fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/log2.h>
  38 #include <linux/crc16.h>
  39 #include <asm/uaccess.h>
  40
  41 #include "ext4.h"
  42 #include "ext4_jbd2.h"
  43 #include "xattr.h"
  44 #include "acl.h"
  45 #include "namei.h"
  46 #include "group.h"
  47
  48 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  49                              unsigned long journal_devnum);
  50 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  51                                unsigned int);
  52 static void ext4_commit_super(struct super_block *sb,
  53                               struct ext4_super_block *es, int sync);
  54 static void ext4_mark_recovery_complete(struct super_block *sb,
  55                                         struct ext4_super_block *es);
  56 static void ext4_clear_journal_err(struct super_block *sb,
  57                                    struct ext4_super_block *es);
  58 static int ext4_sync_fs(struct super_block *sb, int wait);
  59 static const char *ext4_decode_error(struct super_block *sb, int errno,
  60                                      char nbuf[16]);
  61 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  62 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  63 static void ext4_unlockfs(struct super_block *sb);
  64 static void ext4_write_super(struct super_block *sb);
  65 static void ext4_write_super_lockfs(struct super_block *sb);
  66
  67
  68 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  69                                struct ext4_group_desc *bg)
  70 {
  71         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  72                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  73                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  74 }
  75
  76 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  77                                struct ext4_group_desc *bg)
  78 {
  79         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  80                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  81                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  82 }
  83
  84 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  85                               struct ext4_group_desc *bg)
  86 {
  87         return le32_to_cpu(bg->bg_inode_table_lo) |
  88                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  89                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  90 }
  91
  92 void ext4_block_bitmap_set(struct super_block *sb,
  93                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  94 {
  95         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
  96         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  97                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  98 }
  99
 100 void ext4_inode_bitmap_set(struct super_block *sb,
 101                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 102 {
 103         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 104         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 105                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 106 }
 107
 108 void ext4_inode_table_set(struct super_block *sb,
 109                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 110 {
 111         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 112         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 113                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 114 }
 115
 116 /*
 117  * Wrappers for jbd2_journal_start/end.
 118  *
 119  * The only special thing we need to do here is to make sure that all
 120  * journal_end calls result in the superblock being marked dirty, so
 121  * that sync() will call the filesystem's write_super callback if
 122  * appropriate.
 123  */
 124 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 125 {
 126         journal_t *journal;
 127
 128         if (sb->s_flags & MS_RDONLY)
 129                 return ERR_PTR(-EROFS);
 130
 131         /* Special case here: if the journal has aborted behind our
 132          * backs (eg. EIO in the commit thread), then we still need to
 133          * take the FS itself readonly cleanly. */
 134         journal = EXT4_SB(sb)->s_journal;
 135         if (is_journal_aborted(journal)) {
 136                 ext4_abort(sb, __func__,
 137                            "Detected aborted journal");
 138                 return ERR_PTR(-EROFS);
 139         }
 140
 141         return jbd2_journal_start(journal, nblocks);
 142 }
 143
 144 /*
 145  * The only special thing we need to do here is to make sure that all
 146  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 147  * that sync() will call the filesystem's write_super callback if
 148  * appropriate.
 149  */
 150 int __ext4_journal_stop(const char *where, handle_t *handle)
 151 {
 152         struct super_block *sb;
 153         int err;
 154         int rc;
 155
 156         sb = handle->h_transaction->t_journal->j_private;
 157         err = handle->h_err;
 158         rc = jbd2_journal_stop(handle);
 159
 160         if (!err)
 161                 err = rc;
 162         if (err)
 163                 __ext4_std_error(sb, where, err);
 164         return err;
 165 }
 166
 167 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 168                 struct buffer_head *bh, handle_t *handle, int err)
 169 {
 170         char nbuf[16];
 171         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 172
 173         if (bh)
 174                 BUFFER_TRACE(bh, "abort");
 175
 176         if (!handle->h_err)
 177                 handle->h_err = err;
 178
 179         if (is_handle_aborted(handle))
 180                 return;
 181
 182         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 183                caller, errstr, err_fn);
 184
 185         jbd2_journal_abort_handle(handle);
 186 }
 187
 188 /* Deal with the reporting of failure conditions on a filesystem such as
 189  * inconsistencies detected or read IO failures.
 190  *
 191  * On ext2, we can store the error state of the filesystem in the
 192  * superblock.  That is not possible on ext4, because we may have other
 193  * write ordering constraints on the superblock which prevent us from
 194  * writing it out straight away; and given that the journal is about to
 195  * be aborted, we can't rely on the current, or future, transactions to
 196  * write out the superblock safely.
 197  *
 198  * We'll just use the jbd2_journal_abort() error code to record an error in
 199  * the journal instead.  On recovery, the journal will compain about
 200  * that error until we've noted it down and cleared it.
 201  */
 202
 203 static void ext4_handle_error(struct super_block *sb)
 204 {
 205         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 206
 207         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 208         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 209
 210         if (sb->s_flags & MS_RDONLY)
 211                 return;
 212
 213         if (!test_opt(sb, ERRORS_CONT)) {
 214                 journal_t *journal = EXT4_SB(sb)->s_journal;
 215
 216                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 217                 if (journal)
 218                         jbd2_journal_abort(journal, -EIO);
 219         }
 220         if (test_opt(sb, ERRORS_RO)) {
 221                 printk(KERN_CRIT "Remounting filesystem read-only\n");
 222                 sb->s_flags |= MS_RDONLY;
 223         }
 224         ext4_commit_super(sb, es, 1);
 225         if (test_opt(sb, ERRORS_PANIC))
 226                 panic("EXT4-fs (device %s): panic forced after error\n",
 227                         sb->s_id);
 228 }
 229
 230 void ext4_error(struct super_block *sb, const char *function,
 231                 const char *fmt, ...)
 232 {
 233         va_list args;
 234
 235         va_start(args, fmt);
 236         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 237         vprintk(fmt, args);
 238         printk("\n");
 239         va_end(args);
 240
 241         ext4_handle_error(sb);
 242 }
 243
 244 static const char *ext4_decode_error(struct super_block *sb, int errno,
 245                                      char nbuf[16])
 246 {
 247         char *errstr = NULL;
 248
 249         switch (errno) {
 250         case -EIO:
 251                 errstr = "IO failure";
 252                 break;
 253         case -ENOMEM:
 254                 errstr = "Out of memory";
 255                 break;
 256         case -EROFS:
 257                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 258                         errstr = "Journal has aborted";
 259                 else
 260                         errstr = "Readonly filesystem";
 261                 break;
 262         default:
 263                 /* If the caller passed in an extra buffer for unknown
 264                  * errors, textualise them now.  Else we just return
 265                  * NULL. */
 266                 if (nbuf) {
 267                         /* Check for truncated error codes... */
 268                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 269                                 errstr = nbuf;
 270                 }
 271                 break;
 272         }
 273
 274         return errstr;
 275 }
 276
 277 /* __ext4_std_error decodes expected errors from journaling functions
 278  * automatically and invokes the appropriate error response.  */
 279
 280 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 281 {
 282         char nbuf[16];
 283         const char *errstr;
 284
 285         /* Special case: if the error is EROFS, and we're not already
 286          * inside a transaction, then there's really no point in logging
 287          * an error. */
 288         if (errno == -EROFS && journal_current_handle() == NULL &&
 289             (sb->s_flags & MS_RDONLY))
 290                 return;
 291
 292         errstr = ext4_decode_error(sb, errno, nbuf);
 293         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 294                sb->s_id, function, errstr);
 295
 296         ext4_handle_error(sb);
 297 }
 298
 299 /*
 300  * ext4_abort is a much stronger failure handler than ext4_error.  The
 301  * abort function may be used to deal with unrecoverable failures such
 302  * as journal IO errors or ENOMEM at a critical moment in log management.
 303  *
 304  * We unconditionally force the filesystem into an ABORT|READONLY state,
 305  * unless the error response on the fs has been set to panic in which
 306  * case we take the easy way out and panic immediately.
 307  */
 308
 309 void ext4_abort(struct super_block *sb, const char *function,
 310                 const char *fmt, ...)
 311 {
 312         va_list args;
 313
 314         printk(KERN_CRIT "ext4_abort called.\n");
 315
 316         va_start(args, fmt);
 317         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 318         vprintk(fmt, args);
 319         printk("\n");
 320         va_end(args);
 321
 322         if (test_opt(sb, ERRORS_PANIC))
 323                 panic("EXT4-fs panic from previous error\n");
 324
 325         if (sb->s_flags & MS_RDONLY)
 326                 return;
 327
 328         printk(KERN_CRIT "Remounting filesystem read-only\n");
 329         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 330         sb->s_flags |= MS_RDONLY;
 331         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 332         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 333 }
 334
 335 void ext4_warning(struct super_block *sb, const char *function,
 336                   const char *fmt, ...)
 337 {
 338         va_list args;
 339
 340         va_start(args, fmt);
 341         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 342                sb->s_id, function);
 343         vprintk(fmt, args);
 344         printk("\n");
 345         va_end(args);
 346 }
 347
 348 void ext4_update_dynamic_rev(struct super_block *sb)
 349 {
 350         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 351
 352         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 353                 return;
 354
 355         ext4_warning(sb, __func__,
 356                      "updating to rev %d because of new feature flag, "
 357                      "running e2fsck is recommended",
 358                      EXT4_DYNAMIC_REV);
 359
 360         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 361         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 362         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 363         /* leave es->s_feature_*compat flags alone */
 364         /* es->s_uuid will be set by e2fsck if empty */
 365
 366         /*
 367          * The rest of the superblock fields should be zero, and if not it
 368          * means they are likely already in use, so leave them alone.  We
 369          * can leave it up to e2fsck to clean up any inconsistencies there.
