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/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <asm/uaccess.h>
  41
  42 #include "ext4.h"
  43 #include "ext4_jbd2.h"
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 struct proc_dir_entry *ext4_proc_root;
  50
  51 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  52                              unsigned long journal_devnum);
  53 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  54                                unsigned int);
  55 static void ext4_commit_super(struct super_block *sb,
  56                               struct ext4_super_block *es, int sync);
  57 static void ext4_mark_recovery_complete(struct super_block *sb,
  58                                         struct ext4_super_block *es);
  59 static void ext4_clear_journal_err(struct super_block *sb,
  60                                    struct ext4_super_block *es);
  61 static int ext4_sync_fs(struct super_block *sb, int wait);
  62 static const char *ext4_decode_error(struct super_block *sb, int errno,
  63                                      char nbuf[16]);
  64 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  65 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  66 static void ext4_unlockfs(struct super_block *sb);
  67 static void ext4_write_super(struct super_block *sb);
  68 static void ext4_write_super_lockfs(struct super_block *sb);
  69
  70
  71 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  72                                struct ext4_group_desc *bg)
  73 {
  74         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  75                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  76                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  77 }
  78
  79 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  80                                struct ext4_group_desc *bg)
  81 {
  82         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  83                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  84                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  85 }
  86
  87 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  88                               struct ext4_group_desc *bg)
  89 {
  90         return le32_to_cpu(bg->bg_inode_table_lo) |
  91                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  92                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  93 }
  94
  95 void ext4_block_bitmap_set(struct super_block *sb,
  96                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  97 {
  98         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  99         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 100                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 101 }
 102
 103 void ext4_inode_bitmap_set(struct super_block *sb,
 104                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 105 {
 106         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 107         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 108                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 109 }
 110
 111 void ext4_inode_table_set(struct super_block *sb,
 112                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 113 {
 114         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 115         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 116                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 117 }
 118
 119 /*
 120  * Wrappers for jbd2_journal_start/end.
 121  *
 122  * The only special thing we need to do here is to make sure that all
 123  * journal_end calls result in the superblock being marked dirty, so
 124  * that sync() will call the filesystem's write_super callback if
 125  * appropriate.
 126  */
 127 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 128 {
 129         journal_t *journal;
 130
 131         if (sb->s_flags & MS_RDONLY)
 132                 return ERR_PTR(-EROFS);
 133
 134         /* Special case here: if the journal has aborted behind our
 135          * backs (eg. EIO in the commit thread), then we still need to
 136          * take the FS itself readonly cleanly. */
 137         journal = EXT4_SB(sb)->s_journal;
 138         if (is_journal_aborted(journal)) {
 139                 ext4_abort(sb, __func__,
 140                            "Detected aborted journal");
 141                 return ERR_PTR(-EROFS);
 142         }
 143
 144         return jbd2_journal_start(journal, nblocks);
 145 }
 146
 147 /*
 148  * The only special thing we need to do here is to make sure that all
 149  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 150  * that sync() will call the filesystem's write_super callback if
 151  * appropriate.
 152  */
 153 int __ext4_journal_stop(const char *where, handle_t *handle)
 154 {
 155         struct super_block *sb;
 156         int err;
 157         int rc;
 158
 159         sb = handle->h_transaction->t_journal->j_private;
 160         err = handle->h_err;
 161         rc = jbd2_journal_stop(handle);
 162
 163         if (!err)
 164                 err = rc;
 165         if (err)
 166                 __ext4_std_error(sb, where, err);
 167         return err;
 168 }
 169
 170 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 171                 struct buffer_head *bh, handle_t *handle, int err)
 172 {
 173         char nbuf[16];
 174         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 175
 176         if (bh)
 177                 BUFFER_TRACE(bh, "abort");
 178
 179         if (!handle->h_err)
 180                 handle->h_err = err;
 181
 182         if (is_handle_aborted(handle))
 183                 return;
 184
 185         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 186                caller, errstr, err_fn);
 187
 188         jbd2_journal_abort_handle(handle);
 189 }
 190
 191 /* Deal with the reporting of failure conditions on a filesystem such as
 192  * inconsistencies detected or read IO failures.
 193  *
 194  * On ext2, we can store the error state of the filesystem in the
 195  * superblock.  That is not possible on ext4, because we may have other
 196  * write ordering constraints on the superblock which prevent us from
 197  * writing it out straight away; and given that the journal is about to
 198  * be aborted, we can't rely on the current, or future, transactions to
 199  * write out the superblock safely.
 200  *
 201  * We'll just use the jbd2_journal_abort() error code to record an error in
 202  * the journal instead.  On recovery, the journal will compain about
 203  * that error until we've noted it down and cleared it.
 204  */
 205
 206 static void ext4_handle_error(struct super_block *sb)
 207 {
 208         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 209
 210         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 211         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 212
 213         if (sb->s_flags & MS_RDONLY)
 214                 return;
 215
 216         if (!test_opt(sb, ERRORS_CONT)) {
 217                 journal_t *journal = EXT4_SB(sb)->s_journal;
 218
 219                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 220                 if (journal)
 221                         jbd2_journal_abort(journal, -EIO);
 222         }
 223         if (test_opt(sb, ERRORS_RO)) {
 224                 printk(KERN_CRIT "Remounting filesystem read-only\n");
 225                 sb->s_flags |= MS_RDONLY;
 226         }
 227         ext4_commit_super(sb, es, 1);
 228         if (test_opt(sb, ERRORS_PANIC))
 229                 panic("EXT4-fs (device %s): panic forced after error\n",
 230                         sb->s_id);
 231 }
 232
 233 void ext4_error(struct super_block *sb, const char *function,
 234                 const char *fmt, ...)
 235 {
 236         va_list args;
 237
 238         va_start(args, fmt);
 239         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 240         vprintk(fmt, args);
 241         printk("\n");
 242         va_end(args);
 243
 244         ext4_handle_error(sb);
 245 }
 246
 247 static const char *ext4_decode_error(struct super_block *sb, int errno,
 248                                      char nbuf[16])
 249 {
 250         char *errstr = NULL;
 251
 252         switch (errno) {
 253         case -EIO:
 254                 errstr = "IO failure";
 255                 break;
 256         case -ENOMEM:
 257                 errstr = "Out of memory";
 258                 break;
 259         case -EROFS:
 260                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 261                         errstr = "Journal has aborted";
 262                 else
 263                         errstr = "Readonly filesystem";
 264                 break;
 265         default:
 266                 /* If the caller passed in an extra buffer for unknown
 267                  * errors, textualise them now.  Else we just return
 268                  * NULL. */
 269                 if (nbuf) {
 270                         /* Check for truncated error codes... */
 271                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 272                                 errstr = nbuf;
 273                 }
 274                 break;
 275         }
 276
 277         return errstr;
 278 }
 279
 280 /* __ext4_std_error decodes expected errors from journaling functions
 281  * automatically and invokes the appropriate error response.  */
 282
 283 void __ext4_std_error(struct super_block *sb, const char *function, 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, __func__,
 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 int ext4_update_compat_feature(handle_t *handle,
 377                                         struct super_block *sb, __u32 compat)
 378 {
 379         int err = 0;
 380         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 381                 err = ext4_journal_get_write_access(handle,
 382                                 EXT4_SB(sb)->s_sbh);
 383                 if (err)
 384                         return err;
 385                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 386                 sb->s_dirt = 1;
 387                 handle->h_sync = 1;
 388                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 389                                         "call ext4_journal_dirty_met adata");
 390                 err = ext4_journal_dirty_metadata(handle,
 391                                 EXT4_SB(sb)->s_sbh);
 392         }
 393         return err;
 394 }
 395
 396 int ext4_update_rocompat_feature(handle_t *handle,
 397                                         struct super_block *sb, __u32 rocompat)
 398 {
 399         int err = 0;
 400         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 401                 err = ext4_journal_get_write_access(handle,
 402                                 EXT4_SB(sb)->s_sbh);
 403                 if (err)
 404                         return err;
 405                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 406                 sb->s_dirt = 1;
 407                 handle->h_sync = 1;
 408                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 409                                         "call ext4_journal_dirty_met adata");
 410                 err = ext4_journal_dirty_metadata(handle,
 411                                 EXT4_SB(sb)->s_sbh);
 412         }
 413         return err;
 414 }
 415
 416 int ext4_update_incompat_feature(handle_t *handle,
 417                                         struct super_block *sb, __u32 incompat)
 418 {
 419         int err = 0;
 420         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 421                 err = ext4_journal_get_write_access(handle,
 422                                 EXT4_SB(sb)->s_sbh);
 423                 if (err)
 424                         return err;
 425                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 426                 sb->s_dirt = 1;
 427                 handle->h_sync = 1;
 428                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 429                                         "call ext4_journal_dirty_met adata");
 430                 err = ext4_journal_dirty_metadata(handle,
 431                                 EXT4_SB(sb)->s_sbh);
 432         }
 433         return err;
 434 }
 435
 436 /*
 437  * Open the external journal device
 438  */
 439 static struct block_device *ext4_blkdev_get(dev_t dev)
 440 {
 441         struct block_device *bdev;
 442         char b[BDEVNAME_SIZE];
 443
 444         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 445         if (IS_ERR(bdev))
 446                 goto fail;
 447         return bdev;
 448
 449 fail:
 450         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 451                         __bdevname(dev, b), PTR_ERR(bdev));
 452         return NULL;
 453 }
 454
 455 /*
 456  * Release the journal device
 457  */
 458 static int ext4_blkdev_put(struct block_device *bdev)
 459 {
 460         bd_release(bdev);
 461         return blkdev_put(bdev);
 462 }
 463
 464 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 465 {
 466         struct block_device *bdev;
 467         int ret = -ENODEV;
 468
 469         bdev = sbi->journal_bdev;
 470         if (bdev) {
 471                 ret = ext4_blkdev_put(bdev);
 472                 sbi->journal_bdev = NULL;
 473         }
 474         return ret;
 475 }
 476
 477 static inline struct inode *orphan_list_entry(struct list_head *l)
 478 {
 479         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 480 }
 481
 482 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 483 {
 484         struct list_head *l;
 485
 486         printk(KERN_ERR "sb orphan head is %d\n",
 487                le32_to_cpu(sbi->s_es->s_last_orphan));
 488
 489         printk(KERN_ERR "sb_info orphan list:\n");
 490         list_for_each(l, &sbi->s_orphan) {
 491                 struct inode *inode = orphan_list_entry(l);
 492                 printk(KERN_ERR "  "
 493                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 494                        inode->i_sb->s_id, inode->i_ino, inode,
 495                        inode->i_mode, inode->i_nlink,
 496                        NEXT_ORPHAN(inode));
 497         }
 498 }
 499
 500 static void ext4_put_super(struct super_block *sb)
 501 {
 502         struct ext4_sb_info *sbi = EXT4_SB(sb);
 503         struct ext4_super_block *es = sbi->s_es;
 504         int i;
 505
 506         ext4_mb_release(sb);
 507         ext4_ext_release(sb);
 508         ext4_xattr_put_super(sb);
 509         jbd2_journal_destroy(sbi->s_journal);
 510         sbi->s_journal = NULL;
 511         if (!(sb->s_flags & MS_RDONLY)) {
 512                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 513                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 514                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 515                 mark_buffer_dirty(sbi->s_sbh);
 516                 ext4_commit_super(sb, es, 1);
 517         }
 518         if (sbi->s_proc) {
 519                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
 520                 remove_proc_entry(sb->s_id, ext4_proc_root);
 521         }
 522
 523         for (i = 0; i < sbi->s_gdb_count; i++)
 524                 brelse(sbi->s_group_desc[i]);
 525         kfree(sbi->s_group_desc);
 526         kfree(sbi->s_flex_groups);
 527         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 528         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 529         percpu_counter_destroy(&sbi->s_dirs_counter);
 530         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 531         brelse(sbi->s_sbh);
 532 #ifdef CONFIG_QUOTA
 533         for (i = 0; i < MAXQUOTAS; i++)
 534                 kfree(sbi->s_qf_names[i]);
 535 #endif
 536
 537         /* Debugging code just in case the in-memory inode orphan list
 538          * isn't empty.  The on-disk one can be non-empty if we've
 539          * detected an error and taken the fs readonly, but the
 540          * in-memory list had better be clean by this point. */
 541         if (!list_empty(&sbi->s_orphan))
 542                 dump_orphan_list(sb, sbi);
 543         J_ASSERT(list_empty(&sbi->s_orphan));
 544
 545         invalidate_bdev(sb->s_bdev);
 546         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 547                 /*
 548                  * Invalidate the journal device's buffers.  We don't want them
 549                  * floating about in memory - the physical journal device may
 550                  * hotswapped, and it breaks the `ro-after' testing code.
