err.no Git - linux-2.6/blob - 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/log2.h>
  38 #include <linux/crc16.h>
  39 #include <asm/uaccess.h>
  40
  41 #include "ext4.h"
  42 #include "ext4_jbd2.h"
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46 #include "group.h"
  47
  48 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  49                              unsigned long journal_devnum);
  50 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  51                                unsigned int);
  52 static void ext4_commit_super (struct super_block * sb,
  53                                struct ext4_super_block * es,
  54                                int sync);
  55 static void ext4_mark_recovery_complete(struct super_block * sb,
  56                                         struct ext4_super_block * es);
  57 static void ext4_clear_journal_err(struct super_block * sb,
  58                                    struct ext4_super_block * es);
  59 static int ext4_sync_fs(struct super_block *sb, int wait);
  60 static const char *ext4_decode_error(struct super_block * sb, int errno,
  61                                      char nbuf[16]);
  62 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  63 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  64 static void ext4_unlockfs(struct super_block *sb);
  65 static void ext4_write_super (struct super_block * sb);
  66 static void ext4_write_super_lockfs(struct super_block *sb);
  67
  68
  69 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  70                                struct ext4_group_desc *bg)
  71 {
  72         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  73                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  74                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  75 }
  76
  77 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  78                                struct ext4_group_desc *bg)
  79 {
  80         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  81                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  82                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  83 }
  84
  85 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  86                               struct ext4_group_desc *bg)
  87 {
  88         return le32_to_cpu(bg->bg_inode_table_lo) |
  89                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  90                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  91 }
  92
  93 void ext4_block_bitmap_set(struct super_block *sb,
  94                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  95 {
  96         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  97         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  98                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  99 }
 100
 101 void ext4_inode_bitmap_set(struct super_block *sb,
 102                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 103 {
 104         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 105         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 106                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 107 }
 108
 109 void ext4_inode_table_set(struct super_block *sb,
 110                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 111 {
 112         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 113         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 114                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 115 }
 116
 117 /*
 118  * Wrappers for jbd2_journal_start/end.
 119  *
 120  * The only special thing we need to do here is to make sure that all
 121  * journal_end calls result in the superblock being marked dirty, so
 122  * that sync() will call the filesystem's write_super callback if
 123  * appropriate.
 124  */
 125 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 126 {
 127         journal_t *journal;
 128
 129         if (sb->s_flags & MS_RDONLY)
 130                 return ERR_PTR(-EROFS);
 131
 132         /* Special case here: if the journal has aborted behind our
 133          * backs (eg. EIO in the commit thread), then we still need to
 134          * take the FS itself readonly cleanly. */
 135         journal = EXT4_SB(sb)->s_journal;
 136         if (is_journal_aborted(journal)) {
 137                 ext4_abort(sb, __func__,
 138                            "Detected aborted journal");
 139                 return ERR_PTR(-EROFS);
 140         }
 141
 142         return jbd2_journal_start(journal, nblocks);
 143 }
 144
 145 /*
 146  * The only special thing we need to do here is to make sure that all
 147  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 148  * that sync() will call the filesystem's write_super callback if
 149  * appropriate.
 150  */
 151 int __ext4_journal_stop(const char *where, handle_t *handle)
 152 {
 153         struct super_block *sb;
 154         int err;
 155         int rc;
 156
 157         sb = handle->h_transaction->t_journal->j_private;
 158         err = handle->h_err;
 159         rc = jbd2_journal_stop(handle);
 160
 161         if (!err)
 162                 err = rc;
 163         if (err)
 164                 __ext4_std_error(sb, where, err);
 165         return err;
 166 }
 167
 168 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 169                 struct buffer_head *bh, handle_t *handle, int err)
 170 {
 171         char nbuf[16];
 172         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 173
 174         if (bh)
 175                 BUFFER_TRACE(bh, "abort");
 176
 177         if (!handle->h_err)
 178                 handle->h_err = err;
 179
 180         if (is_handle_aborted(handle))
 181                 return;
 182
 183         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 184                caller, errstr, err_fn);
 185
 186         jbd2_journal_abort_handle(handle);
 187 }
 188
 189 /* Deal with the reporting of failure conditions on a filesystem such as
 190  * inconsistencies detected or read IO failures.
 191  *
 192  * On ext2, we can store the error state of the filesystem in the
 193  * superblock.  That is not possible on ext4, because we may have other
 194  * write ordering constraints on the superblock which prevent us from
 195  * writing it out straight away; and given that the journal is about to
 196  * be aborted, we can't rely on the current, or future, transactions to
 197  * write out the superblock safely.
 198  *
 199  * We'll just use the jbd2_journal_abort() error code to record an error in
 200  * the journal instead.  On recovery, the journal will compain about
 201  * that error until we've noted it down and cleared it.
 202  */
 203
 204 static void ext4_handle_error(struct super_block *sb)
 205 {
 206         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 207
 208         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 209         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 210
 211         if (sb->s_flags & MS_RDONLY)
 212                 return;
 213
 214         if (!test_opt (sb, ERRORS_CONT)) {
 215                 journal_t *journal = EXT4_SB(sb)->s_journal;
 216
 217                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 218                 if (journal)
 219                         jbd2_journal_abort(journal, -EIO);
 220         }
 221         if (test_opt (sb, ERRORS_RO)) {
 222                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 223                 sb->s_flags |= MS_RDONLY;
 224         }
 225         ext4_commit_super(sb, es, 1);
 226         if (test_opt(sb, ERRORS_PANIC))
 227                 panic("EXT4-fs (device %s): panic forced after error\n",
 228                         sb->s_id);
 229 }
 230
 231 void ext4_error (struct super_block * sb, const char * function,
 232                  const char * fmt, ...)
 233 {
 234         va_list args;
 235
 236         va_start(args, fmt);
 237         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 238         vprintk(fmt, args);
 239         printk("\n");
 240         va_end(args);
 241
 242         ext4_handle_error(sb);
 243 }
 244
 245 static const char *ext4_decode_error(struct super_block * sb, int errno,
 246                                      char nbuf[16])
 247 {
 248         char *errstr = NULL;
 249
 250         switch (errno) {
 251         case -EIO:
 252                 errstr = "IO failure";
 253                 break;
 254         case -ENOMEM:
 255                 errstr = "Out of memory";
 256                 break;
 257         case -EROFS:
 258                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 259                         errstr = "Journal has aborted";
 260                 else
 261                         errstr = "Readonly filesystem";
 262                 break;
 263         default:
 264                 /* If the caller passed in an extra buffer for unknown
 265                  * errors, textualise them now.  Else we just return
 266                  * NULL. */
 267                 if (nbuf) {
 268                         /* Check for truncated error codes... */
 269                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 270                                 errstr = nbuf;
 271                 }
 272                 break;
 273         }
 274
 275         return errstr;
 276 }
 277
 278 /* __ext4_std_error decodes expected errors from journaling functions
 279  * automatically and invokes the appropriate error response.  */
 280
 281 void __ext4_std_error (struct super_block * sb, const char * function,
 282                        int errno)
 283 {
 284         char nbuf[16];
 285         const char *errstr;
 286
 287         /* Special case: if the error is EROFS, and we're not already
 288          * inside a transaction, then there's really no point in logging
 289          * an error. */
 290         if (errno == -EROFS && journal_current_handle() == NULL &&
 291             (sb->s_flags & MS_RDONLY))
 292                 return;
 293
 294         errstr = ext4_decode_error(sb, errno, nbuf);
 295         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 296                 sb->s_id, function, errstr);
 297
 298         ext4_handle_error(sb);
 299 }
 300
 301 /*
 302  * ext4_abort is a much stronger failure handler than ext4_error.  The
 303  * abort function may be used to deal with unrecoverable failures such
 304  * as journal IO errors or ENOMEM at a critical moment in log management.
 305  *
 306  * We unconditionally force the filesystem into an ABORT|READONLY state,
 307  * unless the error response on the fs has been set to panic in which
 308  * case we take the easy way out and panic immediately.
 309  */
 310
 311 void ext4_abort (struct super_block * sb, const char * function,
 312                  const char * fmt, ...)
 313 {
 314         va_list args;
 315
 316         printk (KERN_CRIT "ext4_abort called.\n");
 317
 318         va_start(args, fmt);
 319         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 320         vprintk(fmt, args);
 321         printk("\n");
 322         va_end(args);
 323
 324         if (test_opt(sb, ERRORS_PANIC))
 325                 panic("EXT4-fs panic from previous error\n");
 326
 327         if (sb->s_flags & MS_RDONLY)
 328                 return;
 329
 330         printk(KERN_CRIT "Remounting filesystem read-only\n");
 331         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 332         sb->s_flags |= MS_RDONLY;
 333         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 334         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 335 }
 336
 337 void ext4_warning (struct super_block * sb, const char * function,
 338                    const char * fmt, ...)
 339 {
 340         va_list args;
 341
 342         va_start(args, fmt);
 343         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 344                sb->s_id, function);
 345         vprintk(fmt, args);
 346         printk("\n");
 347         va_end(args);
 348 }
 349
 350 void ext4_update_dynamic_rev(struct super_block *sb)
 351 {
 352         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 353
 354         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 355                 return;
 356
 357         ext4_warning(sb, __func__,
 358                      "updating to rev %d because of new feature flag, "
 359                      "running e2fsck is recommended",
 360                      EXT4_DYNAMIC_REV);
 361
 362         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 363         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 364         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 365         /* leave es->s_feature_*compat flags alone */
 366         /* es->s_uuid will be set by e2fsck if empty */
 367
 368         /*
 369          * The rest of the superblock fields should be zero, and if not it
 370          * means they are likely already in use, so leave them alone.  We
 371          * can leave it up to e2fsck to clean up any inconsistencies there.
 372          */
 373 }
 374
 375 int ext4_update_compat_feature(handle_t *handle,
 376                                         struct super_block *sb, __u32 compat)
 377 {
 378         int err = 0;
 379         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 380                 err = ext4_journal_get_write_access(handle,
 381                                 EXT4_SB(sb)->s_sbh);
 382                 if (err)
 383                         return err;
 384                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 385                 sb->s_dirt = 1;
 386                 handle->h_sync = 1;
 387                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 388                                         "call ext4_journal_dirty_met adata");
 389                 err = ext4_journal_dirty_metadata(handle,
 390                                 EXT4_SB(sb)->s_sbh);
 391         }
 392         return err;
 393 }
 394
 395 int ext4_update_rocompat_feature(handle_t *handle,
 396                                         struct super_block *sb, __u32 rocompat)
 397 {
 398         int err = 0;
 399         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 400                 err = ext4_journal_get_write_access(handle,
 401                                 EXT4_SB(sb)->s_sbh);
 402                 if (err)
 403                         return err;
 404                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 405                 sb->s_dirt = 1;
 406                 handle->h_sync = 1;
 407                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 408                                         "call ext4_journal_dirty_met adata");
 409                 err = ext4_journal_dirty_metadata(handle,
 410                                 EXT4_SB(sb)->s_sbh);
 411         }
 412         return err;
 413 }
 414
 415 int ext4_update_incompat_feature(handle_t *handle,
 416                                         struct super_block *sb, __u32 incompat)
 417 {
 418         int err = 0;
 419         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 420                 err = ext4_journal_get_write_access(handle,
 421                                 EXT4_SB(sb)->s_sbh);
 422                 if (err)
 423                         return err;
 424                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 425                 sb->s_dirt = 1;
 426                 handle->h_sync = 1;
 427                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 428                                         "call ext4_journal_dirty_met adata");
 429                 err = ext4_journal_dirty_metadata(handle,
 430                                 EXT4_SB(sb)->s_sbh);
 431         }
 432         return err;
 433 }
 434
 435 /*
 436  * Open the external journal device
 437  */
 438 static struct block_device *ext4_blkdev_get(dev_t dev)
 439 {
 440         struct block_device *bdev;
 441         char b[BDEVNAME_SIZE];
 442
 443         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 444         if (IS_ERR(bdev))
 445                 goto fail;
 446         return bdev;
 447
 448 fail:
 449         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 450                         __bdevname(dev, b), PTR_ERR(bdev));
 451         return NULL;
 452 }
 453
 454 /*
 455  * Release the journal device
 456  */
 457 static int ext4_blkdev_put(struct block_device *bdev)
 458 {
 459         bd_release(bdev);
 460         return blkdev_put(bdev);
 461 }
 462
 463 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 464 {
 465         struct block_device *bdev;
 466         int ret = -ENODEV;
 467
 468         bdev = sbi->journal_bdev;
 469         if (bdev) {
 470                 ret = ext4_blkdev_put(bdev);
 471                 sbi->journal_bdev = NULL;
 472         }
 473         return ret;
 474 }
 475
 476 static inline struct inode *orphan_list_entry(struct list_head *l)
 477 {
 478         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 479 }
 480
 481 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 482 {
 483         struct list_head *l;
 484
 485         printk(KERN_ERR "sb orphan head is %d\n",
 486                le32_to_cpu(sbi->s_es->s_last_orphan));
 487
 488         printk(KERN_ERR "sb_info orphan list:\n");
 489         list_for_each(l, &sbi->s_orphan) {
 490                 struct inode *inode = orphan_list_entry(l);
 491                 printk(KERN_ERR "  "
 492                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 493                        inode->i_sb->s_id, inode->i_ino, inode,
 494                        inode->i_mode, inode->i_nlink,
 495                        NEXT_ORPHAN(inode));
 496         }
 497 }
 498
 499 static void ext4_put_super (struct super_block * sb)
 500 {
 501         struct ext4_sb_info *sbi = EXT4_SB(sb);
 502         struct ext4_super_block *es = sbi->s_es;
 503         int i;
 504
 505         ext4_mb_release(sb);
 506         ext4_ext_release(sb);
 507         ext4_xattr_put_super(sb);
 508         jbd2_journal_destroy(sbi->s_journal);
 509         sbi->s_journal = NULL;
 510         if (!(sb->s_flags & MS_RDONLY)) {
 511                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 512                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 513                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 514                 mark_buffer_dirty(sbi->s_sbh);
 515                 ext4_commit_super(sb, es, 1);
 516         }
 517
 518         for (i = 0; i < sbi->s_gdb_count; i++)
 519                 brelse(sbi->s_group_desc[i]);
 520         kfree(sbi->s_group_desc);
 521         kfree(sbi->s_flex_groups);
 522         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 523         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 524         percpu_counter_destroy(&sbi->s_dirs_counter);
 525         brelse(sbi->s_sbh);
 526 #ifdef CONFIG_QUOTA
 527         for (i = 0; i < MAXQUOTAS; i++)
 528                 kfree(sbi->s_qf_names[i]);
 529 #endif
 530
 531         /* Debugging code just in case the in-memory inode orphan list
 532          * isn't empty.  The on-disk one can be non-empty if we've
 533          * detected an error and taken the fs readonly, but the
 534          * in-memory list had better be clean by this point. */
 535         if (!list_empty(&sbi->s_orphan))
 536                 dump_orphan_list(sb, sbi);
 537         J_ASSERT(list_empty(&sbi->s_orphan));
 538
 539         invalidate_bdev(sb->s_bdev);
 540         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 541                 /*
 542                  * Invalidate the journal device's buffers.  We don't want them
 543                  * floating about in memory - the physical journal device may
 544                  * hotswapped, and it breaks the `ro-after' testing code.
