fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "xattr.h"
  45 #include "acl.h"
  46 #include "namei.h"
  47 #include "group.h"
  48
  49 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  50                              unsigned long journal_devnum);
  51 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  52                                unsigned int);
  53 static void ext4_commit_super (struct super_block * sb,
  54                                struct ext4_super_block * es,
  55                                int sync);
  56 static void ext4_mark_recovery_complete(struct super_block * sb,
  57                                         struct ext4_super_block * es);
  58 static void ext4_clear_journal_err(struct super_block * sb,
  59                                    struct ext4_super_block * es);
  60 static int ext4_sync_fs(struct super_block *sb, int wait);
  61 static const char *ext4_decode_error(struct super_block * sb, int errno,
  62                                      char nbuf[16]);
  63 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  64 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  65 static void ext4_unlockfs(struct super_block *sb);
  66 static void ext4_write_super (struct super_block * sb);
  67 static void ext4_write_super_lockfs(struct super_block *sb);
  68
  69
  70 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  71                                struct ext4_group_desc *bg)
  72 {
  73         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  74                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  75                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  76 }
  77
  78 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  79                                struct ext4_group_desc *bg)
  80 {
  81         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  82                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  83                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  84 }
  85
  86 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  87                               struct ext4_group_desc *bg)
  88 {
  89         return le32_to_cpu(bg->bg_inode_table_lo) |
  90                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  91                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  92 }
  93
  94 void ext4_block_bitmap_set(struct super_block *sb,
  95                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  96 {
  97         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  98         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  99                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 100 }
 101
 102 void ext4_inode_bitmap_set(struct super_block *sb,
 103                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 104 {
 105         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 106         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 107                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 108 }
 109
 110 void ext4_inode_table_set(struct super_block *sb,
 111                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 112 {
 113         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 114         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 115                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 116 }
 117
 118 /*
 119  * Wrappers for jbd2_journal_start/end.
 120  *
 121  * The only special thing we need to do here is to make sure that all
 122  * journal_end calls result in the superblock being marked dirty, so
 123  * that sync() will call the filesystem's write_super callback if
 124  * appropriate.
 125  */
 126 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 127 {
 128         journal_t *journal;
 129
 130         if (sb->s_flags & MS_RDONLY)
 131                 return ERR_PTR(-EROFS);
 132
 133         /* Special case here: if the journal has aborted behind our
 134          * backs (eg. EIO in the commit thread), then we still need to
 135          * take the FS itself readonly cleanly. */
 136         journal = EXT4_SB(sb)->s_journal;
 137         if (is_journal_aborted(journal)) {
 138                 ext4_abort(sb, __FUNCTION__,
 139                            "Detected aborted journal");
 140                 return ERR_PTR(-EROFS);
 141         }
 142
 143         return jbd2_journal_start(journal, nblocks);
 144 }
 145
 146 /*
 147  * The only special thing we need to do here is to make sure that all
 148  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 149  * that sync() will call the filesystem's write_super callback if
 150  * appropriate.
 151  */
 152 int __ext4_journal_stop(const char *where, handle_t *handle)
 153 {
 154         struct super_block *sb;
 155         int err;
 156         int rc;
 157
 158         sb = handle->h_transaction->t_journal->j_private;
 159         err = handle->h_err;
 160         rc = jbd2_journal_stop(handle);
 161
 162         if (!err)
 163                 err = rc;
 164         if (err)
 165                 __ext4_std_error(sb, where, err);
 166         return err;
 167 }
 168
 169 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 170                 struct buffer_head *bh, handle_t *handle, int err)
 171 {
 172         char nbuf[16];
 173         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 174
 175         if (bh)
 176                 BUFFER_TRACE(bh, "abort");
 177
 178         if (!handle->h_err)
 179                 handle->h_err = err;
 180
 181         if (is_handle_aborted(handle))
 182                 return;
 183
 184         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 185                caller, errstr, err_fn);
 186
 187         jbd2_journal_abort_handle(handle);
 188 }
 189
 190 /* Deal with the reporting of failure conditions on a filesystem such as
 191  * inconsistencies detected or read IO failures.
 192  *
 193  * On ext2, we can store the error state of the filesystem in the
 194  * superblock.  That is not possible on ext4, because we may have other
 195  * write ordering constraints on the superblock which prevent us from
 196  * writing it out straight away; and given that the journal is about to
 197  * be aborted, we can't rely on the current, or future, transactions to
 198  * write out the superblock safely.
 199  *
 200  * We'll just use the jbd2_journal_abort() error code to record an error in
 201  * the journal instead.  On recovery, the journal will compain about
 202  * that error until we've noted it down and cleared it.
 203  */
 204
 205 static void ext4_handle_error(struct super_block *sb)
 206 {
 207         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 208
 209         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 210         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 211
 212         if (sb->s_flags & MS_RDONLY)
 213                 return;
 214
 215         if (!test_opt (sb, ERRORS_CONT)) {
 216                 journal_t *journal = EXT4_SB(sb)->s_journal;
 217
 218                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 219                 if (journal)
 220                         jbd2_journal_abort(journal, -EIO);
 221         }
 222         if (test_opt (sb, ERRORS_RO)) {
 223                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 224                 sb->s_flags |= MS_RDONLY;
 225         }
 226         ext4_commit_super(sb, es, 1);
 227         if (test_opt(sb, ERRORS_PANIC))
 228                 panic("EXT4-fs (device %s): panic forced after error\n",
 229                         sb->s_id);
 230 }
 231
 232 void ext4_error (struct super_block * sb, const char * function,
 233                  const char * fmt, ...)
 234 {
 235         va_list args;
 236
 237         va_start(args, fmt);
 238         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 239         vprintk(fmt, args);
 240         printk("\n");
 241         va_end(args);
 242
 243         ext4_handle_error(sb);
 244 }
 245
 246 static const char *ext4_decode_error(struct super_block * sb, int errno,
 247                                      char nbuf[16])
 248 {
 249         char *errstr = NULL;
 250
 251         switch (errno) {
 252         case -EIO:
 253                 errstr = "IO failure";
 254                 break;
 255         case -ENOMEM:
 256                 errstr = "Out of memory";
 257                 break;
 258         case -EROFS:
 259                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 260                         errstr = "Journal has aborted";
 261                 else
 262                         errstr = "Readonly filesystem";
 263                 break;
 264         default:
 265                 /* If the caller passed in an extra buffer for unknown
 266                  * errors, textualise them now.  Else we just return
 267                  * NULL. */
 268                 if (nbuf) {
 269                         /* Check for truncated error codes... */
 270                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 271                                 errstr = nbuf;
 272                 }
 273                 break;
 274         }
 275
 276         return errstr;
 277 }
 278
 279 /* __ext4_std_error decodes expected errors from journaling functions
 280  * automatically and invokes the appropriate error response.  */
 281
 282 void __ext4_std_error (struct super_block * sb, const char * function,
 283                        int errno)
 284 {
 285         char nbuf[16];
 286         const char *errstr;
 287
 288         /* Special case: if the error is EROFS, and we're not already
 289          * inside a transaction, then there's really no point in logging
 290          * an error. */
 291         if (errno == -EROFS && journal_current_handle() == NULL &&
 292             (sb->s_flags & MS_RDONLY))
 293                 return;
 294
 295         errstr = ext4_decode_error(sb, errno, nbuf);
 296         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 297                 sb->s_id, function, errstr);
 298
 299         ext4_handle_error(sb);
 300 }
 301
 302 /*
 303  * ext4_abort is a much stronger failure handler than ext4_error.  The
 304  * abort function may be used to deal with unrecoverable failures such
 305  * as journal IO errors or ENOMEM at a critical moment in log management.
 306  *
 307  * We unconditionally force the filesystem into an ABORT|READONLY state,
 308  * unless the error response on the fs has been set to panic in which
 309  * case we take the easy way out and panic immediately.
 310  */
 311
 312 void ext4_abort (struct super_block * sb, const char * function,
 313                  const char * fmt, ...)
 314 {
 315         va_list args;
 316
 317         printk (KERN_CRIT "ext4_abort called.\n");
 318
 319         va_start(args, fmt);
 320         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 321         vprintk(fmt, args);
 322         printk("\n");
 323         va_end(args);
 324
 325         if (test_opt(sb, ERRORS_PANIC))
 326                 panic("EXT4-fs panic from previous error\n");
 327
 328         if (sb->s_flags & MS_RDONLY)
 329                 return;
 330
 331         printk(KERN_CRIT "Remounting filesystem read-only\n");
 332         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 333         sb->s_flags |= MS_RDONLY;
 334         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 335         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 336 }
 337
 338 void ext4_warning (struct super_block * sb, const char * function,
 339                    const char * fmt, ...)
 340 {
 341         va_list args;
 342
 343         va_start(args, fmt);
 344         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 345                sb->s_id, function);
 346         vprintk(fmt, args);
 347         printk("\n");
 348         va_end(args);
 349 }
 350
 351 void ext4_update_dynamic_rev(struct super_block *sb)
 352 {
 353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 354
 355         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 356                 return;
 357
 358         ext4_warning(sb, __FUNCTION__,
 359                      "updating to rev %d because of new feature flag, "
 360                      "running e2fsck is recommended",
 361                      EXT4_DYNAMIC_REV);
 362
 363         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 364         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 365         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 366         /* leave es->s_feature_*compat flags alone */
 367         /* es->s_uuid will be set by e2fsck if empty */
 368
 369         /*
 370          * The rest of the superblock fields should be zero, and if not it
 371          * means they are likely already in use, so leave them alone.  We
 372          * can leave it up to e2fsck to clean up any inconsistencies there.
 373          */
 374 }
 375
 376 int ext4_update_compat_feature(handle_t *handle,
 377                                         struct super_block *sb, __u32 compat)
 378 {
 379         int err = 0;
 380         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 381                 err = ext4_journal_get_write_access(handle,
 382                                 EXT4_SB(sb)->s_sbh);
 383                 if (err)
 384                         return err;
 385                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 386                 sb->s_dirt = 1;
 387                 handle->h_sync = 1;
 388                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 389                                         "call ext4_journal_dirty_met adata");
 390                 err = ext4_journal_dirty_metadata(handle,
 391                                 EXT4_SB(sb)->s_sbh);
 392         }
 393         return err;
 394 }
 395
 396 int ext4_update_rocompat_feature(handle_t *handle,
 397                                         struct super_block *sb, __u32 rocompat)
 398 {
 399         int err = 0;
 400         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 401                 err = ext4_journal_get_write_access(handle,
 402                                 EXT4_SB(sb)->s_sbh);
 403                 if (err)
 404                         return err;
 405                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 406                 sb->s_dirt = 1;
 407                 handle->h_sync = 1;
 408                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 409                                         "call ext4_journal_dirty_met adata");
 410                 err = ext4_journal_dirty_metadata(handle,
 411                                 EXT4_SB(sb)->s_sbh);
 412         }
 413         return err;
 414 }
 415
 416 int ext4_update_incompat_feature(handle_t *handle,
 417                                         struct super_block *sb, __u32 incompat)
 418 {
 419         int err = 0;
 420         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 421                 err = ext4_journal_get_write_access(handle,
 422                                 EXT4_SB(sb)->s_sbh);
 423                 if (err)
 424                         return err;
 425                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 426                 sb->s_dirt = 1;
 427                 handle->h_sync = 1;
 428                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 429                                         "call ext4_journal_dirty_met adata");
 430                 err = ext4_journal_dirty_metadata(handle,
 431                                 EXT4_SB(sb)->s_sbh);
 432         }
 433         return err;
 434 }
 435
 436 /*
 437  * Open the external journal device
 438  */
 439 static struct block_device *ext4_blkdev_get(dev_t dev)
 440 {
 441         struct block_device *bdev;
 442         char b[BDEVNAME_SIZE];
 443
 444         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 445         if (IS_ERR(bdev))
 446                 goto fail;
 447         return bdev;
 448
 449 fail:
 450         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 451                         __bdevname(dev, b), PTR_ERR(bdev));
 452         return NULL;
 453 }
 454
 455 /*
 456  * Release the journal device
 457  */
 458 static int ext4_blkdev_put(struct block_device *bdev)
 459 {
 460         bd_release(bdev);
 461         return blkdev_put(bdev);
 462 }
 463
 464 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 465 {
 466         struct block_device *bdev;
 467         int ret = -ENODEV;
 468
 469         bdev = sbi->journal_bdev;
 470         if (bdev) {
 471                 ret = ext4_blkdev_put(bdev);
 472                 sbi->journal_bdev = NULL;
 473         }
 474         return ret;
 475 }
 476
 477 static inline struct inode *orphan_list_entry(struct list_head *l)
 478 {
 479         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 480 }
 481
 482 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 483 {
 484         struct list_head *l;
 485
 486         printk(KERN_ERR "sb orphan head is %d\n",
 487                le32_to_cpu(sbi->s_es->s_last_orphan));
 488
 489         printk(KERN_ERR "sb_info orphan list:\n");
 490         list_for_each(l, &sbi->s_orphan) {
 491                 struct inode *inode = orphan_list_entry(l);
 492                 printk(KERN_ERR "  "
 493                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 494                        inode->i_sb->s_id, inode->i_ino, inode,
 495                        inode->i_mode, inode->i_nlink,
 496                        NEXT_ORPHAN(inode));
 497         }
 498 }
 499
 500 static void ext4_put_super (struct super_block * sb)
 501 {
 502         struct ext4_sb_info *sbi = EXT4_SB(sb);
 503         struct ext4_super_block *es = sbi->s_es;
 504         int i;
 505
 506         ext4_ext_release(sb);
 507         ext4_xattr_put_super(sb);
 508         jbd2_journal_destroy(sbi->s_journal);
 509         if (!(sb->s_flags & MS_RDONLY)) {
 510                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 511                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 512                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 513                 mark_buffer_dirty(sbi->s_sbh);
 514                 ext4_commit_super(sb, es, 1);
 515         }
 516
 517         for (i = 0; i < sbi->s_gdb_count; i++)
 518                 brelse(sbi->s_group_desc[i]);
 519         kfree(sbi->s_group_desc);
 520         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 521         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 522         percpu_counter_destroy(&sbi->s_dirs_counter);
 523         brelse(sbi->s_sbh);
 524 #ifdef CONFIG_QUOTA
 525         for (i = 0; i < MAXQUOTAS; i++)
 526                 kfree(sbi->s_qf_names[i]);
 527 #endif
 528
 529         /* Debugging code just in case the in-memory inode orphan list
 530          * isn't empty.  The on-disk one can be non-empty if we've
 531          * detected an error and taken the fs readonly, but the
 532          * in-memory list had better be clean by this point. */
 533         if (!list_empty(&sbi->s_orphan))
 534                 dump_orphan_list(sb, sbi);
 535         J_ASSERT(list_empty(&sbi->s_orphan));
 536
 537         invalidate_bdev(sb->s_bdev);
 538         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 539                 /*
 540                  * Invalidate the journal device's buffers.  We don't want them
 541                  * floating about in memory - the physical journal device may
 542                  * hotswapped, and it breaks the `ro-after' testing code.
