fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  50                              unsigned long journal_devnum);
  51 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  52                                unsigned int);
  53 static void ext4_commit_super (struct super_block * sb,
  54                                struct ext4_super_block * es,
  55                                int sync);
  56 static void ext4_mark_recovery_complete(struct super_block * sb,
  57                                         struct ext4_super_block * es);
  58 static void ext4_clear_journal_err(struct super_block * sb,
  59                                    struct ext4_super_block * es);
  60 static int ext4_sync_fs(struct super_block *sb, int wait);
  61 static const char *ext4_decode_error(struct super_block * sb, int errno,
  62                                      char nbuf[16]);
  63 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  64 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  65 static void ext4_unlockfs(struct super_block *sb);
  66 static void ext4_write_super (struct super_block * sb);
  67 static void ext4_write_super_lockfs(struct super_block *sb);
  68
  69
  70 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  71                                struct ext4_group_desc *bg)
  72 {
  73         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  74                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  75                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  76 }
  77
  78 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  79                                struct ext4_group_desc *bg)
  80 {
  81         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  82                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  83                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  84 }
  85
  86 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  87                               struct ext4_group_desc *bg)
  88 {
  89         return le32_to_cpu(bg->bg_inode_table_lo) |
  90                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  91                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  92 }
  93
  94 void ext4_block_bitmap_set(struct super_block *sb,
  95                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  96 {
  97         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  98         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  99                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 100 }
 101
 102 void ext4_inode_bitmap_set(struct super_block *sb,
 103                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 104 {
 105         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 106         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 107                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 108 }
 109
 110 void ext4_inode_table_set(struct super_block *sb,
 111                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 112 {
 113         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 114         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 115                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 116 }
 117
 118 /*
 119  * Wrappers for jbd2_journal_start/end.
 120  *
 121  * The only special thing we need to do here is to make sure that all
 122  * journal_end calls result in the superblock being marked dirty, so
 123  * that sync() will call the filesystem's write_super callback if
 124  * appropriate.
 125  */
 126 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 127 {
 128         journal_t *journal;
 129
 130         if (sb->s_flags & MS_RDONLY)
 131                 return ERR_PTR(-EROFS);
 132
 133         /* Special case here: if the journal has aborted behind our
 134          * backs (eg. EIO in the commit thread), then we still need to
 135          * take the FS itself readonly cleanly. */
 136         journal = EXT4_SB(sb)->s_journal;
 137         if (is_journal_aborted(journal)) {
 138                 ext4_abort(sb, __FUNCTION__,
 139                            "Detected aborted journal");
 140                 return ERR_PTR(-EROFS);
 141         }
 142
 143         return jbd2_journal_start(journal, nblocks);
 144 }
 145
 146 /*
 147  * The only special thing we need to do here is to make sure that all
 148  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 149  * that sync() will call the filesystem's write_super callback if
 150  * appropriate.
 151  */
 152 int __ext4_journal_stop(const char *where, handle_t *handle)
 153 {
 154         struct super_block *sb;
 155         int err;
 156         int rc;
 157
 158         sb = handle->h_transaction->t_journal->j_private;
 159         err = handle->h_err;
 160         rc = jbd2_journal_stop(handle);
 161
 162         if (!err)
 163                 err = rc;
 164         if (err)
 165                 __ext4_std_error(sb, where, err);
 166         return err;
 167 }
 168
 169 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 170                 struct buffer_head *bh, handle_t *handle, int err)
 171 {
 172         char nbuf[16];
 173         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 174
 175         if (bh)
 176                 BUFFER_TRACE(bh, "abort");
 177
 178         if (!handle->h_err)
 179                 handle->h_err = err;
 180
 181         if (is_handle_aborted(handle))
 182                 return;
 183
 184         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 185                caller, errstr, err_fn);
 186
 187         jbd2_journal_abort_handle(handle);
 188 }
 189
 190 /* Deal with the reporting of failure conditions on a filesystem such as
 191  * inconsistencies detected or read IO failures.
 192  *
 193  * On ext2, we can store the error state of the filesystem in the
 194  * superblock.  That is not possible on ext4, because we may have other
 195  * write ordering constraints on the superblock which prevent us from
 196  * writing it out straight away; and given that the journal is about to
 197  * be aborted, we can't rely on the current, or future, transactions to
 198  * write out the superblock safely.
 199  *
 200  * We'll just use the jbd2_journal_abort() error code to record an error in
 201  * the journal instead.  On recovery, the journal will compain about
 202  * that error until we've noted it down and cleared it.
 203  */
 204
 205 static void ext4_handle_error(struct super_block *sb)
 206 {
 207         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 208
 209         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 210         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 211
 212         if (sb->s_flags & MS_RDONLY)
 213                 return;
 214
 215         if (!test_opt (sb, ERRORS_CONT)) {
 216                 journal_t *journal = EXT4_SB(sb)->s_journal;
 217
 218                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 219                 if (journal)
 220                         jbd2_journal_abort(journal, -EIO);
 221         }
 222         if (test_opt (sb, ERRORS_RO)) {
 223                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 224                 sb->s_flags |= MS_RDONLY;
 225         }
 226         ext4_commit_super(sb, es, 1);
 227         if (test_opt(sb, ERRORS_PANIC))
 228                 panic("EXT4-fs (device %s): panic forced after error\n",
 229                         sb->s_id);
 230 }
 231
 232 void ext4_error (struct super_block * sb, const char * function,
 233                  const char * fmt, ...)
 234 {
 235         va_list args;
 236
 237         va_start(args, fmt);
 238         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 239         vprintk(fmt, args);
 240         printk("\n");
 241         va_end(args);
 242
 243         ext4_handle_error(sb);
 244 }
 245
 246 static const char *ext4_decode_error(struct super_block * sb, int errno,
 247                                      char nbuf[16])
 248 {
 249         char *errstr = NULL;
 250
 251         switch (errno) {
 252         case -EIO:
 253                 errstr = "IO failure";
 254                 break;
 255         case -ENOMEM:
 256                 errstr = "Out of memory";
 257                 break;
 258         case -EROFS:
 259                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 260                         errstr = "Journal has aborted";
 261                 else
 262                         errstr = "Readonly filesystem";
 263                 break;
 264         default:
 265                 /* If the caller passed in an extra buffer for unknown
 266                  * errors, textualise them now.  Else we just return
 267                  * NULL. */
 268                 if (nbuf) {
 269                         /* Check for truncated error codes... */
 270                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 271                                 errstr = nbuf;
 272                 }
 273                 break;
 274         }
 275
 276         return errstr;
 277 }
 278
 279 /* __ext4_std_error decodes expected errors from journaling functions
 280  * automatically and invokes the appropriate error response.  */
 281
 282 void __ext4_std_error (struct super_block * sb, const char * function,
 283                        int errno)
 284 {
 285         char nbuf[16];
 286         const char *errstr;
 287
 288         /* Special case: if the error is EROFS, and we're not already
 289          * inside a transaction, then there's really no point in logging
 290          * an error. */
 291         if (errno == -EROFS && journal_current_handle() == NULL &&
 292             (sb->s_flags & MS_RDONLY))
 293                 return;
 294
 295         errstr = ext4_decode_error(sb, errno, nbuf);
 296         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 297                 sb->s_id, function, errstr);
 298
 299         ext4_handle_error(sb);
 300 }
 301
 302 /*
 303  * ext4_abort is a much stronger failure handler than ext4_error.  The
 304  * abort function may be used to deal with unrecoverable failures such
 305  * as journal IO errors or ENOMEM at a critical moment in log management.
 306  *
 307  * We unconditionally force the filesystem into an ABORT|READONLY state,
 308  * unless the error response on the fs has been set to panic in which
 309  * case we take the easy way out and panic immediately.
 310  */
 311
 312 void ext4_abort (struct super_block * sb, const char * function,
 313                  const char * fmt, ...)
 314 {
 315         va_list args;
 316
 317         printk (KERN_CRIT "ext4_abort called.\n");
 318
 319         va_start(args, fmt);
 320         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 321         vprintk(fmt, args);
 322         printk("\n");
 323         va_end(args);
 324
 325         if (test_opt(sb, ERRORS_PANIC))
 326                 panic("EXT4-fs panic from previous error\n");
 327
 328         if (sb->s_flags & MS_RDONLY)
 329                 return;
 330
 331         printk(KERN_CRIT "Remounting filesystem read-only\n");
 332         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 333         sb->s_flags |= MS_RDONLY;
 334         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 335         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 336 }
 337
 338 void ext4_warning (struct super_block * sb, const char * function,
 339                    const char * fmt, ...)
 340 {
 341         va_list args;
 342
 343         va_start(args, fmt);
 344         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 345                sb->s_id, function);
 346         vprintk(fmt, args);
 347         printk("\n");
 348         va_end(args);
 349 }
 350
 351 void ext4_update_dynamic_rev(struct super_block *sb)
 352 {
 353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 354
 355         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 356                 return;
 357
 358         ext4_warning(sb, __FUNCTION__,
 359                      "updating to rev %d because of new feature flag, "
 360                      "running e2fsck is recommended",
 361                      EXT4_DYNAMIC_REV);
 362
 363         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 364         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 365         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 366         /* leave es->s_feature_*compat flags alone */
 367         /* es->s_uuid will be set by e2fsck if empty */
 368
 369         /*
 370          * The rest of the superblock fields should be zero, and if not it
 371          * means they are likely already in use, so leave them alone.  We
 372          * can leave it up to e2fsck to clean up any inconsistencies there.
 373          */
 374 }
 375
 376 /*
 377  * Open the external journal device
 378  */
 379 static struct block_device *ext4_blkdev_get(dev_t dev)
 380 {
 381         struct block_device *bdev;
 382         char b[BDEVNAME_SIZE];
 383
 384         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 385         if (IS_ERR(bdev))
 386                 goto fail;
 387         return bdev;
 388
 389 fail:
 390         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 391                         __bdevname(dev, b), PTR_ERR(bdev));
 392         return NULL;
 393 }
 394
 395 /*
 396  * Release the journal device
 397  */
 398 static int ext4_blkdev_put(struct block_device *bdev)
 399 {
 400         bd_release(bdev);
 401         return blkdev_put(bdev);
 402 }
 403
 404 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 405 {
 406         struct block_device *bdev;
 407         int ret = -ENODEV;
 408
 409         bdev = sbi->journal_bdev;
 410         if (bdev) {
 411                 ret = ext4_blkdev_put(bdev);
 412                 sbi->journal_bdev = NULL;
 413         }
 414         return ret;
 415 }
 416
 417 static inline struct inode *orphan_list_entry(struct list_head *l)
 418 {
 419         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 420 }
 421
 422 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 423 {
 424         struct list_head *l;
 425
 426         printk(KERN_ERR "sb orphan head is %d\n",
 427                le32_to_cpu(sbi->s_es->s_last_orphan));
 428
 429         printk(KERN_ERR "sb_info orphan list:\n");
 430         list_for_each(l, &sbi->s_orphan) {
 431                 struct inode *inode = orphan_list_entry(l);
 432                 printk(KERN_ERR "  "
 433                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 434                        inode->i_sb->s_id, inode->i_ino, inode,
 435                        inode->i_mode, inode->i_nlink,
 436                        NEXT_ORPHAN(inode));
 437         }
 438 }
 439
 440 static void ext4_put_super (struct super_block * sb)
 441 {
 442         struct ext4_sb_info *sbi = EXT4_SB(sb);
 443         struct ext4_super_block *es = sbi->s_es;
 444         int i;
 445
 446         ext4_ext_release(sb);
 447         ext4_xattr_put_super(sb);
 448         jbd2_journal_destroy(sbi->s_journal);
 449         if (!(sb->s_flags & MS_RDONLY)) {
 450                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 451                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 452                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 453                 mark_buffer_dirty(sbi->s_sbh);
 454                 ext4_commit_super(sb, es, 1);
 455         }
 456
 457         for (i = 0; i < sbi->s_gdb_count; i++)
 458                 brelse(sbi->s_group_desc[i]);
 459         kfree(sbi->s_group_desc);
 460         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 461         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 462         percpu_counter_destroy(&sbi->s_dirs_counter);
 463         brelse(sbi->s_sbh);
 464 #ifdef CONFIG_QUOTA
 465         for (i = 0; i < MAXQUOTAS; i++)
 466                 kfree(sbi->s_qf_names[i]);
 467 #endif
 468
 469         /* Debugging code just in case the in-memory inode orphan list
 470          * isn't empty.  The on-disk one can be non-empty if we've
 471          * detected an error and taken the fs readonly, but the
 472          * in-memory list had better be clean by this point. */
 473         if (!list_empty(&sbi->s_orphan))
 474                 dump_orphan_list(sb, sbi);
 475         J_ASSERT(list_empty(&sbi->s_orphan));
 476
 477         invalidate_bdev(sb->s_bdev);
 478         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 479                 /*
 480                  * Invalidate the journal device's buffers.  We don't want them
 481                  * floating about in memory - the physical journal device may
 482                  * hotswapped, and it breaks the `ro-after' testing code.
