Pileus Git - ~andy/linux/blob - fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/smp_lock.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/marker.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <asm/uaccess.h>
  42
  43 #include "ext4.h"
  44 #include "ext4_jbd2.h"
  45 #include "xattr.h"
  46 #include "acl.h"
  47 #include "namei.h"
  48 #include "group.h"
  49
  50 struct proc_dir_entry *ext4_proc_root;
  51
  52 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  53                              unsigned long journal_devnum);
  54 static void ext4_commit_super(struct super_block *sb,
  55                               struct ext4_super_block *es, int sync);
  56 static void ext4_mark_recovery_complete(struct super_block *sb,
  57                                         struct ext4_super_block *es);
  58 static void ext4_clear_journal_err(struct super_block *sb,
  59                                    struct ext4_super_block *es);
  60 static int ext4_sync_fs(struct super_block *sb, int wait);
  61 static const char *ext4_decode_error(struct super_block *sb, int errno,
  62                                      char nbuf[16]);
  63 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  64 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  65 static void ext4_unlockfs(struct super_block *sb);
  66 static void ext4_write_super(struct super_block *sb);
  67 static void ext4_write_super_lockfs(struct super_block *sb);
  68
  69
  70 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  71                                struct ext4_group_desc *bg)
  72 {
  73         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  74                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  75                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  76 }
  77
  78 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  79                                struct ext4_group_desc *bg)
  80 {
  81         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  82                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  83                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  84 }
  85
  86 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  87                               struct ext4_group_desc *bg)
  88 {
  89         return le32_to_cpu(bg->bg_inode_table_lo) |
  90                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  91                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  92 }
  93
  94 __u32 ext4_free_blks_count(struct super_block *sb,
  95                               struct ext4_group_desc *bg)
  96 {
  97         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
  98                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  99                 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 100 }
 101
 102 __u32 ext4_free_inodes_count(struct super_block *sb,
 103                               struct ext4_group_desc *bg)
 104 {
 105         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 106                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 107                 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 108 }
 109
 110 __u32 ext4_used_dirs_count(struct super_block *sb,
 111                               struct ext4_group_desc *bg)
 112 {
 113         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 114                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 115                 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 116 }
 117
 118 __u32 ext4_itable_unused_count(struct super_block *sb,
 119                               struct ext4_group_desc *bg)
 120 {
 121         return le16_to_cpu(bg->bg_itable_unused_lo) |
 122                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 123                 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 124 }
 125
 126 void ext4_block_bitmap_set(struct super_block *sb,
 127                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 128 {
 129         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 130         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 131                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 132 }
 133
 134 void ext4_inode_bitmap_set(struct super_block *sb,
 135                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 136 {
 137         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 138         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 139                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 140 }
 141
 142 void ext4_inode_table_set(struct super_block *sb,
 143                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 144 {
 145         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 146         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 147                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 148 }
 149
 150 void ext4_free_blks_set(struct super_block *sb,
 151                           struct ext4_group_desc *bg, __u32 count)
 152 {
 153         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 154         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 155                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 156 }
 157
 158 void ext4_free_inodes_set(struct super_block *sb,
 159                           struct ext4_group_desc *bg, __u32 count)
 160 {
 161         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 162         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 163                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 164 }
 165
 166 void ext4_used_dirs_set(struct super_block *sb,
 167                           struct ext4_group_desc *bg, __u32 count)
 168 {
 169         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 170         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 171                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 172 }
 173
 174 void ext4_itable_unused_set(struct super_block *sb,
 175                           struct ext4_group_desc *bg, __u32 count)
 176 {
 177         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 178         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 179                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 180 }
 181
 182 /*
 183  * Wrappers for jbd2_journal_start/end.
 184  *
 185  * The only special thing we need to do here is to make sure that all
 186  * journal_end calls result in the superblock being marked dirty, so
 187  * that sync() will call the filesystem's write_super callback if
 188  * appropriate.
 189  */
 190 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 191 {
 192         journal_t *journal;
 193
 194         if (sb->s_flags & MS_RDONLY)
 195                 return ERR_PTR(-EROFS);
 196
 197         /* Special case here: if the journal has aborted behind our
 198          * backs (eg. EIO in the commit thread), then we still need to
 199          * take the FS itself readonly cleanly. */
 200         journal = EXT4_SB(sb)->s_journal;
 201         if (journal) {
 202                 if (is_journal_aborted(journal)) {
 203                         ext4_abort(sb, __func__,
 204                                    "Detected aborted journal");
 205                         return ERR_PTR(-EROFS);
 206                 }
 207                 return jbd2_journal_start(journal, nblocks);
 208         }
 209         /*
 210          * We're not journaling, return the appropriate indication.
 211          */
 212         current->journal_info = EXT4_NOJOURNAL_HANDLE;
 213         return current->journal_info;
 214 }
 215
 216 /*
 217  * The only special thing we need to do here is to make sure that all
 218  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 219  * that sync() will call the filesystem's write_super callback if
 220  * appropriate.
 221  */
 222 int __ext4_journal_stop(const char *where, handle_t *handle)
 223 {
 224         struct super_block *sb;
 225         int err;
 226         int rc;
 227
 228         if (!ext4_handle_valid(handle)) {
 229                 /*
 230                  * Do this here since we don't call jbd2_journal_stop() in
 231                  * no-journal mode.
 232                  */
 233                 current->journal_info = NULL;
 234                 return 0;
 235         }
 236         sb = handle->h_transaction->t_journal->j_private;
 237         err = handle->h_err;
 238         rc = jbd2_journal_stop(handle);
 239
 240         if (!err)
 241                 err = rc;
 242         if (err)
 243                 __ext4_std_error(sb, where, err);
 244         return err;
 245 }
 246
 247 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 248                 struct buffer_head *bh, handle_t *handle, int err)
 249 {
 250         char nbuf[16];
 251         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 252
 253         BUG_ON(!ext4_handle_valid(handle));
 254
 255         if (bh)
 256                 BUFFER_TRACE(bh, "abort");
 257
 258         if (!handle->h_err)
 259                 handle->h_err = err;
 260
 261         if (is_handle_aborted(handle))
 262                 return;
 263
 264         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 265                caller, errstr, err_fn);
 266
 267         jbd2_journal_abort_handle(handle);
 268 }
 269
 270 /* Deal with the reporting of failure conditions on a filesystem such as
 271  * inconsistencies detected or read IO failures.
 272  *
 273  * On ext2, we can store the error state of the filesystem in the
 274  * superblock.  That is not possible on ext4, because we may have other
 275  * write ordering constraints on the superblock which prevent us from
 276  * writing it out straight away; and given that the journal is about to
 277  * be aborted, we can't rely on the current, or future, transactions to
 278  * write out the superblock safely.
 279  *
 280  * We'll just use the jbd2_journal_abort() error code to record an error in
 281  * the journal instead.  On recovery, the journal will compain about
 282  * that error until we've noted it down and cleared it.
 283  */
 284
 285 static void ext4_handle_error(struct super_block *sb)
 286 {
 287         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 288
 289         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 290         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 291
 292         if (sb->s_flags & MS_RDONLY)
 293                 return;
 294
 295         if (!test_opt(sb, ERRORS_CONT)) {
 296                 journal_t *journal = EXT4_SB(sb)->s_journal;
 297
 298                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 299                 if (journal)
 300                         jbd2_journal_abort(journal, -EIO);
 301         }
 302         if (test_opt(sb, ERRORS_RO)) {
 303                 printk(KERN_CRIT "Remounting filesystem read-only\n");
 304                 sb->s_flags |= MS_RDONLY;
 305         }
 306         ext4_commit_super(sb, es, 1);
 307         if (test_opt(sb, ERRORS_PANIC))
 308                 panic("EXT4-fs (device %s): panic forced after error\n",
 309                         sb->s_id);
 310 }
 311
 312 void ext4_error(struct super_block *sb, const char *function,
 313                 const char *fmt, ...)
 314 {
 315         va_list args;
 316
 317         va_start(args, fmt);
 318         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 319         vprintk(fmt, args);
 320         printk("\n");
 321         va_end(args);
 322
 323         ext4_handle_error(sb);
 324 }
 325
 326 static const char *ext4_decode_error(struct super_block *sb, int errno,
 327                                      char nbuf[16])
 328 {
 329         char *errstr = NULL;
 330
 331         switch (errno) {
 332         case -EIO:
 333                 errstr = "IO failure";
 334                 break;
 335         case -ENOMEM:
 336                 errstr = "Out of memory";
 337                 break;
 338         case -EROFS:
 339                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 340                         errstr = "Journal has aborted";
 341                 else
 342                         errstr = "Readonly filesystem";
 343                 break;
 344         default:
 345                 /* If the caller passed in an extra buffer for unknown
 346                  * errors, textualise them now.  Else we just return
 347                  * NULL. */
 348                 if (nbuf) {
 349                         /* Check for truncated error codes... */
 350                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 351                                 errstr = nbuf;
 352                 }
 353                 break;
 354         }
 355
 356         return errstr;
 357 }
 358
 359 /* __ext4_std_error decodes expected errors from journaling functions
 360  * automatically and invokes the appropriate error response.  */
 361
 362 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 363 {
 364         char nbuf[16];
 365         const char *errstr;
 366
 367         /* Special case: if the error is EROFS, and we're not already
 368          * inside a transaction, then there's really no point in logging
 369          * an error. */
 370         if (errno == -EROFS && journal_current_handle() == NULL &&
 371             (sb->s_flags & MS_RDONLY))
 372                 return;
 373
 374         errstr = ext4_decode_error(sb, errno, nbuf);
 375         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 376                sb->s_id, function, errstr);
 377
 378         ext4_handle_error(sb);
 379 }
 380
 381 /*
 382  * ext4_abort is a much stronger failure handler than ext4_error.  The
 383  * abort function may be used to deal with unrecoverable failures such
 384  * as journal IO errors or ENOMEM at a critical moment in log management.
 385  *
 386  * We unconditionally force the filesystem into an ABORT|READONLY state,
 387  * unless the error response on the fs has been set to panic in which
 388  * case we take the easy way out and panic immediately.
 389  */
 390
 391 void ext4_abort(struct super_block *sb, const char *function,
 392                 const char *fmt, ...)
 393 {
 394         va_list args;
 395
 396         printk(KERN_CRIT "ext4_abort called.\n");
 397
 398         va_start(args, fmt);
 399         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 400         vprintk(fmt, args);
 401         printk("\n");
 402         va_end(args);
 403
 404         if (test_opt(sb, ERRORS_PANIC))
 405                 panic("EXT4-fs panic from previous error\n");
 406
 407         if (sb->s_flags & MS_RDONLY)
 408                 return;
 409
 410         printk(KERN_CRIT "Remounting filesystem read-only\n");
 411         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 412         sb->s_flags |= MS_RDONLY;
 413         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 414         if (EXT4_SB(sb)->s_journal)
 415                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 416 }
 417
 418 void ext4_warning(struct super_block *sb, const char *function,
 419                   const char *fmt, ...)
 420 {
 421         va_list args;
 422
 423         va_start(args, fmt);
 424         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 425                sb->s_id, function);
 426         vprintk(fmt, args);
 427         printk("\n");
 428         va_end(args);
 429 }
 430
 431 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
 432                                 const char *function, const char *fmt, ...)
 433 __releases(bitlock)
 434 __acquires(bitlock)
 435 {
 436         va_list args;
 437         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 438
 439         va_start(args, fmt);
 440         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 441         vprintk(fmt, args);
 442         printk("\n");
 443         va_end(args);
 444
 445         if (test_opt(sb, ERRORS_CONT)) {
 446                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 447                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 448                 ext4_commit_super(sb, es, 0);
 449                 return;
 450         }
 451         ext4_unlock_group(sb, grp);
 452         ext4_handle_error(sb);
 453         /*
 454          * We only get here in the ERRORS_RO case; relocking the group
 455          * may be dangerous, but nothing bad will happen since the
 456          * filesystem will have already been marked read/only and the
 457          * journal has been aborted.  We return 1 as a hint to callers
 458          * who might what to use the return value from
 459          * ext4_grp_locked_error() to distinguish beween the
 460          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 461          * aggressively from the ext4 function in question, with a
 462          * more appropriate error code.
 463          */
 464         ext4_lock_group(sb, grp);
 465         return;
 466 }
 467
 468
 469 void ext4_update_dynamic_rev(struct super_block *sb)
 470 {
 471         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 472
 473         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 474                 return;
 475
 476         ext4_warning(sb, __func__,
 477                      "updating to rev %d because of new feature flag, "
 478                      "running e2fsck is recommended",
 479                      EXT4_DYNAMIC_REV);
 480
 481         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 482         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 483         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 484         /* leave es->s_feature_*compat flags alone */
 485         /* es->s_uuid will be set by e2fsck if empty */
 486
 487         /*
 488          * The rest of the superblock fields should be zero, and if not it
 489          * means they are likely already in use, so leave them alone.  We
 490          * can leave it up to e2fsck to clean up any inconsistencies there.