 543                  */
 544                 sync_blockdev(sbi->journal_bdev);
 545                 invalidate_bdev(sbi->journal_bdev);
 546                 ext4_blkdev_remove(sbi);
 547         }
 548         sb->s_fs_info = NULL;
 549         kfree(sbi);
 550         return;
 551 }
 552
 553 static struct kmem_cache *ext4_inode_cachep;
 554
 555 /*
 556  * Called inside transaction, so use GFP_NOFS
 557  */
 558 static struct inode *ext4_alloc_inode(struct super_block *sb)
 559 {
 560         struct ext4_inode_info *ei;
 561
 562         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 563         if (!ei)
 564                 return NULL;
 565 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 566         ei->i_acl = EXT4_ACL_NOT_CACHED;
 567         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 568 #endif
 569         ei->i_block_alloc_info = NULL;
 570         ei->vfs_inode.i_version = 1;
 571         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 572         return &ei->vfs_inode;
 573 }
 574
 575 static void ext4_destroy_inode(struct inode *inode)
 576 {
 577         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 578                 printk("EXT4 Inode %p: orphan list check failed!\n",
 579                         EXT4_I(inode));
 580                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 581                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 582                                 true);
 583                 dump_stack();
 584         }
 585         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 586 }
 587
 588 static void init_once(struct kmem_cache *cachep, void *foo)
 589 {
 590         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 591
 592         INIT_LIST_HEAD(&ei->i_orphan);
 593 #ifdef CONFIG_EXT4DEV_FS_XATTR
 594         init_rwsem(&ei->xattr_sem);
 595 #endif
 596         init_rwsem(&ei->i_data_sem);
 597         inode_init_once(&ei->vfs_inode);
 598 }
 599
 600 static int init_inodecache(void)
 601 {
 602         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 603                                              sizeof(struct ext4_inode_info),
 604                                              0, (SLAB_RECLAIM_ACCOUNT|
 605                                                 SLAB_MEM_SPREAD),
 606                                              init_once);
 607         if (ext4_inode_cachep == NULL)
 608                 return -ENOMEM;
 609         return 0;
 610 }
 611
 612 static void destroy_inodecache(void)
 613 {
 614         kmem_cache_destroy(ext4_inode_cachep);
 615 }
 616
 617 static void ext4_clear_inode(struct inode *inode)
 618 {
 619         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 620 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 621         if (EXT4_I(inode)->i_acl &&
 622                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 623                 posix_acl_release(EXT4_I(inode)->i_acl);
 624                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 625         }
 626         if (EXT4_I(inode)->i_default_acl &&
 627                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 628                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 629                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 630         }
 631 #endif
 632         ext4_discard_reservation(inode);
 633         EXT4_I(inode)->i_block_alloc_info = NULL;
 634         if (unlikely(rsv))
 635                 kfree(rsv);
 636 }
 637
 638 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 639 {
 640 #if defined(CONFIG_QUOTA)
 641         struct ext4_sb_info *sbi = EXT4_SB(sb);
 642
 643         if (sbi->s_jquota_fmt)
 644                 seq_printf(seq, ",jqfmt=%s",
 645                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 646
 647         if (sbi->s_qf_names[USRQUOTA])
 648                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 649
 650         if (sbi->s_qf_names[GRPQUOTA])
 651                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 652
 653         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 654                 seq_puts(seq, ",usrquota");
 655
 656         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 657                 seq_puts(seq, ",grpquota");
 658 #endif
 659 }
 660
 661 /*
 662  * Show an option if
 663  *  - it's set to a non-default value OR
 664  *  - if the per-sb default is different from the global default
 665  */
 666 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 667 {
 668         struct super_block *sb = vfs->mnt_sb;
 669         struct ext4_sb_info *sbi = EXT4_SB(sb);
 670         struct ext4_super_block *es = sbi->s_es;
 671         unsigned long def_mount_opts;
 672
 673         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 674
 675         if (sbi->s_sb_block != 1)
 676                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 677         if (test_opt(sb, MINIX_DF))
 678                 seq_puts(seq, ",minixdf");
 679         if (test_opt(sb, GRPID))
 680                 seq_puts(seq, ",grpid");
 681         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 682                 seq_puts(seq, ",nogrpid");
 683         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 684             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 685                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 686         }
 687         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 688             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 689                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 690         }
 691         if (test_opt(sb, ERRORS_RO)) {
 692                 int def_errors = le16_to_cpu(es->s_errors);
 693
 694                 if (def_errors == EXT4_ERRORS_PANIC ||
 695                     def_errors == EXT4_ERRORS_CONTINUE) {
 696                         seq_puts(seq, ",errors=remount-ro");
 697                 }
 698         }
 699         if (test_opt(sb, ERRORS_CONT))
 700                 seq_puts(seq, ",errors=continue");
 701         if (test_opt(sb, ERRORS_PANIC))
 702                 seq_puts(seq, ",errors=panic");
 703         if (test_opt(sb, NO_UID32))
 704                 seq_puts(seq, ",nouid32");
 705         if (test_opt(sb, DEBUG))
 706                 seq_puts(seq, ",debug");
 707         if (test_opt(sb, OLDALLOC))
 708                 seq_puts(seq, ",oldalloc");
 709 #ifdef CONFIG_EXT4DEV_FS_XATTR
 710         if (test_opt(sb, XATTR_USER))
 711                 seq_puts(seq, ",user_xattr");
 712         if (!test_opt(sb, XATTR_USER) &&
 713             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 714                 seq_puts(seq, ",nouser_xattr");
 715         }
 716 #endif
 717 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 718         if (test_opt(sb, POSIX_ACL))
 719                 seq_puts(seq, ",acl");
 720         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 721                 seq_puts(seq, ",noacl");
 722 #endif
 723         if (!test_opt(sb, RESERVATION))
 724                 seq_puts(seq, ",noreservation");
 725         if (sbi->s_commit_interval) {
 726                 seq_printf(seq, ",commit=%u",
 727                            (unsigned) (sbi->s_commit_interval / HZ));
 728         }
 729         if (test_opt(sb, BARRIER))
 730                 seq_puts(seq, ",barrier=1");
 731         if (test_opt(sb, NOBH))
 732                 seq_puts(seq, ",nobh");
 733         if (!test_opt(sb, EXTENTS))
 734                 seq_puts(seq, ",noextents");
 735
 736         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 737                 seq_puts(seq, ",data=journal");
 738         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 739                 seq_puts(seq, ",data=ordered");
 740         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 741                 seq_puts(seq, ",data=writeback");
 742
 743         ext4_show_quota_options(seq, sb);
 744
 745         return 0;
 746 }
 747
 748
 749 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 750                 u64 ino, u32 generation)
 751 {
 752         struct inode *inode;
 753
 754         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 755                 return ERR_PTR(-ESTALE);
 756         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 757                 return ERR_PTR(-ESTALE);
 758
 759         /* iget isn't really right if the inode is currently unallocated!!
 760          *
 761          * ext4_read_inode will return a bad_inode if the inode had been
 762          * deleted, so we should be safe.