 483                  */
 484                 sync_blockdev(sbi->journal_bdev);
 485                 invalidate_bdev(sbi->journal_bdev);
 486                 ext4_blkdev_remove(sbi);
 487         }
 488         sb->s_fs_info = NULL;
 489         kfree(sbi);
 490         return;
 491 }
 492
 493 static struct kmem_cache *ext4_inode_cachep;
 494
 495 /*
 496  * Called inside transaction, so use GFP_NOFS
 497  */
 498 static struct inode *ext4_alloc_inode(struct super_block *sb)
 499 {
 500         struct ext4_inode_info *ei;
 501
 502         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 503         if (!ei)
 504                 return NULL;
 505 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 506         ei->i_acl = EXT4_ACL_NOT_CACHED;
 507         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 508 #endif
 509         ei->i_block_alloc_info = NULL;
 510         ei->vfs_inode.i_version = 1;
 511         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 512         return &ei->vfs_inode;
 513 }
 514
 515 static void ext4_destroy_inode(struct inode *inode)
 516 {
 517         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 518                 printk("EXT4 Inode %p: orphan list check failed!\n",
 519                         EXT4_I(inode));
 520                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 521                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 522                                 true);
 523                 dump_stack();
 524         }
 525         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 526 }
 527
 528 static void init_once(struct kmem_cache *cachep, void *foo)
 529 {
 530         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 531
 532         INIT_LIST_HEAD(&ei->i_orphan);
 533 #ifdef CONFIG_EXT4DEV_FS_XATTR
 534         init_rwsem(&ei->xattr_sem);
 535 #endif
 536         mutex_init(&ei->truncate_mutex);
 537         inode_init_once(&ei->vfs_inode);
 538 }
 539
 540 static int init_inodecache(void)
 541 {
 542         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 543                                              sizeof(struct ext4_inode_info),
 544                                              0, (SLAB_RECLAIM_ACCOUNT|
 545                                                 SLAB_MEM_SPREAD),
 546                                              init_once);
 547         if (ext4_inode_cachep == NULL)
 548                 return -ENOMEM;
 549         return 0;
 550 }
 551
 552 static void destroy_inodecache(void)
 553 {
 554         kmem_cache_destroy(ext4_inode_cachep);
 555 }
 556
 557 static void ext4_clear_inode(struct inode *inode)
 558 {
 559         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 560 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 561         if (EXT4_I(inode)->i_acl &&
 562                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 563                 posix_acl_release(EXT4_I(inode)->i_acl);
 564                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 565         }
 566         if (EXT4_I(inode)->i_default_acl &&
 567                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 568                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 569                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 570         }
 571 #endif
 572         ext4_discard_reservation(inode);
 573         EXT4_I(inode)->i_block_alloc_info = NULL;
 574         if (unlikely(rsv))
 575                 kfree(rsv);
 576 }
 577
 578 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 579 {
 580 #if defined(CONFIG_QUOTA)
 581         struct ext4_sb_info *sbi = EXT4_SB(sb);
 582
 583         if (sbi->s_jquota_fmt)
 584                 seq_printf(seq, ",jqfmt=%s",
 585                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 586
 587         if (sbi->s_qf_names[USRQUOTA])
 588                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 589
 590         if (sbi->s_qf_names[GRPQUOTA])
 591                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 592
 593         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 594                 seq_puts(seq, ",usrquota");
 595
 596         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 597                 seq_puts(seq, ",grpquota");
 598 #endif
 599 }
 600
 601 /*
 602  * Show an option if
 603  *  - it's set to a non-default value OR
 604  *  - if the per-sb default is different from the global default
 605  */
 606 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 607 {
 608         struct super_block *sb = vfs->mnt_sb;
 609         struct ext4_sb_info *sbi = EXT4_SB(sb);
 610         struct ext4_super_block *es = sbi->s_es;
 611         unsigned long def_mount_opts;
 612
 613         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 614
 615         if (sbi->s_sb_block != 1)
 616                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 617         if (test_opt(sb, MINIX_DF))
 618                 seq_puts(seq, ",minixdf");
 619         if (test_opt(sb, GRPID))
 620                 seq_puts(seq, ",grpid");
 621         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 622                 seq_puts(seq, ",nogrpid");
 623         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 624             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 625                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 626         }
 627         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 628             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 629                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 630         }
 631         if (test_opt(sb, ERRORS_CONT)) {
 632                 int def_errors = le16_to_cpu(es->s_errors);
 633
 634                 if (def_errors == EXT4_ERRORS_PANIC ||
 635                     def_errors == EXT4_ERRORS_RO) {
 636                         seq_puts(seq, ",errors=continue");
 637                 }
 638         }
 639         if (test_opt(sb, ERRORS_RO))
 640                 seq_puts(seq, ",errors=remount-ro");
 641         if (test_opt(sb, ERRORS_PANIC))
 642                 seq_puts(seq, ",errors=panic");
 643         if (test_opt(sb, NO_UID32))
 644                 seq_puts(seq, ",nouid32");
 645         if (test_opt(sb, DEBUG))
 646                 seq_puts(seq, ",debug");
 647         if (test_opt(sb, OLDALLOC))
 648                 seq_puts(seq, ",oldalloc");
 649 #ifdef CONFIG_EXT4_FS_XATTR
 650         if (test_opt(sb, XATTR_USER))
 651                 seq_puts(seq, ",user_xattr");
 652         if (!test_opt(sb, XATTR_USER) &&
 653             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 654                 seq_puts(seq, ",nouser_xattr");
 655         }
 656 #endif
 657 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 658         if (test_opt(sb, POSIX_ACL))
 659                 seq_puts(seq, ",acl");
 660         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 661                 seq_puts(seq, ",noacl");
 662 #endif
 663         if (!test_opt(sb, RESERVATION))
 664                 seq_puts(seq, ",noreservation");
 665         if (sbi->s_commit_interval) {
 666                 seq_printf(seq, ",commit=%u",
 667                            (unsigned) (sbi->s_commit_interval / HZ));
 668         }
 669         if (test_opt(sb, BARRIER))
 670                 seq_puts(seq, ",barrier=1");
 671         if (test_opt(sb, NOBH))
 672                 seq_puts(seq, ",nobh");
 673         if (!test_opt(sb, EXTENTS))
 674                 seq_puts(seq, ",noextents");
 675
 676         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 677                 seq_puts(seq, ",data=journal");
 678         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 679                 seq_puts(seq, ",data=ordered");
 680         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 681                 seq_puts(seq, ",data=writeback");
 682
 683         ext4_show_quota_options(seq, sb);
 684
 685         return 0;
 686 }
 687
 688
 689 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 690                 u64 ino, u32 generation)
 691 {
 692         struct inode *inode;
 693
 694         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 695                 return ERR_PTR(-ESTALE);
 696         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 697                 return ERR_PTR(-ESTALE);
 698
 699         /* iget isn't really right if the inode is currently unallocated!!
 700          *
 701          * ext4_read_inode will return a bad_inode if the inode had been
 702          * deleted, so we should be safe.
 703          *
 704          * Currently we don't know the generation for parent directory, so
 705          * a generation of 0 means "accept any"
 706          */
 707         inode = iget(sb, ino);
 708         if (inode == NULL)
 709                 return ERR_PTR(-ENOMEM);
 710         if (is_bad_inode(inode) ||
 711             (generation && inode->i_generation != generation)) {
 712                 iput(inode);
 713                 return ERR_PTR(-ESTALE);
 714         }
 715
 716         return inode;
 717 }
 718
 719 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 720                 int fh_len, int fh_type)
 721 {
 722         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 723                                     ext4_nfs_get_inode);
 724 }
 725
 726 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 727                 int fh_len, int fh_type)
 728 {
 729         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 730                                     ext4_nfs_get_inode);
 731 }
 732
 733 #ifdef CONFIG_QUOTA
 734 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 735 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 736
 737 static int ext4_dquot_initialize(struct inode *inode, int type);
 738 static int ext4_dquot_drop(struct inode *inode);
 739 static int ext4_write_dquot(struct dquot *dquot);
 740 static int ext4_acquire_dquot(struct dquot *dquot);
 741 static int ext4_release_dquot(struct dquot *dquot);
 742 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 743 static int ext4_write_info(struct super_block *sb, int type);
 744 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 745 static int ext4_quota_on_mount(struct super_block *sb, int type);
 746 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 747                                size_t len, loff_t off);
 748 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 749                                 const char *data, size_t len, loff_t off);
 750
 751 static struct dquot_operations ext4_quota_operations = {
 752         .initialize     = ext4_dquot_initialize,
 753         .drop           = ext4_dquot_drop,
 754         .alloc_space    = dquot_alloc_space,
 755         .alloc_inode    = dquot_alloc_inode,
 756         .free_space     = dquot_free_space,
 757         .free_inode     = dquot_free_inode,
 758         .transfer       = dquot_transfer,
 759         .write_dquot    = ext4_write_dquot,
 760         .acquire_dquot  = ext4_acquire_dquot,
 761         .release_dquot  = ext4_release_dquot,
 762         .mark_dirty     = ext4_mark_dquot_dirty,
 763         .write_info     = ext4_write_info
 764 };
 765
 766 static struct quotactl_ops ext4_qctl_operations = {
 767         .quota_on       = ext4_quota_on,
 768         .quota_off      = vfs_quota_off,
 769         .quota_sync     = vfs_quota_sync,
 770         .get_info       = vfs_get_dqinfo,
 771         .set_info       = vfs_set_dqinfo,
 772         .get_dqblk      = vfs_get_dqblk,
 773         .set_dqblk      = vfs_set_dqblk
 774 };
 775 #endif
 776
 777 static const struct super_operations ext4_sops = {
 778         .alloc_inode    = ext4_alloc_inode,
 779         .destroy_inode  = ext4_destroy_inode,
 780         .read_inode     = ext4_read_inode,
 781         .write_inode    = ext4_write_inode,
 782         .dirty_inode    = ext4_dirty_inode,
 783         .delete_inode   = ext4_delete_inode,
 784         .put_super      = ext4_put_super,
 785         .write_super    = ext4_write_super,
 786         .sync_fs        = ext4_sync_fs,
 787         .write_super_lockfs = ext4_write_super_lockfs,
 788         .unlockfs       = ext4_unlockfs,
 789         .statfs         = ext4_statfs,
 790         .remount_fs     = ext4_remount,
 791         .clear_inode    = ext4_clear_inode,
 792         .show_options   = ext4_show_options,
 793 #ifdef CONFIG_QUOTA
 794         .quota_read     = ext4_quota_read,
 795         .quota_write    = ext4_quota_write,
 796 #endif
 797 };
 798
 799 static const struct export_operations ext4_export_ops = {
 800         .fh_to_dentry = ext4_fh_to_dentry,
 801         .fh_to_parent = ext4_fh_to_parent,
 802         .get_parent = ext4_get_parent,
 803 };
 804
 805 enum {
 806         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 807         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 808         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 809         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 810         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 811         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 812         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 813         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 814         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 815         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 816         Opt_grpquota, Opt_extents, Opt_noextents,
 817 };
 818
 819 static match_table_t tokens = {
 820         {Opt_bsd_df, "bsddf"},
 821         {Opt_minix_df, "minixdf"},
 822         {Opt_grpid, "grpid"},
 823         {Opt_grpid, "bsdgroups"},
 824         {Opt_nogrpid, "nogrpid"},
 825         {Opt_nogrpid, "sysvgroups"},
 826         {Opt_resgid, "resgid=%u"},
 827         {Opt_resuid, "resuid=%u"},
 828         {Opt_sb, "sb=%u"},
 829         {Opt_err_cont, "errors=continue"},
 830         {Opt_err_panic, "errors=panic"},
 831         {Opt_err_ro, "errors=remount-ro"},
 832         {Opt_nouid32, "nouid32"},
 833         {Opt_nocheck, "nocheck"},
 834         {Opt_nocheck, "check=none"},
 835         {Opt_debug, "debug"},
 836         {Opt_oldalloc, "oldalloc"},
 837         {Opt_orlov, "orlov"},
 838         {Opt_user_xattr, "user_xattr"},
 839         {Opt_nouser_xattr, "nouser_xattr"},
 840         {Opt_acl, "acl"},
 841         {Opt_noacl, "noacl"},
 842         {Opt_reservation, "reservation"},
 843         {Opt_noreservation, "noreservation"},
 844         {Opt_noload, "noload"},
 845         {Opt_nobh, "nobh"},
 846         {Opt_bh, "bh"},
 847         {Opt_commit, "commit=%u"},
 848         {Opt_journal_update, "journal=update"},
 849         {Opt_journal_inum, "journal=%u"},
 850         {Opt_journal_dev, "journal_dev=%u"},
 851         {Opt_abort, "abort"},
 852         {Opt_data_journal, "data=journal"},
 853         {Opt_data_ordered, "data=ordered"},
 854         {Opt_data_writeback, "data=writeback"},
 855         {Opt_offusrjquota, "usrjquota="},
 856         {Opt_usrjquota, "usrjquota=%s"},
 857         {Opt_offgrpjquota, "grpjquota="},
 858         {Opt_grpjquota, "grpjquota=%s"},
 859         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 860         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 861         {Opt_grpquota, "grpquota"},
 862         {Opt_noquota, "noquota"},
 863         {Opt_quota, "quota"},
 864         {Opt_usrquota, "usrquota"},
 865         {Opt_barrier, "barrier=%u"},
 866         {Opt_extents, "extents"},
 867         {Opt_noextents, "noextents"},
 868         {Opt_err, NULL},
 869         {Opt_resize, "resize"},
 870 };
 871
 872 static ext4_fsblk_t get_sb_block(void **data)
 873 {
 874         ext4_fsblk_t    sb_block;
 875         char            *options = (char *) *data;
 876
 877         if (!options || strncmp(options, "sb=", 3) != 0)