 551                  */
 552                 sync_blockdev(sbi->journal_bdev);
 553                 invalidate_bdev(sbi->journal_bdev);
 554                 ext4_blkdev_remove(sbi);
 555         }
 556         sb->s_fs_info = NULL;
 557         kfree(sbi);
 558         return;
 559 }
 560
 561 static struct kmem_cache *ext4_inode_cachep;
 562
 563 /*
 564  * Called inside transaction, so use GFP_NOFS
 565  */
 566 static struct inode *ext4_alloc_inode(struct super_block *sb)
 567 {
 568         struct ext4_inode_info *ei;
 569
 570         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 571         if (!ei)
 572                 return NULL;
 573 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 574         ei->i_acl = EXT4_ACL_NOT_CACHED;
 575         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 576 #endif
 577         ei->i_block_alloc_info = NULL;
 578         ei->vfs_inode.i_version = 1;
 579         ei->vfs_inode.i_data.writeback_index = 0;
 580         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 581         INIT_LIST_HEAD(&ei->i_prealloc_list);
 582         spin_lock_init(&ei->i_prealloc_lock);
 583         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 584         ei->i_reserved_data_blocks = 0;
 585         ei->i_reserved_meta_blocks = 0;
 586         ei->i_allocated_meta_blocks = 0;
 587         ei->i_delalloc_reserved_flag = 0;
 588         spin_lock_init(&(ei->i_block_reservation_lock));
 589         return &ei->vfs_inode;
 590 }
 591
 592 static void ext4_destroy_inode(struct inode *inode)
 593 {
 594         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 595                 printk("EXT4 Inode %p: orphan list check failed!\n",
 596                         EXT4_I(inode));
 597                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 598                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 599                                 true);
 600                 dump_stack();
 601         }
 602         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 603 }
 604
 605 static void init_once(void *foo)
 606 {
 607         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 608
 609         INIT_LIST_HEAD(&ei->i_orphan);
 610 #ifdef CONFIG_EXT4DEV_FS_XATTR
 611         init_rwsem(&ei->xattr_sem);
 612 #endif
 613         init_rwsem(&ei->i_data_sem);
 614         inode_init_once(&ei->vfs_inode);
 615 }
 616
 617 static int init_inodecache(void)
 618 {
 619         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 620                                              sizeof(struct ext4_inode_info),
 621                                              0, (SLAB_RECLAIM_ACCOUNT|
 622                                                 SLAB_MEM_SPREAD),
 623                                              init_once);
 624         if (ext4_inode_cachep == NULL)
 625                 return -ENOMEM;
 626         return 0;
 627 }
 628
 629 static void destroy_inodecache(void)
 630 {
 631         kmem_cache_destroy(ext4_inode_cachep);
 632 }
 633
 634 static void ext4_clear_inode(struct inode *inode)
 635 {
 636         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 637 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 638         if (EXT4_I(inode)->i_acl &&
 639                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 640                 posix_acl_release(EXT4_I(inode)->i_acl);
 641                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 642         }
 643         if (EXT4_I(inode)->i_default_acl &&
 644                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 645                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 646                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 647         }
 648 #endif
 649         ext4_discard_reservation(inode);
 650         EXT4_I(inode)->i_block_alloc_info = NULL;
 651         if (unlikely(rsv))
 652                 kfree(rsv);
 653         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 654                                        &EXT4_I(inode)->jinode);
 655 }
 656
 657 static inline void ext4_show_quota_options(struct seq_file *seq,
 658                                            struct super_block *sb)
 659 {
 660 #if defined(CONFIG_QUOTA)
 661         struct ext4_sb_info *sbi = EXT4_SB(sb);
 662
 663         if (sbi->s_jquota_fmt)
 664                 seq_printf(seq, ",jqfmt=%s",
 665                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold" : "vfsv0");
 666
 667         if (sbi->s_qf_names[USRQUOTA])
 668                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 669
 670         if (sbi->s_qf_names[GRPQUOTA])
 671                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 672
 673         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 674                 seq_puts(seq, ",usrquota");
 675
 676         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 677                 seq_puts(seq, ",grpquota");
 678 #endif
 679 }
 680
 681 /*
 682  * Show an option if
 683  *  - it's set to a non-default value OR
 684  *  - if the per-sb default is different from the global default
 685  */
 686 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 687 {
 688         int def_errors;
 689         unsigned long def_mount_opts;
 690         struct super_block *sb = vfs->mnt_sb;
 691         struct ext4_sb_info *sbi = EXT4_SB(sb);
 692         struct ext4_super_block *es = sbi->s_es;
 693
 694         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 695         def_errors     = le16_to_cpu(es->s_errors);
 696
 697         if (sbi->s_sb_block != 1)
 698                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 699         if (test_opt(sb, MINIX_DF))
 700                 seq_puts(seq, ",minixdf");
 701         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 702                 seq_puts(seq, ",grpid");
 703         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 704                 seq_puts(seq, ",nogrpid");
 705         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 706             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 707                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 708         }
 709         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 710             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 711                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 712         }
 713         if (test_opt(sb, ERRORS_RO)) {
 714                 if (def_errors == EXT4_ERRORS_PANIC ||
 715                     def_errors == EXT4_ERRORS_CONTINUE) {
 716                         seq_puts(seq, ",errors=remount-ro");
 717                 }
 718         }
 719         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 720                 seq_puts(seq, ",errors=continue");
 721         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 722                 seq_puts(seq, ",errors=panic");
 723         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 724                 seq_puts(seq, ",nouid32");
 725         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 726                 seq_puts(seq, ",debug");
 727         if (test_opt(sb, OLDALLOC))
 728                 seq_puts(seq, ",oldalloc");
 729 #ifdef CONFIG_EXT4DEV_FS_XATTR
 730         if (test_opt(sb, XATTR_USER) &&
 731                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 732                 seq_puts(seq, ",user_xattr");
 733         if (!test_opt(sb, XATTR_USER) &&
 734             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 735                 seq_puts(seq, ",nouser_xattr");
 736         }
 737 #endif
 738 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 739         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 740                 seq_puts(seq, ",acl");
 741         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 742                 seq_puts(seq, ",noacl");
 743 #endif
 744         if (!test_opt(sb, RESERVATION))
 745                 seq_puts(seq, ",noreservation");
 746         if (sbi->s_commit_interval) {
 747                 seq_printf(seq, ",commit=%u",
 748                            (unsigned) (sbi->s_commit_interval / HZ));
 749         }
 750         /*
 751          * We're changing the default of barrier mount option, so
 752          * let's always display its mount state so it's clear what its
 753          * status is.
 754          */
 755         seq_puts(seq, ",barrier=");
 756         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 757         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 758                 seq_puts(seq, ",journal_async_commit");
 759         if (test_opt(sb, NOBH))
 760                 seq_puts(seq, ",nobh");
 761         if (!test_opt(sb, EXTENTS))
 762                 seq_puts(seq, ",noextents");
 763         if (!test_opt(sb, MBALLOC))
 764                 seq_puts(seq, ",nomballoc");
 765         if (test_opt(sb, I_VERSION))
 766                 seq_puts(seq, ",i_version");
 767         if (!test_opt(sb, DELALLOC))
 768                 seq_puts(seq, ",nodelalloc");
 769
 770
 771         if (sbi->s_stripe)
 772                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 773         /*
 774          * journal mode get enabled in different ways
 775          * So just print the value even if we didn't specify it
 776          */
 777         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 778                 seq_puts(seq, ",data=journal");
 779         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 780                 seq_puts(seq, ",data=ordered");
 781         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 782                 seq_puts(seq, ",data=writeback");
 783
 784         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 785                 seq_printf(seq, ",inode_readahead_blks=%u",
 786                            sbi->s_inode_readahead_blks);
 787
 788         ext4_show_quota_options(seq, sb);
 789         return 0;
 790 }
 791
 792
 793 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 794                 u64 ino, u32 generation)
 795 {
 796         struct inode *inode;
 797
 798         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 799                 return ERR_PTR(-ESTALE);
 800         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 801                 return ERR_PTR(-ESTALE);
 802
 803         /* iget isn't really right if the inode is currently unallocated!!
 804          *
 805          * ext4_read_inode will return a bad_inode if the inode had been
 806          * deleted, so we should be safe.
 807          *
 808          * Currently we don't know the generation for parent directory, so
 809          * a generation of 0 means "accept any"
 810          */
 811         inode = ext4_iget(sb, ino);
 812         if (IS_ERR(inode))
 813                 return ERR_CAST(inode);
 814         if (generation && inode->i_generation != generation) {
 815                 iput(inode);
 816                 return ERR_PTR(-ESTALE);
 817         }
 818
 819         return inode;
 820 }
 821
 822 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 823                 int fh_len, int fh_type)
 824 {
 825         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 826                                     ext4_nfs_get_inode);
 827 }
 828
 829 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 830                 int fh_len, int fh_type)
 831 {
 832         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 833                                     ext4_nfs_get_inode);
 834 }
 835
 836 #ifdef CONFIG_QUOTA
 837 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
 838 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 839
 840 static int ext4_dquot_initialize(struct inode *inode, int type);
 841 static int ext4_dquot_drop(struct inode *inode);
 842 static int ext4_write_dquot(struct dquot *dquot);
 843 static int ext4_acquire_dquot(struct dquot *dquot);
 844 static int ext4_release_dquot(struct dquot *dquot);
 845 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 846 static int ext4_write_info(struct super_block *sb, int type);
 847 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 848                                 char *path, int remount);
 849 static int ext4_quota_on_mount(struct super_block *sb, int type);
 850 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 851                                size_t len, loff_t off);
 852 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 853                                 const char *data, size_t len, loff_t off);
 854
 855 static struct dquot_operations ext4_quota_operations = {
 856         .initialize     = ext4_dquot_initialize,
 857         .drop           = ext4_dquot_drop,
 858         .alloc_space    = dquot_alloc_space,
 859         .alloc_inode    = dquot_alloc_inode,
 860         .free_space     = dquot_free_space,
 861         .free_inode     = dquot_free_inode,
 862         .transfer       = dquot_transfer,
 863         .write_dquot    = ext4_write_dquot,
 864         .acquire_dquot  = ext4_acquire_dquot,
 865         .release_dquot  = ext4_release_dquot,
 866         .mark_dirty     = ext4_mark_dquot_dirty,
 867         .write_info     = ext4_write_info
 868 };
 869
 870 static struct quotactl_ops ext4_qctl_operations = {
 871         .quota_on       = ext4_quota_on,
 872         .quota_off      = vfs_quota_off,
 873         .quota_sync     = vfs_quota_sync,
 874         .get_info       = vfs_get_dqinfo,
 875         .set_info       = vfs_set_dqinfo,
 876         .get_dqblk      = vfs_get_dqblk,
 877         .set_dqblk      = vfs_set_dqblk
 878 };
 879 #endif
 880
 881 static const struct super_operations ext4_sops = {
 882         .alloc_inode    = ext4_alloc_inode,
 883         .destroy_inode  = ext4_destroy_inode,
 884         .write_inode    = ext4_write_inode,
 885         .dirty_inode    = ext4_dirty_inode,
 886         .delete_inode   = ext4_delete_inode,
 887         .put_super      = ext4_put_super,
 888         .write_super    = ext4_write_super,
 889         .sync_fs        = ext4_sync_fs,
 890         .write_super_lockfs = ext4_write_super_lockfs,
 891         .unlockfs       = ext4_unlockfs,
 892         .statfs         = ext4_statfs,
 893         .remount_fs     = ext4_remount,
 894         .clear_inode    = ext4_clear_inode,
 895         .show_options   = ext4_show_options,
 896 #ifdef CONFIG_QUOTA
 897         .quota_read     = ext4_quota_read,
 898         .quota_write    = ext4_quota_write,
 899 #endif
 900 };
 901
 902 static const struct export_operations ext4_export_ops = {
 903         .fh_to_dentry = ext4_fh_to_dentry,
 904         .fh_to_parent = ext4_fh_to_parent,
 905         .get_parent = ext4_get_parent,
 906 };
 907
 908 enum {
 909         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 910         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 911         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 912         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 913         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 914         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 915         Opt_journal_checksum, Opt_journal_async_commit,
 916         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 917         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 918         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 919         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 920         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 921         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 922         Opt_inode_readahead_blks
 923 };
 924
 925 static match_table_t tokens = {
 926         {Opt_bsd_df, "bsddf"},
 927         {Opt_minix_df, "minixdf"},
 928         {Opt_grpid, "grpid"},
 929         {Opt_grpid, "bsdgroups"},
 930         {Opt_nogrpid, "nogrpid"},
 931         {Opt_nogrpid, "sysvgroups"},
 932         {Opt_resgid, "resgid=%u"},
 933         {Opt_resuid, "resuid=%u"},
 934         {Opt_sb, "sb=%u"},
 935         {Opt_err_cont, "errors=continue"},
 936         {Opt_err_panic, "errors=panic"},
 937         {Opt_err_ro, "errors=remount-ro"},
 938         {Opt_nouid32, "nouid32"},
 939         {Opt_nocheck, "nocheck"},
 940         {Opt_nocheck, "check=none"},
 941         {Opt_debug, "debug"},
 942         {Opt_oldalloc, "oldalloc"},
 943         {Opt_orlov, "orlov"},
 944         {Opt_user_xattr, "user_xattr"},
 945         {Opt_nouser_xattr, "nouser_xattr"},
 946         {Opt_acl, "acl"},
 947         {Opt_noacl, "noacl"},
 948         {Opt_reservation, "reservation"},
 949         {Opt_noreservation, "noreservation"},
 950         {Opt_noload, "noload"},
 951         {Opt_nobh, "nobh"},
 952         {Opt_bh, "bh"},
 953         {Opt_commit, "commit=%u"},
 954         {Opt_journal_update, "journal=update"},
 955         {Opt_journal_inum, "journal=%u"},
 956         {Opt_journal_dev, "journal_dev=%u"},
 957         {Opt_journal_checksum, "journal_checksum"},
 958         {Opt_journal_async_commit, "journal_async_commit"},
 959         {Opt_abort, "abort"},
 960         {Opt_data_journal, "data=journal"},
 961         {Opt_data_ordered, "data=ordered"},
 962         {Opt_data_writeback, "data=writeback"},
 963         {Opt_offusrjquota, "usrjquota="},
 964         {Opt_usrjquota, "usrjquota=%s"},
 965         {Opt_offgrpjquota, "grpjquota="},
 966         {Opt_grpjquota, "grpjquota=%s"},
 967         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 968         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 969         {Opt_grpquota, "grpquota"},
 970         {Opt_noquota, "noquota"},
 971         {Opt_quota, "quota"},
 972         {Opt_usrquota, "usrquota"},
 973         {Opt_barrier, "barrier=%u"},
 974         {Opt_extents, "extents"},
 975         {Opt_noextents, "noextents"},
 976         {Opt_i_version, "i_version"},
 977         {Opt_mballoc, "mballoc"},
 978         {Opt_nomballoc, "nomballoc"},
 979         {Opt_stripe, "stripe=%u"},
 980         {Opt_resize, "resize"},
 981         {Opt_delalloc, "delalloc"},
 982         {Opt_nodelalloc, "nodelalloc"},
 983         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
 984         {Opt_err, NULL},
 985 };
 986
 987 static ext4_fsblk_t get_sb_block(void **data)
 988 {
 989         ext4_fsblk_t    sb_block;
 990         char            *options = (char *) *data;
 991
 992         if (!options || strncmp(options, "sb=", 3) != 0)
 993                 return 1;       /* Default location */
 994         options += 3;
 995         /*todo: use simple_strtoll with >32bit ext4 */
 996         sb_block = simple_strtoul(options, &options, 0);
 997         if (*options && *options != ',') {
 998                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
 999                        (char *) *data);
1000                 return 1;
1001         }
1002         if (*options == ',')
1003                 options++;
1004         *data = (void *) options;
1005         return sb_block;
1006 }
1007
1008 static int parse_options(char *options, struct super_block *sb,
1009                          unsigned int *inum, unsigned long *journal_devnum,
