTA
1069         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1070                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1071                      sbi->s_qf_names[USRQUOTA])
1072                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1073
1074                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1075                      sbi->s_qf_names[GRPQUOTA])
1076                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1077
1078                 if ((sbi->s_qf_names[USRQUOTA] &&
1079                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1080                     (sbi->s_qf_names[GRPQUOTA] &&
1081                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1082                         printk(KERN_ERR "EXT3-fs: old and new quota "
1083                                         "format mixing.\n");
1084                         return 0;
1085                 }
1086
1087                 if (!sbi->s_jquota_fmt) {
1088                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1089                                         "not specified.\n");
1090                         return 0;
1091                 }
1092         } else {
1093                 if (sbi->s_jquota_fmt) {
1094                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1095                                         "specified with no journalling "
1096                                         "enabled.\n");
1097                         return 0;
1098                 }
1099         }
1100 #endif
1101         return 1;
1102 }
1103
1104 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1105                             int read_only)
1106 {
1107         struct ext3_sb_info *sbi = EXT3_SB(sb);
1108         int res = 0;
1109
1110         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1111                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1112                         "forcing read-only mode\n");
1113                 res = MS_RDONLY;
1114         }
1115         if (read_only)
1116                 return res;
1117         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1118                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1119                         "running e2fsck is recommended\n");
1120         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1121                 printk (KERN_WARNING
1122                         "EXT3-fs warning: mounting fs with errors, "
1123                         "running e2fsck is recommended\n");
1124         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1125                  le16_to_cpu(es->s_mnt_count) >=
1126                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1127                 printk (KERN_WARNING
1128                         "EXT3-fs warning: maximal mount count reached, "
1129                         "running e2fsck is recommended\n");
1130         else if (le32_to_cpu(es->s_checkinterval) &&
1131                 (le32_to_cpu(es->s_lastcheck) +
1132                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1133                 printk (KERN_WARNING
1134                         "EXT3-fs warning: checktime reached, "
1135                         "running e2fsck is recommended\n");
1136 #if 0
1137                 /* @@@ We _will_ want to clear the valid bit if we find
1138                    inconsistencies, to force a fsck at reboot.  But for
1139                    a plain journaled filesystem we can keep it set as
1140                    valid forever! :) */
1141         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1142 #endif
1143         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1144                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1145         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1146         es->s_mtime = cpu_to_le32(get_seconds());
1147         ext3_update_dynamic_rev(sb);
1148         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1149
1150         ext3_commit_super(sb, es, 1);
1151         if (test_opt(sb, DEBUG))
1152                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1153                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1154                         sb->s_blocksize,
1155                         sbi->s_groups_count,
1156                         EXT3_BLOCKS_PER_GROUP(sb),
1157                         EXT3_INODES_PER_GROUP(sb),
1158                         sbi->s_mount_opt);
1159
1160         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1161         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1162                 char b[BDEVNAME_SIZE];
1163
1164                 printk("external journal on %s\n",
1165                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1166         } else {
1167                 printk("internal journal\n");
1168         }
1169         return res;
1170 }
1171
1172 /* Called at mount-time, super-block is locked */
1173 static int ext3_check_descriptors (struct super_block * sb)
1174 {
1175         struct ext3_sb_info *sbi = EXT3_SB(sb);
1176         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1177         struct ext3_group_desc * gdp = NULL;
1178         int desc_block = 0;
1179         int i;
1180
1181         ext3_debug ("Checking group descriptors");
1182
1183         for (i = 0; i < sbi->s_groups_count; i++)
1184         {
1185                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1186                         gdp = (struct ext3_group_desc *)
1187                                         sbi->s_group_desc[desc_block++]->b_data;
1188                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1189                     le32_to_cpu(gdp->bg_block_bitmap) >=
1190                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1191                 {
1192                         ext3_error (sb, "ext3_check_descriptors",
1193                                     "Block bitmap for group %d"
1194                                     " not in group (block %lu)!",
1195                                     i, (unsigned long)
1196                                         le32_to_cpu(gdp->bg_block_bitmap));
1197                         return 0;
1198                 }
1199                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1200                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1201                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1202                 {
1203                         ext3_error (sb, "ext3_check_descriptors",
1204                                     "Inode bitmap for group %d"
1205                                     " not in group (block %lu)!",
1206                                     i, (unsigned long)
1207                                         le32_to_cpu(gdp->bg_inode_bitmap));
1208                         return 0;
1209                 }
1210                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1211                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1212                     block + EXT3_BLOCKS_PER_GROUP(sb))
1213                 {
1214                         ext3_error (sb, "ext3_check_descriptors",
1215                                     "Inode table for group %d"
1216                                     " not in group (block %lu)!",
1217                                     i, (unsigned long)
1218                                         le32_to_cpu(gdp->bg_inode_table));
1219                         return 0;
1220                 }
1221                 block += EXT3_BLOCKS_PER_GROUP(sb);
1222                 gdp++;
1223         }
1224
1225         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1226         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1227         return 1;
1228 }
1229
1230
1231 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1232  * the superblock) which were deleted from all directories, but held open by
1233  * a process at the time of a crash.  We walk the list and try to delete these
1234  * inodes at recovery time (only with a read-write filesystem).
1235  *
1236  * In order to keep the orphan inode chain consistent during traversal (in
1237  * case of crash during recovery), we link each inode into the superblock
1238  * orphan list_head and handle it the same way as an inode deletion during
1239  * normal operation (which journals the operations for us).
1240  *
1241  * We only do an iget() and an iput() on each inode, which is very safe if we
1242  * accidentally point at an in-use or already deleted inode.  The worst that
1243  * can happen in this case is that we get a "bit already cleared" message from
1244  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1245  * e2fsck was run on this filesystem, and it must have already done the orphan
1246  * inode cleanup for us, so we can safely abort without any further action.