 878                 return 1;       /* Default location */
 879         options += 3;
 880         /*todo: use simple_strtoll with >32bit ext4 */
 881         sb_block = simple_strtoul(options, &options, 0);
 882         if (*options && *options != ',') {
 883                 printk("EXT4-fs: Invalid sb specification: %s\n",
 884                        (char *) *data);
 885                 return 1;
 886         }
 887         if (*options == ',')
 888                 options++;
 889         *data = (void *) options;
 890         return sb_block;
 891 }
 892
 893 static int parse_options (char *options, struct super_block *sb,
 894                           unsigned int *inum, unsigned long *journal_devnum,
 895                           ext4_fsblk_t *n_blocks_count, int is_remount)
 896 {
 897         struct ext4_sb_info *sbi = EXT4_SB(sb);
 898         char * p;
 899         substring_t args[MAX_OPT_ARGS];
 900         int data_opt = 0;
 901         int option;
 902 #ifdef CONFIG_QUOTA
 903         int qtype;
 904         char *qname;
 905 #endif
 906
 907         if (!options)
 908                 return 1;
 909
 910         while ((p = strsep (&options, ",")) != NULL) {
 911                 int token;
 912                 if (!*p)
 913                         continue;
 914
 915                 token = match_token(p, tokens, args);
 916                 switch (token) {
 917                 case Opt_bsd_df:
 918                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 919                         break;
 920                 case Opt_minix_df:
 921                         set_opt (sbi->s_mount_opt, MINIX_DF);
 922                         break;
 923                 case Opt_grpid:
 924                         set_opt (sbi->s_mount_opt, GRPID);
 925                         break;
 926                 case Opt_nogrpid:
 927                         clear_opt (sbi->s_mount_opt, GRPID);
 928                         break;
 929                 case Opt_resuid:
 930                         if (match_int(&args[0], &option))
 931                                 return 0;
 932                         sbi->s_resuid = option;
 933                         break;
 934                 case Opt_resgid:
 935                         if (match_int(&args[0], &option))
 936                                 return 0;
 937                         sbi->s_resgid = option;
 938                         break;
 939                 case Opt_sb:
 940                         /* handled by get_sb_block() instead of here */
 941                         /* *sb_block = match_int(&args[0]); */
 942                         break;
 943                 case Opt_err_panic:
 944                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 945                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 946                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 947                         break;
 948                 case Opt_err_ro:
 949                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 950                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 951                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 952                         break;
 953                 case Opt_err_cont:
 954                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 955                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 956                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 957                         break;
 958                 case Opt_nouid32:
 959                         set_opt (sbi->s_mount_opt, NO_UID32);
 960                         break;
 961                 case Opt_nocheck:
 962                         clear_opt (sbi->s_mount_opt, CHECK);
 963                         break;
 964                 case Opt_debug:
 965                         set_opt (sbi->s_mount_opt, DEBUG);
 966                         break;
 967                 case Opt_oldalloc:
 968                         set_opt (sbi->s_mount_opt, OLDALLOC);
 969                         break;
 970                 case Opt_orlov:
 971                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 972                         break;
 973 #ifdef CONFIG_EXT4DEV_FS_XATTR
 974                 case Opt_user_xattr:
 975                         set_opt (sbi->s_mount_opt, XATTR_USER);
 976                         break;
 977                 case Opt_nouser_xattr:
 978                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 979                         break;
 980 #else
 981                 case Opt_user_xattr:
 982                 case Opt_nouser_xattr:
 983                         printk("EXT4 (no)user_xattr options not supported\n");
 984                         break;
 985 #endif
 986 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 987                 case Opt_acl:
 988                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 989                         break;
 990                 case Opt_noacl:
 991                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 992                         break;
 993 #else
 994                 case Opt_acl:
 995                 case Opt_noacl:
 996                         printk("EXT4 (no)acl options not supported\n");
 997                         break;
 998 #endif
 999                 case Opt_reservation:
1000                         set_opt(sbi->s_mount_opt, RESERVATION);
1001                         break;
1002                 case Opt_noreservation:
1003                         clear_opt(sbi->s_mount_opt, RESERVATION);
1004                         break;
1005                 case Opt_journal_update:
1006                         /* @@@ FIXME */
1007                         /* Eventually we will want to be able to create
1008                            a journal file here.  For now, only allow the
1009                            user to specify an existing inode to be the
1010                            journal file. */
1011                         if (is_remount) {
1012                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1013                                        "journal on remount\n");
1014                                 return 0;
1015                         }
1016                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1017                         break;
1018                 case Opt_journal_inum:
1019                         if (is_remount) {
1020                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1021                                        "journal on remount\n");
1022                                 return 0;
1023                         }
1024                         if (match_int(&args[0], &option))
1025                                 return 0;
1026                         *inum = option;
1027                         break;
1028                 case Opt_journal_dev:
1029                         if (is_remount) {
1030                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1031                                        "journal on remount\n");
1032                                 return 0;
1033                         }
1034                         if (match_int(&args[0], &option))
1035                                 return 0;
1036                         *journal_devnum = option;
1037                         break;
1038                 case Opt_noload:
1039                         set_opt (sbi->s_mount_opt, NOLOAD);
1040                         break;
1041                 case Opt_commit:
1042                         if (match_int(&args[0], &option))
1043                                 return 0;
1044                         if (option < 0)
1045                                 return 0;
1046                         if (option == 0)
1047                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1048                         sbi->s_commit_interval = HZ * option;
1049                         break;
1050                 case Opt_data_journal:
1051                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1052                         goto datacheck;
1053                 case Opt_data_ordered:
1054                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1055                         goto datacheck;
1056                 case Opt_data_writeback:
1057                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1058                 datacheck:
1059                         if (is_remount) {
1060                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1061                                                 != data_opt) {
1062                                         printk(KERN_ERR
1063                                                 "EXT4-fs: cannot change data "
1064                                                 "mode on remount\n");
1065                                         return 0;
1066                                 }
1067                         } else {
1068                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1069                                 sbi->s_mount_opt |= data_opt;
1070                         }
1071                         break;
1072 #ifdef CONFIG_QUOTA
1073                 case Opt_usrjquota:
1074                         qtype = USRQUOTA;
1075                         goto set_qf_name;
1076                 case Opt_grpjquota:
1077                         qtype = GRPQUOTA;
1078 set_qf_name:
1079                         if (sb_any_quota_enabled(sb)) {
1080                                 printk(KERN_ERR
1081                                         "EXT4-fs: Cannot change journalled "
1082                                         "quota options when quota turned on.\n");
1083                                 return 0;
1084                         }
1085                         qname = match_strdup(&args[0]);
1086                         if (!qname) {
1087                                 printk(KERN_ERR
1088                                         "EXT4-fs: not enough memory for "
1089                                         "storing quotafile name.\n");
1090                                 return 0;
1091                         }
1092                         if (sbi->s_qf_names[qtype] &&
1093                             strcmp(sbi->s_qf_names[qtype], qname)) {
1094                                 printk(KERN_ERR
1095                                         "EXT4-fs: %s quota file already "
1096                                         "specified.\n", QTYPE2NAME(qtype));
1097                                 kfree(qname);
1098                                 return 0;
1099                         }
1100                         sbi->s_qf_names[qtype] = qname;
1101                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1102                                 printk(KERN_ERR
1103                                         "EXT4-fs: quotafile must be on "
1104                                         "filesystem root.\n");
1105                                 kfree(sbi->s_qf_names[qtype]);
1106                                 sbi->s_qf_names[qtype] = NULL;
1107                                 return 0;
1108                         }
1109                         set_opt(sbi->s_mount_opt, QUOTA);
1110                         break;
1111                 case Opt_offusrjquota:
1112                         qtype = USRQUOTA;
1113                         goto clear_qf_name;
1114                 case Opt_offgrpjquota:
1115                         qtype = GRPQUOTA;
1116 clear_qf_name:
1117                         if (sb_any_quota_enabled(sb)) {
1118                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1119                                         "journalled quota options when "
1120                                         "quota turned on.\n");
1121                                 return 0;
1122                         }
1123                         /*
1124                          * The space will be released later when all options
1125                          * are confirmed to be correct
1126                          */
1127                         sbi->s_qf_names[qtype] = NULL;
1128                         break;
1129                 case Opt_jqfmt_vfsold:
1130                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1131                         break;
1132                 case Opt_jqfmt_vfsv0:
1133                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1134                         break;
1135                 case Opt_quota:
1136                 case Opt_usrquota:
1137                         set_opt(sbi->s_mount_opt, QUOTA);
1138                         set_opt(sbi->s_mount_opt, USRQUOTA);
1139                         break;
1140                 case Opt_grpquota:
1141                         set_opt(sbi->s_mount_opt, QUOTA);
1142                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1143                         break;
1144                 case Opt_noquota:
1145                         if (sb_any_quota_enabled(sb)) {
1146                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1147                                         "options when quota turned on.\n");
1148                                 return 0;
1149                         }
1150                         clear_opt(sbi->s_mount_opt, QUOTA);
1151                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1152                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1153                         break;
1154 #else
1155                 case Opt_quota:
1156                 case Opt_usrquota:
1157                 case Opt_grpquota:
1158                 case Opt_usrjquota:
1159                 case Opt_grpjquota:
1160                 case Opt_offusrjquota:
1161                 case Opt_offgrpjquota:
1162                 case Opt_jqfmt_vfsold:
1163                 case Opt_jqfmt_vfsv0:
1164                         printk(KERN_ERR
1165                                 "EXT4-fs: journalled quota options not "
1166                                 "supported.\n");
1167                         break;
1168                 case Opt_noquota:
1169                         break;
1170 #endif
1171                 case Opt_abort:
1172                         set_opt(sbi->s_mount_opt, ABORT);
1173                         break;
1174                 case Opt_barrier:
1175                         if (match_int(&args[0], &option))
1176                                 return 0;
1177                         if (option)
1178                                 set_opt(sbi->s_mount_opt, BARRIER);
1179                         else
1180                                 clear_opt(sbi->s_mount_opt, BARRIER);
1181                         break;
1182                 case Opt_ignore:
1183                         break;
1184                 case Opt_resize:
1185                         if (!is_remount) {
1186                                 printk("EXT4-fs: resize option only available "
1187                                         "for remount\n");
1188                                 return 0;
1189                         }
1190                         if (match_int(&args[0], &option) != 0)
1191                                 return 0;
1192                         *n_blocks_count = option;
1193                         break;
1194                 case Opt_nobh:
1195                         set_opt(sbi->s_mount_opt, NOBH);
1196                         break;
1197                 case Opt_bh:
1198                         clear_opt(sbi->s_mount_opt, NOBH);
1199                         break;
1200                 case Opt_extents:
1201                         set_opt (sbi->s_mount_opt, EXTENTS);
1202                         break;
1203                 case Opt_noextents:
1204                         clear_opt (sbi->s_mount_opt, EXTENTS);
1205                         break;