1010                          ext4_fsblk_t *n_blocks_count, int is_remount)
1011 {
1012         struct ext4_sb_info *sbi = EXT4_SB(sb);
1013         char *p;
1014         substring_t args[MAX_OPT_ARGS];
1015         int data_opt = 0;
1016         int option;
1017 #ifdef CONFIG_QUOTA
1018         int qtype, qfmt;
1019         char *qname;
1020 #endif
1021         ext4_fsblk_t last_block;
1022
1023         if (!options)
1024                 return 1;
1025
1026         while ((p = strsep(&options, ",")) != NULL) {
1027                 int token;
1028                 if (!*p)
1029                         continue;
1030
1031                 token = match_token(p, tokens, args);
1032                 switch (token) {
1033                 case Opt_bsd_df:
1034                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1035                         break;
1036                 case Opt_minix_df:
1037                         set_opt(sbi->s_mount_opt, MINIX_DF);
1038                         break;
1039                 case Opt_grpid:
1040                         set_opt(sbi->s_mount_opt, GRPID);
1041                         break;
1042                 case Opt_nogrpid:
1043                         clear_opt(sbi->s_mount_opt, GRPID);
1044                         break;
1045                 case Opt_resuid:
1046                         if (match_int(&args[0], &option))
1047                                 return 0;
1048                         sbi->s_resuid = option;
1049                         break;
1050                 case Opt_resgid:
1051                         if (match_int(&args[0], &option))
1052                                 return 0;
1053                         sbi->s_resgid = option;
1054                         break;
1055                 case Opt_sb:
1056                         /* handled by get_sb_block() instead of here */
1057                         /* *sb_block = match_int(&args[0]); */
1058                         break;
1059                 case Opt_err_panic:
1060                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1061                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1062                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1063                         break;
1064                 case Opt_err_ro:
1065                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1066                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1067                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1068                         break;
1069                 case Opt_err_cont:
1070                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1071                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1072                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1073                         break;
1074                 case Opt_nouid32:
1075                         set_opt(sbi->s_mount_opt, NO_UID32);
1076                         break;
1077                 case Opt_nocheck:
1078                         clear_opt(sbi->s_mount_opt, CHECK);
1079                         break;
1080                 case Opt_debug:
1081                         set_opt(sbi->s_mount_opt, DEBUG);
1082                         break;
1083                 case Opt_oldalloc:
1084                         set_opt(sbi->s_mount_opt, OLDALLOC);
1085                         break;
1086                 case Opt_orlov:
1087                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1088                         break;
1089 #ifdef CONFIG_EXT4DEV_FS_XATTR
1090                 case Opt_user_xattr:
1091                         set_opt(sbi->s_mount_opt, XATTR_USER);
1092                         break;
1093                 case Opt_nouser_xattr:
1094                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1095                         break;
1096 #else
1097                 case Opt_user_xattr:
1098                 case Opt_nouser_xattr:
1099                         printk(KERN_ERR "EXT4 (no)user_xattr options "
1100                                "not supported\n");
1101                         break;
1102 #endif
1103 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1104                 case Opt_acl:
1105                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1106                         break;
1107                 case Opt_noacl:
1108                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1109                         break;
1110 #else
1111                 case Opt_acl:
1112                 case Opt_noacl:
1113                         printk(KERN_ERR "EXT4 (no)acl options "
1114                                "not supported\n");
1115                         break;
1116 #endif
1117                 case Opt_reservation:
1118                         set_opt(sbi->s_mount_opt, RESERVATION);
1119                         break;
1120                 case Opt_noreservation:
1121                         clear_opt(sbi->s_mount_opt, RESERVATION);
1122                         break;
1123                 case Opt_journal_update:
1124                         /* @@@ FIXME */
1125                         /* Eventually we will want to be able to create
1126                            a journal file here.  For now, only allow the
1127                            user to specify an existing inode to be the
1128                            journal file. */
1129                         if (is_remount) {
1130                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1131                                        "journal on remount\n");
1132                                 return 0;
1133                         }
1134                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1135                         break;
1136                 case Opt_journal_inum:
1137                         if (is_remount) {
1138                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1139                                        "journal on remount\n");
1140                                 return 0;
1141                         }
1142                         if (match_int(&args[0], &option))
1143                                 return 0;
1144                         *inum = option;
1145                         break;
1146                 case Opt_journal_dev:
1147                         if (is_remount) {
1148                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1149                                        "journal on remount\n");
1150                                 return 0;
1151                         }
1152                         if (match_int(&args[0], &option))
1153                                 return 0;
1154                         *journal_devnum = option;
1155                         break;
1156                 case Opt_journal_checksum:
1157                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1158                         break;
1159                 case Opt_journal_async_commit:
1160                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1161                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1162                         break;
1163                 case Opt_noload:
1164                         set_opt(sbi->s_mount_opt, NOLOAD);
1165                         break;
1166                 case Opt_commit:
1167                         if (match_int(&args[0], &option))
1168                                 return 0;
1169                         if (option < 0)
1170                                 return 0;
1171                         if (option == 0)
1172                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1173                         sbi->s_commit_interval = HZ * option;
1174                         break;
1175                 case Opt_data_journal:
1176                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1177                         goto datacheck;
1178                 case Opt_data_ordered:
1179                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1180                         goto datacheck;
1181                 case Opt_data_writeback:
1182                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1183                 datacheck:
1184                         if (is_remount) {
1185                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1186                                                 != data_opt) {
1187                                         printk(KERN_ERR
1188                                                 "EXT4-fs: cannot change data "
1189                                                 "mode on remount\n");
1190                                         return 0;
1191                                 }
1192                         } else {
1193                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1194                                 sbi->s_mount_opt |= data_opt;
1195                         }
1196                         break;
1197 #ifdef CONFIG_QUOTA
1198                 case Opt_usrjquota:
1199                         qtype = USRQUOTA;
1200                         goto set_qf_name;
1201                 case Opt_grpjquota:
1202                         qtype = GRPQUOTA;
1203 set_qf_name:
1204                         if ((sb_any_quota_enabled(sb) ||
1205                              sb_any_quota_suspended(sb)) &&
1206                             !sbi->s_qf_names[qtype]) {
1207                                 printk(KERN_ERR
1208                                        "EXT4-fs: Cannot change journaled "
1209                                        "quota options when quota turned on.\n");
1210                                 return 0;
1211                         }
1212                         qname = match_strdup(&args[0]);
1213                         if (!qname) {
1214                                 printk(KERN_ERR
1215                                         "EXT4-fs: not enough memory for "
1216                                         "storing quotafile name.\n");
1217                                 return 0;
1218                         }
1219                         if (sbi->s_qf_names[qtype] &&
1220                             strcmp(sbi->s_qf_names[qtype], qname)) {
1221                                 printk(KERN_ERR
1222                                         "EXT4-fs: %s quota file already "
1223                                         "specified.\n", QTYPE2NAME(qtype));
1224                                 kfree(qname);
1225                                 return 0;
1226                         }
1227                         sbi->s_qf_names[qtype] = qname;
1228                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1229                                 printk(KERN_ERR
1230                                         "EXT4-fs: quotafile must be on "
1231                                         "filesystem root.\n");
1232                                 kfree(sbi->s_qf_names[qtype]);
1233                                 sbi->s_qf_names[qtype] = NULL;
1234                                 return 0;
1235                         }
1236                         set_opt(sbi->s_mount_opt, QUOTA);
1237                         break;
1238                 case Opt_offusrjquota:
1239                         qtype = USRQUOTA;
1240                         goto clear_qf_name;
1241                 case Opt_offgrpjquota:
1242                         qtype = GRPQUOTA;
1243 clear_qf_name:
1244                         if ((sb_any_quota_enabled(sb) ||
1245                              sb_any_quota_suspended(sb)) &&
1246                             sbi->s_qf_names[qtype]) {
1247                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1248                                         "journaled quota options when "
1249                                         "quota turned on.\n");
1250                                 return 0;
1251                         }
1252                         /*
1253                          * The space will be released later when all options
1254                          * are confirmed to be correct
1255                          */
1256                         sbi->s_qf_names[qtype] = NULL;
1257                         break;
1258                 case Opt_jqfmt_vfsold:
1259                         qfmt = QFMT_VFS_OLD;
1260                         goto set_qf_format;
1261                 case Opt_jqfmt_vfsv0:
1262                         qfmt = QFMT_VFS_V0;
1263 set_qf_format:
1264                         if ((sb_any_quota_enabled(sb) ||
1265                              sb_any_quota_suspended(sb)) &&
1266                             sbi->s_jquota_fmt != qfmt) {
1267                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1268                                         "journaled quota options when "
1269                                         "quota turned on.\n");
1270                                 return 0;
1271                         }
1272                         sbi->s_jquota_fmt = qfmt;
1273                         break;
1274                 case Opt_quota:
1275                 case Opt_usrquota:
1276                         set_opt(sbi->s_mount_opt, QUOTA);
1277                         set_opt(sbi->s_mount_opt, USRQUOTA);
1278                         break;
1279                 case Opt_grpquota:
1280                         set_opt(sbi->s_mount_opt, QUOTA);
1281                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1282                         break;
1283                 case Opt_noquota:
1284                         if (sb_any_quota_enabled(sb)) {
1285                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1286                                         "options when quota turned on.\n");
1287                                 return 0;
1288                         }
1289                         clear_opt(sbi->s_mount_opt, QUOTA);
1290                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1291                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1292                         break;
1293 #else
1294                 case Opt_quota:
1295                 case Opt_usrquota:
1296                 case Opt_grpquota:
1297                         printk(KERN_ERR
1298                                 "EXT4-fs: quota options not supported.\n");
1299                         break;
1300                 case Opt_usrjquota:
1301                 case Opt_grpjquota:
1302                 case Opt_offusrjquota:
1303                 case Opt_offgrpjquota:
1304                 case Opt_jqfmt_vfsold:
1305                 case Opt_jqfmt_vfsv0:
1306                         printk(KERN_ERR
1307                                 "EXT4-fs: journaled quota options not "
1308                                 "supported.\n");
1309                         break;
1310                 case Opt_noquota:
1311                         break;
1312 #endif
1313                 case Opt_abort:
1314                         set_opt(sbi->s_mount_opt, ABORT);
1315                         break;
1316                 case Opt_barrier:
1317                         if (match_int(&args[0], &option))
1318                                 return 0;
1319                         if (option)
1320                                 set_opt(sbi->s_mount_opt, BARRIER);
1321                         else
1322                                 clear_opt(sbi->s_mount_opt, BARRIER);
1323                         break;
1324                 case Opt_ignore:
1325                         break;
1326                 case Opt_resize:
1327                         if (!is_remount) {
1328                                 printk("EXT4-fs: resize option only available "
1329                                         "for remount\n");
1330                                 return 0;
1331                         }
1332                         if (match_int(&args[0], &option) != 0)
1333                                 return 0;
1334                         *n_blocks_count = option;
1335                         break;
1336                 case Opt_nobh:
1337                         set_opt(sbi->s_mount_opt, NOBH);
1338                         break;
1339                 case Opt_bh:
1340                         clear_opt(sbi->s_mount_opt, NOBH);
1341                         break;
1342                 case Opt_extents:
1343                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1344                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1345                                 ext4_warning(sb, __func__,
1346                                         "extents feature not enabled "
1347                                         "on this filesystem, use tune2fs\n");
1348                                 return 0;
1349                         }
1350                         set_opt(sbi->s_mount_opt, EXTENTS);
1351                         break;
1352                 case Opt_noextents:
1353                         /*
1354                          * When e2fsprogs support resizing an already existing
1355                          * ext3 file system to greater than 2**32 we need to
1356                          * add support to block allocator to handle growing
1357                          * already existing block  mapped inode so that blocks
1358                          * allocated for them fall within 2**32
1359                          */
1360                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1361                         if (last_block  > 0xffffffffULL) {
1362                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1363                                                 "large to mount with "
1364                                                 "-o noextents options\n");
1365                                 return 0;
1366                         }
1367                         clear_opt(sbi->s_mount_opt, EXTENTS);
1368                         break;
1369                 case Opt_i_version:
1370                         set_opt(sbi->s_mount_opt, I_VERSION);
1371                         sb->s_flags |= MS_I_VERSION;
1372                         break;
1373                 case Opt_nodelalloc:
1374                         clear_opt(sbi->s_mount_opt, DELALLOC);
1375                         break;
1376                 case Opt_mballoc:
1377                         set_opt(sbi->s_mount_opt, MBALLOC);
1378                         break;
1379                 case Opt_nomballoc:
1380                         clear_opt(sbi->s_mount_opt, MBALLOC);
1381                         break;
1382                 case Opt_stripe:
1383                         if (match_int(&args[0], &option))
1384                                 return 0;
1385                         if (option < 0)
1386                                 return 0;
1387                         sbi->s_stripe = option;
1388                         break;
1389                 case Opt_delalloc:
1390                         set_opt(sbi->s_mount_opt, DELALLOC);
1391                         break;
1392                 case Opt_inode_readahead_blks:
1393                         if (match_int(&args[0], &option))