1247  */
1248 static void ext3_orphan_cleanup (struct super_block * sb,
1249                                  struct ext3_super_block * es)
1250 {
1251         unsigned int s_flags = sb->s_flags;
1252         int nr_orphans = 0, nr_truncates = 0;
1253 #ifdef CONFIG_QUOTA
1254         int i;
1255 #endif
1256         if (!es->s_last_orphan) {
1257                 jbd_debug(4, "no orphan inodes to clean up\n");
1258                 return;
1259         }
1260
1261         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1262                 if (es->s_last_orphan)
1263                         jbd_debug(1, "Errors on filesystem, "
1264                                   "clearing orphan list.\n");
1265                 es->s_last_orphan = 0;
1266                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1267                 return;
1268         }
1269
1270         if (s_flags & MS_RDONLY) {
1271                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1272                        sb->s_id);
1273                 sb->s_flags &= ~MS_RDONLY;
1274         }
1275 #ifdef CONFIG_QUOTA
1276         /* Needed for iput() to work correctly and not trash data */
1277         sb->s_flags |= MS_ACTIVE;
1278         /* Turn on quotas so that they are updated correctly */
1279         for (i = 0; i < MAXQUOTAS; i++) {
1280                 if (EXT3_SB(sb)->s_qf_names[i]) {
1281                         int ret = ext3_quota_on_mount(sb, i);
1282                         if (ret < 0)
1283                                 printk(KERN_ERR
1284                                         "EXT3-fs: Cannot turn on journalled "
1285                                         "quota: error %d\n", ret);
1286                 }
1287         }
1288 #endif
1289
1290         while (es->s_last_orphan) {
1291                 struct inode *inode;
1292
1293                 if (!(inode =
1294                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1295                         es->s_last_orphan = 0;
1296                         break;
1297                 }
1298
1299                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1300                 DQUOT_INIT(inode);
1301                 if (inode->i_nlink) {
1302                         printk(KERN_DEBUG
1303                                 "%s: truncating inode %ld to %Ld bytes\n",
1304                                 __FUNCTION__, inode->i_ino, inode->i_size);
1305                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1306                                   inode->i_ino, inode->i_size);
1307                         ext3_truncate(inode);
1308                         nr_truncates++;
1309                 } else {
1310                         printk(KERN_DEBUG
1311                                 "%s: deleting unreferenced inode %ld\n",
1312                                 __FUNCTION__, inode->i_ino);
1313                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1314                                   inode->i_ino);
1315                         nr_orphans++;
1316                 }
1317                 iput(inode);  /* The delete magic happens here! */
1318         }
1319
1320 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1321
1322         if (nr_orphans)
1323                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1324                        sb->s_id, PLURAL(nr_orphans));
1325         if (nr_truncates)
1326                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1327                        sb->s_id, PLURAL(nr_truncates));
1328 #ifdef CONFIG_QUOTA
1329         /* Turn quotas off */
1330         for (i = 0; i < MAXQUOTAS; i++) {
1331                 if (sb_dqopt(sb)->files[i])
1332                         vfs_quota_off(sb, i);
1333         }
1334 #endif
1335         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1336 }
1337
1338 #define log2(n) ffz(~(n))
1339
1340 /*
1341  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1342  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1343  * We need to be 1 filesystem block less than the 2^32 sector limit.
1344  */
1345 static loff_t ext3_max_size(int bits)
1346 {
1347         loff_t res = EXT3_NDIR_BLOCKS;
1348         /* This constant is calculated to be the largest file size for a
1349          * dense, 4k-blocksize file such that the total number of
1350          * sectors in the file, including data and all indirect blocks,
1351          * does not exceed 2^32. */
1352         const loff_t upper_limit = 0x1ff7fffd000LL;
1353
1354         res += 1LL << (bits-2);
1355         res += 1LL << (2*(bits-2));
1356         res += 1LL << (3*(bits-2));
1357         res <<= bits;
1358         if (res > upper_limit)
1359                 res = upper_limit;
1360         return res;
1361 }
1362
1363 static unsigned long descriptor_loc(struct super_block *sb,
1364                                     unsigned long logic_sb_block,
1365                                     int nr)
1366 {
1367         struct ext3_sb_info *sbi = EXT3_SB(sb);
1368         unsigned long bg, first_data_block, first_meta_bg;
1369         int has_super = 0;
1370
1371         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1372         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1373
1374         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1375             nr < first_meta_bg)
1376                 return (logic_sb_block + nr + 1);
1377         bg = sbi->s_desc_per_block * nr;
1378         if (ext3_bg_has_super(sb, bg))
1379                 has_super = 1;
1380         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1381 }
1382
1383
1384 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1385 {
1386         struct buffer_head * bh;
1387         struct ext3_super_block *es = NULL;
1388         struct ext3_sb_info *sbi;
1389         unsigned long block;
1390         unsigned long sb_block = get_sb_block(&data);
1391         unsigned long logic_sb_block;
1392         unsigned long offset = 0;
1393         unsigned long journal_inum = 0;
1394         unsigned long journal_devnum = 0;
1395         unsigned long def_mount_opts;
1396         struct inode *root;
1397         int blocksize;
1398         int hblock;
1399         int db_count;
1400         int i;
1401         int needs_recovery;
1402         __le32 features;
1403
1404         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1405         if (!sbi)
1406                 return -ENOMEM;
1407         sb->s_fs_info = sbi;
1408         memset(sbi, 0, sizeof(*sbi));
1409         sbi->s_mount_opt = 0;
1410         sbi->s_resuid = EXT3_DEF_RESUID;
1411         sbi->s_resgid = EXT3_DEF_RESGID;
1412
1413         unlock_kernel();
1414
1415         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1416         if (!blocksize) {
1417                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1418                 goto out_fail;
1419         }
1420
1421         /*
1422          * The ext3 superblock will not be buffer aligned for other than 1kB
1423          * block sizes.  We need to calculate the offset from buffer start.
1424          */
1425         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1426                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1427                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1428         } else {
1429                 logic_sb_block = sb_block;
1430         }
1431
1432         if (!(bh = sb_bread(sb, logic_sb_block))) {
1433                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1434                 goto out_fail;
1435         }
1436         /*
1437          * Note: s_es must be initialized as soon as possible because
1438          *       some ext3 macro-instructions depend on its value
1439          */
1440         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1441         sbi->s_es = es;
1442         sb->s_magic = le16_to_cpu(es->s_magic);
1443         if (sb->s_magic != EXT3_SUPER_MAGIC)
1444                 goto cantfind_ext3;
1445
1446         /* Set defaults before we parse the mount options */
1447         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1448         if (def_mount_opts & EXT3_DEFM_DEBUG)
1449                 set_opt(sbi->s_mount_opt, DEBUG);
1450         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1451                 set_opt(sbi->s_mount_opt, GRPID);
1452         if (def_mount_opts & EXT3_DEFM_UID16)
1453                 set_opt(sbi->s_mount_opt, NO_UID32);
1454         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1455                 set_opt(sbi->s_mount_opt, XATTR_USER);
1456         if (def_mount_opts & EXT3_DEFM_ACL)
1457                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1458         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1459                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1460         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1461                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1462         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1463                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1464
1465         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1466                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1467         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1468                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1469
1470         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1471         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1472
1473         set_opt(sbi->s_mount_opt, RESERVATION);
1474
1475         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1476                             NULL, 0))
1477                 goto failed_mount;
1478
1479         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAGXID)
1480                 sb->s_flags |= MS_TAGXID;
1481
1482         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1483                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1484
1485         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1486             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1487              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1488              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1489                 printk(KERN_WARNING
1490                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1491                        "running e2fsck is recommended\n");
1492         /*
1493          * Check feature flags regardless of the revision level, since we
1494          * previously didn't change the revision level when setting the flags,
1495          * so there is a chance incompat flags are set on a rev 0 filesystem.