 370          */
 371 }
 372
 373 int ext4_update_compat_feature(handle_t *handle,
 374                                         struct super_block *sb, __u32 compat)
 375 {
 376         int err = 0;
 377         if (!EXT4_HAS_COMPAT_FEATURE(sb, compat)) {
 378                 err = ext4_journal_get_write_access(handle,
 379                                 EXT4_SB(sb)->s_sbh);
 380                 if (err)
 381                         return err;
 382                 EXT4_SET_COMPAT_FEATURE(sb, compat);
 383                 sb->s_dirt = 1;
 384                 handle->h_sync = 1;
 385                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 386                                         "call ext4_journal_dirty_met adata");
 387                 err = ext4_journal_dirty_metadata(handle,
 388                                 EXT4_SB(sb)->s_sbh);
 389         }
 390         return err;
 391 }
 392
 393 int ext4_update_rocompat_feature(handle_t *handle,
 394                                         struct super_block *sb, __u32 rocompat)
 395 {
 396         int err = 0;
 397         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, rocompat)) {
 398                 err = ext4_journal_get_write_access(handle,
 399                                 EXT4_SB(sb)->s_sbh);
 400                 if (err)
 401                         return err;
 402                 EXT4_SET_RO_COMPAT_FEATURE(sb, rocompat);
 403                 sb->s_dirt = 1;
 404                 handle->h_sync = 1;
 405                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 406                                         "call ext4_journal_dirty_met adata");
 407                 err = ext4_journal_dirty_metadata(handle,
 408                                 EXT4_SB(sb)->s_sbh);
 409         }
 410         return err;
 411 }
 412
 413 int ext4_update_incompat_feature(handle_t *handle,
 414                                         struct super_block *sb, __u32 incompat)
 415 {
 416         int err = 0;
 417         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, incompat)) {
 418                 err = ext4_journal_get_write_access(handle,
 419                                 EXT4_SB(sb)->s_sbh);
 420                 if (err)
 421                         return err;
 422                 EXT4_SET_INCOMPAT_FEATURE(sb, incompat);
 423                 sb->s_dirt = 1;
 424                 handle->h_sync = 1;
 425                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh,
 426                                         "call ext4_journal_dirty_met adata");
 427                 err = ext4_journal_dirty_metadata(handle,
 428                                 EXT4_SB(sb)->s_sbh);
 429         }
 430         return err;
 431 }
 432
 433 /*
 434  * Open the external journal device
 435  */
 436 static struct block_device *ext4_blkdev_get(dev_t dev)
 437 {
 438         struct block_device *bdev;
 439         char b[BDEVNAME_SIZE];
 440
 441         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 442         if (IS_ERR(bdev))
 443                 goto fail;
 444         return bdev;
 445
 446 fail:
 447         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 448                         __bdevname(dev, b), PTR_ERR(bdev));
 449         return NULL;
 450 }
 451
 452 /*
 453  * Release the journal device
 454  */
 455 static int ext4_blkdev_put(struct block_device *bdev)
 456 {
 457         bd_release(bdev);
 458         return blkdev_put(bdev);
 459 }
 460
 461 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 462 {
 463         struct block_device *bdev;
 464         int ret = -ENODEV;
 465
 466         bdev = sbi->journal_bdev;
 467         if (bdev) {
 468                 ret = ext4_blkdev_put(bdev);
 469                 sbi->journal_bdev = NULL;
 470         }
 471         return ret;
 472 }
 473
 474 static inline struct inode *orphan_list_entry(struct list_head *l)
 475 {
 476         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 477 }
 478
 479 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 480 {
 481         struct list_head *l;
 482
 483         printk(KERN_ERR "sb orphan head is %d\n",
 484                le32_to_cpu(sbi->s_es->s_last_orphan));
 485
 486         printk(KERN_ERR "sb_info orphan list:\n");
 487         list_for_each(l, &sbi->s_orphan) {
 488                 struct inode *inode = orphan_list_entry(l);
 489                 printk(KERN_ERR "  "
 490                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 491                        inode->i_sb->s_id, inode->i_ino, inode,
 492                        inode->i_mode, inode->i_nlink,
 493                        NEXT_ORPHAN(inode));
 494         }
 495 }
 496
 497 static void ext4_put_super(struct super_block *sb)
 498 {
 499         struct ext4_sb_info *sbi = EXT4_SB(sb);
 500         struct ext4_super_block *es = sbi->s_es;
 501         int i;
 502
 503         ext4_mb_release(sb);
 504         ext4_ext_release(sb);
 505         ext4_xattr_put_super(sb);
 506         jbd2_journal_destroy(sbi->s_journal);
 507         sbi->s_journal = NULL;
 508         if (!(sb->s_flags & MS_RDONLY)) {
 509                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 510                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 511                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 512                 mark_buffer_dirty(sbi->s_sbh);
 513                 ext4_commit_super(sb, es, 1);
 514         }
 515
 516         for (i = 0; i < sbi->s_gdb_count; i++)
 517                 brelse(sbi->s_group_desc[i]);
 518         kfree(sbi->s_group_desc);
 519         kfree(sbi->s_flex_groups);
 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         ei->vfs_inode.i_data.writeback_index = 0;
 572         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 573         INIT_LIST_HEAD(&ei->i_prealloc_list);
 574         spin_lock_init(&ei->i_prealloc_lock);
 575         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 576         ei->i_reserved_data_blocks = 0;
 577         ei->i_reserved_meta_blocks = 0;
 578         ei->i_allocated_meta_blocks = 0;
 579         ei->i_delalloc_reserved_flag = 0;
 580         spin_lock_init(&(ei->i_block_reservation_lock));
 581         return &ei->vfs_inode;
 582 }
 583
 584 static void ext4_destroy_inode(struct inode *inode)
 585 {
 586         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 587                 printk("EXT4 Inode %p: orphan list check failed!\n",
 588                         EXT4_I(inode));
 589                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 590                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 591                                 true);
 592                 dump_stack();
 593         }
 594         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 595 }
 596
 597 static void init_once(void *foo)
 598 {
 599         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 600
 601         INIT_LIST_HEAD(&ei->i_orphan);
 602 #ifdef CONFIG_EXT4DEV_FS_XATTR
 603         init_rwsem(&ei->xattr_sem);
 604 #endif
 605         init_rwsem(&ei->i_data_sem);
 606         inode_init_once(&ei->vfs_inode);
 607 }
 608
 609 static int init_inodecache(void)
 610 {
 611         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 612                                              sizeof(struct ext4_inode_info),
 613                                              0, (SLAB_RECLAIM_ACCOUNT|
 614                                                 SLAB_MEM_SPREAD),
 615                                              init_once);
 616         if (ext4_inode_cachep == NULL)
 617                 return -ENOMEM;
 618         return 0;
 619 }
 620
 621 static void destroy_inodecache(void)
 622 {
 623         kmem_cache_destroy(ext4_inode_cachep);
 624 }
 625
 626 static void ext4_clear_inode(struct inode *inode)
 627 {
 628         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 629 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 630         if (EXT4_I(inode)->i_acl &&
 631                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 632                 posix_acl_release(EXT4_I(inode)->i_acl);
 633                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 634         }
 635         if (EXT4_I(inode)->i_default_acl &&
 636                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 637                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 638                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 639         }
 640 #endif
 641         ext4_discard_reservation(inode);
 642         EXT4_I(inode)->i_block_alloc_info = NULL;
 643         if (unlikely(rsv))
 644                 kfree(rsv);
 645         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 646                                        &EXT4_I(inode)->jinode);
 647 }
 648
 649 static inline void ext4_show_quota_options(struct seq_file *seq,
 650                                            struct super_block *sb)
 651 {
 652 #if defined(CONFIG_QUOTA)
 653         struct ext4_sb_info *sbi = EXT4_SB(sb);
 654
 655         if (sbi->s_jquota_fmt)
 656                 seq_printf(seq, ",jqfmt=%s",
 657                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 658
 659         if (sbi->s_qf_names[USRQUOTA])
 660                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 661
 662         if (sbi->s_qf_names[GRPQUOTA])
 663                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 664
 665         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 666                 seq_puts(seq, ",usrquota");
 667
 668         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 669                 seq_puts(seq, ",grpquota");
 670 #endif
 671 }
 672
 673 /*
 674  * Show an option if
 675  *  - it's set to a non-default value OR
 676  *  - if the per-sb default is different from the global default
 677  */
 678 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 679 {
 680         int def_errors;
 681         unsigned long def_mount_opts;
 682         struct super_block *sb = vfs->mnt_sb;
 683         struct ext4_sb_info *sbi = EXT4_SB(sb);
 684         struct ext4_super_block *es = sbi->s_es;
 685
 686         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 687         def_errors     = le16_to_cpu(es->s_errors);
 688
 689         if (sbi->s_sb_block != 1)
 690                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 691         if (test_opt(sb, MINIX_DF))
 692                 seq_puts(seq, ",minixdf");
 693         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 694                 seq_puts(seq, ",grpid");
 695         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 696                 seq_puts(seq, ",nogrpid");
 697         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 698             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 699                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 700         }
 701         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 702             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 703                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 704         }
 705         if (test_opt(sb, ERRORS_RO)) {
 706                 if (def_errors == EXT4_ERRORS_PANIC ||
 707                     def_errors == EXT4_ERRORS_CONTINUE) {
 708                         seq_puts(seq, ",errors=remount-ro");
 709                 }
 710         }
 711         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 712                 seq_puts(seq, ",errors=continue");
 713         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 714                 seq_puts(seq, ",errors=panic");
 715         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 716                 seq_puts(seq, ",nouid32");
 717         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 718                 seq_puts(seq, ",debug");
 719         if (test_opt(sb, OLDALLOC))
 720                 seq_puts(seq, ",oldalloc");
 721 #ifdef CONFIG_EXT4DEV_FS_XATTR
 722         if (test_opt(sb, XATTR_USER) &&
 723                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 724                 seq_puts(seq, ",user_xattr");
 725         if (!test_opt(sb, XATTR_USER) &&
 726             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 727                 seq_puts(seq, ",nouser_xattr");
 728         }
 729 #endif
 730 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 731         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 732                 seq_puts(seq, ",acl");
 733         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 734                 seq_puts(seq, ",noacl");
 735 #endif
 736         if (!test_opt(sb, RESERVATION))
 737                 seq_puts(seq, ",noreservation");
 738         if (sbi->s_commit_interval) {
 739                 seq_printf(seq, ",commit=%u",
 740                            (unsigned) (sbi->s_commit_interval / HZ));
 741         }
 742         /*
 743          * We're changing the default of barrier mount option, so
 744          * let's always display its mount state so it's clear what its
 745          * status is.
 746          */
 747         seq_puts(seq, ",barrier=");
 748         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 749         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 750                 seq_puts(seq, ",journal_async_commit");
 751         if (test_opt(sb, NOBH))
 752                 seq_puts(seq, ",nobh");
 753         if (!test_opt(sb, EXTENTS))
 754                 seq_puts(seq, ",noextents");
 755         if (!test_opt(sb, MBALLOC))
 756                 seq_puts(seq, ",nomballoc");
 757         if (test_opt(sb, I_VERSION))
 758                 seq_puts(seq, ",i_version");
 759         if (!test_opt(sb, DELALLOC))
 760                 seq_puts(seq, ",nodelalloc");
 761
 762
 763         if (sbi->s_stripe)
 764                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 765         /*
 766          * journal mode get enabled in different ways
 767          * So just print the value even if we didn't specify it
 768          */
 769         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 770                 seq_puts(seq, ",data=journal");
 771         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 772                 seq_puts(seq, ",data=ordered");
 773         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 774                 seq_puts(seq, ",data=writeback");
 775
 776         ext4_show_quota_options(seq, sb);
 777         return 0;
 778 }
 779
 780
 781 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 782                 u64 ino, u32 generation)
 783 {
 784         struct inode *inode;
 785
 786         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 787                 return ERR_PTR(-ESTALE);
 788         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 789                 return ERR_PTR(-ESTALE);
 790
 791         /* iget isn't really right if the inode is currently unallocated!!
 792          *
 793          * ext4_read_inode will return a bad_inode if the inode had been
 794          * deleted, so we should be safe.
 795          *
 796          * Currently we don't know the generation for parent directory, so
 797          * a generation of 0 means "accept any"
 798          */
 799         inode = ext4_iget(sb, ino);
 800         if (IS_ERR(inode))
 801                 return ERR_CAST(inode);
 802         if (generation && inode->i_generation != generation) {
 803                 iput(inode);
 804                 return ERR_PTR(-ESTALE);
 805         }
 806
 807         return inode;
 808 }
 809
 810 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 811                 int fh_len, int fh_type)
 812 {
 813         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 814                                     ext4_nfs_get_inode);
 815 }
 816
 817 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 818                 int fh_len, int fh_type)
 819 {
 820         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 821                                     ext4_nfs_get_inode);
 822 }
 823
 824 #ifdef CONFIG_QUOTA
 825 #define QTYPE2NAME(t) ((t) == USRQUOTA?"user":"group")
 826 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 827
 828 static int ext4_dquot_initialize(struct inode *inode, int type);
 829 static int ext4_dquot_drop(struct inode *inode);
 830 static int ext4_write_dquot(struct dquot *dquot);
 831 static int ext4_acquire_dquot(struct dquot *dquot);
 832 static int ext4_release_dquot(struct dquot *dquot);
 833 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 834 static int ext4_write_info(struct super_block *sb, int type);
 835 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 836                                 char *path, int remount);
 837 static int ext4_quota_on_mount(struct super_block *sb, int type);
 838 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 839                                size_t len, loff_t off);
 840 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 841                                 const char *data, size_t len, loff_t off);
 842
 843 static struct dquot_operations ext4_quota_operations = {
 844         .initialize     = ext4_dquot_initialize,
 845         .drop           = ext4_dquot_drop,
 846         .alloc_space    = dquot_alloc_space,
 847         .alloc_inode    = dquot_alloc_inode,
 848         .free_space     = dquot_free_space,
 849         .free_inode     = dquot_free_inode,
 850         .transfer       = dquot_transfer,
 851         .write_dquot    = ext4_write_dquot,
 852         .acquire_dquot  = ext4_acquire_dquot,
 853         .release_dquot  = ext4_release_dquot,
 854         .mark_dirty     = ext4_mark_dquot_dirty,
 855         .write_info     = ext4_write_info
 856 };
 857
 858 static struct quotactl_ops ext4_qctl_operations = {
 859         .quota_on       = ext4_quota_on,
 860         .quota_off      = vfs_quota_off,
 861         .quota_sync     = vfs_quota_sync,
 862         .get_info       = vfs_get_dqinfo,
 863         .set_info       = vfs_set_dqinfo,
 864         .get_dqblk      = vfs_get_dqblk,
 865         .set_dqblk      = vfs_set_dqblk
 866 };
 867 #endif
 868
 869 static const struct super_operations ext4_sops = {
 870         .alloc_inode    = ext4_alloc_inode,
 871         .destroy_inode  = ext4_destroy_inode,
 872         .write_inode    = ext4_write_inode,
 873         .dirty_inode    = ext4_dirty_inode,
 874         .delete_inode   = ext4_delete_inode,
 875         .put_super      = ext4_put_super,
 876         .write_super    = ext4_write_super,
 877         .sync_fs        = ext4_sync_fs,
 878         .write_super_lockfs = ext4_write_super_lockfs,
 879         .unlockfs       = ext4_unlockfs,
 880         .statfs         = ext4_statfs,
 881         .remount_fs     = ext4_remount,
 882         .clear_inode    = ext4_clear_inode,
 883         .show_options   = ext4_show_options,
 884 #ifdef CONFIG_QUOTA
 885         .quota_read     = ext4_quota_read,
 886         .quota_write    = ext4_quota_write,
 887 #endif
 888 };
 889
 890 static const struct export_operations ext4_export_ops = {
 891         .fh_to_dentry = ext4_fh_to_dentry,
 892         .fh_to_parent = ext4_fh_to_parent,
 893         .get_parent = ext4_get_parent,
 894 };
 895
 896 enum {
 897         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 898         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 899         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 900         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 901         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 902         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 903         Opt_journal_checksum, Opt_journal_async_commit,
 904         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 905         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 906         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 907         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 908         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 909         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 910 };
