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         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 524         brelse(sbi->s_sbh);
 525 #ifdef CONFIG_QUOTA
 526         for (i = 0; i < MAXQUOTAS; i++)
 527                 kfree(sbi->s_qf_names[i]);
 528 #endif
 529
 530         /* Debugging code just in case the in-memory inode orphan list
 531          * isn't empty.  The on-disk one can be non-empty if we've
 532          * detected an error and taken the fs readonly, but the
 533          * in-memory list had better be clean by this point. */
 534         if (!list_empty(&sbi->s_orphan))
 535                 dump_orphan_list(sb, sbi);
 536         J_ASSERT(list_empty(&sbi->s_orphan));
 537
 538         invalidate_bdev(sb->s_bdev);
 539         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 540                 /*
 541                  * Invalidate the journal device's buffers.  We don't want them
 542                  * floating about in memory - the physical journal device may
 543                  * hotswapped, and it breaks the `ro-after' testing code.
 544                  */
 545                 sync_blockdev(sbi->journal_bdev);
 546                 invalidate_bdev(sbi->journal_bdev);
 547                 ext4_blkdev_remove(sbi);
 548         }
 549         sb->s_fs_info = NULL;
 550         kfree(sbi);
 551         return;
 552 }
 553
 554 static struct kmem_cache *ext4_inode_cachep;
 555
 556 /*
 557  * Called inside transaction, so use GFP_NOFS
 558  */
 559 static struct inode *ext4_alloc_inode(struct super_block *sb)
 560 {
 561         struct ext4_inode_info *ei;
 562
 563         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 564         if (!ei)
 565                 return NULL;
 566 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 567         ei->i_acl = EXT4_ACL_NOT_CACHED;
 568         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 569 #endif
 570         ei->i_block_alloc_info = NULL;
 571         ei->vfs_inode.i_version = 1;
 572         ei->vfs_inode.i_data.writeback_index = 0;
 573         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 574         INIT_LIST_HEAD(&ei->i_prealloc_list);
 575         spin_lock_init(&ei->i_prealloc_lock);
 576         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 577         ei->i_reserved_data_blocks = 0;
 578         ei->i_reserved_meta_blocks = 0;
 579         ei->i_allocated_meta_blocks = 0;
 580         ei->i_delalloc_reserved_flag = 0;
 581         spin_lock_init(&(ei->i_block_reservation_lock));
 582         return &ei->vfs_inode;
 583 }
 584
 585 static void ext4_destroy_inode(struct inode *inode)
 586 {
 587         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 588                 printk("EXT4 Inode %p: orphan list check failed!\n",
 589                         EXT4_I(inode));
 590                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 591                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 592                                 true);
 593                 dump_stack();
 594         }
 595         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 596 }
 597
 598 static void init_once(void *foo)
 599 {
 600         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 601
 602         INIT_LIST_HEAD(&ei->i_orphan);
 603 #ifdef CONFIG_EXT4DEV_FS_XATTR
 604         init_rwsem(&ei->xattr_sem);
 605 #endif
 606         init_rwsem(&ei->i_data_sem);
 607         inode_init_once(&ei->vfs_inode);
 608 }
 609
 610 static int init_inodecache(void)
 611 {
 612         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 613                                              sizeof(struct ext4_inode_info),
 614                                              0, (SLAB_RECLAIM_ACCOUNT|
 615                                                 SLAB_MEM_SPREAD),
 616                                              init_once);
 617         if (ext4_inode_cachep == NULL)
 618                 return -ENOMEM;
 619         return 0;
 620 }
 621
 622 static void destroy_inodecache(void)
 623 {
 624         kmem_cache_destroy(ext4_inode_cachep);
 625 }
 626
 627 static void ext4_clear_inode(struct inode *inode)
 628 {
 629         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 630 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 631         if (EXT4_I(inode)->i_acl &&
 632                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 633                 posix_acl_release(EXT4_I(inode)->i_acl);
 634                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 635         }
 636         if (EXT4_I(inode)->i_default_acl &&
 637                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 638                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 639                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 640         }
 641 #endif
 642         ext4_discard_reservation(inode);
 643         EXT4_I(inode)->i_block_alloc_info = NULL;
 644         if (unlikely(rsv))
 645                 kfree(rsv);
 646         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 647                                        &EXT4_I(inode)->jinode);
 648 }
 649
 650 static inline void ext4_show_quota_options(struct seq_file *seq,
 651                                            struct super_block *sb)
 652 {
 653 #if defined(CONFIG_QUOTA)
 654         struct ext4_sb_info *sbi = EXT4_SB(sb);
 655
 656         if (sbi->s_jquota_fmt)
 657                 seq_printf(seq, ",jqfmt=%s",
 658                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold" : "vfsv0");
 659
 660         if (sbi->s_qf_names[USRQUOTA])
 661                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 662
 663         if (sbi->s_qf_names[GRPQUOTA])
 664                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 665
 666         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 667                 seq_puts(seq, ",usrquota");
 668
 669         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 670                 seq_puts(seq, ",grpquota");
 671 #endif
 672 }
 673
 674 /*
 675  * Show an option if
 676  *  - it's set to a non-default value OR
 677  *  - if the per-sb default is different from the global default
 678  */
 679 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 680 {
 681         int def_errors;
 682         unsigned long def_mount_opts;
 683         struct super_block *sb = vfs->mnt_sb;
 684         struct ext4_sb_info *sbi = EXT4_SB(sb);
 685         struct ext4_super_block *es = sbi->s_es;
 686
 687         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 688         def_errors     = le16_to_cpu(es->s_errors);
 689
 690         if (sbi->s_sb_block != 1)
 691                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 692         if (test_opt(sb, MINIX_DF))
 693                 seq_puts(seq, ",minixdf");
 694         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 695                 seq_puts(seq, ",grpid");
 696         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 697                 seq_puts(seq, ",nogrpid");
 698         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 699             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 700                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 701         }
 702         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 703             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 704                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 705         }
 706         if (test_opt(sb, ERRORS_RO)) {
 707                 if (def_errors == EXT4_ERRORS_PANIC ||
 708                     def_errors == EXT4_ERRORS_CONTINUE) {
 709                         seq_puts(seq, ",errors=remount-ro");
 710                 }
 711         }
 712         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 713                 seq_puts(seq, ",errors=continue");
 714         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 715                 seq_puts(seq, ",errors=panic");
 716         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 717                 seq_puts(seq, ",nouid32");
 718         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 719                 seq_puts(seq, ",debug");
 720         if (test_opt(sb, OLDALLOC))
 721                 seq_puts(seq, ",oldalloc");
 722 #ifdef CONFIG_EXT4DEV_FS_XATTR
 723         if (test_opt(sb, XATTR_USER) &&
 724                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 725                 seq_puts(seq, ",user_xattr");
 726         if (!test_opt(sb, XATTR_USER) &&
 727             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 728                 seq_puts(seq, ",nouser_xattr");
 729         }
 730 #endif
 731 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 732         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 733                 seq_puts(seq, ",acl");
 734         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 735                 seq_puts(seq, ",noacl");
 736 #endif
 737         if (!test_opt(sb, RESERVATION))
 738                 seq_puts(seq, ",noreservation");
 739         if (sbi->s_commit_interval) {
 740                 seq_printf(seq, ",commit=%u",
 741                            (unsigned) (sbi->s_commit_interval / HZ));
 742         }
 743         /*
 744          * We're changing the default of barrier mount option, so
 745          * let's always display its mount state so it's clear what its
 746          * status is.
 747          */
 748         seq_puts(seq, ",barrier=");
 749         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 750         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 751                 seq_puts(seq, ",journal_async_commit");
 752         if (test_opt(sb, NOBH))
 753                 seq_puts(seq, ",nobh");
 754         if (!test_opt(sb, EXTENTS))
 755                 seq_puts(seq, ",noextents");
 756         if (!test_opt(sb, MBALLOC))
 757                 seq_puts(seq, ",nomballoc");
 758         if (test_opt(sb, I_VERSION))
 759                 seq_puts(seq, ",i_version");
 760         if (!test_opt(sb, DELALLOC))
 761                 seq_puts(seq, ",nodelalloc");
 762
 763
 764         if (sbi->s_stripe)
 765                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 766         /*
 767          * journal mode get enabled in different ways
 768          * So just print the value even if we didn't specify it
 769          */
 770         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 771                 seq_puts(seq, ",data=journal");
 772         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 773                 seq_puts(seq, ",data=ordered");
 774         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 775                 seq_puts(seq, ",data=writeback");
 776
 777         ext4_show_quota_options(seq, sb);
 778         return 0;
 779 }
 780
 781
 782 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 783                 u64 ino, u32 generation)
 784 {
 785         struct inode *inode;
 786
 787         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 788                 return ERR_PTR(-ESTALE);
 789         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 790                 return ERR_PTR(-ESTALE);
 791
 792         /* iget isn't really right if the inode is currently unallocated!!
 793          *
 794          * ext4_read_inode will return a bad_inode if the inode had been
 795          * deleted, so we should be safe.