1206                 default:
1207                         printk (KERN_ERR
1208                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1209                                 "or missing value\n", p);
1210                         return 0;
1211                 }
1212         }
1213 #ifdef CONFIG_QUOTA
1214         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1215                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1216                      sbi->s_qf_names[USRQUOTA])
1217                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1218
1219                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1220                      sbi->s_qf_names[GRPQUOTA])
1221                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1222
1223                 if ((sbi->s_qf_names[USRQUOTA] &&
1224                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1225                     (sbi->s_qf_names[GRPQUOTA] &&
1226                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1227                         printk(KERN_ERR "EXT4-fs: old and new quota "
1228                                         "format mixing.\n");
1229                         return 0;
1230                 }
1231
1232                 if (!sbi->s_jquota_fmt) {
1233                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1234                                         "not specified.\n");
1235                         return 0;
1236                 }
1237         } else {
1238                 if (sbi->s_jquota_fmt) {
1239                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1240                                         "specified with no journalling "
1241                                         "enabled.\n");
1242                         return 0;
1243                 }
1244         }
1245 #endif
1246         return 1;
1247 }
1248
1249 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1250                             int read_only)
1251 {
1252         struct ext4_sb_info *sbi = EXT4_SB(sb);
1253         int res = 0;
1254
1255         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1256                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1257                         "forcing read-only mode\n");
1258                 res = MS_RDONLY;
1259         }
1260         if (read_only)
1261                 return res;
1262         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1263                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1264                         "running e2fsck is recommended\n");
1265         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1266                 printk (KERN_WARNING
1267                         "EXT4-fs warning: mounting fs with errors, "
1268                         "running e2fsck is recommended\n");
1269         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1270                  le16_to_cpu(es->s_mnt_count) >=
1271                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1272                 printk (KERN_WARNING
1273                         "EXT4-fs warning: maximal mount count reached, "
1274                         "running e2fsck is recommended\n");
1275         else if (le32_to_cpu(es->s_checkinterval) &&
1276                 (le32_to_cpu(es->s_lastcheck) +
1277                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1278                 printk (KERN_WARNING
1279                         "EXT4-fs warning: checktime reached, "
1280                         "running e2fsck is recommended\n");
1281 #if 0
1282                 /* @@@ We _will_ want to clear the valid bit if we find
1283                  * inconsistencies, to force a fsck at reboot.  But for
1284                  * a plain journaled filesystem we can keep it set as
1285                  * valid forever! :)
1286                  */
1287         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1288 #endif
1289         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1290                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1291         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1292         es->s_mtime = cpu_to_le32(get_seconds());
1293         ext4_update_dynamic_rev(sb);
1294         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1295
1296         ext4_commit_super(sb, es, 1);
1297         if (test_opt(sb, DEBUG))
1298                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1299                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1300                         sb->s_blocksize,
1301                         sbi->s_groups_count,
1302                         EXT4_BLOCKS_PER_GROUP(sb),
1303                         EXT4_INODES_PER_GROUP(sb),
1304                         sbi->s_mount_opt);
1305
1306         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1307         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1308                 char b[BDEVNAME_SIZE];
1309
1310                 printk("external journal on %s\n",
1311                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1312         } else {
1313                 printk("internal journal\n");
1314         }
1315         return res;
1316 }
1317
1318 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1319                             struct ext4_group_desc *gdp)
1320 {
1321         __u16 crc = 0;
1322
1323         if (sbi->s_es->s_feature_ro_compat &
1324             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1325                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1326                 __le32 le_group = cpu_to_le32(block_group);
1327
1328                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1329                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1330                 crc = crc16(crc, (__u8 *)gdp, offset);
1331                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1332                 /* for checksum of struct ext4_group_desc do the rest...*/
1333                 if ((sbi->s_es->s_feature_incompat &
1334                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1335                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1336                         crc = crc16(crc, (__u8 *)gdp + offset,
1337                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1338                                         offset);
1339         }
1340
1341         return cpu_to_le16(crc);
1342 }
1343
1344 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1345                                 struct ext4_group_desc *gdp)
1346 {
1347         if ((sbi->s_es->s_feature_ro_compat &
1348              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1349             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1350                 return 0;
1351
1352         return 1;
1353 }
1354
1355 /* Called at mount-time, super-block is locked */
1356 static int ext4_check_descriptors (struct super_block * sb)
1357 {
1358         struct ext4_sb_info *sbi = EXT4_SB(sb);
1359         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1360         ext4_fsblk_t last_block;
1361         ext4_fsblk_t block_bitmap;
1362         ext4_fsblk_t inode_bitmap;
1363         ext4_fsblk_t inode_table;
1364         struct ext4_group_desc * gdp = NULL;
1365         int desc_block = 0;
1366         int flexbg_flag = 0;
1367         ext4_group_t i;
1368
1369         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1370                 flexbg_flag = 1;
1371
1372         ext4_debug ("Checking group descriptors");
1373
1374         for (i = 0; i < sbi->s_groups_count; i++)
1375         {
1376                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1377                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1378                 else
1379                         last_block = first_block +
1380                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1381
1382                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1383                         gdp = (struct ext4_group_desc *)
1384                                         sbi->s_group_desc[desc_block++]->b_data;
1385                 block_bitmap = ext4_block_bitmap(sb, gdp);
1386                 if (block_bitmap < first_block || block_bitmap > last_block)
1387                 {
1388                         ext4_error (sb, "ext4_check_descriptors",
1389                                     "Block bitmap for group %lu"
1390                                     " not in group (block %llu)!",
1391                                     i, block_bitmap);
1392                         return 0;
1393                 }
1394                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1395                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1396                 {
1397                         ext4_error (sb, "ext4_check_descriptors",
1398                                     "Inode bitmap for group %lu"
1399                                     " not in group (block %llu)!",
1400                                     i, inode_bitmap);
1401                         return 0;
1402                 }
1403                 inode_table = ext4_inode_table(sb, gdp);
1404                 if (inode_table < first_block ||
1405                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1406                 {
1407                         ext4_error (sb, "ext4_check_descriptors",
1408                                     "Inode table for group %lu"
1409                                     " not in group (block %llu)!",
1410                                     i, inode_table);
1411                         return 0;
1412                 }
1413                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1414                         ext4_error(sb, __FUNCTION__,
1415                                    "Checksum for group %lu failed (%u!=%u)\n",
1416                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1417                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1418                         return 0;
1419                 }
1420                 if (!flexbg_flag)
1421                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1422                 gdp = (struct ext4_group_desc *)
1423                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1424         }
1425
1426         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1427         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1428         return 1;
1429 }
1430
1431 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1432  * the superblock) which were deleted from all directories, but held open by
1433  * a process at the time of a crash.  We walk the list and try to delete these
1434  * inodes at recovery time (only with a read-write filesystem).
1435  *
1436  * In order to keep the orphan inode chain consistent during traversal (in
1437  * case of crash during recovery), we link each inode into the superblock
1438  * orphan list_head and handle it the same way as an inode deletion during
1439  * normal operation (which journals the operations for us).
1440  *
1441  * We only do an iget() and an iput() on each inode, which is very safe if we
1442  * accidentally point at an in-use or already deleted inode.  The worst that
1443  * can happen in this case is that we get a "bit already cleared" message from
1444  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1445  * e2fsck was run on this filesystem, and it must have already done the orphan
1446  * inode cleanup for us, so we can safely abort without any further action.
1447  */
1448 static void ext4_orphan_cleanup (struct super_block * sb,
1449                                  struct ext4_super_block * es)
1450 {
1451         unsigned int s_flags = sb->s_flags;
1452         int nr_orphans = 0, nr_truncates = 0;
1453 #ifdef CONFIG_QUOTA
1454         int i;
1455 #endif
1456         if (!es->s_last_orphan) {
1457                 jbd_debug(4, "no orphan inodes to clean up\n");
1458                 return;
1459         }
1460
1461         if (bdev_read_only(sb->s_bdev)) {
1462                 printk(KERN_ERR "EXT4-fs: write access "
1463                         "unavailable, skipping orphan cleanup.\n");
1464                 return;
1465         }
1466
1467         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1468                 if (es->s_last_orphan)
1469                         jbd_debug(1, "Errors on filesystem, "
1470                                   "clearing orphan list.\n");
1471                 es->s_last_orphan = 0;
1472                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1473                 return;
1474         }
1475
1476         if (s_flags & MS_RDONLY) {
1477                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1478                        sb->s_id);
1479                 sb->s_flags &= ~MS_RDONLY;
1480         }
1481 #ifdef CONFIG_QUOTA
1482         /* Needed for iput() to work correctly and not trash data */
1483         sb->s_flags |= MS_ACTIVE;
1484         /* Turn on quotas so that they are updated correctly */
1485         for (i = 0; i < MAXQUOTAS; i++) {
1486                 if (EXT4_SB(sb)->s_qf_names[i]) {
1487                         int ret = ext4_quota_on_mount(sb, i);
1488                         if (ret < 0)
1489                                 printk(KERN_ERR
1490                                         "EXT4-fs: Cannot turn on journalled "
1491                                         "quota: error %d\n", ret);
1492                 }
1493         }
1494 #endif
1495
1496         while (es->s_last_orphan) {
1497                 struct inode *inode;
1498
1499                 if (!(inode =
1500                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1501                         es->s_last_orphan = 0;
1502                         break;
1503                 }
1504
1505                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1506                 DQUOT_INIT(inode);
1507                 if (inode->i_nlink) {
1508                         printk(KERN_DEBUG
1509                                 "%s: truncating inode %lu to %Ld bytes\n",
1510                                 __FUNCTION__, inode->i_ino, inode->i_size);
1511                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1512                                   inode->i_ino, inode->i_size);
1513                         ext4_truncate(inode);
1514                         nr_truncates++;
1515                 } else {
1516                         printk(KERN_DEBUG
1517                                 "%s: deleting unreferenced inode %lu\n",
1518                                 __FUNCTION__, inode->i_ino);
1519                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1520                                   inode->i_ino);
1521                         nr_orphans++;
1522                 }
1523                 iput(inode);  /* The delete magic happens here! */
1524         }
1525
1526 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1527
1528         if (nr_orphans)
1529                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1530                        sb->s_id, PLURAL(nr_orphans));
1531         if (nr_truncates)
1532                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1533                        sb->s_id, PLURAL(nr_truncates));
1534 #ifdef CONFIG_QUOTA
1535         /* Turn quotas off */
1536         for (i = 0; i < MAXQUOTAS; i++) {
1537                 if (sb_dqopt(sb)->files[i])
1538                         vfs_quota_off(sb, i);
1539         }
1540 #endif
1541         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1542 }
1543
1544 /*
1545  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1546  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1547  * We need to be 1 filesystem block less than the 2^32 sector limit.