 763          *
 764          * Currently we don't know the generation for parent directory, so
 765          * a generation of 0 means "accept any"
 766          */
 767         inode = iget(sb, ino);
 768         if (inode == NULL)
 769                 return ERR_PTR(-ENOMEM);
 770         if (is_bad_inode(inode) ||
 771             (generation && inode->i_generation != generation)) {
 772                 iput(inode);
 773                 return ERR_PTR(-ESTALE);
 774         }
 775
 776         return inode;
 777 }
 778
 779 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 780                 int fh_len, int fh_type)
 781 {
 782         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 783                                     ext4_nfs_get_inode);
 784 }
 785
 786 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 787                 int fh_len, int fh_type)
 788 {
 789         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 790                                     ext4_nfs_get_inode);
 791 }
 792
 793 #ifdef CONFIG_QUOTA
 794 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 795 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 796
 797 static int ext4_dquot_initialize(struct inode *inode, int type);
 798 static int ext4_dquot_drop(struct inode *inode);
 799 static int ext4_write_dquot(struct dquot *dquot);
 800 static int ext4_acquire_dquot(struct dquot *dquot);
 801 static int ext4_release_dquot(struct dquot *dquot);
 802 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 803 static int ext4_write_info(struct super_block *sb, int type);
 804 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 805 static int ext4_quota_on_mount(struct super_block *sb, int type);
 806 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 807                                size_t len, loff_t off);
 808 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 809                                 const char *data, size_t len, loff_t off);
 810
 811 static struct dquot_operations ext4_quota_operations = {
 812         .initialize     = ext4_dquot_initialize,
 813         .drop           = ext4_dquot_drop,
 814         .alloc_space    = dquot_alloc_space,
 815         .alloc_inode    = dquot_alloc_inode,
 816         .free_space     = dquot_free_space,
 817         .free_inode     = dquot_free_inode,
 818         .transfer       = dquot_transfer,
 819         .write_dquot    = ext4_write_dquot,
 820         .acquire_dquot  = ext4_acquire_dquot,
 821         .release_dquot  = ext4_release_dquot,
 822         .mark_dirty     = ext4_mark_dquot_dirty,
 823         .write_info     = ext4_write_info
 824 };
 825
 826 static struct quotactl_ops ext4_qctl_operations = {
 827         .quota_on       = ext4_quota_on,
 828         .quota_off      = vfs_quota_off,
 829         .quota_sync     = vfs_quota_sync,
 830         .get_info       = vfs_get_dqinfo,
 831         .set_info       = vfs_set_dqinfo,
 832         .get_dqblk      = vfs_get_dqblk,
 833         .set_dqblk      = vfs_set_dqblk
 834 };
 835 #endif
 836
 837 static const struct super_operations ext4_sops = {
 838         .alloc_inode    = ext4_alloc_inode,
 839         .destroy_inode  = ext4_destroy_inode,
 840         .read_inode     = ext4_read_inode,
 841         .write_inode    = ext4_write_inode,
 842         .dirty_inode    = ext4_dirty_inode,
 843         .delete_inode   = ext4_delete_inode,
 844         .put_super      = ext4_put_super,
 845         .write_super    = ext4_write_super,
 846         .sync_fs        = ext4_sync_fs,
 847         .write_super_lockfs = ext4_write_super_lockfs,
 848         .unlockfs       = ext4_unlockfs,
 849         .statfs         = ext4_statfs,
 850         .remount_fs     = ext4_remount,
 851         .clear_inode    = ext4_clear_inode,
 852         .show_options   = ext4_show_options,
 853 #ifdef CONFIG_QUOTA
 854         .quota_read     = ext4_quota_read,
 855         .quota_write    = ext4_quota_write,
 856 #endif
 857 };
 858
 859 static const struct export_operations ext4_export_ops = {
 860         .fh_to_dentry = ext4_fh_to_dentry,
 861         .fh_to_parent = ext4_fh_to_parent,
 862         .get_parent = ext4_get_parent,
 863 };
 864
 865 enum {
 866         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 867         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 868         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 869         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 870         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 871         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 872         Opt_journal_checksum, Opt_journal_async_commit,
 873         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 874         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 875         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 876         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 877         Opt_grpquota, Opt_extents, Opt_noextents,
 878 };
 879
 880 static match_table_t tokens = {
 881         {Opt_bsd_df, "bsddf"},
 882         {Opt_minix_df, "minixdf"},
 883         {Opt_grpid, "grpid"},
 884         {Opt_grpid, "bsdgroups"},
 885         {Opt_nogrpid, "nogrpid"},
 886         {Opt_nogrpid, "sysvgroups"},
 887         {Opt_resgid, "resgid=%u"},
 888         {Opt_resuid, "resuid=%u"},
 889         {Opt_sb, "sb=%u"},
 890         {Opt_err_cont, "errors=continue"},
 891         {Opt_err_panic, "errors=panic"},
 892         {Opt_err_ro, "errors=remount-ro"},
 893         {Opt_nouid32, "nouid32"},
 894         {Opt_nocheck, "nocheck"},
 895         {Opt_nocheck, "check=none"},
 896         {Opt_debug, "debug"},
 897         {Opt_oldalloc, "oldalloc"},
 898         {Opt_orlov, "orlov"},
 899         {Opt_user_xattr, "user_xattr"},
 900         {Opt_nouser_xattr, "nouser_xattr"},
 901         {Opt_acl, "acl"},
 902         {Opt_noacl, "noacl"},
 903         {Opt_reservation, "reservation"},
 904         {Opt_noreservation, "noreservation"},
 905         {Opt_noload, "noload"},
 906         {Opt_nobh, "nobh"},
 907         {Opt_bh, "bh"},
 908         {Opt_commit, "commit=%u"},
 909         {Opt_journal_update, "journal=update"},
 910         {Opt_journal_inum, "journal=%u"},
 911         {Opt_journal_dev, "journal_dev=%u"},
 912         {Opt_journal_checksum, "journal_checksum"},
 913         {Opt_journal_async_commit, "journal_async_commit"},
 914         {Opt_abort, "abort"},
 915         {Opt_data_journal, "data=journal"},
 916         {Opt_data_ordered, "data=ordered"},
 917         {Opt_data_writeback, "data=writeback"},
 918         {Opt_offusrjquota, "usrjquota="},
 919         {Opt_usrjquota, "usrjquota=%s"},
 920         {Opt_offgrpjquota, "grpjquota="},
 921         {Opt_grpjquota, "grpjquota=%s"},
 922         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 923         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 924         {Opt_grpquota, "grpquota"},
 925         {Opt_noquota, "noquota"},
 926         {Opt_quota, "quota"},
 927         {Opt_usrquota, "usrquota"},
 928         {Opt_barrier, "barrier=%u"},
 929         {Opt_extents, "extents"},
 930         {Opt_noextents, "noextents"},
 931         {Opt_err, NULL},
 932         {Opt_resize, "resize"},
 933 };
 934
 935 static ext4_fsblk_t get_sb_block(void **data)
 936 {
 937         ext4_fsblk_t    sb_block;
 938         char            *options = (char *) *data;
 939
 940         if (!options || strncmp(options, "sb=", 3) != 0)
 941                 return 1;       /* Default location */
 942         options += 3;
 943         /*todo: use simple_strtoll with >32bit ext4 */
 944         sb_block = simple_strtoul(options, &options, 0);
 945         if (*options && *options != ',') {
 946                 printk("EXT4-fs: Invalid sb specification: %s\n",
 947                        (char *) *data);
 948                 return 1;
 949         }
 950         if (*options == ',')
 951                 options++;
 952         *data = (void *) options;
 953         return sb_block;
 954 }
 955
 956 static int parse_options (char *options, struct super_block *sb,
 957                           unsigned int *inum, unsigned long *journal_devnum,
 958                           ext4_fsblk_t *n_blocks_count, int is_remount)
 959 {
 960         struct ext4_sb_info *sbi = EXT4_SB(sb);
 961         char * p;
 962         substring_t args[MAX_OPT_ARGS];
 963         int data_opt = 0;
 964         int option;
 965 #ifdef CONFIG_QUOTA
 966         int qtype;
 967         char *qname;
 968 #endif
 969
 970         if (!options)
 971                 return 1;
 972
 973         while ((p = strsep (&options, ",")) != NULL) {
 974                 int token;
 975                 if (!*p)
 976                         continue;
 977
 978                 token = match_token(p, tokens, args);
 979                 switch (token) {
 980                 case Opt_bsd_df:
 981                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 982                         break;
 983                 case Opt_minix_df:
 984                         set_opt (sbi->s_mount_opt, MINIX_DF);
 985                         break;
 986                 case Opt_grpid:
 987                         set_opt (sbi->s_mount_opt, GRPID);
 988                         break;
 989                 case Opt_nogrpid:
 990                         clear_opt (sbi->s_mount_opt, GRPID);
 991                         break;
 992                 case Opt_resuid:
 993                         if (match_int(&args[0], &option))
 994                                 return 0;
 995                         sbi->s_resuid = option;
 996                         break;
 997                 case Opt_resgid:
 998                         if (match_int(&args[0], &option))
 999                                 return 0;
1000                         sbi->s_resgid = option;
1001                         break;
1002                 case Opt_sb:
1003                         /* handled by get_sb_block() instead of here */
1004                         /* *sb_block = match_int(&args[0]); */
1005                         break;
1006                 case Opt_err_panic:
1007                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1008                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1009                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1010                         break;
1011                 case Opt_err_ro:
1012                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1013                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1014                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1015                         break;
1016                 case Opt_err_cont:
1017                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1018                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1019                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1020                         break;
1021                 case Opt_nouid32:
1022                         set_opt (sbi->s_mount_opt, NO_UID32);
1023                         break;
1024                 case Opt_nocheck:
1025                         clear_opt (sbi->s_mount_opt, CHECK);
1026                         break;
1027                 case Opt_debug:
1028                         set_opt (sbi->s_mount_opt, DEBUG);
1029                         break;
1030                 case Opt_oldalloc:
1031                         set_opt (sbi->s_mount_opt, OLDALLOC);
1032                         break;
1033                 case Opt_orlov:
1034                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1035                         break;
1036 #ifdef CONFIG_EXT4DEV_FS_XATTR
1037                 case Opt_user_xattr:
1038                         set_opt (sbi->s_mount_opt, XATTR_USER);
1039                         break;
1040                 case Opt_nouser_xattr:
1041                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1042                         break;
1043 #else
1044                 case Opt_user_xattr:
1045                 case Opt_nouser_xattr:
1046                         printk("EXT4 (no)user_xattr options not supported\n");
1047                         break;
1048 #endif
1049 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1050                 case Opt_acl:
1051                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1052                         break;
1053                 case Opt_noacl:
1054                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1055                         break;
1056 #else
1057                 case Opt_acl:
1058                 case Opt_noacl:
1059                         printk("EXT4 (no)acl options not supported\n");
1060                         break;
1061 #endif
1062                 case Opt_reservation:
1063                         set_opt(sbi->s_mount_opt, RESERVATION);
1064                         break;
1065                 case Opt_noreservation:
1066                         clear_opt(sbi->s_mount_opt, RESERVATION);
1067                         break;
1068                 case Opt_journal_update:
1069                         /* @@@ FIXME */
1070                         /* Eventually we will want to be able to create
1071                            a journal file here.  For now, only allow the
1072                            user to specify an existing inode to be the
1073                            journal file. */
1074                         if (is_remount) {
1075                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1076                                        "journal on remount\n");
1077                                 return 0;
1078                         }
1079                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1080                         break;
1081                 case Opt_journal_inum:
1082                         if (is_remount) {
1083                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1084                                        "journal on remount\n");
1085                                 return 0;
1086                         }
1087                         if (match_int(&args[0], &option))
1088                                 return 0;
1089                         *inum = option;
1090                         break;
1091                 case Opt_journal_dev:
1092                         if (is_remount) {
1093                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1094                                        "journal on remount\n");
1095                                 return 0;
1096                         }
1097                         if (match_int(&args[0], &option))
1098                                 return 0;
1099                         *journal_devnum = option;
1100                         break;
1101                 case Opt_journal_checksum:
1102                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1103                         break;
1104                 case Opt_journal_async_commit:
1105                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1106                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1107                         break;
1108                 case Opt_noload:
1109                         set_opt (sbi->s_mount_opt, NOLOAD);
1110                         break;
1111                 case Opt_commit:
1112                         if (match_int(&args[0], &option))
1113                                 return 0;
1114                         if (option < 0)
1115                                 return 0;
1116                         if (option == 0)
1117                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1118                         sbi->s_commit_interval = HZ * option;
1119                         break;
1120                 case Opt_data_journal:
1121                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1122                         goto datacheck;
1123                 case Opt_data_ordered:
1124                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1125                         goto datacheck;
1126                 case Opt_data_writeback:
1127                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1128                 datacheck:
1129                         if (is_remount) {
1130                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1131                                                 != data_opt) {
1132                                         printk(KERN_ERR
1133                                                 "EXT4-fs: cannot change data "
1134                                                 "mode on remount\n");
1135                                         return 0;
1136                                 }
1137                         } else {
1138                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1139                                 sbi->s_mount_opt |= data_opt;
1140                         }
1141                         break;
1142 #ifdef CONFIG_QUOTA
1143                 case Opt_usrjquota:
1144                         qtype = USRQUOTA;
1145                         goto set_qf_name;
1146                 case Opt_grpjquota:
1147                         qtype = GRPQUOTA;
1148 set_qf_name:
1149                         if (sb_any_quota_enabled(sb)) {
1150                                 printk(KERN_ERR
1151                                         "EXT4-fs: Cannot change journalled "
1152                                         "quota options when quota turned on.\n");
1153                                 return 0;
1154                         }
1155                         qname = match_strdup(&args[0]);
1156                         if (!qname) {
1157                                 printk(KERN_ERR
1158                                         "EXT4-fs: not enough memory for "
1159                                         "storing quotafile name.\n");
1160                                 return 0;
1161                         }
1162                         if (sbi->s_qf_names[qtype] &&
1163                             strcmp(sbi->s_qf_names[qtype], qname)) {
1164                                 printk(KERN_ERR
1165                                         "EXT4-fs: %s quota file already "
1166                                         "specified.\n", QTYPE2NAME(qtype));
1167                                 kfree(qname);
1168                                 return 0;
1169                         }
1170                         sbi->s_qf_names[qtype] = qname;
1171                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1172                                 printk(KERN_ERR
1173                                         "EXT4-fs: quotafile must be on "
1174                                         "filesystem root.\n");
1175                                 kfree(sbi->s_qf_names[qtype]);
1176                                 sbi->s_qf_names[qtype] = NULL;
1177                                 return 0;
1178                         }
1179                         set_opt(sbi->s_mount_opt, QUOTA);
1180                         break;
1181                 case Opt_offusrjquota:
1182                         qtype = USRQUOTA;
1183                         goto clear_qf_name;
1184                 case Opt_offgrpjquota:
1185                         qtype = GRPQUOTA;
1186 clear_qf_name:
1187                         if (sb_any_quota_enabled(sb)) {
1188                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1189                                         "journalled quota options when "
1190                                         "quota turned on.\n");
1191                                 return 0;
1192                         }
1193                         /*