 796          *
 797          * Currently we don't know the generation for parent directory, so
 798          * a generation of 0 means "accept any"
 799          */
 800         inode = ext4_iget(sb, ino);
 801         if (IS_ERR(inode))
 802                 return ERR_CAST(inode);
 803         if (generation && inode->i_generation != generation) {
 804                 iput(inode);
 805                 return ERR_PTR(-ESTALE);
 806         }
 807
 808         return inode;
 809 }
 810
 811 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 812                 int fh_len, int fh_type)
 813 {
 814         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 815                                     ext4_nfs_get_inode);
 816 }
 817
 818 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 819                 int fh_len, int fh_type)
 820 {
 821         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 822                                     ext4_nfs_get_inode);
 823 }
 824
 825 #ifdef CONFIG_QUOTA
 826 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
 827 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 828
 829 static int ext4_dquot_initialize(struct inode *inode, int type);
 830 static int ext4_dquot_drop(struct inode *inode);
 831 static int ext4_write_dquot(struct dquot *dquot);
 832 static int ext4_acquire_dquot(struct dquot *dquot);
 833 static int ext4_release_dquot(struct dquot *dquot);
 834 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 835 static int ext4_write_info(struct super_block *sb, int type);
 836 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 837                                 char *path, int remount);
 838 static int ext4_quota_on_mount(struct super_block *sb, int type);
 839 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 840                                size_t len, loff_t off);
 841 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 842                                 const char *data, size_t len, loff_t off);
 843
 844 static struct dquot_operations ext4_quota_operations = {
 845         .initialize     = ext4_dquot_initialize,
 846         .drop           = ext4_dquot_drop,
 847         .alloc_space    = dquot_alloc_space,
 848         .alloc_inode    = dquot_alloc_inode,
 849         .free_space     = dquot_free_space,
 850         .free_inode     = dquot_free_inode,
 851         .transfer       = dquot_transfer,
 852         .write_dquot    = ext4_write_dquot,
 853         .acquire_dquot  = ext4_acquire_dquot,
 854         .release_dquot  = ext4_release_dquot,
 855         .mark_dirty     = ext4_mark_dquot_dirty,
 856         .write_info     = ext4_write_info
 857 };
 858
 859 static struct quotactl_ops ext4_qctl_operations = {
 860         .quota_on       = ext4_quota_on,
 861         .quota_off      = vfs_quota_off,
 862         .quota_sync     = vfs_quota_sync,
 863         .get_info       = vfs_get_dqinfo,
 864         .set_info       = vfs_set_dqinfo,
 865         .get_dqblk      = vfs_get_dqblk,
 866         .set_dqblk      = vfs_set_dqblk
 867 };
 868 #endif
 869
 870 static const struct super_operations ext4_sops = {
 871         .alloc_inode    = ext4_alloc_inode,
 872         .destroy_inode  = ext4_destroy_inode,
 873         .write_inode    = ext4_write_inode,
 874         .dirty_inode    = ext4_dirty_inode,
 875         .delete_inode   = ext4_delete_inode,
 876         .put_super      = ext4_put_super,
 877         .write_super    = ext4_write_super,
 878         .sync_fs        = ext4_sync_fs,
 879         .write_super_lockfs = ext4_write_super_lockfs,
 880         .unlockfs       = ext4_unlockfs,
 881         .statfs         = ext4_statfs,
 882         .remount_fs     = ext4_remount,
 883         .clear_inode    = ext4_clear_inode,
 884         .show_options   = ext4_show_options,
 885 #ifdef CONFIG_QUOTA
 886         .quota_read     = ext4_quota_read,
 887         .quota_write    = ext4_quota_write,
 888 #endif
 889 };
 890
 891 static const struct export_operations ext4_export_ops = {
 892         .fh_to_dentry = ext4_fh_to_dentry,
 893         .fh_to_parent = ext4_fh_to_parent,
 894         .get_parent = ext4_get_parent,
 895 };
 896
 897 enum {
 898         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 899         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 900         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 901         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 902         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 903         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 904         Opt_journal_checksum, Opt_journal_async_commit,
 905         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 906         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 907         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 908         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 909         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
 910         Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
 911 };
 912
 913 static match_table_t tokens = {
 914         {Opt_bsd_df, "bsddf"},
 915         {Opt_minix_df, "minixdf"},
 916         {Opt_grpid, "grpid"},
 917         {Opt_grpid, "bsdgroups"},
 918         {Opt_nogrpid, "nogrpid"},
 919         {Opt_nogrpid, "sysvgroups"},
 920         {Opt_resgid, "resgid=%u"},
 921         {Opt_resuid, "resuid=%u"},
 922         {Opt_sb, "sb=%u"},
 923         {Opt_err_cont, "errors=continue"},
 924         {Opt_err_panic, "errors=panic"},
 925         {Opt_err_ro, "errors=remount-ro"},
 926         {Opt_nouid32, "nouid32"},
 927         {Opt_nocheck, "nocheck"},
 928         {Opt_nocheck, "check=none"},
 929         {Opt_debug, "debug"},
 930         {Opt_oldalloc, "oldalloc"},
 931         {Opt_orlov, "orlov"},
 932         {Opt_user_xattr, "user_xattr"},
 933         {Opt_nouser_xattr, "nouser_xattr"},
 934         {Opt_acl, "acl"},
 935         {Opt_noacl, "noacl"},
 936         {Opt_reservation, "reservation"},
 937         {Opt_noreservation, "noreservation"},
 938         {Opt_noload, "noload"},
 939         {Opt_nobh, "nobh"},
 940         {Opt_bh, "bh"},
 941         {Opt_commit, "commit=%u"},
 942         {Opt_journal_update, "journal=update"},
 943         {Opt_journal_inum, "journal=%u"},
 944         {Opt_journal_dev, "journal_dev=%u"},
 945         {Opt_journal_checksum, "journal_checksum"},
 946         {Opt_journal_async_commit, "journal_async_commit"},
 947         {Opt_abort, "abort"},
 948         {Opt_data_journal, "data=journal"},
 949         {Opt_data_ordered, "data=ordered"},
 950         {Opt_data_writeback, "data=writeback"},
 951         {Opt_offusrjquota, "usrjquota="},
 952         {Opt_usrjquota, "usrjquota=%s"},
 953         {Opt_offgrpjquota, "grpjquota="},
 954         {Opt_grpjquota, "grpjquota=%s"},
 955         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 956         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 957         {Opt_grpquota, "grpquota"},
 958         {Opt_noquota, "noquota"},
 959         {Opt_quota, "quota"},
 960         {Opt_usrquota, "usrquota"},
 961         {Opt_barrier, "barrier=%u"},
 962         {Opt_extents, "extents"},
 963         {Opt_noextents, "noextents"},
 964         {Opt_i_version, "i_version"},
 965         {Opt_mballoc, "mballoc"},
 966         {Opt_nomballoc, "nomballoc"},
 967         {Opt_stripe, "stripe=%u"},
 968         {Opt_resize, "resize"},
 969         {Opt_delalloc, "delalloc"},
 970         {Opt_nodelalloc, "nodelalloc"},
 971         {Opt_err, NULL},
 972 };
 973
 974 static ext4_fsblk_t get_sb_block(void **data)
 975 {
 976         ext4_fsblk_t    sb_block;
 977         char            *options = (char *) *data;
 978
 979         if (!options || strncmp(options, "sb=", 3) != 0)
 980                 return 1;       /* Default location */
 981         options += 3;
 982         /*todo: use simple_strtoll with >32bit ext4 */
 983         sb_block = simple_strtoul(options, &options, 0);
 984         if (*options && *options != ',') {
 985                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
 986                        (char *) *data);
 987                 return 1;
 988         }
 989         if (*options == ',')
 990                 options++;
 991         *data = (void *) options;
 992         return sb_block;
 993 }
 994
 995 static int parse_options(char *options, struct super_block *sb,
 996                          unsigned int *inum, unsigned long *journal_devnum,
 997                          ext4_fsblk_t *n_blocks_count, int is_remount)
 998 {
 999         struct ext4_sb_info *sbi = EXT4_SB(sb);
1000         char *p;
1001         substring_t args[MAX_OPT_ARGS];
1002         int data_opt = 0;
1003         int option;
1004 #ifdef CONFIG_QUOTA
1005         int qtype, qfmt;
1006         char *qname;
1007 #endif
1008         ext4_fsblk_t last_block;
1009
1010         if (!options)
1011                 return 1;
1012
1013         while ((p = strsep(&options, ",")) != NULL) {
1014                 int token;
1015                 if (!*p)
1016                         continue;
1017
1018                 token = match_token(p, tokens, args);
1019                 switch (token) {
1020                 case Opt_bsd_df:
1021                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1022                         break;
1023                 case Opt_minix_df:
1024                         set_opt(sbi->s_mount_opt, MINIX_DF);
1025                         break;
1026                 case Opt_grpid:
1027                         set_opt(sbi->s_mount_opt, GRPID);
1028                         break;
1029                 case Opt_nogrpid:
1030                         clear_opt(sbi->s_mount_opt, GRPID);
1031                         break;
1032                 case Opt_resuid:
1033                         if (match_int(&args[0], &option))
1034                                 return 0;
1035                         sbi->s_resuid = option;
1036                         break;
1037                 case Opt_resgid:
1038                         if (match_int(&args[0], &option))
1039                                 return 0;
1040                         sbi->s_resgid = option;
1041                         break;
1042                 case Opt_sb:
1043                         /* handled by get_sb_block() instead of here */
1044                         /* *sb_block = match_int(&args[0]); */
1045                         break;
1046                 case Opt_err_panic:
1047                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1048                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1049                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1050                         break;
1051                 case Opt_err_ro:
1052                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1053                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1054                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1055                         break;
1056                 case Opt_err_cont:
1057                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1058                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1059                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1060                         break;
1061                 case Opt_nouid32:
1062                         set_opt(sbi->s_mount_opt, NO_UID32);
1063                         break;
1064                 case Opt_nocheck:
1065                         clear_opt(sbi->s_mount_opt, CHECK);
1066                         break;
1067                 case Opt_debug:
1068                         set_opt(sbi->s_mount_opt, DEBUG);
1069                         break;
1070                 case Opt_oldalloc:
1071                         set_opt(sbi->s_mount_opt, OLDALLOC);
1072                         break;