1394                                 return 0;
1395                         if (option < 0 || option > (1 << 30))
1396                                 return 0;
1397                         sbi->s_inode_readahead_blks = option;
1398                         break;
1399                 default:
1400                         printk(KERN_ERR
1401                                "EXT4-fs: Unrecognized mount option \"%s\" "
1402                                "or missing value\n", p);
1403                         return 0;
1404                 }
1405         }
1406 #ifdef CONFIG_QUOTA
1407         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1408                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1409                      sbi->s_qf_names[USRQUOTA])
1410                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1411
1412                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1413                      sbi->s_qf_names[GRPQUOTA])
1414                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1415
1416                 if ((sbi->s_qf_names[USRQUOTA] &&
1417                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1418                     (sbi->s_qf_names[GRPQUOTA] &&
1419                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1420                         printk(KERN_ERR "EXT4-fs: old and new quota "
1421                                         "format mixing.\n");
1422                         return 0;
1423                 }
1424
1425                 if (!sbi->s_jquota_fmt) {
1426                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1427                                         "not specified.\n");
1428                         return 0;
1429                 }
1430         } else {
1431                 if (sbi->s_jquota_fmt) {
1432                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1433                                         "specified with no journaling "
1434                                         "enabled.\n");
1435                         return 0;
1436                 }
1437         }
1438 #endif
1439         return 1;
1440 }
1441
1442 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1443                             int read_only)
1444 {
1445         struct ext4_sb_info *sbi = EXT4_SB(sb);
1446         int res = 0;
1447
1448         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1449                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1450                        "forcing read-only mode\n");
1451                 res = MS_RDONLY;
1452         }
1453         if (read_only)
1454                 return res;
1455         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1456                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1457                        "running e2fsck is recommended\n");
1458         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1459                 printk(KERN_WARNING
1460                        "EXT4-fs warning: mounting fs with errors, "
1461                        "running e2fsck is recommended\n");
1462         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1463                  le16_to_cpu(es->s_mnt_count) >=
1464                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1465                 printk(KERN_WARNING
1466                        "EXT4-fs warning: maximal mount count reached, "
1467                        "running e2fsck is recommended\n");
1468         else if (le32_to_cpu(es->s_checkinterval) &&
1469                 (le32_to_cpu(es->s_lastcheck) +
1470                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1471                 printk(KERN_WARNING
1472                        "EXT4-fs warning: checktime reached, "
1473                        "running e2fsck is recommended\n");
1474 #if 0
1475                 /* @@@ We _will_ want to clear the valid bit if we find
1476                  * inconsistencies, to force a fsck at reboot.  But for
1477                  * a plain journaled filesystem we can keep it set as
1478                  * valid forever! :)
1479                  */
1480         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1481 #endif
1482         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1483                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1484         le16_add_cpu(&es->s_mnt_count, 1);
1485         es->s_mtime = cpu_to_le32(get_seconds());
1486         ext4_update_dynamic_rev(sb);
1487         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1488
1489         ext4_commit_super(sb, es, 1);
1490         if (test_opt(sb, DEBUG))
1491                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1492                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1493                         sb->s_blocksize,
1494                         sbi->s_groups_count,
1495                         EXT4_BLOCKS_PER_GROUP(sb),
1496                         EXT4_INODES_PER_GROUP(sb),
1497                         sbi->s_mount_opt);
1498
1499         printk(KERN_INFO "EXT4 FS on %s, %s journal on %s\n",
1500                sb->s_id, EXT4_SB(sb)->s_journal->j_inode ? "internal" :
1501                "external", EXT4_SB(sb)->s_journal->j_devname);
1502         return res;
1503 }
1504
1505 static int ext4_fill_flex_info(struct super_block *sb)
1506 {
1507         struct ext4_sb_info *sbi = EXT4_SB(sb);
1508         struct ext4_group_desc *gdp = NULL;
1509         struct buffer_head *bh;
1510         ext4_group_t flex_group_count;
1511         ext4_group_t flex_group;
1512         int groups_per_flex = 0;
1513         __u64 block_bitmap = 0;
1514         int i;
1515
1516         if (!sbi->s_es->s_log_groups_per_flex) {
1517                 sbi->s_log_groups_per_flex = 0;
1518                 return 1;
1519         }
1520
1521         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1522         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1523
1524         /* We allocate both existing and potentially added groups */
1525         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1526                             ((sbi->s_es->s_reserved_gdt_blocks +1 ) <<
1527                               EXT4_DESC_PER_BLOCK_BITS(sb))) /
1528                            groups_per_flex;
1529         sbi->s_flex_groups = kzalloc(flex_group_count *
1530                                      sizeof(struct flex_groups), GFP_KERNEL);
1531         if (sbi->s_flex_groups == NULL) {
1532                 printk(KERN_ERR "EXT4-fs: not enough memory for "
1533                                 "%lu flex groups\n", flex_group_count);
1534                 goto failed;
1535         }
1536
1537         gdp = ext4_get_group_desc(sb, 1, &bh);
1538         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1539
1540         for (i = 0; i < sbi->s_groups_count; i++) {
1541                 gdp = ext4_get_group_desc(sb, i, &bh);
1542
1543                 flex_group = ext4_flex_group(sbi, i);
1544                 sbi->s_flex_groups[flex_group].free_inodes +=
1545                         le16_to_cpu(gdp->bg_free_inodes_count);
1546                 sbi->s_flex_groups[flex_group].free_blocks +=
1547                         le16_to_cpu(gdp->bg_free_blocks_count);
1548         }
1549
1550         return 1;
1551 failed:
1552         return 0;
1553 }
1554
1555 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1556                             struct ext4_group_desc *gdp)
1557 {
1558         __u16 crc = 0;
1559
1560         if (sbi->s_es->s_feature_ro_compat &
1561             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1562                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1563                 __le32 le_group = cpu_to_le32(block_group);
1564
1565                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1566                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1567                 crc = crc16(crc, (__u8 *)gdp, offset);
1568                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1569                 /* for checksum of struct ext4_group_desc do the rest...*/
1570                 if ((sbi->s_es->s_feature_incompat &
1571                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1572                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1573                         crc = crc16(crc, (__u8 *)gdp + offset,
1574                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1575                                         offset);
1576         }
1577
1578         return cpu_to_le16(crc);
1579 }
1580
1581 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1582                                 struct ext4_group_desc *gdp)
1583 {
1584         if ((sbi->s_es->s_feature_ro_compat &
1585              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1586             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1587                 return 0;
1588
1589         return 1;
1590 }
1591
1592 /* Called at mount-time, super-block is locked */
1593 static int ext4_check_descriptors(struct super_block *sb)
1594 {
1595         struct ext4_sb_info *sbi = EXT4_SB(sb);
1596         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1597         ext4_fsblk_t last_block;
1598         ext4_fsblk_t block_bitmap;
1599         ext4_fsblk_t inode_bitmap;
1600         ext4_fsblk_t inode_table;
1601         int flexbg_flag = 0;
1602         ext4_group_t i;
1603
1604         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1605                 flexbg_flag = 1;
1606
1607         ext4_debug("Checking group descriptors");
1608
1609         for (i = 0; i < sbi->s_groups_count; i++) {
1610                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1611
1612                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1613                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1614                 else
1615                         last_block = first_block +
1616                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1617
1618                 block_bitmap = ext4_block_bitmap(sb, gdp);
1619                 if (block_bitmap < first_block || block_bitmap > last_block) {
1620                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1621                                "Block bitmap for group %lu not in group "
1622                                "(block %llu)!", i, block_bitmap);
1623                         return 0;
1624                 }
1625                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1626                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1627                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1628                                "Inode bitmap for group %lu not in group "
1629                                "(block %llu)!", i, inode_bitmap);
1630                         return 0;
1631                 }
1632                 inode_table = ext4_inode_table(sb, gdp);
1633                 if (inode_table < first_block ||
1634                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1635                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1636                                "Inode table for group %lu not in group "
1637                                "(block %llu)!", i, inode_table);
1638                         return 0;
1639                 }
1640                 spin_lock(sb_bgl_lock(sbi, i));
1641                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1642                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1643                                "Checksum for group %lu failed (%u!=%u)\n",
1644                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1645                                gdp)), le16_to_cpu(gdp->bg_checksum));
1646                         if (!(sb->s_flags & MS_RDONLY)) {
1647                                 spin_unlock(sb_bgl_lock(sbi, i));
1648                                 return 0;
1649                         }
1650                 }
1651                 spin_unlock(sb_bgl_lock(sbi, i));
1652                 if (!flexbg_flag)
1653                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1654         }
1655
1656         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1657         sbi->s_es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
1658         return 1;
1659 }
1660
1661 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1662  * the superblock) which were deleted from all directories, but held open by
1663  * a process at the time of a crash.  We walk the list and try to delete these
1664  * inodes at recovery time (only with a read-write filesystem).
1665  *
1666  * In order to keep the orphan inode chain consistent during traversal (in
1667  * case of crash during recovery), we link each inode into the superblock
1668  * orphan list_head and handle it the same way as an inode deletion during
1669  * normal operation (which journals the operations for us).
1670  *
1671  * We only do an iget() and an iput() on each inode, which is very safe if we
1672  * accidentally point at an in-use or already deleted inode.  The worst that
1673  * can happen in this case is that we get a "bit already cleared" message from
1674  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1675  * e2fsck was run on this filesystem, and it must have already done the orphan
1676  * inode cleanup for us, so we can safely abort without any further action.
1677  */
1678 static void ext4_orphan_cleanup(struct super_block *sb,
1679                                 struct ext4_super_block *es)
1680 {
1681         unsigned int s_flags = sb->s_flags;
1682         int nr_orphans = 0, nr_truncates = 0;
1683 #ifdef CONFIG_QUOTA
1684         int i;
1685 #endif
1686         if (!es->s_last_orphan) {
1687                 jbd_debug(4, "no orphan inodes to clean up\n");
1688                 return;
1689         }
1690
1691         if (bdev_read_only(sb->s_bdev)) {
1692                 printk(KERN_ERR "EXT4-fs: write access "
1693                         "unavailable, skipping orphan cleanup.\n");
1694                 return;
1695         }
1696
1697         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1698                 if (es->s_last_orphan)
1699                         jbd_debug(1, "Errors on filesystem, "
1700                                   "clearing orphan list.\n");
1701                 es->s_last_orphan = 0;
1702                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1703                 return;
1704         }
1705
1706         if (s_flags & MS_RDONLY) {
1707                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1708                        sb->s_id);
1709                 sb->s_flags &= ~MS_RDONLY;
1710         }
1711 #ifdef CONFIG_QUOTA
1712         /* Needed for iput() to work correctly and not trash data */
1713         sb->s_flags |= MS_ACTIVE;
1714         /* Turn on quotas so that they are updated correctly */
1715         for (i = 0; i < MAXQUOTAS; i++) {
1716                 if (EXT4_SB(sb)->s_qf_names[i]) {
1717                         int ret = ext4_quota_on_mount(sb, i);
1718                         if (ret < 0)
1719                                 printk(KERN_ERR
1720                                         "EXT4-fs: Cannot turn on journaled "
1721                                         "quota: error %d\n", ret);
1722                 }
1723         }
1724 #endif
1725
1726         while (es->s_last_orphan) {
1727                 struct inode *inode;
1728
1729                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1730                 if (IS_ERR(inode)) {
1731                         es->s_last_orphan = 0;
1732                         break;
1733                 }
1734
1735                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1736                 DQUOT_INIT(inode);
1737                 if (inode->i_nlink) {
1738                         printk(KERN_DEBUG
1739                                 "%s: truncating inode %lu to %lld bytes\n",
1740                                 __func__, inode->i_ino, inode->i_size);
1741                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1742                                   inode->i_ino, inode->i_size);
1743                         ext4_truncate(inode);
1744                         nr_truncates++;
1745                 } else {
1746                         printk(KERN_DEBUG
1747                                 "%s: deleting unreferenced inode %lu\n",
1748                                 __func__, inode->i_ino);
1749                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1750                                   inode->i_ino);
1751                         nr_orphans++;
1752                 }
1753                 iput(inode);  /* The delete magic happens here! */
1754         }
1755
1756 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1757
1758         if (nr_orphans)
1759                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1760                        sb->s_id, PLURAL(nr_orphans));
1761         if (nr_truncates)
1762                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1763                        sb->s_id, PLURAL(nr_truncates));
1764 #ifdef CONFIG_QUOTA
1765         /* Turn quotas off */
1766         for (i = 0; i < MAXQUOTAS; i++) {
1767                 if (sb_dqopt(sb)->files[i])
1768                         vfs_quota_off(sb, i, 0);
1769         }
1770 #endif
1771         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1772 }
1773 /*
1774  * Maximal extent format file size.
1775  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1776  * extent format containers, within a sector_t, and within i_blocks
1777  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1778  * so that won't be a limiting factor.
1779  *
1780  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1781  */
1782 static loff_t ext4_max_size(int blkbits)
1783 {
1784         loff_t res;
1785         loff_t upper_limit = MAX_LFS_FILESIZE;
1786
1787         /* small i_blocks in vfs inode? */
1788         if (sizeof(blkcnt_t) < sizeof(u64)) {
1789                 /*
1790                  * CONFIG_LSF is not enabled implies the inode
1791                  * i_block represent total blocks in 512 bytes
1792                  * 32 == size of vfs inode i_blocks * 8
1793                  */
1794                 upper_limit = (1LL << 32) - 1;
1795
1796                 /* total blocks in file system block size */
1797                 upper_limit >>= (blkbits - 9);
1798                 upper_limit <<= blkbits;
1799         }
1800
1801         /* 32-bit extent-start container, ee_block */
1802         res = 1LL << 32;
1803         res <<= blkbits;
1804         res -= 1;
1805
1806         /* Sanity check against vm- & vfs- imposed limits */
1807         if (res > upper_limit)
1808                 res = upper_limit;
1809
1810         return res;
1811 }
1812
1813 /*
1814  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1815  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1816  * We need to be 1 filesystem block less than the 2^48 sector limit.
