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/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/proc_fs.h>
  39 #include <linux/ctype.h>
  40 #include <linux/log2.h>
  41 #include <linux/crc16.h>
  42 #include <asm/uaccess.h>
  43
  44 #include "ext4.h"
  45 #include "ext4_jbd2.h"
  46 #include "xattr.h"
  47 #include "acl.h"
  48 #include "mballoc.h"
  49
  50 #define CREATE_TRACE_POINTS
  51 #include <trace/events/ext4.h>
  52
  53 struct proc_dir_entry *ext4_proc_root;
  54 static struct kset *ext4_kset;
  55
  56 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  57                              unsigned long journal_devnum);
  58 static int ext4_commit_super(struct super_block *sb, int sync);
  59 static void ext4_mark_recovery_complete(struct super_block *sb,
  60                                         struct ext4_super_block *es);
  61 static void ext4_clear_journal_err(struct super_block *sb,
  62                                    struct ext4_super_block *es);
  63 static int ext4_sync_fs(struct super_block *sb, int wait);
  64 static const char *ext4_decode_error(struct super_block *sb, int errno,
  65                                      char nbuf[16]);
  66 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  67 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  68 static int ext4_unfreeze(struct super_block *sb);
  69 static void ext4_write_super(struct super_block *sb);
  70 static int ext4_freeze(struct super_block *sb);
  71
  72
  73 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  74                                struct ext4_group_desc *bg)
  75 {
  76         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  77                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  78                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  79 }
  80
  81 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  82                                struct ext4_group_desc *bg)
  83 {
  84         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  85                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  86                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  87 }
  88
  89 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  90                               struct ext4_group_desc *bg)
  91 {
  92         return le32_to_cpu(bg->bg_inode_table_lo) |
  93                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  94                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  95 }
  96
  97 __u32 ext4_free_blks_count(struct super_block *sb,
  98                               struct ext4_group_desc *bg)
  99 {
 100         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 101                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 102                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 103 }
 104
 105 __u32 ext4_free_inodes_count(struct super_block *sb,
 106                               struct ext4_group_desc *bg)
 107 {
 108         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 109                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 110                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 111 }
 112
 113 __u32 ext4_used_dirs_count(struct super_block *sb,
 114                               struct ext4_group_desc *bg)
 115 {
 116         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 117                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 118                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 119 }
 120
 121 __u32 ext4_itable_unused_count(struct super_block *sb,
 122                               struct ext4_group_desc *bg)
 123 {
 124         return le16_to_cpu(bg->bg_itable_unused_lo) |
 125                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 126                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 127 }
 128
 129 void ext4_block_bitmap_set(struct super_block *sb,
 130                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 131 {
 132         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 133         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 134                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 135 }
 136
 137 void ext4_inode_bitmap_set(struct super_block *sb,
 138                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 139 {
 140         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 141         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 142                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 143 }
 144
 145 void ext4_inode_table_set(struct super_block *sb,
 146                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 147 {
 148         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 149         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 150                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 151 }
 152
 153 void ext4_free_blks_set(struct super_block *sb,
 154                           struct ext4_group_desc *bg, __u32 count)
 155 {
 156         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 157         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 158                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 159 }
 160
 161 void ext4_free_inodes_set(struct super_block *sb,
 162                           struct ext4_group_desc *bg, __u32 count)
 163 {
 164         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 165         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 166                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 167 }
 168
 169 void ext4_used_dirs_set(struct super_block *sb,
 170                           struct ext4_group_desc *bg, __u32 count)
 171 {
 172         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 173         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 174                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 175 }
 176
 177 void ext4_itable_unused_set(struct super_block *sb,
 178                           struct ext4_group_desc *bg, __u32 count)
 179 {
 180         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 181         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 182                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 183 }
 184
 185
 186 /* Just increment the non-pointer handle value */
 187 static handle_t *ext4_get_nojournal(void)
 188 {
 189         handle_t *handle = current->journal_info;
 190         unsigned long ref_cnt = (unsigned long)handle;
 191
 192         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 193
 194         ref_cnt++;
 195         handle = (handle_t *)ref_cnt;
 196
 197         current->journal_info = handle;
 198         return handle;
 199 }
 200
 201
 202 /* Decrement the non-pointer handle value */
 203 static void ext4_put_nojournal(handle_t *handle)
 204 {
 205         unsigned long ref_cnt = (unsigned long)handle;
 206
 207         BUG_ON(ref_cnt == 0);
 208
 209         ref_cnt--;
 210         handle = (handle_t *)ref_cnt;
 211
 212         current->journal_info = handle;
 213 }
 214
 215 /*
 216  * Wrappers for jbd2_journal_start/end.