 491          */
 492 }
 493
 494 /*
 495  * Open the external journal device
 496  */
 497 static struct block_device *ext4_blkdev_get(dev_t dev)
 498 {
 499         struct block_device *bdev;
 500         char b[BDEVNAME_SIZE];
 501
 502         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 503         if (IS_ERR(bdev))
 504                 goto fail;
 505         return bdev;
 506
 507 fail:
 508         printk(KERN_ERR "EXT4-fs: failed to open journal device %s: %ld\n",
 509                         __bdevname(dev, b), PTR_ERR(bdev));
 510         return NULL;
 511 }
 512
 513 /*
 514  * Release the journal device
 515  */
 516 static int ext4_blkdev_put(struct block_device *bdev)
 517 {
 518         bd_release(bdev);
 519         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 520 }
 521
 522 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 523 {
 524         struct block_device *bdev;
 525         int ret = -ENODEV;
 526
 527         bdev = sbi->journal_bdev;
 528         if (bdev) {
 529                 ret = ext4_blkdev_put(bdev);
 530                 sbi->journal_bdev = NULL;
 531         }
 532         return ret;
 533 }
 534
 535 static inline struct inode *orphan_list_entry(struct list_head *l)
 536 {
 537         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 538 }
 539
 540 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 541 {
 542         struct list_head *l;
 543
 544         printk(KERN_ERR "sb orphan head is %d\n",
 545                le32_to_cpu(sbi->s_es->s_last_orphan));
 546
 547         printk(KERN_ERR "sb_info orphan list:\n");
 548         list_for_each(l, &sbi->s_orphan) {
 549                 struct inode *inode = orphan_list_entry(l);
 550                 printk(KERN_ERR "  "
 551                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 552                        inode->i_sb->s_id, inode->i_ino, inode,
 553                        inode->i_mode, inode->i_nlink,
 554                        NEXT_ORPHAN(inode));
 555         }
 556 }
 557
 558 static void ext4_put_super(struct super_block *sb)
 559 {
 560         struct ext4_sb_info *sbi = EXT4_SB(sb);
 561         struct ext4_super_block *es = sbi->s_es;
 562         int i, err;
 563
 564         ext4_mb_release(sb);
 565         ext4_ext_release(sb);
 566         ext4_xattr_put_super(sb);
 567         if (sbi->s_journal) {
 568                 err = jbd2_journal_destroy(sbi->s_journal);
 569                 sbi->s_journal = NULL;
 570                 if (err < 0)
 571                         ext4_abort(sb, __func__,
 572                                    "Couldn't clean up the journal");
 573         }
 574         if (!(sb->s_flags & MS_RDONLY)) {
 575                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 576                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 577                 ext4_commit_super(sb, es, 1);
 578         }
 579         if (sbi->s_proc) {
 580                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
 581                 remove_proc_entry(sb->s_id, ext4_proc_root);
 582         }
 583
 584         for (i = 0; i < sbi->s_gdb_count; i++)
 585                 brelse(sbi->s_group_desc[i]);
 586         kfree(sbi->s_group_desc);
 587         kfree(sbi->s_flex_groups);
 588         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 589         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 590         percpu_counter_destroy(&sbi->s_dirs_counter);
 591         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 592         brelse(sbi->s_sbh);
 593 #ifdef CONFIG_QUOTA
 594         for (i = 0; i < MAXQUOTAS; i++)
 595                 kfree(sbi->s_qf_names[i]);
 596 #endif
 597
 598         /* Debugging code just in case the in-memory inode orphan list
 599          * isn't empty.  The on-disk one can be non-empty if we've
 600          * detected an error and taken the fs readonly, but the
 601          * in-memory list had better be clean by this point. */
 602         if (!list_empty(&sbi->s_orphan))
 603                 dump_orphan_list(sb, sbi);
 604         J_ASSERT(list_empty(&sbi->s_orphan));
 605
 606         invalidate_bdev(sb->s_bdev);
 607         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 608                 /*
 609                  * Invalidate the journal device's buffers.  We don't want them
 610                  * floating about in memory - the physical journal device may
 611                  * hotswapped, and it breaks the `ro-after' testing code.
 612                  */
 613                 sync_blockdev(sbi->journal_bdev);
 614                 invalidate_bdev(sbi->journal_bdev);
 615                 ext4_blkdev_remove(sbi);
 616         }
 617         sb->s_fs_info = NULL;
 618         kfree(sbi);
 619         return;
 620 }
 621
 622 static struct kmem_cache *ext4_inode_cachep;
 623
 624 /*
 625  * Called inside transaction, so use GFP_NOFS
 626  */
 627 static struct inode *ext4_alloc_inode(struct super_block *sb)
 628 {
 629         struct ext4_inode_info *ei;
 630
 631         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 632         if (!ei)
 633                 return NULL;
 634 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 635         ei->i_acl = EXT4_ACL_NOT_CACHED;
 636         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 637 #endif
 638         ei->vfs_inode.i_version = 1;
 639         ei->vfs_inode.i_data.writeback_index = 0;
 640         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 641         INIT_LIST_HEAD(&ei->i_prealloc_list);
 642         spin_lock_init(&ei->i_prealloc_lock);
 643         /*
 644          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 645          * therefore it can be null here.  Don't check it, just initialize
 646          * jinode.
 647          */
 648         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 649         ei->i_reserved_data_blocks = 0;
 650         ei->i_reserved_meta_blocks = 0;
 651         ei->i_allocated_meta_blocks = 0;
 652         ei->i_delalloc_reserved_flag = 0;
 653         spin_lock_init(&(ei->i_block_reservation_lock));
 654         return &ei->vfs_inode;
 655 }
 656
 657 static void ext4_destroy_inode(struct inode *inode)
 658 {
 659         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 660                 printk("EXT4 Inode %p: orphan list check failed!\n",
 661                         EXT4_I(inode));
 662                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 663                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 664                                 true);
 665                 dump_stack();
 666         }
 667         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 668 }
 669
 670 static void init_once(void *foo)
 671 {
 672         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 673
 674         INIT_LIST_HEAD(&ei->i_orphan);
 675 #ifdef CONFIG_EXT4_FS_XATTR
 676         init_rwsem(&ei->xattr_sem);
 677 #endif
 678         init_rwsem(&ei->i_data_sem);
 679         inode_init_once(&ei->vfs_inode);
 680 }
 681
 682 static int init_inodecache(void)
 683 {
 684         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 685                                              sizeof(struct ext4_inode_info),
 686                                              0, (SLAB_RECLAIM_ACCOUNT|
 687                                                 SLAB_MEM_SPREAD),
 688                                              init_once);
 689         if (ext4_inode_cachep == NULL)
 690                 return -ENOMEM;
 691         return 0;
 692 }
 693
 694 static void destroy_inodecache(void)
 695 {
 696         kmem_cache_destroy(ext4_inode_cachep);
 697 }
 698
 699 static void ext4_clear_inode(struct inode *inode)
 700 {
 701 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 702         if (EXT4_I(inode)->i_acl &&
 703                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 704                 posix_acl_release(EXT4_I(inode)->i_acl);
 705                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 706         }
 707         if (EXT4_I(inode)->i_default_acl &&
 708                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 709                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 710                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 711         }
 712 #endif
 713         ext4_discard_preallocations(inode);
 714         if (EXT4_JOURNAL(inode))
 715                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 716                                        &EXT4_I(inode)->jinode);
 717 }
 718
 719 static inline void ext4_show_quota_options(struct seq_file *seq,
 720                                            struct super_block *sb)
 721 {
 722 #if defined(CONFIG_QUOTA)
 723         struct ext4_sb_info *sbi = EXT4_SB(sb);
 724
 725         if (sbi->s_jquota_fmt)
 726                 seq_printf(seq, ",jqfmt=%s",
 727                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold" : "vfsv0");
 728
 729         if (sbi->s_qf_names[USRQUOTA])
 730                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 731
 732         if (sbi->s_qf_names[GRPQUOTA])
 733                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 734
 735         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 736                 seq_puts(seq, ",usrquota");
 737
 738         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 739                 seq_puts(seq, ",grpquota");
 740 #endif
 741 }
 742
 743 /*
 744  * Show an option if
 745  *  - it's set to a non-default value OR
 746  *  - if the per-sb default is different from the global default
 747  */
 748 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 749 {
 750         int def_errors;
 751         unsigned long def_mount_opts;
 752         struct super_block *sb = vfs->mnt_sb;
 753         struct ext4_sb_info *sbi = EXT4_SB(sb);
 754         struct ext4_super_block *es = sbi->s_es;
 755
 756         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 757         def_errors     = le16_to_cpu(es->s_errors);
 758
 759         if (sbi->s_sb_block != 1)
 760                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 761         if (test_opt(sb, MINIX_DF))
 762                 seq_puts(seq, ",minixdf");
 763         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 764                 seq_puts(seq, ",grpid");
 765         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 766                 seq_puts(seq, ",nogrpid");
 767         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 768             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 769                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 770         }
 771         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 772             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 773                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 774         }
 775         if (test_opt(sb, ERRORS_RO)) {
 776                 if (def_errors == EXT4_ERRORS_PANIC ||
 777                     def_errors == EXT4_ERRORS_CONTINUE) {
 778                         seq_puts(seq, ",errors=remount-ro");
 779                 }
 780         }
 781         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 782                 seq_puts(seq, ",errors=continue");
 783         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 784                 seq_puts(seq, ",errors=panic");
 785         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 786                 seq_puts(seq, ",nouid32");
 787         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 788                 seq_puts(seq, ",debug");
 789         if (test_opt(sb, OLDALLOC))
 790                 seq_puts(seq, ",oldalloc");
 791 #ifdef CONFIG_EXT4_FS_XATTR
 792         if (test_opt(sb, XATTR_USER) &&
 793                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 794                 seq_puts(seq, ",user_xattr");
 795         if (!test_opt(sb, XATTR_USER) &&
 796             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 797                 seq_puts(seq, ",nouser_xattr");
 798         }
 799 #endif
 800 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 801         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 802                 seq_puts(seq, ",acl");
 803         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 804                 seq_puts(seq, ",noacl");
 805 #endif
 806         if (!test_opt(sb, RESERVATION))
 807                 seq_puts(seq, ",noreservation");
 808         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 809                 seq_printf(seq, ",commit=%u",
 810                            (unsigned) (sbi->s_commit_interval / HZ));
 811         }
 812         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 813                 seq_printf(seq, ",min_batch_time=%u",
 814                            (unsigned) sbi->s_min_batch_time);
 815         }
 816         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 817                 seq_printf(seq, ",max_batch_time=%u",
 818                            (unsigned) sbi->s_min_batch_time);
 819         }
 820
 821         /*
 822          * We're changing the default of barrier mount option, so
 823          * let's always display its mount state so it's clear what its
 824          * status is.
 825          */
 826         seq_puts(seq, ",barrier=");
 827         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 828         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 829                 seq_puts(seq, ",journal_async_commit");
 830         if (test_opt(sb, NOBH))
 831                 seq_puts(seq, ",nobh");
 832         if (test_opt(sb, I_VERSION))
 833                 seq_puts(seq, ",i_version");
 834         if (!test_opt(sb, DELALLOC))
 835                 seq_puts(seq, ",nodelalloc");
 836
 837
 838         if (sbi->s_stripe)
 839                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 840         /*
 841          * journal mode get enabled in different ways
 842          * So just print the value even if we didn't specify it
 843          */
 844         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 845                 seq_puts(seq, ",data=journal");
 846         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 847                 seq_puts(seq, ",data=ordered");
 848         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 849                 seq_puts(seq, ",data=writeback");
 850
 851         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 852                 seq_printf(seq, ",inode_readahead_blks=%u",
 853                            sbi->s_inode_readahead_blks);
 854
 855         if (test_opt(sb, DATA_ERR_ABORT))
 856                 seq_puts(seq, ",data_err=abort");
 857
 858         ext4_show_quota_options(seq, sb);
 859         return 0;
 860 }
 861
 862
 863 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 864                 u64 ino, u32 generation)
 865 {
 866         struct inode *inode;
 867
 868         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 869                 return ERR_PTR(-ESTALE);
 870         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 871                 return ERR_PTR(-ESTALE);
 872
 873         /* iget isn't really right if the inode is currently unallocated!!
 874          *
 875          * ext4_read_inode will return a bad_inode if the inode had been
 876          * deleted, so we should be safe.
 877          *
 878          * Currently we don't know the generation for parent directory, so
 879          * a generation of 0 means "accept any"
 880          */
 881         inode = ext4_iget(sb, ino);
 882         if (IS_ERR(inode))
 883                 return ERR_CAST(inode);
 884         if (generation && inode->i_generation != generation) {
 885                 iput(inode);
 886                 return ERR_PTR(-ESTALE);
 887         }
 888
 889         return inode;
 890 }
 891
 892 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 893                 int fh_len, int fh_type)
 894 {
 895         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 896                                     ext4_nfs_get_inode);
 897 }
 898
 899 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 900                 int fh_len, int fh_type)
 901 {
 902         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 903                                     ext4_nfs_get_inode);
 904 }
 905
 906 /*
 907  * Try to release metadata pages (indirect blocks, directories) which are
 908  * mapped via the block device.  Since these pages could have journal heads
 909  * which would prevent try_to_free_buffers() from freeing them, we must use
 910  * jbd2 layer's try_to_free_buffers() function to release them.
