fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <asm/uaccess.h>
  41
  42 #include "ext4.h"
  43 #include "ext4_jbd2.h"
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 struct proc_dir_entry *ext4_proc_root;
  50
  51 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  52                              unsigned long journal_devnum);
  53 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  54                                unsigned int);
  55 static void ext4_commit_super(struct super_block *sb,
  56                               struct ext4_super_block *es, int sync);
  57 static void ext4_mark_recovery_complete(struct super_block *sb,
  58                                         struct ext4_super_block *es);
  59 static void ext4_clear_journal_err(struct super_block *sb,
  60                                    struct ext4_super_block *es);
  61 static int ext4_sync_fs(struct super_block *sb, int wait);
  62 static const char *ext4_decode_error(struct super_block *sb, int errno,
  63                                      char nbuf[16]);
  64 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  65 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  66 static void ext4_unlockfs(struct super_block *sb);
  67 static void ext4_write_super(struct super_block *sb);
  68 static void ext4_write_super_lockfs(struct super_block *sb);
  69
  70
  71 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  72                                struct ext4_group_desc *bg)
  73 {
  74         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  75                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  76                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  77 }
  78
  79 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  80                                struct ext4_group_desc *bg)
  81 {
  82         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  83                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  84                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  85 }
  86
  87 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  88                               struct ext4_group_desc *bg)
  89 {
  90         return le32_to_cpu(bg->bg_inode_table_lo) |
  91                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  92                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  93 }
  94
  95 void ext4_block_bitmap_set(struct super_block *sb,
  96                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  97 {
  98         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  99         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 100                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 101 }
 102
 103 void ext4_inode_bitmap_set(struct super_block *sb,
 104                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 105 {
 106         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 107         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 108                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 109 }
 110
 111 void ext4_inode_table_set(struct super_block *sb,
 112                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 113 {
 114         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 115         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 116                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 117 }
 118
 119 /*
 120  * Wrappers for jbd2_journal_start/end.
 121  *
 122  * The only special thing we need to do here is to make sure that all
 123  * journal_end calls result in the superblock being marked dirty, so
 124  * that sync() will call the filesystem's write_super callback if
 125  * appropriate.
 126  */
 127 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 128 {
 129         journal_t *journal;
 130
 131         if (sb->s_flags & MS_RDONLY)
 132                 return ERR_PTR(-EROFS);
 133
 134         /* Special case here: if the journal has aborted behind our
 135          * backs (eg. EIO in the commit thread), then we still need to
 136          * take the FS itself readonly cleanly. */
 137         journal = EXT4_SB(sb)->s_journal;
 138         if (is_journal_aborted(journal)) {
 139                 ext4_abort(sb, __func__,
 140                            "Detected aborted journal");
 141                 return ERR_PTR(-EROFS);
 142         }
 143
 144         return jbd2_journal_start(journal, nblocks);
 145 }
 146
 147 /*
 148  * The only special thing we need to do here is to make sure that all
 149  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 150  * that sync() will call the filesystem's write_super callback if
 151  * appropriate.
 152  */
 153 int __ext4_journal_stop(const char *where, handle_t *handle)
 154 {
 155         struct super_block *sb;
 156         int err;
 157         int rc;
 158
 159         sb = handle->h_transaction->t_journal->j_private;
 160         err = handle->h_err;
 161         rc = jbd2_journal_stop(handle);
 162
 163         if (!err)
 164                 err = rc;
 165         if (err)
 166                 __ext4_std_error(sb, where, err);
 167         return err;
 168 }
 169
 170 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 171                 struct buffer_head *bh, handle_t *handle, int err)
 172 {
 173         char nbuf[16];
 174         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 175
 176         if (bh)
 177                 BUFFER_TRACE(bh, "abort");
 178
 179         if (!handle->h_err)
 180                 handle->h_err = err;
 181
 182         if (is_handle_aborted(handle))
 183                 return;
 184
 185         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 186                caller, errstr, err_fn);
 187
 188         jbd2_journal_abort_handle(handle);
 189 }
 190
 191 /* Deal with the reporting of failure conditions on a filesystem such as
 192  * inconsistencies detected or read IO failures.
 193  *
 194  * On ext2, we can store the error state of the filesystem in the
 195  * superblock.  That is not possible on ext4, because we may have other
 196  * write ordering constraints on the superblock which prevent us from
 197  * writing it out straight away; and given that the journal is about to
 198  * be aborted, we can't rely on the current, or future, transactions to
 199  * write out the superblock safely.
 200  *
 201  * We'll just use the jbd2_journal_abort() error code to record an error in
 202  * the journal instead.  On recovery, the journal will compain about
 203  * that error until we've noted it down and cleared it.
 204  */
 205
 206 static void ext4_handle_error(struct super_block *sb)
 207 {
 208         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 209
 210         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 211         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 212
 213         if (sb->s_flags & MS_RDONLY)
 214                 return;
 215
 216         if (!test_opt(sb, ERRORS_CONT)) {
 217                 journal_t *journal = EXT4_SB(sb)->s_journal;
 218
 219                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 220                 if (journal)
 221                         jbd2_journal_abort(journal, -EIO);
 222         }
 223         if (test_opt(sb, ERRORS_RO)) {
 224                 printk(KERN_CRIT "Remounting filesystem read-only\n");
 225                 sb->s_flags |= MS_RDONLY;
 226         }
 227         ext4_commit_super(sb, es, 1);
 228         if (test_opt(sb, ERRORS_PANIC))
 229                 panic("EXT4-fs (device %s): panic forced after error\n",
 230                         sb->s_id);
 231 }
 232
 233 void ext4_error(struct super_block *sb, const char *function,
 234                 const char *fmt, ...)
 235 {
 236         va_list args;
 237
 238         va_start(args, fmt);
 239         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 240         vprintk(fmt, args);
 241         printk("\n");
 242         va_end(args);
 243
 244         ext4_handle_error(sb);
 245 }
 246
 247 static const char *ext4_decode_error(struct super_block *sb, int errno,
 248                                      char nbuf[16])
 249 {
 250         char *errstr = NULL;
 251
 252         switch (errno) {
 253         case -EIO:
 254                 errstr = "IO failure";
 255                 break;
 256         case -ENOMEM:
 257                 errstr = "Out of memory";
 258                 break;
 259         case -EROFS:
 260                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 261                         errstr = "Journal has aborted";
 262                 else
 263                         errstr = "Readonly filesystem";
 264                 break;
 265         default:
 266                 /* If the caller passed in an extra buffer for unknown
 267                  * errors, textualise them now.  Else we just return
 268                  * NULL. */
 269                 if (nbuf) {
 270                         /* Check for truncated error codes... */
 271                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 272                                 errstr = nbuf;
 273                 }
 274                 break;
 275         }
 276
 277         return errstr;
 278 }
 279
 280 /* __ext4_std_error decodes expected errors from journaling functions
 281  * automatically and invokes the appropriate error response.  */
 282
 283 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 284 {
 285         char nbuf[16];
 286         const char *errstr;
 287
 288         /* Special case: if the error is EROFS, and we're not already
 289          * inside a transaction, then there's really no point in logging
 290          * an error. */
 291         if (errno == -EROFS && journal_current_handle() == NULL &&
 292             (sb->s_flags & MS_RDONLY))
 293                 return;
 294
 295         errstr = ext4_decode_error(sb, errno, nbuf);
 296         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 297                sb->s_id, function, errstr);
 298
 299         ext4_handle_error(sb);
 300 }
 301
 302 /*
 303  * ext4_abort is a much stronger failure handler than ext4_error.  The
 304  * abort function may be used to deal with unrecoverable failures such
 305  * as journal IO errors or ENOMEM at a critical moment in log management.
 306  *
 307  * We unconditionally force the filesystem into an ABORT|READONLY state,
 308  * unless the error response on the fs has been set to panic in which
 309  * case we take the easy way out and panic immediately.
 310  */
 311
 312 void ext4_abort(struct super_block *sb, const char *function,
 313                 const char *fmt, ...)
 314 {
 315         va_list args;
 316
 317         printk(KERN_CRIT "ext4_abort called.\n");
 318
 319         va_start(args, fmt);
 320         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 321         vprintk(fmt, args);
 322         printk("\n");
 323         va_end(args);
 324
 325         if (test_opt(sb, ERRORS_PANIC))
 326                 panic("EXT4-fs panic from previous error\n");
 327
 328         if (sb->s_flags & MS_RDONLY)
 329                 return;
 330
 331         printk(KERN_CRIT "Remounting filesystem read-only\n");
 332         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 333         sb->s_flags |= MS_RDONLY;
 334         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 335         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 336 }
 337
 338 void ext4_warning(struct super_block *sb, const char *function,
 339                   const char *fmt, ...)
 340 {
 341         va_list args;
 342
 343         va_start(args, fmt);
 344         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 345                sb->s_id, function);
 346         vprintk(fmt, args);
 347         printk("\n");
 348         va_end(args);
 349 }
 350
 351 void ext4_update_dynamic_rev(struct super_block *sb)
 352 {
 353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 354
 355         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 356                 return;
 357
 358         ext4_warning(sb, __func__,
 359                      "updating to rev %d because of new feature flag, "
 360                      "running e2fsck is recommended",
 361                      EXT4_DYNAMIC_REV);
 362
 363         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 364         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 365         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 366         /* leave es->s_feature_*compat flags alone */
 367         /* es->s_uuid will be set by e2fsck if empty */
 368
 369         /*
 370          * The rest of the superblock fields should be zero, and if not it
 371          * means they are likely already in use, so leave them alone.  We
 372          * can leave it up to e2fsck to clean up any inconsistencies there.
 373          */
 374 }
 375
 376 int ext4_update_compat_feature(handle_t *handle,
 377                                         struct super_block *sb, __u32 compat)
 378 {
 379         int err = 0;
 380         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 381                 err = ext4_journal_get_write_access(handle,
 382                                 EXT4_SB(sb)->s_sbh);
 383                 if (err)
 384                         return err;
 385                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 386                 sb->s_dirt = 1;
 387                 handle->h_sync = 1;
 388                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 389                                         "call ext4_journal_dirty_met adata");
 390                 err = ext4_journal_dirty_metadata(handle,
 391                                 EXT4_SB(sb)->s_sbh);
 392         }
 393         return err;
 394 }
 395
 396 int ext4_update_rocompat_feature(handle_t *handle,
 397                                         struct super_block *sb, __u32 rocompat)
 398 {
 399         int err = 0;
 400         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 401                 err = ext4_journal_get_write_access(handle,
 402                                 EXT4_SB(sb)->s_sbh);
 403                 if (err)
 404                         return err;
 405                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 406                 sb->s_dirt = 1;
 407                 handle->h_sync = 1;
 408                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 409                                         "call ext4_journal_dirty_met adata");
 410                 err = ext4_journal_dirty_metadata(handle,
 411                                 EXT4_SB(sb)->s_sbh);
 412         }
 413         return err;
 414 }
 415
 416 int ext4_update_incompat_feature(handle_t *handle,
 417                                         struct super_block *sb, __u32 incompat)
 418 {
 419         int err = 0;
 420         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 421                 err = ext4_journal_get_write_access(handle,
 422                                 EXT4_SB(sb)->s_sbh);
 423                 if (err)
 424                         return err;
 425                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 426                 sb->s_dirt = 1;
 427                 handle->h_sync = 1;
 428                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 429                                         "call ext4_journal_dirty_met adata");
 430                 err = ext4_journal_dirty_metadata(handle,
 431                                 EXT4_SB(sb)->s_sbh);
 432         }
 433         return err;
 434 }
 435
 436 /*
 437  * Open the external journal device
 438  */
 439 static struct block_device *ext4_blkdev_get(dev_t dev)
 440 {
 441         struct block_device *bdev;
 442         char b[BDEVNAME_SIZE];
 443
 444         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 445         if (IS_ERR(bdev))
 446                 goto fail;
 447         return bdev;
 448
 449 fail:
 450         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 451                         __bdevname(dev, b), PTR_ERR(bdev));
 452         return NULL;
 453 }
 454
 455 /*
 456  * Release the journal device
 457  */
 458 static int ext4_blkdev_put(struct block_device *bdev)
 459 {
 460         bd_release(bdev);
 461         return blkdev_put(bdev);
 462 }
 463
 464 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 465 {
 466         struct block_device *bdev;
 467         int ret = -ENODEV;
 468
 469         bdev = sbi->journal_bdev;
 470         if (bdev) {
 471                 ret = ext4_blkdev_put(bdev);
 472                 sbi->journal_bdev = NULL;
 473         }
 474         return ret;
 475 }
 476
 477 static inline struct inode *orphan_list_entry(struct list_head *l)
 478 {
 479         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 480 }
 481
 482 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 483 {
 484         struct list_head *l;
 485
 486         printk(KERN_ERR "sb orphan head is %d\n",
 487                le32_to_cpu(sbi->s_es->s_last_orphan));
 488
 489         printk(KERN_ERR "sb_info orphan list:\n");
 490         list_for_each(l, &sbi->s_orphan) {
 491                 struct inode *inode = orphan_list_entry(l);
 492                 printk(KERN_ERR "  "
 493                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 494                        inode->i_sb->s_id, inode->i_ino, inode,
 495                        inode->i_mode, inode->i_nlink,
 496                        NEXT_ORPHAN(inode));
 497         }
 498 }
 499
 500 static void ext4_put_super(struct super_block *sb)
 501 {
 502         struct ext4_sb_info *sbi = EXT4_SB(sb);
 503         struct ext4_super_block *es = sbi->s_es;
 504         int i;
 505
 506         ext4_mb_release(sb);
 507         ext4_ext_release(sb);
 508         ext4_xattr_put_super(sb);
 509         jbd2_journal_destroy(sbi->s_journal);
 510         sbi->s_journal = NULL;
 511         if (!(sb->s_flags & MS_RDONLY)) {
 512                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 513                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 514                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 515                 mark_buffer_dirty(sbi->s_sbh);
 516                 ext4_commit_super(sb, es, 1);
 517         }
 518         if (sbi->s_proc) {
 519                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
 520                 remove_proc_entry(sb->s_id, ext4_proc_root);
 521         }
 522
 523         for (i = 0; i < sbi->s_gdb_count; i++)
 524                 brelse(sbi->s_group_desc[i]);
 525         kfree(sbi->s_group_desc);
 526         kfree(sbi->s_flex_groups);
 527         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 528         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 529         percpu_counter_destroy(&sbi->s_dirs_counter);
 530         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 531         brelse(sbi->s_sbh);
 532 #ifdef CONFIG_QUOTA
 533         for (i = 0; i < MAXQUOTAS; i++)
 534                 kfree(sbi->s_qf_names[i]);
 535 #endif
 536
 537         /* Debugging code just in case the in-memory inode orphan list
 538          * isn't empty.  The on-disk one can be non-empty if we've
 539          * detected an error and taken the fs readonly, but the
 540          * in-memory list had better be clean by this point. */
 541         if (!list_empty(&sbi->s_orphan))
 542                 dump_orphan_list(sb, sbi);
 543         J_ASSERT(list_empty(&sbi->s_orphan));
 544
 545         invalidate_bdev(sb->s_bdev);
 546         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 547                 /*
 548                  * Invalidate the journal device's buffers.  We don't want them
 549                  * floating about in memory - the physical journal device may
 550                  * hotswapped, and it breaks the `ro-after' testing code.