1817  */
1818 static loff_t ext4_max_bitmap_size(int bits)
1819 {
1820         loff_t res = EXT4_NDIR_BLOCKS;
1821         int meta_blocks;
1822         loff_t upper_limit;
1823         /* This is calculated to be the largest file size for a
1824          * dense, bitmapped file such that the total number of
1825          * sectors in the file, including data and all indirect blocks,
1826          * does not exceed 2^48 -1
1827          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1828          * total number of  512 bytes blocks of the file
1829          */
1830
1831         if (sizeof(blkcnt_t) < sizeof(u64)) {
1832                 /*
1833                  * CONFIG_LSF is not enabled implies the inode
1834                  * i_block represent total blocks in 512 bytes
1835                  * 32 == size of vfs inode i_blocks * 8
1836                  */
1837                 upper_limit = (1LL << 32) - 1;
1838
1839                 /* total blocks in file system block size */
1840                 upper_limit >>= (bits - 9);
1841
1842         } else {
1843                 /*
1844                  * We use 48 bit ext4_inode i_blocks
1845                  * With EXT4_HUGE_FILE_FL set the i_blocks
1846                  * represent total number of blocks in
1847                  * file system block size
1848                  */
1849                 upper_limit = (1LL << 48) - 1;
1850
1851         }
1852
1853         /* indirect blocks */
1854         meta_blocks = 1;
1855         /* double indirect blocks */
1856         meta_blocks += 1 + (1LL << (bits-2));
1857         /* tripple indirect blocks */
1858         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1859
1860         upper_limit -= meta_blocks;
1861         upper_limit <<= bits;
1862
1863         res += 1LL << (bits-2);
1864         res += 1LL << (2*(bits-2));
1865         res += 1LL << (3*(bits-2));
1866         res <<= bits;
1867         if (res > upper_limit)
1868                 res = upper_limit;
1869
1870         if (res > MAX_LFS_FILESIZE)
1871                 res = MAX_LFS_FILESIZE;
1872
1873         return res;
1874 }
1875
1876 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1877                                 ext4_fsblk_t logical_sb_block, int nr)
1878 {
1879         struct ext4_sb_info *sbi = EXT4_SB(sb);
1880         ext4_group_t bg, first_meta_bg;
1881         int has_super = 0;
1882
1883         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1884
1885         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1886             nr < first_meta_bg)
1887                 return logical_sb_block + nr + 1;
1888         bg = sbi->s_desc_per_block * nr;
1889         if (ext4_bg_has_super(sb, bg))
1890                 has_super = 1;
1891         return (has_super + ext4_group_first_block_no(sb, bg));
1892 }
1893
1894 /**
1895  * ext4_get_stripe_size: Get the stripe size.
1896  * @sbi: In memory super block info
1897  *
1898  * If we have specified it via mount option, then
1899  * use the mount option value. If the value specified at mount time is
1900  * greater than the blocks per group use the super block value.
1901  * If the super block value is greater than blocks per group return 0.
1902  * Allocator needs it be less than blocks per group.
1903  *
1904  */
1905 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1906 {
1907         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1908         unsigned long stripe_width =
1909                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1910
1911         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1912                 return sbi->s_stripe;
1913
1914         if (stripe_width <= sbi->s_blocks_per_group)
1915                 return stripe_width;
1916
1917         if (stride <= sbi->s_blocks_per_group)
1918                 return stride;
1919
1920         return 0;
1921 }
1922
1923 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
1924                                 __releases(kernel_lock)
1925                                 __acquires(kernel_lock)
1926
1927 {
1928         struct buffer_head *bh;
1929         struct ext4_super_block *es = NULL;
1930         struct ext4_sb_info *sbi;
1931         ext4_fsblk_t block;
1932         ext4_fsblk_t sb_block = get_sb_block(&data);
1933         ext4_fsblk_t logical_sb_block;
1934         unsigned long offset = 0;
1935         unsigned int journal_inum = 0;
1936         unsigned long journal_devnum = 0;
1937         unsigned long def_mount_opts;
1938         struct inode *root;
1939         char *cp;
1940         int ret = -EINVAL;
1941         int blocksize;
1942         int db_count;
1943         int i;
1944         int needs_recovery;
1945         __le32 features;
1946         __u64 blocks_count;
1947         int err;
1948
1949         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1950         if (!sbi)
1951                 return -ENOMEM;
1952         sb->s_fs_info = sbi;
1953         sbi->s_mount_opt = 0;
1954         sbi->s_resuid = EXT4_DEF_RESUID;
1955         sbi->s_resgid = EXT4_DEF_RESGID;
1956         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
1957         sbi->s_sb_block = sb_block;
1958
1959         unlock_kernel();
1960
1961         /* Cleanup superblock name */
1962         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
1963                 *cp = '!';
1964
1965         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1966         if (!blocksize) {
1967                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1968                 goto out_fail;
1969         }
1970
1971         /*
1972          * The ext4 superblock will not be buffer aligned for other than 1kB
1973          * block sizes.  We need to calculate the offset from buffer start.
1974          */
1975         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1976                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1977                 offset = do_div(logical_sb_block, blocksize);
1978         } else {
1979                 logical_sb_block = sb_block;
1980         }
1981
1982         if (!(bh = sb_bread(sb, logical_sb_block))) {
1983                 printk(KERN_ERR "EXT4-fs: unable to read superblock\n");
1984                 goto out_fail;
1985         }
1986         /*
1987          * Note: s_es must be initialized as soon as possible because
1988          *       some ext4 macro-instructions depend on its value
1989          */
1990         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1991         sbi->s_es = es;
1992         sb->s_magic = le16_to_cpu(es->s_magic);
1993         if (sb->s_magic != EXT4_SUPER_MAGIC)
1994                 goto cantfind_ext4;
1995
1996         /* Set defaults before we parse the mount options */
1997         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1998         if (def_mount_opts & EXT4_DEFM_DEBUG)
1999                 set_opt(sbi->s_mount_opt, DEBUG);
2000         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
2001                 set_opt(sbi->s_mount_opt, GRPID);
2002         if (def_mount_opts & EXT4_DEFM_UID16)
2003                 set_opt(sbi->s_mount_opt, NO_UID32);
2004 #ifdef CONFIG_EXT4DEV_FS_XATTR
2005         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2006                 set_opt(sbi->s_mount_opt, XATTR_USER);
2007 #endif
2008 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2009         if (def_mount_opts & EXT4_DEFM_ACL)
2010                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2011 #endif
2012         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2013                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
2014         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2015                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
2016         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2017                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
2018
2019         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2020                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2021         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2022                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2023         else
2024                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2025
2026         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2027         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2028
2029         set_opt(sbi->s_mount_opt, RESERVATION);
2030         set_opt(sbi->s_mount_opt, BARRIER);
2031
2032         /*
2033          * turn on extents feature by default in ext4 filesystem
2034          * only if feature flag already set by mkfs or tune2fs.
2035          * Use -o noextents to turn it off
2036          */
2037         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2038                 set_opt(sbi->s_mount_opt, EXTENTS);
2039         else
2040                 ext4_warning(sb, __func__,
2041                         "extents feature not enabled on this filesystem, "
2042                         "use tune2fs.\n");
2043         /*
2044          * turn on mballoc code by default in ext4 filesystem
2045          * Use -o nomballoc to turn it off
2046          */
2047         set_opt(sbi->s_mount_opt, MBALLOC);
2048
2049         /*
2050          * enable delayed allocation by default
2051          * Use -o nodelalloc to turn it off
2052          */
2053         set_opt(sbi->s_mount_opt, DELALLOC);
2054
2055
2056         if (!parse_options((char *) data, sb, &journal_inum, &journal_devnum,
2057                            NULL, 0))
2058                 goto failed_mount;
2059
2060         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2061                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2062
2063         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2064             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2065              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2066              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2067                 printk(KERN_WARNING
2068                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2069                        "running e2fsck is recommended\n");
2070
2071         /*
2072          * Since ext4 is still considered development code, we require
2073          * that the TEST_FILESYS flag in s->flags be set.
2074          */
2075         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
2076                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
2077                        "OK to use with test code.\n", sb->s_id);
2078                 goto failed_mount;
2079         }
2080
2081         /*
2082          * Check feature flags regardless of the revision level, since we
2083          * previously didn't change the revision level when setting the flags,
2084          * so there is a chance incompat flags are set on a rev 0 filesystem.
2085          */
2086         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2087         if (features) {
2088                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2089                        "unsupported optional features (%x).\n",
2090                        sb->s_id, le32_to_cpu(features));
2091                 goto failed_mount;
2092         }
2093         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2094         if (!(sb->s_flags & MS_RDONLY) && features) {
2095                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2096                        "unsupported optional features (%x).\n",
2097                        sb->s_id, le32_to_cpu(features));
2098                 goto failed_mount;
2099         }
2100         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2101                 /*
2102                  * Large file size enabled file system can only be
2103                  * mount if kernel is build with CONFIG_LSF
2104                  */
2105                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2106                                 !(sb->s_flags & MS_RDONLY)) {
2107                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2108                                         "files cannot be mounted read-write "
2109                                         "without CONFIG_LSF.\n", sb->s_id);
2110                         goto failed_mount;
2111                 }
2112         }
2113         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2114
2115         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2116             blocksize > EXT4_MAX_BLOCK_SIZE) {
2117                 printk(KERN_ERR
2118                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2119                        blocksize, sb->s_id);
2120                 goto failed_mount;
2121         }
2122
2123         if (sb->s_blocksize != blocksize) {
2124
2125                 /* Validate the filesystem blocksize */
2126                 if (!sb_set_blocksize(sb, blocksize)) {
2127                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2128                                         blocksize);
2129                         goto failed_mount;
2130                 }
2131
2132                 brelse(bh);
2133                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2134                 offset = do_div(logical_sb_block, blocksize);
2135                 bh = sb_bread(sb, logical_sb_block);
2136                 if (!bh) {
2137                         printk(KERN_ERR
2138                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2139                         goto failed_mount;
2140                 }
2141                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2142                 sbi->s_es = es;
2143                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2144                         printk(KERN_ERR
2145                                "EXT4-fs: Magic mismatch, very weird !\n");
2146                         goto failed_mount;
2147                 }
2148         }
2149
2150         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2151         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2152
2153         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2154                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2155                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2156         } else {
2157                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2158                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2159                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2160                     (!is_power_of_2(sbi->s_inode_size)) ||
2161                     (sbi->s_inode_size > blocksize)) {
2162                         printk(KERN_ERR
2163                                "EXT4-fs: unsupported inode size: %d\n",
2164                                sbi->s_inode_size);
2165                         goto failed_mount;
2166                 }
2167                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2168                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2169         }
2170         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2171         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2172                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2173                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2174                     !is_power_of_2(sbi->s_desc_size)) {
2175                         printk(KERN_ERR
2176                                "EXT4-fs: unsupported descriptor size %lu\n",
2177                                sbi->s_desc_size);
2178                         goto failed_mount;
2179                 }
2180         } else
2181                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2182         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2183         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2184         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2185                 goto cantfind_ext4;
2186         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2187         if (sbi->s_inodes_per_block == 0)
2188                 goto cantfind_ext4;
2189         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2190                                         sbi->s_inodes_per_block;
2191         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2192         sbi->s_sbh = bh;
2193         sbi->s_mount_state = le16_to_cpu(es->s_state);
2194         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2195         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2196         for (i = 0; i < 4; i++)
2197                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2198         sbi->s_def_hash_version = es->s_def_hash_version;
2199
2200         if (sbi->s_blocks_per_group > blocksize * 8) {
2201                 printk(KERN_ERR
2202                        "EXT4-fs: #blocks per group too big: %lu\n",
2203                        sbi->s_blocks_per_group);
2204                 goto failed_mount;
2205         }
2206         if (sbi->s_inodes_per_group > blocksize * 8) {
2207                 printk(KERN_ERR
2208                        "EXT4-fs: #inodes per group too big: %lu\n",
2209                        sbi->s_inodes_per_group);
2210                 goto failed_mount;
2211         }
2212
2213         if (ext4_blocks_count(es) >
2214                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2215                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2216                         " too large to mount safely\n", sb->s_id);
2217                 if (sizeof(sector_t) < 8)
2218                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2219                                         "enabled\n");
2220                 goto failed_mount;
2221         }
2222
2223         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2224                 goto cantfind_ext4;
2225
2226         /* ensure blocks_count calculation below doesn't sign-extend */
2227         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2228             le32_to_cpu(es->s_first_data_block) + 1) {
2229                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2230                        "first data block %u, blocks per group %lu\n",
2231                         ext4_blocks_count(es),
2232                         le32_to_cpu(es->s_first_data_block),
2233                         EXT4_BLOCKS_PER_GROUP(sb));
2234                 goto failed_mount;
2235         }
2236         blocks_count = (ext4_blocks_count(es) -
2237                         le32_to_cpu(es->s_first_data_block) +
2238                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2239         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2240         sbi->s_groups_count = blocks_count;
2241         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2242                    EXT4_DESC_PER_BLOCK(sb);
2243         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2244                                     GFP_KERNEL);
2245         if (sbi->s_group_desc == NULL) {
2246                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
2247                 goto failed_mount;
2248         }
2249
2250         if (ext4_proc_root)
2251                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2252
2253         if (sbi->s_proc)
2254                 proc_create_data("inode_readahead_blks", 0644, sbi->s_proc,
2255                                  &ext4_ui_proc_fops,
2256                                  &sbi->s_inode_readahead_blks);
2257
2258         bgl_lock_init(&sbi->s_blockgroup_lock);
2259
2260         for (i = 0; i < db_count; i++) {
2261                 block = descriptor_loc(sb, logical_sb_block, i);
2262                 sbi->s_group_desc[i] = sb_bread(sb, block);
2263                 if (!sbi->s_group_desc[i]) {
2264                         printk(KERN_ERR "EXT4-fs: "
2265                                "can't read group descriptor %d\n", i);
2266                         db_count = i;
2267                         goto failed_mount2;
2268                 }
2269         }
2270         if (!ext4_check_descriptors(sb)) {
2271                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2272                 goto failed_mount2;
2273         }
2274         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2275                 if (!ext4_fill_flex_info(sb)) {
2276                         printk(KERN_ERR
2277                                "EXT4-fs: unable to initialize "
2278                                "flex_bg meta info!\n");
2279                         goto failed_mount2;
2280                 }
2281
2282         sbi->s_gdb_count = db_count;
2283         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2284         spin_lock_init(&sbi->s_next_gen_lock);
2285
2286         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2287                         ext4_count_free_blocks(sb));
2288         if (!err) {
2289                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2290                                 ext4_count_free_inodes(sb));
2291         }
2292         if (!err) {
2293                 err = percpu_counter_init(&sbi->s_dirs_counter,
2294                                 ext4_count_dirs(sb));
2295         }
2296         if (!err) {
2297                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2298         }
2299         if (err) {
2300                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2301                 goto failed_mount3;
2302         }
2303
2304         /* per fileystem reservation list head & lock */
2305         spin_lock_init(&sbi->s_rsv_window_lock);
2306         sbi->s_rsv_window_root = RB_ROOT;
2307         /* Add a single, static dummy reservation to the start of the
2308          * reservation window list --- it gives us a placeholder for
2309          * append-at-start-of-list which makes the allocation logic
2310          * _much_ simpler. */
2311         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2312         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2313         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2314         sbi->s_rsv_window_head.rsv_goal_size = 0;
2315         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2316
2317         sbi->s_stripe = ext4_get_stripe_size(sbi);
2318
2319         /*
2320          * set up enough so that it can read an inode
2321          */
2322         sb->s_op = &ext4_sops;
2323         sb->s_export_op = &ext4_export_ops;
2324         sb->s_xattr = ext4_xattr_handlers;
2325 #ifdef CONFIG_QUOTA
2326         sb->s_qcop = &ext4_qctl_operations;
2327         sb->dq_op = &ext4_quota_operations;
2328 #endif
2329         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2330
2331         sb->s_root = NULL;
2332
2333         needs_recovery = (es->s_last_orphan != 0 ||
2334                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2335                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2336
2337         /*
2338          * The first inode we look at is the journal inode.  Don't try
2339          * root first: it may be modified in the journal!