 911  */
 912 static int bdev_try_to_free_page(struct super_block *sb, struct page *page, gfp_t wait)
 913 {
 914         journal_t *journal = EXT4_SB(sb)->s_journal;
 915
 916         WARN_ON(PageChecked(page));
 917         if (!page_has_buffers(page))
 918                 return 0;
 919         if (journal)
 920                 return jbd2_journal_try_to_free_buffers(journal, page,
 921                                                         wait & ~__GFP_WAIT);
 922         return try_to_free_buffers(page);
 923 }
 924
 925 #ifdef CONFIG_QUOTA
 926 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
 927 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 928
 929 static int ext4_dquot_initialize(struct inode *inode, int type);
 930 static int ext4_dquot_drop(struct inode *inode);
 931 static int ext4_write_dquot(struct dquot *dquot);
 932 static int ext4_acquire_dquot(struct dquot *dquot);
 933 static int ext4_release_dquot(struct dquot *dquot);
 934 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 935 static int ext4_write_info(struct super_block *sb, int type);
 936 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
 937                                 char *path, int remount);
 938 static int ext4_quota_on_mount(struct super_block *sb, int type);
 939 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 940                                size_t len, loff_t off);
 941 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 942                                 const char *data, size_t len, loff_t off);
 943
 944 static struct dquot_operations ext4_quota_operations = {
 945         .initialize     = ext4_dquot_initialize,
 946         .drop           = ext4_dquot_drop,
 947         .alloc_space    = dquot_alloc_space,
 948         .alloc_inode    = dquot_alloc_inode,
 949         .free_space     = dquot_free_space,
 950         .free_inode     = dquot_free_inode,
 951         .transfer       = dquot_transfer,
 952         .write_dquot    = ext4_write_dquot,
 953         .acquire_dquot  = ext4_acquire_dquot,
 954         .release_dquot  = ext4_release_dquot,
 955         .mark_dirty     = ext4_mark_dquot_dirty,
 956         .write_info     = ext4_write_info
 957 };
 958
 959 static struct quotactl_ops ext4_qctl_operations = {
 960         .quota_on       = ext4_quota_on,
 961         .quota_off      = vfs_quota_off,
 962         .quota_sync     = vfs_quota_sync,
 963         .get_info       = vfs_get_dqinfo,
 964         .set_info       = vfs_set_dqinfo,
 965         .get_dqblk      = vfs_get_dqblk,
 966         .set_dqblk      = vfs_set_dqblk
 967 };
 968 #endif
 969
 970 static const struct super_operations ext4_sops = {
 971         .alloc_inode    = ext4_alloc_inode,
 972         .destroy_inode  = ext4_destroy_inode,
 973         .write_inode    = ext4_write_inode,
 974         .dirty_inode    = ext4_dirty_inode,
 975         .delete_inode   = ext4_delete_inode,
 976         .put_super      = ext4_put_super,
 977         .write_super    = ext4_write_super,
 978         .sync_fs        = ext4_sync_fs,
 979         .write_super_lockfs = ext4_write_super_lockfs,
 980         .unlockfs       = ext4_unlockfs,
 981         .statfs         = ext4_statfs,
 982         .remount_fs     = ext4_remount,
 983         .clear_inode    = ext4_clear_inode,
 984         .show_options   = ext4_show_options,
 985 #ifdef CONFIG_QUOTA
 986         .quota_read     = ext4_quota_read,
 987         .quota_write    = ext4_quota_write,
 988 #endif
 989         .bdev_try_to_free_page = bdev_try_to_free_page,
 990 };
 991
 992 static const struct export_operations ext4_export_ops = {
 993         .fh_to_dentry = ext4_fh_to_dentry,
 994         .fh_to_parent = ext4_fh_to_parent,
 995         .get_parent = ext4_get_parent,
 996 };
 997
 998 enum {
 999         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1000         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1001         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1002         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1003         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
1004         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1005         Opt_journal_update, Opt_journal_dev,
1006         Opt_journal_checksum, Opt_journal_async_commit,
1007         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1008         Opt_data_err_abort, Opt_data_err_ignore,
1009         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1010         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
1011         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
1012         Opt_grpquota, Opt_i_version,
1013         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1014         Opt_inode_readahead_blks, Opt_journal_ioprio
1015 };
1016
1017 static const match_table_t tokens = {
1018         {Opt_bsd_df, "bsddf"},
1019         {Opt_minix_df, "minixdf"},
1020         {Opt_grpid, "grpid"},
1021         {Opt_grpid, "bsdgroups"},
1022         {Opt_nogrpid, "nogrpid"},
1023         {Opt_nogrpid, "sysvgroups"},
1024         {Opt_resgid, "resgid=%u"},
1025         {Opt_resuid, "resuid=%u"},
1026         {Opt_sb, "sb=%u"},
1027         {Opt_err_cont, "errors=continue"},
1028         {Opt_err_panic, "errors=panic"},
1029         {Opt_err_ro, "errors=remount-ro"},
1030         {Opt_nouid32, "nouid32"},
1031         {Opt_debug, "debug"},
1032         {Opt_oldalloc, "oldalloc"},
1033         {Opt_orlov, "orlov"},
1034         {Opt_user_xattr, "user_xattr"},
1035         {Opt_nouser_xattr, "nouser_xattr"},
1036         {Opt_acl, "acl"},
1037         {Opt_noacl, "noacl"},
1038         {Opt_reservation, "reservation"},
1039         {Opt_noreservation, "noreservation"},
1040         {Opt_noload, "noload"},
1041         {Opt_nobh, "nobh"},
1042         {Opt_bh, "bh"},
1043         {Opt_commit, "commit=%u"},
1044         {Opt_min_batch_time, "min_batch_time=%u"},
1045         {Opt_max_batch_time, "max_batch_time=%u"},
1046         {Opt_journal_update, "journal=update"},
1047         {Opt_journal_dev, "journal_dev=%u"},
1048         {Opt_journal_checksum, "journal_checksum"},
1049         {Opt_journal_async_commit, "journal_async_commit"},
1050         {Opt_abort, "abort"},
1051         {Opt_data_journal, "data=journal"},
1052         {Opt_data_ordered, "data=ordered"},
1053         {Opt_data_writeback, "data=writeback"},
1054         {Opt_data_err_abort, "data_err=abort"},
1055         {Opt_data_err_ignore, "data_err=ignore"},
1056         {Opt_offusrjquota, "usrjquota="},
1057         {Opt_usrjquota, "usrjquota=%s"},
1058         {Opt_offgrpjquota, "grpjquota="},
1059         {Opt_grpjquota, "grpjquota=%s"},
1060         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1061         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1062         {Opt_grpquota, "grpquota"},
1063         {Opt_noquota, "noquota"},
1064         {Opt_quota, "quota"},
1065         {Opt_usrquota, "usrquota"},
1066         {Opt_barrier, "barrier=%u"},
1067         {Opt_i_version, "i_version"},
1068         {Opt_stripe, "stripe=%u"},
1069         {Opt_resize, "resize"},
1070         {Opt_delalloc, "delalloc"},
1071         {Opt_nodelalloc, "nodelalloc"},
1072         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1073         {Opt_journal_ioprio, "journal_ioprio=%u"},
1074         {Opt_err, NULL},
1075 };
1076
1077 static ext4_fsblk_t get_sb_block(void **data)
1078 {
1079         ext4_fsblk_t    sb_block;
1080         char            *options = (char *) *data;
1081
1082         if (!options || strncmp(options, "sb=", 3) != 0)
1083                 return 1;       /* Default location */
1084         options += 3;
1085         /*todo: use simple_strtoll with >32bit ext4 */
1086         sb_block = simple_strtoul(options, &options, 0);
1087         if (*options && *options != ',') {
1088                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1089                        (char *) *data);
1090                 return 1;
1091         }
1092         if (*options == ',')
1093                 options++;
1094         *data = (void *) options;
1095         return sb_block;
1096 }
1097
1098 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1099
1100 static int parse_options(char *options, struct super_block *sb,
1101                          unsigned long *journal_devnum,
1102                          unsigned int *journal_ioprio,
1103                          ext4_fsblk_t *n_blocks_count, int is_remount)
1104 {
1105         struct ext4_sb_info *sbi = EXT4_SB(sb);
1106         char *p;
1107         substring_t args[MAX_OPT_ARGS];
1108         int data_opt = 0;
1109         int option;
1110 #ifdef CONFIG_QUOTA
1111         int qtype, qfmt;
1112         char *qname;
1113 #endif
1114
1115         if (!options)
1116                 return 1;
1117
1118         while ((p = strsep(&options, ",")) != NULL) {
1119                 int token;
1120                 if (!*p)
1121                         continue;
1122
1123                 token = match_token(p, tokens, args);
1124                 switch (token) {
1125                 case Opt_bsd_df:
1126                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1127                         break;
1128                 case Opt_minix_df:
1129                         set_opt(sbi->s_mount_opt, MINIX_DF);
1130                         break;
1131                 case Opt_grpid:
1132                         set_opt(sbi->s_mount_opt, GRPID);
1133                         break;
1134                 case Opt_nogrpid:
1135                         clear_opt(sbi->s_mount_opt, GRPID);
1136                         break;
1137                 case Opt_resuid:
1138                         if (match_int(&args[0], &option))
1139                                 return 0;
1140                         sbi->s_resuid = option;
1141                         break;
1142                 case Opt_resgid:
1143                         if (match_int(&args[0], &option))
1144                                 return 0;
1145                         sbi->s_resgid = option;
1146                         break;
1147                 case Opt_sb:
1148                         /* handled by get_sb_block() instead of here */
1149                         /* *sb_block = match_int(&args[0]); */
1150                         break;
1151                 case Opt_err_panic:
1152                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1153                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1154                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1155                         break;
1156                 case Opt_err_ro:
1157                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1158                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1159                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1160                         break;
1161                 case Opt_err_cont:
1162                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1163                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1164                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1165                         break;
1166                 case Opt_nouid32:
1167                         set_opt(sbi->s_mount_opt, NO_UID32);
1168                         break;
1169                 case Opt_debug:
1170                         set_opt(sbi->s_mount_opt, DEBUG);
1171                         break;
1172                 case Opt_oldalloc:
1173                         set_opt(sbi->s_mount_opt, OLDALLOC);
1174                         break;
1175                 case Opt_orlov:
1176                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1177                         break;
1178 #ifdef CONFIG_EXT4_FS_XATTR
1179                 case Opt_user_xattr:
1180                         set_opt(sbi->s_mount_opt, XATTR_USER);
1181                         break;
1182                 case Opt_nouser_xattr:
1183                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1184                         break;
1185 #else
1186                 case Opt_user_xattr:
1187                 case Opt_nouser_xattr:
1188                         printk(KERN_ERR "EXT4 (no)user_xattr options "
1189                                "not supported\n");
1190                         break;
1191 #endif
1192 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1193                 case Opt_acl:
1194                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1195                         break;
1196                 case Opt_noacl:
1197                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1198                         break;
1199 #else
1200                 case Opt_acl:
1201                 case Opt_noacl:
1202                         printk(KERN_ERR "EXT4 (no)acl options "
1203                                "not supported\n");
1204                         break;
1205 #endif
1206                 case Opt_reservation:
1207                         set_opt(sbi->s_mount_opt, RESERVATION);
1208                         break;
1209                 case Opt_noreservation:
1210                         clear_opt(sbi->s_mount_opt, RESERVATION);
1211                         break;
1212                 case Opt_journal_update:
1213                         /* @@@ FIXME */
1214                         /* Eventually we will want to be able to create
1215                            a journal file here.  For now, only allow the
1216                            user to specify an existing inode to be the
1217                            journal file. */
1218                         if (is_remount) {
1219                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1220                                        "journal on remount\n");
1221                                 return 0;
1222                         }
1223                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1224                         break;
1225                 case Opt_journal_dev:
1226                         if (is_remount) {
1227                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1228                                        "journal on remount\n");
1229                                 return 0;
1230                         }
1231                         if (match_int(&args[0], &option))
1232                                 return 0;
1233                         *journal_devnum = option;
1234                         break;
1235                 case Opt_journal_checksum:
1236                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1237                         break;
1238                 case Opt_journal_async_commit:
1239                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1240                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1241                         break;
1242                 case Opt_noload:
1243                         set_opt(sbi->s_mount_opt, NOLOAD);
1244                         break;
1245                 case Opt_commit:
1246                         if (match_int(&args[0], &option))
1247                                 return 0;
1248                         if (option < 0)
1249                                 return 0;
1250                         if (option == 0)
1251                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1252                         sbi->s_commit_interval = HZ * option;
1253                         break;
1254                 case Opt_max_batch_time:
1255                         if (match_int(&args[0], &option))
1256                                 return 0;
1257                         if (option < 0)
1258                                 return 0;
1259                         if (option == 0)
1260                                 option = EXT4_DEF_MAX_BATCH_TIME;
1261                         sbi->s_max_batch_time = option;
1262                         break;
1263                 case Opt_min_batch_time:
1264                         if (match_int(&args[0], &option))
1265                                 return 0;
1266                         if (option < 0)
1267                                 return 0;
1268                         sbi->s_min_batch_time = option;
1269                         break;
1270                 case Opt_data_journal:
1271                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1272                         goto datacheck;
1273                 case Opt_data_ordered:
1274                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1275                         goto datacheck;
1276                 case Opt_data_writeback:
1277                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1278                 datacheck:
1279                         if (is_remount) {
1280                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1281                                                 != data_opt) {
1282                                         printk(KERN_ERR
1283                                                 "EXT4-fs: cannot change data "
1284                                                 "mode on remount\n");
1285                                         return 0;
1286                                 }
1287                         } else {
1288                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1289                                 sbi->s_mount_opt |= data_opt;
1290                         }
1291                         break;
1292                 case Opt_data_err_abort:
1293                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1294                         break;
1295                 case Opt_data_err_ignore:
1296                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1297                         break;
1298 #ifdef CONFIG_QUOTA
1299                 case Opt_usrjquota:
1300                         qtype = USRQUOTA;
1301                         goto set_qf_name;
1302                 case Opt_grpjquota:
1303                         qtype = GRPQUOTA;
1304 set_qf_name:
1305                         if ((sb_any_quota_enabled(sb) ||
1306                              sb_any_quota_suspended(sb)) &&
1307                             !sbi->s_qf_names[qtype]) {
1308                                 printk(KERN_ERR
1309                                        "EXT4-fs: Cannot change journaled "
1310                                        "quota options when quota turned on.\n");
1311                                 return 0;
1312                         }
1313                         qname = match_strdup(&args[0]);
1314                         if (!qname) {
1315                                 printk(KERN_ERR
1316                                         "EXT4-fs: not enough memory for "
1317                                         "storing quotafile name.\n");
1318                                 return 0;
1319                         }
1320                         if (sbi->s_qf_names[qtype] &&
1321                             strcmp(sbi->s_qf_names[qtype], qname)) {
1322                                 printk(KERN_ERR
1323                                         "EXT4-fs: %s quota file already "
1324                                         "specified.\n", QTYPE2NAME(qtype));
1325                                 kfree(qname);
1326                                 return 0;
1327                         }
1328                         sbi->s_qf_names[qtype] = qname;
1329                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1330                                 printk(KERN_ERR
1331                                         "EXT4-fs: quotafile must be on "
1332                                         "filesystem root.\n");
1333                                 kfree(sbi->s_qf_names[qtype]);
1334                                 sbi->s_qf_names[qtype] = NULL;
1335                                 return 0;
1336                         }
1337                         set_opt(sbi->s_mount_opt, QUOTA);
1338                         break;
1339                 case Opt_offusrjquota:
1340                         qtype = USRQUOTA;
1341                         goto clear_qf_name;
1342                 case Opt_offgrpjquota:
1343                         qtype = GRPQUOTA;
1344 clear_qf_name:
1345                         if ((sb_any_quota_enabled(sb) ||
1346                              sb_any_quota_suspended(sb)) &&
1347                             sbi->s_qf_names[qtype]) {
1348                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1349                                         "journaled quota options when "
1350                                         "quota turned on.\n");
1351                                 return 0;
1352                         }
1353                         /*
1354                          * The space will be released later when all options
1355                          * are confirmed to be correct
1356                          */
1357                         sbi->s_qf_names[qtype] = NULL;
1358                         break;
1359                 case Opt_jqfmt_vfsold:
1360                         qfmt = QFMT_VFS_OLD;
1361                         goto set_qf_format;
1362                 case Opt_jqfmt_vfsv0:
1363                         qfmt = QFMT_VFS_V0;
1364 set_qf_format:
1365                         if ((sb_any_quota_enabled(sb) ||
1366                              sb_any_quota_suspended(sb)) &&
1367                             sbi->s_jquota_fmt != qfmt) {
1368                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1369                                         "journaled quota options when "
1370                                         "quota turned on.\n");
1371                                 return 0;
1372                         }
1373                         sbi->s_jquota_fmt = qfmt;
1374                         break;
1375                 case Opt_quota:
1376                 case Opt_usrquota:
1377                         set_opt(sbi->s_mount_opt, QUOTA);
1378                         set_opt(sbi->s_mount_opt, USRQUOTA);
1379                         break;
1380                 case Opt_grpquota:
1381                         set_opt(sbi->s_mount_opt, QUOTA);
1382                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1383                         break;
1384                 case Opt_noquota:
1385                         if (sb_any_quota_enabled(sb)) {
1386                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1387                                         "options when quota turned on.\n");
1388                                 return 0;
1389                         }
1390                         clear_opt(sbi->s_mount_opt, QUOTA);