 217  *
 218  * The only special thing we need to do here is to make sure that all
 219  * journal_end calls result in the superblock being marked dirty, so
 220  * that sync() will call the filesystem's write_super callback if
 221  * appropriate.
 222  */
 223 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 224 {
 225         journal_t *journal;
 226
 227         if (sb->s_flags & MS_RDONLY)
 228                 return ERR_PTR(-EROFS);
 229
 230         /* Special case here: if the journal has aborted behind our
 231          * backs (eg. EIO in the commit thread), then we still need to
 232          * take the FS itself readonly cleanly. */
 233         journal = EXT4_SB(sb)->s_journal;
 234         if (journal) {
 235                 if (is_journal_aborted(journal)) {
 236                         ext4_abort(sb, __func__, "Detected aborted journal");
 237                         return ERR_PTR(-EROFS);
 238                 }
 239                 return jbd2_journal_start(journal, nblocks);
 240         }
 241         return ext4_get_nojournal();
 242 }
 243
 244 /*
 245  * The only special thing we need to do here is to make sure that all
 246  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 247  * that sync() will call the filesystem's write_super callback if
 248  * appropriate.
 249  */
 250 int __ext4_journal_stop(const char *where, handle_t *handle)
 251 {
 252         struct super_block *sb;
 253         int err;
 254         int rc;
 255
 256         if (!ext4_handle_valid(handle)) {
 257                 ext4_put_nojournal(handle);
 258                 return 0;
 259         }
 260         sb = handle->h_transaction->t_journal->j_private;
 261         err = handle->h_err;
 262         rc = jbd2_journal_stop(handle);
 263
 264         if (!err)
 265                 err = rc;
 266         if (err)
 267                 __ext4_std_error(sb, where, err);
 268         return err;
 269 }
 270
 271 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 272                 struct buffer_head *bh, handle_t *handle, int err)
 273 {
 274         char nbuf[16];
 275         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 276
 277         BUG_ON(!ext4_handle_valid(handle));
 278
 279         if (bh)
 280                 BUFFER_TRACE(bh, "abort");
 281
 282         if (!handle->h_err)
 283                 handle->h_err = err;
 284
 285         if (is_handle_aborted(handle))
 286                 return;
 287
 288         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 289                caller, errstr, err_fn);
 290
 291         jbd2_journal_abort_handle(handle);
 292 }
 293
 294 /* Deal with the reporting of failure conditions on a filesystem such as
 295  * inconsistencies detected or read IO failures.
 296  *
 297  * On ext2, we can store the error state of the filesystem in the
 298  * superblock.  That is not possible on ext4, because we may have other
 299  * write ordering constraints on the superblock which prevent us from
 300  * writing it out straight away; and given that the journal is about to
 301  * be aborted, we can't rely on the current, or future, transactions to
 302  * write out the superblock safely.
 303  *
 304  * We'll just use the jbd2_journal_abort() error code to record an error in
 305  * the journal instead.  On recovery, the journal will compain about
 306  * that error until we've noted it down and cleared it.