 551                  */
 552                 sync_blockdev(sbi->journal_bdev);
 553                 invalidate_bdev(sbi->journal_bdev);
 554                 ext4_blkdev_remove(sbi);
 555         }
 556         sb->s_fs_info = NULL;
 557         kfree(sbi);
 558         return;
 559 }
 560
 561 static struct kmem_cache *ext4_inode_cachep;
 562
 563 /*
 564  * Called inside transaction, so use GFP_NOFS
 565  */
 566 static struct inode *ext4_alloc_inode(struct super_block *sb)
 567 {
 568         struct ext4_inode_info *ei;
 569
 570         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 571         if (!ei)
 572                 return NULL;
 573 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 574         ei->i_acl = EXT4_ACL_NOT_CACHED;
 575         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 576 #endif
 577         ei->vfs_inode.i_version = 1;
 578         ei->vfs_inode.i_data.writeback_index = 0;
 579         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 580         INIT_LIST_HEAD(&ei->i_prealloc_list);
 581         spin_lock_init(&ei->i_prealloc_lock);
 582         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 583         ei->i_reserved_data_blocks = 0;
 584         ei->i_reserved_meta_blocks = 0;
 585         ei->i_allocated_meta_blocks = 0;
 586         ei->i_delalloc_reserved_flag = 0;
 587         spin_lock_init(&(ei->i_block_reservation_lock));
 588         return &ei->vfs_inode;
 589 }
 590
 591 static void ext4_destroy_inode(struct inode *inode)
 592 {
 593         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 594                 printk("EXT4 Inode %p: orphan list check failed!\n",
 595                         EXT4_I(inode));
 596                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 597                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 598                                 true);
 599                 dump_stack();
 600         }
 601         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 602 }
 603
 604 static void init_once(void *foo)
 605 {
 606         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 607
 608         INIT_LIST_HEAD(&ei->i_orphan);
 609 #ifdef CONFIG_EXT4DEV_FS_XATTR
 610         init_rwsem(&ei->xattr_sem);
 611 #endif
 612         init_rwsem(&ei->i_data_sem);
 613         inode_init_once(&ei->vfs_inode);
 614 }
 615
 616 static int init_inodecache(void)
 617 {
 618         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 619                                              sizeof(struct ext4_inode_info),
 620                                              0, (SLAB_RECLAIM_ACCOUNT|
 621                                                 SLAB_MEM_SPREAD),
 622                                              init_once);
 623         if (ext4_inode_cachep == NULL)
 624                 return -ENOMEM;
 625         return 0;
 626 }
 627
 628 static void destroy_inodecache(void)
 629 {
 630         kmem_cache_destroy(ext4_inode_cachep);
 631 }
 632
 633 static void ext4_clear_inode(struct inode *inode)
 634 {
 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_preallocations(inode);
 648         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 649                                        &EXT4_I(inode)->jinode);
 650 }
 651
 652 static inline void ext4_show_quota_options(struct seq_file *seq,
 653                                            struct super_block *sb)
 654 {
 655 #if defined(CONFIG_QUOTA)
 656         struct ext4_sb_info *sbi = EXT4_SB(sb);
 657
 658         if (sbi->s_jquota_fmt)
 659                 seq_printf(seq, ",jqfmt=%s",
 660                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold" : "vfsv0");
 661
 662         if (sbi->s_qf_names[USRQUOTA])
 663                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 664
 665         if (sbi->s_qf_names[GRPQUOTA])
 666                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 667
 668         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 669                 seq_puts(seq, ",usrquota");
 670
 671         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 672                 seq_puts(seq, ",grpquota");
 673 #endif
 674 }
 675
 676 /*
 677  * Show an option if
 678  *  - it's set to a non-default value OR
 679  *  - if the per-sb default is different from the global default
 680  */
 681 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 682 {
 683         int def_errors;
 684         unsigned long def_mount_opts;
 685         struct super_block *sb = vfs->mnt_sb;
 686         struct ext4_sb_info *sbi = EXT4_SB(sb);
 687         struct ext4_super_block *es = sbi->s_es;
 688
 689         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 690         def_errors     = le16_to_cpu(es->s_errors);
 691
 692         if (sbi->s_sb_block != 1)
 693                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 694         if (test_opt(sb, MINIX_DF))
 695                 seq_puts(seq, ",minixdf");
 696         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 697                 seq_puts(seq, ",grpid");
 698         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 699                 seq_puts(seq, ",nogrpid");
 700         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 701             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 702                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 703         }
 704         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 705             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 706                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 707         }
 708         if (test_opt(sb, ERRORS_RO)) {
 709                 if (def_errors == EXT4_ERRORS_PANIC ||
 710                     def_errors == EXT4_ERRORS_CONTINUE) {
 711                         seq_puts(seq, ",errors=remount-ro");
 712                 }
 713         }
 714         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 715                 seq_puts(seq, ",errors=continue");
 716         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 717                 seq_puts(seq, ",errors=panic");
 718         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 719                 seq_puts(seq, ",nouid32");
 720         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 721                 seq_puts(seq, ",debug");
 722         if (test_opt(sb, OLDALLOC))
 723                 seq_puts(seq, ",oldalloc");
 724 #ifdef CONFIG_EXT4DEV_FS_XATTR
 725         if (test_opt(sb, XATTR_USER) &&
 726                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 727                 seq_puts(seq, ",user_xattr");
 728         if (!test_opt(sb, XATTR_USER) &&
 729             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 730                 seq_puts(seq, ",nouser_xattr");
 731         }
 732 #endif
 733 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 734         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 735                 seq_puts(seq, ",acl");
 736         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 737                 seq_puts(seq, ",noacl");
 738 #endif
 739         if (!test_opt(sb, RESERVATION))
 740                 seq_puts(seq, ",noreservation");
 741         if (sbi->s_commit_interval) {
 742                 seq_printf(seq, ",commit=%u",
 743                            (unsigned) (sbi->s_commit_interval / HZ));
 744         }
 745         /*
 746          * We're changing the default of barrier mount option, so
 747          * let's always display its mount state so it's clear what its
 748          * status is.
 749          */
 750         seq_puts(seq, ",barrier=");
 751         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 752         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 753                 seq_puts(seq, ",journal_async_commit");
 754         if (test_opt(sb, NOBH))
 755                 seq_puts(seq, ",nobh");
 756         if (!test_opt(sb, EXTENTS))
 757                 seq_puts(seq, ",noextents");
 758         if (test_opt(sb, I_VERSION))
 759                 seq_puts(seq, ",i_version");
 760         if (!test_opt(sb, DELALLOC))
 761                 seq_puts(seq, ",nodelalloc");
 762
 763
 764         if (sbi->s_stripe)
 765                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 766         /*
 767          * journal mode get enabled in different ways
 768          * So just print the value even if we didn't specify it
 769          */
 770         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 771                 seq_puts(seq, ",data=journal");
 772         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 773                 seq_puts(seq, ",data=ordered");
 774         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 775                 seq_puts(seq, ",data=writeback");
 776
 777         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 778                 seq_printf(seq, ",inode_readahead_blks=%u",
 779                            sbi->s_inode_readahead_blks);
 780
 781         ext4_show_quota_options(seq, sb);
 782         return 0;
 783 }
 784
 785
 786 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 787                 u64 ino, u32 generation)
 788 {
 789         struct inode *inode;
 790
 791         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 792                 return ERR_PTR(-ESTALE);
 793         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 794                 return ERR_PTR(-ESTALE);
 795
 796         /* iget isn't really right if the inode is currently unallocated!!
 797          *
 798          * ext4_read_inode will return a bad_inode if the inode had been
 799          * deleted, so we should be safe.
 800          *
 801          * Currently we don't know the generation for parent directory, so
 802          * a generation of 0 means "accept any"
 803          */
 804         inode = ext4_iget(sb, ino);
 805         if (IS_ERR(inode))
 806                 return ERR_CAST(inode);
 807         if (generation && inode->i_generation != generation) {
 808                 iput(inode);
 809                 return ERR_PTR(-ESTALE);
 810         }
 811
 812         return inode;
 813 }
 814
 815 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 816                 int fh_len, int fh_type)
 817 {
 818         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 819                                     ext4_nfs_get_inode);
 820 }
 821
 822 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 823                 int fh_len, int fh_type)
 824 {
 825         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 826                                     ext4_nfs_get_inode);
 827 }
 828
 829 #ifdef CONFIG_QUOTA
 830 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
 831 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 832
 833 static int ext4_dquot_initialize(struct inode *inode, int type);
 834 static int ext4_dquot_drop(struct inode *inode);
 835 static int ext4_write_dquot(struct dquot *dquot);
 836 static int ext4_acquire_dquot(struct dquot *dquot);
 837 static int ext4_release_dquot(struct dquot *dquot);
 838 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 839 static int ext4_write_info(struct super_block *sb, int type);
 840 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 841                                 char *path, int remount);
 842 static int ext4_quota_on_mount(struct super_block *sb, int type);
 843 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 844                                size_t len, loff_t off);
 845 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 846                                 const char *data, size_t len, loff_t off);
 847
 848 static struct dquot_operations ext4_quota_operations = {
 849         .initialize     = ext4_dquot_initialize,
 850         .drop           = ext4_dquot_drop,
 851         .alloc_space    = dquot_alloc_space,
 852         .alloc_inode    = dquot_alloc_inode,
 853         .free_space     = dquot_free_space,
 854         .free_inode     = dquot_free_inode,
 855         .transfer       = dquot_transfer,
 856         .write_dquot    = ext4_write_dquot,
 857         .acquire_dquot  = ext4_acquire_dquot,
 858         .release_dquot  = ext4_release_dquot,
 859         .mark_dirty     = ext4_mark_dquot_dirty,
 860         .write_info     = ext4_write_info
 861 };
 862
 863 static struct quotactl_ops ext4_qctl_operations = {
 864         .quota_on       = ext4_quota_on,
 865         .quota_off      = vfs_quota_off,
 866         .quota_sync     = vfs_quota_sync,
 867         .get_info       = vfs_get_dqinfo,
 868         .set_info       = vfs_set_dqinfo,
 869         .get_dqblk      = vfs_get_dqblk,
 870         .set_dqblk      = vfs_set_dqblk
 871 };
 872 #endif
 873
 874 static const struct super_operations ext4_sops = {
 875         .alloc_inode    = ext4_alloc_inode,
 876         .destroy_inode  = ext4_destroy_inode,
 877         .write_inode    = ext4_write_inode,
 878         .dirty_inode    = ext4_dirty_inode,
 879         .delete_inode   = ext4_delete_inode,
 880         .put_super      = ext4_put_super,
 881         .write_super    = ext4_write_super,
 882         .sync_fs        = ext4_sync_fs,
 883         .write_super_lockfs = ext4_write_super_lockfs,
 884         .unlockfs       = ext4_unlockfs,
 885         .statfs         = ext4_statfs,
 886         .remount_fs     = ext4_remount,
 887         .clear_inode    = ext4_clear_inode,
 888         .show_options   = ext4_show_options,
 889 #ifdef CONFIG_QUOTA
 890         .quota_read     = ext4_quota_read,
 891         .quota_write    = ext4_quota_write,
 892 #endif
 893 };
 894
 895 static const struct export_operations ext4_export_ops = {
 896         .fh_to_dentry = ext4_fh_to_dentry,
 897         .fh_to_parent = ext4_fh_to_parent,
 898         .get_parent = ext4_get_parent,
 899 };
 900
 901 enum {
 902         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 903         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 904         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 905         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 906         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 907         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 908         Opt_journal_checksum, Opt_journal_async_commit,
 909         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 910         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 911         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 912         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 913         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 914         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 915         Opt_inode_readahead_blks
 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_inode_readahead_blks, "inode_readahead_blks=%u"},
 977         {Opt_err, NULL},
 978 };
 979
 980 static ext4_fsblk_t get_sb_block(void **data)
 981 {
 982         ext4_fsblk_t    sb_block;
 983         char            *options = (char *) *data;
 984
 985         if (!options || strncmp(options, "sb=", 3) != 0)
 986                 return 1;       /* Default location */
 987         options += 3;
 988         /*todo: use simple_strtoll with >32bit ext4 */
 989         sb_block = simple_strtoul(options, &options, 0);
 990         if (*options && *options != ',') {
 991                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
 992                        (char *) *data);
 993                 return 1;
 994         }
 995         if (*options == ',')
 996                 options++;
 997         *data = (void *) options;
 998         return sb_block;
 999 }
1000
1001 static int parse_options(char *options, struct super_block *sb,
1002                          unsigned int *inum, unsigned long *journal_devnum,
1003                          ext4_fsblk_t *n_blocks_count, int is_remount)
1004 {
1005         struct ext4_sb_info *sbi = EXT4_SB(sb);
1006         char *p;
1007         substring_t args[MAX_OPT_ARGS];
1008         int data_opt = 0;
1009         int option;
1010 #ifdef CONFIG_QUOTA
1011         int qtype, qfmt;
1012         char *qname;
1013 #endif
1014         ext4_fsblk_t last_block;
1015
1016         if (!options)
1017                 return 1;
1018
1019         while ((p = strsep(&options, ",")) != NULL) {
1020                 int token;
1021                 if (!*p)
1022                         continue;
1023
1024                 token = match_token(p, tokens, args);
1025                 switch (token) {
1026                 case Opt_bsd_df:
1027                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1028                         break;
1029                 case Opt_minix_df:
1030                         set_opt(sbi->s_mount_opt, MINIX_DF);
1031                         break;
1032                 case Opt_grpid:
1033                         set_opt(sbi->s_mount_opt, GRPID);
1034                         break;
1035                 case Opt_nogrpid:
1036                         clear_opt(sbi->s_mount_opt, GRPID);
1037                         break;
1038                 case Opt_resuid:
1039                         if (match_int(&args[0], &option))
1040                                 return 0;
1041                         sbi->s_resuid = option;
1042                         break;
1043                 case Opt_resgid:
1044                         if (match_int(&args[0], &option))
1045                                 return 0;
1046                         sbi->s_resgid = option;
1047                         break;
1048                 case Opt_sb:
1049                         /* handled by get_sb_block() instead of here */
1050                         /* *sb_block = match_int(&args[0]); */
1051                         break;
1052                 case Opt_err_panic:
1053                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1054                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1055                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1056                         break;
1057                 case Opt_err_ro:
1058                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1059                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1060                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1061                         break;
1062                 case Opt_err_cont:
1063                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1064                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1065                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1066                         break;
1067                 case Opt_nouid32:
1068                         set_opt(sbi->s_mount_opt, NO_UID32);
1069                         break;
1070                 case Opt_nocheck:
1071                         clear_opt(sbi->s_mount_opt, CHECK);
1072                         break;
1073                 case Opt_debug:
1074                         set_opt(sbi->s_mount_opt, DEBUG);
1075                         break;
1076                 case Opt_oldalloc:
1077                         set_opt(sbi->s_mount_opt, OLDALLOC);
1078                         break;
1079                 case Opt_orlov:
1080                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1081                         break;
1082 #ifdef CONFIG_EXT4DEV_FS_XATTR
1083                 case Opt_user_xattr:
1084                         set_opt(sbi->s_mount_opt, XATTR_USER);
1085                         break;
1086                 case Opt_nouser_xattr:
1087                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1088                         break;
1089 #else
1090                 case Opt_user_xattr:
1091                 case Opt_nouser_xattr:
1092                         printk(KERN_ERR "EXT4 (no)user_xattr options "
1093                                "not supported\n");
1094                         break;
1095 #endif
1096 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1097                 case Opt_acl:
1098                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1099                         break;
1100                 case Opt_noacl:
1101                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1102                         break;
1103 #else
1104                 case Opt_acl:
1105                 case Opt_noacl:
1106                         printk(KERN_ERR "EXT4 (no)acl options "
1107                                "not supported\n");
1108                         break;
1109 #endif
1110                 case Opt_reservation:
1111                         set_opt(sbi->s_mount_opt, RESERVATION);
1112                         break;
1113                 case Opt_noreservation:
1114                         clear_opt(sbi->s_mount_opt, RESERVATION);
1115                         break;
1116                 case Opt_journal_update:
1117                         /* @@@ FIXME */
1118                         /* Eventually we will want to be able to create
1119                            a journal file here.  For now, only allow the
1120                            user to specify an existing inode to be the
1121                            journal file. */
1122                         if (is_remount) {
1123                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1124                                        "journal on remount\n");
1125                                 return 0;
1126                         }
1127                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1128                         break;
1129                 case Opt_journal_inum:
1130                         if (is_remount) {
1131                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1132                                        "journal on remount\n");
1133                                 return 0;
1134                         }
1135                         if (match_int(&args[0], &option))
1136                                 return 0;
1137                         *inum = option;
1138                         break;
1139                 case Opt_journal_dev:
1140                         if (is_remount) {
1141                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1142                                        "journal on remount\n");
1143                                 return 0;
1144                         }
1145                         if (match_int(&args[0], &option))
1146                                 return 0;
1147                         *journal_devnum = option;
1148                         break;
1149                 case Opt_journal_checksum:
1150                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1151                         break;
1152                 case Opt_journal_async_commit:
1153                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1154                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1155                         break;
1156                 case Opt_noload:
1157                         set_opt(sbi->s_mount_opt, NOLOAD);
1158                         break;
1159                 case Opt_commit:
1160                         if (match_int(&args[0], &option))
1161                                 return 0;
1162                         if (option < 0)
1163                                 return 0;
1164                         if (option == 0)
1165                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1166                         sbi->s_commit_interval = HZ * option;
1167                         break;
1168                 case Opt_data_journal:
1169                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1170                         goto datacheck;
1171                 case Opt_data_ordered:
1172                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1173                         goto datacheck;
1174                 case Opt_data_writeback:
1175                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1176                 datacheck:
1177                         if (is_remount) {
1178                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1179                                                 != data_opt) {
1180                                         printk(KERN_ERR
1181                                                 "EXT4-fs: cannot change data "
1182                                                 "mode on remount\n");
1183                                         return 0;
1184                                 }
1185                         } else {
1186                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1187                                 sbi->s_mount_opt |= data_opt;
1188                         }
1189                         break;
1190 #ifdef CONFIG_QUOTA
1191                 case Opt_usrjquota:
1192                         qtype = USRQUOTA;
1193                         goto set_qf_name;
1194                 case Opt_grpjquota:
1195                         qtype = GRPQUOTA;
1196 set_qf_name:
1197                         if ((sb_any_quota_enabled(sb) ||
1198                              sb_any_quota_suspended(sb)) &&
1199                             !sbi->s_qf_names[qtype]) {
1200                                 printk(KERN_ERR
1201                                        "EXT4-fs: Cannot change journaled "
1202                                        "quota options when quota turned on.\n");
1203                                 return 0;
1204                         }
1205                         qname = match_strdup(&args[0]);
1206                         if (!qname) {
1207                                 printk(KERN_ERR
1208                                         "EXT4-fs: not enough memory for "
1209                                         "storing quotafile name.\n");
1210                                 return 0;
1211                         }
1212                         if (sbi->s_qf_names[qtype] &&
1213                             strcmp(sbi->s_qf_names[qtype], qname)) {
1214                                 printk(KERN_ERR
1215                                         "EXT4-fs: %s quota file already "
1216                                         "specified.\n", QTYPE2NAME(qtype));
1217                                 kfree(qname);
1218                                 return 0;
1219                         }
1220                         sbi->s_qf_names[qtype] = qname;
1221                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1222                                 printk(KERN_ERR
1223                                         "EXT4-fs: quotafile must be on "
1224                                         "filesystem root.\n");
1225                                 kfree(sbi->s_qf_names[qtype]);
1226                                 sbi->s_qf_names[qtype] = NULL;
1227                                 return 0;
1228                         }
1229                         set_opt(sbi->s_mount_opt, QUOTA);
1230                         break;
1231                 case Opt_offusrjquota:
1232                         qtype = USRQUOTA;
1233                         goto clear_qf_name;
1234                 case Opt_offgrpjquota:
1235                         qtype = GRPQUOTA;
1236 clear_qf_name:
1237                         if ((sb_any_quota_enabled(sb) ||
1238                              sb_any_quota_suspended(sb)) &&
1239                             sbi->s_qf_names[qtype]) {
1240                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1241                                         "journaled quota options when "
1242                                         "quota turned on.\n");
1243                                 return 0;
1244                         }
1245                         /*
1246                          * The space will be released later when all options
1247                          * are confirmed to be correct
1248                          */
1249                         sbi->s_qf_names[qtype] = NULL;
1250                         break;
1251                 case Opt_jqfmt_vfsold:
1252                         qfmt = QFMT_VFS_OLD;
1253                         goto set_qf_format;
1254                 case Opt_jqfmt_vfsv0:
1255                         qfmt = QFMT_VFS_V0;
1256 set_qf_format:
1257                         if ((sb_any_quota_enabled(sb) ||
1258                              sb_any_quota_suspended(sb)) &&
1259                             sbi->s_jquota_fmt != qfmt) {
1260                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1261                                         "journaled quota options when "
1262                                         "quota turned on.\n");
1263                                 return 0;
1264                         }
1265                         sbi->s_jquota_fmt = qfmt;
1266                         break;
1267                 case Opt_quota:
1268                 case Opt_usrquota:
1269                         set_opt(sbi->s_mount_opt, QUOTA);
1270                         set_opt(sbi->s_mount_opt, USRQUOTA);
1271                         break;
1272                 case Opt_grpquota:
1273                         set_opt(sbi->s_mount_opt, QUOTA);
1274                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1275                         break;
1276                 case Opt_noquota:
1277                         if (sb_any_quota_enabled(sb)) {
1278                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1279                                         "options when quota turned on.\n");
1280                                 return 0;
1281                         }
1282                         clear_opt(sbi->s_mount_opt, QUOTA);
1283                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1284                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1285                         break;
1286 #else
1287                 case Opt_quota:
1288                 case Opt_usrquota:
1289                 case Opt_grpquota:
1290                         printk(KERN_ERR
1291                                 "EXT4-fs: quota options not supported.\n");
1292                         break;
1293                 case Opt_usrjquota:
1294                 case Opt_grpjquota:
1295                 case Opt_offusrjquota:
1296                 case Opt_offgrpjquota:
1297                 case Opt_jqfmt_vfsold:
1298                 case Opt_jqfmt_vfsv0:
1299                         printk(KERN_ERR
1300                                 "EXT4-fs: journaled quota options not "
1301                                 "supported.\n");
1302                         break;
1303                 case Opt_noquota:
1304                         break;
1305 #endif
1306                 case Opt_abort:
1307                         set_opt(sbi->s_mount_opt, ABORT);
1308                         break;
1309                 case Opt_barrier:
1310                         if (match_int(&args[0], &option))
1311                                 return 0;
1312                         if (option)
1313                                 set_opt(sbi->s_mount_opt, BARRIER);
1314                         else
1315                                 clear_opt(sbi->s_mount_opt, BARRIER);
1316                         break;
1317                 case Opt_ignore:
1318                         break;
1319                 case Opt_resize:
1320                         if (!is_remount) {
1321                                 printk("EXT4-fs: resize option only available "
1322                                         "for remount\n");
1323                                 return 0;
1324                         }
1325                         if (match_int(&args[0], &option) != 0)
1326                                 return 0;
1327                         *n_blocks_count = option;
1328                         break;
1329                 case Opt_nobh:
1330                         set_opt(sbi->s_mount_opt, NOBH);
1331                         break;
1332                 case Opt_bh:
1333                         clear_opt(sbi->s_mount_opt, NOBH);
1334                         break;
1335                 case Opt_extents:
1336                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1337                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1338                                 ext4_warning(sb, __func__,
1339                                         "extents feature not enabled "
1340                                         "on this filesystem, use tune2fs\n");
1341                                 return 0;
1342                         }
1343                         set_opt(sbi->s_mount_opt, EXTENTS);
1344                         break;
1345                 case Opt_noextents:
1346                         /*
1347                          * When e2fsprogs support resizing an already existing
1348                          * ext3 file system to greater than 2**32 we need to
1349                          * add support to block allocator to handle growing
1350                          * already existing block  mapped inode so that blocks
1351                          * allocated for them fall within 2**32
1352                          */
1353                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1354                         if (last_block  > 0xffffffffULL) {
1355                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1356                                                 "large to mount with "
1357                                                 "-o noextents options\n");
1358                                 return 0;
1359                         }
1360                         clear_opt(sbi->s_mount_opt, EXTENTS);
1361                         break;
1362                 case Opt_i_version:
1363                         set_opt(sbi->s_mount_opt, I_VERSION);
1364                         sb->s_flags |= MS_I_VERSION;
1365                         break;
1366                 case Opt_nodelalloc:
1367                         clear_opt(sbi->s_mount_opt, DELALLOC);
1368                         break;
1369                 case Opt_stripe:
1370                         if (match_int(&args[0], &option))
1371                                 return 0;
1372                         if (option < 0)
1373                                 return 0;
1374                         sbi->s_stripe = option;
1375                         break;
1376                 case Opt_delalloc:
1377                         set_opt(sbi->s_mount_opt, DELALLOC);
1378                         break;
1379                 case Opt_inode_readahead_blks:
1380                         if (match_int(&args[0], &option))
1381                                 return 0;
1382                         if (option < 0 || option > (1 << 30))
1383                                 return 0;
1384                         sbi->s_inode_readahead_blks = option;
1385                         break;
1386                 default:
1387                         printk(KERN_ERR
1388                                "EXT4-fs: Unrecognized mount option \"%s\" "
1389                                "or missing value\n", p);
1390                         return 0;
1391                 }
1392         }
1393 #ifdef CONFIG_QUOTA
1394         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1395                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1396                      sbi->s_qf_names[USRQUOTA])
1397                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1398
1399                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1400                      sbi->s_qf_names[GRPQUOTA])
1401                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1402
1403                 if ((sbi->s_qf_names[USRQUOTA] &&
1404                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1405                     (sbi->s_qf_names[GRPQUOTA] &&