1496          */
1497         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1498         if (features) {
1499                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1500                        "unsupported optional features (%x).\n",
1501                        sb->s_id, le32_to_cpu(features));
1502                 goto failed_mount;
1503         }
1504         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1505         if (!(sb->s_flags & MS_RDONLY) && features) {
1506                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1507                        "unsupported optional features (%x).\n",
1508                        sb->s_id, le32_to_cpu(features));
1509                 goto failed_mount;
1510         }
1511         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1512
1513         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1514             blocksize > EXT3_MAX_BLOCK_SIZE) {
1515                 printk(KERN_ERR
1516                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1517                        blocksize, sb->s_id);
1518                 goto failed_mount;
1519         }
1520
1521         hblock = bdev_hardsect_size(sb->s_bdev);
1522         if (sb->s_blocksize != blocksize) {
1523                 /*
1524                  * Make sure the blocksize for the filesystem is larger
1525                  * than the hardware sectorsize for the machine.
1526                  */
1527                 if (blocksize < hblock) {
1528                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1529                                "device blocksize %d.\n", blocksize, hblock);
1530                         goto failed_mount;
1531                 }
1532
1533                 brelse (bh);
1534                 sb_set_blocksize(sb, blocksize);
1535                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1536                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1537                 bh = sb_bread(sb, logic_sb_block);
1538                 if (!bh) {
1539                         printk(KERN_ERR
1540                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1541                         goto failed_mount;
1542                 }
1543                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1544                 sbi->s_es = es;
1545                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1546                         printk (KERN_ERR
1547                                 "EXT3-fs: Magic mismatch, very weird !\n");
1548                         goto failed_mount;
1549                 }
1550         }
1551
1552         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1553
1554         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1555                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1556                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1557         } else {
1558                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1559                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1560                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1561                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1562                     (sbi->s_inode_size > blocksize)) {
1563                         printk (KERN_ERR
1564                                 "EXT3-fs: unsupported inode size: %d\n",
1565                                 sbi->s_inode_size);
1566                         goto failed_mount;
1567                 }
1568         }
1569         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1570                                    le32_to_cpu(es->s_log_frag_size);
1571         if (blocksize != sbi->s_frag_size) {
1572                 printk(KERN_ERR
1573                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1574                        sbi->s_frag_size, blocksize);
1575                 goto failed_mount;
1576         }
1577         sbi->s_frags_per_block = 1;
1578         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1579         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1580         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1581         if (EXT3_INODE_SIZE(sb) == 0)
1582                 goto cantfind_ext3;
1583         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1584         if (sbi->s_inodes_per_block == 0)
1585                 goto cantfind_ext3;
1586         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1587                                         sbi->s_inodes_per_block;
1588         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1589         sbi->s_sbh = bh;
1590         sbi->s_mount_state = le16_to_cpu(es->s_state);
1591         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1592         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1593         for (i=0; i < 4; i++)
1594                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1595         sbi->s_def_hash_version = es->s_def_hash_version;
1596
1597         if (sbi->s_blocks_per_group > blocksize * 8) {
1598                 printk (KERN_ERR
1599                         "EXT3-fs: #blocks per group too big: %lu\n",
1600                         sbi->s_blocks_per_group);
1601                 goto failed_mount;
1602         }
1603         if (sbi->s_frags_per_group > blocksize * 8) {
1604                 printk (KERN_ERR
1605                         "EXT3-fs: #fragments per group too big: %lu\n",
1606                         sbi->s_frags_per_group);
1607                 goto failed_mount;
1608         }
1609         if (sbi->s_inodes_per_group > blocksize * 8) {
1610                 printk (KERN_ERR
1611                         "EXT3-fs: #inodes per group too big: %lu\n",
1612                         sbi->s_inodes_per_group);
1613                 goto failed_mount;
1614         }
1615
1616         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1617                 goto cantfind_ext3;
1618         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1619                                le32_to_cpu(es->s_first_data_block) +
1620                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1621                               EXT3_BLOCKS_PER_GROUP(sb);
1622         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1623                    EXT3_DESC_PER_BLOCK(sb);
1624         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1625                                     GFP_KERNEL);
1626         if (sbi->s_group_desc == NULL) {
1627                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1628                 goto failed_mount;
1629         }
1630
1631         percpu_counter_init(&sbi->s_freeblocks_counter);
1632         percpu_counter_init(&sbi->s_freeinodes_counter);
1633         percpu_counter_init(&sbi->s_dirs_counter);
1634         bgl_lock_init(&sbi->s_blockgroup_lock);
1635
1636         for (i = 0; i < db_count; i++) {
1637                 block = descriptor_loc(sb, logic_sb_block, i);
1638                 sbi->s_group_desc[i] = sb_bread(sb, block);
1639                 if (!sbi->s_group_desc[i]) {
1640                         printk (KERN_ERR "EXT3-fs: "
1641                                 "can't read group descriptor %d\n", i);
1642                         db_count = i;
1643                         goto failed_mount2;
1644                 }
1645         }
1646         if (!ext3_check_descriptors (sb)) {
1647                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1648                 goto failed_mount2;
1649         }
1650         sbi->s_gdb_count = db_count;
1651         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1652         spin_lock_init(&sbi->s_next_gen_lock);
1653         /* per fileystem reservation list head & lock */
1654         spin_lock_init(&sbi->s_rsv_window_lock);
1655         sbi->s_rsv_window_root = RB_ROOT;
1656         /* Add a single, static dummy reservation to the start of the
1657          * reservation window list --- it gives us a placeholder for
1658          * append-at-start-of-list which makes the allocation logic
1659          * _much_ simpler. */
1660         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1661         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1662         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1663         sbi->s_rsv_window_head.rsv_goal_size = 0;
1664         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1665
1666         /*
1667          * set up enough so that it can read an inode
1668          */
1669         sb->s_op = &ext3_sops;
1670         sb->s_export_op = &ext3_export_ops;
1671         sb->s_xattr = ext3_xattr_handlers;
1672 #ifdef CONFIG_QUOTA
1673         sb->s_qcop = &ext3_qctl_operations;
1674         sb->dq_op = &ext3_quota_operations;
1675 #endif
1676         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1677
1678         sb->s_root = NULL;
1679
1680         needs_recovery = (es->s_last_orphan != 0 ||
1681                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1682                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1683
1684         /*
1685          * The first inode we look at is the journal inode.  Don't try
1686          * root first: it may be modified in the journal!