1548  */
1549 static loff_t ext4_max_size(int bits)
1550 {
1551         loff_t res = EXT4_NDIR_BLOCKS;
1552         /* This constant is calculated to be the largest file size for a
1553          * dense, 4k-blocksize file such that the total number of
1554          * sectors in the file, including data and all indirect blocks,
1555          * does not exceed 2^32. */
1556         const loff_t upper_limit = 0x1ff7fffd000LL;
1557
1558         res += 1LL << (bits-2);
1559         res += 1LL << (2*(bits-2));
1560         res += 1LL << (3*(bits-2));
1561         res <<= bits;
1562         if (res > upper_limit)
1563                 res = upper_limit;
1564         return res;
1565 }
1566
1567 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1568                                 ext4_fsblk_t logical_sb_block, int nr)
1569 {
1570         struct ext4_sb_info *sbi = EXT4_SB(sb);
1571         ext4_group_t bg, first_meta_bg;
1572         int has_super = 0;
1573
1574         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1575
1576         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1577             nr < first_meta_bg)
1578                 return logical_sb_block + nr + 1;
1579         bg = sbi->s_desc_per_block * nr;
1580         if (ext4_bg_has_super(sb, bg))
1581                 has_super = 1;
1582         return (has_super + ext4_group_first_block_no(sb, bg));
1583 }
1584
1585
1586 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1587 {
1588         struct buffer_head * bh;
1589         struct ext4_super_block *es = NULL;
1590         struct ext4_sb_info *sbi;
1591         ext4_fsblk_t block;
1592         ext4_fsblk_t sb_block = get_sb_block(&data);
1593         ext4_fsblk_t logical_sb_block;
1594         unsigned long offset = 0;
1595         unsigned int journal_inum = 0;
1596         unsigned long journal_devnum = 0;
1597         unsigned long def_mount_opts;
1598         struct inode *root;
1599         int blocksize;
1600         int hblock;
1601         int db_count;
1602         int i;
1603         int needs_recovery;
1604         __le32 features;
1605         __u64 blocks_count;
1606         int err;
1607
1608         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1609         if (!sbi)
1610                 return -ENOMEM;
1611         sb->s_fs_info = sbi;
1612         sbi->s_mount_opt = 0;
1613         sbi->s_resuid = EXT4_DEF_RESUID;
1614         sbi->s_resgid = EXT4_DEF_RESGID;
1615         sbi->s_sb_block = sb_block;
1616
1617         unlock_kernel();
1618
1619         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1620         if (!blocksize) {
1621                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1622                 goto out_fail;
1623         }
1624
1625         if (!sb_set_blocksize(sb, blocksize)) {
1626                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1627                 goto out_fail;
1628         }
1629
1630         /*
1631          * The ext4 superblock will not be buffer aligned for other than 1kB
1632          * block sizes.  We need to calculate the offset from buffer start.
1633          */
1634         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1635                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1636                 offset = do_div(logical_sb_block, blocksize);
1637         } else {
1638                 logical_sb_block = sb_block;
1639         }
1640
1641         if (!(bh = sb_bread(sb, logical_sb_block))) {
1642                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1643                 goto out_fail;
1644         }
1645         /*
1646          * Note: s_es must be initialized as soon as possible because
1647          *       some ext4 macro-instructions depend on its value
1648          */
1649         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1650         sbi->s_es = es;
1651         sb->s_magic = le16_to_cpu(es->s_magic);
1652         if (sb->s_magic != EXT4_SUPER_MAGIC)
1653                 goto cantfind_ext4;
1654
1655         /* Set defaults before we parse the mount options */
1656         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1657         if (def_mount_opts & EXT4_DEFM_DEBUG)
1658                 set_opt(sbi->s_mount_opt, DEBUG);
1659         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1660                 set_opt(sbi->s_mount_opt, GRPID);
1661         if (def_mount_opts & EXT4_DEFM_UID16)
1662                 set_opt(sbi->s_mount_opt, NO_UID32);
1663 #ifdef CONFIG_EXT4DEV_FS_XATTR
1664         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1665                 set_opt(sbi->s_mount_opt, XATTR_USER);
1666 #endif
1667 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1668         if (def_mount_opts & EXT4_DEFM_ACL)
1669                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1670 #endif
1671         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1672                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1673         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1674                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1675         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1676                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1677
1678         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1679                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1680         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1681                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1682         else
1683                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1684
1685         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1686         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1687
1688         set_opt(sbi->s_mount_opt, RESERVATION);
1689
1690         /*
1691          * turn on extents feature by default in ext4 filesystem
1692          * User -o noextents to turn it off
1693          */
1694         set_opt(sbi->s_mount_opt, EXTENTS);
1695
1696         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1697                             NULL, 0))
1698                 goto failed_mount;
1699
1700         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1701                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1702
1703         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1704             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1705              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1706              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1707                 printk(KERN_WARNING
1708                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1709                        "running e2fsck is recommended\n");
1710         /*
1711          * Check feature flags regardless of the revision level, since we
1712          * previously didn't change the revision level when setting the flags,
1713          * so there is a chance incompat flags are set on a rev 0 filesystem.
1714          */
1715         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1716         if (features) {
1717                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1718                        "unsupported optional features (%x).\n",
1719                        sb->s_id, le32_to_cpu(features));
1720                 goto failed_mount;
1721         }
1722         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1723         if (!(sb->s_flags & MS_RDONLY) && features) {
1724                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1725                        "unsupported optional features (%x).\n",
1726                        sb->s_id, le32_to_cpu(features));
1727                 goto failed_mount;
1728         }
1729         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1730
1731         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1732             blocksize > EXT4_MAX_BLOCK_SIZE) {
1733                 printk(KERN_ERR
1734                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1735                        blocksize, sb->s_id);
1736                 goto failed_mount;
1737         }
1738
1739         hblock = bdev_hardsect_size(sb->s_bdev);
1740         if (sb->s_blocksize != blocksize) {
1741                 /*
1742                  * Make sure the blocksize for the filesystem is larger
1743                  * than the hardware sectorsize for the machine.
1744                  */
1745                 if (blocksize < hblock) {
1746                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1747                                "device blocksize %d.\n", blocksize, hblock);
1748                         goto failed_mount;
1749                 }
1750
1751                 brelse (bh);
1752                 sb_set_blocksize(sb, blocksize);
1753                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1754                 offset = do_div(logical_sb_block, blocksize);
1755                 bh = sb_bread(sb, logical_sb_block);
1756                 if (!bh) {
1757                         printk(KERN_ERR
1758                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1759                         goto failed_mount;
1760                 }
1761                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1762                 sbi->s_es = es;
1763                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1764                         printk (KERN_ERR
1765                                 "EXT4-fs: Magic mismatch, very weird !\n");
1766                         goto failed_mount;
1767                 }
1768         }
1769
1770         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1771
1772         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1773                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1774                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1775         } else {
1776                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1777                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1778                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1779                     (!is_power_of_2(sbi->s_inode_size)) ||
1780                     (sbi->s_inode_size > blocksize)) {
1781                         printk (KERN_ERR
1782                                 "EXT4-fs: unsupported inode size: %d\n",
1783                                 sbi->s_inode_size);
1784                         goto failed_mount;
1785                 }
1786                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
1787                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
1788         }
1789         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1790         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1791                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1792                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1793                     !is_power_of_2(sbi->s_desc_size)) {
1794                         printk(KERN_ERR
1795                                "EXT4-fs: unsupported descriptor size %lu\n",
1796                                sbi->s_desc_size);
1797                         goto failed_mount;
1798                 }
1799         } else
1800                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1801         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1802         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1803         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
1804                 goto cantfind_ext4;
1805         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1806         if (sbi->s_inodes_per_block == 0)
1807                 goto cantfind_ext4;
1808         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1809                                         sbi->s_inodes_per_block;
1810         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1811         sbi->s_sbh = bh;
1812         sbi->s_mount_state = le16_to_cpu(es->s_state);
1813         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
1814         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
1815         for (i=0; i < 4; i++)
1816                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1817         sbi->s_def_hash_version = es->s_def_hash_version;
1818
1819         if (sbi->s_blocks_per_group > blocksize * 8) {
1820                 printk (KERN_ERR
1821                         "EXT4-fs: #blocks per group too big: %lu\n",
1822                         sbi->s_blocks_per_group);
1823                 goto failed_mount;
1824         }
1825         if (sbi->s_inodes_per_group > blocksize * 8) {
1826                 printk (KERN_ERR
1827                         "EXT4-fs: #inodes per group too big: %lu\n",
1828                         sbi->s_inodes_per_group);
1829                 goto failed_mount;
1830         }
1831
1832         if (ext4_blocks_count(es) >
1833                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1834                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1835                         " too large to mount safely\n", sb->s_id);
1836                 if (sizeof(sector_t) < 8)
1837                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1838                                         "enabled\n");
1839                 goto failed_mount;
1840         }
1841
1842         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1843                 goto cantfind_ext4;
1844         blocks_count = (ext4_blocks_count(es) -
1845                         le32_to_cpu(es->s_first_data_block) +
1846                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1847         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1848         sbi->s_groups_count = blocks_count;
1849         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1850                    EXT4_DESC_PER_BLOCK(sb);
1851         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1852                                     GFP_KERNEL);
1853         if (sbi->s_group_desc == NULL) {
1854                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1855                 goto failed_mount;
1856         }
1857
1858         bgl_lock_init(&sbi->s_blockgroup_lock);
1859
1860         for (i = 0; i < db_count; i++) {
1861                 block = descriptor_loc(sb, logical_sb_block, i);
1862                 sbi->s_group_desc[i] = sb_bread(sb, block);
1863                 if (!sbi->s_group_desc[i]) {
1864                         printk (KERN_ERR "EXT4-fs: "
1865                                 "can't read group descriptor %d\n", i);
1866                         db_count = i;
1867                         goto failed_mount2;
1868                 }
1869         }
1870         if (!ext4_check_descriptors (sb)) {
1871                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1872                 goto failed_mount2;
1873         }
1874         sbi->s_gdb_count = db_count;
1875         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1876         spin_lock_init(&sbi->s_next_gen_lock);
1877
1878         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1879                         ext4_count_free_blocks(sb));
1880         if (!err) {
1881                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1882                                 ext4_count_free_inodes(sb));
1883         }
1884         if (!err) {
1885                 err = percpu_counter_init(&sbi->s_dirs_counter,
1886                                 ext4_count_dirs(sb));
1887         }
1888         if (err) {
1889                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
1890                 goto failed_mount3;
1891         }
1892
1893         /* per fileystem reservation list head & lock */
1894         spin_lock_init(&sbi->s_rsv_window_lock);
1895         sbi->s_rsv_window_root = RB_ROOT;
1896         /* Add a single, static dummy reservation to the start of the
1897          * reservation window list --- it gives us a placeholder for
1898          * append-at-start-of-list which makes the allocation logic
1899          * _much_ simpler. */
1900         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1901         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1902         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1903         sbi->s_rsv_window_head.rsv_goal_size = 0;
1904         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1905
1906         /*
1907          * set up enough so that it can read an inode
1908          */
1909         sb->s_op = &ext4_sops;
1910         sb->s_export_op = &ext4_export_ops;
1911         sb->s_xattr = ext4_xattr_handlers;
1912 #ifdef CONFIG_QUOTA
1913         sb->s_qcop = &ext4_qctl_operations;
1914         sb->dq_op = &ext4_quota_operations;
1915 #endif
1916         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1917
1918         sb->s_root = NULL;
1919
1920         needs_recovery = (es->s_last_orphan != 0 ||
1921                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1922                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1923
1924         /*
1925          * The first inode we look at is the journal inode.  Don't try
1926          * root first: it may be modified in the journal!