1391                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1392                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1393                         break;
1394 #else
1395                 case Opt_quota:
1396                 case Opt_usrquota:
1397                 case Opt_grpquota:
1398                         printk(KERN_ERR
1399                                 "EXT4-fs: quota options not supported.\n");
1400                         break;
1401                 case Opt_usrjquota:
1402                 case Opt_grpjquota:
1403                 case Opt_offusrjquota:
1404                 case Opt_offgrpjquota:
1405                 case Opt_jqfmt_vfsold:
1406                 case Opt_jqfmt_vfsv0:
1407                         printk(KERN_ERR
1408                                 "EXT4-fs: journaled quota options not "
1409                                 "supported.\n");
1410                         break;
1411                 case Opt_noquota:
1412                         break;
1413 #endif
1414                 case Opt_abort:
1415                         set_opt(sbi->s_mount_opt, ABORT);
1416                         break;
1417                 case Opt_barrier:
1418                         if (match_int(&args[0], &option))
1419                                 return 0;
1420                         if (option)
1421                                 set_opt(sbi->s_mount_opt, BARRIER);
1422                         else
1423                                 clear_opt(sbi->s_mount_opt, BARRIER);
1424                         break;
1425                 case Opt_ignore:
1426                         break;
1427                 case Opt_resize:
1428                         if (!is_remount) {
1429                                 printk("EXT4-fs: resize option only available "
1430                                         "for remount\n");
1431                                 return 0;
1432                         }
1433                         if (match_int(&args[0], &option) != 0)
1434                                 return 0;
1435                         *n_blocks_count = option;
1436                         break;
1437                 case Opt_nobh:
1438                         set_opt(sbi->s_mount_opt, NOBH);
1439                         break;
1440                 case Opt_bh:
1441                         clear_opt(sbi->s_mount_opt, NOBH);
1442                         break;
1443                 case Opt_i_version:
1444                         set_opt(sbi->s_mount_opt, I_VERSION);
1445                         sb->s_flags |= MS_I_VERSION;
1446                         break;
1447                 case Opt_nodelalloc:
1448                         clear_opt(sbi->s_mount_opt, DELALLOC);
1449                         break;
1450                 case Opt_stripe:
1451                         if (match_int(&args[0], &option))
1452                                 return 0;
1453                         if (option < 0)
1454                                 return 0;
1455                         sbi->s_stripe = option;
1456                         break;
1457                 case Opt_delalloc:
1458                         set_opt(sbi->s_mount_opt, DELALLOC);
1459                         break;
1460                 case Opt_inode_readahead_blks:
1461                         if (match_int(&args[0], &option))
1462                                 return 0;
1463                         if (option < 0 || option > (1 << 30))
1464                                 return 0;
1465                         sbi->s_inode_readahead_blks = option;
1466                         break;
1467                 case Opt_journal_ioprio:
1468                         if (match_int(&args[0], &option))
1469                                 return 0;
1470                         if (option < 0 || option > 7)
1471                                 break;
1472                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1473                                                             option);
1474                         break;
1475                 default:
1476                         printk(KERN_ERR
1477                                "EXT4-fs: Unrecognized mount option \"%s\" "
1478                                "or missing value\n", p);
1479                         return 0;
1480                 }
1481         }
1482 #ifdef CONFIG_QUOTA
1483         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1484                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1485                      sbi->s_qf_names[USRQUOTA])
1486                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1487
1488                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1489                      sbi->s_qf_names[GRPQUOTA])
1490                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1491
1492                 if ((sbi->s_qf_names[USRQUOTA] &&
1493                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1494                     (sbi->s_qf_names[GRPQUOTA] &&
1495                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1496                         printk(KERN_ERR "EXT4-fs: old and new quota "
1497                                         "format mixing.\n");
1498                         return 0;
1499                 }
1500
1501                 if (!sbi->s_jquota_fmt) {
1502                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1503                                         "not specified.\n");
1504                         return 0;
1505                 }
1506         } else {
1507                 if (sbi->s_jquota_fmt) {
1508                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1509                                         "specified with no journaling "
1510                                         "enabled.\n");
1511                         return 0;
1512                 }
1513         }
1514 #endif
1515         return 1;
1516 }
1517
1518 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1519                             int read_only)
1520 {
1521         struct ext4_sb_info *sbi = EXT4_SB(sb);
1522         int res = 0;
1523
1524         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1525                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1526                        "forcing read-only mode\n");
1527                 res = MS_RDONLY;
1528         }
1529         if (read_only)
1530                 return res;
1531         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1532                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1533                        "running e2fsck is recommended\n");
1534         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1535                 printk(KERN_WARNING
1536                        "EXT4-fs warning: mounting fs with errors, "
1537                        "running e2fsck is recommended\n");
1538         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1539                  le16_to_cpu(es->s_mnt_count) >=
1540                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1541                 printk(KERN_WARNING
1542                        "EXT4-fs warning: maximal mount count reached, "
1543                        "running e2fsck is recommended\n");
1544         else if (le32_to_cpu(es->s_checkinterval) &&
1545                 (le32_to_cpu(es->s_lastcheck) +
1546                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1547                 printk(KERN_WARNING
1548                        "EXT4-fs warning: checktime reached, "
1549                        "running e2fsck is recommended\n");
1550         if (!sbi->s_journal)
1551                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1552         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1553                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1554         le16_add_cpu(&es->s_mnt_count, 1);
1555         es->s_mtime = cpu_to_le32(get_seconds());
1556         ext4_update_dynamic_rev(sb);
1557         if (sbi->s_journal)
1558                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1559
1560         ext4_commit_super(sb, es, 1);
1561         if (test_opt(sb, DEBUG))
1562                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1563                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1564                         sb->s_blocksize,
1565                         sbi->s_groups_count,
1566                         EXT4_BLOCKS_PER_GROUP(sb),
1567                         EXT4_INODES_PER_GROUP(sb),
1568                         sbi->s_mount_opt);
1569
1570         if (EXT4_SB(sb)->s_journal) {
1571                 printk(KERN_INFO "EXT4 FS on %s, %s journal on %s\n",
1572                        sb->s_id, EXT4_SB(sb)->s_journal->j_inode ? "internal" :
1573                        "external", EXT4_SB(sb)->s_journal->j_devname);
1574         } else {
1575                 printk(KERN_INFO "EXT4 FS on %s, no journal\n", sb->s_id);
1576         }
1577         return res;
1578 }
1579
1580 static int ext4_fill_flex_info(struct super_block *sb)
1581 {
1582         struct ext4_sb_info *sbi = EXT4_SB(sb);
1583         struct ext4_group_desc *gdp = NULL;
1584         struct buffer_head *bh;
1585         ext4_group_t flex_group_count;
1586         ext4_group_t flex_group;
1587         int groups_per_flex = 0;
1588         int i;
1589
1590         if (!sbi->s_es->s_log_groups_per_flex) {
1591                 sbi->s_log_groups_per_flex = 0;
1592                 return 1;
1593         }
1594
1595         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1596         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1597
1598         /* We allocate both existing and potentially added groups */
1599         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1600                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1601                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1602         sbi->s_flex_groups = kzalloc(flex_group_count *
1603                                      sizeof(struct flex_groups), GFP_KERNEL);
1604         if (sbi->s_flex_groups == NULL) {
1605                 printk(KERN_ERR "EXT4-fs: not enough memory for "
1606                                 "%u flex groups\n", flex_group_count);
1607                 goto failed;
1608         }
1609
1610         for (i = 0; i < sbi->s_groups_count; i++) {
1611                 gdp = ext4_get_group_desc(sb, i, &bh);
1612
1613                 flex_group = ext4_flex_group(sbi, i);
1614                 sbi->s_flex_groups[flex_group].free_inodes +=
1615                         ext4_free_inodes_count(sb, gdp);
1616                 sbi->s_flex_groups[flex_group].free_blocks +=
1617                         ext4_free_blks_count(sb, gdp);
1618         }
1619
1620         return 1;
1621 failed:
1622         return 0;
1623 }
1624
1625 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1626                             struct ext4_group_desc *gdp)
1627 {
1628         __u16 crc = 0;
1629
1630         if (sbi->s_es->s_feature_ro_compat &
1631             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1632                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1633                 __le32 le_group = cpu_to_le32(block_group);
1634
1635                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1636                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1637                 crc = crc16(crc, (__u8 *)gdp, offset);
1638                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1639                 /* for checksum of struct ext4_group_desc do the rest...*/
1640                 if ((sbi->s_es->s_feature_incompat &
1641                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1642                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1643                         crc = crc16(crc, (__u8 *)gdp + offset,
1644                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1645                                         offset);
1646         }
1647
1648         return cpu_to_le16(crc);
1649 }
1650
1651 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1652                                 struct ext4_group_desc *gdp)
1653 {
1654         if ((sbi->s_es->s_feature_ro_compat &
1655              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1656             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1657                 return 0;
1658
1659         return 1;
1660 }
1661
1662 /* Called at mount-time, super-block is locked */
1663 static int ext4_check_descriptors(struct super_block *sb)
1664 {
1665         struct ext4_sb_info *sbi = EXT4_SB(sb);
1666         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1667         ext4_fsblk_t last_block;
1668         ext4_fsblk_t block_bitmap;
1669         ext4_fsblk_t inode_bitmap;
1670         ext4_fsblk_t inode_table;
1671         int flexbg_flag = 0;
1672         ext4_group_t i;
1673
1674         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1675                 flexbg_flag = 1;
1676
1677         ext4_debug("Checking group descriptors");
1678
1679         for (i = 0; i < sbi->s_groups_count; i++) {
1680                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1681
1682                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1683                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1684                 else
1685                         last_block = first_block +
1686                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1687
1688                 block_bitmap = ext4_block_bitmap(sb, gdp);
1689                 if (block_bitmap < first_block || block_bitmap > last_block) {
1690                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1691                                "Block bitmap for group %u not in group "
1692                                "(block %llu)!\n", i, block_bitmap);
1693                         return 0;
1694                 }
1695                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1696                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1697                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1698                                "Inode bitmap for group %u not in group "
1699                                "(block %llu)!\n", i, inode_bitmap);
1700                         return 0;
1701                 }
1702                 inode_table = ext4_inode_table(sb, gdp);
1703                 if (inode_table < first_block ||
1704                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1705                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1706                                "Inode table for group %u not in group "
1707                                "(block %llu)!\n", i, inode_table);
1708                         return 0;
1709                 }
1710                 spin_lock(sb_bgl_lock(sbi, i));
1711                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1712                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1713                                "Checksum for group %u failed (%u!=%u)\n",
1714                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1715                                gdp)), le16_to_cpu(gdp->bg_checksum));
1716                         if (!(sb->s_flags & MS_RDONLY)) {
1717                                 spin_unlock(sb_bgl_lock(sbi, i));
1718                                 return 0;
1719                         }
1720                 }
1721                 spin_unlock(sb_bgl_lock(sbi, i));
1722                 if (!flexbg_flag)
1723                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1724         }
1725
1726         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1727         sbi->s_es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
1728         return 1;
1729 }
1730
1731 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1732  * the superblock) which were deleted from all directories, but held open by
1733  * a process at the time of a crash.  We walk the list and try to delete these
1734  * inodes at recovery time (only with a read-write filesystem).
1735  *
1736  * In order to keep the orphan inode chain consistent during traversal (in
1737  * case of crash during recovery), we link each inode into the superblock
1738  * orphan list_head and handle it the same way as an inode deletion during
1739  * normal operation (which journals the operations for us).
1740  *
1741  * We only do an iget() and an iput() on each inode, which is very safe if we
1742  * accidentally point at an in-use or already deleted inode.  The worst that
1743  * can happen in this case is that we get a "bit already cleared" message from
1744  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1745  * e2fsck was run on this filesystem, and it must have already done the orphan
1746  * inode cleanup for us, so we can safely abort without any further action.
1747  */
1748 static void ext4_orphan_cleanup(struct super_block *sb,
1749                                 struct ext4_super_block *es)
1750 {
1751         unsigned int s_flags = sb->s_flags;
1752         int nr_orphans = 0, nr_truncates = 0;
1753 #ifdef CONFIG_QUOTA
1754         int i;
1755 #endif
1756         if (!es->s_last_orphan) {
1757                 jbd_debug(4, "no orphan inodes to clean up\n");
1758                 return;
1759         }
1760
1761         if (bdev_read_only(sb->s_bdev)) {
1762                 printk(KERN_ERR "EXT4-fs: write access "
1763                         "unavailable, skipping orphan cleanup.\n");
1764                 return;
1765         }
1766
1767         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1768                 if (es->s_last_orphan)
1769                         jbd_debug(1, "Errors on filesystem, "
1770                                   "clearing orphan list.\n");
1771                 es->s_last_orphan = 0;
1772                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1773                 return;
1774         }
1775
1776         if (s_flags & MS_RDONLY) {
1777                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1778                        sb->s_id);
1779                 sb->s_flags &= ~MS_RDONLY;
1780         }
1781 #ifdef CONFIG_QUOTA
1782         /* Needed for iput() to work correctly and not trash data */
1783         sb->s_flags |= MS_ACTIVE;
1784         /* Turn on quotas so that they are updated correctly */
1785         for (i = 0; i < MAXQUOTAS; i++) {
1786                 if (EXT4_SB(sb)->s_qf_names[i]) {
1787                         int ret = ext4_quota_on_mount(sb, i);
1788                         if (ret < 0)
1789                                 printk(KERN_ERR
1790                                         "EXT4-fs: Cannot turn on journaled "
1791                                         "quota: error %d\n", ret);
1792                 }
1793         }
1794 #endif
1795
1796         while (es->s_last_orphan) {
1797                 struct inode *inode;
1798
1799                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1800                 if (IS_ERR(inode)) {
1801                         es->s_last_orphan = 0;
1802                         break;
1803                 }
1804
1805                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1806                 DQUOT_INIT(inode);
1807                 if (inode->i_nlink) {
1808                         printk(KERN_DEBUG
1809                                 "%s: truncating inode %lu to %lld bytes\n",
1810                                 __func__, inode->i_ino, inode->i_size);
1811                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1812                                   inode->i_ino, inode->i_size);
1813                         ext4_truncate(inode);
1814                         nr_truncates++;
1815                 } else {
1816                         printk(KERN_DEBUG
1817                                 "%s: deleting unreferenced inode %lu\n",
1818                                 __func__, inode->i_ino);
1819                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1820                                   inode->i_ino);
1821                         nr_orphans++;
1822                 }
1823                 iput(inode);  /* The delete magic happens here! */
1824         }
1825
1826 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1827
1828         if (nr_orphans)
1829                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1830                        sb->s_id, PLURAL(nr_orphans));
1831         if (nr_truncates)
1832                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1833                        sb->s_id, PLURAL(nr_truncates));
1834 #ifdef CONFIG_QUOTA
1835         /* Turn quotas off */
1836         for (i = 0; i < MAXQUOTAS; i++) {
1837                 if (sb_dqopt(sb)->files[i])
1838                         vfs_quota_off(sb, i, 0);
1839         }
1840 #endif
1841         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1842 }
1843 /*
1844  * Maximal extent format file size.
1845  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1846  * extent format containers, within a sector_t, and within i_blocks
1847  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1848  * so that won't be a limiting factor.
1849  *
1850  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1851  */
1852 static loff_t ext4_max_size(int blkbits, int has_huge_files)
1853 {
1854         loff_t res;
1855         loff_t upper_limit = MAX_LFS_FILESIZE;
1856
1857         /* small i_blocks in vfs inode? */
1858         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
1859                 /*
1860                  * CONFIG_LBD is not enabled implies the inode
1861                  * i_block represent total blocks in 512 bytes
1862                  * 32 == size of vfs inode i_blocks * 8
1863                  */
1864                 upper_limit = (1LL << 32) - 1;
1865
1866                 /* total blocks in file system block size */
1867                 upper_limit >>= (blkbits - 9);
1868                 upper_limit <<= blkbits;
1869         }
1870
1871         /* 32-bit extent-start container, ee_block */
1872         res = 1LL << 32;
1873         res <<= blkbits;
1874         res -= 1;
1875
1876         /* Sanity check against vm- & vfs- imposed limits */
1877         if (res > upper_limit)
1878                 res = upper_limit;
1879
1880         return res;
1881 }
1882
1883 /*
1884  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1885  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1886  * We need to be 1 filesystem block less than the 2^48 sector limit.