1687          */
1688         if (!test_opt(sb, NOLOAD) &&
1689             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1690                 if (ext3_load_journal(sb, es, journal_devnum))
1691                         goto failed_mount2;
1692         } else if (journal_inum) {
1693                 if (ext3_create_journal(sb, es, journal_inum))
1694                         goto failed_mount2;
1695         } else {
1696                 if (!silent)
1697                         printk (KERN_ERR
1698                                 "ext3: No journal on filesystem on %s\n",
1699                                 sb->s_id);
1700                 goto failed_mount2;
1701         }
1702
1703         /* We have now updated the journal if required, so we can
1704          * validate the data journaling mode. */
1705         switch (test_opt(sb, DATA_FLAGS)) {
1706         case 0:
1707                 /* No mode set, assume a default based on the journal
1708                    capabilities: ORDERED_DATA if the journal can
1709                    cope, else JOURNAL_DATA */
1710                 if (journal_check_available_features
1711                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1712                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1713                 else
1714                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1715                 break;
1716
1717         case EXT3_MOUNT_ORDERED_DATA:
1718         case EXT3_MOUNT_WRITEBACK_DATA:
1719                 if (!journal_check_available_features
1720                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1721                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1722                                "requested data journaling mode\n");
1723                         goto failed_mount3;
1724                 }
1725         default:
1726                 break;
1727         }
1728
1729         if (test_opt(sb, NOBH)) {
1730                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1731                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1732                                 "its supported only with writeback mode\n");
1733                         clear_opt(sbi->s_mount_opt, NOBH);
1734                 }
1735         }
1736         /*
1737          * The journal_load will have done any necessary log recovery,
1738          * so we can safely mount the rest of the filesystem now.
1739          */
1740
1741         root = iget(sb, EXT3_ROOT_INO);
1742         sb->s_root = d_alloc_root(root);
1743         if (!sb->s_root) {
1744                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1745                 iput(root);
1746                 goto failed_mount3;
1747         }
1748         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1749                 dput(sb->s_root);
1750                 sb->s_root = NULL;
1751                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1752                 goto failed_mount3;
1753         }
1754
1755         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1756         /*
1757          * akpm: core read_super() calls in here with the superblock locked.
1758          * That deadlocks, because orphan cleanup needs to lock the superblock
1759          * in numerous places.  Here we just pop the lock - it's relatively
1760          * harmless, because we are now ready to accept write_super() requests,
1761          * and aviro says that's the only reason for hanging onto the
1762          * superblock lock.
1763          */
1764         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1765         ext3_orphan_cleanup(sb, es);
1766         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1767         if (needs_recovery)
1768                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1769         ext3_mark_recovery_complete(sb, es);
1770         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1771                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1772                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1773                 "writeback");
1774
1775         percpu_counter_mod(&sbi->s_freeblocks_counter,
1776                 ext3_count_free_blocks(sb));
1777         percpu_counter_mod(&sbi->s_freeinodes_counter,
1778                 ext3_count_free_inodes(sb));
1779         percpu_counter_mod(&sbi->s_dirs_counter,
1780                 ext3_count_dirs(sb));
1781
1782         lock_kernel();
1783         return 0;
1784
1785 cantfind_ext3:
1786         if (!silent)
1787                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1788                        sb->s_id);
1789         goto failed_mount;
1790
1791 failed_mount3:
1792         journal_destroy(sbi->s_journal);
1793 failed_mount2:
1794         for (i = 0; i < db_count; i++)
1795                 brelse(sbi->s_group_desc[i]);
1796         kfree(sbi->s_group_desc);
1797 failed_mount:
1798 #ifdef CONFIG_QUOTA
1799         for (i = 0; i < MAXQUOTAS; i++)
1800                 kfree(sbi->s_qf_names[i]);
1801 #endif
1802         ext3_blkdev_remove(sbi);
1803         brelse(bh);
1804 out_fail:
1805         sb->s_fs_info = NULL;
1806         kfree(sbi);
1807         lock_kernel();
1808         return -EINVAL;
1809 }
1810
1811 /*
1812  * Setup any per-fs journal parameters now.  We'll do this both on
1813  * initial mount, once the journal has been initialised but before we've
1814  * done any recovery; and again on any subsequent remount.