 911
 912 static match_table_t tokens = {
 913         {Opt_bsd_df, "bsddf"},
 914         {Opt_minix_df, "minixdf"},
 915         {Opt_grpid, "grpid"},
 916         {Opt_grpid, "bsdgroups"},
 917         {Opt_nogrpid, "nogrpid"},
 918         {Opt_nogrpid, "sysvgroups"},
 919         {Opt_resgid, "resgid=%u"},
 920         {Opt_resuid, "resuid=%u"},
 921         {Opt_sb, "sb=%u"},
 922         {Opt_err_cont, "errors=continue"},
 923         {Opt_err_panic, "errors=panic"},
 924         {Opt_err_ro, "errors=remount-ro"},
 925         {Opt_nouid32, "nouid32"},
 926         {Opt_nocheck, "nocheck"},
 927         {Opt_nocheck, "check=none"},
 928         {Opt_debug, "debug"},
 929         {Opt_oldalloc, "oldalloc"},
 930         {Opt_orlov, "orlov"},
 931         {Opt_user_xattr, "user_xattr"},
 932         {Opt_nouser_xattr, "nouser_xattr"},
 933         {Opt_acl, "acl"},
 934         {Opt_noacl, "noacl"},
 935         {Opt_reservation, "reservation"},
 936         {Opt_noreservation, "noreservation"},
 937         {Opt_noload, "noload"},
 938         {Opt_nobh, "nobh"},
 939         {Opt_bh, "bh"},
 940         {Opt_commit, "commit=%u"},
 941         {Opt_journal_update, "journal=update"},
 942         {Opt_journal_inum, "journal=%u"},
 943         {Opt_journal_dev, "journal_dev=%u"},
 944         {Opt_journal_checksum, "journal_checksum"},
 945         {Opt_journal_async_commit, "journal_async_commit"},
 946         {Opt_abort, "abort"},
 947         {Opt_data_journal, "data=journal"},
 948         {Opt_data_ordered, "data=ordered"},
 949         {Opt_data_writeback, "data=writeback"},
 950         {Opt_offusrjquota, "usrjquota="},
 951         {Opt_usrjquota, "usrjquota=%s"},
 952         {Opt_offgrpjquota, "grpjquota="},
 953         {Opt_grpjquota, "grpjquota=%s"},
 954         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 955         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 956         {Opt_grpquota, "grpquota"},
 957         {Opt_noquota, "noquota"},
 958         {Opt_quota, "quota"},
 959         {Opt_usrquota, "usrquota"},
 960         {Opt_barrier, "barrier=%u"},
 961         {Opt_extents, "extents"},
 962         {Opt_noextents, "noextents"},
 963         {Opt_i_version, "i_version"},
 964         {Opt_mballoc, "mballoc"},
 965         {Opt_nomballoc, "nomballoc"},
 966         {Opt_stripe, "stripe=%u"},
 967         {Opt_resize, "resize"},
 968         {Opt_delalloc, "delalloc"},
 969         {Opt_nodelalloc, "nodelalloc"},
 970         {Opt_err, NULL},
 971 };
 972
 973 static ext4_fsblk_t get_sb_block(void **data)
 974 {
 975         ext4_fsblk_t    sb_block;
 976         char            *options = (char *) *data;
 977
 978         if (!options || strncmp(options, "sb=", 3) != 0)
 979                 return 1;       /* Default location */
 980         options += 3;
 981         /*todo: use simple_strtoll with >32bit ext4 */
 982         sb_block = simple_strtoul(options, &options, 0);
 983         if (*options && *options != ',') {
 984                 printk("EXT4-fs: Invalid sb specification: %s\n",
 985                        (char *) *data);
 986                 return 1;
 987         }
 988         if (*options == ',')
 989                 options++;
 990         *data = (void *) options;
 991         return sb_block;
 992 }
 993
 994 static int parse_options(char *options, struct super_block *sb,
 995                          unsigned int *inum, unsigned long *journal_devnum,
 996                          ext4_fsblk_t *n_blocks_count, int is_remount)
 997 {
 998         struct ext4_sb_info *sbi = EXT4_SB(sb);
 999         char *p;
1000         substring_t args[MAX_OPT_ARGS];
1001         int data_opt = 0;
1002         int option;
1003 #ifdef CONFIG_QUOTA
1004         int qtype, qfmt;
1005         char *qname;
1006 #endif
1007         ext4_fsblk_t last_block;
1008
1009         if (!options)
1010                 return 1;
1011
1012         while ((p = strsep(&options, ",")) != NULL) {
1013                 int token;
1014                 if (!*p)
1015                         continue;
1016
1017                 token = match_token(p, tokens, args);
1018                 switch (token) {
1019                 case Opt_bsd_df:
1020                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1021                         break;
1022                 case Opt_minix_df:
1023                         set_opt(sbi->s_mount_opt, MINIX_DF);
1024                         break;
1025                 case Opt_grpid:
1026                         set_opt(sbi->s_mount_opt, GRPID);
1027                         break;
1028                 case Opt_nogrpid:
1029                         clear_opt(sbi->s_mount_opt, GRPID);
1030                         break;
1031                 case Opt_resuid:
1032                         if (match_int(&args[0], &option))
1033                                 return 0;
1034                         sbi->s_resuid = option;
1035                         break;
1036                 case Opt_resgid:
1037                         if (match_int(&args[0], &option))
1038                                 return 0;
1039                         sbi->s_resgid = option;
1040                         break;
1041                 case Opt_sb:
1042                         /* handled by get_sb_block() instead of here */
1043                         /* *sb_block = match_int(&args[0]); */
1044                         break;
1045                 case Opt_err_panic:
1046                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1047                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1048                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1049                         break;
1050                 case Opt_err_ro:
1051                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1052                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1053                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1054                         break;
1055                 case Opt_err_cont:
1056                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1057                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1058                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1059                         break;
1060                 case Opt_nouid32:
1061                         set_opt(sbi->s_mount_opt, NO_UID32);
1062                         break;
1063                 case Opt_nocheck:
1064                         clear_opt(sbi->s_mount_opt, CHECK);
1065                         break;
1066                 case Opt_debug:
1067                         set_opt(sbi->s_mount_opt, DEBUG);
1068                         break;
1069                 case Opt_oldalloc:
1070                         set_opt(sbi->s_mount_opt, OLDALLOC);
1071                         break;
1072                 case Opt_orlov:
1073                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1074                         break;
1075 #ifdef CONFIG_EXT4DEV_FS_XATTR
1076                 case Opt_user_xattr:
1077                         set_opt(sbi->s_mount_opt, XATTR_USER);
1078                         break;
1079                 case Opt_nouser_xattr:
1080                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1081                         break;
1082 #else
1083                 case Opt_user_xattr:
1084                 case Opt_nouser_xattr:
1085                         printk("EXT4 (no)user_xattr options not supported\n");
1086                         break;
1087 #endif
1088 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1089                 case Opt_acl:
1090                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1091                         break;
1092                 case Opt_noacl:
1093                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1094                         break;
1095 #else
1096                 case Opt_acl:
1097                 case Opt_noacl:
1098                         printk("EXT4 (no)acl options not supported\n");
1099                         break;
1100 #endif
1101                 case Opt_reservation:
1102                         set_opt(sbi->s_mount_opt, RESERVATION);
1103                         break;
1104                 case Opt_noreservation:
1105                         clear_opt(sbi->s_mount_opt, RESERVATION);
1106                         break;
1107                 case Opt_journal_update:
1108                         /* @@@ FIXME */
1109                         /* Eventually we will want to be able to create
1110                            a journal file here.  For now, only allow the
1111                            user to specify an existing inode to be the
1112                            journal file. */
1113                         if (is_remount) {
1114                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1115                                        "journal on remount\n");
1116                                 return 0;
1117                         }
1118                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1119                         break;
1120                 case Opt_journal_inum:
1121                         if (is_remount) {
1122                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1123                                        "journal on remount\n");
1124                                 return 0;
1125                         }
1126                         if (match_int(&args[0], &option))
1127                                 return 0;
1128                         *inum = option;
1129                         break;
1130                 case Opt_journal_dev:
1131                         if (is_remount) {
1132                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1133                                        "journal on remount\n");
1134                                 return 0;
1135                         }
1136                         if (match_int(&args[0], &option))
1137                                 return 0;
1138                         *journal_devnum = option;
1139                         break;
1140                 case Opt_journal_checksum:
1141                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1142                         break;
1143                 case Opt_journal_async_commit:
1144                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1145                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1146                         break;
1147                 case Opt_noload:
1148                         set_opt(sbi->s_mount_opt, NOLOAD);
1149                         break;
1150                 case Opt_commit:
1151                         if (match_int(&args[0], &option))
1152                                 return 0;
1153                         if (option < 0)
1154                                 return 0;
1155                         if (option == 0)
1156                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1157                         sbi->s_commit_interval = HZ * option;
1158                         break;
1159                 case Opt_data_journal:
1160                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1161                         goto datacheck;
1162                 case Opt_data_ordered:
1163                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1164                         goto datacheck;
1165                 case Opt_data_writeback:
1166                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1167                 datacheck:
1168                         if (is_remount) {
1169                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1170                                                 != data_opt) {
1171                                         printk(KERN_ERR
1172                                                 "EXT4-fs: cannot change data "
1173                                                 "mode on remount\n");
1174                                         return 0;
1175                                 }
1176                         } else {
1177                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1178                                 sbi->s_mount_opt |= data_opt;
1179                         }
1180                         break;
1181 #ifdef CONFIG_QUOTA
1182                 case Opt_usrjquota:
1183                         qtype = USRQUOTA;
1184                         goto set_qf_name;
1185                 case Opt_grpjquota:
1186                         qtype = GRPQUOTA;
1187 set_qf_name:
1188                         if ((sb_any_quota_enabled(sb) ||
1189                              sb_any_quota_suspended(sb)) &&
1190                             !sbi->s_qf_names[qtype]) {
1191                                 printk(KERN_ERR
1192                                         "EXT4-fs: Cannot change journaled "
1193                                         "quota options when quota turned on.\n");
1194                                 return 0;
1195                         }
1196                         qname = match_strdup(&args[0]);
1197                         if (!qname) {
1198                                 printk(KERN_ERR
1199                                         "EXT4-fs: not enough memory for "
1200                                         "storing quotafile name.\n");
1201                                 return 0;
1202                         }
1203                         if (sbi->s_qf_names[qtype] &&
1204                             strcmp(sbi->s_qf_names[qtype], qname)) {
1205                                 printk(KERN_ERR
1206                                         "EXT4-fs: %s quota file already "
1207                                         "specified.\n", QTYPE2NAME(qtype));
1208                                 kfree(qname);
1209                                 return 0;
1210                         }
1211                         sbi->s_qf_names[qtype] = qname;
1212                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1213                                 printk(KERN_ERR
1214                                         "EXT4-fs: quotafile must be on "
1215                                         "filesystem root.\n");
1216                                 kfree(sbi->s_qf_names[qtype]);
1217                                 sbi->s_qf_names[qtype] = NULL;
1218                                 return 0;
1219                         }
1220                         set_opt(sbi->s_mount_opt, QUOTA);
1221                         break;
1222                 case Opt_offusrjquota:
1223                         qtype = USRQUOTA;
1224                         goto clear_qf_name;
1225                 case Opt_offgrpjquota:
1226                         qtype = GRPQUOTA;
1227 clear_qf_name:
1228                         if ((sb_any_quota_enabled(sb) ||
1229                              sb_any_quota_suspended(sb)) &&
1230                             sbi->s_qf_names[qtype]) {
1231                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1232                                         "journaled quota options when "
1233                                         "quota turned on.\n");
1234                                 return 0;
1235                         }
1236                         /*
1237                          * The space will be released later when all options
1238                          * are confirmed to be correct
1239                          */
1240                         sbi->s_qf_names[qtype] = NULL;
1241                         break;
1242                 case Opt_jqfmt_vfsold:
1243                         qfmt = QFMT_VFS_OLD;
1244                         goto set_qf_format;
1245                 case Opt_jqfmt_vfsv0:
1246                         qfmt = QFMT_VFS_V0;
1247 set_qf_format:
1248                         if ((sb_any_quota_enabled(sb) ||
1249                              sb_any_quota_suspended(sb)) &&
1250                             sbi->s_jquota_fmt != qfmt) {
1251                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1252                                         "journaled quota options when "
1253                                         "quota turned on.\n");
1254                                 return 0;
1255                         }
1256                         sbi->s_jquota_fmt = qfmt;
1257                         break;
1258                 case Opt_quota:
1259                 case Opt_usrquota:
1260                         set_opt(sbi->s_mount_opt, QUOTA);
1261                         set_opt(sbi->s_mount_opt, USRQUOTA);
1262                         break;
1263                 case Opt_grpquota:
1264                         set_opt(sbi->s_mount_opt, QUOTA);
1265                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1266                         break;
1267                 case Opt_noquota:
1268                         if (sb_any_quota_enabled(sb)) {
1269                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1270                                         "options when quota turned on.\n");
1271                                 return 0;
1272                         }
1273                         clear_opt(sbi->s_mount_opt, QUOTA);
1274                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1275                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1276                         break;
1277 #else
1278                 case Opt_quota:
1279                 case Opt_usrquota:
1280                 case Opt_grpquota:
1281                         printk(KERN_ERR
1282                                 "EXT4-fs: quota options not supported.\n");
1283                         break;
1284                 case Opt_usrjquota:
1285                 case Opt_grpjquota:
1286                 case Opt_offusrjquota:
1287                 case Opt_offgrpjquota:
1288                 case Opt_jqfmt_vfsold:
1289                 case Opt_jqfmt_vfsv0:
1290                         printk(KERN_ERR
1291                                 "EXT4-fs: journaled quota options not "
1292                                 "supported.\n");
1293                         break;
1294                 case Opt_noquota:
1295                         break;
1296 #endif
1297                 case Opt_abort:
1298                         set_opt(sbi->s_mount_opt, ABORT);
1299                         break;
1300                 case Opt_barrier:
1301                         if (match_int(&args[0], &option))
1302                                 return 0;
1303                         if (option)
1304                                 set_opt(sbi->s_mount_opt, BARRIER);
1305                         else
1306                                 clear_opt(sbi->s_mount_opt, BARRIER);
1307                         break;
1308                 case Opt_ignore:
1309                         break;
1310                 case Opt_resize:
1311                         if (!is_remount) {
1312                                 printk("EXT4-fs: resize option only available "
1313                                         "for remount\n");
1314                                 return 0;
1315                         }
1316                         if (match_int(&args[0], &option) != 0)
1317                                 return 0;
1318                         *n_blocks_count = option;
1319                         break;
1320                 case Opt_nobh:
1321                         set_opt(sbi->s_mount_opt, NOBH);
1322                         break;
1323                 case Opt_bh:
1324                         clear_opt(sbi->s_mount_opt, NOBH);
1325                         break;
1326                 case Opt_extents:
1327                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1328                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1329                                 ext4_warning(sb, __func__,
1330                                         "extents feature not enabled "
1331                                         "on this filesystem, use tune2fs\n");
1332                                 return 0;
1333                         }
1334                         set_opt(sbi->s_mount_opt, EXTENTS);
1335                         break;
1336                 case Opt_noextents:
1337                         /*
1338                          * When e2fsprogs support resizing an already existing
1339                          * ext3 file system to greater than 2**32 we need to
1340                          * add support to block allocator to handle growing
1341                          * already existing block  mapped inode so that blocks
1342                          * allocated for them fall within 2**32
1343                          */
1344                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1345                         if (last_block  > 0xffffffffULL) {
1346                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1347                                                 "large to mount with "
1348                                                 "-o noextents options\n");
1349                                 return 0;
1350                         }
1351                         clear_opt(sbi->s_mount_opt, EXTENTS);
1352                         break;
1353                 case Opt_i_version:
1354                         set_opt(sbi->s_mount_opt, I_VERSION);
1355                         sb->s_flags |= MS_I_VERSION;
1356                         break;
1357                 case Opt_nodelalloc:
1358                         clear_opt(sbi->s_mount_opt, DELALLOC);
1359                         break;
1360                 case Opt_mballoc:
1361                         set_opt(sbi->s_mount_opt, MBALLOC);
1362                         break;
1363                 case Opt_nomballoc:
1364                         clear_opt(sbi->s_mount_opt, MBALLOC);
1365                         break;
1366                 case Opt_stripe:
1367                         if (match_int(&args[0], &option))
1368                                 return 0;
1369                         if (option < 0)
1370                                 return 0;
1371                         sbi->s_stripe = option;
1372                         break;
1373                 case Opt_delalloc:
1374                         set_opt(sbi->s_mount_opt, DELALLOC);
1375                         break;
1376                 default:
1377                         printk(KERN_ERR
1378                                "EXT4-fs: Unrecognized mount option \"%s\" "
1379                                "or missing value\n", p);
1380                         return 0;
1381                 }
1382         }
1383 #ifdef CONFIG_QUOTA
1384         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1385                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1386                      sbi->s_qf_names[USRQUOTA])
1387                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1388
1389                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1390                      sbi->s_qf_names[GRPQUOTA])
1391                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1392
1393                 if ((sbi->s_qf_names[USRQUOTA] &&
1394                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1395                     (sbi->s_qf_names[GRPQUOTA] &&
1396                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1397                         printk(KERN_ERR "EXT4-fs: old and new quota "
1398                                         "format mixing.\n");
1399                         return 0;
1400                 }
1401
1402                 if (!sbi->s_jquota_fmt) {
1403                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1404                                         "not specified.\n");
1405                         return 0;
1406                 }
1407         } else {
1408                 if (sbi->s_jquota_fmt) {
1409                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1410                                         "specified with no journaling "
1411                                         "enabled.\n");
1412                         return 0;
1413                 }
1414         }
1415 #endif
1416         return 1;
1417 }
1418
1419 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1420                             int read_only)
1421 {
1422         struct ext4_sb_info *sbi = EXT4_SB(sb);
1423         int res = 0;
1424
1425         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1426                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1427                        "forcing read-only mode\n");
1428                 res = MS_RDONLY;
1429         }
1430         if (read_only)
1431                 return res;
1432         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1433                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1434                        "running e2fsck is recommended\n");
1435         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1436                 printk(KERN_WARNING
1437                        "EXT4-fs warning: mounting fs with errors, "
1438                        "running e2fsck is recommended\n");
1439         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1440                  le16_to_cpu(es->s_mnt_count) >=
1441                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1442                 printk(KERN_WARNING
1443                        "EXT4-fs warning: maximal mount count reached, "
1444                        "running e2fsck is recommended\n");
1445         else if (le32_to_cpu(es->s_checkinterval) &&
1446                 (le32_to_cpu(es->s_lastcheck) +
1447                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1448                 printk(KERN_WARNING
1449                        "EXT4-fs warning: checktime reached, "
1450                        "running e2fsck is recommended\n");
1451 #if 0
1452                 /* @@@ We _will_ want to clear the valid bit if we find
1453                  * inconsistencies, to force a fsck at reboot.  But for
1454                  * a plain journaled filesystem we can keep it set as
1455                  * valid forever! :)
1456                  */
1457         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1458 #endif
1459         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1460                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1461         le16_add_cpu(&es->s_mnt_count, 1);
1462         es->s_mtime = cpu_to_le32(get_seconds());
1463         ext4_update_dynamic_rev(sb);
1464         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1465
1466         ext4_commit_super(sb, es, 1);
1467         if (test_opt(sb, DEBUG))
1468                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1469                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1470                         sb->s_blocksize,
1471                         sbi->s_groups_count,
1472                         EXT4_BLOCKS_PER_GROUP(sb),
1473                         EXT4_INODES_PER_GROUP(sb),
1474                         sbi->s_mount_opt);
1475
1476         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1477         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1478                 char b[BDEVNAME_SIZE];
1479
1480                 printk("external journal on %s\n",
1481                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1482         } else {
1483                 printk("internal journal\n");
1484         }
1485         return res;
1486 }
1487
1488 static int ext4_fill_flex_info(struct super_block *sb)
1489 {
1490         struct ext4_sb_info *sbi = EXT4_SB(sb);
1491         struct ext4_group_desc *gdp = NULL;
1492         struct buffer_head *bh;
1493         ext4_group_t flex_group_count;
1494         ext4_group_t flex_group;
1495         int groups_per_flex = 0;
1496         __u64 block_bitmap = 0;
1497         int i;
1498
1499         if (!sbi->s_es->s_log_groups_per_flex) {
1500                 sbi->s_log_groups_per_flex = 0;
1501                 return 1;
1502         }
1503
1504         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1505         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1506
1507         flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
1508                 groups_per_flex;
1509         sbi->s_flex_groups = kzalloc(flex_group_count *
1510                                      sizeof(struct flex_groups), GFP_KERNEL);
1511         if (sbi->s_flex_groups == NULL) {
1512                 printk(KERN_ERR "EXT4-fs: not enough memory for "
1513                                 "%lu flex groups\n", flex_group_count);
1514                 goto failed;
1515         }
1516
1517         gdp = ext4_get_group_desc(sb, 1, &bh);
1518         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1519
1520         for (i = 0; i < sbi->s_groups_count; i++) {
1521                 gdp = ext4_get_group_desc(sb, i, &bh);
1522
1523                 flex_group = ext4_flex_group(sbi, i);
1524                 sbi->s_flex_groups[flex_group].free_inodes +=
1525                         le16_to_cpu(gdp->bg_free_inodes_count);
1526                 sbi->s_flex_groups[flex_group].free_blocks +=
1527                         le16_to_cpu(gdp->bg_free_blocks_count);
1528         }
1529
1530         return 1;
1531 failed:
1532         return 0;
1533 }
1534
1535 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1536                             struct ext4_group_desc *gdp)
1537 {
1538         __u16 crc = 0;
1539
1540         if (sbi->s_es->s_feature_ro_compat &
1541             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1542                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1543                 __le32 le_group = cpu_to_le32(block_group);
1544
1545                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1546                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1547                 crc = crc16(crc, (__u8 *)gdp, offset);
1548                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1549                 /* for checksum of struct ext4_group_desc do the rest...*/
1550                 if ((sbi->s_es->s_feature_incompat &
1551                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1552                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1553                         crc = crc16(crc, (__u8 *)gdp + offset,
1554                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1555                                         offset);
1556         }
1557
1558         return cpu_to_le16(crc);
1559 }
1560
1561 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1562                                 struct ext4_group_desc *gdp)
1563 {
1564         if ((sbi->s_es->s_feature_ro_compat &
1565              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1566             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1567                 return 0;
1568
1569         return 1;
1570 }
1571
1572 /* Called at mount-time, super-block is locked */
1573 static int ext4_check_descriptors(struct super_block *sb)
1574 {
1575         struct ext4_sb_info *sbi = EXT4_SB(sb);
1576         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1577         ext4_fsblk_t last_block;
1578         ext4_fsblk_t block_bitmap;
1579         ext4_fsblk_t inode_bitmap;
1580         ext4_fsblk_t inode_table;
1581         int flexbg_flag = 0;
1582         ext4_group_t i;
1583
1584         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1585                 flexbg_flag = 1;
1586
1587         ext4_debug ("Checking group descriptors");
1588
1589         for (i = 0; i < sbi->s_groups_count; i++) {
1590                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1591
1592                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1593                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1594                 else
1595                         last_block = first_block +
1596                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1597
1598                 block_bitmap = ext4_block_bitmap(sb, gdp);
1599                 if (block_bitmap < first_block || block_bitmap > last_block) {
1600                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1601                                "Block bitmap for group %lu not in group "
1602                                "(block %llu)!", i, block_bitmap);
1603                         return 0;
1604                 }
1605                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1606                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1607                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1608                                "Inode bitmap for group %lu not in group "
1609                                "(block %llu)!", i, inode_bitmap);
1610                         return 0;
1611                 }
1612                 inode_table = ext4_inode_table(sb, gdp);
1613                 if (inode_table < first_block ||
1614                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1615                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1616                                "Inode table for group %lu not in group "
1617                                "(block %llu)!", i, inode_table);
1618                         return 0;
1619                 }
1620                 spin_lock(sb_bgl_lock(sbi, i));
1621                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1622                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1623                                "Checksum for group %lu failed (%u!=%u)\n",
1624                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1625                                gdp)), le16_to_cpu(gdp->bg_checksum));
1626                         if (!(sb->s_flags & MS_RDONLY))
1627                                 return 0;
1628                 }
1629                 spin_unlock(sb_bgl_lock(sbi, i));
1630                 if (!flexbg_flag)
1631                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1632         }
1633
1634         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1635         sbi->s_es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
1636         return 1;
1637 }
1638
1639 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1640  * the superblock) which were deleted from all directories, but held open by
1641  * a process at the time of a crash.  We walk the list and try to delete these
1642  * inodes at recovery time (only with a read-write filesystem).
1643  *
1644  * In order to keep the orphan inode chain consistent during traversal (in
1645  * case of crash during recovery), we link each inode into the superblock
1646  * orphan list_head and handle it the same way as an inode deletion during
1647  * normal operation (which journals the operations for us).
1648  *
1649  * We only do an iget() and an iput() on each inode, which is very safe if we
1650  * accidentally point at an in-use or already deleted inode.  The worst that
1651  * can happen in this case is that we get a "bit already cleared" message from
1652  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1653  * e2fsck was run on this filesystem, and it must have already done the orphan
1654  * inode cleanup for us, so we can safely abort without any further action.
1655  */
1656 static void ext4_orphan_cleanup(struct super_block *sb,
1657                                 struct ext4_super_block *es)
1658 {
1659         unsigned int s_flags = sb->s_flags;
1660         int nr_orphans = 0, nr_truncates = 0;
1661 #ifdef CONFIG_QUOTA
1662         int i;
1663 #endif
1664         if (!es->s_last_orphan) {
1665                 jbd_debug(4, "no orphan inodes to clean up\n");
1666                 return;
1667         }
1668
1669         if (bdev_read_only(sb->s_bdev)) {
1670                 printk(KERN_ERR "EXT4-fs: write access "
1671                         "unavailable, skipping orphan cleanup.\n");
1672                 return;
1673         }
1674
1675         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1676                 if (es->s_last_orphan)
1677                         jbd_debug(1, "Errors on filesystem, "
1678                                   "clearing orphan list.\n");
1679                 es->s_last_orphan = 0;
1680                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1681                 return;
1682         }
1683
1684         if (s_flags & MS_RDONLY) {
1685                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1686                        sb->s_id);
1687                 sb->s_flags &= ~MS_RDONLY;
1688         }
1689 #ifdef CONFIG_QUOTA
1690         /* Needed for iput() to work correctly and not trash data */
1691         sb->s_flags |= MS_ACTIVE;
1692         /* Turn on quotas so that they are updated correctly */
1693         for (i = 0; i < MAXQUOTAS; i++) {
1694                 if (EXT4_SB(sb)->s_qf_names[i]) {
1695                         int ret = ext4_quota_on_mount(sb, i);
1696                         if (ret < 0)
1697                                 printk(KERN_ERR
1698                                         "EXT4-fs: Cannot turn on journaled "
1699                                         "quota: error %d\n", ret);
1700                 }
1701         }
1702 #endif
1703
1704         while (es->s_last_orphan) {
1705                 struct inode *inode;
1706
1707                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1708                 if (IS_ERR(inode)) {
1709                         es->s_last_orphan = 0;
1710                         break;
1711                 }
1712
1713                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1714                 DQUOT_INIT(inode);
1715                 if (inode->i_nlink) {
1716                         printk(KERN_DEBUG
1717                                 "%s: truncating inode %lu to %Ld bytes\n",
1718                                 __func__, inode->i_ino, inode->i_size);
1719                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1720                                   inode->i_ino, inode->i_size);
1721                         ext4_truncate(inode);
1722                         nr_truncates++;
1723                 } else {
1724                         printk(KERN_DEBUG
1725                                 "%s: deleting unreferenced inode %lu\n",
1726                                 __func__, inode->i_ino);
1727                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1728                                   inode->i_ino);
1729                         nr_orphans++;
1730                 }
1731                 iput(inode);  /* The delete magic happens here! */
1732         }
1733
1734 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1735
1736         if (nr_orphans)
1737                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1738                        sb->s_id, PLURAL(nr_orphans));
1739         if (nr_truncates)
1740                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1741                        sb->s_id, PLURAL(nr_truncates));
1742 #ifdef CONFIG_QUOTA
1743         /* Turn quotas off */
1744         for (i = 0; i < MAXQUOTAS; i++) {
1745                 if (sb_dqopt(sb)->files[i])
1746                         vfs_quota_off(sb, i, 0);
1747         }
1748 #endif
1749         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1750 }
1751 /*
1752  * Maximal extent format file size.
1753  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1754  * extent format containers, within a sector_t, and within i_blocks
1755  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1756  * so that won't be a limiting factor.