1194                          * The space will be released later when all options
1195                          * are confirmed to be correct
1196                          */
1197                         sbi->s_qf_names[qtype] = NULL;
1198                         break;
1199                 case Opt_jqfmt_vfsold:
1200                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1201                         break;
1202                 case Opt_jqfmt_vfsv0:
1203                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1204                         break;
1205                 case Opt_quota:
1206                 case Opt_usrquota:
1207                         set_opt(sbi->s_mount_opt, QUOTA);
1208                         set_opt(sbi->s_mount_opt, USRQUOTA);
1209                         break;
1210                 case Opt_grpquota:
1211                         set_opt(sbi->s_mount_opt, QUOTA);
1212                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1213                         break;
1214                 case Opt_noquota:
1215                         if (sb_any_quota_enabled(sb)) {
1216                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1217                                         "options when quota turned on.\n");
1218                                 return 0;
1219                         }
1220                         clear_opt(sbi->s_mount_opt, QUOTA);
1221                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1222                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1223                         break;
1224 #else
1225                 case Opt_quota:
1226                 case Opt_usrquota:
1227                 case Opt_grpquota:
1228                 case Opt_usrjquota:
1229                 case Opt_grpjquota:
1230                 case Opt_offusrjquota:
1231                 case Opt_offgrpjquota:
1232                 case Opt_jqfmt_vfsold:
1233                 case Opt_jqfmt_vfsv0:
1234                         printk(KERN_ERR
1235                                 "EXT4-fs: journalled quota options not "
1236                                 "supported.\n");
1237                         break;
1238                 case Opt_noquota:
1239                         break;
1240 #endif
1241                 case Opt_abort:
1242                         set_opt(sbi->s_mount_opt, ABORT);
1243                         break;
1244                 case Opt_barrier:
1245                         if (match_int(&args[0], &option))
1246                                 return 0;
1247                         if (option)
1248                                 set_opt(sbi->s_mount_opt, BARRIER);
1249                         else
1250                                 clear_opt(sbi->s_mount_opt, BARRIER);
1251                         break;
1252                 case Opt_ignore:
1253                         break;
1254                 case Opt_resize:
1255                         if (!is_remount) {
1256                                 printk("EXT4-fs: resize option only available "
1257                                         "for remount\n");
1258                                 return 0;
1259                         }
1260                         if (match_int(&args[0], &option) != 0)
1261                                 return 0;
1262                         *n_blocks_count = option;
1263                         break;
1264                 case Opt_nobh:
1265                         set_opt(sbi->s_mount_opt, NOBH);
1266                         break;
1267                 case Opt_bh:
1268                         clear_opt(sbi->s_mount_opt, NOBH);
1269                         break;
1270                 case Opt_extents:
1271                         set_opt (sbi->s_mount_opt, EXTENTS);
1272                         break;
1273                 case Opt_noextents:
1274                         clear_opt (sbi->s_mount_opt, EXTENTS);
1275                         break;
1276                 default:
1277                         printk (KERN_ERR
1278                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1279                                 "or missing value\n", p);
1280                         return 0;
1281                 }
1282         }
1283 #ifdef CONFIG_QUOTA
1284         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1285                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1286                      sbi->s_qf_names[USRQUOTA])
1287                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1288
1289                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1290                      sbi->s_qf_names[GRPQUOTA])
1291                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1292
1293                 if ((sbi->s_qf_names[USRQUOTA] &&
1294                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1295                     (sbi->s_qf_names[GRPQUOTA] &&
1296                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1297                         printk(KERN_ERR "EXT4-fs: old and new quota "
1298                                         "format mixing.\n");
1299                         return 0;
1300                 }
1301
1302                 if (!sbi->s_jquota_fmt) {
1303                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1304                                         "not specified.\n");
1305                         return 0;
1306                 }
1307         } else {
1308                 if (sbi->s_jquota_fmt) {
1309                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1310                                         "specified with no journalling "
1311                                         "enabled.\n");
1312                         return 0;
1313                 }
1314         }
1315 #endif
1316         return 1;
1317 }
1318
1319 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1320                             int read_only)
1321 {
1322         struct ext4_sb_info *sbi = EXT4_SB(sb);
1323         int res = 0;
1324
1325         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1326                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1327                         "forcing read-only mode\n");
1328                 res = MS_RDONLY;
1329         }
1330         if (read_only)
1331                 return res;
1332         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1333                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1334                         "running e2fsck is recommended\n");
1335         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1336                 printk (KERN_WARNING
1337                         "EXT4-fs warning: mounting fs with errors, "
1338                         "running e2fsck is recommended\n");
1339         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1340                  le16_to_cpu(es->s_mnt_count) >=
1341                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1342                 printk (KERN_WARNING
1343                         "EXT4-fs warning: maximal mount count reached, "
1344                         "running e2fsck is recommended\n");
1345         else if (le32_to_cpu(es->s_checkinterval) &&
1346                 (le32_to_cpu(es->s_lastcheck) +
1347                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1348                 printk (KERN_WARNING
1349                         "EXT4-fs warning: checktime reached, "
1350                         "running e2fsck is recommended\n");
1351 #if 0
1352                 /* @@@ We _will_ want to clear the valid bit if we find
1353                  * inconsistencies, to force a fsck at reboot.  But for
1354                  * a plain journaled filesystem we can keep it set as
1355                  * valid forever! :)
1356                  */
1357         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1358 #endif
1359         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1360                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1361         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1362         es->s_mtime = cpu_to_le32(get_seconds());
1363         ext4_update_dynamic_rev(sb);
1364         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1365
1366         ext4_commit_super(sb, es, 1);
1367         if (test_opt(sb, DEBUG))
1368                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1369                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1370                         sb->s_blocksize,
1371                         sbi->s_groups_count,
1372                         EXT4_BLOCKS_PER_GROUP(sb),
1373                         EXT4_INODES_PER_GROUP(sb),
1374                         sbi->s_mount_opt);
1375
1376         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1377         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1378                 char b[BDEVNAME_SIZE];
1379
1380                 printk("external journal on %s\n",
1381                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1382         } else {
1383                 printk("internal journal\n");
1384         }
1385         return res;
1386 }
1387
1388 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1389                             struct ext4_group_desc *gdp)
1390 {
1391         __u16 crc = 0;
1392
1393         if (sbi->s_es->s_feature_ro_compat &
1394             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1395                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1396                 __le32 le_group = cpu_to_le32(block_group);
1397
1398                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1399                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1400                 crc = crc16(crc, (__u8 *)gdp, offset);
1401                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1402                 /* for checksum of struct ext4_group_desc do the rest...*/
1403                 if ((sbi->s_es->s_feature_incompat &
1404                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1405                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1406                         crc = crc16(crc, (__u8 *)gdp + offset,
1407                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1408                                         offset);
1409         }
1410
1411         return cpu_to_le16(crc);
1412 }
1413
1414 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1415                                 struct ext4_group_desc *gdp)
1416 {
1417         if ((sbi->s_es->s_feature_ro_compat &
1418              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1419             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1420                 return 0;
1421
1422         return 1;
1423 }
1424
1425 /* Called at mount-time, super-block is locked */
1426 static int ext4_check_descriptors (struct super_block * sb)
1427 {
1428         struct ext4_sb_info *sbi = EXT4_SB(sb);
1429         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1430         ext4_fsblk_t last_block;
1431         ext4_fsblk_t block_bitmap;
1432         ext4_fsblk_t inode_bitmap;
1433         ext4_fsblk_t inode_table;
1434         struct ext4_group_desc * gdp = NULL;
1435         int desc_block = 0;
1436         int flexbg_flag = 0;
1437         ext4_group_t i;
1438
1439         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1440                 flexbg_flag = 1;
1441
1442         ext4_debug ("Checking group descriptors");
1443
1444         for (i = 0; i < sbi->s_groups_count; i++)
1445         {
1446                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1447                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1448                 else
1449                         last_block = first_block +
1450                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1451
1452                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1453                         gdp = (struct ext4_group_desc *)
1454                                         sbi->s_group_desc[desc_block++]->b_data;
1455                 block_bitmap = ext4_block_bitmap(sb, gdp);
1456                 if (block_bitmap < first_block || block_bitmap > last_block)
1457                 {
1458                         ext4_error (sb, "ext4_check_descriptors",
1459                                     "Block bitmap for group %lu"
1460                                     " not in group (block %llu)!",
1461                                     i, block_bitmap);
1462                         return 0;
1463                 }
1464                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1465                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1466                 {
1467                         ext4_error (sb, "ext4_check_descriptors",
1468                                     "Inode bitmap for group %lu"
1469                                     " not in group (block %llu)!",
1470                                     i, inode_bitmap);
1471                         return 0;
1472                 }
1473                 inode_table = ext4_inode_table(sb, gdp);
1474                 if (inode_table < first_block ||
1475                     inode_table + sbi->s_itb_per_group - 1 > last_block)
1476                 {
1477                         ext4_error (sb, "ext4_check_descriptors",
1478                                     "Inode table for group %lu"
1479                                     " not in group (block %llu)!",
1480                                     i, inode_table);
1481                         return 0;
1482                 }
1483                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1484                         ext4_error(sb, __FUNCTION__,
1485                                    "Checksum for group %lu failed (%u!=%u)\n",
1486                                     i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1487                                     gdp)), le16_to_cpu(gdp->bg_checksum));
1488                         return 0;
1489                 }
1490                 if (!flexbg_flag)
1491                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1492                 gdp = (struct ext4_group_desc *)
1493                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1494         }
1495
1496         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1497         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1498         return 1;
1499 }
1500
1501 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1502  * the superblock) which were deleted from all directories, but held open by
1503  * a process at the time of a crash.  We walk the list and try to delete these
1504  * inodes at recovery time (only with a read-write filesystem).
1505  *
1506  * In order to keep the orphan inode chain consistent during traversal (in
1507  * case of crash during recovery), we link each inode into the superblock
1508  * orphan list_head and handle it the same way as an inode deletion during
1509  * normal operation (which journals the operations for us).
1510  *
1511  * We only do an iget() and an iput() on each inode, which is very safe if we
1512  * accidentally point at an in-use or already deleted inode.  The worst that
1513  * can happen in this case is that we get a "bit already cleared" message from
1514  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1515  * e2fsck was run on this filesystem, and it must have already done the orphan
1516  * inode cleanup for us, so we can safely abort without any further action.
1517  */
1518 static void ext4_orphan_cleanup (struct super_block * sb,
1519                                  struct ext4_super_block * es)
1520 {
1521         unsigned int s_flags = sb->s_flags;
1522         int nr_orphans = 0, nr_truncates = 0;
1523 #ifdef CONFIG_QUOTA
1524         int i;
1525 #endif
1526         if (!es->s_last_orphan) {
1527                 jbd_debug(4, "no orphan inodes to clean up\n");
1528                 return;
1529         }
1530
1531         if (bdev_read_only(sb->s_bdev)) {
1532                 printk(KERN_ERR "EXT4-fs: write access "
1533                         "unavailable, skipping orphan cleanup.\n");
1534                 return;
1535         }
1536
1537         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1538                 if (es->s_last_orphan)
1539                         jbd_debug(1, "Errors on filesystem, "
1540                                   "clearing orphan list.\n");
1541                 es->s_last_orphan = 0;
1542                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1543                 return;
1544         }
1545
1546         if (s_flags & MS_RDONLY) {
1547                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1548                        sb->s_id);
1549                 sb->s_flags &= ~MS_RDONLY;
1550         }
1551 #ifdef CONFIG_QUOTA
1552         /* Needed for iput() to work correctly and not trash data */
1553         sb->s_flags |= MS_ACTIVE;
1554         /* Turn on quotas so that they are updated correctly */
1555         for (i = 0; i < MAXQUOTAS; i++) {
1556                 if (EXT4_SB(sb)->s_qf_names[i]) {
1557                         int ret = ext4_quota_on_mount(sb, i);
1558                         if (ret < 0)
1559                                 printk(KERN_ERR
1560                                         "EXT4-fs: Cannot turn on journalled "
1561                                         "quota: error %d\n", ret);
1562                 }
1563         }
1564 #endif
1565
1566         while (es->s_last_orphan) {
1567                 struct inode *inode;
1568
1569                 if (!(inode =
1570                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1571                         es->s_last_orphan = 0;
1572                         break;
1573                 }
1574
1575                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1576                 DQUOT_INIT(inode);
1577                 if (inode->i_nlink) {
1578                         printk(KERN_DEBUG
1579                                 "%s: truncating inode %lu to %Ld bytes\n",
1580                                 __FUNCTION__, inode->i_ino, inode->i_size);
1581                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1582                                   inode->i_ino, inode->i_size);
1583                         ext4_truncate(inode);
1584                         nr_truncates++;
1585                 } else {
1586                         printk(KERN_DEBUG
1587                                 "%s: deleting unreferenced inode %lu\n",
1588                                 __FUNCTION__, inode->i_ino);
1589                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1590                                   inode->i_ino);
1591                         nr_orphans++;
1592                 }
1593                 iput(inode);  /* The delete magic happens here! */
1594         }
1595
1596 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1597
1598         if (nr_orphans)
1599                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1600                        sb->s_id, PLURAL(nr_orphans));
1601         if (nr_truncates)
1602                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1603                        sb->s_id, PLURAL(nr_truncates));
1604 #ifdef CONFIG_QUOTA
1605         /* Turn quotas off */
1606         for (i = 0; i < MAXQUOTAS; i++) {
1607                 if (sb_dqopt(sb)->files[i])
1608                         vfs_quota_off(sb, i);
1609         }
1610 #endif
1611         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1612 }
1613 /*
1614  * Maximal extent format file size.