1815  */
1816 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1817 {
1818         struct ext3_sb_info *sbi = EXT3_SB(sb);
1819
1820         if (sbi->s_commit_interval)
1821                 journal->j_commit_interval = sbi->s_commit_interval;
1822         /* We could also set up an ext3-specific default for the commit
1823          * interval here, but for now we'll just fall back to the jbd
1824          * default. */
1825
1826         spin_lock(&journal->j_state_lock);
1827         if (test_opt(sb, BARRIER))
1828                 journal->j_flags |= JFS_BARRIER;
1829         else
1830                 journal->j_flags &= ~JFS_BARRIER;
1831         spin_unlock(&journal->j_state_lock);
1832 }
1833
1834 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1835 {
1836         struct inode *journal_inode;
1837         journal_t *journal;
1838
1839         /* First, test for the existence of a valid inode on disk.  Bad
1840          * things happen if we iget() an unused inode, as the subsequent
1841          * iput() will try to delete it. */
1842
1843         journal_inode = iget(sb, journal_inum);
1844         if (!journal_inode) {
1845                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1846                 return NULL;
1847         }
1848         if (!journal_inode->i_nlink) {
1849                 make_bad_inode(journal_inode);
1850                 iput(journal_inode);
1851                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1852                 return NULL;
1853         }
1854
1855         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1856                   journal_inode, journal_inode->i_size);
1857         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1858                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1859                 iput(journal_inode);
1860                 return NULL;
1861         }
1862
1863         journal = journal_init_inode(journal_inode);
1864         if (!journal) {
1865                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1866                 iput(journal_inode);
1867                 return NULL;
1868         }
1869         journal->j_private = sb;
1870         ext3_init_journal_params(sb, journal);
1871         return journal;
1872 }
1873
1874 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1875                                        dev_t j_dev)
1876 {
1877         struct buffer_head * bh;
1878         journal_t *journal;
1879         int start;
1880         int len;
1881         int hblock, blocksize;
1882         unsigned long sb_block;
1883         unsigned long offset;
1884         struct ext3_super_block * es;
1885         struct block_device *bdev;
1886
1887         bdev = ext3_blkdev_get(j_dev);
1888         if (bdev == NULL)
1889                 return NULL;
1890
1891         if (bd_claim(bdev, sb)) {
1892                 printk(KERN_ERR
1893                         "EXT3: failed to claim external journal device.\n");
1894                 blkdev_put(bdev);
1895                 return NULL;
1896         }
1897
1898         blocksize = sb->s_blocksize;
1899         hblock = bdev_hardsect_size(bdev);
1900         if (blocksize < hblock) {
1901                 printk(KERN_ERR
1902                         "EXT3-fs: blocksize too small for journal device.\n");
1903                 goto out_bdev;
1904         }
1905
1906         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1907         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1908         set_blocksize(bdev, blocksize);
1909         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1910                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1911                        "external journal\n");
1912                 goto out_bdev;
1913         }
1914
1915         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1916         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1917             !(le32_to_cpu(es->s_feature_incompat) &
1918               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1919                 printk(KERN_ERR "EXT3-fs: external journal has "
1920                                         "bad superblock\n");
1921                 brelse(bh);
1922                 goto out_bdev;
1923         }
1924
1925         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1926                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1927                 brelse(bh);
1928                 goto out_bdev;
1929         }
1930
1931         len = le32_to_cpu(es->s_blocks_count);
1932         start = sb_block + 1;
1933         brelse(bh);     /* we're done with the superblock */
1934
1935         journal = journal_init_dev(bdev, sb->s_bdev,
1936                                         start, len, blocksize);
1937         if (!journal) {
1938                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1939                 goto out_bdev;
1940         }
1941         journal->j_private = sb;
1942         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1943         wait_on_buffer(journal->j_sb_buffer);
1944         if (!buffer_uptodate(journal->j_sb_buffer)) {
1945                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1946                 goto out_journal;
1947         }
1948         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1949                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1950                                         "user (unsupported) - %d\n",
1951                         be32_to_cpu(journal->j_superblock->s_nr_users));
1952                 goto out_journal;
1953         }
1954         EXT3_SB(sb)->journal_bdev = bdev;
1955         ext3_init_journal_params(sb, journal);
1956         return journal;
1957 out_journal:
1958         journal_destroy(journal);
1959 out_bdev:
1960         ext3_blkdev_put(bdev);
1961         return NULL;
1962 }
1963
1964 static int ext3_load_journal(struct super_block *sb,
1965                              struct ext3_super_block *es,
1966                              unsigned long journal_devnum)
1967 {
1968         journal_t *journal;
1969         int journal_inum = le32_to_cpu(es->s_journal_inum);
1970         dev_t journal_dev;
1971         int err = 0;
1972         int really_read_only;
1973
1974         if (journal_devnum &&
1975             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1976                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1977                         "numbers have changed\n");
1978                 journal_dev = new_decode_dev(journal_devnum);
1979         } else
1980                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1981
1982         really_read_only = bdev_read_only(sb->s_bdev);
1983
1984         /*
1985          * Are we loading a blank journal or performing recovery after a
1986          * crash?  For recovery, we need to check in advance whether we
1987          * can get read-write access to the device.
1988          */
1989
1990         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1991                 if (sb->s_flags & MS_RDONLY) {
1992                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1993                                         "required on readonly filesystem.\n");
1994                         if (really_read_only) {
1995                                 printk(KERN_ERR "EXT3-fs: write access "
1996                                         "unavailable, cannot proceed.\n");
1997                                 return -EROFS;
1998                         }
1999                         printk (KERN_INFO "EXT3-fs: write access will "
2000                                         "be enabled during recovery.\n");
2001                 }
2002         }
2003
2004         if (journal_inum && journal_dev) {
2005                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2006                        "and inode journals!\n");
2007                 return -EINVAL;
2008         }
2009
2010         if (journal_inum) {
2011                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2012                         return -EINVAL;
2013         } else {
2014                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2015                         return -EINVAL;
2016         }
2017
2018         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2019                 err = journal_update_format(journal);
2020                 if (err)  {
2021                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2022                         journal_destroy(journal);
2023                         return err;
2024                 }
2025         }
2026
2027         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2028                 err = journal_wipe(journal, !really_read_only);
2029         if (!err)
2030                 err = journal_load(journal);
2031
2032         if (err) {
2033                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2034                 journal_destroy(journal);
2035                 return err;
2036         }
2037
2038         EXT3_SB(sb)->s_journal = journal;
2039         ext3_clear_journal_err(sb, es);
2040
2041         if (journal_devnum &&
2042             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2043                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2044                 sb->s_dirt = 1;
2045
2046                 /* Make sure we flush the recovery flag to disk. */
2047                 ext3_commit_super(sb, es, 1);
2048         }
2049
2050         return 0;
2051 }
2052
2053 static int ext3_create_journal(struct super_block * sb,
2054                                struct ext3_super_block * es,
2055                                int journal_inum)
2056 {
2057         journal_t *journal;
2058
2059         if (sb->s_flags & MS_RDONLY) {
2060                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2061                                 "create journal.\n");
2062                 return -EROFS;
2063         }
2064
2065         if (!(journal = ext3_get_journal(sb, journal_inum)))
2066                 return -EINVAL;
2067
2068         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2069                journal_inum);
2070
2071         if (journal_create(journal)) {
2072                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2073                 journal_destroy(journal);
2074                 return -EIO;
2075         }
2076
2077         EXT3_SB(sb)->s_journal = journal;
2078
2079         ext3_update_dynamic_rev(sb);
2080         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2081         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2082
2083         es->s_journal_inum = cpu_to_le32(journal_inum);
2084         sb->s_dirt = 1;
2085
2086         /* Make sure we flush the recovery flag to disk. */
2087         ext3_commit_super(sb, es, 1);
2088
2089         return 0;
2090 }
2091
2092 static void ext3_commit_super (struct super_block * sb,
2093                                struct ext3_super_block * es,
2094                                int sync)
2095 {
2096         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2097
2098         if (!sbh)
2099                 return;
2100         es->s_wtime = cpu_to_le32(get_seconds());
2101         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2102         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2103         BUFFER_TRACE(sbh, "marking dirty");
2104         mark_buffer_dirty(sbh);
2105         if (sync)
2106                 sync_dirty_buffer(sbh);
2107 }
2108
2109
2110 /*
2111  * Have we just finished recovery?  If so, and if we are mounting (or
2112  * remounting) the filesystem readonly, then we will end up with a
2113  * consistent fs on disk.  Record that fact.