1406                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1407                         printk(KERN_ERR "EXT4-fs: old and new quota "
1408                                         "format mixing.\n");
1409                         return 0;
1410                 }
1411
1412                 if (!sbi->s_jquota_fmt) {
1413                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1414                                         "not specified.\n");
1415                         return 0;
1416                 }
1417         } else {
1418                 if (sbi->s_jquota_fmt) {
1419                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1420                                         "specified with no journaling "
1421                                         "enabled.\n");
1422                         return 0;
1423                 }
1424         }
1425 #endif
1426         return 1;
1427 }
1428
1429 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1430                             int read_only)
1431 {
1432         struct ext4_sb_info *sbi = EXT4_SB(sb);
1433         int res = 0;
1434
1435         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1436                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1437                        "forcing read-only mode\n");
1438                 res = MS_RDONLY;
1439         }
1440         if (read_only)
1441                 return res;
1442         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1443                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1444                        "running e2fsck is recommended\n");
1445         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1446                 printk(KERN_WARNING
1447                        "EXT4-fs warning: mounting fs with errors, "
1448                        "running e2fsck is recommended\n");
1449         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1450                  le16_to_cpu(es->s_mnt_count) >=
1451                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1452                 printk(KERN_WARNING
1453                        "EXT4-fs warning: maximal mount count reached, "
1454                        "running e2fsck is recommended\n");
1455         else if (le32_to_cpu(es->s_checkinterval) &&
1456                 (le32_to_cpu(es->s_lastcheck) +
1457                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1458                 printk(KERN_WARNING
1459                        "EXT4-fs warning: checktime reached, "
1460                        "running e2fsck is recommended\n");
1461 #if 0
1462                 /* @@@ We _will_ want to clear the valid bit if we find
1463                  * inconsistencies, to force a fsck at reboot.  But for
1464                  * a plain journaled filesystem we can keep it set as
1465                  * valid forever! :)
1466                  */
1467         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1468 #endif
1469         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1470                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1471         le16_add_cpu(&es->s_mnt_count, 1);
1472         es->s_mtime = cpu_to_le32(get_seconds());
1473         ext4_update_dynamic_rev(sb);
1474         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1475
1476         ext4_commit_super(sb, es, 1);
1477         if (test_opt(sb, DEBUG))
1478                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1479                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1480                         sb->s_blocksize,
1481                         sbi->s_groups_count,
1482                         EXT4_BLOCKS_PER_GROUP(sb),
1483                         EXT4_INODES_PER_GROUP(sb),
1484                         sbi->s_mount_opt);
1485
1486         printk(KERN_INFO "EXT4 FS on %s, %s journal on %s\n",
1487                sb->s_id, EXT4_SB(sb)->s_journal->j_inode ? "internal" :
1488                "external", EXT4_SB(sb)->s_journal->j_devname);
1489         return res;
1490 }
1491
1492 static int ext4_fill_flex_info(struct super_block *sb)
1493 {
1494         struct ext4_sb_info *sbi = EXT4_SB(sb);
1495         struct ext4_group_desc *gdp = NULL;
1496         struct buffer_head *bh;
1497         ext4_group_t flex_group_count;
1498         ext4_group_t flex_group;
1499         int groups_per_flex = 0;
1500         __u64 block_bitmap = 0;
1501         int i;
1502
1503         if (!sbi->s_es->s_log_groups_per_flex) {
1504                 sbi->s_log_groups_per_flex = 0;
1505                 return 1;
1506         }
1507
1508         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1509         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1510
1511         /* We allocate both existing and potentially added groups */
1512         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1513                             ((sbi->s_es->s_reserved_gdt_blocks +1 ) <<
1514                               EXT4_DESC_PER_BLOCK_BITS(sb))) /
1515                            groups_per_flex;
1516         sbi->s_flex_groups = kzalloc(flex_group_count *
1517                                      sizeof(struct flex_groups), GFP_KERNEL);
1518         if (sbi->s_flex_groups == NULL) {
1519                 printk(KERN_ERR "EXT4-fs: not enough memory for "
1520                                 "%lu flex groups\n", flex_group_count);
1521                 goto failed;
1522         }
1523
1524         gdp = ext4_get_group_desc(sb, 1, &bh);
1525         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1526
1527         for (i = 0; i < sbi->s_groups_count; i++) {
1528                 gdp = ext4_get_group_desc(sb, i, &bh);
1529
1530                 flex_group = ext4_flex_group(sbi, i);
1531                 sbi->s_flex_groups[flex_group].free_inodes +=
1532                         le16_to_cpu(gdp->bg_free_inodes_count);
1533                 sbi->s_flex_groups[flex_group].free_blocks +=
1534                         le16_to_cpu(gdp->bg_free_blocks_count);
1535         }
1536
1537         return 1;
1538 failed:
1539         return 0;
1540 }
1541
1542 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1543                             struct ext4_group_desc *gdp)
1544 {
1545         __u16 crc = 0;
1546
1547         if (sbi->s_es->s_feature_ro_compat &
1548             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1549                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1550                 __le32 le_group = cpu_to_le32(block_group);
1551
1552                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1553                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1554                 crc = crc16(crc, (__u8 *)gdp, offset);
1555                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1556                 /* for checksum of struct ext4_group_desc do the rest...*/
1557                 if ((sbi->s_es->s_feature_incompat &
1558                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1559                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1560                         crc = crc16(crc, (__u8 *)gdp + offset,
1561                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1562                                         offset);
1563         }
1564
1565         return cpu_to_le16(crc);
1566 }
1567
1568 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1569                                 struct ext4_group_desc *gdp)
1570 {
1571         if ((sbi->s_es->s_feature_ro_compat &
1572              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1573             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1574                 return 0;
1575
1576         return 1;
1577 }
1578
1579 /* Called at mount-time, super-block is locked */
1580 static int ext4_check_descriptors(struct super_block *sb)
1581 {
1582         struct ext4_sb_info *sbi = EXT4_SB(sb);
1583         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1584         ext4_fsblk_t last_block;
1585         ext4_fsblk_t block_bitmap;
1586         ext4_fsblk_t inode_bitmap;
1587         ext4_fsblk_t inode_table;
1588         int flexbg_flag = 0;
1589         ext4_group_t i;
1590
1591         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1592                 flexbg_flag = 1;
1593
1594         ext4_debug("Checking group descriptors");
1595
1596         for (i = 0; i < sbi->s_groups_count; i++) {
1597                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1598
1599                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1600                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1601                 else
1602                         last_block = first_block +
1603                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1604
1605                 block_bitmap = ext4_block_bitmap(sb, gdp);
1606                 if (block_bitmap < first_block || block_bitmap > last_block) {
1607                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1608                                "Block bitmap for group %lu not in group "
1609                                "(block %llu)!", i, block_bitmap);
1610                         return 0;
1611                 }
1612                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1613                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1614                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1615                                "Inode bitmap for group %lu not in group "
1616                                "(block %llu)!", i, inode_bitmap);
1617                         return 0;
1618                 }
1619                 inode_table = ext4_inode_table(sb, gdp);
1620                 if (inode_table < first_block ||
1621                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1622                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1623                                "Inode table for group %lu not in group "
1624                                "(block %llu)!", i, inode_table);
1625                         return 0;
1626                 }
1627                 spin_lock(sb_bgl_lock(sbi, i));
1628                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1629                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1630                                "Checksum for group %lu failed (%u!=%u)\n",
1631                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1632                                gdp)), le16_to_cpu(gdp->bg_checksum));
1633                         if (!(sb->s_flags & MS_RDONLY)) {
1634                                 spin_unlock(sb_bgl_lock(sbi, i));
1635                                 return 0;
1636                         }
1637                 }
1638                 spin_unlock(sb_bgl_lock(sbi, i));
1639                 if (!flexbg_flag)
1640                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1641         }
1642
1643         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1644         sbi->s_es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
1645         return 1;
1646 }
1647
1648 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1649  * the superblock) which were deleted from all directories, but held open by
1650  * a process at the time of a crash.  We walk the list and try to delete these
1651  * inodes at recovery time (only with a read-write filesystem).
1652  *
1653  * In order to keep the orphan inode chain consistent during traversal (in
1654  * case of crash during recovery), we link each inode into the superblock
1655  * orphan list_head and handle it the same way as an inode deletion during
1656  * normal operation (which journals the operations for us).
1657  *
1658  * We only do an iget() and an iput() on each inode, which is very safe if we
1659  * accidentally point at an in-use or already deleted inode.  The worst that
1660  * can happen in this case is that we get a "bit already cleared" message from
1661  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1662  * e2fsck was run on this filesystem, and it must have already done the orphan
1663  * inode cleanup for us, so we can safely abort without any further action.
1664  */
1665 static void ext4_orphan_cleanup(struct super_block *sb,
1666                                 struct ext4_super_block *es)
1667 {
1668         unsigned int s_flags = sb->s_flags;
1669         int nr_orphans = 0, nr_truncates = 0;
1670 #ifdef CONFIG_QUOTA
1671         int i;
1672 #endif
1673         if (!es->s_last_orphan) {
1674                 jbd_debug(4, "no orphan inodes to clean up\n");
1675                 return;
1676         }
1677
1678         if (bdev_read_only(sb->s_bdev)) {
1679                 printk(KERN_ERR "EXT4-fs: write access "
1680                         "unavailable, skipping orphan cleanup.\n");
1681                 return;
1682         }
1683
1684         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1685                 if (es->s_last_orphan)
1686                         jbd_debug(1, "Errors on filesystem, "
1687                                   "clearing orphan list.\n");
1688                 es->s_last_orphan = 0;
1689                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1690                 return;
1691         }
1692
1693         if (s_flags & MS_RDONLY) {
1694                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1695                        sb->s_id);
1696                 sb->s_flags &= ~MS_RDONLY;
1697         }
1698 #ifdef CONFIG_QUOTA
1699         /* Needed for iput() to work correctly and not trash data */
1700         sb->s_flags |= MS_ACTIVE;
1701         /* Turn on quotas so that they are updated correctly */
1702         for (i = 0; i < MAXQUOTAS; i++) {
1703                 if (EXT4_SB(sb)->s_qf_names[i]) {
1704                         int ret = ext4_quota_on_mount(sb, i);
1705                         if (ret < 0)
1706                                 printk(KERN_ERR
1707                                         "EXT4-fs: Cannot turn on journaled "
1708                                         "quota: error %d\n", ret);
1709                 }
1710         }
1711 #endif
1712
1713         while (es->s_last_orphan) {
1714                 struct inode *inode;
1715
1716                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1717                 if (IS_ERR(inode)) {
1718                         es->s_last_orphan = 0;
1719                         break;
1720                 }
1721
1722                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1723                 DQUOT_INIT(inode);
1724                 if (inode->i_nlink) {
1725                         printk(KERN_DEBUG
1726                                 "%s: truncating inode %lu to %lld bytes\n",
1727                                 __func__, inode->i_ino, inode->i_size);
1728                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1729                                   inode->i_ino, inode->i_size);
1730                         ext4_truncate(inode);
1731                         nr_truncates++;
1732                 } else {
1733                         printk(KERN_DEBUG
1734                                 "%s: deleting unreferenced inode %lu\n",
1735                                 __func__, inode->i_ino);
1736                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1737                                   inode->i_ino);
1738                         nr_orphans++;
1739                 }
1740                 iput(inode);  /* The delete magic happens here! */
1741         }
1742
1743 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1744
1745         if (nr_orphans)
1746                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1747                        sb->s_id, PLURAL(nr_orphans));
1748         if (nr_truncates)
1749                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1750                        sb->s_id, PLURAL(nr_truncates));
1751 #ifdef CONFIG_QUOTA
1752         /* Turn quotas off */
1753         for (i = 0; i < MAXQUOTAS; i++) {
1754                 if (sb_dqopt(sb)->files[i])
1755                         vfs_quota_off(sb, i, 0);
1756         }
1757 #endif
1758         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1759 }
1760 /*
1761  * Maximal extent format file size.
1762  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1763  * extent format containers, within a sector_t, and within i_blocks
1764  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1765  * so that won't be a limiting factor.
1766  *
1767  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1768  */
1769 static loff_t ext4_max_size(int blkbits)
1770 {
1771         loff_t res;
1772         loff_t upper_limit = MAX_LFS_FILESIZE;
1773
1774         /* small i_blocks in vfs inode? */
1775         if (sizeof(blkcnt_t) < sizeof(u64)) {
1776                 /*
1777                  * CONFIG_LSF is not enabled implies the inode
1778                  * i_block represent total blocks in 512 bytes
1779                  * 32 == size of vfs inode i_blocks * 8
1780                  */
1781                 upper_limit = (1LL << 32) - 1;
1782
1783                 /* total blocks in file system block size */
1784                 upper_limit >>= (blkbits - 9);
1785                 upper_limit <<= blkbits;
1786         }
1787
1788         /* 32-bit extent-start container, ee_block */
1789         res = 1LL << 32;
1790         res <<= blkbits;
1791         res -= 1;
1792
1793         /* Sanity check against vm- & vfs- imposed limits */
1794         if (res > upper_limit)
1795                 res = upper_limit;
1796
1797         return res;
1798 }
1799
1800 /*
1801  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1802  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1803  * We need to be 1 filesystem block less than the 2^48 sector limit.