 307  */
 308
 309 static void ext4_handle_error(struct super_block *sb)
 310 {
 311         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 312
 313         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 314         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 315
 316         if (sb->s_flags & MS_RDONLY)
 317                 return;
 318
 319         if (!test_opt(sb, ERRORS_CONT)) {
 320                 journal_t *journal = EXT4_SB(sb)->s_journal;
 321
 322                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 323                 if (journal)
 324                         jbd2_journal_abort(journal, -EIO);
 325         }
 326         if (test_opt(sb, ERRORS_RO)) {
 327                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 328                 sb->s_flags |= MS_RDONLY;
 329         }
 330         ext4_commit_super(sb, 1);
 331         if (test_opt(sb, ERRORS_PANIC))
 332                 panic("EXT4-fs (device %s): panic forced after error\n",
 333                         sb->s_id);
 334 }
 335
 336 void ext4_error(struct super_block *sb, const char *function,
 337                 const char *fmt, ...)
 338 {
 339         va_list args;
 340
 341         va_start(args, fmt);
 342         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 343         vprintk(fmt, args);
 344         printk("\n");
 345         va_end(args);
 346
 347         ext4_handle_error(sb);
 348 }
 349
 350 static const char *ext4_decode_error(struct super_block *sb, int errno,
 351                                      char nbuf[16])
 352 {
 353         char *errstr = NULL;
 354
 355         switch (errno) {
 356         case -EIO:
 357                 errstr = "IO failure";
 358                 break;
 359         case -ENOMEM:
 360                 errstr = "Out of memory";
 361                 break;
 362         case -EROFS:
 363                 if (!sb || (EXT4_SB(sb)->s_journal &&
 364                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 365                         errstr = "Journal has aborted";
 366                 else
 367                         errstr = "Readonly filesystem";
 368                 break;
 369         default:
 370                 /* If the caller passed in an extra buffer for unknown
 371                  * errors, textualise them now.  Else we just return
 372                  * NULL. */
 373                 if (nbuf) {
 374                         /* Check for truncated error codes... */
 375                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 376                                 errstr = nbuf;
 377                 }
 378                 break;
 379         }
 380
 381         return errstr;
 382 }
 383
 384 /* __ext4_std_error decodes expected errors from journaling functions
 385  * automatically and invokes the appropriate error response.  */
 386
 387 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 388 {
 389         char nbuf[16];
 390         const char *errstr;
 391
 392         /* Special case: if the error is EROFS, and we're not already
 393          * inside a transaction, then there's really no point in logging
 394          * an error. */
 395         if (errno == -EROFS && journal_current_handle() == NULL &&
 396             (sb->s_flags & MS_RDONLY))
 397                 return;
 398
 399         errstr = ext4_decode_error(sb, errno, nbuf);
 400         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 401                sb->s_id, function, errstr);
 402
 403         ext4_handle_error(sb);
 404 }
 405
 406 /*
 407  * ext4_abort is a much stronger failure handler than ext4_error.  The
 408  * abort function may be used to deal with unrecoverable failures such
 409  * as journal IO errors or ENOMEM at a critical moment in log management.
 410  *
 411  * We unconditionally force the filesystem into an ABORT|READONLY state,
 412  * unless the error response on the fs has been set to panic in which
 413  * case we take the easy way out and panic immediately.
 414  */
 415
 416 void ext4_abort(struct super_block *sb, const char *function,
 417                 const char *fmt, ...)
 418 {
 419         va_list args;
 420
 421         va_start(args, fmt);
 422         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 423         vprintk(fmt, args);
 424         printk("\n");
 425         va_end(args);
 426
 427         if (test_opt(sb, ERRORS_PANIC))
 428                 panic("EXT4-fs panic from previous error\n");
 429
 430         if (sb->s_flags & MS_RDONLY)
 431                 return;
 432
 433         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 434         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 435         sb->s_flags |= MS_RDONLY;
 436         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 437         if (EXT4_SB(sb)->s_journal)
 438                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 439 }
 440
 441 void ext4_msg (struct super_block * sb, const char *prefix,
 442                    const char *fmt, ...)
 443 {
 444         va_list args;
 445
 446         va_start(args, fmt);
 447         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 448         vprintk(fmt, args);
 449         printk("\n");
 450         va_end(args);
 451 }
 452
 453 void ext4_warning(struct super_block *sb, const char *function,
 454                   const char *fmt, ...)