1887  */
1888 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
1889 {
1890         loff_t res = EXT4_NDIR_BLOCKS;
1891         int meta_blocks;
1892         loff_t upper_limit;
1893         /* This is calculated to be the largest file size for a
1894          * dense, bitmapped file such that the total number of
1895          * sectors in the file, including data and all indirect blocks,
1896          * does not exceed 2^48 -1
1897          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1898          * total number of  512 bytes blocks of the file
1899          */
1900
1901         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
1902                 /*
1903                  * !has_huge_files or CONFIG_LBD is not enabled
1904                  * implies the inode i_block represent total blocks in
1905                  * 512 bytes 32 == size of vfs inode i_blocks * 8
1906                  */
1907                 upper_limit = (1LL << 32) - 1;
1908
1909                 /* total blocks in file system block size */
1910                 upper_limit >>= (bits - 9);
1911
1912         } else {
1913                 /*
1914                  * We use 48 bit ext4_inode i_blocks
1915                  * With EXT4_HUGE_FILE_FL set the i_blocks
1916                  * represent total number of blocks in
1917                  * file system block size
1918                  */
1919                 upper_limit = (1LL << 48) - 1;
1920
1921         }
1922
1923         /* indirect blocks */
1924         meta_blocks = 1;
1925         /* double indirect blocks */
1926         meta_blocks += 1 + (1LL << (bits-2));
1927         /* tripple indirect blocks */
1928         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1929
1930         upper_limit -= meta_blocks;
1931         upper_limit <<= bits;
1932
1933         res += 1LL << (bits-2);
1934         res += 1LL << (2*(bits-2));
1935         res += 1LL << (3*(bits-2));
1936         res <<= bits;
1937         if (res > upper_limit)
1938                 res = upper_limit;
1939
1940         if (res > MAX_LFS_FILESIZE)
1941                 res = MAX_LFS_FILESIZE;
1942
1943         return res;
1944 }
1945
1946 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1947                                 ext4_fsblk_t logical_sb_block, int nr)
1948 {
1949         struct ext4_sb_info *sbi = EXT4_SB(sb);
1950         ext4_group_t bg, first_meta_bg;
1951         int has_super = 0;
1952
1953         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1954
1955         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1956             nr < first_meta_bg)
1957                 return logical_sb_block + nr + 1;
1958         bg = sbi->s_desc_per_block * nr;
1959         if (ext4_bg_has_super(sb, bg))
1960                 has_super = 1;
1961         return (has_super + ext4_group_first_block_no(sb, bg));
1962 }
1963
1964 /**
1965  * ext4_get_stripe_size: Get the stripe size.
1966  * @sbi: In memory super block info
1967  *
1968  * If we have specified it via mount option, then
1969  * use the mount option value. If the value specified at mount time is
1970  * greater than the blocks per group use the super block value.
1971  * If the super block value is greater than blocks per group return 0.
1972  * Allocator needs it be less than blocks per group.
1973  *
1974  */
1975 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1976 {
1977         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1978         unsigned long stripe_width =
1979                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1980
1981         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1982                 return sbi->s_stripe;
1983
1984         if (stripe_width <= sbi->s_blocks_per_group)
1985                 return stripe_width;
1986
1987         if (stride <= sbi->s_blocks_per_group)
1988                 return stride;
1989
1990         return 0;
1991 }
1992
1993 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
1994                                 __releases(kernel_lock)
1995                                 __acquires(kernel_lock)
1996
1997 {
1998         struct buffer_head *bh;
1999         struct ext4_super_block *es = NULL;
2000         struct ext4_sb_info *sbi;
2001         ext4_fsblk_t block;
2002         ext4_fsblk_t sb_block = get_sb_block(&data);
2003         ext4_fsblk_t logical_sb_block;
2004         unsigned long offset = 0;
2005         unsigned long journal_devnum = 0;
2006         unsigned long def_mount_opts;
2007         struct inode *root;
2008         char *cp;
2009         const char *descr;
2010         int ret = -EINVAL;
2011         int blocksize;
2012         unsigned int db_count;
2013         unsigned int i;
2014         int needs_recovery, has_huge_files;
2015         int features;
2016         __u64 blocks_count;
2017         int err;
2018         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2019
2020         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2021         if (!sbi)
2022                 return -ENOMEM;
2023         sb->s_fs_info = sbi;
2024         sbi->s_mount_opt = 0;
2025         sbi->s_resuid = EXT4_DEF_RESUID;
2026         sbi->s_resgid = EXT4_DEF_RESGID;
2027         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2028         sbi->s_sb_block = sb_block;
2029
2030         unlock_kernel();
2031
2032         /* Cleanup superblock name */
2033         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2034                 *cp = '!';
2035
2036         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2037         if (!blocksize) {
2038                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
2039                 goto out_fail;
2040         }
2041
2042         /*
2043          * The ext4 superblock will not be buffer aligned for other than 1kB
2044          * block sizes.  We need to calculate the offset from buffer start.
2045          */
2046         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2047                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2048                 offset = do_div(logical_sb_block, blocksize);
2049         } else {
2050                 logical_sb_block = sb_block;
2051         }
2052
2053         if (!(bh = sb_bread(sb, logical_sb_block))) {
2054                 printk(KERN_ERR "EXT4-fs: unable to read superblock\n");
2055                 goto out_fail;
2056         }
2057         /*
2058          * Note: s_es must be initialized as soon as possible because
2059          *       some ext4 macro-instructions depend on its value
2060          */
2061         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2062         sbi->s_es = es;
2063         sb->s_magic = le16_to_cpu(es->s_magic);
2064         if (sb->s_magic != EXT4_SUPER_MAGIC)
2065                 goto cantfind_ext4;
2066
2067         /* Set defaults before we parse the mount options */
2068         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2069         if (def_mount_opts & EXT4_DEFM_DEBUG)
2070                 set_opt(sbi->s_mount_opt, DEBUG);
2071         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
2072                 set_opt(sbi->s_mount_opt, GRPID);
2073         if (def_mount_opts & EXT4_DEFM_UID16)
2074                 set_opt(sbi->s_mount_opt, NO_UID32);
2075 #ifdef CONFIG_EXT4_FS_XATTR
2076         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2077                 set_opt(sbi->s_mount_opt, XATTR_USER);
2078 #endif
2079 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2080         if (def_mount_opts & EXT4_DEFM_ACL)
2081                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2082 #endif
2083         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2084                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
2085         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2086                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
2087         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2088                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
2089
2090         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2091                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2092         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2093                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2094         else
2095                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2096
2097         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2098         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2099         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2100         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2101         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2102
2103         set_opt(sbi->s_mount_opt, RESERVATION);
2104         set_opt(sbi->s_mount_opt, BARRIER);
2105
2106         /*
2107          * enable delayed allocation by default
2108          * Use -o nodelalloc to turn it off
2109          */
2110         set_opt(sbi->s_mount_opt, DELALLOC);
2111
2112
2113         if (!parse_options((char *) data, sb, &journal_devnum,
2114                            &journal_ioprio, NULL, 0))
2115                 goto failed_mount;
2116
2117         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2118                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2119
2120         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2121             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2122              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2123              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2124                 printk(KERN_WARNING
2125                        "EXT4-fs warning: feature flags set on rev 0 fs, "
2126                        "running e2fsck is recommended\n");
2127
2128         /*
2129          * Check feature flags regardless of the revision level, since we
2130          * previously didn't change the revision level when setting the flags,
2131          * so there is a chance incompat flags are set on a rev 0 filesystem.
2132          */
2133         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2134         if (features) {
2135                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2136                        "unsupported optional features (%x).\n", sb->s_id,
2137                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2138                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2139                 goto failed_mount;
2140         }
2141         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2142         if (!(sb->s_flags & MS_RDONLY) && features) {
2143                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2144                        "unsupported optional features (%x).\n", sb->s_id,
2145                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2146                         ~EXT4_FEATURE_RO_COMPAT_SUPP));
2147                 goto failed_mount;
2148         }
2149         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2150                                     EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2151         if (has_huge_files) {
2152                 /*
2153                  * Large file size enabled file system can only be
2154                  * mount if kernel is build with CONFIG_LBD
2155                  */
2156                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2157                                 !(sb->s_flags & MS_RDONLY)) {
2158                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2159                                         "files cannot be mounted read-write "
2160                                         "without CONFIG_LBD.\n", sb->s_id);
2161                         goto failed_mount;
2162                 }
2163         }
2164         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2165
2166         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2167             blocksize > EXT4_MAX_BLOCK_SIZE) {
2168                 printk(KERN_ERR
2169                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2170                        blocksize, sb->s_id);
2171                 goto failed_mount;
2172         }
2173
2174         if (sb->s_blocksize != blocksize) {
2175
2176                 /* Validate the filesystem blocksize */
2177                 if (!sb_set_blocksize(sb, blocksize)) {
2178                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2179                                         blocksize);
2180                         goto failed_mount;
2181                 }
2182
2183                 brelse(bh);
2184                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2185                 offset = do_div(logical_sb_block, blocksize);
2186                 bh = sb_bread(sb, logical_sb_block);
2187                 if (!bh) {
2188                         printk(KERN_ERR
2189                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2190                         goto failed_mount;
2191                 }
2192                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2193                 sbi->s_es = es;
2194                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2195                         printk(KERN_ERR
2196                                "EXT4-fs: Magic mismatch, very weird !\n");
2197                         goto failed_mount;
2198                 }
2199         }
2200
2201         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2202                                                       has_huge_files);
2203         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2204
2205         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2206                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2207                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2208         } else {
2209                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2210                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2211                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2212                     (!is_power_of_2(sbi->s_inode_size)) ||
2213                     (sbi->s_inode_size > blocksize)) {
2214                         printk(KERN_ERR
2215                                "EXT4-fs: unsupported inode size: %d\n",
2216                                sbi->s_inode_size);
2217                         goto failed_mount;
2218                 }
2219                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2220                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2221         }
2222         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2223         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2224                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2225                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2226                     !is_power_of_2(sbi->s_desc_size)) {
2227                         printk(KERN_ERR
2228                                "EXT4-fs: unsupported descriptor size %lu\n",
2229                                sbi->s_desc_size);
2230                         goto failed_mount;
2231                 }
2232         } else
2233                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2234         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2235         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2236         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2237                 goto cantfind_ext4;
2238         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2239         if (sbi->s_inodes_per_block == 0)
2240                 goto cantfind_ext4;
2241         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2242                                         sbi->s_inodes_per_block;
2243         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2244         sbi->s_sbh = bh;
2245         sbi->s_mount_state = le16_to_cpu(es->s_state);
2246         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2247         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2248         for (i = 0; i < 4; i++)
2249                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2250         sbi->s_def_hash_version = es->s_def_hash_version;
2251         i = le32_to_cpu(es->s_flags);
2252         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2253                 sbi->s_hash_unsigned = 3;
2254         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2255 #ifdef __CHAR_UNSIGNED__
2256                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2257                 sbi->s_hash_unsigned = 3;
2258 #else
2259                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2260 #endif
2261                 sb->s_dirt = 1;
2262         }
2263
2264         if (sbi->s_blocks_per_group > blocksize * 8) {
2265                 printk(KERN_ERR
2266                        "EXT4-fs: #blocks per group too big: %lu\n",
2267                        sbi->s_blocks_per_group);
2268                 goto failed_mount;
2269         }
2270         if (sbi->s_inodes_per_group > blocksize * 8) {
2271                 printk(KERN_ERR
2272                        "EXT4-fs: #inodes per group too big: %lu\n",
2273                        sbi->s_inodes_per_group);
2274                 goto failed_mount;
2275         }
2276
2277         if (ext4_blocks_count(es) >
2278                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2279                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2280                         " too large to mount safely\n", sb->s_id);
2281                 if (sizeof(sector_t) < 8)
2282                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2283                                         "enabled\n");
2284                 goto failed_mount;
2285         }
2286
2287         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2288                 goto cantfind_ext4;
2289
2290         /*
2291          * It makes no sense for the first data block to be beyond the end
2292          * of the filesystem.
2293          */
2294         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2295                 printk(KERN_WARNING "EXT4-fs: bad geometry: first data"
2296                        "block %u is beyond end of filesystem (%llu)\n",
2297                        le32_to_cpu(es->s_first_data_block),
2298                        ext4_blocks_count(es));
2299                 goto failed_mount;
2300         }
2301         blocks_count = (ext4_blocks_count(es) -
2302                         le32_to_cpu(es->s_first_data_block) +
2303                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2304         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2305         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2306                 printk(KERN_WARNING "EXT4-fs: groups count too large: %u "
2307                        "(block count %llu, first data block %u, "
2308                        "blocks per group %lu)\n", sbi->s_groups_count,
2309                        ext4_blocks_count(es),
2310                        le32_to_cpu(es->s_first_data_block),
2311                        EXT4_BLOCKS_PER_GROUP(sb));
2312                 goto failed_mount;
2313         }
2314         sbi->s_groups_count = blocks_count;
2315         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2316                    EXT4_DESC_PER_BLOCK(sb);
2317         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2318                                     GFP_KERNEL);
2319         if (sbi->s_group_desc == NULL) {
2320                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
2321                 goto failed_mount;
2322         }
2323
2324 #ifdef CONFIG_PROC_FS
2325         if (ext4_proc_root)
2326                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2327
2328         if (sbi->s_proc)
2329                 proc_create_data("inode_readahead_blks", 0644, sbi->s_proc,
2330                                  &ext4_ui_proc_fops,
2331                                  &sbi->s_inode_readahead_blks);
2332 #endif
2333
2334         bgl_lock_init(&sbi->s_blockgroup_lock);
2335
2336         for (i = 0; i < db_count; i++) {
2337                 block = descriptor_loc(sb, logical_sb_block, i);
2338                 sbi->s_group_desc[i] = sb_bread(sb, block);
2339                 if (!sbi->s_group_desc[i]) {
2340                         printk(KERN_ERR "EXT4-fs: "
2341                                "can't read group descriptor %d\n", i);
2342                         db_count = i;
2343                         goto failed_mount2;
2344                 }
2345         }
2346         if (!ext4_check_descriptors(sb)) {
2347                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2348                 goto failed_mount2;
2349         }
2350         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2351                 if (!ext4_fill_flex_info(sb)) {
2352                         printk(KERN_ERR
2353                                "EXT4-fs: unable to initialize "
2354                                "flex_bg meta info!\n");
2355                         goto failed_mount2;
2356                 }
2357
2358         sbi->s_gdb_count = db_count;
2359         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2360         spin_lock_init(&sbi->s_next_gen_lock);
2361
2362         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2363                         ext4_count_free_blocks(sb));
2364         if (!err) {
2365                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2366                                 ext4_count_free_inodes(sb));
2367         }
2368         if (!err) {
2369                 err = percpu_counter_init(&sbi->s_dirs_counter,
2370                                 ext4_count_dirs(sb));
2371         }
2372         if (!err) {
2373                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2374         }
2375         if (err) {
2376                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2377                 goto failed_mount3;
2378         }
2379
2380         sbi->s_stripe = ext4_get_stripe_size(sbi);
2381
2382         /*
2383          * set up enough so that it can read an inode
2384          */
2385         sb->s_op = &ext4_sops;
2386         sb->s_export_op = &ext4_export_ops;
2387         sb->s_xattr = ext4_xattr_handlers;
2388 #ifdef CONFIG_QUOTA
2389         sb->s_qcop = &ext4_qctl_operations;
2390         sb->dq_op = &ext4_quota_operations;
2391 #endif
2392         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2393
2394         sb->s_root = NULL;
2395
2396         needs_recovery = (es->s_last_orphan != 0 ||
2397                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2398                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2399
2400         /*
2401          * The first inode we look at is the journal inode.  Don't try
2402          * root first: it may be modified in the journal!