1804  */
1805 static loff_t ext4_max_bitmap_size(int bits)
1806 {
1807         loff_t res = EXT4_NDIR_BLOCKS;
1808         int meta_blocks;
1809         loff_t upper_limit;
1810         /* This is calculated to be the largest file size for a
1811          * dense, bitmapped file such that the total number of
1812          * sectors in the file, including data and all indirect blocks,
1813          * does not exceed 2^48 -1
1814          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1815          * total number of  512 bytes blocks of the file
1816          */
1817
1818         if (sizeof(blkcnt_t) < sizeof(u64)) {
1819                 /*
1820                  * CONFIG_LSF is not enabled implies the inode
1821                  * i_block represent total blocks in 512 bytes
1822                  * 32 == size of vfs inode i_blocks * 8
1823                  */
1824                 upper_limit = (1LL << 32) - 1;
1825
1826                 /* total blocks in file system block size */
1827                 upper_limit >>= (bits - 9);
1828
1829         } else {
1830                 /*
1831                  * We use 48 bit ext4_inode i_blocks
1832                  * With EXT4_HUGE_FILE_FL set the i_blocks
1833                  * represent total number of blocks in
1834                  * file system block size
1835                  */
1836                 upper_limit = (1LL << 48) - 1;
1837
1838         }
1839
1840         /* indirect blocks */
1841         meta_blocks = 1;
1842         /* double indirect blocks */
1843         meta_blocks += 1 + (1LL << (bits-2));
1844         /* tripple indirect blocks */
1845         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1846
1847         upper_limit -= meta_blocks;
1848         upper_limit <<= bits;
1849
1850         res += 1LL << (bits-2);
1851         res += 1LL << (2*(bits-2));
1852         res += 1LL << (3*(bits-2));
1853         res <<= bits;
1854         if (res > upper_limit)
1855                 res = upper_limit;
1856
1857         if (res > MAX_LFS_FILESIZE)
1858                 res = MAX_LFS_FILESIZE;
1859
1860         return res;
1861 }
1862
1863 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1864                                 ext4_fsblk_t logical_sb_block, int nr)
1865 {
1866         struct ext4_sb_info *sbi = EXT4_SB(sb);
1867         ext4_group_t bg, first_meta_bg;
1868         int has_super = 0;
1869
1870         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1871
1872         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1873             nr < first_meta_bg)
1874                 return logical_sb_block + nr + 1;
1875         bg = sbi->s_desc_per_block * nr;
1876         if (ext4_bg_has_super(sb, bg))
1877                 has_super = 1;
1878         return (has_super + ext4_group_first_block_no(sb, bg));
1879 }
1880
1881 /**
1882  * ext4_get_stripe_size: Get the stripe size.
1883  * @sbi: In memory super block info
1884  *
1885  * If we have specified it via mount option, then
1886  * use the mount option value. If the value specified at mount time is
1887  * greater than the blocks per group use the super block value.
1888  * If the super block value is greater than blocks per group return 0.
1889  * Allocator needs it be less than blocks per group.
1890  *
1891  */
1892 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1893 {
1894         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1895         unsigned long stripe_width =
1896                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1897
1898         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1899                 return sbi->s_stripe;
1900
1901         if (stripe_width <= sbi->s_blocks_per_group)
1902                 return stripe_width;
1903
1904         if (stride <= sbi->s_blocks_per_group)
1905                 return stride;
1906
1907         return 0;
1908 }
1909
1910 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
1911                                 __releases(kernel_lock)
1912                                 __acquires(kernel_lock)
1913
1914 {
1915         struct buffer_head *bh;
1916         struct ext4_super_block *es = NULL;
1917         struct ext4_sb_info *sbi;
1918         ext4_fsblk_t block;
1919         ext4_fsblk_t sb_block = get_sb_block(&data);
1920         ext4_fsblk_t logical_sb_block;
1921         unsigned long offset = 0;
1922         unsigned int journal_inum = 0;
1923         unsigned long journal_devnum = 0;
1924         unsigned long def_mount_opts;
1925         struct inode *root;
1926         char *cp;
1927         int ret = -EINVAL;
1928         int blocksize;
1929         int db_count;
1930         int i;
1931         int needs_recovery;
1932         __le32 features;
1933         __u64 blocks_count;
1934         int err;
1935
1936         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1937         if (!sbi)
1938                 return -ENOMEM;
1939         sb->s_fs_info = sbi;
1940         sbi->s_mount_opt = 0;
1941         sbi->s_resuid = EXT4_DEF_RESUID;
1942         sbi->s_resgid = EXT4_DEF_RESGID;
1943         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
1944         sbi->s_sb_block = sb_block;
1945
1946         unlock_kernel();
1947
1948         /* Cleanup superblock name */
1949         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
1950                 *cp = '!';
1951
1952         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1953         if (!blocksize) {
1954                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1955                 goto out_fail;
1956         }
1957
1958         /*
1959          * The ext4 superblock will not be buffer aligned for other than 1kB
1960          * block sizes.  We need to calculate the offset from buffer start.
1961          */
1962         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1963                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1964                 offset = do_div(logical_sb_block, blocksize);
1965         } else {
1966                 logical_sb_block = sb_block;
1967         }
1968
1969         if (!(bh = sb_bread(sb, logical_sb_block))) {
1970                 printk(KERN_ERR "EXT4-fs: unable to read superblock\n");
1971                 goto out_fail;
1972         }
1973         /*
1974          * Note: s_es must be initialized as soon as possible because
1975          *       some ext4 macro-instructions depend on its value
1976          */
1977         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1978         sbi->s_es = es;
1979         sb->s_magic = le16_to_cpu(es->s_magic);
1980         if (sb->s_magic != EXT4_SUPER_MAGIC)
1981                 goto cantfind_ext4;
1982
1983         /* Set defaults before we parse the mount options */
1984         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1985         if (def_mount_opts & EXT4_DEFM_DEBUG)
1986                 set_opt(sbi->s_mount_opt, DEBUG);
1987         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1988                 set_opt(sbi->s_mount_opt, GRPID);
1989         if (def_mount_opts & EXT4_DEFM_UID16)
1990                 set_opt(sbi->s_mount_opt, NO_UID32);
1991 #ifdef CONFIG_EXT4DEV_FS_XATTR
1992         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1993                 set_opt(sbi->s_mount_opt, XATTR_USER);
1994 #endif
1995 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1996         if (def_mount_opts & EXT4_DEFM_ACL)
1997                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1998 #endif
1999         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2000                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
2001         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2002                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
2003         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2004                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
2005
2006         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2007                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2008         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2009                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2010         else
2011                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2012
2013         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2014         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2015
2016         set_opt(sbi->s_mount_opt, RESERVATION);
2017         set_opt(sbi->s_mount_opt, BARRIER);
2018
2019         /*
2020          * turn on extents feature by default in ext4 filesystem
2021          * only if feature flag already set by mkfs or tune2fs.
2022          * Use -o noextents to turn it off
2023          */
2024         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2025                 set_opt(sbi->s_mount_opt, EXTENTS);
2026         else
2027                 ext4_warning(sb, __func__,
2028                         "extents feature not enabled on this filesystem, "
2029                         "use tune2fs.\n");
2030
2031         /*
2032          * enable delayed allocation by default
2033          * Use -o nodelalloc to turn it off
2034          */
2035         set_opt(sbi->s_mount_opt, DELALLOC);
2036
2037
2038         if (!parse_options((char *) data, sb, &journal_inum, &journal_devnum,
2039                            NULL, 0))
2040                 goto failed_mount;
2041
2042         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2043                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2044
2045         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2046             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2047              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2048              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2049                 printk(KERN_WARNING
2050                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2051                        "running e2fsck is recommended\n");
2052
2053         /*
2054          * Since ext4 is still considered development code, we require
2055          * that the TEST_FILESYS flag in s->flags be set.
2056          */
2057         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
2058                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
2059                        "OK to use with test code.\n", sb->s_id);
2060                 goto failed_mount;
2061         }
2062
2063         /*
2064          * Check feature flags regardless of the revision level, since we
2065          * previously didn't change the revision level when setting the flags,
2066          * so there is a chance incompat flags are set on a rev 0 filesystem.
2067          */
2068         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2069         if (features) {
2070                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2071                        "unsupported optional features (%x).\n",
2072                        sb->s_id, le32_to_cpu(features));
2073                 goto failed_mount;
2074         }
2075         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2076         if (!(sb->s_flags & MS_RDONLY) && features) {
2077                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2078                        "unsupported optional features (%x).\n",
2079                        sb->s_id, le32_to_cpu(features));
2080                 goto failed_mount;
2081         }
2082         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2083                 /*
2084                  * Large file size enabled file system can only be
2085                  * mount if kernel is build with CONFIG_LSF
2086                  */
2087                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2088                                 !(sb->s_flags & MS_RDONLY)) {
2089                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2090                                         "files cannot be mounted read-write "
2091                                         "without CONFIG_LSF.\n", sb->s_id);
2092                         goto failed_mount;
2093                 }
2094         }
2095         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2096
2097         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2098             blocksize > EXT4_MAX_BLOCK_SIZE) {
2099                 printk(KERN_ERR
2100                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2101                        blocksize, sb->s_id);
2102                 goto failed_mount;
2103         }
2104
2105         if (sb->s_blocksize != blocksize) {
2106
2107                 /* Validate the filesystem blocksize */
2108                 if (!sb_set_blocksize(sb, blocksize)) {
2109                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2110                                         blocksize);
2111                         goto failed_mount;
2112                 }
2113
2114                 brelse(bh);
2115                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2116                 offset = do_div(logical_sb_block, blocksize);
2117                 bh = sb_bread(sb, logical_sb_block);
2118                 if (!bh) {
2119                         printk(KERN_ERR
2120                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2121                         goto failed_mount;
2122                 }
2123                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2124                 sbi->s_es = es;
2125                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2126                         printk(KERN_ERR
2127                                "EXT4-fs: Magic mismatch, very weird !\n");
2128                         goto failed_mount;
2129                 }
2130         }
2131
2132         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2133         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2134
2135         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2136                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2137                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2138         } else {
2139                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2140                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2141                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2142                     (!is_power_of_2(sbi->s_inode_size)) ||
2143                     (sbi->s_inode_size > blocksize)) {
2144                         printk(KERN_ERR
2145                                "EXT4-fs: unsupported inode size: %d\n",
2146                                sbi->s_inode_size);
2147                         goto failed_mount;
2148                 }
2149                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2150                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2151         }
2152         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2153         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2154                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2155                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2156                     !is_power_of_2(sbi->s_desc_size)) {
2157                         printk(KERN_ERR
2158                                "EXT4-fs: unsupported descriptor size %lu\n",
2159                                sbi->s_desc_size);
2160                         goto failed_mount;
2161                 }
2162         } else
2163                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2164         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2165         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2166         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2167                 goto cantfind_ext4;
2168         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2169         if (sbi->s_inodes_per_block == 0)
2170                 goto cantfind_ext4;
2171         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2172                                         sbi->s_inodes_per_block;
2173         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2174         sbi->s_sbh = bh;
2175         sbi->s_mount_state = le16_to_cpu(es->s_state);
2176         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2177         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2178         for (i = 0; i < 4; i++)
2179                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2180         sbi->s_def_hash_version = es->s_def_hash_version;
2181
2182         if (sbi->s_blocks_per_group > blocksize * 8) {
2183                 printk(KERN_ERR
2184                        "EXT4-fs: #blocks per group too big: %lu\n",
2185                        sbi->s_blocks_per_group);
2186                 goto failed_mount;
2187         }
2188         if (sbi->s_inodes_per_group > blocksize * 8) {
2189                 printk(KERN_ERR
2190                        "EXT4-fs: #inodes per group too big: %lu\n",
2191                        sbi->s_inodes_per_group);
2192                 goto failed_mount;
2193         }
2194
2195         if (ext4_blocks_count(es) >
2196                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2197                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2198                         " too large to mount safely\n", sb->s_id);
2199                 if (sizeof(sector_t) < 8)
2200                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2201                                         "enabled\n");
2202                 goto failed_mount;
2203         }
2204
2205         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2206                 goto cantfind_ext4;
2207
2208         /* ensure blocks_count calculation below doesn't sign-extend */
2209         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2210             le32_to_cpu(es->s_first_data_block) + 1) {
2211                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2212                        "first data block %u, blocks per group %lu\n",
2213                         ext4_blocks_count(es),
2214                         le32_to_cpu(es->s_first_data_block),
2215                         EXT4_BLOCKS_PER_GROUP(sb));
2216                 goto failed_mount;
2217         }
2218         blocks_count = (ext4_blocks_count(es) -
2219                         le32_to_cpu(es->s_first_data_block) +
2220                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2221         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2222         sbi->s_groups_count = blocks_count;
2223         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2224                    EXT4_DESC_PER_BLOCK(sb);
2225         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2226                                     GFP_KERNEL);
2227         if (sbi->s_group_desc == NULL) {
2228                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
2229                 goto failed_mount;
2230         }
2231
2232         if (ext4_proc_root)
2233                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2234
2235         if (sbi->s_proc)
2236                 proc_create_data("inode_readahead_blks", 0644, sbi->s_proc,
2237                                  &ext4_ui_proc_fops,
2238                                  &sbi->s_inode_readahead_blks);
2239
2240         bgl_lock_init(&sbi->s_blockgroup_lock);
2241
2242         for (i = 0; i < db_count; i++) {
2243                 block = descriptor_loc(sb, logical_sb_block, i);
2244                 sbi->s_group_desc[i] = sb_bread(sb, block);
2245                 if (!sbi->s_group_desc[i]) {
2246                         printk(KERN_ERR "EXT4-fs: "
2247                                "can't read group descriptor %d\n", i);
2248                         db_count = i;
2249                         goto failed_mount2;
2250                 }
2251         }
2252         if (!ext4_check_descriptors(sb)) {
2253                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2254                 goto failed_mount2;
2255         }
2256         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2257                 if (!ext4_fill_flex_info(sb)) {
2258                         printk(KERN_ERR
2259                                "EXT4-fs: unable to initialize "
2260                                "flex_bg meta info!\n");
2261                         goto failed_mount2;
2262                 }
2263
2264         sbi->s_gdb_count = db_count;
2265         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2266         spin_lock_init(&sbi->s_next_gen_lock);
2267
2268         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2269                         ext4_count_free_blocks(sb));
2270         if (!err) {
2271                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2272                                 ext4_count_free_inodes(sb));
2273         }
2274         if (!err) {
2275                 err = percpu_counter_init(&sbi->s_dirs_counter,
2276                                 ext4_count_dirs(sb));
2277         }
2278         if (!err) {
2279                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2280         }
2281         if (err) {
2282                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2283                 goto failed_mount3;
2284         }
2285
2286         sbi->s_stripe = ext4_get_stripe_size(sbi);
2287
2288         /*
2289          * set up enough so that it can read an inode
2290          */
2291         sb->s_op = &ext4_sops;
2292         sb->s_export_op = &ext4_export_ops;
2293         sb->s_xattr = ext4_xattr_handlers;
2294 #ifdef CONFIG_QUOTA
2295         sb->s_qcop = &ext4_qctl_operations;
2296         sb->dq_op = &ext4_quota_operations;
2297 #endif
2298         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2299
2300         sb->s_root = NULL;
2301
2302         needs_recovery = (es->s_last_orphan != 0 ||
2303                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2304                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2305
2306         /*
2307          * The first inode we look at is the journal inode.  Don't try
2308          * root first: it may be modified in the journal!