 545                  */
 546                 sync_blockdev(sbi->journal_bdev);
 547                 invalidate_bdev(sbi->journal_bdev);
 548                 ext4_blkdev_remove(sbi);
 549         }
 550         sb->s_fs_info = NULL;
 551         kfree(sbi);
 552         return;
 553 }
 554
 555 static struct kmem_cache *ext4_inode_cachep;
 556
 557 /*
 558  * Called inside transaction, so use GFP_NOFS
 559  */
 560 static struct inode *ext4_alloc_inode(struct super_block *sb)
 561 {
 562         struct ext4_inode_info *ei;
 563
 564         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 565         if (!ei)
 566                 return NULL;
 567 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 568         ei->i_acl = EXT4_ACL_NOT_CACHED;
 569         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 570 #endif
 571         ei->i_block_alloc_info = NULL;
 572         ei->vfs_inode.i_version = 1;
 573         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 574         INIT_LIST_HEAD(&ei->i_prealloc_list);
 575         spin_lock_init(&ei->i_prealloc_lock);
 576         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 577         ei->i_reserved_data_blocks = 0;
 578         ei->i_reserved_meta_blocks = 0;
 579         ei->i_allocated_meta_blocks = 0;
 580         ei->i_delalloc_reserved_flag = 0;
 581         spin_lock_init(&(ei->i_block_reservation_lock));
 582         return &ei->vfs_inode;
 583 }
 584
 585 static void ext4_destroy_inode(struct inode *inode)
 586 {
 587         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 588                 printk("EXT4 Inode %p: orphan list check failed!\n",
 589                         EXT4_I(inode));
 590                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 591                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 592                                 true);
 593                 dump_stack();
 594         }
 595         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 596 }
 597
 598 static void init_once(struct kmem_cache *cachep, void *foo)
 599 {
 600         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 601
 602         INIT_LIST_HEAD(&ei->i_orphan);
 603 #ifdef CONFIG_EXT4DEV_FS_XATTR
 604         init_rwsem(&ei->xattr_sem);
 605 #endif
 606         init_rwsem(&ei->i_data_sem);
 607         inode_init_once(&ei->vfs_inode);
 608 }
 609
 610 static int init_inodecache(void)
 611 {
 612         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 613                                              sizeof(struct ext4_inode_info),
 614                                              0, (SLAB_RECLAIM_ACCOUNT|
 615                                                 SLAB_MEM_SPREAD),
 616                                              init_once);
 617         if (ext4_inode_cachep == NULL)
 618                 return -ENOMEM;
 619         return 0;
 620 }
 621
 622 static void destroy_inodecache(void)
 623 {
 624         kmem_cache_destroy(ext4_inode_cachep);
 625 }
 626
 627 static void ext4_clear_inode(struct inode *inode)
 628 {
 629         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 630 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 631         if (EXT4_I(inode)->i_acl &&
 632                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 633                 posix_acl_release(EXT4_I(inode)->i_acl);
 634                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 635         }
 636         if (EXT4_I(inode)->i_default_acl &&
 637                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 638                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 639                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 640         }
 641 #endif
 642         ext4_discard_reservation(inode);
 643         EXT4_I(inode)->i_block_alloc_info = NULL;
 644         if (unlikely(rsv))
 645                 kfree(rsv);
 646         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 647                                        &EXT4_I(inode)->jinode);
 648 }
 649
 650 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 651 {
 652 #if defined(CONFIG_QUOTA)
 653         struct ext4_sb_info *sbi = EXT4_SB(sb);
 654
 655         if (sbi->s_jquota_fmt)
 656                 seq_printf(seq, ",jqfmt=%s",
 657                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 658
 659         if (sbi->s_qf_names[USRQUOTA])
 660                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 661
 662         if (sbi->s_qf_names[GRPQUOTA])
 663                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 664
 665         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 666                 seq_puts(seq, ",usrquota");
 667
 668         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 669                 seq_puts(seq, ",grpquota");
 670 #endif
 671 }
 672
 673 /*
 674  * Show an option if
 675  *  - it's set to a non-default value OR
 676  *  - if the per-sb default is different from the global default
 677  */
 678 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 679 {
 680         int def_errors;
 681         unsigned long def_mount_opts;
 682         struct super_block *sb = vfs->mnt_sb;
 683         struct ext4_sb_info *sbi = EXT4_SB(sb);
 684         struct ext4_super_block *es = sbi->s_es;
 685
 686         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 687         def_errors     = le16_to_cpu(es->s_errors);
 688
 689         if (sbi->s_sb_block != 1)
 690                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 691         if (test_opt(sb, MINIX_DF))
 692                 seq_puts(seq, ",minixdf");
 693         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 694                 seq_puts(seq, ",grpid");
 695         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 696                 seq_puts(seq, ",nogrpid");
 697         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 698             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 699                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 700         }
 701         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 702             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 703                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 704         }
 705         if (test_opt(sb, ERRORS_RO)) {
 706                 if (def_errors == EXT4_ERRORS_PANIC ||
 707                     def_errors == EXT4_ERRORS_CONTINUE) {
 708                         seq_puts(seq, ",errors=remount-ro");
 709                 }
 710         }
 711         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 712                 seq_puts(seq, ",errors=continue");
 713         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 714                 seq_puts(seq, ",errors=panic");
 715         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 716                 seq_puts(seq, ",nouid32");
 717         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 718                 seq_puts(seq, ",debug");
 719         if (test_opt(sb, OLDALLOC))
 720                 seq_puts(seq, ",oldalloc");
 721 #ifdef CONFIG_EXT4DEV_FS_XATTR
 722         if (test_opt(sb, XATTR_USER) &&
 723                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 724                 seq_puts(seq, ",user_xattr");
 725         if (!test_opt(sb, XATTR_USER) &&
 726             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 727                 seq_puts(seq, ",nouser_xattr");
 728         }
 729 #endif
 730 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 731         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 732                 seq_puts(seq, ",acl");
 733         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 734                 seq_puts(seq, ",noacl");
 735 #endif
 736         if (!test_opt(sb, RESERVATION))
 737                 seq_puts(seq, ",noreservation");
 738         if (sbi->s_commit_interval) {
 739                 seq_printf(seq, ",commit=%u",
 740                            (unsigned) (sbi->s_commit_interval / HZ));
 741         }
 742         /*
 743          * We're changing the default of barrier mount option, so
 744          * let's always display its mount state so it's clear what its
 745          * status is.
 746          */
 747         seq_puts(seq, ",barrier=");
 748         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 749         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 750                 seq_puts(seq, ",journal_async_commit");
 751         if (test_opt(sb, NOBH))
 752                 seq_puts(seq, ",nobh");
 753         if (!test_opt(sb, EXTENTS))
 754                 seq_puts(seq, ",noextents");
 755         if (!test_opt(sb, MBALLOC))
 756                 seq_puts(seq, ",nomballoc");
 757         if (test_opt(sb, I_VERSION))
 758                 seq_puts(seq, ",i_version");
 759         if (!test_opt(sb, DELALLOC))
 760                 seq_puts(seq, ",nodelalloc");
 761
 762
 763         if (sbi->s_stripe)
 764                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 765         /*
 766          * journal mode get enabled in different ways
 767          * So just print the value even if we didn't specify it
 768          */
 769         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 770                 seq_puts(seq, ",data=journal");
 771         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 772                 seq_puts(seq, ",data=ordered");
 773         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 774                 seq_puts(seq, ",data=writeback");
 775
 776         ext4_show_quota_options(seq, sb);
 777         return 0;
 778 }
 779
 780
 781 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 782                 u64 ino, u32 generation)
 783 {
 784         struct inode *inode;
 785
 786         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 787                 return ERR_PTR(-ESTALE);
 788         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 789                 return ERR_PTR(-ESTALE);
 790
 791         /* iget isn't really right if the inode is currently unallocated!!
 792          *
 793          * ext4_read_inode will return a bad_inode if the inode had been
 794          * deleted, so we should be safe.
 795          *
 796          * Currently we don't know the generation for parent directory, so
 797          * a generation of 0 means "accept any"
 798          */
 799         inode = ext4_iget(sb, ino);
 800         if (IS_ERR(inode))
 801                 return ERR_CAST(inode);
 802         if (generation && inode->i_generation != generation) {
 803                 iput(inode);
 804                 return ERR_PTR(-ESTALE);
 805         }
 806
 807         return inode;
 808 }
 809
 810 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 811                 int fh_len, int fh_type)
 812 {
 813         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 814                                     ext4_nfs_get_inode);
 815 }
 816
 817 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 818                 int fh_len, int fh_type)
 819 {
 820         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 821                                     ext4_nfs_get_inode);
 822 }
 823
 824 #ifdef CONFIG_QUOTA
 825 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 826 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 827
 828 static int ext4_dquot_initialize(struct inode *inode, int type);
 829 static int ext4_dquot_drop(struct inode *inode);
 830 static int ext4_write_dquot(struct dquot *dquot);
 831 static int ext4_acquire_dquot(struct dquot *dquot);
 832 static int ext4_release_dquot(struct dquot *dquot);
 833 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 834 static int ext4_write_info(struct super_block *sb, int type);
 835 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 836                                 char *path, int remount);
 837 static int ext4_quota_on_mount(struct super_block *sb, int type);
 838 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 839                                size_t len, loff_t off);
 840 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 841                                 const char *data, size_t len, loff_t off);
 842
 843 static struct dquot_operations ext4_quota_operations = {
 844         .initialize     = ext4_dquot_initialize,
 845         .drop           = ext4_dquot_drop,
 846         .alloc_space    = dquot_alloc_space,
 847         .alloc_inode    = dquot_alloc_inode,
 848         .free_space     = dquot_free_space,
 849         .free_inode     = dquot_free_inode,
 850         .transfer       = dquot_transfer,
 851         .write_dquot    = ext4_write_dquot,
 852         .acquire_dquot  = ext4_acquire_dquot,
 853         .release_dquot  = ext4_release_dquot,
 854         .mark_dirty     = ext4_mark_dquot_dirty,
 855         .write_info     = ext4_write_info
 856 };
 857
 858 static struct quotactl_ops ext4_qctl_operations = {
 859         .quota_on       = ext4_quota_on,
 860         .quota_off      = vfs_quota_off,
 861         .quota_sync     = vfs_quota_sync,
 862         .get_info       = vfs_get_dqinfo,
 863         .set_info       = vfs_set_dqinfo,
 864         .get_dqblk      = vfs_get_dqblk,
 865         .set_dqblk      = vfs_set_dqblk
 866 };
 867 #endif
 868
 869 static const struct super_operations ext4_sops = {
 870         .alloc_inode    = ext4_alloc_inode,
 871         .destroy_inode  = ext4_destroy_inode,
 872         .write_inode    = ext4_write_inode,
 873         .dirty_inode    = ext4_dirty_inode,
 874         .delete_inode   = ext4_delete_inode,
 875         .put_super      = ext4_put_super,
 876         .write_super    = ext4_write_super,
 877         .sync_fs        = ext4_sync_fs,
 878         .write_super_lockfs = ext4_write_super_lockfs,
 879         .unlockfs       = ext4_unlockfs,
 880         .statfs         = ext4_statfs,
 881         .remount_fs     = ext4_remount,
 882         .clear_inode    = ext4_clear_inode,
 883         .show_options   = ext4_show_options,
 884 #ifdef CONFIG_QUOTA
 885         .quota_read     = ext4_quota_read,
 886         .quota_write    = ext4_quota_write,
 887 #endif
 888 };
 889
 890 static const struct export_operations ext4_export_ops = {
 891         .fh_to_dentry = ext4_fh_to_dentry,
 892         .fh_to_parent = ext4_fh_to_parent,
 893         .get_parent = ext4_get_parent,
 894 };
 895
 896 enum {
 897         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 898         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 899         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 900         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 901         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 902         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 903         Opt_journal_checksum, Opt_journal_async_commit,
 904         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 905         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 906         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 907         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 908         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 909         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 910 };
 911
 912 static match_table_t tokens = {
 913         {Opt_bsd_df, "bsddf"},
 914         {Opt_minix_df, "minixdf"},
 915         {Opt_grpid, "grpid"},
 916         {Opt_grpid, "bsdgroups"},
 917         {Opt_nogrpid, "nogrpid"},
 918         {Opt_nogrpid, "sysvgroups"},
 919         {Opt_resgid, "resgid=%u"},
 920         {Opt_resuid, "resuid=%u"},
 921         {Opt_sb, "sb=%u"},
 922         {Opt_err_cont, "errors=continue"},
 923         {Opt_err_panic, "errors=panic"},
 924         {Opt_err_ro, "errors=remount-ro"},
 925         {Opt_nouid32, "nouid32"},
 926         {Opt_nocheck, "nocheck"},
 927         {Opt_nocheck, "check=none"},
 928         {Opt_debug, "debug"},
 929         {Opt_oldalloc, "oldalloc"},
 930         {Opt_orlov, "orlov"},
 931         {Opt_user_xattr, "user_xattr"},
 932         {Opt_nouser_xattr, "nouser_xattr"},
 933         {Opt_acl, "acl"},
 934         {Opt_noacl, "noacl"},
 935         {Opt_reservation, "reservation"},
 936         {Opt_noreservation, "noreservation"},
 937         {Opt_noload, "noload"},
 938         {Opt_nobh, "nobh"},
 939         {Opt_bh, "bh"},
 940         {Opt_commit, "commit=%u"},
 941         {Opt_journal_update, "journal=update"},
 942         {Opt_journal_inum, "journal=%u"},
 943         {Opt_journal_dev, "journal_dev=%u"},
 944         {Opt_journal_checksum, "journal_checksum"},
 945         {Opt_journal_async_commit, "journal_async_commit"},
 946         {Opt_abort, "abort"},
 947         {Opt_data_journal, "data=journal"},
 948         {Opt_data_ordered, "data=ordered"},
 949         {Opt_data_writeback, "data=writeback"},
 950         {Opt_offusrjquota, "usrjquota="},
 951         {Opt_usrjquota, "usrjquota=%s"},
 952         {Opt_offgrpjquota, "grpjquota="},
 953         {Opt_grpjquota, "grpjquota=%s"},
 954         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 955         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 956         {Opt_grpquota, "grpquota"},
 957         {Opt_noquota, "noquota"},
 958         {Opt_quota, "quota"},
 959         {Opt_usrquota, "usrquota"},
 960         {Opt_barrier, "barrier=%u"},
 961         {Opt_extents, "extents"},
 962         {Opt_noextents, "noextents"},
 963         {Opt_i_version, "i_version"},
 964         {Opt_mballoc, "mballoc"},
 965         {Opt_nomballoc, "nomballoc"},
 966         {Opt_stripe, "stripe=%u"},
 967         {Opt_resize, "resize"},
 968         {Opt_delalloc, "delalloc"},
 969         {Opt_nodelalloc, "nodelalloc"},
 970         {Opt_err, NULL},
 971 };
 972
 973 static ext4_fsblk_t get_sb_block(void **data)
 974 {
 975         ext4_fsblk_t    sb_block;
 976         char            *options = (char *) *data;
 977
 978         if (!options || strncmp(options, "sb=", 3) != 0)
 979                 return 1;       /* Default location */