1757  *
1758  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1759  */
1760 static loff_t ext4_max_size(int blkbits)
1761 {
1762         loff_t res;
1763         loff_t upper_limit = MAX_LFS_FILESIZE;
1764
1765         /* small i_blocks in vfs inode? */
1766         if (sizeof(blkcnt_t) < sizeof(u64)) {
1767                 /*
1768                  * CONFIG_LSF is not enabled implies the inode
1769                  * i_block represent total blocks in 512 bytes
1770                  * 32 == size of vfs inode i_blocks * 8
1771                  */
1772                 upper_limit = (1LL << 32) - 1;
1773
1774                 /* total blocks in file system block size */
1775                 upper_limit >>= (blkbits - 9);
1776                 upper_limit <<= blkbits;
1777         }
1778
1779         /* 32-bit extent-start container, ee_block */
1780         res = 1LL << 32;
1781         res <<= blkbits;
1782         res -= 1;
1783
1784         /* Sanity check against vm- & vfs- imposed limits */
1785         if (res > upper_limit)
1786                 res = upper_limit;
1787
1788         return res;
1789 }
1790
1791 /*
1792  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1793  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1794  * We need to be 1 filesystem block less than the 2^48 sector limit.
1795  */
1796 static loff_t ext4_max_bitmap_size(int bits)
1797 {
1798         loff_t res = EXT4_NDIR_BLOCKS;
1799         int meta_blocks;
1800         loff_t upper_limit;
1801         /* This is calculated to be the largest file size for a
1802          * dense, bitmapped file such that the total number of
1803          * sectors in the file, including data and all indirect blocks,
1804          * does not exceed 2^48 -1
1805          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1806          * total number of  512 bytes blocks of the file
1807          */
1808
1809         if (sizeof(blkcnt_t) < sizeof(u64)) {
1810                 /*
1811                  * CONFIG_LSF is not enabled implies the inode
1812                  * i_block represent total blocks in 512 bytes
1813                  * 32 == size of vfs inode i_blocks * 8
1814                  */
1815                 upper_limit = (1LL << 32) - 1;
1816
1817                 /* total blocks in file system block size */
1818                 upper_limit >>= (bits - 9);
1819
1820         } else {
1821                 /*
1822                  * We use 48 bit ext4_inode i_blocks
1823                  * With EXT4_HUGE_FILE_FL set the i_blocks
1824                  * represent total number of blocks in
1825                  * file system block size
1826                  */
1827                 upper_limit = (1LL << 48) - 1;
1828
1829         }
1830
1831         /* indirect blocks */
1832         meta_blocks = 1;
1833         /* double indirect blocks */
1834         meta_blocks += 1 + (1LL << (bits-2));
1835         /* tripple indirect blocks */
1836         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1837
1838         upper_limit -= meta_blocks;
1839         upper_limit <<= bits;
1840
1841         res += 1LL << (bits-2);
1842         res += 1LL << (2*(bits-2));
1843         res += 1LL << (3*(bits-2));
1844         res <<= bits;
1845         if (res > upper_limit)
1846                 res = upper_limit;
1847
1848         if (res > MAX_LFS_FILESIZE)
1849                 res = MAX_LFS_FILESIZE;
1850
1851         return res;
1852 }
1853
1854 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1855                                 ext4_fsblk_t logical_sb_block, int nr)
1856 {
1857         struct ext4_sb_info *sbi = EXT4_SB(sb);
1858         ext4_group_t bg, first_meta_bg;
1859         int has_super = 0;
1860
1861         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1862
1863         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1864             nr < first_meta_bg)
1865                 return logical_sb_block + nr + 1;
1866         bg = sbi->s_desc_per_block * nr;
1867         if (ext4_bg_has_super(sb, bg))
1868                 has_super = 1;
1869         return (has_super + ext4_group_first_block_no(sb, bg));
1870 }
1871
1872 /**
1873  * ext4_get_stripe_size: Get the stripe size.
1874  * @sbi: In memory super block info
1875  *
1876  * If we have specified it via mount option, then
1877  * use the mount option value. If the value specified at mount time is
1878  * greater than the blocks per group use the super block value.
1879  * If the super block value is greater than blocks per group return 0.
1880  * Allocator needs it be less than blocks per group.
1881  *
1882  */
1883 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1884 {
1885         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1886         unsigned long stripe_width =
1887                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1888
1889         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1890                 return sbi->s_stripe;
1891
1892         if (stripe_width <= sbi->s_blocks_per_group)
1893                 return stripe_width;
1894
1895         if (stride <= sbi->s_blocks_per_group)
1896                 return stride;
1897
1898         return 0;
1899 }
1900
1901 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
1902                                 __releases(kernel_lock)
1903                                 __acquires(kernel_lock)
1904
1905 {
1906         struct buffer_head *bh;
1907         struct ext4_super_block *es = NULL;
1908         struct ext4_sb_info *sbi;
1909         ext4_fsblk_t block;
1910         ext4_fsblk_t sb_block = get_sb_block(&data);
1911         ext4_fsblk_t logical_sb_block;
1912         unsigned long offset = 0;
1913         unsigned int journal_inum = 0;
1914         unsigned long journal_devnum = 0;
1915         unsigned long def_mount_opts;
1916         struct inode *root;
1917         int ret = -EINVAL;
1918         int blocksize;
1919         int db_count;
1920         int i;
1921         int needs_recovery;
1922         __le32 features;
1923         __u64 blocks_count;
1924         int err;
1925
1926         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1927         if (!sbi)
1928                 return -ENOMEM;
1929         sb->s_fs_info = sbi;
1930         sbi->s_mount_opt = 0;
1931         sbi->s_resuid = EXT4_DEF_RESUID;
1932         sbi->s_resgid = EXT4_DEF_RESGID;
1933         sbi->s_sb_block = sb_block;
1934
1935         unlock_kernel();
1936
1937         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1938         if (!blocksize) {
1939                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1940                 goto out_fail;
1941         }
1942
1943         /*
1944          * The ext4 superblock will not be buffer aligned for other than 1kB
1945          * block sizes.  We need to calculate the offset from buffer start.
1946          */
1947         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1948                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1949                 offset = do_div(logical_sb_block, blocksize);
1950         } else {
1951                 logical_sb_block = sb_block;
1952         }
1953
1954         if (!(bh = sb_bread(sb, logical_sb_block))) {
1955                 printk(KERN_ERR "EXT4-fs: unable to read superblock\n");
1956                 goto out_fail;
1957         }
1958         /*
1959          * Note: s_es must be initialized as soon as possible because
1960          *       some ext4 macro-instructions depend on its value
1961          */
1962         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1963         sbi->s_es = es;
1964         sb->s_magic = le16_to_cpu(es->s_magic);
1965         if (sb->s_magic != EXT4_SUPER_MAGIC)
1966                 goto cantfind_ext4;
1967
1968         /* Set defaults before we parse the mount options */
1969         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1970         if (def_mount_opts & EXT4_DEFM_DEBUG)
1971                 set_opt(sbi->s_mount_opt, DEBUG);
1972         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1973                 set_opt(sbi->s_mount_opt, GRPID);
1974         if (def_mount_opts & EXT4_DEFM_UID16)
1975                 set_opt(sbi->s_mount_opt, NO_UID32);
1976 #ifdef CONFIG_EXT4DEV_FS_XATTR
1977         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1978                 set_opt(sbi->s_mount_opt, XATTR_USER);
1979 #endif
1980 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1981         if (def_mount_opts & EXT4_DEFM_ACL)
1982                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1983 #endif
1984         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1985                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1986         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1987                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1988         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1989                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1990
1991         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1992                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1993         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1994                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1995         else
1996                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1997
1998         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1999         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2000
2001         set_opt(sbi->s_mount_opt, RESERVATION);
2002         set_opt(sbi->s_mount_opt, BARRIER);
2003
2004         /*
2005          * turn on extents feature by default in ext4 filesystem
2006          * only if feature flag already set by mkfs or tune2fs.
2007          * Use -o noextents to turn it off
2008          */
2009         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2010                 set_opt(sbi->s_mount_opt, EXTENTS);
2011         else
2012                 ext4_warning(sb, __func__,
2013                         "extents feature not enabled on this filesystem, "
2014                         "use tune2fs.\n");
2015         /*
2016          * turn on mballoc code by default in ext4 filesystem
2017          * Use -o nomballoc to turn it off
2018          */
2019         set_opt(sbi->s_mount_opt, MBALLOC);
2020
2021         /*
2022          * enable delayed allocation by default
2023          * Use -o nodelalloc to turn it off
2024          */
2025         set_opt(sbi->s_mount_opt, DELALLOC);
2026
2027
2028         if (!parse_options((char *) data, sb, &journal_inum, &journal_devnum,
2029                            NULL, 0))
2030                 goto failed_mount;
2031
2032         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2033                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2034
2035         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2036             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2037              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2038              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2039                 printk(KERN_WARNING
2040                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2041                        "running e2fsck is recommended\n");
2042
2043         /*
2044          * Since ext4 is still considered development code, we require
2045          * that the TEST_FILESYS flag in s->flags be set.
2046          */
2047         if (!(le32_to_cpu(es->s_flags) & EXT2_FLAGS_TEST_FILESYS)) {
2048                 printk(KERN_WARNING "EXT4-fs: %s: not marked "
2049                        "OK to use with test code.\n", sb->s_id);
2050                 goto failed_mount;
2051         }
2052
2053         /*
2054          * Check feature flags regardless of the revision level, since we
2055          * previously didn't change the revision level when setting the flags,
2056          * so there is a chance incompat flags are set on a rev 0 filesystem.
2057          */
2058         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2059         if (features) {
2060                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2061                        "unsupported optional features (%x).\n",
2062                        sb->s_id, le32_to_cpu(features));
2063                 goto failed_mount;
2064         }
2065         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2066         if (!(sb->s_flags & MS_RDONLY) && features) {
2067                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2068                        "unsupported optional features (%x).\n",
2069                        sb->s_id, le32_to_cpu(features));
2070                 goto failed_mount;
2071         }
2072         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2073                 /*
2074                  * Large file size enabled file system can only be
2075                  * mount if kernel is build with CONFIG_LSF
2076                  */
2077                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2078                                 !(sb->s_flags & MS_RDONLY)) {
2079                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2080                                         "files cannot be mounted read-write "
2081                                         "without CONFIG_LSF.\n", sb->s_id);
2082                         goto failed_mount;
2083                 }
2084         }
2085         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2086
2087         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2088             blocksize > EXT4_MAX_BLOCK_SIZE) {
2089                 printk(KERN_ERR
2090                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2091                        blocksize, sb->s_id);
2092                 goto failed_mount;
2093         }
2094
2095         if (sb->s_blocksize != blocksize) {
2096
2097                 /* Validate the filesystem blocksize */
2098                 if (!sb_set_blocksize(sb, blocksize)) {
2099                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2100                                         blocksize);
2101                         goto failed_mount;
2102                 }
2103
2104                 brelse(bh);
2105                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2106                 offset = do_div(logical_sb_block, blocksize);
2107                 bh = sb_bread(sb, logical_sb_block);
2108                 if (!bh) {
2109                         printk(KERN_ERR
2110                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2111                         goto failed_mount;
2112                 }
2113                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2114                 sbi->s_es = es;
2115                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2116                         printk(KERN_ERR
2117                                "EXT4-fs: Magic mismatch, very weird !\n");
2118                         goto failed_mount;
2119                 }
2120         }
2121
2122         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits);
2123         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
2124
2125         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2126                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2127                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2128         } else {
2129                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2130                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2131                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2132                     (!is_power_of_2(sbi->s_inode_size)) ||
2133                     (sbi->s_inode_size > blocksize)) {
2134                         printk(KERN_ERR
2135                                "EXT4-fs: unsupported inode size: %d\n",
2136                                sbi->s_inode_size);
2137                         goto failed_mount;
2138                 }
2139                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2140                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2141         }
2142         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2143         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2144                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2145                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2146                     !is_power_of_2(sbi->s_desc_size)) {
2147                         printk(KERN_ERR
2148                                "EXT4-fs: unsupported descriptor size %lu\n",
2149                                sbi->s_desc_size);
2150                         goto failed_mount;
2151                 }
2152         } else
2153                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2154         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2155         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2156         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2157                 goto cantfind_ext4;
2158         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2159         if (sbi->s_inodes_per_block == 0)
2160                 goto cantfind_ext4;
2161         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2162                                         sbi->s_inodes_per_block;
2163         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2164         sbi->s_sbh = bh;
2165         sbi->s_mount_state = le16_to_cpu(es->s_state);
2166         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2167         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2168         for (i = 0; i < 4; i++)
2169                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2170         sbi->s_def_hash_version = es->s_def_hash_version;
2171
2172         if (sbi->s_blocks_per_group > blocksize * 8) {
2173                 printk(KERN_ERR
2174                        "EXT4-fs: #blocks per group too big: %lu\n",
2175                        sbi->s_blocks_per_group);
2176                 goto failed_mount;
2177         }
2178         if (sbi->s_inodes_per_group > blocksize * 8) {
2179                 printk(KERN_ERR
2180                        "EXT4-fs: #inodes per group too big: %lu\n",
2181                        sbi->s_inodes_per_group);
2182                 goto failed_mount;
2183         }
2184
2185         if (ext4_blocks_count(es) >
2186                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2187                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2188                         " too large to mount safely\n", sb->s_id);
2189                 if (sizeof(sector_t) < 8)
2190                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2191                                         "enabled\n");
2192                 goto failed_mount;
2193         }
2194
2195         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2196                 goto cantfind_ext4;
2197
2198         /* ensure blocks_count calculation below doesn't sign-extend */
2199         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2200             le32_to_cpu(es->s_first_data_block) + 1) {
2201                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2202                        "first data block %u, blocks per group %lu\n",
2203                         ext4_blocks_count(es),
2204                         le32_to_cpu(es->s_first_data_block),
2205                         EXT4_BLOCKS_PER_GROUP(sb));
2206                 goto failed_mount;
2207         }
2208         blocks_count = (ext4_blocks_count(es) -
2209                         le32_to_cpu(es->s_first_data_block) +
2210                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2211         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2212         sbi->s_groups_count = blocks_count;
2213         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2214                    EXT4_DESC_PER_BLOCK(sb);
2215         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2216                                     GFP_KERNEL);
2217         if (sbi->s_group_desc == NULL) {
2218                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
2219                 goto failed_mount;
2220         }
2221
2222         bgl_lock_init(&sbi->s_blockgroup_lock);
2223
2224         for (i = 0; i < db_count; i++) {
2225                 block = descriptor_loc(sb, logical_sb_block, i);
2226                 sbi->s_group_desc[i] = sb_bread(sb, block);
2227                 if (!sbi->s_group_desc[i]) {
2228                         printk(KERN_ERR "EXT4-fs: "
2229                                "can't read group descriptor %d\n", i);
2230                         db_count = i;
2231                         goto failed_mount2;
2232                 }
2233         }
2234         if (!ext4_check_descriptors(sb)) {
2235                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2236                 goto failed_mount2;
2237         }
2238         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2239                 if (!ext4_fill_flex_info(sb)) {
2240                         printk(KERN_ERR
2241                                "EXT4-fs: unable to initialize "
2242                                "flex_bg meta info!\n");
2243                         goto failed_mount2;
2244                 }
2245
2246         sbi->s_gdb_count = db_count;
2247         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2248         spin_lock_init(&sbi->s_next_gen_lock);
2249
2250         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2251                         ext4_count_free_blocks(sb));
2252         if (!err) {
2253                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2254                                 ext4_count_free_inodes(sb));
2255         }
2256         if (!err) {
2257                 err = percpu_counter_init(&sbi->s_dirs_counter,
2258                                 ext4_count_dirs(sb));
2259         }
2260         if (err) {
2261                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2262                 goto failed_mount3;
2263         }
2264
2265         /* per fileystem reservation list head & lock */
2266         spin_lock_init(&sbi->s_rsv_window_lock);
2267         sbi->s_rsv_window_root = RB_ROOT;
2268         /* Add a single, static dummy reservation to the start of the
2269          * reservation window list --- it gives us a placeholder for
2270          * append-at-start-of-list which makes the allocation logic
2271          * _much_ simpler. */
2272         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2273         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2274         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2275         sbi->s_rsv_window_head.rsv_goal_size = 0;
2276         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2277
2278         sbi->s_stripe = ext4_get_stripe_size(sbi);
2279
2280         /*
2281          * set up enough so that it can read an inode
2282          */
2283         sb->s_op = &ext4_sops;
2284         sb->s_export_op = &ext4_export_ops;
2285         sb->s_xattr = ext4_xattr_handlers;
2286 #ifdef CONFIG_QUOTA
2287         sb->s_qcop = &ext4_qctl_operations;
2288         sb->dq_op = &ext4_quota_operations;
2289 #endif
2290         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2291
2292         sb->s_root = NULL;
2293
2294         needs_recovery = (es->s_last_orphan != 0 ||
2295                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2296                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2297
2298         /*
2299          * The first inode we look at is the journal inode.  Don't try
2300          * root first: it may be modified in the journal!