2309          */
2310         if (!test_opt(sb, NOLOAD) &&
2311             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2312                 if (ext4_load_journal(sb, es, journal_devnum))
2313                         goto failed_mount3;
2314                 if (!(sb->s_flags & MS_RDONLY) &&
2315                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2316                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2317                                "ext4_fill_super: Journal transaction "
2318                                "%u is corrupt\n", sb->s_id,
2319                                EXT4_SB(sb)->s_journal->j_failed_commit);
2320                         if (test_opt(sb, ERRORS_RO)) {
2321                                 printk(KERN_CRIT
2322                                        "Mounting filesystem read-only\n");
2323                                 sb->s_flags |= MS_RDONLY;
2324                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2325                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2326                         }
2327                         if (test_opt(sb, ERRORS_PANIC)) {
2328                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2329                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2330                                 ext4_commit_super(sb, es, 1);
2331                                 printk(KERN_CRIT
2332                                        "EXT4-fs (device %s): mount failed\n",
2333                                       sb->s_id);
2334                                 goto failed_mount4;
2335                         }
2336                 }
2337         } else if (journal_inum) {
2338                 if (ext4_create_journal(sb, es, journal_inum))
2339                         goto failed_mount3;
2340         } else {
2341                 if (!silent)
2342                         printk(KERN_ERR
2343                                "ext4: No journal on filesystem on %s\n",
2344                                sb->s_id);
2345                 goto failed_mount3;
2346         }
2347
2348         if (ext4_blocks_count(es) > 0xffffffffULL &&
2349             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2350                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2351                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2352                 goto failed_mount4;
2353         }
2354
2355         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2356                 jbd2_journal_set_features(sbi->s_journal,
2357                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2358                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2359         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2360                 jbd2_journal_set_features(sbi->s_journal,
2361                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2362                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2363                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2364         } else {
2365                 jbd2_journal_clear_features(sbi->s_journal,
2366                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2367                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2368         }
2369
2370         /* We have now updated the journal if required, so we can
2371          * validate the data journaling mode. */
2372         switch (test_opt(sb, DATA_FLAGS)) {
2373         case 0:
2374                 /* No mode set, assume a default based on the journal
2375                  * capabilities: ORDERED_DATA if the journal can
2376                  * cope, else JOURNAL_DATA
2377                  */
2378                 if (jbd2_journal_check_available_features
2379                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2380                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2381                 else
2382                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2383                 break;
2384
2385         case EXT4_MOUNT_ORDERED_DATA:
2386         case EXT4_MOUNT_WRITEBACK_DATA:
2387                 if (!jbd2_journal_check_available_features
2388                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2389                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2390                                "requested data journaling mode\n");
2391                         goto failed_mount4;
2392                 }
2393         default:
2394                 break;
2395         }
2396
2397         if (test_opt(sb, NOBH)) {
2398                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2399                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2400                                 "its supported only with writeback mode\n");
2401                         clear_opt(sbi->s_mount_opt, NOBH);
2402                 }
2403         }
2404         /*
2405          * The jbd2_journal_load will have done any necessary log recovery,
2406          * so we can safely mount the rest of the filesystem now.
2407          */
2408
2409         root = ext4_iget(sb, EXT4_ROOT_INO);
2410         if (IS_ERR(root)) {
2411                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2412                 ret = PTR_ERR(root);
2413                 goto failed_mount4;
2414         }
2415         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2416                 iput(root);
2417                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2418                 goto failed_mount4;
2419         }
2420         sb->s_root = d_alloc_root(root);
2421         if (!sb->s_root) {
2422                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2423                 iput(root);
2424                 ret = -ENOMEM;
2425                 goto failed_mount4;
2426         }
2427
2428         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2429
2430         /* determine the minimum size of new large inodes, if present */
2431         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2432                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2433                                                      EXT4_GOOD_OLD_INODE_SIZE;
2434                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2435                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2436                         if (sbi->s_want_extra_isize <
2437                             le16_to_cpu(es->s_want_extra_isize))
2438                                 sbi->s_want_extra_isize =
2439                                         le16_to_cpu(es->s_want_extra_isize);
2440                         if (sbi->s_want_extra_isize <
2441                             le16_to_cpu(es->s_min_extra_isize))
2442                                 sbi->s_want_extra_isize =
2443                                         le16_to_cpu(es->s_min_extra_isize);
2444                 }
2445         }
2446         /* Check if enough inode space is available */
2447         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2448                                                         sbi->s_inode_size) {
2449                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2450                                                        EXT4_GOOD_OLD_INODE_SIZE;
2451                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2452                         "available.\n");
2453         }
2454
2455         /*
2456          * akpm: core read_super() calls in here with the superblock locked.
2457          * That deadlocks, because orphan cleanup needs to lock the superblock
2458          * in numerous places.  Here we just pop the lock - it's relatively
2459          * harmless, because we are now ready to accept write_super() requests,
2460          * and aviro says that's the only reason for hanging onto the
2461          * superblock lock.
2462          */
2463         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2464         ext4_orphan_cleanup(sb, es);
2465         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2466         if (needs_recovery)
2467                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2468         ext4_mark_recovery_complete(sb, es);
2469         printk(KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2470                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2471                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2472                "writeback");
2473
2474         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2475                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2476                                 "requested data journaling mode\n");
2477                 clear_opt(sbi->s_mount_opt, DELALLOC);
2478         } else if (test_opt(sb, DELALLOC))
2479                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2480
2481         ext4_ext_init(sb);
2482         err = ext4_mb_init(sb, needs_recovery);
2483         if (err) {
2484                 printk(KERN_ERR "EXT4-fs: failed to initalize mballoc (%d)\n",
2485                        err);
2486                 goto failed_mount4;
2487         }
2488
2489         lock_kernel();
2490         return 0;
2491
2492 cantfind_ext4:
2493         if (!silent)
2494                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2495                        sb->s_id);
2496         goto failed_mount;
2497
2498 failed_mount4:
2499         jbd2_journal_destroy(sbi->s_journal);
2500         sbi->s_journal = NULL;
2501 failed_mount3:
2502         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2503         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2504         percpu_counter_destroy(&sbi->s_dirs_counter);
2505         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2506 failed_mount2:
2507         for (i = 0; i < db_count; i++)
2508                 brelse(sbi->s_group_desc[i]);
2509         kfree(sbi->s_group_desc);
2510 failed_mount:
2511         if (sbi->s_proc) {
2512                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
2513                 remove_proc_entry(sb->s_id, ext4_proc_root);
2514         }
2515 #ifdef CONFIG_QUOTA
2516         for (i = 0; i < MAXQUOTAS; i++)
2517                 kfree(sbi->s_qf_names[i]);
2518 #endif
2519         ext4_blkdev_remove(sbi);
2520         brelse(bh);
2521 out_fail:
2522         sb->s_fs_info = NULL;
2523         kfree(sbi);
2524         lock_kernel();
2525         return ret;
2526 }
2527
2528 /*
2529  * Setup any per-fs journal parameters now.  We'll do this both on
2530  * initial mount, once the journal has been initialised but before we've
2531  * done any recovery; and again on any subsequent remount.
2532  */
2533 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2534 {
2535         struct ext4_sb_info *sbi = EXT4_SB(sb);
2536
2537         if (sbi->s_commit_interval)
2538                 journal->j_commit_interval = sbi->s_commit_interval;
2539         /* We could also set up an ext4-specific default for the commit
2540          * interval here, but for now we'll just fall back to the jbd
2541          * default. */
2542
2543         spin_lock(&journal->j_state_lock);
2544         if (test_opt(sb, BARRIER))
2545                 journal->j_flags |= JBD2_BARRIER;
2546         else
2547                 journal->j_flags &= ~JBD2_BARRIER;
2548         spin_unlock(&journal->j_state_lock);
2549 }
2550
2551 static journal_t *ext4_get_journal(struct super_block *sb,
2552                                    unsigned int journal_inum)
2553 {
2554         struct inode *journal_inode;
2555         journal_t *journal;
2556
2557         /* First, test for the existence of a valid inode on disk.  Bad
2558          * things happen if we iget() an unused inode, as the subsequent
2559          * iput() will try to delete it. */
2560
2561         journal_inode = ext4_iget(sb, journal_inum);
2562         if (IS_ERR(journal_inode)) {
2563                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2564                 return NULL;
2565         }
2566         if (!journal_inode->i_nlink) {
2567                 make_bad_inode(journal_inode);
2568                 iput(journal_inode);
2569                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2570                 return NULL;
2571         }
2572
2573         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
2574                   journal_inode, journal_inode->i_size);
2575         if (!S_ISREG(journal_inode->i_mode)) {
2576                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2577                 iput(journal_inode);
2578                 return NULL;
2579         }
2580
2581         journal = jbd2_journal_init_inode(journal_inode);
2582         if (!journal) {
2583                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2584                 iput(journal_inode);
2585                 return NULL;
2586         }
2587         journal->j_private = sb;
2588         ext4_init_journal_params(sb, journal);
2589         return journal;
2590 }
2591
2592 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2593                                        dev_t j_dev)
2594 {
2595         struct buffer_head *bh;
2596         journal_t *journal;
2597         ext4_fsblk_t start;
2598         ext4_fsblk_t len;
2599         int hblock, blocksize;
2600         ext4_fsblk_t sb_block;
2601         unsigned long offset;
2602         struct ext4_super_block *es;
2603         struct block_device *bdev;
2604
2605         bdev = ext4_blkdev_get(j_dev);
2606         if (bdev == NULL)
2607                 return NULL;
2608
2609         if (bd_claim(bdev, sb)) {
2610                 printk(KERN_ERR
2611                         "EXT4: failed to claim external journal device.\n");
2612                 blkdev_put(bdev);
2613                 return NULL;
2614         }
2615
2616         blocksize = sb->s_blocksize;
2617         hblock = bdev_hardsect_size(bdev);
2618         if (blocksize < hblock) {
2619                 printk(KERN_ERR
2620                         "EXT4-fs: blocksize too small for journal device.\n");
2621                 goto out_bdev;
2622         }
2623
2624         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2625         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2626         set_blocksize(bdev, blocksize);
2627         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2628                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2629                        "external journal\n");
2630                 goto out_bdev;
2631         }
2632
2633         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2634         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2635             !(le32_to_cpu(es->s_feature_incompat) &
2636               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2637                 printk(KERN_ERR "EXT4-fs: external journal has "
2638                                         "bad superblock\n");
2639                 brelse(bh);
2640                 goto out_bdev;
2641         }
2642
2643         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2644                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2645                 brelse(bh);
2646                 goto out_bdev;
2647         }
2648
2649         len = ext4_blocks_count(es);
2650         start = sb_block + 1;
2651         brelse(bh);     /* we're done with the superblock */
2652
2653         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2654                                         start, len, blocksize);
2655         if (!journal) {
2656                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2657                 goto out_bdev;
2658         }
2659         journal->j_private = sb;
2660         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2661         wait_on_buffer(journal->j_sb_buffer);
2662         if (!buffer_uptodate(journal->j_sb_buffer)) {
2663                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2664                 goto out_journal;
2665         }
2666         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2667                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2668                                         "user (unsupported) - %d\n",
2669                         be32_to_cpu(journal->j_superblock->s_nr_users));
2670                 goto out_journal;
2671         }
2672         EXT4_SB(sb)->journal_bdev = bdev;
2673         ext4_init_journal_params(sb, journal);
2674         return journal;
2675 out_journal:
2676         jbd2_journal_destroy(journal);
2677 out_bdev:
2678         ext4_blkdev_put(bdev);
2679         return NULL;
2680 }
2681
2682 static int ext4_load_journal(struct super_block *sb,
2683                              struct ext4_super_block *es,
2684                              unsigned long journal_devnum)
2685 {
2686         journal_t *journal;
2687         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2688         dev_t journal_dev;
2689         int err = 0;
2690         int really_read_only;
2691
2692         if (journal_devnum &&
2693             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2694                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2695                         "numbers have changed\n");
2696                 journal_dev = new_decode_dev(journal_devnum);
2697         } else
2698                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2699
2700         really_read_only = bdev_read_only(sb->s_bdev);
2701
2702         /*
2703          * Are we loading a blank journal or performing recovery after a
2704          * crash?  For recovery, we need to check in advance whether we
2705          * can get read-write access to the device.