1073                 case Opt_orlov:
1074                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1075                         break;
1076 #ifdef CONFIG_EXT4DEV_FS_XATTR
1077                 case Opt_user_xattr:
1078                         set_opt(sbi->s_mount_opt, XATTR_USER);
1079                         break;
1080                 case Opt_nouser_xattr:
1081                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1082                         break;
1083 #else
1084                 case Opt_user_xattr:
1085                 case Opt_nouser_xattr:
1086                         printk(KERN_ERR "EXT4 (no)user_xattr options "
1087                                "not supported\n");
1088                         break;
1089 #endif
1090 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1091                 case Opt_acl:
1092                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1093                         break;
1094                 case Opt_noacl:
1095                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1096                         break;
1097 #else
1098                 case Opt_acl:
1099                 case Opt_noacl:
1100                         printk(KERN_ERR "EXT4 (no)acl options "
1101                                "not supported\n");
1102                         break;
1103 #endif
1104                 case Opt_reservation:
1105                         set_opt(sbi->s_mount_opt, RESERVATION);
1106                         break;
1107                 case Opt_noreservation:
1108                         clear_opt(sbi->s_mount_opt, RESERVATION);
1109                         break;
1110                 case Opt_journal_update:
1111                         /* @@@ FIXME */
1112                         /* Eventually we will want to be able to create
1113                            a journal file here.  For now, only allow the
1114                            user to specify an existing inode to be the
1115                            journal file. */
1116                         if (is_remount) {
1117                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1118                                        "journal on remount\n");
1119                                 return 0;
1120                         }
1121                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1122                         break;
1123                 case Opt_journal_inum:
1124                         if (is_remount) {
1125                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1126                                        "journal on remount\n");
1127                                 return 0;
1128                         }
1129                         if (match_int(&args[0], &option))
1130                                 return 0;
1131                         *inum = option;
1132                         break;
1133                 case Opt_journal_dev:
1134                         if (is_remount) {
1135                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1136                                        "journal on remount\n");
1137                                 return 0;
1138                         }
1139                         if (match_int(&args[0], &option))
1140                                 return 0;
1141                         *journal_devnum = option;
1142                         break;
1143                 case Opt_journal_checksum:
1144                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1145                         break;
1146                 case Opt_journal_async_commit:
1147                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1148                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1149                         break;
1150                 case Opt_noload:
1151                         set_opt(sbi->s_mount_opt, NOLOAD);
1152                         break;
1153                 case Opt_commit:
1154                         if (match_int(&args[0], &option))
1155                                 return 0;
1156                         if (option < 0)
1157                                 return 0;
1158                         if (option == 0)
1159                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1160                         sbi->s_commit_interval = HZ * option;
1161                         break;
1162                 case Opt_data_journal:
1163                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1164                         goto datacheck;
1165                 case Opt_data_ordered:
1166                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1167                         goto datacheck;
1168                 case Opt_data_writeback:
1169                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1170                 datacheck:
1171                         if (is_remount) {
1172                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1173                                                 != data_opt) {
1174                                         printk(KERN_ERR
1175                                                 "EXT4-fs: cannot change data "
1176                                                 "mode on remount\n");
1177                                         return 0;
1178                                 }
1179                         } else {
1180                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1181                                 sbi->s_mount_opt |= data_opt;
1182                         }
1183                         break;
1184 #ifdef CONFIG_QUOTA
1185                 case Opt_usrjquota:
1186                         qtype = USRQUOTA;
1187                         goto set_qf_name;
1188                 case Opt_grpjquota:
1189                         qtype = GRPQUOTA;
1190 set_qf_name:
1191                         if ((sb_any_quota_enabled(sb) ||
1192                              sb_any_quota_suspended(sb)) &&
1193                             !sbi->s_qf_names[qtype]) {
1194                                 printk(KERN_ERR
1195                                        "EXT4-fs: Cannot change journaled "
1196                                        "quota options when quota turned on.\n");
1197                                 return 0;
1198                         }
1199                         qname = match_strdup(&args[0]);
1200                         if (!qname) {
1201                                 printk(KERN_ERR
1202                                         "EXT4-fs: not enough memory for "
1203                                         "storing quotafile name.\n");
1204                                 return 0;
1205                         }
1206                         if (sbi->s_qf_names[qtype] &&
1207                             strcmp(sbi->s_qf_names[qtype], qname)) {
1208                                 printk(KERN_ERR
1209                                         "EXT4-fs: %s quota file already "
1210                                         "specified.\n", QTYPE2NAME(qtype));
1211                                 kfree(qname);
1212                                 return 0;
1213                         }
1214                         sbi->s_qf_names[qtype] = qname;
1215                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1216                                 printk(KERN_ERR
1217                                         "EXT4-fs: quotafile must be on "
1218                                         "filesystem root.\n");
1219                                 kfree(sbi->s_qf_names[qtype]);
1220                                 sbi->s_qf_names[qtype] = NULL;
1221                                 return 0;
1222                         }
1223                         set_opt(sbi->s_mount_opt, QUOTA);
1224                         break;
1225                 case Opt_offusrjquota:
1226                         qtype = USRQUOTA;
1227                         goto clear_qf_name;
1228                 case Opt_offgrpjquota:
1229                         qtype = GRPQUOTA;
1230 clear_qf_name:
1231                         if ((sb_any_quota_enabled(sb) ||
1232                              sb_any_quota_suspended(sb)) &&
1233                             sbi->s_qf_names[qtype]) {
1234                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1235                                         "journaled quota options when "
1236                                         "quota turned on.\n");
1237                                 return 0;
1238                         }
1239                         /*
1240                          * The space will be released later when all options
1241                          * are confirmed to be correct
1242                          */
1243                         sbi->s_qf_names[qtype] = NULL;
1244                         break;
1245                 case Opt_jqfmt_vfsold:
1246                         qfmt = QFMT_VFS_OLD;
1247                         goto set_qf_format;
1248                 case Opt_jqfmt_vfsv0:
1249                         qfmt = QFMT_VFS_V0;
1250 set_qf_format:
1251                         if ((sb_any_quota_enabled(sb) ||
1252                              sb_any_quota_suspended(sb)) &&
1253                             sbi->s_jquota_fmt != qfmt) {
1254                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1255                                         "journaled quota options when "
1256                                         "quota turned on.\n");
1257                                 return 0;
1258                         }
1259                         sbi->s_jquota_fmt = qfmt;
1260                         break;
1261                 case Opt_quota:
1262                 case Opt_usrquota:
1263                         set_opt(sbi->s_mount_opt, QUOTA);
1264                         set_opt(sbi->s_mount_opt, USRQUOTA);
1265                         break;
1266                 case Opt_grpquota:
1267                         set_opt(sbi->s_mount_opt, QUOTA);
1268                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1269                         break;
1270                 case Opt_noquota:
1271                         if (sb_any_quota_enabled(sb)) {
1272                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1273                                         "options when quota turned on.\n");
1274                                 return 0;
1275                         }
1276                         clear_opt(sbi->s_mount_opt, QUOTA);
1277                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1278                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1279                         break;
1280 #else
1281                 case Opt_quota:
1282                 case Opt_usrquota:
1283                 case Opt_grpquota:
1284                         printk(KERN_ERR
1285                                 "EXT4-fs: quota options not supported.\n");
1286                         break;
1287                 case Opt_usrjquota:
1288                 case Opt_grpjquota:
1289                 case Opt_offusrjquota:
1290                 case Opt_offgrpjquota:
1291                 case Opt_jqfmt_vfsold:
1292                 case Opt_jqfmt_vfsv0:
1293                         printk(KERN_ERR
1294                                 "EXT4-fs: journaled quota options not "
1295                                 "supported.\n");
1296                         break;
1297                 case Opt_noquota:
1298                         break;
1299 #endif
1300                 case Opt_abort:
1301                         set_opt(sbi->s_mount_opt, ABORT);
1302                         break;
1303                 case Opt_barrier:
1304                         if (match_int(&args[0], &option))
1305                                 return 0;
1306                         if (option)