2403          */
2404         if (!test_opt(sb, NOLOAD) &&
2405             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2406                 if (ext4_load_journal(sb, es, journal_devnum))
2407                         goto failed_mount3;
2408                 if (!(sb->s_flags & MS_RDONLY) &&
2409                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2410                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2411                                "ext4_fill_super: Journal transaction "
2412                                "%u is corrupt\n", sb->s_id,
2413                                EXT4_SB(sb)->s_journal->j_failed_commit);
2414                         if (test_opt(sb, ERRORS_RO)) {
2415                                 printk(KERN_CRIT
2416                                        "Mounting filesystem read-only\n");
2417                                 sb->s_flags |= MS_RDONLY;
2418                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2419                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2420                         }
2421                         if (test_opt(sb, ERRORS_PANIC)) {
2422                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2423                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2424                                 ext4_commit_super(sb, es, 1);
2425                                 goto failed_mount4;
2426                         }
2427                 }
2428         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2429               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2430                 printk(KERN_ERR "EXT4-fs: required journal recovery "
2431                        "suppressed and not mounted read-only\n");
2432                 goto failed_mount4;
2433         } else {
2434                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2435                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2436                 sbi->s_journal = NULL;
2437                 needs_recovery = 0;
2438                 goto no_journal;
2439         }
2440
2441         if (ext4_blocks_count(es) > 0xffffffffULL &&
2442             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2443                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2444                 printk(KERN_ERR "EXT4-fs: Failed to set 64-bit journal feature\n");
2445                 goto failed_mount4;
2446         }
2447
2448         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2449                 jbd2_journal_set_features(sbi->s_journal,
2450                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2451                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2452         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2453                 jbd2_journal_set_features(sbi->s_journal,
2454                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2455                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2456                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2457         } else {
2458                 jbd2_journal_clear_features(sbi->s_journal,
2459                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2460                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2461         }
2462
2463         /* We have now updated the journal if required, so we can
2464          * validate the data journaling mode. */
2465         switch (test_opt(sb, DATA_FLAGS)) {
2466         case 0:
2467                 /* No mode set, assume a default based on the journal
2468                  * capabilities: ORDERED_DATA if the journal can
2469                  * cope, else JOURNAL_DATA
2470                  */
2471                 if (jbd2_journal_check_available_features
2472                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2473                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2474                 else
2475                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2476                 break;
2477
2478         case EXT4_MOUNT_ORDERED_DATA:
2479         case EXT4_MOUNT_WRITEBACK_DATA:
2480                 if (!jbd2_journal_check_available_features
2481                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2482                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2483                                "requested data journaling mode\n");
2484                         goto failed_mount4;
2485                 }
2486         default:
2487                 break;
2488         }
2489         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2490
2491 no_journal:
2492
2493         if (test_opt(sb, NOBH)) {
2494                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2495                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2496                                 "its supported only with writeback mode\n");
2497                         clear_opt(sbi->s_mount_opt, NOBH);
2498                 }
2499         }
2500         /*
2501          * The jbd2_journal_load will have done any necessary log recovery,
2502          * so we can safely mount the rest of the filesystem now.
2503          */
2504
2505         root = ext4_iget(sb, EXT4_ROOT_INO);
2506         if (IS_ERR(root)) {
2507                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2508                 ret = PTR_ERR(root);
2509                 goto failed_mount4;
2510         }
2511         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2512                 iput(root);
2513                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2514                 goto failed_mount4;
2515         }
2516         sb->s_root = d_alloc_root(root);
2517         if (!sb->s_root) {
2518                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2519                 iput(root);
2520                 ret = -ENOMEM;
2521                 goto failed_mount4;
2522         }
2523
2524         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2525
2526         /* determine the minimum size of new large inodes, if present */
2527         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2528                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2529                                                      EXT4_GOOD_OLD_INODE_SIZE;
2530                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2531                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2532                         if (sbi->s_want_extra_isize <
2533                             le16_to_cpu(es->s_want_extra_isize))
2534                                 sbi->s_want_extra_isize =
2535                                         le16_to_cpu(es->s_want_extra_isize);
2536                         if (sbi->s_want_extra_isize <
2537                             le16_to_cpu(es->s_min_extra_isize))
2538                                 sbi->s_want_extra_isize =
2539                                         le16_to_cpu(es->s_min_extra_isize);
2540                 }
2541         }
2542         /* Check if enough inode space is available */
2543         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2544                                                         sbi->s_inode_size) {
2545                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2546                                                        EXT4_GOOD_OLD_INODE_SIZE;
2547                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2548                         "available.\n");
2549         }
2550
2551         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2552                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2553                                 "requested data journaling mode\n");
2554                 clear_opt(sbi->s_mount_opt, DELALLOC);
2555         } else if (test_opt(sb, DELALLOC))
2556                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2557
2558         ext4_ext_init(sb);
2559         err = ext4_mb_init(sb, needs_recovery);
2560         if (err) {
2561                 printk(KERN_ERR "EXT4-fs: failed to initalize mballoc (%d)\n",
2562                        err);
2563                 goto failed_mount4;
2564         }
2565
2566         /*
2567          * akpm: core read_super() calls in here with the superblock locked.
2568          * That deadlocks, because orphan cleanup needs to lock the superblock
2569          * in numerous places.  Here we just pop the lock - it's relatively
2570          * harmless, because we are now ready to accept write_super() requests,
2571          * and aviro says that's the only reason for hanging onto the
2572          * superblock lock.
2573          */
2574         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2575         ext4_orphan_cleanup(sb, es);
2576         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2577         if (needs_recovery) {
2578                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2579                 ext4_mark_recovery_complete(sb, es);
2580         }
2581         if (EXT4_SB(sb)->s_journal) {
2582                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2583                         descr = " journalled data mode";
2584                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2585                         descr = " ordered data mode";
2586                 else
2587                         descr = " writeback data mode";
2588         } else
2589                 descr = "out journal";
2590
2591         printk(KERN_INFO "EXT4-fs: mounted filesystem %s with%s\n",
2592                sb->s_id, descr);
2593
2594         lock_kernel();
2595         return 0;
2596
2597 cantfind_ext4:
2598         if (!silent)
2599                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2600                        sb->s_id);
2601         goto failed_mount;
2602
2603 failed_mount4:
2604         printk(KERN_ERR "EXT4-fs (device %s): mount failed\n", sb->s_id);
2605         if (sbi->s_journal) {
2606                 jbd2_journal_destroy(sbi->s_journal);
2607                 sbi->s_journal = NULL;
2608         }
2609 failed_mount3:
2610         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2611         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2612         percpu_counter_destroy(&sbi->s_dirs_counter);
2613         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2614 failed_mount2:
2615         for (i = 0; i < db_count; i++)
2616                 brelse(sbi->s_group_desc[i]);
2617         kfree(sbi->s_group_desc);
2618 failed_mount:
2619         if (sbi->s_proc) {
2620                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
2621                 remove_proc_entry(sb->s_id, ext4_proc_root);
2622         }
2623 #ifdef CONFIG_QUOTA
2624         for (i = 0; i < MAXQUOTAS; i++)
2625                 kfree(sbi->s_qf_names[i]);
2626 #endif
2627         ext4_blkdev_remove(sbi);
2628         brelse(bh);
2629 out_fail:
2630         sb->s_fs_info = NULL;
2631         kfree(sbi);
2632         lock_kernel();
2633         return ret;
2634 }
2635
2636 /*
2637  * Setup any per-fs journal parameters now.  We'll do this both on
2638  * initial mount, once the journal has been initialised but before we've
2639  * done any recovery; and again on any subsequent remount.
2640  */
2641 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2642 {
2643         struct ext4_sb_info *sbi = EXT4_SB(sb);
2644
2645         journal->j_commit_interval = sbi->s_commit_interval;
2646         journal->j_min_batch_time = sbi->s_min_batch_time;
2647         journal->j_max_batch_time = sbi->s_max_batch_time;
2648
2649         spin_lock(&journal->j_state_lock);
2650         if (test_opt(sb, BARRIER))
2651                 journal->j_flags |= JBD2_BARRIER;
2652         else
2653                 journal->j_flags &= ~JBD2_BARRIER;
2654         if (test_opt(sb, DATA_ERR_ABORT))
2655                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
2656         else
2657                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
2658         spin_unlock(&journal->j_state_lock);
2659 }
2660
2661 static journal_t *ext4_get_journal(struct super_block *sb,
2662                                    unsigned int journal_inum)
2663 {
2664         struct inode *journal_inode;
2665         journal_t *journal;
2666
2667         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
2668
2669         /* First, test for the existence of a valid inode on disk.  Bad
2670          * things happen if we iget() an unused inode, as the subsequent
2671          * iput() will try to delete it. */
2672
2673         journal_inode = ext4_iget(sb, journal_inum);
2674         if (IS_ERR(journal_inode)) {
2675                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2676                 return NULL;
2677         }
2678         if (!journal_inode->i_nlink) {
2679                 make_bad_inode(journal_inode);
2680                 iput(journal_inode);
2681                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2682                 return NULL;
2683         }
2684
2685         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
2686                   journal_inode, journal_inode->i_size);
2687         if (!S_ISREG(journal_inode->i_mode)) {
2688                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2689                 iput(journal_inode);
2690                 return NULL;
2691         }
2692
2693         journal = jbd2_journal_init_inode(journal_inode);
2694         if (!journal) {
2695                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2696                 iput(journal_inode);
2697                 return NULL;
2698         }
2699         journal->j_private = sb;
2700         ext4_init_journal_params(sb, journal);
2701         return journal;
2702 }
2703
2704 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2705                                        dev_t j_dev)
2706 {
2707         struct buffer_head *bh;
2708         journal_t *journal;
2709         ext4_fsblk_t start;
2710         ext4_fsblk_t len;
2711         int hblock, blocksize;
2712         ext4_fsblk_t sb_block;
2713         unsigned long offset;
2714         struct ext4_super_block *es;
2715         struct block_device *bdev;
2716
2717         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
2718
2719         bdev = ext4_blkdev_get(j_dev);
2720         if (bdev == NULL)
2721                 return NULL;
2722
2723         if (bd_claim(bdev, sb)) {
2724                 printk(KERN_ERR
2725                         "EXT4-fs: failed to claim external journal device.\n");
2726                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2727                 return NULL;
2728         }
2729
2730         blocksize = sb->s_blocksize;
2731         hblock = bdev_hardsect_size(bdev);
2732         if (blocksize < hblock) {
2733                 printk(KERN_ERR
2734                         "EXT4-fs: blocksize too small for journal device.\n");
2735                 goto out_bdev;
2736         }
2737
2738         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2739         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2740         set_blocksize(bdev, blocksize);
2741         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2742                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2743                        "external journal\n");
2744                 goto out_bdev;
2745         }
2746
2747         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2748         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2749             !(le32_to_cpu(es->s_feature_incompat) &
2750               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2751                 printk(KERN_ERR "EXT4-fs: external journal has "
2752                                         "bad superblock\n");
2753                 brelse(bh);
2754                 goto out_bdev;
2755         }
2756
2757         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2758                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2759                 brelse(bh);
2760                 goto out_bdev;
2761         }
2762
2763         len = ext4_blocks_count(es);
2764         start = sb_block + 1;
2765         brelse(bh);     /* we're done with the superblock */
2766
2767         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2768                                         start, len, blocksize);
2769         if (!journal) {
2770                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2771                 goto out_bdev;
2772         }
2773         journal->j_private = sb;
2774         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2775         wait_on_buffer(journal->j_sb_buffer);
2776         if (!buffer_uptodate(journal->j_sb_buffer)) {
2777                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2778                 goto out_journal;
2779         }
2780         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2781                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2782                                         "user (unsupported) - %d\n",
2783                         be32_to_cpu(journal->j_superblock->s_nr_users));
2784                 goto out_journal;
2785         }
2786         EXT4_SB(sb)->journal_bdev = bdev;
2787         ext4_init_journal_params(sb, journal);
2788         return journal;
2789 out_journal:
2790         jbd2_journal_destroy(journal);
2791 out_bdev:
2792         ext4_blkdev_put(bdev);
2793         return NULL;
2794 }
2795
2796 static int ext4_load_journal(struct super_block *sb,
2797                              struct ext4_super_block *es,
2798                              unsigned long journal_devnum)
2799 {
2800         journal_t *journal;
2801         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2802         dev_t journal_dev;
2803         int err = 0;
2804         int really_read_only;
2805
2806         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
2807
2808         if (journal_devnum &&
2809             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2810                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2811                         "numbers have changed\n");
2812                 journal_dev = new_decode_dev(journal_devnum);
2813         } else
2814                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2815
2816         really_read_only = bdev_read_only(sb->s_bdev);
2817
2818         /*
2819          * Are we loading a blank journal or performing recovery after a
2820          * crash?  For recovery, we need to check in advance whether we
2821          * can get read-write access to the device.