2340          */
2341         if (!test_opt(sb, NOLOAD) &&
2342             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2343                 if (ext4_load_journal(sb, es, journal_devnum))
2344                         goto failed_mount3;
2345                 if (!(sb->s_flags & MS_RDONLY) &&
2346                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2347                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2348                                "ext4_fill_super: Journal transaction "
2349                                "%u is corrupt\n", sb->s_id,
2350                                EXT4_SB(sb)->s_journal->j_failed_commit);
2351                         if (test_opt(sb, ERRORS_RO)) {
2352                                 printk(KERN_CRIT
2353                                        "Mounting filesystem read-only\n");
2354                                 sb->s_flags |= MS_RDONLY;
2355                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2356                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2357                         }
2358                         if (test_opt(sb, ERRORS_PANIC)) {
2359                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2360                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2361                                 ext4_commit_super(sb, es, 1);
2362                                 printk(KERN_CRIT
2363                                        "EXT4-fs (device %s): mount failed\n",
2364                                       sb->s_id);
2365                                 goto failed_mount4;
2366                         }
2367                 }
2368         } else if (journal_inum) {
2369                 if (ext4_create_journal(sb, es, journal_inum))
2370                         goto failed_mount3;
2371         } else {
2372                 if (!silent)
2373                         printk(KERN_ERR
2374                                "ext4: No journal on filesystem on %s\n",
2375                                sb->s_id);
2376                 goto failed_mount3;
2377         }
2378
2379         if (ext4_blocks_count(es) > 0xffffffffULL &&
2380             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2381                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2382                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2383                 goto failed_mount4;
2384         }
2385
2386         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2387                 jbd2_journal_set_features(sbi->s_journal,
2388                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2389                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2390         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2391                 jbd2_journal_set_features(sbi->s_journal,
2392                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2393                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2394                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2395         } else {
2396                 jbd2_journal_clear_features(sbi->s_journal,
2397                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2398                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2399         }
2400
2401         /* We have now updated the journal if required, so we can
2402          * validate the data journaling mode. */
2403         switch (test_opt(sb, DATA_FLAGS)) {
2404         case 0:
2405                 /* No mode set, assume a default based on the journal
2406                  * capabilities: ORDERED_DATA if the journal can
2407                  * cope, else JOURNAL_DATA
2408                  */
2409                 if (jbd2_journal_check_available_features
2410                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2411                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2412                 else
2413                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2414                 break;
2415
2416         case EXT4_MOUNT_ORDERED_DATA:
2417         case EXT4_MOUNT_WRITEBACK_DATA:
2418                 if (!jbd2_journal_check_available_features
2419                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2420                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2421                                "requested data journaling mode\n");
2422                         goto failed_mount4;
2423                 }
2424         default:
2425                 break;
2426         }
2427
2428         if (test_opt(sb, NOBH)) {
2429                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2430                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2431                                 "its supported only with writeback mode\n");
2432                         clear_opt(sbi->s_mount_opt, NOBH);
2433                 }
2434         }
2435         /*
2436          * The jbd2_journal_load will have done any necessary log recovery,
2437          * so we can safely mount the rest of the filesystem now.
2438          */
2439
2440         root = ext4_iget(sb, EXT4_ROOT_INO);
2441         if (IS_ERR(root)) {
2442                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2443                 ret = PTR_ERR(root);
2444                 goto failed_mount4;
2445         }
2446         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2447                 iput(root);
2448                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2449                 goto failed_mount4;
2450         }
2451         sb->s_root = d_alloc_root(root);
2452         if (!sb->s_root) {
2453                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2454                 iput(root);
2455                 ret = -ENOMEM;
2456                 goto failed_mount4;
2457         }
2458
2459         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2460
2461         /* determine the minimum size of new large inodes, if present */
2462         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2463                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2464                                                      EXT4_GOOD_OLD_INODE_SIZE;
2465                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2466                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2467                         if (sbi->s_want_extra_isize <
2468                             le16_to_cpu(es->s_want_extra_isize))
2469                                 sbi->s_want_extra_isize =
2470                                         le16_to_cpu(es->s_want_extra_isize);
2471                         if (sbi->s_want_extra_isize <
2472                             le16_to_cpu(es->s_min_extra_isize))
2473                                 sbi->s_want_extra_isize =
2474                                         le16_to_cpu(es->s_min_extra_isize);
2475                 }
2476         }
2477         /* Check if enough inode space is available */
2478         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2479                                                         sbi->s_inode_size) {
2480                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2481                                                        EXT4_GOOD_OLD_INODE_SIZE;
2482                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2483                         "available.\n");
2484         }
2485
2486         /*
2487          * akpm: core read_super() calls in here with the superblock locked.
2488          * That deadlocks, because orphan cleanup needs to lock the superblock
2489          * in numerous places.  Here we just pop the lock - it's relatively
2490          * harmless, because we are now ready to accept write_super() requests,
2491          * and aviro says that's the only reason for hanging onto the
2492          * superblock lock.
2493          */
2494         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2495         ext4_orphan_cleanup(sb, es);
2496         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2497         if (needs_recovery)
2498                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2499         ext4_mark_recovery_complete(sb, es);
2500         printk(KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2501                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2502                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2503                "writeback");
2504
2505         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2506                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2507                                 "requested data journaling mode\n");
2508                 clear_opt(sbi->s_mount_opt, DELALLOC);
2509         } else if (test_opt(sb, DELALLOC))
2510                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2511
2512         ext4_ext_init(sb);
2513         ext4_mb_init(sb, needs_recovery);
2514
2515         lock_kernel();
2516         return 0;
2517
2518 cantfind_ext4:
2519         if (!silent)
2520                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2521                        sb->s_id);
2522         goto failed_mount;
2523
2524 failed_mount4:
2525         jbd2_journal_destroy(sbi->s_journal);
2526         sbi->s_journal = NULL;
2527 failed_mount3:
2528         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2529         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2530         percpu_counter_destroy(&sbi->s_dirs_counter);
2531         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2532 failed_mount2:
2533         for (i = 0; i < db_count; i++)
2534                 brelse(sbi->s_group_desc[i]);
2535         kfree(sbi->s_group_desc);
2536 failed_mount:
2537         if (sbi->s_proc) {
2538                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
2539                 remove_proc_entry(sb->s_id, ext4_proc_root);
2540         }
2541 #ifdef CONFIG_QUOTA
2542         for (i = 0; i < MAXQUOTAS; i++)
2543                 kfree(sbi->s_qf_names[i]);
2544 #endif
2545         ext4_blkdev_remove(sbi);
2546         brelse(bh);
2547 out_fail:
2548         sb->s_fs_info = NULL;
2549         kfree(sbi);
2550         lock_kernel();
2551         return ret;
2552 }
2553
2554 /*
2555  * Setup any per-fs journal parameters now.  We'll do this both on
2556  * initial mount, once the journal has been initialised but before we've
2557  * done any recovery; and again on any subsequent remount.
2558  */
2559 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2560 {
2561         struct ext4_sb_info *sbi = EXT4_SB(sb);
2562
2563         if (sbi->s_commit_interval)
2564                 journal->j_commit_interval = sbi->s_commit_interval;
2565         /* We could also set up an ext4-specific default for the commit
2566          * interval here, but for now we'll just fall back to the jbd
2567          * default. */
2568
2569         spin_lock(&journal->j_state_lock);
2570         if (test_opt(sb, BARRIER))
2571                 journal->j_flags |= JBD2_BARRIER;
2572         else
2573                 journal->j_flags &= ~JBD2_BARRIER;
2574         spin_unlock(&journal->j_state_lock);
2575 }
2576
2577 static journal_t *ext4_get_journal(struct super_block *sb,
2578                                    unsigned int journal_inum)
2579 {
2580         struct inode *journal_inode;
2581         journal_t *journal;
2582
2583         /* First, test for the existence of a valid inode on disk.  Bad
2584          * things happen if we iget() an unused inode, as the subsequent
2585          * iput() will try to delete it. */
2586
2587         journal_inode = ext4_iget(sb, journal_inum);
2588         if (IS_ERR(journal_inode)) {
2589                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2590                 return NULL;
2591         }
2592         if (!journal_inode->i_nlink) {
2593                 make_bad_inode(journal_inode);
2594                 iput(journal_inode);
2595                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2596                 return NULL;
2597         }
2598
2599         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
2600                   journal_inode, journal_inode->i_size);
2601         if (!S_ISREG(journal_inode->i_mode)) {
2602                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2603                 iput(journal_inode);
2604                 return NULL;
2605         }
2606
2607         journal = jbd2_journal_init_inode(journal_inode);
2608         if (!journal) {
2609                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2610                 iput(journal_inode);
2611                 return NULL;
2612         }
2613         journal->j_private = sb;
2614         ext4_init_journal_params(sb, journal);
2615         return journal;
2616 }
2617
2618 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2619                                        dev_t j_dev)
2620 {
2621         struct buffer_head *bh;
2622         journal_t *journal;
2623         ext4_fsblk_t start;
2624         ext4_fsblk_t len;
2625         int hblock, blocksize;
2626         ext4_fsblk_t sb_block;
2627         unsigned long offset;
2628         struct ext4_super_block *es;
2629         struct block_device *bdev;
2630
2631         bdev = ext4_blkdev_get(j_dev);
2632         if (bdev == NULL)
2633                 return NULL;
2634
2635         if (bd_claim(bdev, sb)) {
2636                 printk(KERN_ERR
2637                         "EXT4: failed to claim external journal device.\n");
2638                 blkdev_put(bdev);
2639                 return NULL;
2640         }
2641
2642         blocksize = sb->s_blocksize;
2643         hblock = bdev_hardsect_size(bdev);
2644         if (blocksize < hblock) {
2645                 printk(KERN_ERR
2646                         "EXT4-fs: blocksize too small for journal device.\n");
2647                 goto out_bdev;
2648         }
2649
2650         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2651         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2652         set_blocksize(bdev, blocksize);
2653         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2654                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2655                        "external journal\n");
2656                 goto out_bdev;
2657         }
2658
2659         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2660         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2661             !(le32_to_cpu(es->s_feature_incompat) &
2662               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2663                 printk(KERN_ERR "EXT4-fs: external journal has "
2664                                         "bad superblock\n");
2665                 brelse(bh);
2666                 goto out_bdev;
2667         }
2668
2669         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2670                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2671                 brelse(bh);
2672                 goto out_bdev;
2673         }
2674
2675         len = ext4_blocks_count(es);
2676         start = sb_block + 1;
2677         brelse(bh);     /* we're done with the superblock */
2678
2679         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2680                                         start, len, blocksize);
2681         if (!journal) {
2682                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2683                 goto out_bdev;
2684         }
2685         journal->j_private = sb;
2686         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2687         wait_on_buffer(journal->j_sb_buffer);
2688         if (!buffer_uptodate(journal->j_sb_buffer)) {
2689                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2690                 goto out_journal;
2691         }
2692         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2693                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2694                                         "user (unsupported) - %d\n",
2695                         be32_to_cpu(journal->j_superblock->s_nr_users));
2696                 goto out_journal;
2697         }
2698         EXT4_SB(sb)->journal_bdev = bdev;
2699         ext4_init_journal_params(sb, journal);
2700         return journal;
2701 out_journal:
2702         jbd2_journal_destroy(journal);
2703 out_bdev:
2704         ext4_blkdev_put(bdev);
2705         return NULL;
2706 }
2707
2708 static int ext4_load_journal(struct super_block *sb,
2709                              struct ext4_super_block *es,
2710                              unsigned long journal_devnum)
2711 {
2712         journal_t *journal;
2713         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2714         dev_t journal_dev;
2715         int err = 0;
2716         int really_read_only;
2717
2718         if (journal_devnum &&
2719             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2720                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2721                         "numbers have changed\n");
2722                 journal_dev = new_decode_dev(journal_devnum);
2723         } else
2724                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2725
2726         really_read_only = bdev_read_only(sb->s_bdev);
2727
2728         /*
2729          * Are we loading a blank journal or performing recovery after a
2730          * crash?  For recovery, we need to check in advance whether we
2731          * can get read-write access to the device.