1307                                 set_opt(sbi->s_mount_opt, BARRIER);
1308                         else
1309                                 clear_opt(sbi->s_mount_opt, BARRIER);
1310                         break;
1311                 case Opt_ignore:
1312                         break;
1313                 case Opt_resize:
1314                         if (!is_remount) {
1315                                 printk("EXT4-fs: resize option only available "
1316                                         "for remount\n");
1317                                 return 0;
1318                         }
1319                         if (match_int(&args[0], &option) != 0)
1320                                 return 0;
1321                         *n_blocks_count = option;
1322                         break;
1323                 case Opt_nobh:
1324                         set_opt(sbi->s_mount_opt, NOBH);
1325                         break;
1326                 case Opt_bh:
1327                         clear_opt(sbi->s_mount_opt, NOBH);
1328                         break;
1329                 case Opt_extents:
1330                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1331                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1332                                 ext4_warning(sb, __func__,
1333                                         "extents feature not enabled "
1334                                         "on this filesystem, use tune2fs\n");
1335                                 return 0;
1336                         }
1337                         set_opt(sbi->s_mount_opt, EXTENTS);
1338                         break;
1339                 case Opt_noextents:
1340                         /*
1341                          * When e2fsprogs support resizing an already existing
1342                          * ext3 file system to greater than 2**32 we need to
1343                          * add support to block allocator to handle growing
1344                          * already existing block  mapped inode so that blocks
1345                          * allocated for them fall within 2**32
1346                          */
1347                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1348                         if (last_block  > 0xffffffffULL) {
1349                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1350                                                 "large to mount with "
1351                                                 "-o noextents options\n");
1352                                 return 0;
1353                         }
1354                         clear_opt(sbi->s_mount_opt, EXTENTS);
1355                         break;
1356                 case Opt_i_version:
1357                         set_opt(sbi->s_mount_opt, I_VERSION);
1358                         sb->s_flags |= MS_I_VERSION;
1359                         break;
1360                 case Opt_nodelalloc:
1361                         clear_opt(sbi->s_mount_opt, DELALLOC);
1362                         break;
1363                 case Opt_mballoc:
1364                         set_opt(sbi->s_mount_opt, MBALLOC);
1365                         break;
1366                 case Opt_nomballoc:
1367                         clear_opt(sbi->s_mount_opt, MBALLOC);
1368                         break;
1369                 case Opt_stripe:
1370                         if (match_int(&args[0], &option))
1371                                 return 0;
1372                         if (option < 0)
1373                                 return 0;
1374                         sbi->s_stripe = option;
1375                         break;
1376                 case Opt_delalloc:
1377                         set_opt(sbi->s_mount_opt, DELALLOC);
1378                         break;
1379                 default:
1380                         printk(KERN_ERR
1381                                "EXT4-fs: Unrecognized mount option \"%s\" "
1382                                "or missing value\n", p);
1383                         return 0;
1384                 }
1385         }
1386 #ifdef CONFIG_QUOTA
1387         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1388                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1389                      sbi->s_qf_names[USRQUOTA])
1390                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1391
1392                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1393                      sbi->s_qf_names[GRPQUOTA])
1394                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1395
1396                 if ((sbi->s_qf_names[USRQUOTA] &&
1397                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1398                     (sbi->s_qf_names[GRPQUOTA] &&
1399                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1400                         printk(KERN_ERR "EXT4-fs: old and new quota "
1401                                         "format mixing.\n");
1402                         return 0;
1403                 }
1404
1405                 if (!sbi->s_jquota_fmt) {
1406                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1407                                         "not specified.\n");
1408                         return 0;
1409                 }
1410         } else {
1411                 if (sbi->s_jquota_fmt) {
1412                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1413                                         "specified with no journaling "
1414                                         "enabled.\n");
1415                         return 0;
1416                 }
1417         }
1418 #endif
1419         return 1;
1420 }
1421
1422 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1423                             int read_only)
1424 {
1425         struct ext4_sb_info *sbi = EXT4_SB(sb);
1426         int res = 0;
1427
1428         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1429                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1430                        "forcing read-only mode\n");
1431                 res = MS_RDONLY;
1432         }
1433         if (read_only)
1434                 return res;
1435         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1436                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1437                        "running e2fsck is recommended\n");
1438         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1439                 printk(KERN_WARNING
1440                        "EXT4-fs warning: mounting fs with errors, "
1441                        "running e2fsck is recommended\n");
1442         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1443                  le16_to_cpu(es->s_mnt_count) >=
1444                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1445                 printk(KERN_WARNING
1446                        "EXT4-fs warning: maximal mount count reached, "
1447                        "running e2fsck is recommended\n");
1448         else if (le32_to_cpu(es->s_checkinterval) &&
1449                 (le32_to_cpu(es->s_lastcheck) +
1450                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1451                 printk(KERN_WARNING
1452                        "EXT4-fs warning: checktime reached, "
1453                        "running e2fsck is recommended\n");
1454 #if 0
1455                 /* @@@ We _will_ want to clear the valid bit if we find
1456                  * inconsistencies, to force a fsck at reboot.  But for
1457                  * a plain journaled filesystem we can keep it set as
1458                  * valid forever! :)
1459                  */
1460         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1461 #endif
1462         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1463                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1464         le16_add_cpu(&es->s_mnt_count, 1);
1465         es->s_mtime = cpu_to_le32(get_seconds());
1466         ext4_update_dynamic_rev(sb);
1467         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1468
1469         ext4_commit_super(sb, es, 1);
1470         if (test_opt(sb, DEBUG))
1471                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1472                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1473                         sb->s_blocksize,
1474                         sbi->s_groups_count,
1475                         EXT4_BLOCKS_PER_GROUP(sb),
1476                         EXT4_INODES_PER_GROUP(sb),
1477                         sbi->s_mount_opt);
1478
1479         printk(KERN_INFO "EXT4 FS on %s, %s journal on %s\n",
1480                sb->s_id, EXT4_SB(sb)->s_journal->j_inode ? "internal" :
1481                "external", EXT4_SB(sb)->s_journal->j_devname);
1482         return res;
1483 }
1484
1485 static int ext4_fill_flex_info(struct super_block *sb)
1486 {
1487         struct ext4_sb_info *sbi = EXT4_SB(sb);
1488         struct ext4_group_desc *gdp = NULL;
1489         struct buffer_head *bh;
1490         ext4_group_t flex_group_count;
1491         ext4_group_t flex_group;
1492         int groups_per_flex = 0;
1493         __u64 block_bitmap = 0;
1494         int i;
1495
1496         if (!sbi->s_es->s_log_groups_per_flex) {
1497                 sbi->s_log_groups_per_flex = 0;
1498                 return 1;
1499         }
1500
1501         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1502         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1503
1504         /* We allocate both existing and potentially added groups */
1505         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1506                             ((sbi->s_es->s_reserved_gdt_blocks +1 ) <<
1507                               EXT4_DESC_PER_BLOCK_BITS(sb))) /
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 %lld bytes\n",
1718                                 __func__, inode->i_ino, inode->i_size);
1719                         jbd_debug(2, "truncating inode %lu to %lld 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                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2262         }
2263         if (err) {
2264                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2265                 goto failed_mount3;
2266         }
2267
2268         /* per fileystem reservation list head & lock */
2269         spin_lock_init(&sbi->s_rsv_window_lock);
2270         sbi->s_rsv_window_root = RB_ROOT;
2271         /* Add a single, static dummy reservation to the start of the
2272          * reservation window list --- it gives us a placeholder for
2273          * append-at-start-of-list which makes the allocation logic
2274          * _much_ simpler. */
2275         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2276         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
2277         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
2278         sbi->s_rsv_window_head.rsv_goal_size = 0;
2279         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
2280
2281         sbi->s_stripe = ext4_get_stripe_size(sbi);
2282
2283         /*
2284          * set up enough so that it can read an inode
2285          */
2286         sb->s_op = &ext4_sops;
2287         sb->s_export_op = &ext4_export_ops;
2288         sb->s_xattr = ext4_xattr_handlers;
2289 #ifdef CONFIG_QUOTA
2290         sb->s_qcop = &ext4_qctl_operations;
2291         sb->dq_op = &ext4_quota_operations;
2292 #endif
2293         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2294
2295         sb->s_root = NULL;
2296
2297         needs_recovery = (es->s_last_orphan != 0 ||
2298                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2299                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2300
2301         /*
2302          * The first inode we look at is the journal inode.  Don't try
2303          * root first: it may be modified in the journal!