 980         options += 3;
 981         /*todo: use simple_strtoll with >32bit ext4 */
 982         sb_block = simple_strtoul(options, &options, 0);
 983         if (*options && *options != ',') {
 984                 printk("EXT4-fs: Invalid sb specification: %s\n",
 985                        (char *) *data);
 986                 return 1;
 987         }
 988         if (*options == ',')
 989                 options++;
 990         *data = (void *) options;
 991         return sb_block;
 992 }
 993
 994 static int parse_options (char *options, struct super_block *sb,
 995                           unsigned int *inum, unsigned long *journal_devnum,
 996                           ext4_fsblk_t *n_blocks_count, int is_remount)
 997 {
 998         struct ext4_sb_info *sbi = EXT4_SB(sb);
 999         char * p;
1000         substring_t args[MAX_OPT_ARGS];
1001         int data_opt = 0;
1002         int option;
1003 #ifdef CONFIG_QUOTA
1004         int qtype, qfmt;
1005         char *qname;
1006 #endif
1007         ext4_fsblk_t last_block;
1008
1009         if (!options)
1010                 return 1;
1011
1012         while ((p = strsep (&options, ",")) != NULL) {
1013                 int token;
1014                 if (!*p)
1015                         continue;
1016
1017                 token = match_token(p, tokens, args);
1018                 switch (token) {
1019                 case Opt_bsd_df:
1020                         clear_opt (sbi->s_mount_opt, MINIX_DF);
1021                         break;
1022                 case Opt_minix_df:
1023                         set_opt (sbi->s_mount_opt, MINIX_DF);
1024                         break;
1025                 case Opt_grpid:
1026                         set_opt (sbi->s_mount_opt, GRPID);
1027                         break;
1028                 case Opt_nogrpid:
1029                         clear_opt (sbi->s_mount_opt, GRPID);
1030                         break;
1031                 case Opt_resuid:
1032                         if (match_int(&args[0], &option))
1033                                 return 0;
1034                         sbi->s_resuid = option;
1035                         break;
1036                 case Opt_resgid:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         sbi->s_resgid = option;
1040                         break;
1041                 case Opt_sb:
1042                         /* handled by get_sb_block() instead of here */
1043                         /* *sb_block = match_int(&args[0]); */
1044                         break;
1045                 case Opt_err_panic:
1046                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1047                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1048                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1049                         break;
1050                 case Opt_err_ro:
1051                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1052                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1053                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1054                         break;
1055                 case Opt_err_cont:
1056                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1057                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1058                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1059                         break;
1060                 case Opt_nouid32:
1061                         set_opt (sbi->s_mount_opt, NO_UID32);
1062                         break;
1063                 case Opt_nocheck:
1064                         clear_opt (sbi->s_mount_opt, CHECK);
1065                         break;
1066                 case Opt_debug:
1067                         set_opt (sbi->s_mount_opt, DEBUG);
1068                         break;
1069                 case Opt_oldalloc:
1070                         set_opt (sbi->s_mount_opt, OLDALLOC);
1071                         break;
1072                 case Opt_orlov:
1073                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1074                         break;
1075 #ifdef CONFIG_EXT4DEV_FS_XATTR
1076                 case Opt_user_xattr:
1077                         set_opt (sbi->s_mount_opt, XATTR_USER);
1078                         break;
1079                 case Opt_nouser_xattr:
1080                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1081                         break;
1082 #else
1083                 case Opt_user_xattr:
1084                 case Opt_nouser_xattr:
1085                         printk("EXT4 (no)user_xattr options not supported\n");
1086                         break;
1087 #endif
1088 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1089                 case Opt_acl:
1090                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1091                         break;
1092                 case Opt_noacl:
1093                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1094                         break;
1095 #else
1096                 case Opt_acl:
1097                 case Opt_noacl:
1098                         printk("EXT4 (no)acl options not supported\n");
1099                         break;
1100 #endif
1101                 case Opt_reservation:
1102                         set_opt(sbi->s_mount_opt, RESERVATION);
1103                         break;
1104                 case Opt_noreservation:
1105                         clear_opt(sbi->s_mount_opt, RESERVATION);
1106                         break;
1107                 case Opt_journal_update:
1108                         /* @@@ FIXME */
1109                         /* Eventually we will want to be able to create
1110                            a journal file here.  For now, only allow the
1111                            user to specify an existing inode to be the
1112                            journal file. */
1113                         if (is_remount) {
1114                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1115                                        "journal on remount\n");
1116                                 return 0;
1117                         }
1118                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1119                         break;
1120                 case Opt_journal_inum:
1121                         if (is_remount) {
1122                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1123                                        "journal on remount\n");
1124                                 return 0;
1125                         }
1126                         if (match_int(&args[0], &option))
1127                                 return 0;
1128                         *inum = option;
1129                         break;
1130                 case Opt_journal_dev:
1131                         if (is_remount) {
1132                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1133                                        "journal on remount\n");
1134                                 return 0;
1135                         }
1136                         if (match_int(&args[0], &option))
1137                                 return 0;
1138                         *journal_devnum = option;
1139                         break;
1140                 case Opt_journal_checksum:
1141                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1142                         break;
1143                 case Opt_journal_async_commit:
1144                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1145                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1146                         break;
1147                 case Opt_noload:
1148                         set_opt (sbi->s_mount_opt, NOLOAD);
1149                         break;
1150                 case Opt_commit:
1151                         if (match_int(&args[0], &option))
1152                                 return 0;
1153                         if (option < 0)
1154                                 return 0;
1155                         if (option == 0)
1156                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1157                         sbi->s_commit_interval = HZ * option;
1158                         break;
1159                 case Opt_data_journal:
1160                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1161                         goto datacheck;
1162                 case Opt_data_ordered:
1163                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1164                         goto datacheck;
1165                 case Opt_data_writeback:
1166                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1167                 datacheck:
1168                         if (is_remount) {
1169                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1170                                                 != data_opt) {
1171                                         printk(KERN_ERR
1172                                                 "EXT4-fs: cannot change data "
1173                                                 "mode on remount\n");
1174                                         return 0;
1175                                 }
1176                         } else {
1177                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1178                                 sbi->s_mount_opt |= data_opt;
1179                         }
1180                         break;
1181 #ifdef CONFIG_QUOTA
1182                 case Opt_usrjquota:
1183                         qtype = USRQUOTA;
1184                         goto set_qf_name;
1185                 case Opt_grpjquota:
1186                         qtype = GRPQUOTA;
1187 set_qf_name:
1188                         if ((sb_any_quota_enabled(sb) ||
1189                              sb_any_quota_suspended(sb)) &&
1190                             !sbi->s_qf_names[qtype]) {
1191                                 printk(KERN_ERR
1192                                         "EXT4-fs: Cannot change journaled "
1193                                         "quota options when quota turned on.\n");
1194                                 return 0;
1195                         }
1196                         qname = match_strdup(&args[0]);
1197                         if (!qname) {
1198                                 printk(KERN_ERR
1199                                         "EXT4-fs: not enough memory for "
1200                                         "storing quotafile name.\n");
1201                                 return 0;
1202                         }
1203                         if (sbi->s_qf_names[qtype] &&
1204                             strcmp(sbi->s_qf_names[qtype], qname)) {
1205                                 printk(KERN_ERR
1206                                         "EXT4-fs: %s quota file already "
1207                                         "specified.\n", QTYPE2NAME(qtype));
1208                                 kfree(qname);
1209                                 return 0;
1210                         }
1211                         sbi->s_qf_names[qtype] = qname;
1212                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1213                                 printk(KERN_ERR
1214                                         "EXT4-fs: quotafile must be on "
1215                                         "filesystem root.\n");
1216                                 kfree(sbi->s_qf_names[qtype]);
1217                                 sbi->s_qf_names[qtype] = NULL;
1218                                 return 0;
1219                         }
1220                         set_opt(sbi->s_mount_opt, QUOTA);
1221                         break;
1222                 case Opt_offusrjquota:
1223                         qtype = USRQUOTA;
1224                         goto clear_qf_name;
1225                 case Opt_offgrpjquota:
1226                         qtype = GRPQUOTA;
1227 clear_qf_name:
1228                         if ((sb_any_quota_enabled(sb) ||
1229                              sb_any_quota_suspended(sb)) &&
1230                             sbi->s_qf_names[qtype]) {
1231                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1232                                         "journaled quota options when "
1233                                         "quota turned on.\n");
1234                                 return 0;
1235                         }
1236                         /*
1237                          * The space will be released later when all options
1238                          * are confirmed to be correct
1239                          */
1240                         sbi->s_qf_names[qtype] = NULL;
1241                         break;
1242                 case Opt_jqfmt_vfsold:
1243                         qfmt = QFMT_VFS_OLD;
1244                         goto set_qf_format;
1245                 case Opt_jqfmt_vfsv0:
1246                         qfmt = QFMT_VFS_V0;
1247 set_qf_format:
1248                         if ((sb_any_quota_enabled(sb) ||
1249                              sb_any_quota_suspended(sb)) &&
1250                             sbi->s_jquota_fmt != qfmt) {
1251                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1252                                         "journaled quota options when "
1253                                         "quota turned on.\n");
1254                                 return 0;
1255                         }
1256                         sbi->s_jquota_fmt = qfmt;
1257                         break;
1258                 case Opt_quota:
1259                 case Opt_usrquota:
1260                         set_opt(sbi->s_mount_opt, QUOTA);
1261                         set_opt(sbi->s_mount_opt, USRQUOTA);
1262                         break;
1263                 case Opt_grpquota:
1264                         set_opt(sbi->s_mount_opt, QUOTA);
1265                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1266                         break;
1267                 case Opt_noquota:
1268                         if (sb_any_quota_enabled(sb)) {
1269                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1270                                         "options when quota turned on.\n");
1271                                 return 0;
1272                         }
1273                         clear_opt(sbi->s_mount_opt, QUOTA);
1274                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1275                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1276                         break;
1277 #else
1278                 case Opt_quota:
1279                 case Opt_usrquota:
1280                 case Opt_grpquota:
1281                         printk(KERN_ERR
1282                                 "EXT4-fs: quota options not supported.\n");
1283                         break;
1284                 case Opt_usrjquota:
1285                 case Opt_grpjquota:
1286                 case Opt_offusrjquota:
1287                 case Opt_offgrpjquota:
1288                 case Opt_jqfmt_vfsold:
1289                 case Opt_jqfmt_vfsv0:
1290                         printk(KERN_ERR
1291                                 "EXT4-fs: journaled quota options not "
1292                                 "supported.\n");
1293                         break;
1294                 case Opt_noquota:
1295                         break;
1296 #endif
1297                 case Opt_abort:
1298                         set_opt(sbi->s_mount_opt, ABORT);
1299                         break;
1300                 case Opt_barrier:
1301                         if (match_int(&args[0], &option))
1302                                 return 0;
1303                         if (option)
1304                                 set_opt(sbi->s_mount_opt, BARRIER);
1305                         else
1306                                 clear_opt(sbi->s_mount_opt, BARRIER);
1307                         break;
1308                 case Opt_ignore:
1309                         break;
1310                 case Opt_resize:
1311                         if (!is_remount) {
1312                                 printk("EXT4-fs: resize option only available "
1313                                         "for remount\n");
1314                                 return 0;
1315                         }
1316                         if (match_int(&args[0], &option) != 0)
1317                                 return 0;
1318                         *n_blocks_count = option;
1319                         break;
1320                 case Opt_nobh:
1321                         set_opt(sbi->s_mount_opt, NOBH);
1322                         break;
1323                 case Opt_bh:
1324                         clear_opt(sbi->s_mount_opt, NOBH);
1325                         break;
1326                 case Opt_extents:
1327                         set_opt (sbi->s_mount_opt, EXTENTS);
1328                         break;
1329                 case Opt_noextents:
1330                         /*
1331                          * When e2fsprogs support resizing an already existing
1332                          * ext3 file system to greater than 2**32 we need to
1333                          * add support to block allocator to handle growing
1334                          * already existing block  mapped inode so that blocks
1335                          * allocated for them fall within 2**32
1336                          */
1337                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1338                         if (last_block  > 0xffffffffULL) {
1339                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1340                                                 "large to mount with "
1341                                                 "-o noextents options\n");
1342                                 return 0;
1343                         }
1344                         clear_opt (sbi->s_mount_opt, EXTENTS);
1345                         break;
1346                 case Opt_i_version:
1347                         set_opt(sbi->s_mount_opt, I_VERSION);