2301          */
2302         if (!test_opt(sb, NOLOAD) &&
2303             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2304                 if (ext4_load_journal(sb, es, journal_devnum))
2305                         goto failed_mount3;
2306                 if (!(sb->s_flags & MS_RDONLY) &&
2307                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2308                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2309                                "ext4_fill_super: Journal transaction "
2310                                "%u is corrupt\n", sb->s_id,
2311                                EXT4_SB(sb)->s_journal->j_failed_commit);
2312                         if (test_opt(sb, ERRORS_RO)) {
2313                                 printk(KERN_CRIT
2314                                        "Mounting filesystem read-only\n");
2315                                 sb->s_flags |= MS_RDONLY;
2316                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2317                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2318                         }
2319                         if (test_opt(sb, ERRORS_PANIC)) {
2320                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2321                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2322                                 ext4_commit_super(sb, es, 1);
2323                                 printk(KERN_CRIT
2324                                        "EXT4-fs (device %s): mount failed\n",
2325                                       sb->s_id);
2326                                 goto failed_mount4;
2327                         }
2328                 }
2329         } else if (journal_inum) {
2330                 if (ext4_create_journal(sb, es, journal_inum))
2331                         goto failed_mount3;
2332         } else {
2333                 if (!silent)
2334                         printk(KERN_ERR
2335                                "ext4: No journal on filesystem on %s\n",
2336                                sb->s_id);
2337                 goto failed_mount3;
2338         }
2339
2340         if (ext4_blocks_count(es) > 0xffffffffULL &&
2341             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2342                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2343                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2344                 goto failed_mount4;
2345         }
2346
2347         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2348                 jbd2_journal_set_features(sbi->s_journal,
2349                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2350                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2351         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2352                 jbd2_journal_set_features(sbi->s_journal,
2353                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2354                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2355                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2356         } else {
2357                 jbd2_journal_clear_features(sbi->s_journal,
2358                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2359                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2360         }
2361
2362         /* We have now updated the journal if required, so we can
2363          * validate the data journaling mode. */
2364         switch (test_opt(sb, DATA_FLAGS)) {
2365         case 0:
2366                 /* No mode set, assume a default based on the journal
2367                  * capabilities: ORDERED_DATA if the journal can
2368                  * cope, else JOURNAL_DATA
2369                  */
2370                 if (jbd2_journal_check_available_features
2371                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2372                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2373                 else
2374                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2375                 break;
2376
2377         case EXT4_MOUNT_ORDERED_DATA:
2378         case EXT4_MOUNT_WRITEBACK_DATA:
2379                 if (!jbd2_journal_check_available_features
2380                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2381                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2382                                "requested data journaling mode\n");
2383                         goto failed_mount4;
2384                 }
2385         default:
2386                 break;
2387         }
2388
2389         if (test_opt(sb, NOBH)) {
2390                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2391                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2392                                 "its supported only with writeback mode\n");
2393                         clear_opt(sbi->s_mount_opt, NOBH);
2394                 }
2395         }
2396         /*
2397          * The jbd2_journal_load will have done any necessary log recovery,
2398          * so we can safely mount the rest of the filesystem now.
2399          */
2400
2401         root = ext4_iget(sb, EXT4_ROOT_INO);
2402         if (IS_ERR(root)) {
2403                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2404                 ret = PTR_ERR(root);
2405                 goto failed_mount4;
2406         }
2407         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2408                 iput(root);
2409                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2410                 goto failed_mount4;
2411         }
2412         sb->s_root = d_alloc_root(root);
2413         if (!sb->s_root) {
2414                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2415                 iput(root);
2416                 ret = -ENOMEM;
2417                 goto failed_mount4;
2418         }
2419
2420         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2421
2422         /* determine the minimum size of new large inodes, if present */
2423         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2424                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2425                                                      EXT4_GOOD_OLD_INODE_SIZE;
2426                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2427                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2428                         if (sbi->s_want_extra_isize <
2429                             le16_to_cpu(es->s_want_extra_isize))
2430                                 sbi->s_want_extra_isize =
2431                                         le16_to_cpu(es->s_want_extra_isize);
2432                         if (sbi->s_want_extra_isize <
2433                             le16_to_cpu(es->s_min_extra_isize))
2434                                 sbi->s_want_extra_isize =
2435                                         le16_to_cpu(es->s_min_extra_isize);
2436                 }
2437         }
2438         /* Check if enough inode space is available */
2439         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2440                                                         sbi->s_inode_size) {
2441                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2442                                                        EXT4_GOOD_OLD_INODE_SIZE;
2443                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2444                         "available.\n");
2445         }
2446
2447         /*
2448          * akpm: core read_super() calls in here with the superblock locked.
2449          * That deadlocks, because orphan cleanup needs to lock the superblock
2450          * in numerous places.  Here we just pop the lock - it's relatively
2451          * harmless, because we are now ready to accept write_super() requests,
2452          * and aviro says that's the only reason for hanging onto the
2453          * superblock lock.
2454          */
2455         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2456         ext4_orphan_cleanup(sb, es);
2457         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2458         if (needs_recovery)
2459                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2460         ext4_mark_recovery_complete(sb, es);
2461         printk(KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2462                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2463                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2464                "writeback");
2465
2466         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2467                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2468                                 "requested data journaling mode\n");
2469                 clear_opt(sbi->s_mount_opt, DELALLOC);
2470         } else if (test_opt(sb, DELALLOC))
2471                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2472
2473         ext4_ext_init(sb);
2474         ext4_mb_init(sb, needs_recovery);
2475
2476         lock_kernel();
2477         return 0;
2478
2479 cantfind_ext4:
2480         if (!silent)
2481                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2482                        sb->s_id);
2483         goto failed_mount;
2484
2485 failed_mount4:
2486         jbd2_journal_destroy(sbi->s_journal);
2487         sbi->s_journal = NULL;
2488 failed_mount3:
2489         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2490         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2491         percpu_counter_destroy(&sbi->s_dirs_counter);
2492 failed_mount2:
2493         for (i = 0; i < db_count; i++)
2494                 brelse(sbi->s_group_desc[i]);
2495         kfree(sbi->s_group_desc);
2496 failed_mount:
2497 #ifdef CONFIG_QUOTA
2498         for (i = 0; i < MAXQUOTAS; i++)
2499                 kfree(sbi->s_qf_names[i]);
2500 #endif
2501         ext4_blkdev_remove(sbi);
2502         brelse(bh);
2503 out_fail:
2504         sb->s_fs_info = NULL;
2505         kfree(sbi);
2506         lock_kernel();
2507         return ret;
2508 }
2509
2510 /*
2511  * Setup any per-fs journal parameters now.  We'll do this both on
2512  * initial mount, once the journal has been initialised but before we've
2513  * done any recovery; and again on any subsequent remount.
2514  */
2515 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2516 {
2517         struct ext4_sb_info *sbi = EXT4_SB(sb);
2518
2519         if (sbi->s_commit_interval)
2520                 journal->j_commit_interval = sbi->s_commit_interval;
2521         /* We could also set up an ext4-specific default for the commit
2522          * interval here, but for now we'll just fall back to the jbd
2523          * default. */
2524
2525         spin_lock(&journal->j_state_lock);
2526         if (test_opt(sb, BARRIER))
2527                 journal->j_flags |= JBD2_BARRIER;
2528         else
2529                 journal->j_flags &= ~JBD2_BARRIER;
2530         spin_unlock(&journal->j_state_lock);
2531 }
2532
2533 static journal_t *ext4_get_journal(struct super_block *sb,
2534                                    unsigned int journal_inum)
2535 {
2536         struct inode *journal_inode;
2537         journal_t *journal;
2538
2539         /* First, test for the existence of a valid inode on disk.  Bad
2540          * things happen if we iget() an unused inode, as the subsequent
2541          * iput() will try to delete it. */
2542
2543         journal_inode = ext4_iget(sb, journal_inum);
2544         if (IS_ERR(journal_inode)) {
2545                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2546                 return NULL;
2547         }
2548         if (!journal_inode->i_nlink) {
2549                 make_bad_inode(journal_inode);
2550                 iput(journal_inode);
2551                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2552                 return NULL;
2553         }
2554
2555         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2556                   journal_inode, journal_inode->i_size);
2557         if (!S_ISREG(journal_inode->i_mode)) {
2558                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2559                 iput(journal_inode);
2560                 return NULL;
2561         }
2562
2563         journal = jbd2_journal_init_inode(journal_inode);
2564         if (!journal) {
2565                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2566                 iput(journal_inode);
2567                 return NULL;
2568         }
2569         journal->j_private = sb;
2570         ext4_init_journal_params(sb, journal);
2571         return journal;
2572 }
2573
2574 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2575                                        dev_t j_dev)
2576 {
2577         struct buffer_head *bh;
2578         journal_t *journal;
2579         ext4_fsblk_t start;
2580         ext4_fsblk_t len;
2581         int hblock, blocksize;
2582         ext4_fsblk_t sb_block;
2583         unsigned long offset;
2584         struct ext4_super_block *es;
2585         struct block_device *bdev;
2586
2587         bdev = ext4_blkdev_get(j_dev);
2588         if (bdev == NULL)
2589                 return NULL;
2590
2591         if (bd_claim(bdev, sb)) {
2592                 printk(KERN_ERR
2593                         "EXT4: failed to claim external journal device.\n");
2594                 blkdev_put(bdev);
2595                 return NULL;
2596         }
2597
2598         blocksize = sb->s_blocksize;
2599         hblock = bdev_hardsect_size(bdev);
2600         if (blocksize < hblock) {
2601                 printk(KERN_ERR
2602                         "EXT4-fs: blocksize too small for journal device.\n");
2603                 goto out_bdev;
2604         }
2605
2606         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2607         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2608         set_blocksize(bdev, blocksize);
2609         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2610                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2611                        "external journal\n");
2612                 goto out_bdev;
2613         }
2614
2615         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2616         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2617             !(le32_to_cpu(es->s_feature_incompat) &
2618               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2619                 printk(KERN_ERR "EXT4-fs: external journal has "
2620                                         "bad superblock\n");
2621                 brelse(bh);
2622                 goto out_bdev;
2623         }
2624
2625         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2626                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2627                 brelse(bh);
2628                 goto out_bdev;
2629         }
2630
2631         len = ext4_blocks_count(es);
2632         start = sb_block + 1;
2633         brelse(bh);     /* we're done with the superblock */
2634
2635         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2636                                         start, len, blocksize);
2637         if (!journal) {
2638                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2639                 goto out_bdev;
2640         }
2641         journal->j_private = sb;
2642         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2643         wait_on_buffer(journal->j_sb_buffer);
2644         if (!buffer_uptodate(journal->j_sb_buffer)) {
2645                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2646                 goto out_journal;
2647         }
2648         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2649                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2650                                         "user (unsupported) - %d\n",
2651                         be32_to_cpu(journal->j_superblock->s_nr_users));
2652                 goto out_journal;
2653         }
2654         EXT4_SB(sb)->journal_bdev = bdev;
2655         ext4_init_journal_params(sb, journal);
2656         return journal;
2657 out_journal:
2658         jbd2_journal_destroy(journal);
2659 out_bdev:
2660         ext4_blkdev_put(bdev);
2661         return NULL;
2662 }
2663
2664 static int ext4_load_journal(struct super_block *sb,
2665                              struct ext4_super_block *es,
2666                              unsigned long journal_devnum)
2667 {
2668         journal_t *journal;
2669         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2670         dev_t journal_dev;
2671         int err = 0;
2672         int really_read_only;
2673
2674         if (journal_devnum &&
2675             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2676                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2677                         "numbers have changed\n");
2678                 journal_dev = new_decode_dev(journal_devnum);
2679         } else
2680                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2681
2682         really_read_only = bdev_read_only(sb->s_bdev);
2683
2684         /*
2685          * Are we loading a blank journal or performing recovery after a
2686          * crash?  For recovery, we need to check in advance whether we
2687          * can get read-write access to the device.