2114  */
2115 static void ext3_mark_recovery_complete(struct super_block * sb,
2116                                         struct ext3_super_block * es)
2117 {
2118         journal_t *journal = EXT3_SB(sb)->s_journal;
2119
2120         journal_lock_updates(journal);
2121         journal_flush(journal);
2122         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2123             sb->s_flags & MS_RDONLY) {
2124                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2125                 sb->s_dirt = 0;
2126                 ext3_commit_super(sb, es, 1);
2127         }
2128         journal_unlock_updates(journal);
2129 }
2130
2131 /*
2132  * If we are mounting (or read-write remounting) a filesystem whose journal
2133  * has recorded an error from a previous lifetime, move that error to the
2134  * main filesystem now.
2135  */
2136 static void ext3_clear_journal_err(struct super_block * sb,
2137                                    struct ext3_super_block * es)
2138 {
2139         journal_t *journal;
2140         int j_errno;
2141         const char *errstr;
2142
2143         journal = EXT3_SB(sb)->s_journal;
2144
2145         /*
2146          * Now check for any error status which may have been recorded in the
2147          * journal by a prior ext3_error() or ext3_abort()
2148          */
2149
2150         j_errno = journal_errno(journal);
2151         if (j_errno) {
2152                 char nbuf[16];
2153
2154                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2155                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2156                              "from previous mount: %s", errstr);
2157                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2158                              "filesystem check.");
2159
2160                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2161                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2162                 ext3_commit_super (sb, es, 1);
2163
2164                 journal_clear_err(journal);
2165         }
2166 }
2167
2168 /*
2169  * Force the running and committing transactions to commit,
2170  * and wait on the commit.
2171  */
2172 int ext3_force_commit(struct super_block *sb)
2173 {
2174         journal_t *journal;
2175         int ret;
2176
2177         if (sb->s_flags & MS_RDONLY)
2178                 return 0;
2179
2180         journal = EXT3_SB(sb)->s_journal;
2181         sb->s_dirt = 0;
2182         ret = ext3_journal_force_commit(journal);
2183         return ret;
2184 }
2185
2186 /*
2187  * Ext3 always journals updates to the superblock itself, so we don't
2188  * have to propagate any other updates to the superblock on disk at this
2189  * point.  Just start an async writeback to get the buffers on their way
2190  * to the disk.
2191  *
2192  * This implicitly triggers the writebehind on sync().
2193  */
2194
2195 static void ext3_write_super (struct super_block * sb)
2196 {
2197         if (mutex_trylock(&sb->s_lock) != 0)
2198                 BUG();
2199         sb->s_dirt = 0;
2200 }
2201
2202 static int ext3_sync_fs(struct super_block *sb, int wait)
2203 {
2204         tid_t target;
2205
2206         sb->s_dirt = 0;
2207         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2208                 if (wait)
2209                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2210         }
2211         return 0;
2212 }
2213
2214 /*
2215  * LVM calls this function before a (read-only) snapshot is created.  This
2216  * gives us a chance to flush the journal completely and mark the fs clean.
2217  */
2218 static void ext3_write_super_lockfs(struct super_block *sb)
2219 {
2220         sb->s_dirt = 0;
2221
2222         if (!(sb->s_flags & MS_RDONLY)) {
2223                 journal_t *journal = EXT3_SB(sb)->s_journal;
2224
2225                 /* Now we set up the journal barrier. */
2226                 journal_lock_updates(journal);
2227                 journal_flush(journal);
2228
2229                 /* Journal blocked and flushed, clear needs_recovery flag. */
2230                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2231                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2232         }
2233 }
2234
2235 /*
2236  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2237  * flag here, even though the filesystem is not technically dirty yet.
2238  */
2239 static void ext3_unlockfs(struct super_block *sb)
2240 {
2241         if (!(sb->s_flags & MS_RDONLY)) {
2242                 lock_super(sb);
2243                 /* Reser the needs_recovery flag before the fs is unlocked. */
2244                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2245                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2246                 unlock_super(sb);
2247                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2248         }
2249 }
2250
2251 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2252 {
2253         struct ext3_super_block * es;
2254         struct ext3_sb_info *sbi = EXT3_SB(sb);
2255         unsigned long n_blocks_count = 0;
2256         unsigned long old_sb_flags;
2257         struct ext3_mount_options old_opts;
2258         int err;
2259 #ifdef CONFIG_QUOTA
2260         int i;
2261 #endif
2262
2263         /* Store the original options */
2264         old_sb_flags = sb->s_flags;
2265         old_opts.s_mount_opt = sbi->s_mount_opt;
2266         old_opts.s_resuid = sbi->s_resuid;
2267         old_opts.s_resgid = sbi->s_resgid;
2268         old_opts.s_commit_interval = sbi->s_commit_interval;
2269 #ifdef CONFIG_QUOTA
2270         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2271         for (i = 0; i < MAXQUOTAS; i++)
2272                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2273 #endif
2274
2275         /*
2276          * Allow the "check" option to be passed as a remount option.
2277          */
2278         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2279                 err = -EINVAL;
2280                 goto restore_opts;
2281         }
2282
2283         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2284                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2285         if ((sbi->s_mount_opt & EXT3_MOUNT_TAGXID) &&
2286                 !(sb->s_flags & MS_TAGXID)) {
2287                 printk("EXT3-fs: %s: tagxid not permitted on remount.\n",
2288                         sb->s_id);
2289                 return -EINVAL;
2290         }
2291
2292         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2293                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2294
2295         es = sbi->s_es;
2296
2297         ext3_init_journal_params(sb, sbi->s_journal);
2298
2299         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2300                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2301                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2302                         err = -EROFS;
2303                         goto restore_opts;
2304                 }
2305
2306                 if (*flags & MS_RDONLY) {
2307                         /*
2308                          * First of all, the unconditional stuff we have to do
2309                          * to disable replay of the journal when we next remount
2310                          */
2311                         sb->s_flags |= MS_RDONLY;
2312
2313                         /*
2314                          * OK, test if we are remounting a valid rw partition
2315                          * readonly, and if so set the rdonly flag and then
2316                          * mark the partition as valid again.