1348                         sb->s_flags |= MS_I_VERSION;
1349                         break;
1350                 case Opt_nodelalloc:
1351                         clear_opt(sbi->s_mount_opt, DELALLOC);
1352                         break;
1353                 case Opt_mballoc:
1354                         set_opt(sbi->s_mount_opt, MBALLOC);
1355                         break;
1356                 case Opt_nomballoc:
1357                         clear_opt(sbi->s_mount_opt, MBALLOC);
1358                         break;
1359                 case Opt_stripe:
1360                         if (match_int(&args[0], &option))
1361                                 return 0;
1362                         if (option < 0)
1363                                 return 0;
1364                         sbi->s_stripe = option;
1365                         break;
1366                 case Opt_delalloc:
1367                         set_opt(sbi->s_mount_opt, DELALLOC);
1368                         break;
1369                 default:
1370                         printk (KERN_ERR
1371                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1372                                 "or missing value\n", p);
1373                         return 0;
1374                 }
1375         }
1376 #ifdef CONFIG_QUOTA
1377         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1378                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1379                      sbi->s_qf_names[USRQUOTA])
1380                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1381
1382                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1383                      sbi->s_qf_names[GRPQUOTA])
1384                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1385
1386                 if ((sbi->s_qf_names[USRQUOTA] &&
1387                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1388                     (sbi->s_qf_names[GRPQUOTA] &&
1389                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1390                         printk(KERN_ERR "EXT4-fs: old and new quota "
1391                                         "format mixing.\n");
1392                         return 0;
1393                 }
1394
1395                 if (!sbi->s_jquota_fmt) {
1396                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1397                                         "not specified.\n");
1398                         return 0;
1399                 }
1400         } else {
1401                 if (sbi->s_jquota_fmt) {
1402                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1403                                         "specified with no journaling "
1404                                         "enabled.\n");
1405                         return 0;
1406                 }
1407         }
1408 #endif
1409         return 1;
1410 }
1411
1412 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1413                             int read_only)
1414 {
1415         struct ext4_sb_info *sbi = EXT4_SB(sb);
1416         int res = 0;
1417
1418         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1419                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1420                         "forcing read-only mode\n");
1421                 res = MS_RDONLY;
1422         }
1423         if (read_only)
1424                 return res;
1425         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1426                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1427                         "running e2fsck is recommended\n");
1428         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1429                 printk (KERN_WARNING
1430                         "EXT4-fs warning: mounting fs with errors, "
1431                         "running e2fsck is recommended\n");
1432         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1433                  le16_to_cpu(es->s_mnt_count) >=
1434                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1435                 printk (KERN_WARNING
1436                         "EXT4-fs warning: maximal mount count reached, "
1437                         "running e2fsck is recommended\n");
1438         else if (le32_to_cpu(es->s_checkinterval) &&
1439                 (le32_to_cpu(es->s_lastcheck) +
1440                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1441                 printk (KERN_WARNING
1442                         "EXT4-fs warning: checktime reached, "
1443                         "running e2fsck is recommended\n");
1444 #if 0
1445                 /* @@@ We _will_ want to clear the valid bit if we find
1446                  * inconsistencies, to force a fsck at reboot.  But for
1447                  * a plain journaled filesystem we can keep it set as
1448                  * valid forever! :)
1449                  */
1450         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1451 #endif
1452         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1453                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1454         le16_add_cpu(&es->s_mnt_count, 1);
1455         es->s_mtime = cpu_to_le32(get_seconds());
1456         ext4_update_dynamic_rev(sb);
1457         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1458
1459         ext4_commit_super(sb, es, 1);
1460         if (test_opt(sb, DEBUG))
1461                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1462                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1463                         sb->s_blocksize,
1464                         sbi->s_groups_count,
1465                         EXT4_BLOCKS_PER_GROUP(sb),
1466                         EXT4_INODES_PER_GROUP(sb),
1467                         sbi->s_mount_opt);
1468
1469         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1470         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1471                 char b[BDEVNAME_SIZE];
1472
1473                 printk("external journal on %s\n",
1474                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1475         } else {
1476                 printk("internal journal\n");
1477         }
1478         return res;
1479 }
1480
1481 static int ext4_fill_flex_info(struct super_block *sb)
1482 {
1483         struct ext4_sb_info *sbi = EXT4_SB(sb);
1484         struct ext4_group_desc *gdp = NULL;
1485         struct buffer_head *bh;
1486         ext4_group_t flex_group_count;
1487         ext4_group_t flex_group;
1488         int groups_per_flex = 0;
1489         __u64 block_bitmap = 0;
1490         int i;
1491
1492         if (!sbi->s_es->s_log_groups_per_flex) {
1493                 sbi->s_log_groups_per_flex = 0;
1494                 return 1;
1495         }
1496
1497         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1498         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1499
1500         flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
1501                 groups_per_flex;
1502         sbi->s_flex_groups = kmalloc(flex_group_count *
1503                                      sizeof(struct flex_groups), GFP_KERNEL);
1504         if (sbi->s_flex_groups == NULL) {
1505                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
1506                 goto failed;
1507         }
1508         memset(sbi->s_flex_groups, 0, flex_group_count *
1509                sizeof(struct flex_groups));
1510
1511         gdp = ext4_get_group_desc(sb, 1, &bh);
1512         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1513
1514         for (i = 0; i < sbi->s_groups_count; i++) {
1515                 gdp = ext4_get_group_desc(sb, i, &bh);
1516
1517                 flex_group = ext4_flex_group(sbi, i);
1518                 sbi->s_flex_groups[flex_group].free_inodes +=
1519                         le16_to_cpu(gdp->bg_free_inodes_count);
1520                 sbi->s_flex_groups[flex_group].free_blocks +=
1521                         le16_to_cpu(gdp->bg_free_blocks_count);
1522         }
1523
1524         return 1;
1525 failed:
1526         return 0;
1527 }
1528
1529 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1530                             struct ext4_group_desc *gdp)
1531 {
1532         __u16 crc = 0;
1533
1534         if (sbi->s_es->s_feature_ro_compat &
1535             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1536                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1537                 __le32 le_group = cpu_to_le32(block_group);
1538
1539                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1540                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1541                 crc = crc16(crc, (__u8 *)gdp, offset);
1542                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1543                 /* for checksum of struct ext4_group_desc do the rest...*/
1544                 if ((sbi->s_es->s_feature_incompat &
1545                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1546                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1547                         crc = crc16(crc, (__u8 *)gdp + offset,
1548                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1549                                         offset);
1550         }
1551
1552         return cpu_to_le16(crc);
1553 }
1554
1555 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1556                                 struct ext4_group_desc *gdp)
1557 {
1558         if ((sbi->s_es->s_feature_ro_compat &
1559              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1560             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1561                 return 0;
1562
1563         return 1;
1564 }
1565
1566 /* Called at mount-time, super-block is locked */
1567 static int ext4_check_descriptors(struct super_block *sb)
1568 {
1569         struct ext4_sb_info *sbi = EXT4_SB(sb);
1570         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1571         ext4_fsblk_t last_block;
1572         ext4_fsblk_t block_bitmap;
1573         ext4_fsblk_t inode_bitmap;
1574         ext4_fsblk_t inode_table;
1575         int flexbg_flag = 0;
1576         ext4_group_t i;
1577
1578         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1579                 flexbg_flag = 1;
1580
1581         ext4_debug ("Checking group descriptors");
1582
1583         for (i = 0; i < sbi->s_groups_count; i++) {
1584                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1585
1586                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1587                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1588                 else
1589                         last_block = first_block +
1590                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1591
1592                 block_bitmap = ext4_block_bitmap(sb, gdp);
1593                 if (block_bitmap < first_block || block_bitmap > last_block)
1594                 {
1595                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1596                                "Block bitmap for group %lu not in group "
1597                                "(block %llu)!", i, block_bitmap);
1598                         return 0;
1599                 }
1600                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1601                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1602                 {
1603                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1604                                "Inode bitmap for group %lu not in group "
1605                                "(block %llu)!", i, inode_bitmap);
1606                         return 0;
1607                 }
1608                 inode_table = ext4_inode_table(sb, gdp);
1609                 if (inode_table < first_block ||
1610                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1611                 {
1612                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1613                                "Inode table for group %lu not in group "
1614                                "(block %llu)!", i, inode_table);
1615                         return 0;
1616                 }
1617                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1618                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1619                                "Checksum for group %lu failed (%u!=%u)\n",
1620                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1621                                gdp)), le16_to_cpu(gdp->bg_checksum));
1622                         return 0;
1623                 }
1624                 if (!flexbg_flag)
1625                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1626         }
1627
1628         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1629         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1630         return 1;
1631 }
1632
1633 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1634  * the superblock) which were deleted from all directories, but held open by
1635  * a process at the time of a crash.  We walk the list and try to delete these
1636  * inodes at recovery time (only with a read-write filesystem).
1637  *
1638  * In order to keep the orphan inode chain consistent during traversal (in
1639  * case of crash during recovery), we link each inode into the superblock
1640  * orphan list_head and handle it the same way as an inode deletion during
1641  * normal operation (which journals the operations for us).
1642  *
1643  * We only do an iget() and an iput() on each inode, which is very safe if we
1644  * accidentally point at an in-use or already deleted inode.  The worst that
1645  * can happen in this case is that we get a "bit already cleared" message from
1646  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1647  * e2fsck was run on this filesystem, and it must have already done the orphan
1648  * inode cleanup for us, so we can safely abort without any further action.
1649  */
1650 static void ext4_orphan_cleanup (struct super_block * sb,
1651                                  struct ext4_super_block * es)
1652 {
1653         unsigned int s_flags = sb->s_flags;
1654         int nr_orphans = 0, nr_truncates = 0;
1655 #ifdef CONFIG_QUOTA
1656         int i;
1657 #endif
1658         if (!es->s_last_orphan) {
1659                 jbd_debug(4, "no orphan inodes to clean up\n");
1660                 return;
1661         }
1662
1663         if (bdev_read_only(sb->s_bdev)) {
1664                 printk(KERN_ERR "EXT4-fs: write access "
1665                         "unavailable, skipping orphan cleanup.\n");
1666                 return;
1667         }
1668
1669         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1670                 if (es->s_last_orphan)
1671                         jbd_debug(1, "Errors on filesystem, "
1672                                   "clearing orphan list.\n");
1673                 es->s_last_orphan = 0;
1674                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1675                 return;
1676         }
1677
1678         if (s_flags & MS_RDONLY) {
1679                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1680                        sb->s_id);
1681                 sb->s_flags &= ~MS_RDONLY;
1682         }
1683 #ifdef CONFIG_QUOTA
1684         /* Needed for iput() to work correctly and not trash data */
1685         sb->s_flags |= MS_ACTIVE;
1686         /* Turn on quotas so that they are updated correctly */
1687         for (i = 0; i < MAXQUOTAS; i++) {
1688                 if (EXT4_SB(sb)->s_qf_names[i]) {
1689                         int ret = ext4_quota_on_mount(sb, i);
1690                         if (ret < 0)
1691                                 printk(KERN_ERR
1692                                         "EXT4-fs: Cannot turn on journaled "
1693                                         "quota: error %d\n", ret);
1694                 }
1695         }
1696 #endif
1697
1698         while (es->s_last_orphan) {
1699                 struct inode *inode;
1700
1701                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1702                 if (IS_ERR(inode)) {
1703                         es->s_last_orphan = 0;
1704                         break;
1705                 }
1706
1707                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1708                 DQUOT_INIT(inode);
1709                 if (inode->i_nlink) {
1710                         printk(KERN_DEBUG
1711                                 "%s: truncating inode %lu to %Ld bytes\n",
1712                                 __func__, inode->i_ino, inode->i_size);
1713                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1714                                   inode->i_ino, inode->i_size);
1715                         ext4_truncate(inode);
1716                         nr_truncates++;
1717                 } else {
1718                         printk(KERN_DEBUG
1719                                 "%s: deleting unreferenced inode %lu\n",
1720                                 __func__, inode->i_ino);
1721                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1722                                   inode->i_ino);
1723                         nr_orphans++;
1724                 }
1725                 iput(inode);  /* The delete magic happens here! */
1726         }
1727
1728 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1729
1730         if (nr_orphans)
1731                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1732                        sb->s_id, PLURAL(nr_orphans));
1733         if (nr_truncates)
1734                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1735                        sb->s_id, PLURAL(nr_truncates));
1736 #ifdef CONFIG_QUOTA
1737         /* Turn quotas off */
1738         for (i = 0; i < MAXQUOTAS; i++) {
1739                 if (sb_dqopt(sb)->files[i])
1740                         vfs_quota_off(sb, i, 0);
1741         }
1742 #endif
1743         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1744 }
1745 /*
1746  * Maximal extent format file size.