1615  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1616  * extent format containers, within a sector_t, and within i_blocks
1617  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1618  * so that won't be a limiting factor.
1619  *
1620  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1621  */
1622 static loff_t ext4_max_size(int blkbits)
1623 {
1624         loff_t res;
1625         loff_t upper_limit = MAX_LFS_FILESIZE;
1626
1627         /* small i_blocks in vfs inode? */
1628         if (sizeof(blkcnt_t) < sizeof(u64)) {
1629                 /*
1630                  * CONFIG_LSF is not enabled implies the inode
1631                  * i_block represent total blocks in 512 bytes
1632                  * 32 == size of vfs inode i_blocks * 8
1633                  */
1634                 upper_limit = (1LL << 32) - 1;
1635
1636                 /* total blocks in file system block size */
1637                 upper_limit >>= (blkbits - 9);
1638                 upper_limit <<= blkbits;
1639         }
1640
1641         /* 32-bit extent-start container, ee_block */
1642         res = 1LL << 32;
1643         res <<= blkbits;
1644         res -= 1;
1645
1646         /* Sanity check against vm- & vfs- imposed limits */
1647         if (res > upper_limit)
1648                 res = upper_limit;
1649
1650         return res;
1651 }
1652
1653 /*
1654  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1655  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1656  * We need to be 1 filesystem block less than the 2^48 sector limit.
1657  */
1658 static loff_t ext4_max_bitmap_size(int bits)
1659 {
1660         loff_t res = EXT4_NDIR_BLOCKS;
1661         int meta_blocks;
1662         loff_t upper_limit;
1663         /* This is calculated to be the largest file size for a
1664          * dense, bitmapped file such that the total number of
1665          * sectors in the file, including data and all indirect blocks,
1666          * does not exceed 2^48 -1
1667          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1668          * total number of  512 bytes blocks of the file
1669          */
1670
1671         if (sizeof(blkcnt_t) < sizeof(u64)) {
1672                 /*
1673                  * CONFIG_LSF is not enabled implies the inode
1674                  * i_block represent total blocks in 512 bytes
1675                  * 32 == size of vfs inode i_blocks * 8
1676                  */
1677                 upper_limit = (1LL << 32) - 1;
1678
1679                 /* total blocks in file system block size */
1680                 upper_limit >>= (bits - 9);
1681
1682         } else {
1683                 /*
1684                  * We use 48 bit ext4_inode i_blocks
1685                  * With EXT4_HUGE_FILE_FL set the i_blocks
1686                  * represent total number of blocks in
1687                  * file system block size
1688                  */
1689                 upper_limit = (1LL << 48) - 1;
1690
1691         }
1692
1693         /* indirect blocks */
1694         meta_blocks = 1;
1695         /* double indirect blocks */
1696         meta_blocks += 1 + (1LL << (bits-2));
1697         /* tripple indirect blocks */
1698         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1699
1700         upper_limit -= meta_blocks;
1701         upper_limit <<= bits;
1702
1703         res += 1LL << (bits-2);
1704         res += 1LL << (2*(bits-2));
1705         res += 1LL << (3*(bits-2));
1706         res <<= bits;
1707         if (res > upper_limit)
1708                 res = upper_limit;
1709
1710         if (res > MAX_LFS_FILESIZE)
1711                 res = MAX_LFS_FILESIZE;
1712
1713         return res;
1714 }
1715
1716 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1717                                 ext4_fsblk_t logical_sb_block, int nr)
1718 {
1719         struct ext4_sb_info *sbi = EXT4_SB(sb);
1720         ext4_group_t bg, first_meta_bg;
1721         int has_super = 0;
1722
1723         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1724
1725         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1726             nr < first_meta_bg)
1727                 return logical_sb_block + nr + 1;
1728         bg = sbi->s_desc_per_block * nr;
1729         if (ext4_bg_has_super(sb, bg))
1730                 has_super = 1;
1731         return (has_super + ext4_group_first_block_no(sb, bg));
1732 }
1733
1734
1735 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1736                                 __releases(kernel_sem)
1737                                 __acquires(kernel_sem)
1738
1739 {
1740         struct buffer_head * bh;
1741         struct ext4_super_block *es = NULL;
1742         struct ext4_sb_info *sbi;
1743         ext4_fsblk_t block;
1744         ext4_fsblk_t sb_block = get_sb_block(&data);
1745         ext4_fsblk_t logical_sb_block;
1746         unsigned long offset = 0;
1747         unsigned int journal_inum = 0;
1748         unsigned long journal_devnum = 0;
1749         unsigned long def_mount_opts;
1750         struct inode *root;
1751         int blocksize;
1752         int hblock;
1753         int db_count;
1754         int i;
1755         int needs_recovery;
1756         __le32 features;
1757         __u64 blocks_count;
1758         int err;
1759
1760         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1761         if (!sbi)
1762                 return -ENOMEM;
1763         sb->s_fs_info = sbi;
1764         sbi->s_mount_opt = 0;
1765         sbi->s_resuid = EXT4_DEF_RESUID;
1766         sbi->s_resgid = EXT4_DEF_RESGID;
1767         sbi->s_sb_block = sb_block;
1768
1769         unlock_kernel();
1770
1771         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1772         if (!blocksize) {
1773                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1774                 goto out_fail;
1775         }
1776
1777         if (!sb_set_blocksize(sb, blocksize)) {
1778                 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1779                 goto out_fail;
1780         }
1781
1782         /*
1783          * The ext4 superblock will not be buffer aligned for other than 1kB
1784          * block sizes.  We need to calculate the offset from buffer start.
1785          */
1786         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1787                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1788                 offset = do_div(logical_sb_block, blocksize);
1789         } else {
1790                 logical_sb_block = sb_block;
1791         }
1792
1793         if (!(bh = sb_bread(sb, logical_sb_block))) {
1794                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1795                 goto out_fail;
1796         }
1797         /*
1798          * Note: s_es must be initialized as soon as possible because
1799          *       some ext4 macro-instructions depend on its value
1800          */
1801         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1802         sbi->s_es = es;
1803         sb->s_magic = le16_to_cpu(es->s_magic);
1804         if (sb->s_magic != EXT4_SUPER_MAGIC)
1805                 goto cantfind_ext4;
1806
1807         /* Set defaults before we parse the mount options */
1808         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1809         if (def_mount_opts & EXT4_DEFM_DEBUG)
1810                 set_opt(sbi->s_mount_opt, DEBUG);
1811         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1812                 set_opt(sbi->s_mount_opt, GRPID);
1813         if (def_mount_opts & EXT4_DEFM_UID16)
1814                 set_opt(sbi->s_mount_opt, NO_UID32);
1815 #ifdef CONFIG_EXT4DEV_FS_XATTR
1816         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1817                 set_opt(sbi->s_mount_opt, XATTR_USER);
1818 #endif
1819 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1820         if (def_mount_opts & EXT4_DEFM_ACL)
1821                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1822 #endif
1823         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1824                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1825         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1826                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1827         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1828                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1829
1830         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1831                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1832         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1833                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1834         else
1835                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1836
1837         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1838         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1839
1840         set_opt(sbi->s_mount_opt, RESERVATION);
1841
1842         /*
1843          * turn on extents feature by default in ext4 filesystem
1844          * User -o noextents to turn it off
1845          */
1846         set_opt(sbi->s_mount_opt, EXTENTS);
1847
1848         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1849                             NULL, 0))
1850                 goto failed_mount;
1851
1852         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1853                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1854
1855         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1856             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1857              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1858              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1859                 printk(KERN_WARNING
1860                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1861                        "running e2fsck is recommended\n");
1862         /*
1863          * Check feature flags regardless of the revision level, since we
1864          * previously didn't change the revision level when setting the flags,
1865          * so there is a chance incompat flags are set on a rev 0 filesystem.
1866          */
1867         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1868         if (features) {
1869                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1870                        "unsupported optional features (%x).\n",
1871                        sb->s_id, le32_to_cpu(features));
1872                 goto failed_mount;
1873         }
1874         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1875         if (!(sb->s_flags & MS_RDONLY) && features) {
1876                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1877                        "unsupported optional features (%x).\n",
1878                        sb->s_id, le32_to_cpu(features));
1879                 goto failed_mount;
1880         }
1881         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
1882                 /*
1883                  * Large file size enabled file system can only be
1884                  * mount if kernel is build with CONFIG_LSF
1885                  */
1886                 if (sizeof(root->i_blocks) < sizeof(u64) &&
1887                                 !(sb->s_flags & MS_RDONLY)) {
1888                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
1889                                         "files cannot be mounted read-write "
1890                                         "without CONFIG_LSF.\n", sb->s_id);
1891                         goto failed_mount;
1892                 }
1893         }
1894         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1895
1896         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1897             blocksize > EXT4_MAX_BLOCK_SIZE) {
1898                 printk(KERN_ERR
1899                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1900                        blocksize, sb->s_id);
1901                 goto failed_mount;
1902         }
1903
1904         hblock = bdev_hardsect_size(sb->s_bdev);
1905         if (sb->s_blocksize != blocksize) {
1906                 /*
1907                  * Make sure the blocksize for the filesystem is larger
1908                  * than the hardware sectorsize for the machine.