1927          */
1928         if (!test_opt(sb, NOLOAD) &&
1929             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1930                 if (ext4_load_journal(sb, es, journal_devnum))
1931                         goto failed_mount3;
1932         } else if (journal_inum) {
1933                 if (ext4_create_journal(sb, es, journal_inum))
1934                         goto failed_mount3;
1935         } else {
1936                 if (!silent)
1937                         printk (KERN_ERR
1938                                 "ext4: No journal on filesystem on %s\n",
1939                                 sb->s_id);
1940                 goto failed_mount3;
1941         }
1942
1943         if (ext4_blocks_count(es) > 0xffffffffULL &&
1944             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
1945                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
1946                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
1947                 goto failed_mount4;
1948         }
1949
1950         /* We have now updated the journal if required, so we can
1951          * validate the data journaling mode. */
1952         switch (test_opt(sb, DATA_FLAGS)) {
1953         case 0:
1954                 /* No mode set, assume a default based on the journal
1955                  * capabilities: ORDERED_DATA if the journal can
1956                  * cope, else JOURNAL_DATA
1957                  */
1958                 if (jbd2_journal_check_available_features
1959                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
1960                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1961                 else
1962                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1963                 break;
1964
1965         case EXT4_MOUNT_ORDERED_DATA:
1966         case EXT4_MOUNT_WRITEBACK_DATA:
1967                 if (!jbd2_journal_check_available_features
1968                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
1969                         printk(KERN_ERR "EXT4-fs: Journal does not support "
1970                                "requested data journaling mode\n");
1971                         goto failed_mount4;
1972                 }
1973         default:
1974                 break;
1975         }
1976
1977         if (test_opt(sb, NOBH)) {
1978                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
1979                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
1980                                 "its supported only with writeback mode\n");
1981                         clear_opt(sbi->s_mount_opt, NOBH);
1982                 }
1983         }
1984         /*
1985          * The jbd2_journal_load will have done any necessary log recovery,
1986          * so we can safely mount the rest of the filesystem now.
1987          */
1988
1989         root = iget(sb, EXT4_ROOT_INO);
1990         sb->s_root = d_alloc_root(root);
1991         if (!sb->s_root) {
1992                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
1993                 iput(root);
1994                 goto failed_mount4;
1995         }
1996         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1997                 dput(sb->s_root);
1998                 sb->s_root = NULL;
1999                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2000                 goto failed_mount4;
2001         }
2002
2003         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2004
2005         /* determine the minimum size of new large inodes, if present */
2006         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2007                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2008                                                      EXT4_GOOD_OLD_INODE_SIZE;
2009                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2010                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2011                         if (sbi->s_want_extra_isize <
2012                             le16_to_cpu(es->s_want_extra_isize))
2013                                 sbi->s_want_extra_isize =
2014                                         le16_to_cpu(es->s_want_extra_isize);
2015                         if (sbi->s_want_extra_isize <
2016                             le16_to_cpu(es->s_min_extra_isize))
2017                                 sbi->s_want_extra_isize =
2018                                         le16_to_cpu(es->s_min_extra_isize);
2019                 }
2020         }
2021         /* Check if enough inode space is available */
2022         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2023                                                         sbi->s_inode_size) {
2024                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2025                                                        EXT4_GOOD_OLD_INODE_SIZE;
2026                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2027                         "available.\n");
2028         }
2029
2030         /*
2031          * akpm: core read_super() calls in here with the superblock locked.
2032          * That deadlocks, because orphan cleanup needs to lock the superblock
2033          * in numerous places.  Here we just pop the lock - it's relatively
2034          * harmless, because we are now ready to accept write_super() requests,
2035          * and aviro says that's the only reason for hanging onto the
2036          * superblock lock.
2037          */
2038         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2039         ext4_orphan_cleanup(sb, es);
2040         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2041         if (needs_recovery)
2042                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2043         ext4_mark_recovery_complete(sb, es);
2044         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2045                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2046                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2047                 "writeback");
2048
2049         ext4_ext_init(sb);
2050
2051         lock_kernel();
2052         return 0;
2053
2054 cantfind_ext4:
2055         if (!silent)
2056                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2057                        sb->s_id);
2058         goto failed_mount;
2059
2060 failed_mount4:
2061         jbd2_journal_destroy(sbi->s_journal);
2062 failed_mount3:
2063         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2064         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2065         percpu_counter_destroy(&sbi->s_dirs_counter);
2066 failed_mount2:
2067         for (i = 0; i < db_count; i++)
2068                 brelse(sbi->s_group_desc[i]);
2069         kfree(sbi->s_group_desc);
2070 failed_mount:
2071 #ifdef CONFIG_QUOTA
2072         for (i = 0; i < MAXQUOTAS; i++)
2073                 kfree(sbi->s_qf_names[i]);
2074 #endif
2075         ext4_blkdev_remove(sbi);
2076         brelse(bh);
2077 out_fail:
2078         sb->s_fs_info = NULL;
2079         kfree(sbi);
2080         lock_kernel();
2081         return -EINVAL;
2082 }
2083
2084 /*
2085  * Setup any per-fs journal parameters now.  We'll do this both on
2086  * initial mount, once the journal has been initialised but before we've
2087  * done any recovery; and again on any subsequent remount.
2088  */
2089 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2090 {
2091         struct ext4_sb_info *sbi = EXT4_SB(sb);
2092
2093         if (sbi->s_commit_interval)
2094                 journal->j_commit_interval = sbi->s_commit_interval;
2095         /* We could also set up an ext4-specific default for the commit
2096          * interval here, but for now we'll just fall back to the jbd
2097          * default. */
2098
2099         spin_lock(&journal->j_state_lock);
2100         if (test_opt(sb, BARRIER))
2101                 journal->j_flags |= JBD2_BARRIER;
2102         else
2103                 journal->j_flags &= ~JBD2_BARRIER;
2104         spin_unlock(&journal->j_state_lock);
2105 }
2106
2107 static journal_t *ext4_get_journal(struct super_block *sb,
2108                                    unsigned int journal_inum)
2109 {
2110         struct inode *journal_inode;
2111         journal_t *journal;
2112
2113         /* First, test for the existence of a valid inode on disk.  Bad
2114          * things happen if we iget() an unused inode, as the subsequent
2115          * iput() will try to delete it. */
2116
2117         journal_inode = iget(sb, journal_inum);
2118         if (!journal_inode) {
2119                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2120                 return NULL;
2121         }
2122         if (!journal_inode->i_nlink) {
2123                 make_bad_inode(journal_inode);
2124                 iput(journal_inode);
2125                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2126                 return NULL;
2127         }
2128
2129         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2130                   journal_inode, journal_inode->i_size);
2131         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2132                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2133                 iput(journal_inode);
2134                 return NULL;
2135         }
2136
2137         journal = jbd2_journal_init_inode(journal_inode);
2138         if (!journal) {
2139                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2140                 iput(journal_inode);
2141                 return NULL;
2142         }
2143         journal->j_private = sb;
2144         ext4_init_journal_params(sb, journal);
2145         return journal;
2146 }
2147
2148 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2149                                        dev_t j_dev)
2150 {
2151         struct buffer_head * bh;
2152         journal_t *journal;
2153         ext4_fsblk_t start;
2154         ext4_fsblk_t len;
2155         int hblock, blocksize;
2156         ext4_fsblk_t sb_block;
2157         unsigned long offset;
2158         struct ext4_super_block * es;
2159         struct block_device *bdev;
2160
2161         bdev = ext4_blkdev_get(j_dev);
2162         if (bdev == NULL)
2163                 return NULL;
2164
2165         if (bd_claim(bdev, sb)) {
2166                 printk(KERN_ERR
2167                         "EXT4: failed to claim external journal device.\n");
2168                 blkdev_put(bdev);
2169                 return NULL;
2170         }
2171
2172         blocksize = sb->s_blocksize;
2173         hblock = bdev_hardsect_size(bdev);
2174         if (blocksize < hblock) {
2175                 printk(KERN_ERR
2176                         "EXT4-fs: blocksize too small for journal device.\n");
2177                 goto out_bdev;
2178         }
2179
2180         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2181         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2182         set_blocksize(bdev, blocksize);
2183         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2184                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2185                        "external journal\n");
2186                 goto out_bdev;
2187         }
2188
2189         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2190         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2191             !(le32_to_cpu(es->s_feature_incompat) &
2192               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2193                 printk(KERN_ERR "EXT4-fs: external journal has "
2194                                         "bad superblock\n");
2195                 brelse(bh);
2196                 goto out_bdev;
2197         }
2198
2199         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2200                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2201                 brelse(bh);
2202                 goto out_bdev;
2203         }
2204
2205         len = ext4_blocks_count(es);
2206         start = sb_block + 1;
2207         brelse(bh);     /* we're done with the superblock */
2208
2209         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2210                                         start, len, blocksize);
2211         if (!journal) {
2212                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2213                 goto out_bdev;
2214         }
2215         journal->j_private = sb;
2216         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2217         wait_on_buffer(journal->j_sb_buffer);
2218         if (!buffer_uptodate(journal->j_sb_buffer)) {
2219                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2220                 goto out_journal;
2221         }
2222         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2223                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2224                                         "user (unsupported) - %d\n",
2225                         be32_to_cpu(journal->j_superblock->s_nr_users));
2226                 goto out_journal;
2227         }
2228         EXT4_SB(sb)->journal_bdev = bdev;
2229         ext4_init_journal_params(sb, journal);
2230         return journal;
2231 out_journal:
2232         jbd2_journal_destroy(journal);
2233 out_bdev:
2234         ext4_blkdev_put(bdev);
2235         return NULL;
2236 }
2237
2238 static int ext4_load_journal(struct super_block *sb,
2239                              struct ext4_super_block *es,
2240                              unsigned long journal_devnum)
2241 {
2242         journal_t *journal;
2243         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2244         dev_t journal_dev;
2245         int err = 0;
2246         int really_read_only;
2247
2248         if (journal_devnum &&
2249             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2250                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2251                         "numbers have changed\n");
2252                 journal_dev = new_decode_dev(journal_devnum);
2253         } else
2254                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2255
2256         really_read_only = bdev_read_only(sb->s_bdev);
2257
2258         /*
2259          * Are we loading a blank journal or performing recovery after a
2260          * crash?  For recovery, we need to check in advance whether we
2261          * can get read-write access to the device.