2822          */
2823
2824         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2825                 if (sb->s_flags & MS_RDONLY) {
2826                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2827                                         "required on readonly filesystem.\n");
2828                         if (really_read_only) {
2829                                 printk(KERN_ERR "EXT4-fs: write access "
2830                                         "unavailable, cannot proceed.\n");
2831                                 return -EROFS;
2832                         }
2833                         printk(KERN_INFO "EXT4-fs: write access will "
2834                                "be enabled during recovery.\n");
2835                 }
2836         }
2837
2838         if (journal_inum && journal_dev) {
2839                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2840                        "and inode journals!\n");
2841                 return -EINVAL;
2842         }
2843
2844         if (journal_inum) {
2845                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2846                         return -EINVAL;
2847         } else {
2848                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2849                         return -EINVAL;
2850         }
2851
2852         if (journal->j_flags & JBD2_BARRIER)
2853                 printk(KERN_INFO "EXT4-fs: barriers enabled\n");
2854         else
2855                 printk(KERN_INFO "EXT4-fs: barriers disabled\n");
2856
2857         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2858                 err = jbd2_journal_update_format(journal);
2859                 if (err)  {
2860                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2861                         jbd2_journal_destroy(journal);
2862                         return err;
2863                 }
2864         }
2865
2866         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2867                 err = jbd2_journal_wipe(journal, !really_read_only);
2868         if (!err)
2869                 err = jbd2_journal_load(journal);
2870
2871         if (err) {
2872                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2873                 jbd2_journal_destroy(journal);
2874                 return err;
2875         }
2876
2877         EXT4_SB(sb)->s_journal = journal;
2878         ext4_clear_journal_err(sb, es);
2879
2880         if (journal_devnum &&
2881             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2882                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2883                 sb->s_dirt = 1;
2884
2885                 /* Make sure we flush the recovery flag to disk. */
2886                 ext4_commit_super(sb, es, 1);
2887         }
2888
2889         return 0;
2890 }
2891
2892 static void ext4_commit_super(struct super_block *sb,
2893                               struct ext4_super_block *es, int sync)
2894 {
2895         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2896
2897         if (!sbh)
2898                 return;
2899         if (buffer_write_io_error(sbh)) {
2900                 /*
2901                  * Oh, dear.  A previous attempt to write the
2902                  * superblock failed.  This could happen because the
2903                  * USB device was yanked out.  Or it could happen to
2904                  * be a transient write error and maybe the block will
2905                  * be remapped.  Nothing we can do but to retry the
2906                  * write and hope for the best.
2907                  */
2908                 printk(KERN_ERR "EXT4-fs: previous I/O error to "
2909                        "superblock detected for %s.\n", sb->s_id);
2910                 clear_buffer_write_io_error(sbh);
2911                 set_buffer_uptodate(sbh);
2912         }
2913         es->s_wtime = cpu_to_le32(get_seconds());
2914         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
2915                                         &EXT4_SB(sb)->s_freeblocks_counter));
2916         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
2917                                         &EXT4_SB(sb)->s_freeinodes_counter));
2918
2919         BUFFER_TRACE(sbh, "marking dirty");
2920         mark_buffer_dirty(sbh);
2921         if (sync) {
2922                 sync_dirty_buffer(sbh);
2923                 if (buffer_write_io_error(sbh)) {
2924                         printk(KERN_ERR "EXT4-fs: I/O error while writing "
2925                                "superblock for %s.\n", sb->s_id);
2926                         clear_buffer_write_io_error(sbh);
2927                         set_buffer_uptodate(sbh);
2928                 }
2929         }
2930 }
2931
2932
2933 /*
2934  * Have we just finished recovery?  If so, and if we are mounting (or
2935  * remounting) the filesystem readonly, then we will end up with a
2936  * consistent fs on disk.  Record that fact.
2937  */
2938 static void ext4_mark_recovery_complete(struct super_block *sb,
2939                                         struct ext4_super_block *es)
2940 {
2941         journal_t *journal = EXT4_SB(sb)->s_journal;
2942
2943         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2944                 BUG_ON(journal != NULL);
2945                 return;
2946         }
2947         jbd2_journal_lock_updates(journal);
2948         if (jbd2_journal_flush(journal) < 0)
2949                 goto out;
2950
2951         lock_super(sb);
2952         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2953             sb->s_flags & MS_RDONLY) {
2954                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2955                 sb->s_dirt = 0;
2956                 ext4_commit_super(sb, es, 1);
2957         }
2958         unlock_super(sb);
2959
2960 out:
2961         jbd2_journal_unlock_updates(journal);
2962 }
2963
2964 /*
2965  * If we are mounting (or read-write remounting) a filesystem whose journal
2966  * has recorded an error from a previous lifetime, move that error to the
2967  * main filesystem now.
2968  */
2969 static void ext4_clear_journal_err(struct super_block *sb,
2970                                    struct ext4_super_block *es)
2971 {
2972         journal_t *journal;
2973         int j_errno;
2974         const char *errstr;
2975
2976         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
2977
2978         journal = EXT4_SB(sb)->s_journal;
2979
2980         /*
2981          * Now check for any error status which may have been recorded in the
2982          * journal by a prior ext4_error() or ext4_abort()
2983          */
2984
2985         j_errno = jbd2_journal_errno(journal);
2986         if (j_errno) {
2987                 char nbuf[16];
2988
2989                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2990                 ext4_warning(sb, __func__, "Filesystem error recorded "
2991                              "from previous mount: %s", errstr);
2992                 ext4_warning(sb, __func__, "Marking fs in need of "
2993                              "filesystem check.");
2994
2995                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2996                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2997                 ext4_commit_super(sb, es, 1);
2998
2999                 jbd2_journal_clear_err(journal);
3000         }
3001 }
3002
3003 /*
3004  * Force the running and committing transactions to commit,
3005  * and wait on the commit.
3006  */
3007 int ext4_force_commit(struct super_block *sb)
3008 {
3009         journal_t *journal;
3010         int ret = 0;
3011
3012         if (sb->s_flags & MS_RDONLY)
3013                 return 0;
3014
3015         journal = EXT4_SB(sb)->s_journal;
3016         if (journal) {
3017                 sb->s_dirt = 0;
3018                 ret = ext4_journal_force_commit(journal);
3019         }
3020
3021         return ret;
3022 }
3023
3024 /*
3025  * Ext4 always journals updates to the superblock itself, so we don't
3026  * have to propagate any other updates to the superblock on disk at this
3027  * point.  (We can probably nuke this function altogether, and remove
3028  * any mention to sb->s_dirt in all of fs/ext4; eventual cleanup...)
3029  */
3030 static void ext4_write_super(struct super_block *sb)
3031 {
3032         if (EXT4_SB(sb)->s_journal) {
3033                 if (mutex_trylock(&sb->s_lock) != 0)
3034                         BUG();
3035                 sb->s_dirt = 0;
3036         } else {
3037                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
3038         }
3039 }
3040
3041 static int ext4_sync_fs(struct super_block *sb, int wait)
3042 {
3043         int ret = 0;
3044
3045         trace_mark(ext4_sync_fs, "dev %s wait %d", sb->s_id, wait);
3046         sb->s_dirt = 0;
3047         if (EXT4_SB(sb)->s_journal) {
3048                 if (wait)
3049                         ret = ext4_force_commit(sb);
3050                 else
3051                         jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, NULL);
3052         } else {
3053                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, wait);
3054         }
3055         return ret;
3056 }
3057
3058 /*
3059  * LVM calls this function before a (read-only) snapshot is created.  This
3060  * gives us a chance to flush the journal completely and mark the fs clean.
3061  */
3062 static void ext4_write_super_lockfs(struct super_block *sb)
3063 {
3064         sb->s_dirt = 0;
3065
3066         if (!(sb->s_flags & MS_RDONLY)) {
3067                 journal_t *journal = EXT4_SB(sb)->s_journal;
3068
3069                 if (journal) {
3070                         /* Now we set up the journal barrier. */
3071                         jbd2_journal_lock_updates(journal);
3072
3073                         /*
3074                          * We don't want to clear needs_recovery flag when we
3075                          * failed to flush the journal.
3076                          */
3077                         if (jbd2_journal_flush(journal) < 0)
3078                                 return;
3079                 }
3080
3081                 /* Journal blocked and flushed, clear needs_recovery flag. */
3082                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3083                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
3084         }
3085 }
3086
3087 /*
3088  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3089  * flag here, even though the filesystem is not technically dirty yet.
3090  */
3091 static void ext4_unlockfs(struct super_block *sb)
3092 {
3093         if (EXT4_SB(sb)->s_journal && !(sb->s_flags & MS_RDONLY)) {
3094                 lock_super(sb);
3095                 /* Reser the needs_recovery flag before the fs is unlocked. */
3096                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3097                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
3098                 unlock_super(sb);
3099                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3100         }
3101 }
3102
3103 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3104 {
3105         struct ext4_super_block *es;
3106         struct ext4_sb_info *sbi = EXT4_SB(sb);
3107         ext4_fsblk_t n_blocks_count = 0;
3108         unsigned long old_sb_flags;
3109         struct ext4_mount_options old_opts;
3110         ext4_group_t g;
3111         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3112         int err;
3113 #ifdef CONFIG_QUOTA
3114         int i;
3115 #endif
3116
3117         /* Store the original options */
3118         old_sb_flags = sb->s_flags;
3119         old_opts.s_mount_opt = sbi->s_mount_opt;
3120         old_opts.s_resuid = sbi->s_resuid;
3121         old_opts.s_resgid = sbi->s_resgid;
3122         old_opts.s_commit_interval = sbi->s_commit_interval;
3123         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3124         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3125 #ifdef CONFIG_QUOTA
3126         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3127         for (i = 0; i < MAXQUOTAS; i++)
3128                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3129 #endif
3130         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3131                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3132
3133         /*
3134          * Allow the "check" option to be passed as a remount option.
3135          */
3136         if (!parse_options(data, sb, NULL, &journal_ioprio,
3137                            &n_blocks_count, 1)) {
3138                 err = -EINVAL;
3139                 goto restore_opts;
3140         }
3141
3142         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
3143                 ext4_abort(sb, __func__, "Abort forced by user");
3144
3145         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3146                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3147
3148         es = sbi->s_es;
3149
3150         if (sbi->s_journal) {
3151                 ext4_init_journal_params(sb, sbi->s_journal);
3152                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3153         }
3154
3155         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3156                 n_blocks_count > ext4_blocks_count(es)) {
3157                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
3158                         err = -EROFS;
3159                         goto restore_opts;
3160                 }
3161
3162                 if (*flags & MS_RDONLY) {
3163                         /*
3164                          * First of all, the unconditional stuff we have to do
3165                          * to disable replay of the journal when we next remount
3166                          */
3167                         sb->s_flags |= MS_RDONLY;
3168
3169                         /*
3170                          * OK, test if we are remounting a valid rw partition
3171                          * readonly, and if so set the rdonly flag and then
3172                          * mark the partition as valid again.
3173                          */
3174                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3175                             (sbi->s_mount_state & EXT4_VALID_FS))
3176                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3177
3178                         /*
3179                          * We have to unlock super so that we can wait for
3180                          * transactions.
3181                          */
3182                         if (sbi->s_journal) {
3183                                 unlock_super(sb);
3184                                 ext4_mark_recovery_complete(sb, es);
3185                                 lock_super(sb);
3186                         }
3187                 } else {
3188                         int ret;
3189                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3190                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3191                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3192                                        "remount RDWR because of unsupported "
3193                                        "optional features (%x).\n", sb->s_id,
3194                                 (le32_to_cpu(sbi->s_es->s_feature_ro_compat) &
3195                                         ~EXT4_FEATURE_RO_COMPAT_SUPP));
3196                                 err = -EROFS;
3197                                 goto restore_opts;
3198                         }
3199
3200                         /*
3201                          * Make sure the group descriptor checksums
3202                          * are sane.  If they aren't, refuse to
3203                          * remount r/w.
3204                          */
3205                         for (g = 0; g < sbi->s_groups_count; g++) {
3206                                 struct ext4_group_desc *gdp =
3207                                         ext4_get_group_desc(sb, g, NULL);
3208
3209                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3210                                         printk(KERN_ERR
3211                "EXT4-fs: ext4_remount: "
3212                 "Checksum for group %u failed (%u!=%u)\n",
3213                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3214                                                le16_to_cpu(gdp->bg_checksum));
3215                                         err = -EINVAL;
3216                                         goto restore_opts;
3217                                 }
3218                         }
3219
3220                         /*
3221                          * If we have an unprocessed orphan list hanging
3222                          * around from a previously readonly bdev mount,
3223                          * require a full umount/remount for now.
3224                          */
3225                         if (es->s_last_orphan) {
3226                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3227                                        "remount RDWR because of unprocessed "
3228                                        "orphan inode list.  Please "
3229                                        "umount/remount instead.\n",
3230                                        sb->s_id);
3231                                 err = -EINVAL;
3232                                 goto restore_opts;
3233                         }
3234
3235                         /*
3236                          * Mounting a RDONLY partition read-write, so reread
3237                          * and store the current valid flag.  (It may have
3238                          * been changed by e2fsck since we originally mounted
3239                          * the partition.)
3240                          */
3241                         if (sbi->s_journal)
3242                                 ext4_clear_journal_err(sb, es);
3243                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3244                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3245                                 goto restore_opts;
3246                         if (!ext4_setup_super(sb, es, 0))
3247                                 sb->s_flags &= ~MS_RDONLY;
3248                 }
3249         }
3250         if (sbi->s_journal == NULL)
3251                 ext4_commit_super(sb, es, 1);
3252
3253 #ifdef CONFIG_QUOTA
3254         /* Release old quota file names */
3255         for (i = 0; i < MAXQUOTAS; i++)
3256                 if (old_opts.s_qf_names[i] &&
3257                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3258                         kfree(old_opts.s_qf_names[i]);
3259 #endif
3260         return 0;
3261 restore_opts:
3262         sb->s_flags = old_sb_flags;
3263         sbi->s_mount_opt = old_opts.s_mount_opt;
3264         sbi->s_resuid = old_opts.s_resuid;
3265         sbi->s_resgid = old_opts.s_resgid;
3266         sbi->s_commit_interval = old_opts.s_commit_interval;
3267         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3268         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3269 #ifdef CONFIG_QUOTA
3270         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3271         for (i = 0; i < MAXQUOTAS; i++) {
3272                 if (sbi->s_qf_names[i] &&
3273                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3274                         kfree(sbi->s_qf_names[i]);
3275                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3276         }
3277 #endif
3278         return err;
3279 }
3280
3281 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3282 {
3283         struct super_block *sb = dentry->d_sb;
3284         struct ext4_sb_info *sbi = EXT4_SB(sb);
3285         struct ext4_super_block *es = sbi->s_es;
3286         u64 fsid;
3287
3288         if (test_opt(sb, MINIX_DF)) {
3289                 sbi->s_overhead_last = 0;
3290         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3291                 ext4_group_t ngroups = sbi->s_groups_count, i;
3292                 ext4_fsblk_t overhead = 0;
3293                 smp_rmb();
3294
3295                 /*
3296                  * Compute the overhead (FS structures).  This is constant
3297                  * for a given filesystem unless the number of block groups
3298                  * changes so we cache the previous value until it does.
3299                  */
3300
3301                 /*
3302                  * All of the blocks before first_data_block are
3303                  * overhead
3304                  */
3305                 overhead = le32_to_cpu(es->s_first_data_block);
3306
3307                 /*
3308                  * Add the overhead attributed to the superblock and
3309                  * block group descriptors.  If the sparse superblocks
3310                  * feature is turned on, then not all groups have this.
3311                  */
3312                 for (i = 0; i < ngroups; i++) {
3313                         overhead += ext4_bg_has_super(sb, i) +
3314                                 ext4_bg_num_gdb(sb, i);
3315                         cond_resched();
3316                 }
3317
3318                 /*
3319                  * Every block group has an inode bitmap, a block
3320                  * bitmap, and an inode table.