2304          */
2305         if (!test_opt(sb, NOLOAD) &&
2306             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2307                 if (ext4_load_journal(sb, es, journal_devnum))
2308                         goto failed_mount3;
2309                 if (!(sb->s_flags & MS_RDONLY) &&
2310                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2311                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2312                                "ext4_fill_super: Journal transaction "
2313                                "%u is corrupt\n", sb->s_id,
2314                                EXT4_SB(sb)->s_journal->j_failed_commit);
2315                         if (test_opt(sb, ERRORS_RO)) {
2316                                 printk(KERN_CRIT
2317                                        "Mounting filesystem read-only\n");
2318                                 sb->s_flags |= MS_RDONLY;
2319                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2320                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2321                         }
2322                         if (test_opt(sb, ERRORS_PANIC)) {
2323                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2324                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2325                                 ext4_commit_super(sb, es, 1);
2326                                 printk(KERN_CRIT
2327                                        "EXT4-fs (device %s): mount failed\n",
2328                                       sb->s_id);
2329                                 goto failed_mount4;
2330                         }
2331                 }
2332         } else if (journal_inum) {
2333                 if (ext4_create_journal(sb, es, journal_inum))
2334                         goto failed_mount3;
2335         } else {
2336                 if (!silent)
2337                         printk(KERN_ERR
2338                                "ext4: No journal on filesystem on %s\n",
2339                                sb->s_id);
2340                 goto failed_mount3;
2341         }
2342
2343         if (ext4_blocks_count(es) > 0xffffffffULL &&
2344             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2345                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2346                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2347                 goto failed_mount4;
2348         }
2349
2350         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2351                 jbd2_journal_set_features(sbi->s_journal,
2352                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2353                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2354         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2355                 jbd2_journal_set_features(sbi->s_journal,
2356                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2357                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2358                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2359         } else {
2360                 jbd2_journal_clear_features(sbi->s_journal,
2361                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2362                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2363         }
2364
2365         /* We have now updated the journal if required, so we can
2366          * validate the data journaling mode. */
2367         switch (test_opt(sb, DATA_FLAGS)) {
2368         case 0:
2369                 /* No mode set, assume a default based on the journal
2370                  * capabilities: ORDERED_DATA if the journal can
2371                  * cope, else JOURNAL_DATA
2372                  */
2373                 if (jbd2_journal_check_available_features
2374                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2375                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2376                 else
2377                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2378                 break;
2379
2380         case EXT4_MOUNT_ORDERED_DATA:
2381         case EXT4_MOUNT_WRITEBACK_DATA:
2382                 if (!jbd2_journal_check_available_features
2383                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2384                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2385                                "requested data journaling mode\n");
2386                         goto failed_mount4;
2387                 }
2388         default:
2389                 break;
2390         }
2391
2392         if (test_opt(sb, NOBH)) {
2393                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2394                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2395                                 "its supported only with writeback mode\n");
2396                         clear_opt(sbi->s_mount_opt, NOBH);
2397                 }
2398         }
2399         /*
2400          * The jbd2_journal_load will have done any necessary log recovery,
2401          * so we can safely mount the rest of the filesystem now.
2402          */
2403
2404         root = ext4_iget(sb, EXT4_ROOT_INO);
2405         if (IS_ERR(root)) {
2406                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2407                 ret = PTR_ERR(root);
2408                 goto failed_mount4;
2409         }
2410         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2411                 iput(root);
2412                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2413                 goto failed_mount4;
2414         }
2415         sb->s_root = d_alloc_root(root);
2416         if (!sb->s_root) {
2417                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2418                 iput(root);
2419                 ret = -ENOMEM;
2420                 goto failed_mount4;
2421         }
2422
2423         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2424
2425         /* determine the minimum size of new large inodes, if present */
2426         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2427                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2428                                                      EXT4_GOOD_OLD_INODE_SIZE;
2429                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2430                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2431                         if (sbi->s_want_extra_isize <
2432                             le16_to_cpu(es->s_want_extra_isize))
2433                                 sbi->s_want_extra_isize =
2434                                         le16_to_cpu(es->s_want_extra_isize);
2435                         if (sbi->s_want_extra_isize <
2436                             le16_to_cpu(es->s_min_extra_isize))
2437                                 sbi->s_want_extra_isize =
2438                                         le16_to_cpu(es->s_min_extra_isize);
2439                 }
2440         }
2441         /* Check if enough inode space is available */
2442         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2443                                                         sbi->s_inode_size) {
2444                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2445                                                        EXT4_GOOD_OLD_INODE_SIZE;
2446                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2447                         "available.\n");
2448         }
2449
2450         /*
2451          * akpm: core read_super() calls in here with the superblock locked.
2452          * That deadlocks, because orphan cleanup needs to lock the superblock
2453          * in numerous places.  Here we just pop the lock - it's relatively
2454          * harmless, because we are now ready to accept write_super() requests,
2455          * and aviro says that's the only reason for hanging onto the
2456          * superblock lock.
2457          */
2458         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2459         ext4_orphan_cleanup(sb, es);
2460         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2461         if (needs_recovery)
2462                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2463         ext4_mark_recovery_complete(sb, es);
2464         printk(KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2465                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2466                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2467                "writeback");
2468
2469         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2470                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2471                                 "requested data journaling mode\n");
2472                 clear_opt(sbi->s_mount_opt, DELALLOC);
2473         } else if (test_opt(sb, DELALLOC))
2474                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2475
2476         ext4_ext_init(sb);
2477         ext4_mb_init(sb, needs_recovery);
2478
2479         lock_kernel();
2480         return 0;
2481
2482 cantfind_ext4:
2483         if (!silent)
2484                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2485                        sb->s_id);
2486         goto failed_mount;
2487
2488 failed_mount4:
2489         jbd2_journal_destroy(sbi->s_journal);
2490         sbi->s_journal = NULL;
2491 failed_mount3:
2492         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2493         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2494         percpu_counter_destroy(&sbi->s_dirs_counter);
2495         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2496 failed_mount2:
2497         for (i = 0; i < db_count; i++)
2498                 brelse(sbi->s_group_desc[i]);
2499         kfree(sbi->s_group_desc);
2500 failed_mount:
2501 #ifdef CONFIG_QUOTA
2502         for (i = 0; i < MAXQUOTAS; i++)
2503                 kfree(sbi->s_qf_names[i]);
2504 #endif
2505         ext4_blkdev_remove(sbi);
2506         brelse(bh);
2507 out_fail:
2508         sb->s_fs_info = NULL;
2509         kfree(sbi);
2510         lock_kernel();
2511         return ret;
2512 }
2513
2514 /*
2515  * Setup any per-fs journal parameters now.  We'll do this both on
2516  * initial mount, once the journal has been initialised but before we've
2517  * done any recovery; and again on any subsequent remount.
2518  */
2519 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2520 {
2521         struct ext4_sb_info *sbi = EXT4_SB(sb);
2522
2523         if (sbi->s_commit_interval)
2524                 journal->j_commit_interval = sbi->s_commit_interval;
2525         /* We could also set up an ext4-specific default for the commit
2526          * interval here, but for now we'll just fall back to the jbd
2527          * default. */
2528
2529         spin_lock(&journal->j_state_lock);
2530         if (test_opt(sb, BARRIER))
2531                 journal->j_flags |= JBD2_BARRIER;
2532         else
2533                 journal->j_flags &= ~JBD2_BARRIER;
2534         spin_unlock(&journal->j_state_lock);
2535 }
2536
2537 static journal_t *ext4_get_journal(struct super_block *sb,
2538                                    unsigned int journal_inum)
2539 {
2540         struct inode *journal_inode;
2541         journal_t *journal;
2542
2543         /* First, test for the existence of a valid inode on disk.  Bad
2544          * things happen if we iget() an unused inode, as the subsequent
2545          * iput() will try to delete it. */
2546
2547         journal_inode = ext4_iget(sb, journal_inum);
2548         if (IS_ERR(journal_inode)) {
2549                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2550                 return NULL;
2551         }
2552         if (!journal_inode->i_nlink) {
2553                 make_bad_inode(journal_inode);
2554                 iput(journal_inode);
2555                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2556                 return NULL;
2557         }
2558
2559         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
2560                   journal_inode, journal_inode->i_size);
2561         if (!S_ISREG(journal_inode->i_mode)) {
2562                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2563                 iput(journal_inode);
2564                 return NULL;
2565         }
2566
2567         journal = jbd2_journal_init_inode(journal_inode);
2568         if (!journal) {
2569                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2570                 iput(journal_inode);
2571                 return NULL;
2572         }
2573         journal->j_private = sb;
2574         ext4_init_journal_params(sb, journal);
2575         return journal;
2576 }
2577
2578 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2579                                        dev_t j_dev)
2580 {
2581         struct buffer_head *bh;
2582         journal_t *journal;
2583         ext4_fsblk_t start;
2584         ext4_fsblk_t len;
2585         int hblock, blocksize;
2586         ext4_fsblk_t sb_block;
2587         unsigned long offset;
2588         struct ext4_super_block *es;
2589         struct block_device *bdev;
2590
2591         bdev = ext4_blkdev_get(j_dev);
2592         if (bdev == NULL)
2593                 return NULL;
2594
2595         if (bd_claim(bdev, sb)) {
2596                 printk(KERN_ERR
2597                         "EXT4: failed to claim external journal device.\n");
2598                 blkdev_put(bdev);
2599                 return NULL;
2600         }
2601
2602         blocksize = sb->s_blocksize;
2603         hblock = bdev_hardsect_size(bdev);
2604         if (blocksize < hblock) {
2605                 printk(KERN_ERR
2606                         "EXT4-fs: blocksize too small for journal device.\n");
2607                 goto out_bdev;
2608         }
2609
2610         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2611         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2612         set_blocksize(bdev, blocksize);
2613         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2614                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2615                        "external journal\n");
2616                 goto out_bdev;
2617         }
2618
2619         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2620         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2621             !(le32_to_cpu(es->s_feature_incompat) &
2622               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2623                 printk(KERN_ERR "EXT4-fs: external journal has "
2624                                         "bad superblock\n");
2625                 brelse(bh);
2626                 goto out_bdev;
2627         }
2628
2629         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2630                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2631                 brelse(bh);
2632                 goto out_bdev;
2633         }
2634
2635         len = ext4_blocks_count(es);
2636         start = sb_block + 1;
2637         brelse(bh);     /* we're done with the superblock */
2638
2639         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2640                                         start, len, blocksize);
2641         if (!journal) {
2642                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2643                 goto out_bdev;
2644         }
2645         journal->j_private = sb;
2646         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2647         wait_on_buffer(journal->j_sb_buffer);
2648         if (!buffer_uptodate(journal->j_sb_buffer)) {
2649                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2650                 goto out_journal;
2651         }
2652         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2653                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2654                                         "user (unsupported) - %d\n",
2655                         be32_to_cpu(journal->j_superblock->s_nr_users));
2656                 goto out_journal;
2657         }
2658         EXT4_SB(sb)->journal_bdev = bdev;
2659         ext4_init_journal_params(sb, journal);
2660         return journal;
2661 out_journal:
2662         jbd2_journal_destroy(journal);
2663 out_bdev:
2664         ext4_blkdev_put(bdev);
2665         return NULL;
2666 }
2667
2668 static int ext4_load_journal(struct super_block *sb,
2669                              struct ext4_super_block *es,
2670                              unsigned long journal_devnum)
2671 {
2672         journal_t *journal;
2673         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2674         dev_t journal_dev;
2675         int err = 0;
2676         int really_read_only;
2677
2678         if (journal_devnum &&
2679             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2680                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2681                         "numbers have changed\n");
2682                 journal_dev = new_decode_dev(journal_devnum);
2683         } else
2684                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2685
2686         really_read_only = bdev_read_only(sb->s_bdev);
2687
2688         /*
2689          * Are we loading a blank journal or performing recovery after a
2690          * crash?  For recovery, we need to check in advance whether we
2691          * can get read-write access to the device.
