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