2688          */
2689
2690         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2691                 if (sb->s_flags & MS_RDONLY) {
2692                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2693                                         "required on readonly filesystem.\n");
2694                         if (really_read_only) {
2695                                 printk(KERN_ERR "EXT4-fs: write access "
2696                                         "unavailable, cannot proceed.\n");
2697                                 return -EROFS;
2698                         }
2699                         printk(KERN_INFO "EXT4-fs: write access will "
2700                                "be enabled during recovery.\n");
2701                 }
2702         }
2703
2704         if (journal_inum && journal_dev) {
2705                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2706                        "and inode journals!\n");
2707                 return -EINVAL;
2708         }
2709
2710         if (journal_inum) {
2711                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2712                         return -EINVAL;
2713         } else {
2714                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2715                         return -EINVAL;
2716         }
2717
2718         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2719                 err = jbd2_journal_update_format(journal);
2720                 if (err)  {
2721                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2722                         jbd2_journal_destroy(journal);
2723                         return err;
2724                 }
2725         }
2726
2727         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2728                 err = jbd2_journal_wipe(journal, !really_read_only);
2729         if (!err)
2730                 err = jbd2_journal_load(journal);
2731
2732         if (err) {
2733                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2734                 jbd2_journal_destroy(journal);
2735                 return err;
2736         }
2737
2738         EXT4_SB(sb)->s_journal = journal;
2739         ext4_clear_journal_err(sb, es);
2740
2741         if (journal_devnum &&
2742             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2743                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2744                 sb->s_dirt = 1;
2745
2746                 /* Make sure we flush the recovery flag to disk. */
2747                 ext4_commit_super(sb, es, 1);
2748         }
2749
2750         return 0;
2751 }
2752
2753 static int ext4_create_journal(struct super_block *sb,
2754                                struct ext4_super_block *es,
2755                                unsigned int journal_inum)
2756 {
2757         journal_t *journal;
2758         int err;
2759
2760         if (sb->s_flags & MS_RDONLY) {
2761                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2762                                 "create journal.\n");
2763                 return -EROFS;
2764         }
2765
2766         journal = ext4_get_journal(sb, journal_inum);
2767         if (!journal)
2768                 return -EINVAL;
2769
2770         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2771                journal_inum);
2772
2773         err = jbd2_journal_create(journal);
2774         if (err) {
2775                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2776                 jbd2_journal_destroy(journal);
2777                 return -EIO;
2778         }
2779
2780         EXT4_SB(sb)->s_journal = journal;
2781
2782         ext4_update_dynamic_rev(sb);
2783         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2784         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2785
2786         es->s_journal_inum = cpu_to_le32(journal_inum);
2787         sb->s_dirt = 1;
2788
2789         /* Make sure we flush the recovery flag to disk. */
2790         ext4_commit_super(sb, es, 1);
2791
2792         return 0;
2793 }
2794
2795 static void ext4_commit_super(struct super_block *sb,
2796                               struct ext4_super_block *es, int sync)
2797 {
2798         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2799
2800         if (!sbh)
2801                 return;
2802         es->s_wtime = cpu_to_le32(get_seconds());
2803         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2804         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2805         BUFFER_TRACE(sbh, "marking dirty");
2806         mark_buffer_dirty(sbh);
2807         if (sync)
2808                 sync_dirty_buffer(sbh);
2809 }
2810
2811
2812 /*
2813  * Have we just finished recovery?  If so, and if we are mounting (or
2814  * remounting) the filesystem readonly, then we will end up with a
2815  * consistent fs on disk.  Record that fact.
2816  */
2817 static void ext4_mark_recovery_complete(struct super_block *sb,
2818                                         struct ext4_super_block *es)
2819 {
2820         journal_t *journal = EXT4_SB(sb)->s_journal;
2821
2822         jbd2_journal_lock_updates(journal);
2823         jbd2_journal_flush(journal);
2824         lock_super(sb);
2825         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2826             sb->s_flags & MS_RDONLY) {
2827                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2828                 sb->s_dirt = 0;
2829                 ext4_commit_super(sb, es, 1);
2830         }
2831         unlock_super(sb);
2832         jbd2_journal_unlock_updates(journal);
2833 }
2834
2835 /*
2836  * If we are mounting (or read-write remounting) a filesystem whose journal
2837  * has recorded an error from a previous lifetime, move that error to the
2838  * main filesystem now.
2839  */
2840 static void ext4_clear_journal_err(struct super_block *sb,
2841                                    struct ext4_super_block *es)
2842 {
2843         journal_t *journal;
2844         int j_errno;
2845         const char *errstr;
2846
2847         journal = EXT4_SB(sb)->s_journal;
2848
2849         /*
2850          * Now check for any error status which may have been recorded in the
2851          * journal by a prior ext4_error() or ext4_abort()
2852          */
2853
2854         j_errno = jbd2_journal_errno(journal);
2855         if (j_errno) {
2856                 char nbuf[16];
2857
2858                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2859                 ext4_warning(sb, __func__, "Filesystem error recorded "
2860                              "from previous mount: %s", errstr);
2861                 ext4_warning(sb, __func__, "Marking fs in need of "
2862                              "filesystem check.");
2863
2864                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2865                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2866                 ext4_commit_super(sb, es, 1);
2867
2868                 jbd2_journal_clear_err(journal);
2869         }
2870 }
2871
2872 /*
2873  * Force the running and committing transactions to commit,
2874  * and wait on the commit.
2875  */
2876 int ext4_force_commit(struct super_block *sb)
2877 {
2878         journal_t *journal;
2879         int ret;
2880
2881         if (sb->s_flags & MS_RDONLY)
2882                 return 0;
2883
2884         journal = EXT4_SB(sb)->s_journal;
2885         sb->s_dirt = 0;
2886         ret = ext4_journal_force_commit(journal);
2887         return ret;
2888 }
2889
2890 /*
2891  * Ext4 always journals updates to the superblock itself, so we don't
2892  * have to propagate any other updates to the superblock on disk at this
2893  * point.  Just start an async writeback to get the buffers on their way
2894  * to the disk.
2895  *
2896  * This implicitly triggers the writebehind on sync().
2897  */
2898
2899 static void ext4_write_super(struct super_block *sb)
2900 {
2901         if (mutex_trylock(&sb->s_lock) != 0)
2902                 BUG();
2903         sb->s_dirt = 0;
2904 }
2905
2906 static int ext4_sync_fs(struct super_block *sb, int wait)
2907 {
2908         tid_t target;
2909
2910         sb->s_dirt = 0;
2911         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2912                 if (wait)
2913                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2914         }
2915         return 0;
2916 }
2917
2918 /*
2919  * LVM calls this function before a (read-only) snapshot is created.  This
2920  * gives us a chance to flush the journal completely and mark the fs clean.
2921  */
2922 static void ext4_write_super_lockfs(struct super_block *sb)
2923 {
2924         sb->s_dirt = 0;
2925
2926         if (!(sb->s_flags & MS_RDONLY)) {
2927                 journal_t *journal = EXT4_SB(sb)->s_journal;
2928
2929                 /* Now we set up the journal barrier. */
2930                 jbd2_journal_lock_updates(journal);
2931                 jbd2_journal_flush(journal);
2932
2933                 /* Journal blocked and flushed, clear needs_recovery flag. */
2934                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2935                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2936         }
2937 }
2938
2939 /*
2940  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2941  * flag here, even though the filesystem is not technically dirty yet.
2942  */
2943 static void ext4_unlockfs(struct super_block *sb)
2944 {
2945         if (!(sb->s_flags & MS_RDONLY)) {
2946                 lock_super(sb);
2947                 /* Reser the needs_recovery flag before the fs is unlocked. */
2948                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2949                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2950                 unlock_super(sb);
2951                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2952         }
2953 }
2954
2955 static int ext4_remount(struct super_block *sb, int *flags, char *data)
2956 {
2957         struct ext4_super_block *es;
2958         struct ext4_sb_info *sbi = EXT4_SB(sb);
2959         ext4_fsblk_t n_blocks_count = 0;
2960         unsigned long old_sb_flags;
2961         struct ext4_mount_options old_opts;
2962         ext4_group_t g;
2963         int err;
2964 #ifdef CONFIG_QUOTA
2965         int i;
2966 #endif
2967
2968         /* Store the original options */
2969         old_sb_flags = sb->s_flags;
2970         old_opts.s_mount_opt = sbi->s_mount_opt;
2971         old_opts.s_resuid = sbi->s_resuid;
2972         old_opts.s_resgid = sbi->s_resgid;
2973         old_opts.s_commit_interval = sbi->s_commit_interval;
2974 #ifdef CONFIG_QUOTA
2975         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2976         for (i = 0; i < MAXQUOTAS; i++)
2977                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2978 #endif
2979
2980         /*
2981          * Allow the "check" option to be passed as a remount option.
2982          */
2983         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2984                 err = -EINVAL;
2985                 goto restore_opts;
2986         }
2987
2988         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2989                 ext4_abort(sb, __func__, "Abort forced by user");
2990
2991         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2992                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2993
2994         es = sbi->s_es;
2995
2996         ext4_init_journal_params(sb, sbi->s_journal);
2997
2998         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2999                 n_blocks_count > ext4_blocks_count(es)) {
3000                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3001                         err = -EROFS;
3002                         goto restore_opts;
3003                 }
3004
3005                 if (*flags & MS_RDONLY) {
3006                         /*
3007                          * First of all, the unconditional stuff we have to do
3008                          * to disable replay of the journal when we next remount
3009                          */
3010                         sb->s_flags |= MS_RDONLY;
3011
3012                         /*
3013                          * OK, test if we are remounting a valid rw partition
3014                          * readonly, and if so set the rdonly flag and then
3015                          * mark the partition as valid again.
3016                          */
3017                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3018                             (sbi->s_mount_state & EXT4_VALID_FS))
3019                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3020
3021                         /*
3022                          * We have to unlock super so that we can wait for
3023                          * transactions.
3024                          */
3025                         unlock_super(sb);
3026                         ext4_mark_recovery_complete(sb, es);
3027                         lock_super(sb);
3028                 } else {
3029                         __le32 ret;
3030                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3031                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3032                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3033                                        "remount RDWR because of unsupported "
3034                                        "optional features (%x).\n",
3035                                        sb->s_id, le32_to_cpu(ret));
3036                                 err = -EROFS;
3037                                 goto restore_opts;
3038                         }
3039
3040                         /*
3041                          * Make sure the group descriptor checksums
3042                          * are sane.  If they aren't, refuse to
3043                          * remount r/w.
3044                          */
3045                         for (g = 0; g < sbi->s_groups_count; g++) {
3046                                 struct ext4_group_desc *gdp =
3047                                         ext4_get_group_desc(sb, g, NULL);
3048
3049                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3050                                         printk(KERN_ERR
3051                "EXT4-fs: ext4_remount: "
3052                 "Checksum for group %lu failed (%u!=%u)\n",
3053                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3054                                                le16_to_cpu(gdp->bg_checksum));
3055                                         err = -EINVAL;
3056                                         goto restore_opts;
3057                                 }
3058                         }
3059
3060                         /*
3061                          * If we have an unprocessed orphan list hanging
3062                          * around from a previously readonly bdev mount,
3063                          * require a full umount/remount for now.
3064                          */
3065                         if (es->s_last_orphan) {
3066                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3067                                        "remount RDWR because of unprocessed "
3068                                        "orphan inode list.  Please "
3069                                        "umount/remount instead.\n",
3070                                        sb->s_id);
3071                                 err = -EINVAL;
3072                                 goto restore_opts;
3073                         }
3074
3075                         /*
3076                          * Mounting a RDONLY partition read-write, so reread
3077                          * and store the current valid flag.  (It may have
3078                          * been changed by e2fsck since we originally mounted
3079                          * the partition.)