2692          */
2693
2694         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2695                 if (sb->s_flags & MS_RDONLY) {
2696                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2697                                         "required on readonly filesystem.\n");
2698                         if (really_read_only) {
2699                                 printk(KERN_ERR "EXT4-fs: write access "
2700                                         "unavailable, cannot proceed.\n");
2701                                 return -EROFS;
2702                         }
2703                         printk(KERN_INFO "EXT4-fs: write access will "
2704                                "be enabled during recovery.\n");
2705                 }
2706         }
2707
2708         if (journal_inum && journal_dev) {
2709                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2710                        "and inode journals!\n");
2711                 return -EINVAL;
2712         }
2713
2714         if (journal_inum) {
2715                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2716                         return -EINVAL;
2717         } else {
2718                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2719                         return -EINVAL;
2720         }
2721
2722         if (journal->j_flags & JBD2_BARRIER)
2723                 printk(KERN_INFO "EXT4-fs: barriers enabled\n");
2724         else
2725                 printk(KERN_INFO "EXT4-fs: barriers disabled\n");
2726
2727         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2728                 err = jbd2_journal_update_format(journal);
2729                 if (err)  {
2730                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2731                         jbd2_journal_destroy(journal);
2732                         return err;
2733                 }
2734         }
2735
2736         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2737                 err = jbd2_journal_wipe(journal, !really_read_only);
2738         if (!err)
2739                 err = jbd2_journal_load(journal);
2740
2741         if (err) {
2742                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2743                 jbd2_journal_destroy(journal);
2744                 return err;
2745         }
2746
2747         EXT4_SB(sb)->s_journal = journal;
2748         ext4_clear_journal_err(sb, es);
2749
2750         if (journal_devnum &&
2751             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2752                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2753                 sb->s_dirt = 1;
2754
2755                 /* Make sure we flush the recovery flag to disk. */
2756                 ext4_commit_super(sb, es, 1);
2757         }
2758
2759         return 0;
2760 }
2761
2762 static int ext4_create_journal(struct super_block *sb,
2763                                struct ext4_super_block *es,
2764                                unsigned int journal_inum)
2765 {
2766         journal_t *journal;
2767         int err;
2768
2769         if (sb->s_flags & MS_RDONLY) {
2770                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2771                                 "create journal.\n");
2772                 return -EROFS;
2773         }
2774
2775         journal = ext4_get_journal(sb, journal_inum);
2776         if (!journal)
2777                 return -EINVAL;
2778
2779         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2780                journal_inum);
2781
2782         err = jbd2_journal_create(journal);
2783         if (err) {
2784                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2785                 jbd2_journal_destroy(journal);
2786                 return -EIO;
2787         }
2788
2789         EXT4_SB(sb)->s_journal = journal;
2790
2791         ext4_update_dynamic_rev(sb);
2792         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2793         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2794
2795         es->s_journal_inum = cpu_to_le32(journal_inum);
2796         sb->s_dirt = 1;
2797
2798         /* Make sure we flush the recovery flag to disk. */
2799         ext4_commit_super(sb, es, 1);
2800
2801         return 0;
2802 }
2803
2804 static void ext4_commit_super(struct super_block *sb,
2805                               struct ext4_super_block *es, int sync)
2806 {
2807         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2808
2809         if (!sbh)
2810                 return;
2811         es->s_wtime = cpu_to_le32(get_seconds());
2812         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2813         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2814         BUFFER_TRACE(sbh, "marking dirty");
2815         mark_buffer_dirty(sbh);
2816         if (sync)
2817                 sync_dirty_buffer(sbh);
2818 }
2819
2820
2821 /*
2822  * Have we just finished recovery?  If so, and if we are mounting (or
2823  * remounting) the filesystem readonly, then we will end up with a
2824  * consistent fs on disk.  Record that fact.
2825  */
2826 static void ext4_mark_recovery_complete(struct super_block *sb,
2827                                         struct ext4_super_block *es)
2828 {
2829         journal_t *journal = EXT4_SB(sb)->s_journal;
2830
2831         jbd2_journal_lock_updates(journal);
2832         jbd2_journal_flush(journal);
2833         lock_super(sb);
2834         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2835             sb->s_flags & MS_RDONLY) {
2836                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2837                 sb->s_dirt = 0;
2838                 ext4_commit_super(sb, es, 1);
2839         }
2840         unlock_super(sb);
2841         jbd2_journal_unlock_updates(journal);
2842 }
2843
2844 /*
2845  * If we are mounting (or read-write remounting) a filesystem whose journal
2846  * has recorded an error from a previous lifetime, move that error to the
2847  * main filesystem now.
2848  */
2849 static void ext4_clear_journal_err(struct super_block *sb,
2850                                    struct ext4_super_block *es)
2851 {
2852         journal_t *journal;
2853         int j_errno;
2854         const char *errstr;
2855
2856         journal = EXT4_SB(sb)->s_journal;
2857
2858         /*
2859          * Now check for any error status which may have been recorded in the
2860          * journal by a prior ext4_error() or ext4_abort()
2861          */
2862
2863         j_errno = jbd2_journal_errno(journal);
2864         if (j_errno) {
2865                 char nbuf[16];
2866
2867                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2868                 ext4_warning(sb, __func__, "Filesystem error recorded "
2869                              "from previous mount: %s", errstr);
2870                 ext4_warning(sb, __func__, "Marking fs in need of "
2871                              "filesystem check.");
2872
2873                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2874                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2875                 ext4_commit_super(sb, es, 1);
2876
2877                 jbd2_journal_clear_err(journal);
2878         }
2879 }
2880
2881 /*
2882  * Force the running and committing transactions to commit,
2883  * and wait on the commit.
2884  */
2885 int ext4_force_commit(struct super_block *sb)
2886 {
2887         journal_t *journal;
2888         int ret;
2889
2890         if (sb->s_flags & MS_RDONLY)
2891                 return 0;
2892
2893         journal = EXT4_SB(sb)->s_journal;
2894         sb->s_dirt = 0;
2895         ret = ext4_journal_force_commit(journal);
2896         return ret;
2897 }
2898
2899 /*
2900  * Ext4 always journals updates to the superblock itself, so we don't
2901  * have to propagate any other updates to the superblock on disk at this
2902  * point.  Just start an async writeback to get the buffers on their way
2903  * to the disk.
2904  *
2905  * This implicitly triggers the writebehind on sync().
2906  */
2907
2908 static void ext4_write_super(struct super_block *sb)
2909 {
2910         if (mutex_trylock(&sb->s_lock) != 0)
2911                 BUG();
2912         sb->s_dirt = 0;
2913 }
2914
2915 static int ext4_sync_fs(struct super_block *sb, int wait)
2916 {
2917         tid_t target;
2918
2919         sb->s_dirt = 0;
2920         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2921                 if (wait)
2922                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2923         }
2924         return 0;
2925 }
2926
2927 /*
2928  * LVM calls this function before a (read-only) snapshot is created.  This
2929  * gives us a chance to flush the journal completely and mark the fs clean.
2930  */
2931 static void ext4_write_super_lockfs(struct super_block *sb)
2932 {
2933         sb->s_dirt = 0;
2934
2935         if (!(sb->s_flags & MS_RDONLY)) {
2936                 journal_t *journal = EXT4_SB(sb)->s_journal;
2937
2938                 /* Now we set up the journal barrier. */
2939                 jbd2_journal_lock_updates(journal);
2940                 jbd2_journal_flush(journal);
2941
2942                 /* Journal blocked and flushed, clear needs_recovery flag. */
2943                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2944                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2945         }
2946 }
2947
2948 /*
2949  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2950  * flag here, even though the filesystem is not technically dirty yet.