3321                  */
3322                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3323                 sbi->s_overhead_last = overhead;
3324                 smp_wmb();
3325                 sbi->s_blocks_last = ext4_blocks_count(es);
3326         }
3327
3328         buf->f_type = EXT4_SUPER_MAGIC;
3329         buf->f_bsize = sb->s_blocksize;
3330         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3331         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3332                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3333         ext4_free_blocks_count_set(es, buf->f_bfree);
3334         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3335         if (buf->f_bfree < ext4_r_blocks_count(es))
3336                 buf->f_bavail = 0;
3337         buf->f_files = le32_to_cpu(es->s_inodes_count);
3338         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3339         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3340         buf->f_namelen = EXT4_NAME_LEN;
3341         fsid = le64_to_cpup((void *)es->s_uuid) ^
3342                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3343         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3344         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3345         return 0;
3346 }
3347
3348 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3349  * is locked for write. Otherwise the are possible deadlocks:
3350  * Process 1                         Process 2
3351  * ext4_create()                     quota_sync()
3352  *   jbd2_journal_start()                   write_dquot()
3353  *   DQUOT_INIT()                        down(dqio_mutex)
3354  *     down(dqio_mutex)                    jbd2_journal_start()
3355  *
3356  */
3357
3358 #ifdef CONFIG_QUOTA
3359
3360 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3361 {
3362         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3363 }
3364
3365 static int ext4_dquot_initialize(struct inode *inode, int type)
3366 {
3367         handle_t *handle;
3368         int ret, err;
3369
3370         /* We may create quota structure so we need to reserve enough blocks */
3371         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3372         if (IS_ERR(handle))
3373                 return PTR_ERR(handle);
3374         ret = dquot_initialize(inode, type);
3375         err = ext4_journal_stop(handle);
3376         if (!ret)
3377                 ret = err;
3378         return ret;
3379 }
3380
3381 static int ext4_dquot_drop(struct inode *inode)
3382 {
3383         handle_t *handle;
3384         int ret, err;
3385
3386         /* We may delete quota structure so we need to reserve enough blocks */
3387         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3388         if (IS_ERR(handle)) {
3389                 /*
3390                  * We call dquot_drop() anyway to at least release references
3391                  * to quota structures so that umount does not hang.
3392                  */
3393                 dquot_drop(inode);
3394                 return PTR_ERR(handle);
3395         }
3396         ret = dquot_drop(inode);
3397         err = ext4_journal_stop(handle);
3398         if (!ret)
3399                 ret = err;
3400         return ret;
3401 }
3402
3403 static int ext4_write_dquot(struct dquot *dquot)
3404 {
3405         int ret, err;
3406         handle_t *handle;
3407         struct inode *inode;
3408
3409         inode = dquot_to_inode(dquot);
3410         handle = ext4_journal_start(inode,
3411                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3412         if (IS_ERR(handle))
3413                 return PTR_ERR(handle);
3414         ret = dquot_commit(dquot);
3415         err = ext4_journal_stop(handle);
3416         if (!ret)
3417                 ret = err;
3418         return ret;
3419 }
3420
3421 static int ext4_acquire_dquot(struct dquot *dquot)
3422 {
3423         int ret, err;
3424         handle_t *handle;
3425
3426         handle = ext4_journal_start(dquot_to_inode(dquot),
3427                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3428         if (IS_ERR(handle))
3429                 return PTR_ERR(handle);
3430         ret = dquot_acquire(dquot);
3431         err = ext4_journal_stop(handle);
3432         if (!ret)
3433                 ret = err;
3434         return ret;
3435 }
3436
3437 static int ext4_release_dquot(struct dquot *dquot)
3438 {
3439         int ret, err;
3440         handle_t *handle;
3441
3442         handle = ext4_journal_start(dquot_to_inode(dquot),
3443                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3444         if (IS_ERR(handle)) {
3445                 /* Release dquot anyway to avoid endless cycle in dqput() */
3446                 dquot_release(dquot);
3447                 return PTR_ERR(handle);
3448         }
3449         ret = dquot_release(dquot);
3450         err = ext4_journal_stop(handle);
3451         if (!ret)
3452                 ret = err;
3453         return ret;
3454 }
3455
3456 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3457 {
3458         /* Are we journaling quotas? */
3459         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3460             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3461                 dquot_mark_dquot_dirty(dquot);
3462                 return ext4_write_dquot(dquot);
3463         } else {
3464                 return dquot_mark_dquot_dirty(dquot);
3465         }
3466 }
3467
3468 static int ext4_write_info(struct super_block *sb, int type)
3469 {
3470         int ret, err;
3471         handle_t *handle;
3472
3473         /* Data block + inode block */
3474         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3475         if (IS_ERR(handle))
3476                 return PTR_ERR(handle);
3477         ret = dquot_commit_info(sb, type);
3478         err = ext4_journal_stop(handle);
3479         if (!ret)
3480                 ret = err;
3481         return ret;
3482 }
3483
3484 /*
3485  * Turn on quotas during mount time - we need to find
3486  * the quota file and such...
3487  */
3488 static int ext4_quota_on_mount(struct super_block *sb, int type)
3489 {
3490         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3491                         EXT4_SB(sb)->s_jquota_fmt, type);
3492 }
3493
3494 /*
3495  * Standard function to be called on quota_on
3496  */
3497 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3498                          char *name, int remount)
3499 {
3500         int err;
3501         struct path path;
3502
3503         if (!test_opt(sb, QUOTA))
3504                 return -EINVAL;
3505         /* When remounting, no checks are needed and in fact, name is NULL */
3506         if (remount)
3507                 return vfs_quota_on(sb, type, format_id, name, remount);
3508
3509         err = kern_path(name, LOOKUP_FOLLOW, &path);
3510         if (err)
3511                 return err;
3512
3513         /* Quotafile not on the same filesystem? */
3514         if (path.mnt->mnt_sb != sb) {
3515                 path_put(&path);
3516                 return -EXDEV;
3517         }
3518         /* Journaling quota? */
3519         if (EXT4_SB(sb)->s_qf_names[type]) {
3520                 /* Quotafile not in fs root? */
3521                 if (path.dentry->d_parent != sb->s_root)
3522                         printk(KERN_WARNING
3523                                 "EXT4-fs: Quota file not on filesystem root. "
3524                                 "Journaled quota will not work.\n");
3525         }
3526
3527         /*
3528          * When we journal data on quota file, we have to flush journal to see
3529          * all updates to the file when we bypass pagecache...
3530          */
3531         if (EXT4_SB(sb)->s_journal &&
3532             ext4_should_journal_data(path.dentry->d_inode)) {
3533                 /*
3534                  * We don't need to lock updates but journal_flush() could
3535                  * otherwise be livelocked...
3536                  */
3537                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3538                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3539                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3540                 if (err) {
3541                         path_put(&path);
3542                         return err;
3543                 }
3544         }
3545
3546         err = vfs_quota_on_path(sb, type, format_id, &path);
3547         path_put(&path);
3548         return err;
3549 }
3550
3551 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3552  * acquiring the locks... As quota files are never truncated and quota code
3553  * itself serializes the operations (and noone else should touch the files)
3554  * we don't have to be afraid of races */
3555 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3556                                size_t len, loff_t off)
3557 {
3558         struct inode *inode = sb_dqopt(sb)->files[type];
3559         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3560         int err = 0;
3561         int offset = off & (sb->s_blocksize - 1);
3562         int tocopy;
3563         size_t toread;
3564         struct buffer_head *bh;
3565         loff_t i_size = i_size_read(inode);
3566
3567         if (off > i_size)
3568                 return 0;
3569         if (off+len > i_size)
3570                 len = i_size-off;
3571         toread = len;
3572         while (toread > 0) {
3573                 tocopy = sb->s_blocksize - offset < toread ?
3574                                 sb->s_blocksize - offset : toread;
3575                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3576                 if (err)
3577                         return err;
3578                 if (!bh)        /* A hole? */
3579                         memset(data, 0, tocopy);
3580                 else
3581                         memcpy(data, bh->b_data+offset, tocopy);
3582                 brelse(bh);
3583                 offset = 0;
3584                 toread -= tocopy;
3585                 data += tocopy;
3586                 blk++;
3587         }
3588         return len;
3589 }
3590
3591 /* Write to quotafile (we know the transaction is already started and has
3592  * enough credits) */
3593 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3594                                 const char *data, size_t len, loff_t off)
3595 {
3596         struct inode *inode = sb_dqopt(sb)->files[type];
3597         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3598         int err = 0;
3599         int offset = off & (sb->s_blocksize - 1);
3600         int tocopy;
3601         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3602         size_t towrite = len;
3603         struct buffer_head *bh;
3604         handle_t *handle = journal_current_handle();
3605
3606         if (EXT4_SB(sb)->s_journal && !handle) {
3607                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%llu, len=%llu)"
3608                         " cancelled because transaction is not started.\n",
3609                         (unsigned long long)off, (unsigned long long)len);
3610                 return -EIO;
3611         }
3612         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3613         while (towrite > 0) {
3614                 tocopy = sb->s_blocksize - offset < towrite ?
3615                                 sb->s_blocksize - offset : towrite;
3616                 bh = ext4_bread(handle, inode, blk, 1, &err);
3617                 if (!bh)
3618                         goto out;
3619                 if (journal_quota) {
3620                         err = ext4_journal_get_write_access(handle, bh);
3621                         if (err) {
3622                                 brelse(bh);
3623                                 goto out;
3624                         }
3625                 }
3626                 lock_buffer(bh);
3627                 memcpy(bh->b_data+offset, data, tocopy);
3628                 flush_dcache_page(bh->b_page);
3629                 unlock_buffer(bh);
3630                 if (journal_quota)
3631                         err = ext4_handle_dirty_metadata(handle, NULL, bh);
3632                 else {
3633                         /* Always do at least ordered writes for quotas */
3634                         err = ext4_jbd2_file_inode(handle, inode);
3635                         mark_buffer_dirty(bh);
3636                 }
3637                 brelse(bh);
3638                 if (err)
3639                         goto out;
3640                 offset = 0;
3641                 towrite -= tocopy;
3642                 data += tocopy;
3643                 blk++;
3644         }
3645 out:
3646         if (len == towrite) {
3647                 mutex_unlock(&inode->i_mutex);
3648                 return err;
3649         }
3650         if (inode->i_size < off+len-towrite) {
3651                 i_size_write(inode, off+len-towrite);
3652                 EXT4_I(inode)->i_disksize = inode->i_size;
3653         }
3654         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3655         ext4_mark_inode_dirty(handle, inode);
3656         mutex_unlock(&inode->i_mutex);
3657         return len - towrite;
3658 }
3659
3660 #endif
3661
3662 static int ext4_get_sb(struct file_system_type *fs_type,
3663         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3664 {
3665         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3666 }
3667
3668 #ifdef CONFIG_PROC_FS
3669 static int ext4_ui_proc_show(struct seq_file *m, void *v)
3670 {
3671         unsigned int *p = m->private;
3672
3673         seq_printf(m, "%u\n", *p);
3674         return 0;
3675 }
3676
3677 static int ext4_ui_proc_open(struct inode *inode, struct file *file)
3678 {
3679         return single_open(file, ext4_ui_proc_show, PDE(inode)->data);
3680 }
3681
3682 static ssize_t ext4_ui_proc_write(struct file *file, const char __user *buf,
3683                                size_t cnt, loff_t *ppos)
3684 {
3685         unsigned long *p = PDE(file->f_path.dentry->d_inode)->data;
3686         char str[32];
3687
3688         if (cnt >= sizeof(str))
3689                 return -EINVAL;
3690         if (copy_from_user(str, buf, cnt))
3691                 return -EFAULT;
3692
3693         *p = simple_strtoul(str, NULL, 0);
3694         return cnt;
3695 }
3696
3697 const struct file_operations ext4_ui_proc_fops = {
3698         .owner          = THIS_MODULE,
3699         .open           = ext4_ui_proc_open,
3700         .read           = seq_read,
3701         .llseek         = seq_lseek,
3702         .release        = single_release,
3703         .write          = ext4_ui_proc_write,
3704 };
3705 #endif
3706
3707 static struct file_system_type ext4_fs_type = {
3708         .owner          = THIS_MODULE,
3709         .name           = "ext4",
3710         .get_sb         = ext4_get_sb,
3711         .kill_sb        = kill_block_super,
3712         .fs_flags       = FS_REQUIRES_DEV,
3713 };
3714
3715 #ifdef CONFIG_EXT4DEV_COMPAT
3716 static int ext4dev_get_sb(struct file_system_type *fs_type,
3717         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3718 {
3719         printk(KERN_WARNING "EXT4-fs: Update your userspace programs "
3720                "to mount using ext4\n");
3721         printk(KERN_WARNING "EXT4-fs: ext4dev backwards compatibility "
3722                "will go away by 2.6.31\n");
3723         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3724 }
3725
3726 static struct file_system_type ext4dev_fs_type = {
3727         .owner          = THIS_MODULE,
3728         .name           = "ext4dev",
3729         .get_sb         = ext4dev_get_sb,
3730         .kill_sb        = kill_block_super,
3731         .fs_flags       = FS_REQUIRES_DEV,
3732 };
3733 MODULE_ALIAS("ext4dev");
3734 #endif
3735
3736 static int __init init_ext4_fs(void)
3737 {
3738         int err;
3739
3740         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
3741         err = init_ext4_mballoc();
3742         if (err)
3743                 return err;
3744
3745         err = init_ext4_xattr();
3746         if (err)
3747                 goto out2;
3748         err = init_inodecache();
3749         if (err)
3750                 goto out1;
3751         err = register_filesystem(&ext4_fs_type);
3752         if (err)
3753                 goto out;
3754 #ifdef CONFIG_EXT4DEV_COMPAT
3755         err = register_filesystem(&ext4dev_fs_type);
3756         if (err) {
3757                 unregister_filesystem(&ext4_fs_type);
3758                 goto out;
3759         }
3760 #endif
3761         return 0;
3762 out:
3763         destroy_inodecache();
3764 out1:
3765         exit_ext4_xattr();
3766 out2:
3767         exit_ext4_mballoc();
3768         return err;
3769 }
3770
3771 static void __exit exit_ext4_fs(void)
3772 {
3773         unregister_filesystem(&ext4_fs_type);
3774 #ifdef CONFIG_EXT4DEV_COMPAT
3775         unregister_filesystem(&ext4dev_fs_type);
3776 #endif
3777         destroy_inodecache();
3778         exit_ext4_xattr();
3779         exit_ext4_mballoc();
3780         remove_proc_entry("fs/ext4", NULL);
3781 }
3782
3783 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3784 MODULE_DESCRIPTION("Fourth Extended Filesystem");
3785 MODULE_LICENSE("GPL");
3786 module_init(init_ext4_fs)
3787 module_exit(exit_ext4_fs)