1747  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1748  * extent format containers, within a sector_t, and within i_blocks
1749  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1750  * so that won't be a limiting factor.
1751  *
1752  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1753  */
1754 static loff_t ext4_max_size(int blkbits)
1755 {
1756         loff_t res;
1757         loff_t upper_limit = MAX_LFS_FILESIZE;
1758
1759         /* small i_blocks in vfs inode? */
1760         if (sizeof(blkcnt_t) < sizeof(u64)) {
1761                 /*
1762                  * CONFIG_LSF is not enabled implies the inode
1763                  * i_block represent total blocks in 512 bytes
1764                  * 32 == size of vfs inode i_blocks * 8
1765                  */
1766                 upper_limit = (1LL << 32) - 1;
1767
1768                 /* total blocks in file system block size */
1769                 upper_limit >>= (blkbits - 9);
1770                 upper_limit <<= blkbits;
1771         }
1772
1773         /* 32-bit extent-start container, ee_block */
1774         res = 1LL << 32;
1775         res <<= blkbits;
1776         res -= 1;
1777
1778         /* Sanity check against vm- & vfs- imposed limits */
1779         if (res > upper_limit)
1780                 res = upper_limit;
1781
1782         return res;
1783 }
1784
1785 /*
1786  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1787  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1788  * We need to be 1 filesystem block less than the 2^48 sector limit.
1789  */
1790 static loff_t ext4_max_bitmap_size(int bits)
1791 {
1792         loff_t res = EXT4_NDIR_BLOCKS;
1793         int meta_blocks;
1794         loff_t upper_limit;
1795         /* This is calculated to be the largest file size for a
1796          * dense, bitmapped file such that the total number of
1797          * sectors in the file, including data and all indirect blocks,
1798          * does not exceed 2^48 -1
1799          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1800          * total number of  512 bytes blocks of the file
1801          */
1802
1803         if (sizeof(blkcnt_t) < sizeof(u64)) {
1804                 /*
1805                  * CONFIG_LSF is not enabled implies the inode
1806                  * i_block represent total blocks in 512 bytes
1807                  * 32 == size of vfs inode i_blocks * 8
1808                  */
1809                 upper_limit = (1LL << 32) - 1;
1810
1811                 /* total blocks in file system block size */
1812                 upper_limit >>= (bits - 9);
1813
1814         } else {
1815                 /*
1816                  * We use 48 bit ext4_inode i_blocks
1817                  * With EXT4_HUGE_FILE_FL set the i_blocks
1818                  * represent total number of blocks in
1819                  * file system block size
1820                  */
1821                 upper_limit = (1LL << 48) - 1;
1822
1823         }
1824
1825         /* indirect blocks */
1826         meta_blocks = 1;
1827         /* double indirect blocks */
1828         meta_blocks += 1 + (1LL << (bits-2));
1829         /* tripple indirect blocks */
1830         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1831
1832         upper_limit -= meta_blocks;
1833         upper_limit <<= bits;
1834
1835         res += 1LL << (bits-2);
1836         res += 1LL << (2*(bits-2));
1837         res += 1LL << (3*(bits-2));
1838         res <<= bits;
1839         if (res > upper_limit)
1840                 res = upper_limit;
1841
1842         if (res > MAX_LFS_FILESIZE)
1843                 res = MAX_LFS_FILESIZE;
1844
1845         return res;
1846 }
1847
1848 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1849                                 ext4_fsblk_t logical_sb_block, int nr)
1850 {
1851         struct ext4_sb_info *sbi = EXT4_SB(sb);
1852         ext4_group_t bg, first_meta_bg;
1853         int has_super = 0;
1854
1855         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1856
1857         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1858             nr < first_meta_bg)
1859                 return logical_sb_block + nr + 1;
1860         bg = sbi->s_desc_per_block * nr;
1861         if (ext4_bg_has_super(sb, bg))
1862                 has_super = 1;
1863         return (has_super + ext4_group_first_block_no(sb, bg));
1864 }
1865
1866 /**
1867  * ext4_get_stripe_size: Get the stripe size.
1868  * @sbi: In memory super block info
1869  *
1870  * If we have specified it via mount option, then
1871  * use the mount option value. If the value specified at mount time is
1872  * greater than the blocks per group use the super block value.
1873  * If the super block value is greater than blocks per group return 0.
1874  * Allocator needs it be less than blocks per group.
1875  *
1876  */
1877 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1878 {
1879         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1880         unsigned long stripe_width =
1881                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1882
1883         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1884                 return sbi->s_stripe;
1885
1886         if (stripe_width <= sbi->s_blocks_per_group)
1887                 return stripe_width;
1888
1889         if (stride <= sbi->s_blocks_per_group)
1890                 return stride;
1891
1892         return 0;
1893 }
1894
1895 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1896                                 __releases(kernel_lock)
1897                                 __acquires(kernel_lock)
1898
1899 {
1900         struct buffer_head * bh;
1901         struct ext4_super_block *es = NULL;
1902         struct ext4_sb_info *sbi;
1903         ext4_fsblk_t block;
1904         ext4_fsblk_t sb_block = get_sb_block(&data);
1905         ext4_fsblk_t logical_sb_block;
1906         unsigned long offset = 0;
1907         unsigned int journal_inum = 0;
1908         unsigned long journal_devnum = 0;
1909         unsigned long def_mount_opts;
1910         struct inode *root;
1911         int ret = -EINVAL;
1912         int blocksize;
1913         int db_count;
1914         int i;
1915         int needs_recovery;
1916         __le32 features;
1917         __u64 blocks_count;
1918         int err;
1919
1920         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1921         if (!sbi)
1922                 return -ENOMEM;
1923         sb->s_fs_info = sbi;
1924         sbi->s_mount_opt = 0;
1925         sbi->s_resuid = EXT4_DEF_RESUID;
1926         sbi->s_resgid = EXT4_DEF_RESGID;
1927         sbi->s_sb_block = sb_block;
1928
1929         unlock_kernel();
1930
1931         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1932         if (!blocksize) {
1933                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1934                 goto out_fail;
1935         }
1936
1937         /*
1938          * The ext4 superblock will not be buffer aligned for other than 1kB
1939          * block sizes.  We need to calculate the offset from buffer start.
1940          */
1941         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1942                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1943                 offset = do_div(logical_sb_block, blocksize);
1944         } else {
1945                 logical_sb_block = sb_block;
1946         }
1947
1948         if (!(bh = sb_bread(sb, logical_sb_block))) {
1949                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1950                 goto out_fail;
1951         }
1952         /*
1953          * Note: s_es must be initialized as soon as possible because
1954          *       some ext4 macro-instructions depend on its value
1955          */
1956         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1957         sbi->s_es = es;
1958         sb->s_magic = le16_to_cpu(es->s_magic);
1959         if (sb->s_magic != EXT4_SUPER_MAGIC)
1960                 goto cantfind_ext4;
1961
1962         /* Set defaults before we parse the mount options */
1963         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1964         if (def_mount_opts & EXT4_DEFM_DEBUG)
1965                 set_opt(sbi->s_mount_opt, DEBUG);
1966         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1967                 set_opt(sbi->s_mount_opt, GRPID);
1968         if (def_mount_opts & EXT4_DEFM_UID16)
1969                 set_opt(sbi->s_mount_opt, NO_UID32);
1970 #ifdef CONFIG_EXT4DEV_FS_XATTR
1971         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1972                 set_opt(sbi->s_mount_opt, XATTR_USER);
1973 #endif
1974 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1975         if (def_mount_opts & EXT4_DEFM_ACL)
1976                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1977 #endif
1978         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1979                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1980         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1981                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1982         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1983                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1984
1985         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1986                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1987         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1988                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1989         else
1990                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1991
1992         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1993         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1994
1995         set_opt(sbi->s_mount_opt, RESERVATION);
1996         set_opt(sbi->s_mount_opt, BARRIER);
1997
1998         /*
1999          * turn on extents feature by default in ext4 filesystem
2000          * User -o noextents to turn it off
2001          */
2002         set_opt(sbi->s_mount_opt, EXTENTS);
2003         /*
2004          * turn on mballoc feature by default in ext4 filesystem
2005          * User -o nomballoc to turn it off
2006          */
2007         set_opt(sbi->s_mount_opt, MBALLOC);
2008
2009         /*
2010          * enable delayed allocation by default
2011          * Use -o nodelalloc to turn it off
2012          */
2013         set_opt(sbi->s_mount_opt, DELALLOC);
2014
2015
2016         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
2017                             NULL, 0))
2018                 goto failed_mount;
2019
2020         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2021                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2022
2023         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2024             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2025              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2026              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2027                 printk(KERN_WARNING
2028                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2029                        "running e2fsck is recommended\n");
2030
2031         /*
2032          * Since ext4 is still considered development code, we require
2033          * that the TEST_FILESYS flag in s->flags be set.
2034          */
2035         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
2036                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
2037                        "OK to use with test code.\n", sb->s_id);
2038                 goto failed_mount;
2039         }
2040
2041         /*
2042          * Check feature flags regardless of the revision level, since we
2043          * previously didn't change the revision level when setting the flags,
2044          * so there is a chance incompat flags are set on a rev 0 filesystem.