2317                          */
2318                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2319                             (sbi->s_mount_state & EXT3_VALID_FS))
2320                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2321
2322                         ext3_mark_recovery_complete(sb, es);
2323                 } else {
2324                         __le32 ret;
2325                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2326                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2327                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2328                                        "remount RDWR because of unsupported "
2329                                        "optional features (%x).\n",
2330                                        sb->s_id, le32_to_cpu(ret));
2331                                 err = -EROFS;
2332                                 goto restore_opts;
2333                         }
2334                         /*
2335                          * Mounting a RDONLY partition read-write, so reread
2336                          * and store the current valid flag.  (It may have
2337                          * been changed by e2fsck since we originally mounted
2338                          * the partition.)
2339                          */
2340                         ext3_clear_journal_err(sb, es);
2341                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2342                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2343                                 err = ret;
2344                                 goto restore_opts;
2345                         }
2346                         if (!ext3_setup_super (sb, es, 0))
2347                                 sb->s_flags &= ~MS_RDONLY;
2348                 }
2349         }
2350 #ifdef CONFIG_QUOTA
2351         /* Release old quota file names */
2352         for (i = 0; i < MAXQUOTAS; i++)
2353                 if (old_opts.s_qf_names[i] &&
2354                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2355                         kfree(old_opts.s_qf_names[i]);
2356 #endif
2357         return 0;
2358 restore_opts:
2359         sb->s_flags = old_sb_flags;
2360         sbi->s_mount_opt = old_opts.s_mount_opt;
2361         sbi->s_resuid = old_opts.s_resuid;
2362         sbi->s_resgid = old_opts.s_resgid;
2363         sbi->s_commit_interval = old_opts.s_commit_interval;
2364 #ifdef CONFIG_QUOTA
2365         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2366         for (i = 0; i < MAXQUOTAS; i++) {
2367                 if (sbi->s_qf_names[i] &&
2368                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2369                         kfree(sbi->s_qf_names[i]);
2370                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2371         }
2372 #endif
2373         return err;
2374 }
2375
2376 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2377 {
2378         struct ext3_sb_info *sbi = EXT3_SB(sb);
2379         struct ext3_super_block *es = sbi->s_es;
2380         unsigned long overhead;
2381         int i;
2382
2383         if (test_opt (sb, MINIX_DF))
2384                 overhead = 0;
2385         else {
2386                 unsigned long ngroups;
2387                 ngroups = EXT3_SB(sb)->s_groups_count;
2388                 smp_rmb();
2389
2390                 /*
2391                  * Compute the overhead (FS structures)
2392                  */
2393
2394                 /*
2395                  * All of the blocks before first_data_block are
2396                  * overhead
2397                  */
2398                 overhead = le32_to_cpu(es->s_first_data_block);
2399
2400                 /*
2401                  * Add the overhead attributed to the superblock and
2402                  * block group descriptors.  If the sparse superblocks
2403                  * feature is turned on, then not all groups have this.
2404                  */
2405                 for (i = 0; i < ngroups; i++) {
2406                         overhead += ext3_bg_has_super(sb, i) +
2407                                 ext3_bg_num_gdb(sb, i);
2408                         cond_resched();
2409                 }
2410
2411                 /*
2412                  * Every block group has an inode bitmap, a block
2413                  * bitmap, and an inode table.
2414                  */
2415                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2416         }
2417
2418         buf->f_type = EXT3_SUPER_MAGIC;
2419         buf->f_bsize = sb->s_blocksize;
2420         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2421         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2422         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2423         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2424                 buf->f_bavail = 0;
2425         buf->f_files = le32_to_cpu(es->s_inodes_count);
2426         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2427         buf->f_namelen = EXT3_NAME_LEN;
2428         return 0;
2429 }
2430
2431 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2432  * is locked for write. Otherwise the are possible deadlocks:
2433  * Process 1                         Process 2
2434  * ext3_create()                     quota_sync()
2435  *   journal_start()                   write_dquot()
2436  *   DQUOT_INIT()                        down(dqio_mutex)
2437  *     down(dqio_mutex)                    journal_start()
2438  *
2439  */
2440
2441 #ifdef CONFIG_QUOTA
2442
2443 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2444 {
2445         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2446 }
2447
2448 static int ext3_dquot_initialize(struct inode *inode, int type)
2449 {
2450         handle_t *handle;
2451         int ret, err;
2452
2453         /* We may create quota structure so we need to reserve enough blocks */
2454         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2455         if (IS_ERR(handle))
2456                 return PTR_ERR(handle);
2457         ret = dquot_initialize(inode, type);
2458         err = ext3_journal_stop(handle);
2459         if (!ret)
2460                 ret = err;
2461         return ret;
2462 }
2463
2464 static int ext3_dquot_drop(struct inode *inode)
2465 {
2466         handle_t *handle;
2467         int ret, err;
2468
2469         /* We may delete quota structure so we need to reserve enough blocks */
2470         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2471         if (IS_ERR(handle))
2472                 return PTR_ERR(handle);
2473         ret = dquot_drop(inode);
2474         err = ext3_journal_stop(handle);
2475         if (!ret)
2476                 ret = err;
2477         return ret;
2478 }
2479
2480 static int ext3_write_dquot(struct dquot *dquot)
2481 {
2482         int ret, err;
2483         handle_t *handle;
2484         struct inode *inode;
2485
2486         inode = dquot_to_inode(dquot);
2487         handle = ext3_journal_start(inode,
2488                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2489         if (IS_ERR(handle))
2490                 return PTR_ERR(handle);
2491         ret = dquot_commit(dquot);
2492         err = ext3_journal_stop(handle);
2493         if (!ret)
2494                 ret = err;
2495         return ret;
2496 }
2497
2498 static int ext3_acquire_dquot(struct dquot *dquot)
2499 {
2500         int ret, err;
2501         handle_t *handle;
2502
2503         handle = ext3_journal_start(dquot_to_inode(dquot),
2504                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2505         if (IS_ERR(handle))
2506                 return PTR_ERR(handle);
2507         ret = dquot_acquire(dquot);
2508         err = ext3_journal_stop(handle);
2509         if (!ret)
2510                 ret = err;
2511         return ret;
2512 }
2513
2514 static int ext3_release_dquot(struct dquot *dquot)
2515 {
2516         int ret, err;
2517         handle_t *handle;
2518
2519         handle = ext3_journal_start(dquot_to_inode(dquot),
2520                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2521         if (IS_ERR(handle))
2522                 return PTR_ERR(handle);
2523         ret = dquot_release(dquot);
2524         err = ext3_journal_stop(handle);
2525         if (!ret)
2526                 ret = err;
2527         return ret;
2528 }
2529
2530 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2531 {
2532         /* Are we journalling quotas? */
2533         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2534             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2535                 dquot_mark_dquot_dirty(dquot);
2536                 return ext3_write_dquot(dquot);
2537         } else {
2538                 return dquot_mark_dquot_dirty(dquot);
2539         }
2540 }
2541
2542 static int ext3_write_info(struct super_block *sb, int type)
2543 {
2544         int ret, err;
2545         handle_t *handle;
2546
2547         /* Data block + inode block */
2548         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2549         if (IS_ERR(handle))
2550                 return PTR_ERR(handle);
2551         ret = dquot_commit_info(sb, type);
2552         err = ext3_journal_stop(handle);
2553         if (!ret)
2554                 ret = err;
2555         return ret;
2556 }
2557
2558 /*
2559  * Turn on quotas during mount time - we need to find
2560  * the quota file and such...