2262          */
2263
2264         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2265                 if (sb->s_flags & MS_RDONLY) {
2266                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2267                                         "required on readonly filesystem.\n");
2268                         if (really_read_only) {
2269                                 printk(KERN_ERR "EXT4-fs: write access "
2270                                         "unavailable, cannot proceed.\n");
2271                                 return -EROFS;
2272                         }
2273                         printk (KERN_INFO "EXT4-fs: write access will "
2274                                         "be enabled during recovery.\n");
2275                 }
2276         }
2277
2278         if (journal_inum && journal_dev) {
2279                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2280                        "and inode journals!\n");
2281                 return -EINVAL;
2282         }
2283
2284         if (journal_inum) {
2285                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2286                         return -EINVAL;
2287         } else {
2288                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2289                         return -EINVAL;
2290         }
2291
2292         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2293                 err = jbd2_journal_update_format(journal);
2294                 if (err)  {
2295                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2296                         jbd2_journal_destroy(journal);
2297                         return err;
2298                 }
2299         }
2300
2301         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2302                 err = jbd2_journal_wipe(journal, !really_read_only);
2303         if (!err)
2304                 err = jbd2_journal_load(journal);
2305
2306         if (err) {
2307                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2308                 jbd2_journal_destroy(journal);
2309                 return err;
2310         }
2311
2312         EXT4_SB(sb)->s_journal = journal;
2313         ext4_clear_journal_err(sb, es);
2314
2315         if (journal_devnum &&
2316             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2317                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2318                 sb->s_dirt = 1;
2319
2320                 /* Make sure we flush the recovery flag to disk. */
2321                 ext4_commit_super(sb, es, 1);
2322         }
2323
2324         return 0;
2325 }
2326
2327 static int ext4_create_journal(struct super_block * sb,
2328                                struct ext4_super_block * es,
2329                                unsigned int journal_inum)
2330 {
2331         journal_t *journal;
2332         int err;
2333
2334         if (sb->s_flags & MS_RDONLY) {
2335                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2336                                 "create journal.\n");
2337                 return -EROFS;
2338         }
2339
2340         journal = ext4_get_journal(sb, journal_inum);
2341         if (!journal)
2342                 return -EINVAL;
2343
2344         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2345                journal_inum);
2346
2347         err = jbd2_journal_create(journal);
2348         if (err) {
2349                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2350                 jbd2_journal_destroy(journal);
2351                 return -EIO;
2352         }
2353
2354         EXT4_SB(sb)->s_journal = journal;
2355
2356         ext4_update_dynamic_rev(sb);
2357         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2358         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2359
2360         es->s_journal_inum = cpu_to_le32(journal_inum);
2361         sb->s_dirt = 1;
2362
2363         /* Make sure we flush the recovery flag to disk. */
2364         ext4_commit_super(sb, es, 1);
2365
2366         return 0;
2367 }
2368
2369 static void ext4_commit_super (struct super_block * sb,
2370                                struct ext4_super_block * es,
2371                                int sync)
2372 {
2373         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2374
2375         if (!sbh)
2376                 return;
2377         es->s_wtime = cpu_to_le32(get_seconds());
2378         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2379         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2380         BUFFER_TRACE(sbh, "marking dirty");
2381         mark_buffer_dirty(sbh);
2382         if (sync)
2383                 sync_dirty_buffer(sbh);
2384 }
2385
2386
2387 /*
2388  * Have we just finished recovery?  If so, and if we are mounting (or
2389  * remounting) the filesystem readonly, then we will end up with a
2390  * consistent fs on disk.  Record that fact.
2391  */
2392 static void ext4_mark_recovery_complete(struct super_block * sb,
2393                                         struct ext4_super_block * es)
2394 {
2395         journal_t *journal = EXT4_SB(sb)->s_journal;
2396
2397         jbd2_journal_lock_updates(journal);
2398         jbd2_journal_flush(journal);
2399         lock_super(sb);
2400         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2401             sb->s_flags & MS_RDONLY) {
2402                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2403                 sb->s_dirt = 0;
2404                 ext4_commit_super(sb, es, 1);
2405         }
2406         unlock_super(sb);
2407         jbd2_journal_unlock_updates(journal);
2408 }
2409
2410 /*
2411  * If we are mounting (or read-write remounting) a filesystem whose journal
2412  * has recorded an error from a previous lifetime, move that error to the
2413  * main filesystem now.
2414  */
2415 static void ext4_clear_journal_err(struct super_block * sb,
2416                                    struct ext4_super_block * es)
2417 {
2418         journal_t *journal;
2419         int j_errno;
2420         const char *errstr;
2421
2422         journal = EXT4_SB(sb)->s_journal;
2423
2424         /*
2425          * Now check for any error status which may have been recorded in the
2426          * journal by a prior ext4_error() or ext4_abort()
2427          */
2428
2429         j_errno = jbd2_journal_errno(journal);
2430         if (j_errno) {
2431                 char nbuf[16];
2432
2433                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2434                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2435                              "from previous mount: %s", errstr);
2436                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2437                              "filesystem check.");
2438
2439                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2440                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2441                 ext4_commit_super (sb, es, 1);
2442
2443                 jbd2_journal_clear_err(journal);
2444         }
2445 }
2446
2447 /*
2448  * Force the running and committing transactions to commit,
2449  * and wait on the commit.
2450  */
2451 int ext4_force_commit(struct super_block *sb)
2452 {
2453         journal_t *journal;
2454         int ret;
2455
2456         if (sb->s_flags & MS_RDONLY)
2457                 return 0;
2458
2459         journal = EXT4_SB(sb)->s_journal;
2460         sb->s_dirt = 0;
2461         ret = ext4_journal_force_commit(journal);
2462         return ret;
2463 }
2464
2465 /*
2466  * Ext4 always journals updates to the superblock itself, so we don't
2467  * have to propagate any other updates to the superblock on disk at this
2468  * point.  Just start an async writeback to get the buffers on their way
2469  * to the disk.
2470  *
2471  * This implicitly triggers the writebehind on sync().
2472  */
2473
2474 static void ext4_write_super (struct super_block * sb)
2475 {
2476         if (mutex_trylock(&sb->s_lock) != 0)
2477                 BUG();
2478         sb->s_dirt = 0;
2479 }
2480
2481 static int ext4_sync_fs(struct super_block *sb, int wait)
2482 {
2483         tid_t target;
2484
2485         sb->s_dirt = 0;
2486         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2487                 if (wait)
2488                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2489         }
2490         return 0;
2491 }
2492
2493 /*
2494  * LVM calls this function before a (read-only) snapshot is created.  This
2495  * gives us a chance to flush the journal completely and mark the fs clean.
2496  */
2497 static void ext4_write_super_lockfs(struct super_block *sb)
2498 {
2499         sb->s_dirt = 0;
2500
2501         if (!(sb->s_flags & MS_RDONLY)) {
2502                 journal_t *journal = EXT4_SB(sb)->s_journal;
2503
2504                 /* Now we set up the journal barrier. */
2505                 jbd2_journal_lock_updates(journal);
2506                 jbd2_journal_flush(journal);
2507
2508                 /* Journal blocked and flushed, clear needs_recovery flag. */
2509                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2510                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2511         }
2512 }
2513
2514 /*
2515  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2516  * flag here, even though the filesystem is not technically dirty yet.
2517  */
2518 static void ext4_unlockfs(struct super_block *sb)
2519 {
2520         if (!(sb->s_flags & MS_RDONLY)) {
2521                 lock_super(sb);
2522                 /* Reser the needs_recovery flag before the fs is unlocked. */
2523                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2524                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2525                 unlock_super(sb);
2526                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2527         }
2528 }
2529
2530 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2531 {
2532         struct ext4_super_block * es;
2533         struct ext4_sb_info *sbi = EXT4_SB(sb);
2534         ext4_fsblk_t n_blocks_count = 0;
2535         unsigned long old_sb_flags;
2536         struct ext4_mount_options old_opts;
2537         int err;
2538 #ifdef CONFIG_QUOTA
2539         int i;
2540 #endif
2541
2542         /* Store the original options */
2543         old_sb_flags = sb->s_flags;
2544         old_opts.s_mount_opt = sbi->s_mount_opt;
2545         old_opts.s_resuid = sbi->s_resuid;
2546         old_opts.s_resgid = sbi->s_resgid;
2547         old_opts.s_commit_interval = sbi->s_commit_interval;
2548 #ifdef CONFIG_QUOTA
2549         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2550         for (i = 0; i < MAXQUOTAS; i++)
2551                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2552 #endif
2553
2554         /*
2555          * Allow the "check" option to be passed as a remount option.
2556          */
2557         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2558                 err = -EINVAL;
2559                 goto restore_opts;
2560         }
2561
2562         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2563                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2564
2565         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2566                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2567
2568         es = sbi->s_es;
2569
2570         ext4_init_journal_params(sb, sbi->s_journal);
2571
2572         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2573                 n_blocks_count > ext4_blocks_count(es)) {
2574                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2575                         err = -EROFS;
2576                         goto restore_opts;
2577                 }
2578
2579                 if (*flags & MS_RDONLY) {
2580                         /*
2581                          * First of all, the unconditional stuff we have to do
2582                          * to disable replay of the journal when we next remount
2583                          */
2584                         sb->s_flags |= MS_RDONLY;
2585
2586                         /*
2587                          * OK, test if we are remounting a valid rw partition
2588                          * readonly, and if so set the rdonly flag and then
2589                          * mark the partition as valid again.
2590                          */
2591                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2592                             (sbi->s_mount_state & EXT4_VALID_FS))
2593                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2594
2595                         /*
2596                          * We have to unlock super so that we can wait for
2597                          * transactions.
2598                          */
2599                         unlock_super(sb);
2600                         ext4_mark_recovery_complete(sb, es);
2601                         lock_super(sb);
2602                 } else {
2603                         __le32 ret;
2604                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2605                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2606                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2607                                        "remount RDWR because of unsupported "
2608                                        "optional features (%x).\n",
2609                                        sb->s_id, le32_to_cpu(ret));
2610                                 err = -EROFS;
2611                                 goto restore_opts;
2612                         }
2613
2614                         /*
2615                          * If we have an unprocessed orphan list hanging
2616                          * around from a previously readonly bdev mount,
2617                          * require a full umount/remount for now.
2618                          */
2619                         if (es->s_last_orphan) {
2620                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2621                                        "remount RDWR because of unprocessed "
2622                                        "orphan inode list.  Please "
2623                                        "umount/remount instead.\n",
2624                                        sb->s_id);
2625                                 err = -EINVAL;
2626                                 goto restore_opts;
2627                         }
2628
2629                         /*
2630                          * Mounting a RDONLY partition read-write, so reread
2631                          * and store the current valid flag.  (It may have
2632                          * been changed by e2fsck since we originally mounted
2633                          * the partition.)
2634                          */
2635                         ext4_clear_journal_err(sb, es);
2636                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2637                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2638                                 goto restore_opts;
2639                         if (!ext4_setup_super (sb, es, 0))
2640                                 sb->s_flags &= ~MS_RDONLY;
2641                 }
2642         }
2643 #ifdef CONFIG_QUOTA
2644         /* Release old quota file names */
2645         for (i = 0; i < MAXQUOTAS; i++)
2646                 if (old_opts.s_qf_names[i] &&
2647                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2648                         kfree(old_opts.s_qf_names[i]);
2649 #endif
2650         return 0;
2651 restore_opts:
2652         sb->s_flags = old_sb_flags;
2653         sbi->s_mount_opt = old_opts.s_mount_opt;
2654         sbi->s_resuid = old_opts.s_resuid;
2655         sbi->s_resgid = old_opts.s_resgid;
2656         sbi->s_commit_interval = old_opts.s_commit_interval;
2657 #ifdef CONFIG_QUOTA
2658         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2659         for (i = 0; i < MAXQUOTAS; i++) {
2660                 if (sbi->s_qf_names[i] &&
2661                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2662                         kfree(sbi->s_qf_names[i]);
2663                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2664         }
2665 #endif
2666         return err;
2667 }
2668
2669 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2670 {
2671         struct super_block *sb = dentry->d_sb;
2672         struct ext4_sb_info *sbi = EXT4_SB(sb);
2673         struct ext4_super_block *es = sbi->s_es;
2674         u64 fsid;
2675
2676         if (test_opt(sb, MINIX_DF)) {
2677                 sbi->s_overhead_last = 0;
2678         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2679                 ext4_group_t ngroups = sbi->s_groups_count, i;
2680                 ext4_fsblk_t overhead = 0;
2681                 smp_rmb();
2682
2683                 /*
2684                  * Compute the overhead (FS structures).  This is constant
2685                  * for a given filesystem unless the number of block groups
2686                  * changes so we cache the previous value until it does.