2732          */
2733
2734         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2735                 if (sb->s_flags & MS_RDONLY) {
2736                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2737                                         "required on readonly filesystem.\n");
2738                         if (really_read_only) {
2739                                 printk(KERN_ERR "EXT4-fs: write access "
2740                                         "unavailable, cannot proceed.\n");
2741                                 return -EROFS;
2742                         }
2743                         printk(KERN_INFO "EXT4-fs: write access will "
2744                                "be enabled during recovery.\n");
2745                 }
2746         }
2747
2748         if (journal_inum && journal_dev) {
2749                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2750                        "and inode journals!\n");
2751                 return -EINVAL;
2752         }
2753
2754         if (journal_inum) {
2755                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2756                         return -EINVAL;
2757         } else {
2758                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2759                         return -EINVAL;
2760         }
2761
2762         if (journal->j_flags & JBD2_BARRIER)
2763                 printk(KERN_INFO "EXT4-fs: barriers enabled\n");
2764         else
2765                 printk(KERN_INFO "EXT4-fs: barriers disabled\n");
2766
2767         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2768                 err = jbd2_journal_update_format(journal);
2769                 if (err)  {
2770                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2771                         jbd2_journal_destroy(journal);
2772                         return err;
2773                 }
2774         }
2775
2776         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2777                 err = jbd2_journal_wipe(journal, !really_read_only);
2778         if (!err)
2779                 err = jbd2_journal_load(journal);
2780
2781         if (err) {
2782                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2783                 jbd2_journal_destroy(journal);
2784                 return err;
2785         }
2786
2787         EXT4_SB(sb)->s_journal = journal;
2788         ext4_clear_journal_err(sb, es);
2789
2790         if (journal_devnum &&
2791             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2792                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2793                 sb->s_dirt = 1;
2794
2795                 /* Make sure we flush the recovery flag to disk. */
2796                 ext4_commit_super(sb, es, 1);
2797         }
2798
2799         return 0;
2800 }
2801
2802 static int ext4_create_journal(struct super_block *sb,
2803                                struct ext4_super_block *es,
2804                                unsigned int journal_inum)
2805 {
2806         journal_t *journal;
2807         int err;
2808
2809         if (sb->s_flags & MS_RDONLY) {
2810                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2811                                 "create journal.\n");
2812                 return -EROFS;
2813         }
2814
2815         journal = ext4_get_journal(sb, journal_inum);
2816         if (!journal)
2817                 return -EINVAL;
2818
2819         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2820                journal_inum);
2821
2822         err = jbd2_journal_create(journal);
2823         if (err) {
2824                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2825                 jbd2_journal_destroy(journal);
2826                 return -EIO;
2827         }
2828
2829         EXT4_SB(sb)->s_journal = journal;
2830
2831         ext4_update_dynamic_rev(sb);
2832         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2833         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2834
2835         es->s_journal_inum = cpu_to_le32(journal_inum);
2836         sb->s_dirt = 1;
2837
2838         /* Make sure we flush the recovery flag to disk. */
2839         ext4_commit_super(sb, es, 1);
2840
2841         return 0;
2842 }
2843
2844 static void ext4_commit_super(struct super_block *sb,
2845                               struct ext4_super_block *es, int sync)
2846 {
2847         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2848
2849         if (!sbh)
2850                 return;
2851         if (buffer_write_io_error(sbh)) {
2852                 /*
2853                  * Oh, dear.  A previous attempt to write the
2854                  * superblock failed.  This could happen because the
2855                  * USB device was yanked out.  Or it could happen to
2856                  * be a transient write error and maybe the block will
2857                  * be remapped.  Nothing we can do but to retry the
2858                  * write and hope for the best.
2859                  */
2860                 printk(KERN_ERR "ext4: previous I/O error to "
2861                        "superblock detected for %s.\n", sb->s_id);
2862                 clear_buffer_write_io_error(sbh);
2863                 set_buffer_uptodate(sbh);
2864         }
2865         es->s_wtime = cpu_to_le32(get_seconds());
2866         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2867         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2868         BUFFER_TRACE(sbh, "marking dirty");
2869         mark_buffer_dirty(sbh);
2870         if (sync) {
2871                 sync_dirty_buffer(sbh);
2872                 if (buffer_write_io_error(sbh)) {
2873                         printk(KERN_ERR "ext4: I/O error while writing "
2874                                "superblock for %s.\n", sb->s_id);
2875                         clear_buffer_write_io_error(sbh);
2876                         set_buffer_uptodate(sbh);
2877                 }
2878         }
2879 }
2880
2881
2882 /*
2883  * Have we just finished recovery?  If so, and if we are mounting (or
2884  * remounting) the filesystem readonly, then we will end up with a
2885  * consistent fs on disk.  Record that fact.
2886  */
2887 static void ext4_mark_recovery_complete(struct super_block *sb,
2888                                         struct ext4_super_block *es)
2889 {
2890         journal_t *journal = EXT4_SB(sb)->s_journal;
2891
2892         jbd2_journal_lock_updates(journal);
2893         jbd2_journal_flush(journal);
2894         lock_super(sb);
2895         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2896             sb->s_flags & MS_RDONLY) {
2897                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2898                 sb->s_dirt = 0;
2899                 ext4_commit_super(sb, es, 1);
2900         }
2901         unlock_super(sb);
2902         jbd2_journal_unlock_updates(journal);
2903 }
2904
2905 /*
2906  * If we are mounting (or read-write remounting) a filesystem whose journal
2907  * has recorded an error from a previous lifetime, move that error to the
2908  * main filesystem now.
2909  */
2910 static void ext4_clear_journal_err(struct super_block *sb,
2911                                    struct ext4_super_block *es)
2912 {
2913         journal_t *journal;
2914         int j_errno;
2915         const char *errstr;
2916
2917         journal = EXT4_SB(sb)->s_journal;
2918
2919         /*
2920          * Now check for any error status which may have been recorded in the
2921          * journal by a prior ext4_error() or ext4_abort()
2922          */
2923
2924         j_errno = jbd2_journal_errno(journal);
2925         if (j_errno) {
2926                 char nbuf[16];
2927
2928                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2929                 ext4_warning(sb, __func__, "Filesystem error recorded "
2930                              "from previous mount: %s", errstr);
2931                 ext4_warning(sb, __func__, "Marking fs in need of "
2932                              "filesystem check.");
2933
2934                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2935                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2936                 ext4_commit_super(sb, es, 1);
2937
2938                 jbd2_journal_clear_err(journal);
2939         }
2940 }
2941
2942 /*
2943  * Force the running and committing transactions to commit,
2944  * and wait on the commit.
2945  */
2946 int ext4_force_commit(struct super_block *sb)
2947 {
2948         journal_t *journal;
2949         int ret;
2950
2951         if (sb->s_flags & MS_RDONLY)
2952                 return 0;
2953
2954         journal = EXT4_SB(sb)->s_journal;
2955         sb->s_dirt = 0;
2956         ret = ext4_journal_force_commit(journal);
2957         return ret;
2958 }
2959
2960 /*
2961  * Ext4 always journals updates to the superblock itself, so we don't
2962  * have to propagate any other updates to the superblock on disk at this
2963  * point.  Just start an async writeback to get the buffers on their way
2964  * to the disk.
2965  *
2966  * This implicitly triggers the writebehind on sync().
2967  */
2968
2969 static void ext4_write_super(struct super_block *sb)
2970 {
2971         if (mutex_trylock(&sb->s_lock) != 0)
2972                 BUG();
2973         sb->s_dirt = 0;
2974 }
2975
2976 static int ext4_sync_fs(struct super_block *sb, int wait)
2977 {
2978         tid_t target;
2979
2980         sb->s_dirt = 0;
2981         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2982                 if (wait)
2983                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2984         }
2985         return 0;
2986 }
2987
2988 /*
2989  * LVM calls this function before a (read-only) snapshot is created.  This
2990  * gives us a chance to flush the journal completely and mark the fs clean.
2991  */
2992 static void ext4_write_super_lockfs(struct super_block *sb)
2993 {
2994         sb->s_dirt = 0;
2995
2996         if (!(sb->s_flags & MS_RDONLY)) {
2997                 journal_t *journal = EXT4_SB(sb)->s_journal;
2998
2999                 /* Now we set up the journal barrier. */
3000                 jbd2_journal_lock_updates(journal);
3001                 jbd2_journal_flush(journal);
3002
3003                 /* Journal blocked and flushed, clear needs_recovery flag. */
3004                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3005                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
3006         }
3007 }
3008
3009 /*
3010  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3011  * flag here, even though the filesystem is not technically dirty yet.
3012  */
3013 static void ext4_unlockfs(struct super_block *sb)
3014 {
3015         if (!(sb->s_flags & MS_RDONLY)) {
3016                 lock_super(sb);
3017                 /* Reser the needs_recovery flag before the fs is unlocked. */
3018                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3019                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
3020                 unlock_super(sb);
3021                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3022         }
3023 }
3024
3025 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3026 {
3027         struct ext4_super_block *es;
3028         struct ext4_sb_info *sbi = EXT4_SB(sb);
3029         ext4_fsblk_t n_blocks_count = 0;
3030         unsigned long old_sb_flags;
3031         struct ext4_mount_options old_opts;
3032         ext4_group_t g;
3033         int err;
3034 #ifdef CONFIG_QUOTA
3035         int i;
3036 #endif
3037
3038         /* Store the original options */
3039         old_sb_flags = sb->s_flags;
3040         old_opts.s_mount_opt = sbi->s_mount_opt;
3041         old_opts.s_resuid = sbi->s_resuid;
3042         old_opts.s_resgid = sbi->s_resgid;
3043         old_opts.s_commit_interval = sbi->s_commit_interval;
3044 #ifdef CONFIG_QUOTA
3045         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3046         for (i = 0; i < MAXQUOTAS; i++)
3047                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3048 #endif
3049
3050         /*
3051          * Allow the "check" option to be passed as a remount option.
3052          */
3053         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
3054                 err = -EINVAL;
3055                 goto restore_opts;
3056         }
3057
3058         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
3059                 ext4_abort(sb, __func__, "Abort forced by user");
3060
3061         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3062                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3063
3064         es = sbi->s_es;
3065
3066         ext4_init_journal_params(sb, sbi->s_journal);
3067
3068         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3069                 n_blocks_count > ext4_blocks_count(es)) {
3070                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3071                         err = -EROFS;
3072                         goto restore_opts;
3073                 }
3074
3075                 if (*flags & MS_RDONLY) {
3076                         /*
3077                          * First of all, the unconditional stuff we have to do
3078                          * to disable replay of the journal when we next remount
3079                          */
3080                         sb->s_flags |= MS_RDONLY;
3081
3082                         /*
3083                          * OK, test if we are remounting a valid rw partition
3084                          * readonly, and if so set the rdonly flag and then
3085                          * mark the partition as valid again.
3086                          */
3087                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3088                             (sbi->s_mount_state & EXT4_VALID_FS))
3089                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3090
3091                         /*
3092                          * We have to unlock super so that we can wait for
3093                          * transactions.
3094                          */
3095                         unlock_super(sb);
3096                         ext4_mark_recovery_complete(sb, es);
3097                         lock_super(sb);
3098                 } else {
3099                         __le32 ret;
3100                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3101                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3102                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3103                                        "remount RDWR because of unsupported "
3104                                        "optional features (%x).\n",
3105                                        sb->s_id, le32_to_cpu(ret));
3106                                 err = -EROFS;
3107                                 goto restore_opts;
3108                         }
3109
3110                         /*
3111                          * Make sure the group descriptor checksums
3112                          * are sane.  If they aren't, refuse to
3113                          * remount r/w.
3114                          */
3115                         for (g = 0; g < sbi->s_groups_count; g++) {
3116                                 struct ext4_group_desc *gdp =
3117                                         ext4_get_group_desc(sb, g, NULL);
3118
3119                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3120                                         printk(KERN_ERR
3121                "EXT4-fs: ext4_remount: "
3122                 "Checksum for group %lu failed (%u!=%u)\n",
3123                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3124                                                le16_to_cpu(gdp->bg_checksum));
3125                                         err = -EINVAL;
3126                                         goto restore_opts;
3127                                 }
3128                         }
3129
3130                         /*
3131                          * If we have an unprocessed orphan list hanging
3132                          * around from a previously readonly bdev mount,
3133                          * require a full umount/remount for now.
3134                          */
3135                         if (es->s_last_orphan) {
3136                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3137                                        "remount RDWR because of unprocessed "
3138                                        "orphan inode list.  Please "
3139                                        "umount/remount instead.\n",
3140                                        sb->s_id);
3141                                 err = -EINVAL;
3142                                 goto restore_opts;
3143                         }
3144
3145                         /*
3146                          * Mounting a RDONLY partition read-write, so reread
3147                          * and store the current valid flag.  (It may have
3148                          * been changed by e2fsck since we originally mounted
3149                          * the partition.)
3150                          */
3151                         ext4_clear_journal_err(sb, es);
3152                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3153                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3154                                 goto restore_opts;
3155                         if (!ext4_setup_super(sb, es, 0))
3156                                 sb->s_flags &= ~MS_RDONLY;
3157                 }
3158         }
3159 #ifdef CONFIG_QUOTA
3160         /* Release old quota file names */
3161         for (i = 0; i < MAXQUOTAS; i++)
3162                 if (old_opts.s_qf_names[i] &&
3163                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3164                         kfree(old_opts.s_qf_names[i]);
3165 #endif
3166         return 0;
3167 restore_opts:
3168         sb->s_flags = old_sb_flags;
3169         sbi->s_mount_opt = old_opts.s_mount_opt;
3170         sbi->s_resuid = old_opts.s_resuid;
3171         sbi->s_resgid = old_opts.s_resgid;
3172         sbi->s_commit_interval = old_opts.s_commit_interval;
3173 #ifdef CONFIG_QUOTA
3174         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3175         for (i = 0; i < MAXQUOTAS; i++) {
3176                 if (sbi->s_qf_names[i] &&
3177                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3178                         kfree(sbi->s_qf_names[i]);
3179                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3180         }
3181 #endif
3182         return err;
3183 }
3184
3185 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3186 {
3187         struct super_block *sb = dentry->d_sb;
3188         struct ext4_sb_info *sbi = EXT4_SB(sb);
3189         struct ext4_super_block *es = sbi->s_es;
3190         u64 fsid;
3191
3192         if (test_opt(sb, MINIX_DF)) {
3193                 sbi->s_overhead_last = 0;
3194         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3195                 ext4_group_t ngroups = sbi->s_groups_count, i;
3196                 ext4_fsblk_t overhead = 0;
3197                 smp_rmb();
3198
3199                 /*
3200                  * Compute the overhead (FS structures).  This is constant
3201                  * for a given filesystem unless the number of block groups
3202                  * changes so we cache the previous value until it does.