2561  */
2562 static int ext3_quota_on_mount(struct super_block *sb, int type)
2563 {
2564         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2565                         EXT3_SB(sb)->s_jquota_fmt, type);
2566 }
2567
2568 /*
2569  * Standard function to be called on quota_on
2570  */
2571 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2572                          char *path)
2573 {
2574         int err;
2575         struct nameidata nd;
2576
2577         if (!test_opt(sb, QUOTA))
2578                 return -EINVAL;
2579         /* Not journalling quota? */
2580         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2581             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2582                 return vfs_quota_on(sb, type, format_id, path);
2583         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2584         if (err)
2585                 return err;
2586         /* Quotafile not on the same filesystem? */
2587         if (nd.mnt->mnt_sb != sb) {
2588                 path_release(&nd);
2589                 return -EXDEV;
2590         }
2591         /* Quotafile not of fs root? */
2592         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2593                 printk(KERN_WARNING
2594                         "EXT3-fs: Quota file not on filesystem root. "
2595                         "Journalled quota will not work.\n");
2596         path_release(&nd);
2597         return vfs_quota_on(sb, type, format_id, path);
2598 }
2599
2600 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2601  * acquiring the locks... As quota files are never truncated and quota code
2602  * itself serializes the operations (and noone else should touch the files)
2603  * we don't have to be afraid of races */
2604 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2605                                size_t len, loff_t off)
2606 {
2607         struct inode *inode = sb_dqopt(sb)->files[type];
2608         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2609         int err = 0;
2610         int offset = off & (sb->s_blocksize - 1);
2611         int tocopy;
2612         size_t toread;
2613         struct buffer_head *bh;
2614         loff_t i_size = i_size_read(inode);
2615
2616         if (off > i_size)
2617                 return 0;
2618         if (off+len > i_size)
2619                 len = i_size-off;
2620         toread = len;
2621         while (toread > 0) {
2622                 tocopy = sb->s_blocksize - offset < toread ?
2623                                 sb->s_blocksize - offset : toread;
2624                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2625                 if (err)
2626                         return err;
2627                 if (!bh)        /* A hole? */
2628                         memset(data, 0, tocopy);
2629                 else
2630                         memcpy(data, bh->b_data+offset, tocopy);
2631                 brelse(bh);
2632                 offset = 0;
2633                 toread -= tocopy;
2634                 data += tocopy;
2635                 blk++;
2636         }
2637         return len;
2638 }
2639
2640 /* Write to quotafile (we know the transaction is already started and has
2641  * enough credits) */
2642 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2643                                 const char *data, size_t len, loff_t off)
2644 {
2645         struct inode *inode = sb_dqopt(sb)->files[type];
2646         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2647         int err = 0;
2648         int offset = off & (sb->s_blocksize - 1);
2649         int tocopy;
2650         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2651         size_t towrite = len;
2652         struct buffer_head *bh;
2653         handle_t *handle = journal_current_handle();
2654
2655         mutex_lock(&inode->i_mutex);
2656         while (towrite > 0) {
2657                 tocopy = sb->s_blocksize - offset < towrite ?
2658                                 sb->s_blocksize - offset : towrite;
2659                 bh = ext3_bread(handle, inode, blk, 1, &err);
2660                 if (!bh)
2661                         goto out;
2662                 if (journal_quota) {
2663                         err = ext3_journal_get_write_access(handle, bh);
2664                         if (err) {
2665                                 brelse(bh);
2666                                 goto out;
2667                         }
2668                 }
2669                 lock_buffer(bh);
2670                 memcpy(bh->b_data+offset, data, tocopy);
2671                 flush_dcache_page(bh->b_page);
2672                 unlock_buffer(bh);
2673                 if (journal_quota)
2674                         err = ext3_journal_dirty_metadata(handle, bh);
2675                 else {
2676                         /* Always do at least ordered writes for quotas */
2677                         err = ext3_journal_dirty_data(handle, bh);
2678                         mark_buffer_dirty(bh);
2679                 }
2680                 brelse(bh);
2681                 if (err)
2682                         goto out;
2683                 offset = 0;
2684                 towrite -= tocopy;
2685                 data += tocopy;
2686                 blk++;
2687         }
2688 out:
2689         if (len == towrite)
2690                 return err;
2691         if (inode->i_size < off+len-towrite) {
2692                 i_size_write(inode, off+len-towrite);
2693                 EXT3_I(inode)->i_disksize = inode->i_size;
2694         }
2695         inode->i_version++;
2696         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2697         ext3_mark_inode_dirty(handle, inode);
2698         mutex_unlock(&inode->i_mutex);
2699         return len - towrite;
2700 }
2701
2702 #endif
2703
2704 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2705         int flags, const char *dev_name, void *data)
2706 {
2707         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2708 }
2709
2710 static struct file_system_type ext3_fs_type = {
2711         .owner          = THIS_MODULE,
2712         .name           = "ext3",
2713         .get_sb         = ext3_get_sb,
2714         .kill_sb        = kill_block_super,
2715         .fs_flags       = FS_REQUIRES_DEV,
2716 };
2717
2718 static int __init init_ext3_fs(void)
2719 {
2720         int err = init_ext3_xattr();
2721         if (err)
2722                 return err;
2723         err = init_inodecache();
2724         if (err)
2725                 goto out1;
2726         err = register_filesystem(&ext3_fs_type);
2727         if (err)
2728                 goto out;
2729         return 0;
2730 out:
2731         destroy_inodecache();
2732 out1:
2733         exit_ext3_xattr();
2734         return err;
2735 }
2736
2737 static void __exit exit_ext3_fs(void)
2738 {
2739         unregister_filesystem(&ext3_fs_type);
2740         destroy_inodecache();
2741         exit_ext3_xattr();
2742 }
2743
2744 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2745 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2746 MODULE_LICENSE("GPL");
2747 module_init(init_ext3_fs)
2748 module_exit(exit_ext3_fs)