2687                  */
2688
2689                 /*
2690                  * All of the blocks before first_data_block are
2691                  * overhead
2692                  */
2693                 overhead = le32_to_cpu(es->s_first_data_block);
2694
2695                 /*
2696                  * Add the overhead attributed to the superblock and
2697                  * block group descriptors.  If the sparse superblocks
2698                  * feature is turned on, then not all groups have this.
2699                  */
2700                 for (i = 0; i < ngroups; i++) {
2701                         overhead += ext4_bg_has_super(sb, i) +
2702                                 ext4_bg_num_gdb(sb, i);
2703                         cond_resched();
2704                 }
2705
2706                 /*
2707                  * Every block group has an inode bitmap, a block
2708                  * bitmap, and an inode table.
2709                  */
2710                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2711                 sbi->s_overhead_last = overhead;
2712                 smp_wmb();
2713                 sbi->s_blocks_last = ext4_blocks_count(es);
2714         }
2715
2716         buf->f_type = EXT4_SUPER_MAGIC;
2717         buf->f_bsize = sb->s_blocksize;
2718         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2719         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2720         ext4_free_blocks_count_set(es, buf->f_bfree);
2721         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2722         if (buf->f_bfree < ext4_r_blocks_count(es))
2723                 buf->f_bavail = 0;
2724         buf->f_files = le32_to_cpu(es->s_inodes_count);
2725         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2726         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2727         buf->f_namelen = EXT4_NAME_LEN;
2728         fsid = le64_to_cpup((void *)es->s_uuid) ^
2729                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2730         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2731         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2732         return 0;
2733 }
2734
2735 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2736  * is locked for write. Otherwise the are possible deadlocks:
2737  * Process 1                         Process 2
2738  * ext4_create()                     quota_sync()
2739  *   jbd2_journal_start()                   write_dquot()
2740  *   DQUOT_INIT()                        down(dqio_mutex)
2741  *     down(dqio_mutex)                    jbd2_journal_start()
2742  *
2743  */
2744
2745 #ifdef CONFIG_QUOTA
2746
2747 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2748 {
2749         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2750 }
2751
2752 static int ext4_dquot_initialize(struct inode *inode, int type)
2753 {
2754         handle_t *handle;
2755         int ret, err;
2756
2757         /* We may create quota structure so we need to reserve enough blocks */
2758         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2759         if (IS_ERR(handle))
2760                 return PTR_ERR(handle);
2761         ret = dquot_initialize(inode, type);
2762         err = ext4_journal_stop(handle);
2763         if (!ret)
2764                 ret = err;
2765         return ret;
2766 }
2767
2768 static int ext4_dquot_drop(struct inode *inode)
2769 {
2770         handle_t *handle;
2771         int ret, err;
2772
2773         /* We may delete quota structure so we need to reserve enough blocks */
2774         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2775         if (IS_ERR(handle))
2776                 return PTR_ERR(handle);
2777         ret = dquot_drop(inode);
2778         err = ext4_journal_stop(handle);
2779         if (!ret)
2780                 ret = err;
2781         return ret;
2782 }
2783
2784 static int ext4_write_dquot(struct dquot *dquot)
2785 {
2786         int ret, err;
2787         handle_t *handle;
2788         struct inode *inode;
2789
2790         inode = dquot_to_inode(dquot);
2791         handle = ext4_journal_start(inode,
2792                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2793         if (IS_ERR(handle))
2794                 return PTR_ERR(handle);
2795         ret = dquot_commit(dquot);
2796         err = ext4_journal_stop(handle);
2797         if (!ret)
2798                 ret = err;
2799         return ret;
2800 }
2801
2802 static int ext4_acquire_dquot(struct dquot *dquot)
2803 {
2804         int ret, err;
2805         handle_t *handle;
2806
2807         handle = ext4_journal_start(dquot_to_inode(dquot),
2808                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2809         if (IS_ERR(handle))
2810                 return PTR_ERR(handle);
2811         ret = dquot_acquire(dquot);
2812         err = ext4_journal_stop(handle);
2813         if (!ret)
2814                 ret = err;
2815         return ret;
2816 }
2817
2818 static int ext4_release_dquot(struct dquot *dquot)
2819 {
2820         int ret, err;
2821         handle_t *handle;
2822
2823         handle = ext4_journal_start(dquot_to_inode(dquot),
2824                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2825         if (IS_ERR(handle)) {
2826                 /* Release dquot anyway to avoid endless cycle in dqput() */
2827                 dquot_release(dquot);
2828                 return PTR_ERR(handle);
2829         }
2830         ret = dquot_release(dquot);
2831         err = ext4_journal_stop(handle);
2832         if (!ret)
2833                 ret = err;
2834         return ret;
2835 }
2836
2837 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2838 {
2839         /* Are we journalling quotas? */
2840         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2841             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2842                 dquot_mark_dquot_dirty(dquot);
2843                 return ext4_write_dquot(dquot);
2844         } else {
2845                 return dquot_mark_dquot_dirty(dquot);
2846         }
2847 }
2848
2849 static int ext4_write_info(struct super_block *sb, int type)
2850 {
2851         int ret, err;
2852         handle_t *handle;
2853
2854         /* Data block + inode block */
2855         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2856         if (IS_ERR(handle))
2857                 return PTR_ERR(handle);
2858         ret = dquot_commit_info(sb, type);
2859         err = ext4_journal_stop(handle);
2860         if (!ret)
2861                 ret = err;
2862         return ret;
2863 }
2864
2865 /*
2866  * Turn on quotas during mount time - we need to find
2867  * the quota file and such...
2868  */
2869 static int ext4_quota_on_mount(struct super_block *sb, int type)
2870 {
2871         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2872                         EXT4_SB(sb)->s_jquota_fmt, type);
2873 }
2874
2875 /*
2876  * Standard function to be called on quota_on
2877  */
2878 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2879                          char *path)
2880 {
2881         int err;
2882         struct nameidata nd;
2883
2884         if (!test_opt(sb, QUOTA))
2885                 return -EINVAL;
2886         /* Not journalling quota? */
2887         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2888             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2889                 return vfs_quota_on(sb, type, format_id, path);
2890         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2891         if (err)
2892                 return err;
2893         /* Quotafile not on the same filesystem? */
2894         if (nd.mnt->mnt_sb != sb) {
2895                 path_release(&nd);
2896                 return -EXDEV;
2897         }
2898         /* Quotafile not of fs root? */
2899         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2900                 printk(KERN_WARNING
2901                         "EXT4-fs: Quota file not on filesystem root. "
2902                         "Journalled quota will not work.\n");
2903         path_release(&nd);
2904         return vfs_quota_on(sb, type, format_id, path);
2905 }
2906
2907 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2908  * acquiring the locks... As quota files are never truncated and quota code
2909  * itself serializes the operations (and noone else should touch the files)
2910  * we don't have to be afraid of races */
2911 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2912                                size_t len, loff_t off)
2913 {
2914         struct inode *inode = sb_dqopt(sb)->files[type];
2915         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2916         int err = 0;
2917         int offset = off & (sb->s_blocksize - 1);
2918         int tocopy;
2919         size_t toread;
2920         struct buffer_head *bh;
2921         loff_t i_size = i_size_read(inode);
2922
2923         if (off > i_size)
2924                 return 0;
2925         if (off+len > i_size)
2926                 len = i_size-off;
2927         toread = len;
2928         while (toread > 0) {
2929                 tocopy = sb->s_blocksize - offset < toread ?
2930                                 sb->s_blocksize - offset : toread;
2931                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2932                 if (err)
2933                         return err;
2934                 if (!bh)        /* A hole? */
2935                         memset(data, 0, tocopy);
2936                 else
2937                         memcpy(data, bh->b_data+offset, tocopy);
2938                 brelse(bh);
2939                 offset = 0;
2940                 toread -= tocopy;
2941                 data += tocopy;
2942                 blk++;
2943         }
2944         return len;
2945 }
2946
2947 /* Write to quotafile (we know the transaction is already started and has
2948  * enough credits) */
2949 static ssize_t ext4_quota_write(struct super_block *sb, int type,
2950                                 const char *data, size_t len, loff_t off)
2951 {
2952         struct inode *inode = sb_dqopt(sb)->files[type];
2953         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2954         int err = 0;
2955         int offset = off & (sb->s_blocksize - 1);
2956         int tocopy;
2957         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
2958         size_t towrite = len;
2959         struct buffer_head *bh;
2960         handle_t *handle = journal_current_handle();
2961
2962         if (!handle) {
2963                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
2964                         " cancelled because transaction is not started.\n",
2965                         (unsigned long long)off, (unsigned long long)len);
2966                 return -EIO;
2967         }
2968         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2969         while (towrite > 0) {
2970                 tocopy = sb->s_blocksize - offset < towrite ?
2971                                 sb->s_blocksize - offset : towrite;
2972                 bh = ext4_bread(handle, inode, blk, 1, &err);
2973                 if (!bh)
2974                         goto out;
2975                 if (journal_quota) {
2976                         err = ext4_journal_get_write_access(handle, bh);
2977                         if (err) {
2978                                 brelse(bh);
2979                                 goto out;
2980                         }
2981                 }
2982                 lock_buffer(bh);
2983                 memcpy(bh->b_data+offset, data, tocopy);
2984                 flush_dcache_page(bh->b_page);
2985                 unlock_buffer(bh);
2986                 if (journal_quota)
2987                         err = ext4_journal_dirty_metadata(handle, bh);
2988                 else {
2989                         /* Always do at least ordered writes for quotas */
2990                         err = ext4_journal_dirty_data(handle, bh);
2991                         mark_buffer_dirty(bh);
2992                 }
2993                 brelse(bh);
2994                 if (err)
2995                         goto out;
2996                 offset = 0;
2997                 towrite -= tocopy;
2998                 data += tocopy;
2999                 blk++;
3000         }
3001 out:
3002         if (len == towrite)
3003                 return err;
3004         if (inode->i_size < off+len-towrite) {
3005                 i_size_write(inode, off+len-towrite);
3006                 EXT4_I(inode)->i_disksize = inode->i_size;
3007         }
3008         inode->i_version++;
3009         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3010         ext4_mark_inode_dirty(handle, inode);
3011         mutex_unlock(&inode->i_mutex);
3012         return len - towrite;
3013 }
3014
3015 #endif
3016
3017 static int ext4_get_sb(struct file_system_type *fs_type,
3018         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3019 {
3020         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3021 }
3022
3023 static struct file_system_type ext4dev_fs_type = {
3024         .owner          = THIS_MODULE,
3025         .name           = "ext4dev",
3026         .get_sb         = ext4_get_sb,
3027         .kill_sb        = kill_block_super,
3028         .fs_flags       = FS_REQUIRES_DEV,
3029 };
3030
3031 static int __init init_ext4_fs(void)
3032 {
3033         int err = init_ext4_xattr();
3034         if (err)
3035                 return err;
3036         err = init_inodecache();
3037         if (err)
3038                 goto out1;
3039         err = register_filesystem(&ext4dev_fs_type);
3040         if (err)
3041                 goto out;
3042         return 0;
3043 out:
3044         destroy_inodecache();
3045 out1:
3046         exit_ext4_xattr();
3047         return err;
3048 }
3049
3050 static void __exit exit_ext4_fs(void)
3051 {
3052         unregister_filesystem(&ext4dev_fs_type);
3053         destroy_inodecache();
3054         exit_ext4_xattr();
3055 }
3056
3057 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3058 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3059 MODULE_LICENSE("GPL");
3060 module_init(init_ext4_fs)
3061 module_exit(exit_ext4_fs)