2951  */
2952 static void ext4_unlockfs(struct super_block *sb)
2953 {
2954         if (!(sb->s_flags & MS_RDONLY)) {
2955                 lock_super(sb);
2956                 /* Reser the needs_recovery flag before the fs is unlocked. */
2957                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2958                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2959                 unlock_super(sb);
2960                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2961         }
2962 }
2963
2964 static int ext4_remount(struct super_block *sb, int *flags, char *data)
2965 {
2966         struct ext4_super_block *es;
2967         struct ext4_sb_info *sbi = EXT4_SB(sb);
2968         ext4_fsblk_t n_blocks_count = 0;
2969         unsigned long old_sb_flags;
2970         struct ext4_mount_options old_opts;
2971         ext4_group_t g;
2972         int err;
2973 #ifdef CONFIG_QUOTA
2974         int i;
2975 #endif
2976
2977         /* Store the original options */
2978         old_sb_flags = sb->s_flags;
2979         old_opts.s_mount_opt = sbi->s_mount_opt;
2980         old_opts.s_resuid = sbi->s_resuid;
2981         old_opts.s_resgid = sbi->s_resgid;
2982         old_opts.s_commit_interval = sbi->s_commit_interval;
2983 #ifdef CONFIG_QUOTA
2984         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2985         for (i = 0; i < MAXQUOTAS; i++)
2986                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2987 #endif
2988
2989         /*
2990          * Allow the "check" option to be passed as a remount option.
2991          */
2992         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2993                 err = -EINVAL;
2994                 goto restore_opts;
2995         }
2996
2997         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2998                 ext4_abort(sb, __func__, "Abort forced by user");
2999
3000         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3001                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3002
3003         es = sbi->s_es;
3004
3005         ext4_init_journal_params(sb, sbi->s_journal);
3006
3007         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3008                 n_blocks_count > ext4_blocks_count(es)) {
3009                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3010                         err = -EROFS;
3011                         goto restore_opts;
3012                 }
3013
3014                 if (*flags & MS_RDONLY) {
3015                         /*
3016                          * First of all, the unconditional stuff we have to do
3017                          * to disable replay of the journal when we next remount
3018                          */
3019                         sb->s_flags |= MS_RDONLY;
3020
3021                         /*
3022                          * OK, test if we are remounting a valid rw partition
3023                          * readonly, and if so set the rdonly flag and then
3024                          * mark the partition as valid again.
3025                          */
3026                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3027                             (sbi->s_mount_state & EXT4_VALID_FS))
3028                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3029
3030                         /*
3031                          * We have to unlock super so that we can wait for
3032                          * transactions.
3033                          */
3034                         unlock_super(sb);
3035                         ext4_mark_recovery_complete(sb, es);
3036                         lock_super(sb);
3037                 } else {
3038                         __le32 ret;
3039                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3040                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3041                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3042                                        "remount RDWR because of unsupported "
3043                                        "optional features (%x).\n",
3044                                        sb->s_id, le32_to_cpu(ret));
3045                                 err = -EROFS;
3046                                 goto restore_opts;
3047                         }
3048
3049                         /*
3050                          * Make sure the group descriptor checksums
3051                          * are sane.  If they aren't, refuse to
3052                          * remount r/w.
3053                          */
3054                         for (g = 0; g < sbi->s_groups_count; g++) {
3055                                 struct ext4_group_desc *gdp =
3056                                         ext4_get_group_desc(sb, g, NULL);
3057
3058                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3059                                         printk(KERN_ERR
3060                "EXT4-fs: ext4_remount: "
3061                 "Checksum for group %lu failed (%u!=%u)\n",
3062                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3063                                                le16_to_cpu(gdp->bg_checksum));
3064                                         err = -EINVAL;
3065                                         goto restore_opts;
3066                                 }
3067                         }
3068
3069                         /*
3070                          * If we have an unprocessed orphan list hanging
3071                          * around from a previously readonly bdev mount,
3072                          * require a full umount/remount for now.
3073                          */
3074                         if (es->s_last_orphan) {
3075                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3076                                        "remount RDWR because of unprocessed "
3077                                        "orphan inode list.  Please "
3078                                        "umount/remount instead.\n",
3079                                        sb->s_id);
3080                                 err = -EINVAL;
3081                                 goto restore_opts;
3082                         }
3083
3084                         /*
3085                          * Mounting a RDONLY partition read-write, so reread
3086                          * and store the current valid flag.  (It may have
3087                          * been changed by e2fsck since we originally mounted
3088                          * the partition.)
3089                          */
3090                         ext4_clear_journal_err(sb, es);
3091                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3092                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3093                                 goto restore_opts;
3094                         if (!ext4_setup_super(sb, es, 0))
3095                                 sb->s_flags &= ~MS_RDONLY;
3096                 }
3097         }
3098 #ifdef CONFIG_QUOTA
3099         /* Release old quota file names */
3100         for (i = 0; i < MAXQUOTAS; i++)
3101                 if (old_opts.s_qf_names[i] &&
3102                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3103                         kfree(old_opts.s_qf_names[i]);
3104 #endif
3105         return 0;
3106 restore_opts:
3107         sb->s_flags = old_sb_flags;
3108         sbi->s_mount_opt = old_opts.s_mount_opt;
3109         sbi->s_resuid = old_opts.s_resuid;
3110         sbi->s_resgid = old_opts.s_resgid;
3111         sbi->s_commit_interval = old_opts.s_commit_interval;
3112 #ifdef CONFIG_QUOTA
3113         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3114         for (i = 0; i < MAXQUOTAS; i++) {
3115                 if (sbi->s_qf_names[i] &&
3116                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3117                         kfree(sbi->s_qf_names[i]);
3118                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3119         }
3120 #endif
3121         return err;
3122 }
3123
3124 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3125 {
3126         struct super_block *sb = dentry->d_sb;
3127         struct ext4_sb_info *sbi = EXT4_SB(sb);
3128         struct ext4_super_block *es = sbi->s_es;
3129         u64 fsid;
3130
3131         if (test_opt(sb, MINIX_DF)) {
3132                 sbi->s_overhead_last = 0;
3133         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3134                 ext4_group_t ngroups = sbi->s_groups_count, i;
3135                 ext4_fsblk_t overhead = 0;
3136                 smp_rmb();
3137
3138                 /*
3139                  * Compute the overhead (FS structures).  This is constant
3140                  * for a given filesystem unless the number of block groups
3141                  * changes so we cache the previous value until it does.
3142                  */
3143
3144                 /*
3145                  * All of the blocks before first_data_block are
3146                  * overhead
3147                  */
3148                 overhead = le32_to_cpu(es->s_first_data_block);
3149
3150                 /*
3151                  * Add the overhead attributed to the superblock and
3152                  * block group descriptors.  If the sparse superblocks
3153                  * feature is turned on, then not all groups have this.
3154                  */
3155                 for (i = 0; i < ngroups; i++) {
3156                         overhead += ext4_bg_has_super(sb, i) +
3157                                 ext4_bg_num_gdb(sb, i);
3158                         cond_resched();
3159                 }
3160
3161                 /*
3162                  * Every block group has an inode bitmap, a block
3163                  * bitmap, and an inode table.
3164                  */
3165                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3166                 sbi->s_overhead_last = overhead;
3167                 smp_wmb();
3168                 sbi->s_blocks_last = ext4_blocks_count(es);
3169         }
3170
3171         buf->f_type = EXT4_SUPER_MAGIC;
3172         buf->f_bsize = sb->s_blocksize;
3173         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3174         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3175                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3176         ext4_free_blocks_count_set(es, buf->f_bfree);
3177         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3178         if (buf->f_bfree < ext4_r_blocks_count(es))
3179                 buf->f_bavail = 0;
3180         buf->f_files = le32_to_cpu(es->s_inodes_count);
3181         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3182         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3183         buf->f_namelen = EXT4_NAME_LEN;
3184         fsid = le64_to_cpup((void *)es->s_uuid) ^
3185                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3186         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3187         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3188         return 0;
3189 }
3190
3191 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3192  * is locked for write. Otherwise the are possible deadlocks:
3193  * Process 1                         Process 2
3194  * ext4_create()                     quota_sync()
3195  *   jbd2_journal_start()                   write_dquot()
3196  *   DQUOT_INIT()                        down(dqio_mutex)
3197  *     down(dqio_mutex)                    jbd2_journal_start()
3198  *
3199  */
3200
3201 #ifdef CONFIG_QUOTA
3202
3203 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3204 {
3205         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3206 }
3207
3208 static int ext4_dquot_initialize(struct inode *inode, int type)
3209 {
3210         handle_t *handle;
3211         int ret, err;
3212
3213         /* We may create quota structure so we need to reserve enough blocks */
3214         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3215         if (IS_ERR(handle))
3216                 return PTR_ERR(handle);
3217         ret = dquot_initialize(inode, type);
3218         err = ext4_journal_stop(handle);
3219         if (!ret)
3220                 ret = err;
3221         return ret;
3222 }
3223
3224 static int ext4_dquot_drop(struct inode *inode)
3225 {
3226         handle_t *handle;
3227         int ret, err;
3228
3229         /* We may delete quota structure so we need to reserve enough blocks */
3230         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3231         if (IS_ERR(handle)) {
3232                 /*
3233                  * We call dquot_drop() anyway to at least release references
3234                  * to quota structures so that umount does not hang.