3080                          */
3081                         ext4_clear_journal_err(sb, es);
3082                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3083                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3084                                 goto restore_opts;
3085                         if (!ext4_setup_super(sb, es, 0))
3086                                 sb->s_flags &= ~MS_RDONLY;
3087                 }
3088         }
3089 #ifdef CONFIG_QUOTA
3090         /* Release old quota file names */
3091         for (i = 0; i < MAXQUOTAS; i++)
3092                 if (old_opts.s_qf_names[i] &&
3093                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3094                         kfree(old_opts.s_qf_names[i]);
3095 #endif
3096         return 0;
3097 restore_opts:
3098         sb->s_flags = old_sb_flags;
3099         sbi->s_mount_opt = old_opts.s_mount_opt;
3100         sbi->s_resuid = old_opts.s_resuid;
3101         sbi->s_resgid = old_opts.s_resgid;
3102         sbi->s_commit_interval = old_opts.s_commit_interval;
3103 #ifdef CONFIG_QUOTA
3104         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3105         for (i = 0; i < MAXQUOTAS; i++) {
3106                 if (sbi->s_qf_names[i] &&
3107                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3108                         kfree(sbi->s_qf_names[i]);
3109                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3110         }
3111 #endif
3112         return err;
3113 }
3114
3115 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3116 {
3117         struct super_block *sb = dentry->d_sb;
3118         struct ext4_sb_info *sbi = EXT4_SB(sb);
3119         struct ext4_super_block *es = sbi->s_es;
3120         u64 fsid;
3121
3122         if (test_opt(sb, MINIX_DF)) {
3123                 sbi->s_overhead_last = 0;
3124         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3125                 ext4_group_t ngroups = sbi->s_groups_count, i;
3126                 ext4_fsblk_t overhead = 0;
3127                 smp_rmb();
3128
3129                 /*
3130                  * Compute the overhead (FS structures).  This is constant
3131                  * for a given filesystem unless the number of block groups
3132                  * changes so we cache the previous value until it does.
3133                  */
3134
3135                 /*
3136                  * All of the blocks before first_data_block are
3137                  * overhead
3138                  */
3139                 overhead = le32_to_cpu(es->s_first_data_block);
3140
3141                 /*
3142                  * Add the overhead attributed to the superblock and
3143                  * block group descriptors.  If the sparse superblocks
3144                  * feature is turned on, then not all groups have this.
3145                  */
3146                 for (i = 0; i < ngroups; i++) {
3147                         overhead += ext4_bg_has_super(sb, i) +
3148                                 ext4_bg_num_gdb(sb, i);
3149                         cond_resched();
3150                 }
3151
3152                 /*
3153                  * Every block group has an inode bitmap, a block
3154                  * bitmap, and an inode table.
3155                  */
3156                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3157                 sbi->s_overhead_last = overhead;
3158                 smp_wmb();
3159                 sbi->s_blocks_last = ext4_blocks_count(es);
3160         }
3161
3162         buf->f_type = EXT4_SUPER_MAGIC;
3163         buf->f_bsize = sb->s_blocksize;
3164         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3165         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
3166         ext4_free_blocks_count_set(es, buf->f_bfree);
3167         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3168         if (buf->f_bfree < ext4_r_blocks_count(es))
3169                 buf->f_bavail = 0;
3170         buf->f_files = le32_to_cpu(es->s_inodes_count);
3171         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3172         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3173         buf->f_namelen = EXT4_NAME_LEN;
3174         fsid = le64_to_cpup((void *)es->s_uuid) ^
3175                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3176         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3177         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3178         return 0;
3179 }
3180
3181 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3182  * is locked for write. Otherwise the are possible deadlocks:
3183  * Process 1                         Process 2
3184  * ext4_create()                     quota_sync()
3185  *   jbd2_journal_start()                   write_dquot()
3186  *   DQUOT_INIT()                        down(dqio_mutex)
3187  *     down(dqio_mutex)                    jbd2_journal_start()
3188  *
3189  */
3190
3191 #ifdef CONFIG_QUOTA
3192
3193 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3194 {
3195         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3196 }
3197
3198 static int ext4_dquot_initialize(struct inode *inode, int type)
3199 {
3200         handle_t *handle;
3201         int ret, err;
3202
3203         /* We may create quota structure so we need to reserve enough blocks */
3204         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3205         if (IS_ERR(handle))
3206                 return PTR_ERR(handle);
3207         ret = dquot_initialize(inode, type);
3208         err = ext4_journal_stop(handle);
3209         if (!ret)
3210                 ret = err;
3211         return ret;
3212 }
3213
3214 static int ext4_dquot_drop(struct inode *inode)
3215 {
3216         handle_t *handle;
3217         int ret, err;
3218
3219         /* We may delete quota structure so we need to reserve enough blocks */
3220         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3221         if (IS_ERR(handle)) {
3222                 /*
3223                  * We call dquot_drop() anyway to at least release references
3224                  * to quota structures so that umount does not hang.
3225                  */
3226                 dquot_drop(inode);
3227                 return PTR_ERR(handle);
3228         }
3229         ret = dquot_drop(inode);
3230         err = ext4_journal_stop(handle);
3231         if (!ret)
3232                 ret = err;
3233         return ret;
3234 }
3235
3236 static int ext4_write_dquot(struct dquot *dquot)
3237 {
3238         int ret, err;
3239         handle_t *handle;
3240         struct inode *inode;
3241
3242         inode = dquot_to_inode(dquot);
3243         handle = ext4_journal_start(inode,
3244                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3245         if (IS_ERR(handle))
3246                 return PTR_ERR(handle);
3247         ret = dquot_commit(dquot);
3248         err = ext4_journal_stop(handle);
3249         if (!ret)
3250                 ret = err;
3251         return ret;
3252 }
3253
3254 static int ext4_acquire_dquot(struct dquot *dquot)
3255 {
3256         int ret, err;
3257         handle_t *handle;
3258
3259         handle = ext4_journal_start(dquot_to_inode(dquot),
3260                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3261         if (IS_ERR(handle))
3262                 return PTR_ERR(handle);
3263         ret = dquot_acquire(dquot);
3264         err = ext4_journal_stop(handle);
3265         if (!ret)
3266                 ret = err;
3267         return ret;
3268 }
3269
3270 static int ext4_release_dquot(struct dquot *dquot)
3271 {
3272         int ret, err;
3273         handle_t *handle;
3274
3275         handle = ext4_journal_start(dquot_to_inode(dquot),
3276                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3277         if (IS_ERR(handle)) {
3278                 /* Release dquot anyway to avoid endless cycle in dqput() */
3279                 dquot_release(dquot);
3280                 return PTR_ERR(handle);
3281         }
3282         ret = dquot_release(dquot);
3283         err = ext4_journal_stop(handle);
3284         if (!ret)
3285                 ret = err;
3286         return ret;
3287 }
3288
3289 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3290 {
3291         /* Are we journaling quotas? */
3292         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3293             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3294                 dquot_mark_dquot_dirty(dquot);
3295                 return ext4_write_dquot(dquot);
3296         } else {
3297                 return dquot_mark_dquot_dirty(dquot);
3298         }
3299 }
3300
3301 static int ext4_write_info(struct super_block *sb, int type)
3302 {
3303         int ret, err;
3304         handle_t *handle;
3305
3306         /* Data block + inode block */
3307         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3308         if (IS_ERR(handle))
3309                 return PTR_ERR(handle);
3310         ret = dquot_commit_info(sb, type);
3311         err = ext4_journal_stop(handle);
3312         if (!ret)
3313                 ret = err;
3314         return ret;
3315 }
3316
3317 /*
3318  * Turn on quotas during mount time - we need to find
3319  * the quota file and such...
3320  */
3321 static int ext4_quota_on_mount(struct super_block *sb, int type)
3322 {
3323         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3324                         EXT4_SB(sb)->s_jquota_fmt, type);
3325 }
3326
3327 /*
3328  * Standard function to be called on quota_on
3329  */
3330 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3331                          char *path, int remount)
3332 {
3333         int err;
3334         struct nameidata nd;
3335
3336         if (!test_opt(sb, QUOTA))
3337                 return -EINVAL;
3338         /* When remounting, no checks are needed and in fact, path is NULL */
3339         if (remount)
3340                 return vfs_quota_on(sb, type, format_id, path, remount);
3341
3342         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3343         if (err)
3344                 return err;
3345
3346         /* Quotafile not on the same filesystem? */
3347         if (nd.path.mnt->mnt_sb != sb) {
3348                 path_put(&nd.path);
3349                 return -EXDEV;
3350         }
3351         /* Journaling quota? */
3352         if (EXT4_SB(sb)->s_qf_names[type]) {
3353                 /* Quotafile not in fs root? */
3354                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3355                         printk(KERN_WARNING
3356                                 "EXT4-fs: Quota file not on filesystem root. "
3357                                 "Journaled quota will not work.\n");
3358         }
3359
3360         /*
3361          * When we journal data on quota file, we have to flush journal to see
3362          * all updates to the file when we bypass pagecache...
3363          */
3364         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3365                 /*
3366                  * We don't need to lock updates but journal_flush() could
3367                  * otherwise be livelocked...
3368                  */
3369                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3370                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3371                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3372         }
3373
3374         err = vfs_quota_on_path(sb, type, format_id, &nd.path);
3375         path_put(&nd.path);
3376         return err;
3377 }
3378
3379 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3380  * acquiring the locks... As quota files are never truncated and quota code
3381  * itself serializes the operations (and noone else should touch the files)
3382  * we don't have to be afraid of races */
3383 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3384                                size_t len, loff_t off)
3385 {
3386         struct inode *inode = sb_dqopt(sb)->files[type];
3387         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3388         int err = 0;
3389         int offset = off & (sb->s_blocksize - 1);
3390         int tocopy;
3391         size_t toread;
3392         struct buffer_head *bh;
3393         loff_t i_size = i_size_read(inode);
3394
3395         if (off > i_size)
3396                 return 0;
3397         if (off+len > i_size)
3398                 len = i_size-off;
3399         toread = len;
3400         while (toread > 0) {
3401                 tocopy = sb->s_blocksize - offset < toread ?
3402                                 sb->s_blocksize - offset : toread;
3403                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3404                 if (err)
3405                         return err;
3406                 if (!bh)        /* A hole? */
3407                         memset(data, 0, tocopy);
3408                 else
3409                         memcpy(data, bh->b_data+offset, tocopy);
3410                 brelse(bh);
3411                 offset = 0;
3412                 toread -= tocopy;
3413                 data += tocopy;
3414                 blk++;
3415         }
3416         return len;
3417 }
3418
3419 /* Write to quotafile (we know the transaction is already started and has
3420  * enough credits) */
3421 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3422                                 const char *data, size_t len, loff_t off)
3423 {
3424         struct inode *inode = sb_dqopt(sb)->files[type];
3425         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3426         int err = 0;
3427         int offset = off & (sb->s_blocksize - 1);
3428         int tocopy;
3429         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3430         size_t towrite = len;
3431         struct buffer_head *bh;
3432         handle_t *handle = journal_current_handle();
3433
3434         if (!handle) {
3435                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%Lu, len=%Lu)"
3436                         " cancelled because transaction is not started.\n",
3437                         (unsigned long long)off, (unsigned long long)len);
3438                 return -EIO;
3439         }
3440         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3441         while (towrite > 0) {
3442                 tocopy = sb->s_blocksize - offset < towrite ?
3443                                 sb->s_blocksize - offset : towrite;
3444                 bh = ext4_bread(handle, inode, blk, 1, &err);
3445                 if (!bh)
3446                         goto out;
3447                 if (journal_quota) {
3448                         err = ext4_journal_get_write_access(handle, bh);
3449                         if (err) {
3450                                 brelse(bh);
3451                                 goto out;
3452                         }
3453                 }
3454                 lock_buffer(bh);
3455                 memcpy(bh->b_data+offset, data, tocopy);
3456                 flush_dcache_page(bh->b_page);
3457                 unlock_buffer(bh);
3458                 if (journal_quota)
3459                         err = ext4_journal_dirty_metadata(handle, bh);
3460                 else {
3461                         /* Always do at least ordered writes for quotas */
3462                         err = ext4_jbd2_file_inode(handle, inode);
3463                         mark_buffer_dirty(bh);
3464                 }
3465                 brelse(bh);
3466                 if (err)
3467                         goto out;
3468                 offset = 0;
3469                 towrite -= tocopy;
3470                 data += tocopy;
3471                 blk++;
3472         }
3473 out:
3474         if (len == towrite) {
3475                 mutex_unlock(&inode->i_mutex);
3476                 return err;
3477         }
3478         if (inode->i_size < off+len-towrite) {
3479                 i_size_write(inode, off+len-towrite);
3480                 EXT4_I(inode)->i_disksize = inode->i_size;
3481         }
3482         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3483         ext4_mark_inode_dirty(handle, inode);
3484         mutex_unlock(&inode->i_mutex);
3485         return len - towrite;
3486 }
3487
3488 #endif
3489
3490 static int ext4_get_sb(struct file_system_type *fs_type,
3491         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3492 {
3493         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3494 }
3495
3496 static struct file_system_type ext4dev_fs_type = {
3497         .owner          = THIS_MODULE,
3498         .name           = "ext4dev",
3499         .get_sb         = ext4_get_sb,
3500         .kill_sb        = kill_block_super,
3501         .fs_flags       = FS_REQUIRES_DEV,
3502 };
3503
3504 static int __init init_ext4_fs(void)
3505 {
3506         int err;
3507
3508         err = init_ext4_mballoc();
3509         if (err)
3510                 return err;
3511
3512         err = init_ext4_xattr();
3513         if (err)
3514                 goto out2;
3515         err = init_inodecache();
3516         if (err)
3517                 goto out1;
3518         err = register_filesystem(&ext4dev_fs_type);
3519         if (err)
3520                 goto out;
3521         return 0;
3522 out:
3523         destroy_inodecache();
3524 out1:
3525         exit_ext4_xattr();
3526 out2:
3527         exit_ext4_mballoc();
3528         return err;
3529 }
3530
3531 static void __exit exit_ext4_fs(void)
3532 {
3533         unregister_filesystem(&ext4dev_fs_type);
3534         destroy_inodecache();
3535         exit_ext4_xattr();
3536         exit_ext4_mballoc();
3537 }
3538
3539 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3540 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3541 MODULE_LICENSE("GPL");
3542 module_init(init_ext4_fs)
3543 module_exit(exit_ext4_fs)