2706          */
2707
2708         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2709                 if (sb->s_flags & MS_RDONLY) {
2710                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2711                                         "required on readonly filesystem.\n");
2712                         if (really_read_only) {
2713                                 printk(KERN_ERR "EXT4-fs: write access "
2714                                         "unavailable, cannot proceed.\n");
2715                                 return -EROFS;
2716                         }
2717                         printk(KERN_INFO "EXT4-fs: write access will "
2718                                "be enabled during recovery.\n");
2719                 }
2720         }
2721
2722         if (journal_inum && journal_dev) {
2723                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2724                        "and inode journals!\n");
2725                 return -EINVAL;
2726         }
2727
2728         if (journal_inum) {
2729                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2730                         return -EINVAL;
2731         } else {
2732                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2733                         return -EINVAL;
2734         }
2735
2736         if (journal->j_flags & JBD2_BARRIER)
2737                 printk(KERN_INFO "EXT4-fs: barriers enabled\n");
2738         else
2739                 printk(KERN_INFO "EXT4-fs: barriers disabled\n");
2740
2741         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2742                 err = jbd2_journal_update_format(journal);
2743                 if (err)  {
2744                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2745                         jbd2_journal_destroy(journal);
2746                         return err;
2747                 }
2748         }
2749
2750         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2751                 err = jbd2_journal_wipe(journal, !really_read_only);
2752         if (!err)
2753                 err = jbd2_journal_load(journal);
2754
2755         if (err) {
2756                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2757                 jbd2_journal_destroy(journal);
2758                 return err;
2759         }
2760
2761         EXT4_SB(sb)->s_journal = journal;
2762         ext4_clear_journal_err(sb, es);
2763
2764         if (journal_devnum &&
2765             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2766                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2767                 sb->s_dirt = 1;
2768
2769                 /* Make sure we flush the recovery flag to disk. */
2770                 ext4_commit_super(sb, es, 1);
2771         }
2772
2773         return 0;
2774 }
2775
2776 static int ext4_create_journal(struct super_block *sb,
2777                                struct ext4_super_block *es,
2778                                unsigned int journal_inum)
2779 {
2780         journal_t *journal;
2781         int err;
2782
2783         if (sb->s_flags & MS_RDONLY) {
2784                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2785                                 "create journal.\n");
2786                 return -EROFS;
2787         }
2788
2789         journal = ext4_get_journal(sb, journal_inum);
2790         if (!journal)
2791                 return -EINVAL;
2792
2793         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2794                journal_inum);
2795
2796         err = jbd2_journal_create(journal);
2797         if (err) {
2798                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2799                 jbd2_journal_destroy(journal);
2800                 return -EIO;
2801         }
2802
2803         EXT4_SB(sb)->s_journal = journal;
2804
2805         ext4_update_dynamic_rev(sb);
2806         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2807         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2808
2809         es->s_journal_inum = cpu_to_le32(journal_inum);
2810         sb->s_dirt = 1;
2811
2812         /* Make sure we flush the recovery flag to disk. */
2813         ext4_commit_super(sb, es, 1);
2814
2815         return 0;
2816 }
2817
2818 static void ext4_commit_super(struct super_block *sb,
2819                               struct ext4_super_block *es, int sync)
2820 {
2821         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2822
2823         if (!sbh)
2824                 return;
2825         if (buffer_write_io_error(sbh)) {
2826                 /*
2827                  * Oh, dear.  A previous attempt to write the
2828                  * superblock failed.  This could happen because the
2829                  * USB device was yanked out.  Or it could happen to
2830                  * be a transient write error and maybe the block will
2831                  * be remapped.  Nothing we can do but to retry the
2832                  * write and hope for the best.
2833                  */
2834                 printk(KERN_ERR "ext4: previous I/O error to "
2835                        "superblock detected for %s.\n", sb->s_id);
2836                 clear_buffer_write_io_error(sbh);
2837                 set_buffer_uptodate(sbh);
2838         }
2839         es->s_wtime = cpu_to_le32(get_seconds());
2840         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2841         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2842         BUFFER_TRACE(sbh, "marking dirty");
2843         mark_buffer_dirty(sbh);
2844         if (sync) {
2845                 sync_dirty_buffer(sbh);
2846                 if (buffer_write_io_error(sbh)) {
2847                         printk(KERN_ERR "ext4: I/O error while writing "
2848                                "superblock for %s.\n", sb->s_id);
2849                         clear_buffer_write_io_error(sbh);
2850                         set_buffer_uptodate(sbh);
2851                 }
2852         }
2853 }
2854
2855
2856 /*
2857  * Have we just finished recovery?  If so, and if we are mounting (or
2858  * remounting) the filesystem readonly, then we will end up with a
2859  * consistent fs on disk.  Record that fact.
2860  */
2861 static void ext4_mark_recovery_complete(struct super_block *sb,
2862                                         struct ext4_super_block *es)
2863 {
2864         journal_t *journal = EXT4_SB(sb)->s_journal;
2865
2866         jbd2_journal_lock_updates(journal);
2867         jbd2_journal_flush(journal);
2868         lock_super(sb);
2869         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2870             sb->s_flags & MS_RDONLY) {
2871                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2872                 sb->s_dirt = 0;
2873                 ext4_commit_super(sb, es, 1);
2874         }
2875         unlock_super(sb);
2876         jbd2_journal_unlock_updates(journal);
2877 }
2878
2879 /*
2880  * If we are mounting (or read-write remounting) a filesystem whose journal
2881  * has recorded an error from a previous lifetime, move that error to the
2882  * main filesystem now.
2883  */
2884 static void ext4_clear_journal_err(struct super_block *sb,
2885                                    struct ext4_super_block *es)
2886 {
2887         journal_t *journal;
2888         int j_errno;
2889         const char *errstr;
2890
2891         journal = EXT4_SB(sb)->s_journal;
2892
2893         /*
2894          * Now check for any error status which may have been recorded in the
2895          * journal by a prior ext4_error() or ext4_abort()
2896          */
2897
2898         j_errno = jbd2_journal_errno(journal);
2899         if (j_errno) {
2900                 char nbuf[16];
2901
2902                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2903                 ext4_warning(sb, __func__, "Filesystem error recorded "
2904                              "from previous mount: %s", errstr);
2905                 ext4_warning(sb, __func__, "Marking fs in need of "
2906                              "filesystem check.");
2907
2908                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2909                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2910                 ext4_commit_super(sb, es, 1);
2911
2912                 jbd2_journal_clear_err(journal);
2913         }
2914 }
2915
2916 /*
2917  * Force the running and committing transactions to commit,
2918  * and wait on the commit.
2919  */
2920 int ext4_force_commit(struct super_block *sb)
2921 {
2922         journal_t *journal;
2923         int ret;
2924
2925         if (sb->s_flags & MS_RDONLY)
2926                 return 0;
2927
2928         journal = EXT4_SB(sb)->s_journal;
2929         sb->s_dirt = 0;
2930         ret = ext4_journal_force_commit(journal);
2931         return ret;
2932 }
2933
2934 /*
2935  * Ext4 always journals updates to the superblock itself, so we don't
2936  * have to propagate any other updates to the superblock on disk at this
2937  * point.  Just start an async writeback to get the buffers on their way
2938  * to the disk.
2939  *
2940  * This implicitly triggers the writebehind on sync().
2941  */
2942
2943 static void ext4_write_super(struct super_block *sb)
2944 {
2945         if (mutex_trylock(&sb->s_lock) != 0)
2946                 BUG();
2947         sb->s_dirt = 0;
2948 }
2949
2950 static int ext4_sync_fs(struct super_block *sb, int wait)
2951 {
2952         tid_t target;
2953
2954         sb->s_dirt = 0;
2955         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2956                 if (wait)
2957                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2958         }
2959         return 0;
2960 }
2961
2962 /*
2963  * LVM calls this function before a (read-only) snapshot is created.  This
2964  * gives us a chance to flush the journal completely and mark the fs clean.
2965  */
2966 static void ext4_write_super_lockfs(struct super_block *sb)
2967 {
2968         sb->s_dirt = 0;
2969
2970         if (!(sb->s_flags & MS_RDONLY)) {
2971                 journal_t *journal = EXT4_SB(sb)->s_journal;
2972
2973                 /* Now we set up the journal barrier. */
2974                 jbd2_journal_lock_updates(journal);
2975                 jbd2_journal_flush(journal);
2976
2977                 /* Journal blocked and flushed, clear needs_recovery flag. */
2978                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2979                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2980         }
2981 }
2982
2983 /*
2984  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2985  * flag here, even though the filesystem is not technically dirty yet.
2986  */
2987 static void ext4_unlockfs(struct super_block *sb)
2988 {
2989         if (!(sb->s_flags & MS_RDONLY)) {
2990                 lock_super(sb);
2991                 /* Reser the needs_recovery flag before the fs is unlocked. */
2992                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2993                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2994                 unlock_super(sb);
2995                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2996         }
2997 }
2998
2999 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3000 {
3001         struct ext4_super_block *es;
3002         struct ext4_sb_info *sbi = EXT4_SB(sb);
3003         ext4_fsblk_t n_blocks_count = 0;
3004         unsigned long old_sb_flags;
3005         struct ext4_mount_options old_opts;
3006         ext4_group_t g;
3007         int err;
3008 #ifdef CONFIG_QUOTA
3009         int i;
3010 #endif
3011
3012         /* Store the original options */
3013         old_sb_flags = sb->s_flags;
3014         old_opts.s_mount_opt = sbi->s_mount_opt;
3015         old_opts.s_resuid = sbi->s_resuid;
3016         old_opts.s_resgid = sbi->s_resgid;
3017         old_opts.s_commit_interval = sbi->s_commit_interval;
3018 #ifdef CONFIG_QUOTA
3019         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3020         for (i = 0; i < MAXQUOTAS; i++)
3021                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3022 #endif
3023
3024         /*
3025          * Allow the "check" option to be passed as a remount option.
3026          */
3027         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
3028                 err = -EINVAL;
3029                 goto restore_opts;
3030         }
3031
3032         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
3033                 ext4_abort(sb, __func__, "Abort forced by user");
3034
3035         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3036                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3037
3038         es = sbi->s_es;
3039
3040         ext4_init_journal_params(sb, sbi->s_journal);
3041
3042         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3043                 n_blocks_count > ext4_blocks_count(es)) {
3044                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3045                         err = -EROFS;
3046                         goto restore_opts;
3047                 }
3048
3049                 if (*flags & MS_RDONLY) {
3050                         /*
3051                          * First of all, the unconditional stuff we have to do
3052                          * to disable replay of the journal when we next remount
3053                          */
3054                         sb->s_flags |= MS_RDONLY;
3055
3056                         /*
3057                          * OK, test if we are remounting a valid rw partition
3058                          * readonly, and if so set the rdonly flag and then
3059                          * mark the partition as valid again.
3060                          */
3061                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3062                             (sbi->s_mount_state & EXT4_VALID_FS))
3063                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3064
3065                         /*
3066                          * We have to unlock super so that we can wait for
3067                          * transactions.
3068                          */
3069                         unlock_super(sb);
3070                         ext4_mark_recovery_complete(sb, es);
3071                         lock_super(sb);
3072                 } else {
3073                         __le32 ret;
3074                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3075                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3076                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3077                                        "remount RDWR because of unsupported "
3078                                        "optional features (%x).\n",
3079                                        sb->s_id, le32_to_cpu(ret));
3080                                 err = -EROFS;
3081                                 goto restore_opts;
3082                         }
3083
3084                         /*
3085                          * Make sure the group descriptor checksums
3086                          * are sane.  If they aren't, refuse to
3087                          * remount r/w.
3088                          */
3089                         for (g = 0; g < sbi->s_groups_count; g++) {
3090                                 struct ext4_group_desc *gdp =
3091                                         ext4_get_group_desc(sb, g, NULL);
3092
3093                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3094                                         printk(KERN_ERR
3095                "EXT4-fs: ext4_remount: "
3096                 "Checksum for group %lu failed (%u!=%u)\n",
3097                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3098                                                le16_to_cpu(gdp->bg_checksum));
3099                                         err = -EINVAL;
3100                                         goto restore_opts;
3101                                 }
3102                         }
3103
3104                         /*
3105                          * If we have an unprocessed orphan list hanging
3106                          * around from a previously readonly bdev mount,
3107                          * require a full umount/remount for now.
3108                          */
3109                         if (es->s_last_orphan) {
3110                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3111                                        "remount RDWR because of unprocessed "
3112                                        "orphan inode list.  Please "
3113                                        "umount/remount instead.\n",
3114                                        sb->s_id);
3115                                 err = -EINVAL;
3116                                 goto restore_opts;
3117                         }
3118
3119                         /*
3120                          * Mounting a RDONLY partition read-write, so reread
3121                          * and store the current valid flag.  (It may have
3122                          * been changed by e2fsck since we originally mounted
3123                          * the partition.)
3124                          */
3125                         ext4_clear_journal_err(sb, es);
3126                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3127                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3128                                 goto restore_opts;
3129                         if (!ext4_setup_super(sb, es, 0))
3130                                 sb->s_flags &= ~MS_RDONLY;
3131                 }
3132         }
3133 #ifdef CONFIG_QUOTA
3134         /* Release old quota file names */
3135         for (i = 0; i < MAXQUOTAS; i++)
3136                 if (old_opts.s_qf_names[i] &&
3137                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3138                         kfree(old_opts.s_qf_names[i]);
3139 #endif
3140         return 0;
3141 restore_opts:
3142         sb->s_flags = old_sb_flags;
3143         sbi->s_mount_opt = old_opts.s_mount_opt;
3144         sbi->s_resuid = old_opts.s_resuid;
3145         sbi->s_resgid = old_opts.s_resgid;
3146         sbi->s_commit_interval = old_opts.s_commit_interval;
3147 #ifdef CONFIG_QUOTA
3148         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3149         for (i = 0; i < MAXQUOTAS; i++) {
3150                 if (sbi->s_qf_names[i] &&
3151                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3152                         kfree(sbi->s_qf_names[i]);
3153                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3154         }
3155 #endif
3156         return err;
3157 }
3158
3159 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3160 {
3161         struct super_block *sb = dentry->d_sb;
3162         struct ext4_sb_info *sbi = EXT4_SB(sb);
3163         struct ext4_super_block *es = sbi->s_es;
3164         u64 fsid;
3165
3166         if (test_opt(sb, MINIX_DF)) {
3167                 sbi->s_overhead_last = 0;
3168         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3169                 ext4_group_t ngroups = sbi->s_groups_count, i;
3170                 ext4_fsblk_t overhead = 0;
3171                 smp_rmb();
3172
3173                 /*
3174                  * Compute the overhead (FS structures).  This is constant
3175                  * for a given filesystem unless the number of block groups
3176                  * changes so we cache the previous value until it does.