1909                  */
1910                 if (blocksize < hblock) {
1911                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1912                                "device blocksize %d.\n", blocksize, hblock);
1913                         goto failed_mount;
1914                 }
1915
1916                 brelse (bh);
1917                 sb_set_blocksize(sb, blocksize);
1918                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1919                 offset = do_div(logical_sb_block, blocksize);
1920                 bh = sb_bread(sb, logical_sb_block);
1921                 if (!bh) {
1922                         printk(KERN_ERR
1923                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1924                         goto failed_mount;
1925                 }
1926                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1927                 sbi->s_es = es;
1928                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1929                         printk (KERN_ERR
1930                                 "EXT4-fs: Magic mismatch, very weird !\n");
1931                         goto failed_mount;
1932                 }
1933         }
1934
1935         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
1936         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1937
1938         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1939                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1940                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1941         } else {
1942                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1943                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1944                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1945                     (!is_power_of_2(sbi->s_inode_size)) ||
1946                     (sbi->s_inode_size > blocksize)) {
1947                         printk (KERN_ERR
1948                                 "EXT4-fs: unsupported inode size: %d\n",
1949                                 sbi->s_inode_size);
1950                         goto failed_mount;
1951                 }
1952                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
1953                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
1954         }
1955         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1956         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1957                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1958                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1959                     !is_power_of_2(sbi->s_desc_size)) {
1960                         printk(KERN_ERR
1961                                "EXT4-fs: unsupported descriptor size %lu\n",
1962                                sbi->s_desc_size);
1963                         goto failed_mount;
1964                 }
1965         } else
1966                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1967         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1968         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1969         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
1970                 goto cantfind_ext4;
1971         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1972         if (sbi->s_inodes_per_block == 0)
1973                 goto cantfind_ext4;
1974         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1975                                         sbi->s_inodes_per_block;
1976         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1977         sbi->s_sbh = bh;
1978         sbi->s_mount_state = le16_to_cpu(es->s_state);
1979         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
1980         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
1981         for (i=0; i < 4; i++)
1982                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1983         sbi->s_def_hash_version = es->s_def_hash_version;
1984
1985         if (sbi->s_blocks_per_group > blocksize * 8) {
1986                 printk (KERN_ERR
1987                         "EXT4-fs: #blocks per group too big: %lu\n",
1988                         sbi->s_blocks_per_group);
1989                 goto failed_mount;
1990         }
1991         if (sbi->s_inodes_per_group > blocksize * 8) {
1992                 printk (KERN_ERR
1993                         "EXT4-fs: #inodes per group too big: %lu\n",
1994                         sbi->s_inodes_per_group);
1995                 goto failed_mount;
1996         }
1997
1998         if (ext4_blocks_count(es) >
1999                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2000                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2001                         " too large to mount safely\n", sb->s_id);
2002                 if (sizeof(sector_t) < 8)
2003                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2004                                         "enabled\n");
2005                 goto failed_mount;
2006         }
2007
2008         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2009                 goto cantfind_ext4;
2010
2011         /* ensure blocks_count calculation below doesn't sign-extend */
2012         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2013             le32_to_cpu(es->s_first_data_block) + 1) {
2014                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2015                        "first data block %u, blocks per group %lu\n",
2016                         ext4_blocks_count(es),
2017                         le32_to_cpu(es->s_first_data_block),
2018                         EXT4_BLOCKS_PER_GROUP(sb));
2019                 goto failed_mount;
2020         }
2021         blocks_count = (ext4_blocks_count(es) -
2022                         le32_to_cpu(es->s_first_data_block) +
2023                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2024         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2025         sbi->s_groups_count = blocks_count;
2026         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2027                    EXT4_DESC_PER_BLOCK(sb);
2028         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
2029                                     GFP_KERNEL);
2030         if (sbi->s_group_desc == NULL) {
2031                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
2032                 goto failed_mount;
2033         }
2034
2035         bgl_lock_init(&sbi->s_blockgroup_lock);
2036
2037         for (i = 0; i < db_count; i++) {
2038                 block = descriptor_loc(sb, logical_sb_block, i);
2039                 sbi->s_group_desc[i] = sb_bread(sb, block);
2040                 if (!sbi->s_group_desc[i]) {
2041                         printk (KERN_ERR "EXT4-fs: "
2042                                 "can't read group descriptor %d\n", i);
2043                         db_count = i;
2044                         goto failed_mount2;
2045                 }
2046         }
2047         if (!ext4_check_descriptors (sb)) {
2048                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2049                 goto failed_mount2;
2050         }
2051         sbi->s_gdb_count = db_count;
2052         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2053         spin_lock_init(&sbi->s_next_gen_lock);
2054
2055         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2056                         ext4_count_free_blocks(sb));
2057         if (!err) {
2058                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2059                                 ext4_count_free_inodes(sb));
2060         }
2061         if (!err) {
2062                 err = percpu_counter_init(&sbi->s_dirs_counter,
2063                                 ext4_count_dirs(sb));
2064         }
2065         if (err) {
2066                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2067                 goto failed_mount3;
2068         }
2069
2070         /* per fileystem reservation list head & lock */
2071         spin_lock_init(&sbi->s_rsv_window_lock);
2072         sbi->s_rsv_window_root = RB_ROOT;
2073         /* Add a single, static dummy reservation to the start of the
2074          * reservation window list --- it gives us a placeholder for
2075          * append-at-start-of-list which makes the allocation logic
2076          * _much_ simpler. */
2077         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2078         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2079         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2080         sbi->s_rsv_window_head.rsv_goal_size = 0;
2081         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2082
2083         /*
2084          * set up enough so that it can read an inode
2085          */
2086         sb->s_op = &ext4_sops;
2087         sb->s_export_op = &ext4_export_ops;
2088         sb->s_xattr = ext4_xattr_handlers;
2089 #ifdef CONFIG_QUOTA
2090         sb->s_qcop = &ext4_qctl_operations;
2091         sb->dq_op = &ext4_quota_operations;
2092 #endif
2093         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2094
2095         sb->s_root = NULL;
2096
2097         needs_recovery = (es->s_last_orphan != 0 ||
2098                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2099                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2100
2101         /*
2102          * The first inode we look at is the journal inode.  Don't try
2103          * root first: it may be modified in the journal!
2104          */
2105         if (!test_opt(sb, NOLOAD) &&
2106             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2107                 if (ext4_load_journal(sb, es, journal_devnum))
2108                         goto failed_mount3;
2109         } else if (journal_inum) {
2110                 if (ext4_create_journal(sb, es, journal_inum))
2111                         goto failed_mount3;
2112         } else {
2113                 if (!silent)
2114                         printk (KERN_ERR
2115                                 "ext4: No journal on filesystem on %s\n",
2116                                 sb->s_id);
2117                 goto failed_mount3;
2118         }
2119
2120         if (ext4_blocks_count(es) > 0xffffffffULL &&
2121             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2122                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2123                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2124                 goto failed_mount4;
2125         }
2126
2127         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2128                 jbd2_journal_set_features(sbi->s_journal,
2129                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2130                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2131         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2132                 jbd2_journal_set_features(sbi->s_journal,
2133                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2134                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2135                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2136         } else {
2137                 jbd2_journal_clear_features(sbi->s_journal,
2138                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2139                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2140         }
2141
2142         /* We have now updated the journal if required, so we can
2143          * validate the data journaling mode. */
2144         switch (test_opt(sb, DATA_FLAGS)) {
2145         case 0:
2146                 /* No mode set, assume a default based on the journal
2147                  * capabilities: ORDERED_DATA if the journal can
2148                  * cope, else JOURNAL_DATA
2149                  */
2150                 if (jbd2_journal_check_available_features
2151                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2152                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2153                 else
2154                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2155                 break;
2156
2157         case EXT4_MOUNT_ORDERED_DATA:
2158         case EXT4_MOUNT_WRITEBACK_DATA:
2159                 if (!jbd2_journal_check_available_features
2160                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2161                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2162                                "requested data journaling mode\n");
2163                         goto failed_mount4;
2164                 }
2165         default:
2166                 break;
2167         }
2168
2169         if (test_opt(sb, NOBH)) {
2170                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2171                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2172                                 "its supported only with writeback mode\n");
2173                         clear_opt(sbi->s_mount_opt, NOBH);
2174                 }
2175         }
2176         /*
2177          * The jbd2_journal_load will have done any necessary log recovery,
2178          * so we can safely mount the rest of the filesystem now.
2179          */
2180
2181         root = iget(sb, EXT4_ROOT_INO);
2182         sb->s_root = d_alloc_root(root);
2183         if (!sb->s_root) {
2184                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2185                 iput(root);
2186                 goto failed_mount4;
2187         }
2188         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2189                 dput(sb->s_root);
2190                 sb->s_root = NULL;
2191                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2192                 goto failed_mount4;
2193         }
2194
2195         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2196
2197         /* determine the minimum size of new large inodes, if present */
2198         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2199                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2200                                                      EXT4_GOOD_OLD_INODE_SIZE;
2201                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2202                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2203                         if (sbi->s_want_extra_isize <
2204                             le16_to_cpu(es->s_want_extra_isize))
2205                                 sbi->s_want_extra_isize =
2206                                         le16_to_cpu(es->s_want_extra_isize);
2207                         if (sbi->s_want_extra_isize <
2208                             le16_to_cpu(es->s_min_extra_isize))
2209                                 sbi->s_want_extra_isize =
2210                                         le16_to_cpu(es->s_min_extra_isize);
2211                 }
2212         }
2213         /* Check if enough inode space is available */
2214         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2215                                                         sbi->s_inode_size) {
2216                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2217                                                        EXT4_GOOD_OLD_INODE_SIZE;
2218                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2219                         "available.\n");
2220         }
2221
2222         /*
2223          * akpm: core read_super() calls in here with the superblock locked.
2224          * That deadlocks, because orphan cleanup needs to lock the superblock
2225          * in numerous places.  Here we just pop the lock - it's relatively
2226          * harmless, because we are now ready to accept write_super() requests,
2227          * and aviro says that's the only reason for hanging onto the
2228          * superblock lock.
2229          */
2230         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2231         ext4_orphan_cleanup(sb, es);
2232         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2233         if (needs_recovery)
2234                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
2235         ext4_mark_recovery_complete(sb, es);
2236         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2237                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2238                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2239                 "writeback");
2240
2241         ext4_ext_init(sb);
2242
2243         lock_kernel();
2244         return 0;
2245
2246 cantfind_ext4:
2247         if (!silent)
2248                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2249                        sb->s_id);
2250         goto failed_mount;
2251
2252 failed_mount4:
2253         jbd2_journal_destroy(sbi->s_journal);
2254 failed_mount3:
2255         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2256         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2257         percpu_counter_destroy(&sbi->s_dirs_counter);
2258 failed_mount2:
2259         for (i = 0; i < db_count; i++)
2260                 brelse(sbi->s_group_desc[i]);
2261         kfree(sbi->s_group_desc);
2262 failed_mount:
2263 #ifdef CONFIG_QUOTA
2264         for (i = 0; i < MAXQUOTAS; i++)
2265                 kfree(sbi->s_qf_names[i]);
2266 #endif
2267         ext4_blkdev_remove(sbi);
2268         brelse(bh);
2269 out_fail:
2270         sb->s_fs_info = NULL;
2271         kfree(sbi);
2272         lock_kernel();
2273         return -EINVAL;
2274 }
2275
2276 /*
2277  * Setup any per-fs journal parameters now.  We'll do this both on
2278  * initial mount, once the journal has been initialised but before we've
2279  * done any recovery; and again on any subsequent remount.
2280  */
2281 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2282 {
2283         struct ext4_sb_info *sbi = EXT4_SB(sb);
2284
2285         if (sbi->s_commit_interval)
2286                 journal->j_commit_interval = sbi->s_commit_interval;
2287         /* We could also set up an ext4-specific default for the commit
2288          * interval here, but for now we'll just fall back to the jbd
2289          * default. */
2290
2291         spin_lock(&journal->j_state_lock);
2292         if (test_opt(sb, BARRIER))
2293                 journal->j_flags |= JBD2_BARRIER;
2294         else
2295                 journal->j_flags &= ~JBD2_BARRIER;
2296         spin_unlock(&journal->j_state_lock);
2297 }
2298
2299 static journal_t *ext4_get_journal(struct super_block *sb,
2300                                    unsigned int journal_inum)
2301 {
2302         struct inode *journal_inode;
2303         journal_t *journal;
2304
2305         /* First, test for the existence of a valid inode on disk.  Bad
2306          * things happen if we iget() an unused inode, as the subsequent
2307          * iput() will try to delete it. */
2308
2309         journal_inode = iget(sb, journal_inum);
2310         if (!journal_inode) {
2311                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2312                 return NULL;
2313         }
2314         if (!journal_inode->i_nlink) {
2315                 make_bad_inode(journal_inode);
2316                 iput(journal_inode);
2317                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2318                 return NULL;
2319         }
2320
2321         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2322                   journal_inode, journal_inode->i_size);
2323         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
2324                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2325                 iput(journal_inode);
2326                 return NULL;
2327         }
2328
2329         journal = jbd2_journal_init_inode(journal_inode);
2330         if (!journal) {
2331                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2332                 iput(journal_inode);
2333                 return NULL;
2334         }
2335         journal->j_private = sb;
2336         ext4_init_journal_params(sb, journal);
2337         return journal;
2338 }
2339
2340 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2341                                        dev_t j_dev)
2342 {
2343         struct buffer_head * bh;
2344         journal_t *journal;
2345         ext4_fsblk_t start;
2346         ext4_fsblk_t len;
2347         int hblock, blocksize;
2348         ext4_fsblk_t sb_block;
2349         unsigned long offset;
2350         struct ext4_super_block * es;
2351         struct block_device *bdev;
2352
2353         bdev = ext4_blkdev_get(j_dev);
2354         if (bdev == NULL)
2355                 return NULL;
2356
2357         if (bd_claim(bdev, sb)) {
2358                 printk(KERN_ERR
2359                         "EXT4: failed to claim external journal device.\n");
2360                 blkdev_put(bdev);
2361                 return NULL;
2362         }
2363
2364         blocksize = sb->s_blocksize;
2365         hblock = bdev_hardsect_size(bdev);
2366         if (blocksize < hblock) {
2367                 printk(KERN_ERR
2368                         "EXT4-fs: blocksize too small for journal device.\n");
2369                 goto out_bdev;
2370         }
2371
2372         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2373         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2374         set_blocksize(bdev, blocksize);
2375         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2376                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2377                        "external journal\n");
2378                 goto out_bdev;
2379         }
2380
2381         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2382         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2383             !(le32_to_cpu(es->s_feature_incompat) &
2384               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2385                 printk(KERN_ERR "EXT4-fs: external journal has "
2386                                         "bad superblock\n");
2387                 brelse(bh);
2388                 goto out_bdev;
2389         }
2390
2391         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2392                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2393                 brelse(bh);
2394                 goto out_bdev;
2395         }
2396
2397         len = ext4_blocks_count(es);
2398         start = sb_block + 1;
2399         brelse(bh);     /* we're done with the superblock */
2400
2401         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2402                                         start, len, blocksize);
2403         if (!journal) {
2404                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2405                 goto out_bdev;
2406         }
2407         journal->j_private = sb;
2408         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2409         wait_on_buffer(journal->j_sb_buffer);
2410         if (!buffer_uptodate(journal->j_sb_buffer)) {
2411                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2412                 goto out_journal;
2413         }
2414         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2415                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2416                                         "user (unsupported) - %d\n",
2417                         be32_to_cpu(journal->j_superblock->s_nr_users));
2418                 goto out_journal;
2419         }
2420         EXT4_SB(sb)->journal_bdev = bdev;
2421         ext4_init_journal_params(sb, journal);
2422         return journal;
2423 out_journal:
2424         jbd2_journal_destroy(journal);
2425 out_bdev:
2426         ext4_blkdev_put(bdev);
2427         return NULL;
2428 }
2429
2430 static int ext4_load_journal(struct super_block *sb,
2431                              struct ext4_super_block *es,
2432                              unsigned long journal_devnum)
2433 {
2434         journal_t *journal;
2435         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2436         dev_t journal_dev;
2437         int err = 0;
2438         int really_read_only;
2439
2440         if (journal_devnum &&
2441             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2442                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2443                         "numbers have changed\n");
2444                 journal_dev = new_decode_dev(journal_devnum);
2445         } else
2446                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2447
2448         really_read_only = bdev_read_only(sb->s_bdev);
2449
2450         /*
2451          * Are we loading a blank journal or performing recovery after a
2452          * crash?  For recovery, we need to check in advance whether we
2453          * can get read-write access to the device.