 455 {
 456         va_list args;
 457
 458         va_start(args, fmt);
 459         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 460                sb->s_id, function);
 461         vprintk(fmt, args);
 462         printk("\n");
 463         va_end(args);
 464 }
 465
 466 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
 467                            const char *function, const char *fmt, ...)
 468 __releases(bitlock)
 469 __acquires(bitlock)
 470 {
 471         va_list args;
 472         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 473
 474         va_start(args, fmt);
 475         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 476         vprintk(fmt, args);
 477         printk("\n");
 478         va_end(args);
 479
 480         if (test_opt(sb, ERRORS_CONT)) {
 481                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 482                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 483                 ext4_commit_super(sb, 0);
 484                 return;
 485         }
 486         ext4_unlock_group(sb, grp);
 487         ext4_handle_error(sb);
 488         /*
 489          * We only get here in the ERRORS_RO case; relocking the group
 490          * may be dangerous, but nothing bad will happen since the
 491          * filesystem will have already been marked read/only and the
 492          * journal has been aborted.  We return 1 as a hint to callers
 493          * who might what to use the return value from
 494          * ext4_grp_locked_error() to distinguish beween the
 495          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 496          * aggressively from the ext4 function in question, with a
 497          * more appropriate error code.
 498          */
 499         ext4_lock_group(sb, grp);
 500         return;
 501 }
 502
 503 void ext4_update_dynamic_rev(struct super_block *sb)
 504 {
 505         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 506
 507         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 508                 return;
 509
 510         ext4_warning(sb, __func__,
 511                      "updating to rev %d because of new feature flag, "
 512                      "running e2fsck is recommended",
 513                      EXT4_DYNAMIC_REV);
 514
 515         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 516         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 517         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 518         /* leave es->s_feature_*compat flags alone */
 519         /* es->s_uuid will be set by e2fsck if empty */
 520
 521         /*
 522          * The rest of the superblock fields should be zero, and if not it
 523          * means they are likely already in use, so leave them alone.  We
 524          * can leave it up to e2fsck to clean up any inconsistencies there.
 525          */
 526 }
 527
 528 /*
 529  * Open the external journal device
 530  */
 531 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 532 {
 533         struct block_device *bdev;
 534         char b[BDEVNAME_SIZE];
 535
 536         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 537         if (IS_ERR(bdev))
 538                 goto fail;
 539         return bdev;
 540
 541 fail:
 542         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 543                         __bdevname(dev, b), PTR_ERR(bdev));
 544         return NULL;
 545 }
 546
 547 /*
 548  * Release the journal device
 549  */
 550 static int ext4_blkdev_put(struct block_device *bdev)
 551 {
 552         bd_release(bdev);
 553         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 554 }
 555
 556 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 557 {
 558         struct block_device *bdev;
 559         int ret = -ENODEV;
 560
 561         bdev = sbi->journal_bdev;
 562         if (bdev) {
 563                 ret = ext4_blkdev_put(bdev);
 564                 sbi->journal_bdev = NULL;
 565         }
 566         return ret;
 567 }
 568
 569 static inline struct inode *orphan_list_entry(struct list_head *l)
 570 {
 571         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 572 }
 573
 574 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 575 {
 576         struct list_head *l;
 577
 578         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 579                  le32_to_cpu(sbi->s_es->s_last_orphan));
 580
 581         printk(KERN_ERR "sb_info orphan list:\n");
 582         list_for_each(l, &sbi->s_orphan) {
 583                 struct inode *inode = orphan_list_entry(l);
 584                 printk(KERN_ERR "  "
 585                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 586                        inode->i_sb->s_id, inode->i_ino, inode,