3177                  */
3178
3179                 /*
3180                  * All of the blocks before first_data_block are
3181                  * overhead
3182                  */
3183                 overhead = le32_to_cpu(es->s_first_data_block);
3184
3185                 /*
3186                  * Add the overhead attributed to the superblock and
3187                  * block group descriptors.  If the sparse superblocks
3188                  * feature is turned on, then not all groups have this.
3189                  */
3190                 for (i = 0; i < ngroups; i++) {
3191                         overhead += ext4_bg_has_super(sb, i) +
3192                                 ext4_bg_num_gdb(sb, i);
3193                         cond_resched();
3194                 }
3195
3196                 /*
3197                  * Every block group has an inode bitmap, a block
3198                  * bitmap, and an inode table.
3199                  */
3200                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3201                 sbi->s_overhead_last = overhead;
3202                 smp_wmb();
3203                 sbi->s_blocks_last = ext4_blocks_count(es);
3204         }
3205
3206         buf->f_type = EXT4_SUPER_MAGIC;
3207         buf->f_bsize = sb->s_blocksize;
3208         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3209         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3210                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3211         ext4_free_blocks_count_set(es, buf->f_bfree);
3212         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3213         if (buf->f_bfree < ext4_r_blocks_count(es))
3214                 buf->f_bavail = 0;
3215         buf->f_files = le32_to_cpu(es->s_inodes_count);
3216         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3217         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3218         buf->f_namelen = EXT4_NAME_LEN;
3219         fsid = le64_to_cpup((void *)es->s_uuid) ^
3220                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3221         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3222         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3223         return 0;
3224 }
3225
3226 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3227  * is locked for write. Otherwise the are possible deadlocks:
3228  * Process 1                         Process 2
3229  * ext4_create()                     quota_sync()
3230  *   jbd2_journal_start()                   write_dquot()
3231  *   DQUOT_INIT()                        down(dqio_mutex)
3232  *     down(dqio_mutex)                    jbd2_journal_start()
3233  *
3234  */
3235
3236 #ifdef CONFIG_QUOTA
3237
3238 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3239 {
3240         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3241 }
3242
3243 static int ext4_dquot_initialize(struct inode *inode, int type)
3244 {
3245         handle_t *handle;
3246         int ret, err;
3247
3248         /* We may create quota structure so we need to reserve enough blocks */
3249         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3250         if (IS_ERR(handle))
3251                 return PTR_ERR(handle);
3252         ret = dquot_initialize(inode, type);
3253         err = ext4_journal_stop(handle);
3254         if (!ret)
3255                 ret = err;
3256         return ret;
3257 }
3258
3259 static int ext4_dquot_drop(struct inode *inode)
3260 {
3261         handle_t *handle;
3262         int ret, err;
3263
3264         /* We may delete quota structure so we need to reserve enough blocks */
3265         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3266         if (IS_ERR(handle)) {
3267                 /*
3268                  * We call dquot_drop() anyway to at least release references
3269                  * to quota structures so that umount does not hang.
3270                  */
3271                 dquot_drop(inode);
3272                 return PTR_ERR(handle);
3273         }
3274         ret = dquot_drop(inode);
3275         err = ext4_journal_stop(handle);
3276         if (!ret)
3277                 ret = err;
3278         return ret;
3279 }
3280
3281 static int ext4_write_dquot(struct dquot *dquot)
3282 {
3283         int ret, err;
3284         handle_t *handle;
3285         struct inode *inode;
3286
3287         inode = dquot_to_inode(dquot);
3288         handle = ext4_journal_start(inode,
3289                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3290         if (IS_ERR(handle))
3291                 return PTR_ERR(handle);
3292         ret = dquot_commit(dquot);
3293         err = ext4_journal_stop(handle);
3294         if (!ret)
3295                 ret = err;
3296         return ret;
3297 }
3298
3299 static int ext4_acquire_dquot(struct dquot *dquot)
3300 {
3301         int ret, err;
3302         handle_t *handle;
3303
3304         handle = ext4_journal_start(dquot_to_inode(dquot),
3305                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3306         if (IS_ERR(handle))
3307                 return PTR_ERR(handle);
3308         ret = dquot_acquire(dquot);
3309         err = ext4_journal_stop(handle);
3310         if (!ret)
3311                 ret = err;
3312         return ret;
3313 }
3314
3315 static int ext4_release_dquot(struct dquot *dquot)
3316 {
3317         int ret, err;
3318         handle_t *handle;
3319
3320         handle = ext4_journal_start(dquot_to_inode(dquot),
3321                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3322         if (IS_ERR(handle)) {
3323                 /* Release dquot anyway to avoid endless cycle in dqput() */
3324                 dquot_release(dquot);
3325                 return PTR_ERR(handle);
3326         }
3327         ret = dquot_release(dquot);
3328         err = ext4_journal_stop(handle);
3329         if (!ret)
3330                 ret = err;
3331         return ret;
3332 }
3333
3334 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3335 {
3336         /* Are we journaling quotas? */
3337         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3338             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3339                 dquot_mark_dquot_dirty(dquot);
3340                 return ext4_write_dquot(dquot);
3341         } else {
3342                 return dquot_mark_dquot_dirty(dquot);
3343         }
3344 }
3345
3346 static int ext4_write_info(struct super_block *sb, int type)
3347 {
3348         int ret, err;
3349         handle_t *handle;
3350
3351         /* Data block + inode block */
3352         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3353         if (IS_ERR(handle))
3354                 return PTR_ERR(handle);
3355         ret = dquot_commit_info(sb, type);
3356         err = ext4_journal_stop(handle);
3357         if (!ret)
3358                 ret = err;
3359         return ret;
3360 }
3361
3362 /*
3363  * Turn on quotas during mount time - we need to find
3364  * the quota file and such...
3365  */
3366 static int ext4_quota_on_mount(struct super_block *sb, int type)
3367 {
3368         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3369                         EXT4_SB(sb)->s_jquota_fmt, type);
3370 }
3371
3372 /*
3373  * Standard function to be called on quota_on
3374  */
3375 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3376                          char *path, int remount)
3377 {
3378         int err;
3379         struct nameidata nd;
3380
3381         if (!test_opt(sb, QUOTA))
3382                 return -EINVAL;
3383         /* When remounting, no checks are needed and in fact, path is NULL */
3384         if (remount)
3385                 return vfs_quota_on(sb, type, format_id, path, remount);
3386
3387         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3388         if (err)
3389                 return err;
3390
3391         /* Quotafile not on the same filesystem? */
3392         if (nd.path.mnt->mnt_sb != sb) {
3393                 path_put(&nd.path);
3394                 return -EXDEV;
3395         }
3396         /* Journaling quota? */
3397         if (EXT4_SB(sb)->s_qf_names[type]) {
3398                 /* Quotafile not in fs root? */
3399                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3400                         printk(KERN_WARNING
3401                                 "EXT4-fs: Quota file not on filesystem root. "
3402                                 "Journaled quota will not work.\n");
3403         }
3404
3405         /*
3406          * When we journal data on quota file, we have to flush journal to see
3407          * all updates to the file when we bypass pagecache...
3408          */
3409         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3410                 /*
3411                  * We don't need to lock updates but journal_flush() could
3412                  * otherwise be livelocked...
3413                  */
3414                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3415                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3416                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3417         }
3418
3419         err = vfs_quota_on_path(sb, type, format_id, &nd.path);
3420         path_put(&nd.path);
3421         return err;
3422 }
3423
3424 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3425  * acquiring the locks... As quota files are never truncated and quota code
3426  * itself serializes the operations (and noone else should touch the files)
3427  * we don't have to be afraid of races */
3428 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3429                                size_t len, loff_t off)
3430 {
3431         struct inode *inode = sb_dqopt(sb)->files[type];
3432         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3433         int err = 0;
3434         int offset = off & (sb->s_blocksize - 1);
3435         int tocopy;
3436         size_t toread;
3437         struct buffer_head *bh;
3438         loff_t i_size = i_size_read(inode);
3439
3440         if (off > i_size)
3441                 return 0;
3442         if (off+len > i_size)
3443                 len = i_size-off;
3444         toread = len;
3445         while (toread > 0) {
3446                 tocopy = sb->s_blocksize - offset < toread ?
3447                                 sb->s_blocksize - offset : toread;
3448                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3449                 if (err)
3450                         return err;
3451                 if (!bh)        /* A hole? */
3452                         memset(data, 0, tocopy);
3453                 else
3454                         memcpy(data, bh->b_data+offset, tocopy);
3455                 brelse(bh);
3456                 offset = 0;
3457                 toread -= tocopy;
3458                 data += tocopy;
3459                 blk++;
3460         }
3461         return len;
3462 }
3463
3464 /* Write to quotafile (we know the transaction is already started and has
3465  * enough credits) */
3466 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3467                                 const char *data, size_t len, loff_t off)
3468 {
3469         struct inode *inode = sb_dqopt(sb)->files[type];
3470         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3471         int err = 0;
3472         int offset = off & (sb->s_blocksize - 1);
3473         int tocopy;
3474         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3475         size_t towrite = len;
3476         struct buffer_head *bh;
3477         handle_t *handle = journal_current_handle();
3478
3479         if (!handle) {
3480                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%llu, len=%llu)"
3481                         " cancelled because transaction is not started.\n",
3482                         (unsigned long long)off, (unsigned long long)len);
3483                 return -EIO;
3484         }
3485         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3486         while (towrite > 0) {
3487                 tocopy = sb->s_blocksize - offset < towrite ?
3488                                 sb->s_blocksize - offset : towrite;
3489                 bh = ext4_bread(handle, inode, blk, 1, &err);
3490                 if (!bh)
3491                         goto out;
3492                 if (journal_quota) {
3493                         err = ext4_journal_get_write_access(handle, bh);
3494                         if (err) {
3495                                 brelse(bh);
3496                                 goto out;
3497                         }
3498                 }
3499                 lock_buffer(bh);
3500                 memcpy(bh->b_data+offset, data, tocopy);
3501                 flush_dcache_page(bh->b_page);
3502                 unlock_buffer(bh);
3503                 if (journal_quota)
3504                         err = ext4_journal_dirty_metadata(handle, bh);
3505                 else {
3506                         /* Always do at least ordered writes for quotas */
3507                         err = ext4_jbd2_file_inode(handle, inode);
3508                         mark_buffer_dirty(bh);
3509                 }
3510                 brelse(bh);
3511                 if (err)
3512                         goto out;
3513                 offset = 0;
3514                 towrite -= tocopy;
3515                 data += tocopy;
3516                 blk++;
3517         }
3518 out:
3519         if (len == towrite) {
3520                 mutex_unlock(&inode->i_mutex);
3521                 return err;
3522         }
3523         if (inode->i_size < off+len-towrite) {
3524                 i_size_write(inode, off+len-towrite);
3525                 EXT4_I(inode)->i_disksize = inode->i_size;
3526         }
3527         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3528         ext4_mark_inode_dirty(handle, inode);
3529         mutex_unlock(&inode->i_mutex);
3530         return len - towrite;
3531 }
3532
3533 #endif
3534
3535 static int ext4_get_sb(struct file_system_type *fs_type,
3536         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3537 {
3538         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3539 }
3540
3541 #ifdef CONFIG_PROC_FS
3542 static int ext4_ui_proc_show(struct seq_file *m, void *v)
3543 {
3544         unsigned int *p = m->private;
3545
3546         seq_printf(m, "%u\n", *p);
3547         return 0;
3548 }
3549
3550 static int ext4_ui_proc_open(struct inode *inode, struct file *file)
3551 {
3552         return single_open(file, ext4_ui_proc_show, PDE(inode)->data);
3553 }
3554
3555 static ssize_t ext4_ui_proc_write(struct file *file, const char __user *buf,
3556                                size_t cnt, loff_t *ppos)
3557 {
3558         unsigned int *p = PDE(file->f_path.dentry->d_inode)->data;
3559         char str[32];
3560         unsigned long value;
3561
3562         if (cnt >= sizeof(str))
3563                 return -EINVAL;
3564         if (copy_from_user(str, buf, cnt))
3565                 return -EFAULT;
3566         value = simple_strtol(str, NULL, 0);
3567         if (value < 0)
3568                 return -ERANGE;
3569         *p = value;
3570         return cnt;
3571 }
3572
3573 const struct file_operations ext4_ui_proc_fops = {
3574         .owner          = THIS_MODULE,
3575         .open           = ext4_ui_proc_open,
3576         .read           = seq_read,
3577         .llseek         = seq_lseek,
3578         .release        = single_release,
3579         .write          = ext4_ui_proc_write,
3580 };
3581 #endif
3582
3583 static struct file_system_type ext4dev_fs_type = {
3584         .owner          = THIS_MODULE,
3585         .name           = "ext4dev",
3586         .get_sb         = ext4_get_sb,
3587         .kill_sb        = kill_block_super,
3588         .fs_flags       = FS_REQUIRES_DEV,
3589 };
3590
3591 static int __init init_ext4_fs(void)
3592 {
3593         int err;
3594
3595         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
3596         err = init_ext4_mballoc();
3597         if (err)
3598                 return err;
3599
3600         err = init_ext4_xattr();
3601         if (err)
3602                 goto out2;
3603         err = init_inodecache();
3604         if (err)
3605                 goto out1;
3606         err = register_filesystem(&ext4dev_fs_type);
3607         if (err)
3608                 goto out;
3609         return 0;
3610 out:
3611         destroy_inodecache();
3612 out1:
3613         exit_ext4_xattr();
3614 out2:
3615         exit_ext4_mballoc();
3616         return err;
3617 }
3618
3619 static void __exit exit_ext4_fs(void)
3620 {
3621         unregister_filesystem(&ext4dev_fs_type);
3622         destroy_inodecache();
3623         exit_ext4_xattr();
3624         exit_ext4_mballoc();
3625         remove_proc_entry("fs/ext4", NULL);
3626 }
3627
3628 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3629 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3630 MODULE_LICENSE("GPL");
3631 module_init(init_ext4_fs)
3632 module_exit(exit_ext4_fs)