2454          */
2455
2456         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2457                 if (sb->s_flags & MS_RDONLY) {
2458                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2459                                         "required on readonly filesystem.\n");
2460                         if (really_read_only) {
2461                                 printk(KERN_ERR "EXT4-fs: write access "
2462                                         "unavailable, cannot proceed.\n");
2463                                 return -EROFS;
2464                         }
2465                         printk (KERN_INFO "EXT4-fs: write access will "
2466                                         "be enabled during recovery.\n");
2467                 }
2468         }
2469
2470         if (journal_inum && journal_dev) {
2471                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2472                        "and inode journals!\n");
2473                 return -EINVAL;
2474         }
2475
2476         if (journal_inum) {
2477                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2478                         return -EINVAL;
2479         } else {
2480                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2481                         return -EINVAL;
2482         }
2483
2484         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2485                 err = jbd2_journal_update_format(journal);
2486                 if (err)  {
2487                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2488                         jbd2_journal_destroy(journal);
2489                         return err;
2490                 }
2491         }
2492
2493         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2494                 err = jbd2_journal_wipe(journal, !really_read_only);
2495         if (!err)
2496                 err = jbd2_journal_load(journal);
2497
2498         if (err) {
2499                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2500                 jbd2_journal_destroy(journal);
2501                 return err;
2502         }
2503
2504         EXT4_SB(sb)->s_journal = journal;
2505         ext4_clear_journal_err(sb, es);
2506
2507         if (journal_devnum &&
2508             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2509                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2510                 sb->s_dirt = 1;
2511
2512                 /* Make sure we flush the recovery flag to disk. */
2513                 ext4_commit_super(sb, es, 1);
2514         }
2515
2516         return 0;
2517 }
2518
2519 static int ext4_create_journal(struct super_block * sb,
2520                                struct ext4_super_block * es,
2521                                unsigned int journal_inum)
2522 {
2523         journal_t *journal;
2524         int err;
2525
2526         if (sb->s_flags & MS_RDONLY) {
2527                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2528                                 "create journal.\n");
2529                 return -EROFS;
2530         }
2531
2532         journal = ext4_get_journal(sb, journal_inum);
2533         if (!journal)
2534                 return -EINVAL;
2535
2536         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2537                journal_inum);
2538
2539         err = jbd2_journal_create(journal);
2540         if (err) {
2541                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2542                 jbd2_journal_destroy(journal);
2543                 return -EIO;
2544         }
2545
2546         EXT4_SB(sb)->s_journal = journal;
2547
2548         ext4_update_dynamic_rev(sb);
2549         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2550         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2551
2552         es->s_journal_inum = cpu_to_le32(journal_inum);
2553         sb->s_dirt = 1;
2554
2555         /* Make sure we flush the recovery flag to disk. */
2556         ext4_commit_super(sb, es, 1);
2557
2558         return 0;
2559 }
2560
2561 static void ext4_commit_super (struct super_block * sb,
2562                                struct ext4_super_block * es,
2563                                int sync)
2564 {
2565         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2566
2567         if (!sbh)
2568                 return;
2569         es->s_wtime = cpu_to_le32(get_seconds());
2570         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2571         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2572         BUFFER_TRACE(sbh, "marking dirty");
2573         mark_buffer_dirty(sbh);
2574         if (sync)
2575                 sync_dirty_buffer(sbh);
2576 }
2577
2578
2579 /*
2580  * Have we just finished recovery?  If so, and if we are mounting (or
2581  * remounting) the filesystem readonly, then we will end up with a
2582  * consistent fs on disk.  Record that fact.
2583  */
2584 static void ext4_mark_recovery_complete(struct super_block * sb,
2585                                         struct ext4_super_block * es)
2586 {
2587         journal_t *journal = EXT4_SB(sb)->s_journal;
2588
2589         jbd2_journal_lock_updates(journal);
2590         jbd2_journal_flush(journal);
2591         lock_super(sb);
2592         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2593             sb->s_flags & MS_RDONLY) {
2594                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2595                 sb->s_dirt = 0;
2596                 ext4_commit_super(sb, es, 1);
2597         }
2598         unlock_super(sb);
2599         jbd2_journal_unlock_updates(journal);
2600 }
2601
2602 /*
2603  * If we are mounting (or read-write remounting) a filesystem whose journal
2604  * has recorded an error from a previous lifetime, move that error to the
2605  * main filesystem now.
2606  */
2607 static void ext4_clear_journal_err(struct super_block * sb,
2608                                    struct ext4_super_block * es)
2609 {
2610         journal_t *journal;
2611         int j_errno;
2612         const char *errstr;
2613
2614         journal = EXT4_SB(sb)->s_journal;
2615
2616         /*
2617          * Now check for any error status which may have been recorded in the
2618          * journal by a prior ext4_error() or ext4_abort()
2619          */
2620
2621         j_errno = jbd2_journal_errno(journal);
2622         if (j_errno) {
2623                 char nbuf[16];
2624
2625                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2626                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2627                              "from previous mount: %s", errstr);
2628                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2629                              "filesystem check.");
2630
2631                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2632                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2633                 ext4_commit_super (sb, es, 1);
2634
2635                 jbd2_journal_clear_err(journal);
2636         }
2637 }
2638
2639 /*
2640  * Force the running and committing transactions to commit,
2641  * and wait on the commit.
2642  */
2643 int ext4_force_commit(struct super_block *sb)
2644 {
2645         journal_t *journal;
2646         int ret;
2647
2648         if (sb->s_flags & MS_RDONLY)
2649                 return 0;
2650
2651         journal = EXT4_SB(sb)->s_journal;
2652         sb->s_dirt = 0;
2653         ret = ext4_journal_force_commit(journal);
2654         return ret;
2655 }
2656
2657 /*
2658  * Ext4 always journals updates to the superblock itself, so we don't
2659  * have to propagate any other updates to the superblock on disk at this
2660  * point.  Just start an async writeback to get the buffers on their way
2661  * to the disk.
2662  *
2663  * This implicitly triggers the writebehind on sync().
2664  */
2665
2666 static void ext4_write_super (struct super_block * sb)
2667 {
2668         if (mutex_trylock(&sb->s_lock) != 0)
2669                 BUG();
2670         sb->s_dirt = 0;
2671 }
2672
2673 static int ext4_sync_fs(struct super_block *sb, int wait)
2674 {
2675         tid_t target;
2676
2677         sb->s_dirt = 0;
2678         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2679                 if (wait)
2680                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2681         }
2682         return 0;
2683 }
2684
2685 /*
2686  * LVM calls this function before a (read-only) snapshot is created.  This
2687  * gives us a chance to flush the journal completely and mark the fs clean.
2688  */
2689 static void ext4_write_super_lockfs(struct super_block *sb)
2690 {
2691         sb->s_dirt = 0;
2692
2693         if (!(sb->s_flags & MS_RDONLY)) {
2694                 journal_t *journal = EXT4_SB(sb)->s_journal;
2695
2696                 /* Now we set up the journal barrier. */
2697                 jbd2_journal_lock_updates(journal);
2698                 jbd2_journal_flush(journal);
2699
2700                 /* Journal blocked and flushed, clear needs_recovery flag. */
2701                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2702                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2703         }
2704 }
2705
2706 /*
2707  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2708  * flag here, even though the filesystem is not technically dirty yet.
2709  */
2710 static void ext4_unlockfs(struct super_block *sb)
2711 {
2712         if (!(sb->s_flags & MS_RDONLY)) {
2713                 lock_super(sb);
2714                 /* Reser the needs_recovery flag before the fs is unlocked. */
2715                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2716                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2717                 unlock_super(sb);
2718                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2719         }
2720 }
2721
2722 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2723 {
2724         struct ext4_super_block * es;
2725         struct ext4_sb_info *sbi = EXT4_SB(sb);
2726         ext4_fsblk_t n_blocks_count = 0;
2727         unsigned long old_sb_flags;
2728         struct ext4_mount_options old_opts;
2729         int err;
2730 #ifdef CONFIG_QUOTA
2731         int i;
2732 #endif
2733
2734         /* Store the original options */
2735         old_sb_flags = sb->s_flags;
2736         old_opts.s_mount_opt = sbi->s_mount_opt;
2737         old_opts.s_resuid = sbi->s_resuid;
2738         old_opts.s_resgid = sbi->s_resgid;
2739         old_opts.s_commit_interval = sbi->s_commit_interval;
2740 #ifdef CONFIG_QUOTA
2741         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2742         for (i = 0; i < MAXQUOTAS; i++)
2743                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2744 #endif
2745
2746         /*
2747          * Allow the "check" option to be passed as a remount option.
2748          */
2749         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2750                 err = -EINVAL;
2751                 goto restore_opts;
2752         }
2753
2754         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2755                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2756
2757         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2758                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2759
2760         es = sbi->s_es;
2761
2762         ext4_init_journal_params(sb, sbi->s_journal);
2763
2764         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2765                 n_blocks_count > ext4_blocks_count(es)) {
2766                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2767                         err = -EROFS;
2768                         goto restore_opts;
2769                 }
2770
2771                 if (*flags & MS_RDONLY) {
2772                         /*
2773                          * First of all, the unconditional stuff we have to do
2774                          * to disable replay of the journal when we next remount
2775                          */
2776                         sb->s_flags |= MS_RDONLY;
2777
2778                         /*
2779                          * OK, test if we are remounting a valid rw partition
2780                          * readonly, and if so set the rdonly flag and then
2781                          * mark the partition as valid again.
2782                          */
2783                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2784                             (sbi->s_mount_state & EXT4_VALID_FS))
2785                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2786
2787                         /*
2788                          * We have to unlock super so that we can wait for
2789                          * transactions.
2790                          */
2791                         unlock_super(sb);
2792                         ext4_mark_recovery_complete(sb, es);
2793                         lock_super(sb);
2794                 } else {
2795                         __le32 ret;
2796                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2797                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2798                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2799                                        "remount RDWR because of unsupported "
2800                                        "optional features (%x).\n",
2801                                        sb->s_id, le32_to_cpu(ret));
2802                                 err = -EROFS;
2803                                 goto restore_opts;
2804                         }
2805
2806                         /*
2807                          * If we have an unprocessed orphan list hanging
2808                          * around from a previously readonly bdev mount,
2809                          * require a full umount/remount for now.
2810                          */
2811                         if (es->s_last_orphan) {
2812                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2813                                        "remount RDWR because of unprocessed "
2814                                        "orphan inode list.  Please "
2815                                        "umount/remount instead.\n",
2816                                        sb->s_id);
2817                                 err = -EINVAL;
2818                                 goto restore_opts;
2819                         }
2820
2821                         /*
2822                          * Mounting a RDONLY partition read-write, so reread
2823                          * and store the current valid flag.  (It may have
2824                          * been changed by e2fsck since we originally mounted
2825                          * the partition.)