3235                  */
3236                 dquot_drop(inode);
3237                 return PTR_ERR(handle);
3238         }
3239         ret = dquot_drop(inode);
3240         err = ext4_journal_stop(handle);
3241         if (!ret)
3242                 ret = err;
3243         return ret;
3244 }
3245
3246 static int ext4_write_dquot(struct dquot *dquot)
3247 {
3248         int ret, err;
3249         handle_t *handle;
3250         struct inode *inode;
3251
3252         inode = dquot_to_inode(dquot);
3253         handle = ext4_journal_start(inode,
3254                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3255         if (IS_ERR(handle))
3256                 return PTR_ERR(handle);
3257         ret = dquot_commit(dquot);
3258         err = ext4_journal_stop(handle);
3259         if (!ret)
3260                 ret = err;
3261         return ret;
3262 }
3263
3264 static int ext4_acquire_dquot(struct dquot *dquot)
3265 {
3266         int ret, err;
3267         handle_t *handle;
3268
3269         handle = ext4_journal_start(dquot_to_inode(dquot),
3270                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3271         if (IS_ERR(handle))
3272                 return PTR_ERR(handle);
3273         ret = dquot_acquire(dquot);
3274         err = ext4_journal_stop(handle);
3275         if (!ret)
3276                 ret = err;
3277         return ret;
3278 }
3279
3280 static int ext4_release_dquot(struct dquot *dquot)
3281 {
3282         int ret, err;
3283         handle_t *handle;
3284
3285         handle = ext4_journal_start(dquot_to_inode(dquot),
3286                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3287         if (IS_ERR(handle)) {
3288                 /* Release dquot anyway to avoid endless cycle in dqput() */
3289                 dquot_release(dquot);
3290                 return PTR_ERR(handle);
3291         }
3292         ret = dquot_release(dquot);
3293         err = ext4_journal_stop(handle);
3294         if (!ret)
3295                 ret = err;
3296         return ret;
3297 }
3298
3299 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3300 {
3301         /* Are we journaling quotas? */
3302         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3303             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3304                 dquot_mark_dquot_dirty(dquot);
3305                 return ext4_write_dquot(dquot);
3306         } else {
3307                 return dquot_mark_dquot_dirty(dquot);
3308         }
3309 }
3310
3311 static int ext4_write_info(struct super_block *sb, int type)
3312 {
3313         int ret, err;
3314         handle_t *handle;
3315
3316         /* Data block + inode block */
3317         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3318         if (IS_ERR(handle))
3319                 return PTR_ERR(handle);
3320         ret = dquot_commit_info(sb, type);
3321         err = ext4_journal_stop(handle);
3322         if (!ret)
3323                 ret = err;
3324         return ret;
3325 }
3326
3327 /*
3328  * Turn on quotas during mount time - we need to find
3329  * the quota file and such...
3330  */
3331 static int ext4_quota_on_mount(struct super_block *sb, int type)
3332 {
3333         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3334                         EXT4_SB(sb)->s_jquota_fmt, type);
3335 }
3336
3337 /*
3338  * Standard function to be called on quota_on
3339  */
3340 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3341                          char *path, int remount)
3342 {
3343         int err;
3344         struct nameidata nd;
3345
3346         if (!test_opt(sb, QUOTA))
3347                 return -EINVAL;
3348         /* When remounting, no checks are needed and in fact, path is NULL */
3349         if (remount)
3350                 return vfs_quota_on(sb, type, format_id, path, remount);
3351
3352         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
3353         if (err)
3354                 return err;
3355
3356         /* Quotafile not on the same filesystem? */
3357         if (nd.path.mnt->mnt_sb != sb) {
3358                 path_put(&nd.path);
3359                 return -EXDEV;
3360         }
3361         /* Journaling quota? */
3362         if (EXT4_SB(sb)->s_qf_names[type]) {
3363                 /* Quotafile not in fs root? */
3364                 if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
3365                         printk(KERN_WARNING
3366                                 "EXT4-fs: Quota file not on filesystem root. "
3367                                 "Journaled quota will not work.\n");
3368         }
3369
3370         /*
3371          * When we journal data on quota file, we have to flush journal to see
3372          * all updates to the file when we bypass pagecache...
3373          */
3374         if (ext4_should_journal_data(nd.path.dentry->d_inode)) {
3375                 /*
3376                  * We don't need to lock updates but journal_flush() could
3377                  * otherwise be livelocked...
3378                  */
3379                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3380                 jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3381                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3382         }
3383
3384         err = vfs_quota_on_path(sb, type, format_id, &nd.path);
3385         path_put(&nd.path);
3386         return err;
3387 }
3388
3389 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3390  * acquiring the locks... As quota files are never truncated and quota code
3391  * itself serializes the operations (and noone else should touch the files)
3392  * we don't have to be afraid of races */
3393 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3394                                size_t len, loff_t off)
3395 {
3396         struct inode *inode = sb_dqopt(sb)->files[type];
3397         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3398         int err = 0;
3399         int offset = off & (sb->s_blocksize - 1);
3400         int tocopy;
3401         size_t toread;
3402         struct buffer_head *bh;
3403         loff_t i_size = i_size_read(inode);
3404
3405         if (off > i_size)
3406                 return 0;
3407         if (off+len > i_size)
3408                 len = i_size-off;
3409         toread = len;
3410         while (toread > 0) {
3411                 tocopy = sb->s_blocksize - offset < toread ?
3412                                 sb->s_blocksize - offset : toread;
3413                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3414                 if (err)
3415                         return err;
3416                 if (!bh)        /* A hole? */
3417                         memset(data, 0, tocopy);
3418                 else
3419                         memcpy(data, bh->b_data+offset, tocopy);
3420                 brelse(bh);
3421                 offset = 0;
3422                 toread -= tocopy;
3423                 data += tocopy;
3424                 blk++;
3425         }
3426         return len;
3427 }
3428
3429 /* Write to quotafile (we know the transaction is already started and has
3430  * enough credits) */
3431 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3432                                 const char *data, size_t len, loff_t off)
3433 {
3434         struct inode *inode = sb_dqopt(sb)->files[type];
3435         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3436         int err = 0;
3437         int offset = off & (sb->s_blocksize - 1);
3438         int tocopy;
3439         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3440         size_t towrite = len;
3441         struct buffer_head *bh;
3442         handle_t *handle = journal_current_handle();
3443
3444         if (!handle) {
3445                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%llu, len=%llu)"
3446                         " cancelled because transaction is not started.\n",
3447                         (unsigned long long)off, (unsigned long long)len);
3448                 return -EIO;
3449         }
3450         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3451         while (towrite > 0) {
3452                 tocopy = sb->s_blocksize - offset < towrite ?
3453                                 sb->s_blocksize - offset : towrite;
3454                 bh = ext4_bread(handle, inode, blk, 1, &err);
3455                 if (!bh)
3456                         goto out;
3457                 if (journal_quota) {
3458                         err = ext4_journal_get_write_access(handle, bh);
3459                         if (err) {
3460                                 brelse(bh);
3461                                 goto out;
3462                         }
3463                 }
3464                 lock_buffer(bh);
3465                 memcpy(bh->b_data+offset, data, tocopy);
3466                 flush_dcache_page(bh->b_page);
3467                 unlock_buffer(bh);
3468                 if (journal_quota)
3469                         err = ext4_journal_dirty_metadata(handle, bh);
3470                 else {
3471                         /* Always do at least ordered writes for quotas */
3472                         err = ext4_jbd2_file_inode(handle, inode);
3473                         mark_buffer_dirty(bh);
3474                 }
3475                 brelse(bh);
3476                 if (err)
3477                         goto out;
3478                 offset = 0;
3479                 towrite -= tocopy;
3480                 data += tocopy;
3481                 blk++;
3482         }
3483 out:
3484         if (len == towrite) {
3485                 mutex_unlock(&inode->i_mutex);
3486                 return err;
3487         }
3488         if (inode->i_size < off+len-towrite) {
3489                 i_size_write(inode, off+len-towrite);
3490                 EXT4_I(inode)->i_disksize = inode->i_size;
3491         }
3492         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3493         ext4_mark_inode_dirty(handle, inode);
3494         mutex_unlock(&inode->i_mutex);
3495         return len - towrite;
3496 }
3497
3498 #endif
3499
3500 static int ext4_get_sb(struct file_system_type *fs_type,
3501         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3502 {
3503         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3504 }
3505
3506 static struct file_system_type ext4dev_fs_type = {
3507         .owner          = THIS_MODULE,
3508         .name           = "ext4dev",
3509         .get_sb         = ext4_get_sb,
3510         .kill_sb        = kill_block_super,
3511         .fs_flags       = FS_REQUIRES_DEV,
3512 };
3513
3514 static int __init init_ext4_fs(void)
3515 {
3516         int err;
3517
3518         err = init_ext4_mballoc();
3519         if (err)
3520                 return err;
3521
3522         err = init_ext4_xattr();
3523         if (err)
3524                 goto out2;
3525         err = init_inodecache();
3526         if (err)
3527                 goto out1;
3528         err = register_filesystem(&ext4dev_fs_type);
3529         if (err)
3530                 goto out;
3531         return 0;
3532 out:
3533         destroy_inodecache();
3534 out1:
3535         exit_ext4_xattr();
3536 out2:
3537         exit_ext4_mballoc();
3538         return err;
3539 }
3540
3541 static void __exit exit_ext4_fs(void)
3542 {
3543         unregister_filesystem(&ext4dev_fs_type);
3544         destroy_inodecache();
3545         exit_ext4_xattr();
3546         exit_ext4_mballoc();
3547 }
3548
3549 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3550 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3551 MODULE_LICENSE("GPL");
3552 module_init(init_ext4_fs)
3553 module_exit(exit_ext4_fs)