3203                  */
3204
3205                 /*
3206                  * All of the blocks before first_data_block are
3207                  * overhead
3208                  */
3209                 overhead = le32_to_cpu(es->s_first_data_block);
3210
3211                 /*
3212                  * Add the overhead attributed to the superblock and
3213                  * block group descriptors.  If the sparse superblocks
3214                  * feature is turned on, then not all groups have this.
3215                  */
3216                 for (i = 0; i < ngroups; i++) {
3217                         overhead += ext4_bg_has_super(sb, i) +
3218                                 ext4_bg_num_gdb(sb, i);
3219                         cond_resched();
3220                 }
3221
3222                 /*
3223                  * Every block group has an inode bitmap, a block
3224                  * bitmap, and an inode table.
3225                  */
3226                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3227                 sbi->s_overhead_last = overhead;
3228                 smp_wmb();
3229                 sbi->s_blocks_last = ext4_blocks_count(es);
3230         }
3231
3232         buf->f_type = EXT4_SUPER_MAGIC;
3233         buf->f_bsize = sb->s_blocksize;
3234         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3235         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3236                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3237         ext4_free_blocks_count_set(es, buf->f_bfree);
3238         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3239         if (buf->f_bfree < ext4_r_blocks_count(es))
3240                 buf->f_bavail = 0;
3241         buf->f_files = le32_to_cpu(es->s_inodes_count);
3242         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3243         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3244         buf->f_namelen = EXT4_NAME_LEN;
3245         fsid = le64_to_cpup((void *)es->s_uuid) ^
3246                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3247         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3248         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3249         return 0;
3250 }
3251
3252 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3253  * is locked for write. Otherwise the are possible deadlocks:
3254  * Process 1                         Process 2
3255  * ext4_create()                     quota_sync()
3256  *   jbd2_journal_start()                   write_dquot()
3257  *   DQUOT_INIT()                        down(dqio_mutex)
3258  *     down(dqio_mutex)                    jbd2_journal_start()
3259  *
3260  */
3261
3262 #ifdef CONFIG_QUOTA
3263
3264 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3265 {
3266         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3267 }
3268
3269 static int ext4_dquot_initialize(struct inode *inode, int type)
3270 {
3271         handle_t *handle;
3272         int ret, err;
3273
3274         /* We may create quota structure so we need to reserve enough blocks */
3275         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3276         if (IS_ERR(handle))
3277                 return PTR_ERR(handle);
3278         ret = dquot_initialize(inode, type);
3279         err = ext4_journal_stop(handle);
3280         if (!ret)
3281                 ret = err;
3282         return ret;
3283 }
3284
3285 static int ext4_dquot_drop(struct inode *inode)
3286 {
3287         handle_t *handle;
3288         int ret, err;
3289
3290         /* We may delete quota structure so we need to reserve enough blocks */
3291         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3292         if (IS_ERR(handle)) {
3293                 /*
3294                  * We call dquot_drop() anyway to at least release references
3295                  * to quota structures so that umount does not hang.
3296                  */
3297                 dquot_drop(inode);
3298                 return PTR_ERR(handle);
3299         }
3300         ret = dquot_drop(inode);
3301         err = ext4_journal_stop(handle);
3302         if (!ret)
3303                 ret = err;
3304         return ret;
3305 }
3306
3307 static int ext4_write_dquot(struct dquot *dquot)
3308 {
3309         int ret, err;
3310         handle_t *handle;
3311         struct inode *inode;
3312
3313         inode = dquot_to_inode(dquot);
3314         handle = ext4_journal_start(inode,
3315                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3316         if (IS_ERR(handle))
3317                 return PTR_ERR(handle);
3318         ret = dquot_commit(dquot);
3319         err = ext4_journal_stop(handle);
3320         if (!ret)
3321                 ret = err;
3322         return ret;
3323 }
3324
3325 static int ext4_acquire_dquot(struct dquot *dquot)
3326 {
3327         int ret, err;
3328         handle_t *handle;
3329
3330         handle = ext4_journal_start(dquot_to_inode(dquot),
3331                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3332         if (IS_ERR(handle))
3333                 return PTR_ERR(handle);
3334         ret = dquot_acquire(dquot);
3335         err = ext4_journal_stop(handle);
3336         if (!ret)
3337                 ret = err;
3338         return ret;
3339 }
3340
3341 static int ext4_release_dquot(struct dquot *dquot)
3342 {
3343         int ret, err;
3344         handle_t *handle;
3345
3346         handle = ext4_journal_start(dquot_to_inode(dquot),
3347                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3348         if (IS_ERR(handle)) {
3349                 /* Release dquot anyway to avoid endless cycle in dqput() */
3350                 dquot_release(dquot);
3351                 return PTR_ERR(handle);
3352         }
3353         ret = dquot_release(dquot);
3354         err = ext4_journal_stop(handle);
3355         if (!ret)
3356                 ret = err;
3357         return ret;
3358 }
3359
3360 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3361 {
3362         /* Are we journaling quotas? */
3363         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3364             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3365                 dquot_mark_dquot_dirty(dquot);
3366                 return ext4_write_dquot(dquot);
3367         } else {
3368                 return dquot_mark_dquot_dirty(dquot);
3369         }
3370 }
3371
3372 static int ext4_write_info(struct super_block *sb, int type)
3373 {
3374         int ret, err;
3375         handle_t *handle;
3376
3377         /* Data block + inode block */
3378         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3379         if (IS_ERR(handle))
3380                 return PTR_ERR(handle);
3381         ret = dquot_commit_info(sb, type);
3382         err = ext4_journal_stop(handle);
3383         if (!ret)
3384                 ret = err;
3385         return ret;
3386 }
3387
3388 /*
3389  * Turn on quotas during mount time - we need to find
3390  * the quota file and such...
3391  */
3392 static int ext4_quota_on_mount(struct super_block *sb, int type)
3393 {
3394         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3395                         EXT4_SB(sb)->s_jquota_fmt, type);
3396 }
3397
3398 /*
3399  * Standard function to be called on quota_on
3400  */
3401 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3402                          char *path, int remount)
3403 {
3404         int err;
3405         struct nameidata nd;
3406
3407         if (!test_opt(sb, QUOTA))
3408                 return -EINVAL;
3409         /* When remounting, no checks are needed and in fact, path is NULL */
3410         if (remount)
3411                 return vfs_quota_on(sb, type, format_id, path, remount);
3412
3413         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3414         if (err)
3415                 return err;
3416
3417         /* Quotafile not on the same filesystem? */
3418         if (nd.path.mnt->mnt_sb != sb) {
3419                 path_put(&nd.path);
3420                 return -EXDEV;
3421         }
3422         /* Journaling quota? */
3423         if (EXT4_SB(sb)->s_qf_names[type]) {
3424                 /* Quotafile not in fs root? */
3425                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3426                         printk(KERN_WARNING
3427                                 "EXT4-fs: Quota file not on filesystem root. "
3428                                 "Journaled quota will not work.\n");
3429         }
3430
3431         /*
3432          * When we journal data on quota file, we have to flush journal to see
3433          * all updates to the file when we bypass pagecache...
3434          */
3435         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3436                 /*
3437                  * We don't need to lock updates but journal_flush() could
3438                  * otherwise be livelocked...
3439                  */
3440                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3441                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3442                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3443         }
3444
3445         err = vfs_quota_on_path(sb, type, format_id, &nd.path);
3446         path_put(&nd.path);
3447         return err;
3448 }
3449
3450 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3451  * acquiring the locks... As quota files are never truncated and quota code
3452  * itself serializes the operations (and noone else should touch the files)
3453  * we don't have to be afraid of races */
3454 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3455                                size_t len, loff_t off)
3456 {
3457         struct inode *inode = sb_dqopt(sb)->files[type];
3458         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3459         int err = 0;
3460         int offset = off & (sb->s_blocksize - 1);
3461         int tocopy;
3462         size_t toread;
3463         struct buffer_head *bh;
3464         loff_t i_size = i_size_read(inode);
3465
3466         if (off > i_size)
3467                 return 0;
3468         if (off+len > i_size)
3469                 len = i_size-off;
3470         toread = len;
3471         while (toread > 0) {
3472                 tocopy = sb->s_blocksize - offset < toread ?
3473                                 sb->s_blocksize - offset : toread;
3474                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3475                 if (err)
3476                         return err;
3477                 if (!bh)        /* A hole? */
3478                         memset(data, 0, tocopy);
3479                 else
3480                         memcpy(data, bh->b_data+offset, tocopy);
3481                 brelse(bh);
3482                 offset = 0;
3483                 toread -= tocopy;
3484                 data += tocopy;
3485                 blk++;
3486         }
3487         return len;
3488 }
3489
3490 /* Write to quotafile (we know the transaction is already started and has
3491  * enough credits) */
3492 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3493                                 const char *data, size_t len, loff_t off)
3494 {
3495         struct inode *inode = sb_dqopt(sb)->files[type];
3496         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3497         int err = 0;
3498         int offset = off & (sb->s_blocksize - 1);
3499         int tocopy;
3500         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3501         size_t towrite = len;
3502         struct buffer_head *bh;
3503         handle_t *handle = journal_current_handle();
3504
3505         if (!handle) {
3506                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%llu, len=%llu)"
3507                         " cancelled because transaction is not started.\n",
3508                         (unsigned long long)off, (unsigned long long)len);
3509                 return -EIO;
3510         }
3511         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3512         while (towrite > 0) {
3513                 tocopy = sb->s_blocksize - offset < towrite ?
3514                                 sb->s_blocksize - offset : towrite;
3515                 bh = ext4_bread(handle, inode, blk, 1, &err);
3516                 if (!bh)
3517                         goto out;
3518                 if (journal_quota) {
3519                         err = ext4_journal_get_write_access(handle, bh);
3520                         if (err) {
3521                                 brelse(bh);
3522                                 goto out;
3523                         }
3524                 }
3525                 lock_buffer(bh);
3526                 memcpy(bh->b_data+offset, data, tocopy);
3527                 flush_dcache_page(bh->b_page);
3528                 unlock_buffer(bh);
3529                 if (journal_quota)
3530                         err = ext4_journal_dirty_metadata(handle, bh);
3531                 else {
3532                         /* Always do at least ordered writes for quotas */
3533                         err = ext4_jbd2_file_inode(handle, inode);
3534                         mark_buffer_dirty(bh);
3535                 }
3536                 brelse(bh);
3537                 if (err)
3538                         goto out;
3539                 offset = 0;
3540                 towrite -= tocopy;
3541                 data += tocopy;
3542                 blk++;
3543         }
3544 out:
3545         if (len == towrite) {
3546                 mutex_unlock(&inode->i_mutex);
3547                 return err;
3548         }
3549         if (inode->i_size < off+len-towrite) {
3550                 i_size_write(inode, off+len-towrite);
3551                 EXT4_I(inode)->i_disksize = inode->i_size;
3552         }
3553         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3554         ext4_mark_inode_dirty(handle, inode);
3555         mutex_unlock(&inode->i_mutex);
3556         return len - towrite;
3557 }
3558
3559 #endif
3560
3561 static int ext4_get_sb(struct file_system_type *fs_type,
3562         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3563 {
3564         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3565 }
3566
3567 #ifdef CONFIG_PROC_FS
3568 static int ext4_ui_proc_show(struct seq_file *m, void *v)
3569 {
3570         unsigned int *p = m->private;
3571
3572         seq_printf(m, "%u\n", *p);
3573         return 0;
3574 }
3575
3576 static int ext4_ui_proc_open(struct inode *inode, struct file *file)
3577 {
3578         return single_open(file, ext4_ui_proc_show, PDE(inode)->data);
3579 }
3580
3581 static ssize_t ext4_ui_proc_write(struct file *file, const char __user *buf,
3582                                size_t cnt, loff_t *ppos)
3583 {
3584         unsigned int *p = PDE(file->f_path.dentry->d_inode)->data;
3585         char str[32];
3586         unsigned long value;
3587
3588         if (cnt >= sizeof(str))
3589                 return -EINVAL;
3590         if (copy_from_user(str, buf, cnt))
3591                 return -EFAULT;
3592         value = simple_strtol(str, NULL, 0);
3593         if (value < 0)
3594                 return -ERANGE;
3595         *p = value;
3596         return cnt;
3597 }
3598
3599 const struct file_operations ext4_ui_proc_fops = {
3600         .owner          = THIS_MODULE,
3601         .open           = ext4_ui_proc_open,
3602         .read           = seq_read,
3603         .llseek         = seq_lseek,
3604         .release        = single_release,
3605         .write          = ext4_ui_proc_write,
3606 };
3607 #endif
3608
3609 static struct file_system_type ext4dev_fs_type = {
3610         .owner          = THIS_MODULE,
3611         .name           = "ext4dev",
3612         .get_sb         = ext4_get_sb,
3613         .kill_sb        = kill_block_super,
3614         .fs_flags       = FS_REQUIRES_DEV,
3615 };
3616
3617 static int __init init_ext4_fs(void)
3618 {
3619         int err;
3620
3621         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
3622         err = init_ext4_mballoc();
3623         if (err)
3624                 return err;
3625
3626         err = init_ext4_xattr();
3627         if (err)
3628                 goto out2;
3629         err = init_inodecache();
3630         if (err)
3631                 goto out1;
3632         err = register_filesystem(&ext4dev_fs_type);
3633         if (err)
3634                 goto out;
3635         return 0;
3636 out:
3637         destroy_inodecache();
3638 out1:
3639         exit_ext4_xattr();
3640 out2:
3641         exit_ext4_mballoc();
3642         return err;
3643 }
3644
3645 static void __exit exit_ext4_fs(void)
3646 {
3647         unregister_filesystem(&ext4dev_fs_type);
3648         destroy_inodecache();
3649         exit_ext4_xattr();
3650         exit_ext4_mballoc();
3651         remove_proc_entry("fs/ext4", NULL);
3652 }
3653
3654 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3655 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3656 MODULE_LICENSE("GPL");
3657 module_init(init_ext4_fs)
3658 module_exit(exit_ext4_fs)