2045          */
2046         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2047         if (features) {
2048                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2049                        "unsupported optional features (%x).\n",
2050                        sb->s_id, le32_to_cpu(features));
2051                 goto failed_mount;
2052         }
2053         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2054         if (!(sb->s_flags & MS_RDONLY) && features) {
2055                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2056                        "unsupported optional features (%x).\n",
2057                        sb->s_id, le32_to_cpu(features));
2058                 goto failed_mount;
2059         }
2060         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2061                 /*
2062                  * Large file size enabled file system can only be
2063                  * mount if kernel is build with CONFIG_LSF
2064                  */
2065                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2066                                 !(sb->s_flags & MS_RDONLY)) {
2067                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2068                                         "files cannot be mounted read-write "
2069                                         "without CONFIG_LSF.\n", sb->s_id);
2070                         goto failed_mount;
2071                 }
2072         }
2073         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2074
2075         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2076             blocksize > EXT4_MAX_BLOCK_SIZE) {
2077                 printk(KERN_ERR
2078                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2079                        blocksize, sb->s_id);
2080                 goto failed_mount;
2081         }
2082
2083         if (sb->s_blocksize != blocksize) {
2084
2085                 /* Validate the filesystem blocksize */
2086                 if (!sb_set_blocksize(sb, blocksize)) {
2087                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2088                                         blocksize);
2089                         goto failed_mount;
2090                 }
2091
2092                 brelse (bh);
2093                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2094                 offset = do_div(logical_sb_block, blocksize);
2095                 bh = sb_bread(sb, logical_sb_block);
2096                 if (!bh) {
2097                         printk(KERN_ERR
2098                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2099                         goto failed_mount;
2100                 }
2101                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2102                 sbi->s_es = es;
2103                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2104                         printk (KERN_ERR
2105                                 "EXT4-fs: Magic mismatch, very weird !\n");
2106                         goto failed_mount;
2107                 }
2108         }
2109
2110         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2111         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2112
2113         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2114                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2115                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2116         } else {
2117                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2118                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2119                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2120                     (!is_power_of_2(sbi->s_inode_size)) ||
2121                     (sbi->s_inode_size > blocksize)) {
2122                         printk (KERN_ERR
2123                                 "EXT4-fs: unsupported inode size: %d\n",
2124                                 sbi->s_inode_size);
2125                         goto failed_mount;
2126                 }
2127                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2128                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2129         }
2130         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2131         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2132                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2133                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2134                     !is_power_of_2(sbi->s_desc_size)) {
2135                         printk(KERN_ERR
2136                                "EXT4-fs: unsupported descriptor size %lu\n",
2137                                sbi->s_desc_size);
2138                         goto failed_mount;
2139                 }
2140         } else
2141                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2142         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2143         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2144         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2145                 goto cantfind_ext4;
2146         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2147         if (sbi->s_inodes_per_block == 0)
2148                 goto cantfind_ext4;
2149         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2150                                         sbi->s_inodes_per_block;
2151         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2152         sbi->s_sbh = bh;
2153         sbi->s_mount_state = le16_to_cpu(es->s_state);
2154         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2155         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2156         for (i=0; i < 4; i++)
2157                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2158         sbi->s_def_hash_version = es->s_def_hash_version;
2159
2160         if (sbi->s_blocks_per_group > blocksize * 8) {
2161                 printk (KERN_ERR
2162                         "EXT4-fs: #blocks per group too big: %lu\n",
2163                         sbi->s_blocks_per_group);
2164                 goto failed_mount;
2165         }
2166         if (sbi->s_inodes_per_group > blocksize * 8) {
2167                 printk (KERN_ERR
2168                         "EXT4-fs: #inodes per group too big: %lu\n",
2169                         sbi->s_inodes_per_group);
2170                 goto failed_mount;
2171         }
2172
2173         if (ext4_blocks_count(es) >
2174                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2175                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2176                         " too large to mount safely\n", sb->s_id);
2177                 if (sizeof(sector_t) < 8)
2178                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2179                                         "enabled\n");
2180                 goto failed_mount;
2181         }
2182
2183         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2184                 goto cantfind_ext4;
2185
2186         /* ensure blocks_count calculation below doesn't sign-extend */
2187         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2188             le32_to_cpu(es->s_first_data_block) + 1) {
2189                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2190                        "first data block %u, blocks per group %lu\n",
2191                         ext4_blocks_count(es),
2192                         le32_to_cpu(es->s_first_data_block),
2193                         EXT4_BLOCKS_PER_GROUP(sb));
2194                 goto failed_mount;
2195         }
2196         blocks_count = (ext4_blocks_count(es) -
2197                         le32_to_cpu(es->s_first_data_block) +
2198                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2199         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2200         sbi->s_groups_count = blocks_count;
2201         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2202                    EXT4_DESC_PER_BLOCK(sb);
2203         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2204                                     GFP_KERNEL);
2205         if (sbi->s_group_desc == NULL) {
2206                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2207                 goto failed_mount;
2208         }
2209
2210         bgl_lock_init(&sbi->s_blockgroup_lock);
2211
2212         for (i = 0; i < db_count; i++) {
2213                 block = descriptor_loc(sb, logical_sb_block, i);
2214                 sbi->s_group_desc[i] = sb_bread(sb, block);
2215                 if (!sbi->s_group_desc[i]) {
2216                         printk (KERN_ERR "EXT4-fs: "
2217                                 "can't read group descriptor %d\n", i);
2218                         db_count = i;
2219                         goto failed_mount2;
2220                 }
2221         }
2222         if (!ext4_check_descriptors (sb)) {
2223                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2224                 goto failed_mount2;
2225         }
2226         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2227                 if (!ext4_fill_flex_info(sb)) {
2228                         printk(KERN_ERR
2229                                "EXT4-fs: unable to initialize "
2230                                "flex_bg meta info!\n");
2231                         goto failed_mount2;
2232                 }
2233
2234         sbi->s_gdb_count = db_count;
2235         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2236         spin_lock_init(&sbi->s_next_gen_lock);
2237
2238         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2239                         ext4_count_free_blocks(sb));
2240         if (!err) {
2241                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2242                                 ext4_count_free_inodes(sb));
2243         }
2244         if (!err) {
2245                 err = percpu_counter_init(&sbi->s_dirs_counter,
2246                                 ext4_count_dirs(sb));
2247         }
2248         if (err) {
2249                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2250                 goto failed_mount3;
2251         }
2252
2253         /* per fileystem reservation list head & lock */
2254         spin_lock_init(&sbi->s_rsv_window_lock);
2255         sbi->s_rsv_window_root = RB_ROOT;
2256         /* Add a single, static dummy reservation to the start of the
2257          * reservation window list --- it gives us a placeholder for
2258          * append-at-start-of-list which makes the allocation logic
2259          * _much_ simpler. */
2260         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2261         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2262         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2263         sbi->s_rsv_window_head.rsv_goal_size = 0;
2264         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2265
2266         sbi->s_stripe = ext4_get_stripe_size(sbi);
2267
2268         /*
2269          * set up enough so that it can read an inode
2270          */
2271         sb->s_op = &ext4_sops;
2272         sb->s_export_op = &ext4_export_ops;
2273         sb->s_xattr = ext4_xattr_handlers;
2274 #ifdef CONFIG_QUOTA
2275         sb->s_qcop = &ext4_qctl_operations;
2276         sb->dq_op = &ext4_quota_operations;
2277 #endif
2278         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2279
2280         sb->s_root = NULL;
2281
2282         needs_recovery = (es->s_last_orphan != 0 ||
2283                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2284                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2285
2286         /*
2287          * The first inode we look at is the journal inode.  Don't try
2288          * root first: it may be modified in the journal!
2289          */
2290         if (!test_opt(sb, NOLOAD) &&
2291             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2292                 if (ext4_load_journal(sb, es, journal_devnum))
2293                         goto failed_mount3;
2294                 if (!(sb->s_flags & MS_RDONLY) &&
2295                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2296                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2297                                "ext4_fill_super: Journal transaction "
2298                                "%u is corrupt\n", sb->s_id,
2299                                EXT4_SB(sb)->s_journal->j_failed_commit);
2300                         if (test_opt (sb, ERRORS_RO)) {
2301                                 printk (KERN_CRIT
2302                                         "Mounting filesystem read-only\n");
2303                                 sb->s_flags |= MS_RDONLY;
2304                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2305                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2306                         }
2307                         if (test_opt(sb, ERRORS_PANIC)) {
2308                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2309                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2310                                 ext4_commit_super(sb, es, 1);
2311                                 printk(KERN_CRIT
2312                                        "EXT4-fs (device %s): mount failed\n",
2313                                       sb->s_id);
2314                                 goto failed_mount4;
2315                         }
2316                 }
2317         } else if (journal_inum) {
2318                 if (ext4_create_journal(sb, es, journal_inum))
2319                         goto failed_mount3;
2320         } else {
2321                 if (!silent)
2322                         printk (KERN_ERR
2323                                 "ext4: No journal on filesystem on %s\n",
2324                                 sb->s_id);
2325                 goto failed_mount3;
2326         }
2327
2328         if (ext4_blocks_count(es) > 0xffffffffULL &&
2329             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2330                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2331                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2332                 goto failed_mount4;
2333         }
2334
2335         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2336                 jbd2_journal_set_features(sbi->s_journal,
2337                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2338                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2339         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2340                 jbd2_journal_set_features(sbi->s_journal,
2341                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2342                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2343                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2344         } else {
2345                 jbd2_journal_clear_features(sbi->s_journal,
2346                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2347                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2348         }
2349
2350         /* We have now updated the journal if required, so we can
2351          * validate the data journaling mode. */
2352         switch (test_opt(sb, DATA_FLAGS)) {
2353         case 0:
2354                 /* No mode set, assume a default based on the journal
2355                  * capabilities: ORDERED_DATA if the journal can
2356                  * cope, else JOURNAL_DATA
2357                  */
2358                 if (jbd2_journal_check_available_features
2359                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2360                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2361                 else
2362                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2363                 break;
2364
2365         case EXT4_MOUNT_ORDERED_DATA:
2366         case EXT4_MOUNT_WRITEBACK_DATA:
2367                 if (!jbd2_journal_check_available_features
2368                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2369                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2370                                "requested data journaling mode\n");
2371                         goto failed_mount4;
2372                 }
2373         default:
2374                 break;
2375         }
2376
2377         if (test_opt(sb, NOBH)) {
2378                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2379                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2380                                 "its supported only with writeback mode\n");
2381                         clear_opt(sbi->s_mount_opt, NOBH);
2382                 }
2383         }
2384         /*
2385          * The jbd2_journal_load will have done any necessary log recovery,
2386          * so we can safely mount the rest of the filesystem now.
2387          */
2388
2389         root = ext4_iget(sb, EXT4_ROOT_INO);
2390         if (IS_ERR(root)) {
2391                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2392                 ret = PTR_ERR(root);
2393                 goto failed_mount4;
2394         }
2395         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2396                 iput(root);
2397                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2398                 goto failed_mount4;
2399         }
2400         sb->s_root = d_alloc_root(root);
2401         if (!sb->s_root) {
2402                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2403                 iput(root);
2404                 ret = -ENOMEM;
2405                 goto failed_mount4;
2406         }
2407
2408         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2409
2410         /* determine the minimum size of new large inodes, if present */
2411         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2412                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2413                                                      EXT4_GOOD_OLD_INODE_SIZE;
2414                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2415                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2416                         if (sbi->s_want_extra_isize <
2417                             le16_to_cpu(es->s_want_extra_isize))
2418                                 sbi->s_want_extra_isize =
2419                                         le16_to_cpu(es->s_want_extra_isize);
2420                         if (sbi->s_want_extra_isize <
2421                             le16_to_cpu(es->s_min_extra_isize))
2422                                 sbi->s_want_extra_isize =
2423                                         le16_to_cpu(es->s_min_extra_isize);
2424                 }
2425         }
2426         /* Check if enough inode space is available */
2427         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2428                                                         sbi->s_inode_size) {
2429                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2430                                                        EXT4_GOOD_OLD_INODE_SIZE;
2431                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2432                         "available.\n");
2433         }
2434
2435         /*
2436          * akpm: core read_super() calls in here with the superblock locked.
2437          * That deadlocks, because orphan cleanup needs to lock the superblock
2438          * in numerous places.  Here we just pop the lock - it's relatively
2439          * harmless, because we are now ready to accept write_super() requests,
2440          * and aviro says that's the only reason for hanging onto the
2441          * superblock lock.
2442          */
2443         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2444         ext4_orphan_cleanup(sb, es);
2445         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2446         if (needs_recovery)
2447                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2448         ext4_mark_recovery_complete(sb, es);
2449         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2450                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2451                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2452                 "writeback");
2453
2454         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2455                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2456                                 "requested data journaling mode\n");
2457                 clear_opt(sbi->s_mount_opt, DELALLOC);
2458         } else if (test_opt(sb, DELALLOC))
2459                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2460
2461         ext4_ext_init(sb);
2462         ext4_mb_init(sb, needs_recovery);
2463
2464         lock_kernel();
2465         return 0;
2466
2467 cantfind_ext4:
2468         if (!silent)
2469                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2470                        sb->s_id);
2471         goto failed_mount;
2472
2473 failed_mount4:
2474         jbd2_journal_destroy(sbi->s_journal);
2475         sbi->s_journal = NULL;
2476 failed_mount3:
2477         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2478         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2479         percpu_counter_destroy(&sbi->s_dirs_counter);
2480 failed_mount2:
2481         for (i = 0; i < db_count; i++)
2482                 brelse(sbi->s_group_desc[i]);
2483         kfree(sbi->s_group_desc);
2484 failed_mount:
2485 #ifdef CONFIG_QUOTA
2486         for (i = 0; i < MAXQUOTAS; i++)
2487                 kfree(sbi->s_qf_names[i]);
2488 #endif
2489         ext4_blkdev_remove(sbi);
2490         brelse(bh);
2491 out_fail:
2492         sb->s_fs_info = NULL;
2493         kfree(sbi);
2494         lock_kernel();
2495         return ret;
2496 }
2497
2498 /*
2499  * Setup any per-fs journal parameters now.  We'll do this both on
2500  * initial mount, once the journal has been initialised but before we've
2501  * done any recovery; and again on any subsequent remount.
2502  */
2503 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2504 {
2505         struct ext4_sb_info *sbi = EXT4_SB(sb);
2506
2507         if (sbi->s_commit_interval)
2508                 journal->j_commit_interval = sbi->s_commit_interval;
2509         /* We could also set up an ext4-specific default for the commit
2510          * interval here, but for now we'll just fall back to the jbd
2511          * default. */
2512
2513         spin_lock(&journal->j_state_lock);
2514         if (test_opt(sb, BARRIER))
2515                 journal->j_flags |= JBD2_BARRIER;
2516         else
2517                 journal->j_flags &= ~JBD2_BARRIER;
2518         spin_unlock(&journal->j_state_lock);
2519 }
2520
2521 static journal_t *ext4_get_journal(struct super_block *sb,
2522                                    unsigned int journal_inum)
2523 {
2524         struct inode *journal_inode;
2525         journal_t *journal;
2526
2527         /* First, test for the existence of a valid inode on disk.  Bad
2528          * things happen if we iget() an unused inode, as the subsequent
2529          * iput() will try to delete it. */
2530
2531         journal_inode = ext4_iget(sb, journal_inum);
2532         if (IS_ERR(journal_inode)) {
2533                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2534                 return NULL;
2535         }
2536         if (!journal_inode->i_nlink) {
2537                 make_bad_inode(journal_inode);
2538                 iput(journal_inode);
2539                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2540                 return NULL;
2541         }
2542
2543         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2544                   journal_inode, journal_inode->i_size);
2545         if (!S_ISREG(journal_inode->i_mode)) {
2546                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2547                 iput(journal_inode);
2548                 return NULL;
2549         }
2550
2551         journal = jbd2_journal_init_inode(journal_inode);
2552         if (!journal) {
2553                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2554                 iput(journal_inode);
2555                 return NULL;
2556         }
2557         journal->j_private = sb;
2558         ext4_init_journal_params(sb, journal);
2559         return journal;
2560 }
2561
2562 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2563                                        dev_t j_dev)
2564 {
2565         struct buffer_head * bh;
2566         journal_t *journal;
2567         ext4_fsblk_t start;
2568         ext4_fsblk_t len;
2569         int hblock, blocksize;
2570         ext4_fsblk_t sb_block;
2571         unsigned long offset;
2572         struct ext4_super_block * es;
2573         struct block_device *bdev;
2574
2575         bdev = ext4_blkdev_get(j_dev);
2576         if (bdev == NULL)
2577                 return NULL;
2578
2579         if (bd_claim(bdev, sb)) {
2580                 printk(KERN_ERR
2581                         "EXT4: failed to claim external journal device.\n");
2582                 blkdev_put(bdev);
2583                 return NULL;
2584         }
2585
2586         blocksize = sb->s_blocksize;
2587         hblock = bdev_hardsect_size(bdev);
2588         if (blocksize < hblock) {
2589                 printk(KERN_ERR
2590                         "EXT4-fs: blocksize too small for journal device.\n");
2591                 goto out_bdev;
2592         }
2593
2594         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2595         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2596         set_blocksize(bdev, blocksize);
2597         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2598                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2599                        "external journal\n");
2600                 goto out_bdev;
2601         }
2602
2603         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2604         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2605             !(le32_to_cpu(es->s_feature_incompat) &
2606               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2607                 printk(KERN_ERR "EXT4-fs: external journal has "
2608                                         "bad superblock\n");
2609                 brelse(bh);
2610                 goto out_bdev;
2611         }
2612
2613         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2614                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2615                 brelse(bh);
2616                 goto out_bdev;
2617         }
2618
2619         len = ext4_blocks_count(es);
2620         start = sb_block + 1;
2621         brelse(bh);     /* we're done with the superblock */
2622
2623         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2624                                         start, len, blocksize);
2625         if (!journal) {
2626                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2627                 goto out_bdev;
2628         }
2629         journal->j_private = sb;
2630         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2631         wait_on_buffer(journal->j_sb_buffer);
2632         if (!buffer_uptodate(journal->j_sb_buffer)) {
2633                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2634                 goto out_journal;
2635         }
2636         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2637                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2638                                         "user (unsupported) - %d\n",
2639                         be32_to_cpu(journal->j_superblock->s_nr_users));
2640                 goto out_journal;
2641         }
2642         EXT4_SB(sb)->journal_bdev = bdev;
2643         ext4_init_journal_params(sb, journal);
2644         return journal;
2645 out_journal:
2646         jbd2_journal_destroy(journal);
2647 out_bdev:
2648         ext4_blkdev_put(bdev);
2649         return NULL;
2650 }
2651
2652 static int ext4_load_journal(struct super_block *sb,
2653                              struct ext4_super_block *es,
2654                              unsigned long journal_devnum)
2655 {
2656         journal_t *journal;
2657         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2658         dev_t journal_dev;
2659         int err = 0;
2660         int really_read_only;
2661
2662         if (journal_devnum &&
2663             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2664                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2665                         "numbers have changed\n");
2666                 journal_dev = new_decode_dev(journal_devnum);
2667         } else
2668                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2669
2670         really_read_only = bdev_read_only(sb->s_bdev);
2671
2672         /*
2673          * Are we loading a blank journal or performing recovery after a
2674          * crash?  For recovery, we need to check in advance whether we
2675          * can get read-write access to the device.