2826                          */
2827                         ext4_clear_journal_err(sb, es);
2828                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2829                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2830                                 goto restore_opts;
2831                         if (!ext4_setup_super (sb, es, 0))
2832                                 sb->s_flags &= ~MS_RDONLY;
2833                 }
2834         }
2835 #ifdef CONFIG_QUOTA
2836         /* Release old quota file names */
2837         for (i = 0; i < MAXQUOTAS; i++)
2838                 if (old_opts.s_qf_names[i] &&
2839                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2840                         kfree(old_opts.s_qf_names[i]);
2841 #endif
2842         return 0;
2843 restore_opts:
2844         sb->s_flags = old_sb_flags;
2845         sbi->s_mount_opt = old_opts.s_mount_opt;
2846         sbi->s_resuid = old_opts.s_resuid;
2847         sbi->s_resgid = old_opts.s_resgid;
2848         sbi->s_commit_interval = old_opts.s_commit_interval;
2849 #ifdef CONFIG_QUOTA
2850         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2851         for (i = 0; i < MAXQUOTAS; i++) {
2852                 if (sbi->s_qf_names[i] &&
2853                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2854                         kfree(sbi->s_qf_names[i]);
2855                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2856         }
2857 #endif
2858         return err;
2859 }
2860
2861 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2862 {
2863         struct super_block *sb = dentry->d_sb;
2864         struct ext4_sb_info *sbi = EXT4_SB(sb);
2865         struct ext4_super_block *es = sbi->s_es;
2866         u64 fsid;
2867
2868         if (test_opt(sb, MINIX_DF)) {
2869                 sbi->s_overhead_last = 0;
2870         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
2871                 ext4_group_t ngroups = sbi->s_groups_count, i;
2872                 ext4_fsblk_t overhead = 0;
2873                 smp_rmb();
2874
2875                 /*
2876                  * Compute the overhead (FS structures).  This is constant
2877                  * for a given filesystem unless the number of block groups
2878                  * changes so we cache the previous value until it does.
2879                  */
2880
2881                 /*
2882                  * All of the blocks before first_data_block are
2883                  * overhead
2884                  */
2885                 overhead = le32_to_cpu(es->s_first_data_block);
2886
2887                 /*
2888                  * Add the overhead attributed to the superblock and
2889                  * block group descriptors.  If the sparse superblocks
2890                  * feature is turned on, then not all groups have this.
2891                  */
2892                 for (i = 0; i < ngroups; i++) {
2893                         overhead += ext4_bg_has_super(sb, i) +
2894                                 ext4_bg_num_gdb(sb, i);
2895                         cond_resched();
2896                 }
2897
2898                 /*
2899                  * Every block group has an inode bitmap, a block
2900                  * bitmap, and an inode table.
2901                  */
2902                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2903                 sbi->s_overhead_last = overhead;
2904                 smp_wmb();
2905                 sbi->s_blocks_last = ext4_blocks_count(es);
2906         }
2907
2908         buf->f_type = EXT4_SUPER_MAGIC;
2909         buf->f_bsize = sb->s_blocksize;
2910         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2911         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2912         ext4_free_blocks_count_set(es, buf->f_bfree);
2913         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2914         if (buf->f_bfree < ext4_r_blocks_count(es))
2915                 buf->f_bavail = 0;
2916         buf->f_files = le32_to_cpu(es->s_inodes_count);
2917         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2918         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2919         buf->f_namelen = EXT4_NAME_LEN;
2920         fsid = le64_to_cpup((void *)es->s_uuid) ^
2921                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2922         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2923         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2924         return 0;
2925 }
2926
2927 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2928  * is locked for write. Otherwise the are possible deadlocks:
2929  * Process 1                         Process 2
2930  * ext4_create()                     quota_sync()
2931  *   jbd2_journal_start()                   write_dquot()
2932  *   DQUOT_INIT()                        down(dqio_mutex)
2933  *     down(dqio_mutex)                    jbd2_journal_start()
2934  *
2935  */
2936
2937 #ifdef CONFIG_QUOTA
2938
2939 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2940 {
2941         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2942 }
2943
2944 static int ext4_dquot_initialize(struct inode *inode, int type)
2945 {
2946         handle_t *handle;
2947         int ret, err;
2948
2949         /* We may create quota structure so we need to reserve enough blocks */
2950         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2951         if (IS_ERR(handle))
2952                 return PTR_ERR(handle);
2953         ret = dquot_initialize(inode, type);
2954         err = ext4_journal_stop(handle);
2955         if (!ret)
2956                 ret = err;
2957         return ret;
2958 }
2959
2960 static int ext4_dquot_drop(struct inode *inode)
2961 {
2962         handle_t *handle;
2963         int ret, err;
2964
2965         /* We may delete quota structure so we need to reserve enough blocks */
2966         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2967         if (IS_ERR(handle))
2968                 return PTR_ERR(handle);
2969         ret = dquot_drop(inode);
2970         err = ext4_journal_stop(handle);
2971         if (!ret)
2972                 ret = err;
2973         return ret;
2974 }
2975
2976 static int ext4_write_dquot(struct dquot *dquot)
2977 {
2978         int ret, err;
2979         handle_t *handle;
2980         struct inode *inode;
2981
2982         inode = dquot_to_inode(dquot);
2983         handle = ext4_journal_start(inode,
2984                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2985         if (IS_ERR(handle))
2986                 return PTR_ERR(handle);
2987         ret = dquot_commit(dquot);
2988         err = ext4_journal_stop(handle);
2989         if (!ret)
2990                 ret = err;
2991         return ret;
2992 }
2993
2994 static int ext4_acquire_dquot(struct dquot *dquot)
2995 {
2996         int ret, err;
2997         handle_t *handle;
2998
2999         handle = ext4_journal_start(dquot_to_inode(dquot),
3000                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3001         if (IS_ERR(handle))
3002                 return PTR_ERR(handle);
3003         ret = dquot_acquire(dquot);
3004         err = ext4_journal_stop(handle);
3005         if (!ret)
3006                 ret = err;
3007         return ret;
3008 }
3009
3010 static int ext4_release_dquot(struct dquot *dquot)
3011 {
3012         int ret, err;
3013         handle_t *handle;
3014
3015         handle = ext4_journal_start(dquot_to_inode(dquot),
3016                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3017         if (IS_ERR(handle)) {
3018                 /* Release dquot anyway to avoid endless cycle in dqput() */
3019                 dquot_release(dquot);
3020                 return PTR_ERR(handle);
3021         }
3022         ret = dquot_release(dquot);
3023         err = ext4_journal_stop(handle);
3024         if (!ret)
3025                 ret = err;
3026         return ret;
3027 }
3028
3029 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3030 {
3031         /* Are we journalling quotas? */
3032         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3033             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3034                 dquot_mark_dquot_dirty(dquot);
3035                 return ext4_write_dquot(dquot);
3036         } else {
3037                 return dquot_mark_dquot_dirty(dquot);
3038         }
3039 }
3040
3041 static int ext4_write_info(struct super_block *sb, int type)
3042 {
3043         int ret, err;
3044         handle_t *handle;
3045
3046         /* Data block + inode block */
3047         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3048         if (IS_ERR(handle))
3049                 return PTR_ERR(handle);
3050         ret = dquot_commit_info(sb, type);
3051         err = ext4_journal_stop(handle);
3052         if (!ret)
3053                 ret = err;
3054         return ret;
3055 }
3056
3057 /*
3058  * Turn on quotas during mount time - we need to find
3059  * the quota file and such...
3060  */
3061 static int ext4_quota_on_mount(struct super_block *sb, int type)
3062 {
3063         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3064                         EXT4_SB(sb)->s_jquota_fmt, type);
3065 }
3066
3067 /*
3068  * Standard function to be called on quota_on
3069  */
3070 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3071                          char *path)
3072 {
3073         int err;
3074         struct nameidata nd;
3075
3076         if (!test_opt(sb, QUOTA))
3077                 return -EINVAL;
3078         /* Not journalling quota? */
3079         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
3080             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
3081                 return vfs_quota_on(sb, type, format_id, path);
3082         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3083         if (err)
3084                 return err;
3085         /* Quotafile not on the same filesystem? */
3086         if (nd.mnt->mnt_sb != sb) {
3087                 path_release(&nd);
3088                 return -EXDEV;
3089         }
3090         /* Quotafile not of fs root? */
3091         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
3092                 printk(KERN_WARNING
3093                         "EXT4-fs: Quota file not on filesystem root. "
3094                         "Journalled quota will not work.\n");
3095         path_release(&nd);
3096         return vfs_quota_on(sb, type, format_id, path);
3097 }
3098
3099 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3100  * acquiring the locks... As quota files are never truncated and quota code
3101  * itself serializes the operations (and noone else should touch the files)
3102  * we don't have to be afraid of races */
3103 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3104                                size_t len, loff_t off)
3105 {
3106         struct inode *inode = sb_dqopt(sb)->files[type];
3107         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3108         int err = 0;
3109         int offset = off & (sb->s_blocksize - 1);
3110         int tocopy;
3111         size_t toread;
3112         struct buffer_head *bh;
3113         loff_t i_size = i_size_read(inode);
3114
3115         if (off > i_size)
3116                 return 0;
3117         if (off+len > i_size)
3118                 len = i_size-off;
3119         toread = len;
3120         while (toread > 0) {
3121                 tocopy = sb->s_blocksize - offset < toread ?
3122                                 sb->s_blocksize - offset : toread;
3123                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3124                 if (err)
3125                         return err;
3126                 if (!bh)        /* A hole? */
3127                         memset(data, 0, tocopy);
3128                 else
3129                         memcpy(data, bh->b_data+offset, tocopy);
3130                 brelse(bh);
3131                 offset = 0;
3132                 toread -= tocopy;
3133                 data += tocopy;
3134                 blk++;
3135         }
3136         return len;
3137 }
3138
3139 /* Write to quotafile (we know the transaction is already started and has
3140  * enough credits) */
3141 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3142                                 const char *data, size_t len, loff_t off)
3143 {
3144         struct inode *inode = sb_dqopt(sb)->files[type];
3145         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3146         int err = 0;
3147         int offset = off & (sb->s_blocksize - 1);
3148         int tocopy;
3149         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3150         size_t towrite = len;
3151         struct buffer_head *bh;
3152         handle_t *handle = journal_current_handle();
3153
3154         if (!handle) {
3155                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3156                         " cancelled because transaction is not started.\n",
3157                         (unsigned long long)off, (unsigned long long)len);
3158                 return -EIO;
3159         }
3160         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3161         while (towrite > 0) {
3162                 tocopy = sb->s_blocksize - offset < towrite ?
3163                                 sb->s_blocksize - offset : towrite;
3164                 bh = ext4_bread(handle, inode, blk, 1, &err);
3165                 if (!bh)
3166                         goto out;
3167                 if (journal_quota) {
3168                         err = ext4_journal_get_write_access(handle, bh);
3169                         if (err) {
3170                                 brelse(bh);
3171                                 goto out;
3172                         }
3173                 }
3174                 lock_buffer(bh);
3175                 memcpy(bh->b_data+offset, data, tocopy);
3176                 flush_dcache_page(bh->b_page);
3177                 unlock_buffer(bh);
3178                 if (journal_quota)
3179                         err = ext4_journal_dirty_metadata(handle, bh);
3180                 else {
3181                         /* Always do at least ordered writes for quotas */
3182                         err = ext4_journal_dirty_data(handle, bh);
3183                         mark_buffer_dirty(bh);
3184                 }
3185                 brelse(bh);
3186                 if (err)
3187                         goto out;
3188                 offset = 0;
3189                 towrite -= tocopy;
3190                 data += tocopy;
3191                 blk++;
3192         }
3193 out:
3194         if (len == towrite)
3195                 return err;
3196         if (inode->i_size < off+len-towrite) {
3197                 i_size_write(inode, off+len-towrite);
3198                 EXT4_I(inode)->i_disksize = inode->i_size;
3199         }
3200         inode->i_version++;
3201         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3202         ext4_mark_inode_dirty(handle, inode);
3203         mutex_unlock(&inode->i_mutex);
3204         return len - towrite;
3205 }
3206
3207 #endif
3208
3209 static int ext4_get_sb(struct file_system_type *fs_type,
3210         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3211 {
3212         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3213 }
3214
3215 static struct file_system_type ext4dev_fs_type = {
3216         .owner          = THIS_MODULE,
3217         .name           = "ext4dev",
3218         .get_sb         = ext4_get_sb,
3219         .kill_sb        = kill_block_super,
3220         .fs_flags       = FS_REQUIRES_DEV,
3221 };
3222
3223 static int __init init_ext4_fs(void)
3224 {
3225         int err = init_ext4_xattr();
3226         if (err)
3227                 return err;
3228         err = init_inodecache();
3229         if (err)
3230                 goto out1;
3231         err = register_filesystem(&ext4dev_fs_type);
3232         if (err)
3233                 goto out;
3234         return 0;
3235 out:
3236         destroy_inodecache();
3237 out1:
3238         exit_ext4_xattr();
3239         return err;
3240 }
3241
3242 static void __exit exit_ext4_fs(void)
3243 {
3244         unregister_filesystem(&ext4dev_fs_type);
3245         destroy_inodecache();
3246         exit_ext4_xattr();
3247 }
3248
3249 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3250 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3251 MODULE_LICENSE("GPL");
3252 module_init(init_ext4_fs)
3253 module_exit(exit_ext4_fs)