2676          */
2677
2678         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2679                 if (sb->s_flags & MS_RDONLY) {
2680                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2681                                         "required on readonly filesystem.\n");
2682                         if (really_read_only) {
2683                                 printk(KERN_ERR "EXT4-fs: write access "
2684                                         "unavailable, cannot proceed.\n");
2685                                 return -EROFS;
2686                         }
2687                         printk (KERN_INFO "EXT4-fs: write access will "
2688                                         "be enabled during recovery.\n");
2689                 }
2690         }
2691
2692         if (journal_inum && journal_dev) {
2693                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2694                        "and inode journals!\n");
2695                 return -EINVAL;
2696         }
2697
2698         if (journal_inum) {
2699                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2700                         return -EINVAL;
2701         } else {
2702                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2703                         return -EINVAL;
2704         }
2705
2706         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2707                 err = jbd2_journal_update_format(journal);
2708                 if (err)  {
2709                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2710                         jbd2_journal_destroy(journal);
2711                         return err;
2712                 }
2713         }
2714
2715         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2716                 err = jbd2_journal_wipe(journal, !really_read_only);
2717         if (!err)
2718                 err = jbd2_journal_load(journal);
2719
2720         if (err) {
2721                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2722                 jbd2_journal_destroy(journal);
2723                 return err;
2724         }
2725
2726         EXT4_SB(sb)->s_journal = journal;
2727         ext4_clear_journal_err(sb, es);
2728
2729         if (journal_devnum &&
2730             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2731                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2732                 sb->s_dirt = 1;
2733
2734                 /* Make sure we flush the recovery flag to disk. */
2735                 ext4_commit_super(sb, es, 1);
2736         }
2737
2738         return 0;
2739 }
2740
2741 static int ext4_create_journal(struct super_block * sb,
2742                                struct ext4_super_block * es,
2743                                unsigned int journal_inum)
2744 {
2745         journal_t *journal;
2746         int err;
2747
2748         if (sb->s_flags & MS_RDONLY) {
2749                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2750                                 "create journal.\n");
2751                 return -EROFS;
2752         }
2753
2754         journal = ext4_get_journal(sb, journal_inum);
2755         if (!journal)
2756                 return -EINVAL;
2757
2758         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2759                journal_inum);
2760
2761         err = jbd2_journal_create(journal);
2762         if (err) {
2763                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2764                 jbd2_journal_destroy(journal);
2765                 return -EIO;
2766         }
2767
2768         EXT4_SB(sb)->s_journal = journal;
2769
2770         ext4_update_dynamic_rev(sb);
2771         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2772         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2773
2774         es->s_journal_inum = cpu_to_le32(journal_inum);
2775         sb->s_dirt = 1;
2776
2777         /* Make sure we flush the recovery flag to disk. */
2778         ext4_commit_super(sb, es, 1);
2779
2780         return 0;
2781 }
2782
2783 static void ext4_commit_super (struct super_block * sb,
2784                                struct ext4_super_block * es,
2785                                int sync)
2786 {
2787         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2788
2789         if (!sbh)
2790                 return;
2791         es->s_wtime = cpu_to_le32(get_seconds());
2792         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2793         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2794         BUFFER_TRACE(sbh, "marking dirty");
2795         mark_buffer_dirty(sbh);
2796         if (sync)
2797                 sync_dirty_buffer(sbh);
2798 }
2799
2800
2801 /*
2802  * Have we just finished recovery?  If so, and if we are mounting (or
2803  * remounting) the filesystem readonly, then we will end up with a
2804  * consistent fs on disk.  Record that fact.
2805  */
2806 static void ext4_mark_recovery_complete(struct super_block * sb,
2807                                         struct ext4_super_block * es)
2808 {
2809         journal_t *journal = EXT4_SB(sb)->s_journal;
2810
2811         jbd2_journal_lock_updates(journal);
2812         jbd2_journal_flush(journal);
2813         lock_super(sb);
2814         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2815             sb->s_flags & MS_RDONLY) {
2816                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2817                 sb->s_dirt = 0;
2818                 ext4_commit_super(sb, es, 1);
2819         }
2820         unlock_super(sb);
2821         jbd2_journal_unlock_updates(journal);
2822 }
2823
2824 /*
2825  * If we are mounting (or read-write remounting) a filesystem whose journal
2826  * has recorded an error from a previous lifetime, move that error to the
2827  * main filesystem now.
2828  */
2829 static void ext4_clear_journal_err(struct super_block * sb,
2830                                    struct ext4_super_block * es)
2831 {
2832         journal_t *journal;
2833         int j_errno;
2834         const char *errstr;
2835
2836         journal = EXT4_SB(sb)->s_journal;
2837
2838         /*
2839          * Now check for any error status which may have been recorded in the
2840          * journal by a prior ext4_error() or ext4_abort()
2841          */
2842
2843         j_errno = jbd2_journal_errno(journal);
2844         if (j_errno) {
2845                 char nbuf[16];
2846
2847                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2848                 ext4_warning(sb, __func__, "Filesystem error recorded "
2849                              "from previous mount: %s", errstr);
2850                 ext4_warning(sb, __func__, "Marking fs in need of "
2851                              "filesystem check.");
2852
2853                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2854                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2855                 ext4_commit_super (sb, es, 1);
2856
2857                 jbd2_journal_clear_err(journal);
2858         }
2859 }
2860
2861 /*
2862  * Force the running and committing transactions to commit,
2863  * and wait on the commit.
2864  */
2865 int ext4_force_commit(struct super_block *sb)
2866 {
2867         journal_t *journal;
2868         int ret;
2869
2870         if (sb->s_flags & MS_RDONLY)
2871                 return 0;
2872
2873         journal = EXT4_SB(sb)->s_journal;
2874         sb->s_dirt = 0;
2875         ret = ext4_journal_force_commit(journal);
2876         return ret;
2877 }
2878
2879 /*
2880  * Ext4 always journals updates to the superblock itself, so we don't
2881  * have to propagate any other updates to the superblock on disk at this
2882  * point.  Just start an async writeback to get the buffers on their way
2883  * to the disk.
2884  *
2885  * This implicitly triggers the writebehind on sync().
2886  */
2887
2888 static void ext4_write_super (struct super_block * sb)
2889 {
2890         if (mutex_trylock(&sb->s_lock) != 0)
2891                 BUG();
2892         sb->s_dirt = 0;
2893 }
2894
2895 static int ext4_sync_fs(struct super_block *sb, int wait)
2896 {
2897         tid_t target;
2898
2899         sb->s_dirt = 0;
2900         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2901                 if (wait)
2902                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2903         }
2904         return 0;
2905 }
2906
2907 /*
2908  * LVM calls this function before a (read-only) snapshot is created.  This
2909  * gives us a chance to flush the journal completely and mark the fs clean.
2910  */
2911 static void ext4_write_super_lockfs(struct super_block *sb)
2912 {
2913         sb->s_dirt = 0;
2914
2915         if (!(sb->s_flags & MS_RDONLY)) {
2916                 journal_t *journal = EXT4_SB(sb)->s_journal;
2917
2918                 /* Now we set up the journal barrier. */
2919                 jbd2_journal_lock_updates(journal);
2920                 jbd2_journal_flush(journal);
2921
2922                 /* Journal blocked and flushed, clear needs_recovery flag. */
2923                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2924                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2925         }
2926 }
2927
2928 /*
2929  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2930  * flag here, even though the filesystem is not technically dirty yet.
2931  */
2932 static void ext4_unlockfs(struct super_block *sb)
2933 {
2934         if (!(sb->s_flags & MS_RDONLY)) {
2935                 lock_super(sb);
2936                 /* Reser the needs_recovery flag before the fs is unlocked. */
2937                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2938                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2939                 unlock_super(sb);
2940                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2941         }
2942 }
2943
2944 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2945 {
2946         struct ext4_super_block * es;
2947         struct ext4_sb_info *sbi = EXT4_SB(sb);
2948         ext4_fsblk_t n_blocks_count = 0;
2949         unsigned long old_sb_flags;
2950         struct ext4_mount_options old_opts;
2951         int err;
2952 #ifdef CONFIG_QUOTA
2953         int i;
2954 #endif
2955
2956         /* Store the original options */
2957         old_sb_flags = sb->s_flags;
2958         old_opts.s_mount_opt = sbi->s_mount_opt;
2959         old_opts.s_resuid = sbi->s_resuid;
2960         old_opts.s_resgid = sbi->s_resgid;
2961         old_opts.s_commit_interval = sbi->s_commit_interval;
2962 #ifdef CONFIG_QUOTA
2963         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2964         for (i = 0; i < MAXQUOTAS; i++)
2965                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2966 #endif
2967
2968         /*
2969          * Allow the "check" option to be passed as a remount option.
2970          */
2971         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2972                 err = -EINVAL;
2973                 goto restore_opts;
2974         }
2975
2976         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2977                 ext4_abort(sb, __func__, "Abort forced by user");
2978
2979         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2980                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2981
2982         es = sbi->s_es;
2983
2984         ext4_init_journal_params(sb, sbi->s_journal);
2985
2986         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2987                 n_blocks_count > ext4_blocks_count(es)) {
2988                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2989                         err = -EROFS;
2990                         goto restore_opts;
2991                 }
2992
2993                 if (*flags & MS_RDONLY) {
2994                         /*
2995                          * First of all, the unconditional stuff we have to do
2996                          * to disable replay of the journal when we next remount
2997                          */
2998                         sb->s_flags |= MS_RDONLY;
2999
3000                         /*
3001                          * OK, test if we are remounting a valid rw partition
3002                          * readonly, and if so set the rdonly flag and then
3003                          * mark the partition as valid again.
3004                          */
3005                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3006                             (sbi->s_mount_state & EXT4_VALID_FS))
3007                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3008
3009                         /*
3010                          * We have to unlock super so that we can wait for
3011                          * transactions.
3012                          */
3013                         unlock_super(sb);
3014                         ext4_mark_recovery_complete(sb, es);
3015                         lock_super(sb);
3016                 } else {
3017                         __le32 ret;
3018                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3019                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3020                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3021                                        "remount RDWR because of unsupported "
3022                                        "optional features (%x).\n",
3023                                        sb->s_id, le32_to_cpu(ret));
3024                                 err = -EROFS;
3025                                 goto restore_opts;
3026                         }
3027
3028                         /*
3029                          * If we have an unprocessed orphan list hanging
3030                          * around from a previously readonly bdev mount,
3031                          * require a full umount/remount for now.
3032                          */
3033                         if (es->s_last_orphan) {
3034                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3035                                        "remount RDWR because of unprocessed "
3036                                        "orphan inode list.  Please "
3037                                        "umount/remount instead.\n",
3038                                        sb->s_id);
3039                                 err = -EINVAL;
3040                                 goto restore_opts;
3041                         }
3042
3043                         /*
3044                          * Mounting a RDONLY partition read-write, so reread
3045                          * and store the current valid flag.  (It may have
3046                          * been changed by e2fsck since we originally mounted
3047                          * the partition.)
3048                          */
3049                         ext4_clear_journal_err(sb, es);
3050                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3051                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3052                                 goto restore_opts;
3053                         if (!ext4_setup_super (sb, es, 0))
3054                                 sb->s_flags &= ~MS_RDONLY;
3055                 }
3056         }
3057 #ifdef CONFIG_QUOTA
3058         /* Release old quota file names */
3059         for (i = 0; i < MAXQUOTAS; i++)
3060                 if (old_opts.s_qf_names[i] &&
3061                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3062                         kfree(old_opts.s_qf_names[i]);
3063 #endif
3064         return 0;
3065 restore_opts:
3066         sb->s_flags = old_sb_flags;
3067         sbi->s_mount_opt = old_opts.s_mount_opt;
3068         sbi->s_resuid = old_opts.s_resuid;
3069         sbi->s_resgid = old_opts.s_resgid;
3070         sbi->s_commit_interval = old_opts.s_commit_interval;
3071 #ifdef CONFIG_QUOTA
3072         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3073         for (i = 0; i < MAXQUOTAS; i++) {
3074                 if (sbi->s_qf_names[i] &&
3075                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3076                         kfree(sbi->s_qf_names[i]);
3077                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3078         }
3079 #endif
3080         return err;
3081 }
3082
3083 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
3084 {
3085         struct super_block *sb = dentry->d_sb;
3086         struct ext4_sb_info *sbi = EXT4_SB(sb);
3087         struct ext4_super_block *es = sbi->s_es;
3088         u64 fsid;
3089
3090         if (test_opt(sb, MINIX_DF)) {
3091                 sbi->s_overhead_last = 0;
3092         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3093                 ext4_group_t ngroups = sbi->s_groups_count, i;
3094                 ext4_fsblk_t overhead = 0;
3095                 smp_rmb();
3096
3097                 /*
3098                  * Compute the overhead (FS structures).  This is constant
3099                  * for a given filesystem unless the number of block groups
3100                  * changes so we cache the previous value until it does.
3101                  */
3102
3103                 /*
3104                  * All of the blocks before first_data_block are
3105                  * overhead
3106                  */
3107                 overhead = le32_to_cpu(es->s_first_data_block);
3108
3109                 /*
3110                  * Add the overhead attributed to the superblock and
3111                  * block group descriptors.  If the sparse superblocks
3112                  * feature is turned on, then not all groups have this.
3113                  */
3114                 for (i = 0; i < ngroups; i++) {
3115                         overhead += ext4_bg_has_super(sb, i) +
3116                                 ext4_bg_num_gdb(sb, i);
3117                         cond_resched();
3118                 }
3119
3120                 /*
3121                  * Every block group has an inode bitmap, a block
3122                  * bitmap, and an inode table.
3123                  */
3124                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3125                 sbi->s_overhead_last = overhead;
3126                 smp_wmb();
3127                 sbi->s_blocks_last = ext4_blocks_count(es);
3128         }
3129
3130         buf->f_type = EXT4_SUPER_MAGIC;
3131         buf->f_bsize = sb->s_blocksize;
3132         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3133         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
3134         ext4_free_blocks_count_set(es, buf->f_bfree);
3135         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3136         if (buf->f_bfree < ext4_r_blocks_count(es))
3137                 buf->f_bavail = 0;
3138         buf->f_files = le32_to_cpu(es->s_inodes_count);
3139         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3140         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3141         buf->f_namelen = EXT4_NAME_LEN;
3142         fsid = le64_to_cpup((void *)es->s_uuid) ^
3143                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3144         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3145         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3146         return 0;
3147 }
3148
3149 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3150  * is locked for write. Otherwise the are possible deadlocks:
3151  * Process 1                         Process 2
3152  * ext4_create()                     quota_sync()
3153  *   jbd2_journal_start()                   write_dquot()
3154  *   DQUOT_INIT()                        down(dqio_mutex)
3155  *     down(dqio_mutex)                    jbd2_journal_start()
3156  *
3157  */
3158
3159 #ifdef CONFIG_QUOTA
3160
3161 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3162 {
3163         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3164 }
3165
3166 static int ext4_dquot_initialize(struct inode *inode, int type)
3167 {
3168         handle_t *handle;
3169         int ret, err;
3170
3171         /* We may create quota structure so we need to reserve enough blocks */
3172         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3173         if (IS_ERR(handle))
3174                 return PTR_ERR(handle);
3175         ret = dquot_initialize(inode, type);
3176         err = ext4_journal_stop(handle);
3177         if (!ret)
3178                 ret = err;
3179         return ret;
3180 }
3181
3182 static int ext4_dquot_drop(struct inode *inode)
3183 {
3184         handle_t *handle;
3185         int ret, err;
3186
3187         /* We may delete quota structure so we need to reserve enough blocks */
3188         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3189         if (IS_ERR(handle)) {
3190                 /*
3191                  * We call dquot_drop() anyway to at least release references
3192                  * to quota structures so that umount does not hang.
3193                  */
3194                 dquot_drop(inode);
3195                 return PTR_ERR(handle);
3196         }
3197         ret = dquot_drop(inode);
3198         err = ext4_journal_stop(handle);
3199         if (!ret)
3200                 ret = err;
3201         return ret;
3202 }
3203
3204 static int ext4_write_dquot(struct dquot *dquot)
3205 {
3206         int ret, err;
3207         handle_t *handle;
3208         struct inode *inode;
3209
3210         inode = dquot_to_inode(dquot);
3211         handle = ext4_journal_start(inode,
3212                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3213         if (IS_ERR(handle))
3214                 return PTR_ERR(handle);
3215         ret = dquot_commit(dquot);
3216         err = ext4_journal_stop(handle);
3217         if (!ret)
3218                 ret = err;
3219         return ret;
3220 }
3221
3222 static int ext4_acquire_dquot(struct dquot *dquot)
3223 {
3224         int ret, err;
3225         handle_t *handle;
3226
3227         handle = ext4_journal_start(dquot_to_inode(dquot),
3228                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3229         if (IS_ERR(handle))
3230                 return PTR_ERR(handle);
3231         ret = dquot_acquire(dquot);
3232         err = ext4_journal_stop(handle);
3233         if (!ret)
3234                 ret = err;
3235         return ret;
3236 }
3237
3238 static int ext4_release_dquot(struct dquot *dquot)
3239 {
3240         int ret, err;
3241         handle_t *handle;
3242
3243         handle = ext4_journal_start(dquot_to_inode(dquot),
3244                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3245         if (IS_ERR(handle)) {
3246                 /* Release dquot anyway to avoid endless cycle in dqput() */
3247                 dquot_release(dquot);
3248                 return PTR_ERR(handle);
3249         }
3250         ret = dquot_release(dquot);
3251         err = ext4_journal_stop(handle);
3252         if (!ret)
3253                 ret = err;
3254         return ret;
3255 }
3256
3257 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3258 {
3259         /* Are we journaling quotas? */
3260         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3261             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3262                 dquot_mark_dquot_dirty(dquot);
3263                 return ext4_write_dquot(dquot);
3264         } else {
3265                 return dquot_mark_dquot_dirty(dquot);
3266         }
3267 }
3268
3269 static int ext4_write_info(struct super_block *sb, int type)
3270 {
3271         int ret, err;
3272         handle_t *handle;
3273
3274         /* Data block + inode block */
3275         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3276         if (IS_ERR(handle))
3277                 return PTR_ERR(handle);
3278         ret = dquot_commit_info(sb, type);
3279         err = ext4_journal_stop(handle);
3280         if (!ret)
3281                 ret = err;
3282         return ret;
3283 }
3284
3285 /*
3286  * Turn on quotas during mount time - we need to find
3287  * the quota file and such...
3288  */
3289 static int ext4_quota_on_mount(struct super_block *sb, int type)
3290 {
3291         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3292                         EXT4_SB(sb)->s_jquota_fmt, type);
3293 }
3294
3295 /*
3296  * Standard function to be called on quota_on
3297  */
3298 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3299                          char *path, int remount)
3300 {
3301         int err;
3302         struct nameidata nd;
3303
3304         if (!test_opt(sb, QUOTA))
3305                 return -EINVAL;
3306         /* When remounting, no checks are needed and in fact, path is NULL */
3307         if (remount)
3308                 return vfs_quota_on(sb, type, format_id, path, remount);
3309
3310         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3311         if (err)
3312                 return err;
3313
3314         /* Quotafile not on the same filesystem? */
3315         if (nd.path.mnt->mnt_sb != sb) {
3316                 path_put(&nd.path);
3317                 return -EXDEV;
3318         }
3319         /* Journaling quota? */
3320         if (EXT4_SB(sb)->s_qf_names[type]) {
3321                 /* Quotafile not of fs root? */
3322                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3323                         printk(KERN_WARNING
3324                                 "EXT4-fs: Quota file not on filesystem root. "
3325                                 "Journaled quota will not work.\n");
3326         }
3327
3328         /*
3329          * When we journal data on quota file, we have to flush journal to see
3330          * all updates to the file when we bypass pagecache...
3331          */
3332         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3333                 /*
3334                  * We don't need to lock updates but journal_flush() could
3335                  * otherwise be livelocked...
3336                  */
3337                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3338                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3339                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3340         }
3341
3342         path_put(&nd.path);
3343         return vfs_quota_on(sb, type, format_id, path, remount);
3344 }
3345
3346 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3347  * acquiring the locks... As quota files are never truncated and quota code
3348  * itself serializes the operations (and noone else should touch the files)
3349  * we don't have to be afraid of races */
3350 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3351                                size_t len, loff_t off)
3352 {
3353         struct inode *inode = sb_dqopt(sb)->files[type];
3354         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3355         int err = 0;
3356         int offset = off & (sb->s_blocksize - 1);
3357         int tocopy;
3358         size_t toread;
3359         struct buffer_head *bh;
3360         loff_t i_size = i_size_read(inode);
3361
3362         if (off > i_size)
3363                 return 0;
3364         if (off+len > i_size)
3365                 len = i_size-off;
3366         toread = len;
3367         while (toread > 0) {
3368                 tocopy = sb->s_blocksize - offset < toread ?
3369                                 sb->s_blocksize - offset : toread;
3370                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3371                 if (err)
3372                         return err;
3373                 if (!bh)        /* A hole? */
3374                         memset(data, 0, tocopy);
3375                 else
3376                         memcpy(data, bh->b_data+offset, tocopy);
3377                 brelse(bh);
3378                 offset = 0;
3379                 toread -= tocopy;
3380                 data += tocopy;
3381                 blk++;
3382         }
3383         return len;
3384 }
3385
3386 /* Write to quotafile (we know the transaction is already started and has
3387  * enough credits) */
3388 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3389                                 const char *data, size_t len, loff_t off)
3390 {
3391         struct inode *inode = sb_dqopt(sb)->files[type];
3392         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3393         int err = 0;
3394         int offset = off & (sb->s_blocksize - 1);
3395         int tocopy;
3396         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3397         size_t towrite = len;
3398         struct buffer_head *bh;
3399         handle_t *handle = journal_current_handle();
3400
3401         if (!handle) {
3402                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3403                         " cancelled because transaction is not started.\n",
3404                         (unsigned long long)off, (unsigned long long)len);
3405                 return -EIO;
3406         }
3407         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3408         while (towrite > 0) {
3409                 tocopy = sb->s_blocksize - offset < towrite ?
3410                                 sb->s_blocksize - offset : towrite;
3411                 bh = ext4_bread(handle, inode, blk, 1, &err);
3412                 if (!bh)
3413                         goto out;
3414                 if (journal_quota) {
3415                         err = ext4_journal_get_write_access(handle, bh);
3416                         if (err) {
3417                                 brelse(bh);
3418                                 goto out;
3419                         }
3420                 }
3421                 lock_buffer(bh);
3422                 memcpy(bh->b_data+offset, data, tocopy);
3423                 flush_dcache_page(bh->b_page);
3424                 unlock_buffer(bh);
3425                 if (journal_quota)
3426                         err = ext4_journal_dirty_metadata(handle, bh);
3427                 else {
3428                         /* Always do at least ordered writes for quotas */
3429                         err = ext4_jbd2_file_inode(handle, inode);
3430                         mark_buffer_dirty(bh);
3431                 }
3432                 brelse(bh);
3433                 if (err)
3434                         goto out;
3435                 offset = 0;
3436                 towrite -= tocopy;
3437                 data += tocopy;
3438                 blk++;
3439         }
3440 out:
3441         if (len == towrite) {
3442                 mutex_unlock(&inode->i_mutex);
3443                 return err;
3444         }
3445         if (inode->i_size < off+len-towrite) {
3446                 i_size_write(inode, off+len-towrite);
3447                 EXT4_I(inode)->i_disksize = inode->i_size;
3448         }
3449         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3450         ext4_mark_inode_dirty(handle, inode);
3451         mutex_unlock(&inode->i_mutex);
3452         return len - towrite;
3453 }
3454
3455 #endif
3456
3457 static int ext4_get_sb(struct file_system_type *fs_type,
3458         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3459 {
3460         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3461 }
3462
3463 static struct file_system_type ext4dev_fs_type = {
3464         .owner          = THIS_MODULE,
3465         .name           = "ext4dev",
3466         .get_sb         = ext4_get_sb,
3467         .kill_sb        = kill_block_super,
3468         .fs_flags       = FS_REQUIRES_DEV,
3469 };
3470
3471 static int __init init_ext4_fs(void)
3472 {
3473         int err;
3474
3475         err = init_ext4_mballoc();
3476         if (err)
3477                 return err;
3478
3479         err = init_ext4_xattr();
3480         if (err)
3481                 goto out2;
3482         err = init_inodecache();
3483         if (err)
3484                 goto out1;
3485         err = register_filesystem(&ext4dev_fs_type);
3486         if (err)
3487                 goto out;
3488         return 0;
3489 out:
3490         destroy_inodecache();
3491 out1:
3492         exit_ext4_xattr();
3493 out2:
3494         exit_ext4_mballoc();
3495         return err;
3496 }
3497
3498 static void __exit exit_ext4_fs(void)
3499 {
3500         unregister_filesystem(&ext4dev_fs_type);
3501         destroy_inodecache();
3502         exit_ext4_xattr();
3503         exit_ext4_mballoc();
3504 }
3505
3506 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3507 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3508 MODULE_LICENSE("GPL");
3509 module_init(init_ext4_fs)
3510 module_exit(exit_ext4_fs)