 587                        inode->i_mode, inode->i_nlink,
 588                        NEXT_ORPHAN(inode));
 589         }
 590 }
 591
 592 static void ext4_put_super(struct super_block *sb)
 593 {
 594         struct ext4_sb_info *sbi = EXT4_SB(sb);
 595         struct ext4_super_block *es = sbi->s_es;
 596         int i, err;
 597
 598         flush_workqueue(sbi->dio_unwritten_wq);
 599         destroy_workqueue(sbi->dio_unwritten_wq);
 600
 601         lock_super(sb);
 602         lock_kernel();
 603         if (sb->s_dirt)
 604                 ext4_commit_super(sb, 1);
 605
 606         if (sbi->s_journal) {
 607                 err = jbd2_journal_destroy(sbi->s_journal);
 608                 sbi->s_journal = NULL;
 609                 if (err < 0)
 610                         ext4_abort(sb, __func__,
 611                                    "Couldn't clean up the journal");
 612         }
 613
 614         ext4_release_system_zone(sb);
 615         ext4_mb_release(sb);
 616         ext4_ext_release(sb);
 617         ext4_xattr_put_super(sb);
 618
 619         if (!(sb->s_flags & MS_RDONLY)) {
 620                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 621                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 622                 ext4_commit_super(sb, 1);
 623         }
 624         if (sbi->s_proc) {
 625                 remove_proc_entry(sb->s_id, ext4_proc_root);
 626         }
 627         kobject_del(&sbi->s_kobj);
 628
 629         for (i = 0; i < sbi->s_gdb_count; i++)
 630                 brelse(sbi->s_group_desc[i]);
 631         kfree(sbi->s_group_desc);
 632         if (is_vmalloc_addr(sbi->s_flex_groups))
 633                 vfree(sbi->s_flex_groups);
 634         else
 635                 kfree(sbi->s_flex_groups);
 636         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 637         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 638         percpu_counter_destroy(&sbi->s_dirs_counter);
 639         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 640         brelse(sbi->s_sbh);
 641 #ifdef CONFIG_QUOTA
 642         for (i = 0; i < MAXQUOTAS; i++)
 643                 kfree(sbi->s_qf_names[i]);
 644 #endif
 645
 646         /* Debugging code just in case the in-memory inode orphan list
 647          * isn't empty.  The on-disk one can be non-empty if we've
 648          * detected an error and taken the fs readonly, but the
 649          * in-memory list had better be clean by this point. */
 650         if (!list_empty(&sbi->s_orphan))
 651                 dump_orphan_list(sb, sbi);
 652         J_ASSERT(list_empty(&sbi->s_orphan));
 653
 654         invalidate_bdev(sb->s_bdev);
 655         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 656                 /*
 657                  * Invalidate the journal device's buffers.  We don't want them
 658                  * floating about in memory - the physical journal device may
 659                  * hotswapped, and it breaks the `ro-after' testing code.
 660                  */
 661                 sync_blockdev(sbi->journal_bdev);
 662                 invalidate_bdev(sbi->journal_bdev);
 663                 ext4_blkdev_remove(sbi);
 664         }
 665         sb->s_fs_info = NULL;
 666         /*
 667          * Now that we are completely done shutting down the
 668          * superblock, we need to actually destroy the kobject.
 669          */
 670         unlock_kernel();
 671         unlock_super(sb);
 672         kobject_put(&sbi->s_kobj);
 673         wait_for_completion(&sbi->s_kobj_unregister);
 674         kfree(sbi->s_blockgroup_lock);
 675         kfree(sbi);
 676 }
 677
 678 static struct kmem_cache *ext4_inode_cachep;
 679
 680 /*
 681  * Called inside transaction, so use GFP_NOFS
 682  */
 683 static struct inode *ext4_alloc_inode(struct super_block *sb)
 684 {
 685         struct ext4_inode_info *ei;
 686
 687         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 688         if (!ei)
 689                 return NULL;
 690
 691         ei->vfs_inode.i_version = 1;
 692         ei->vfs_inode.i_data.writeback_index = 0;
 693         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 694         INIT_LIST_HEAD(&ei->i_prealloc_list);
 695         spin_lock_init(&ei->i_prealloc_lock);
 696         /*
 697          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 698          * therefore it can be null here.  Don't check it, just initialize
 699          * jinode.
 700          */
 701         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 702         ei->i_reserved_data_blocks = 0;
 703         ei->i_reserved_meta_blocks = 0;
 704         ei->i_allocated_meta_blocks = 0;
 705         ei->i_da_metadata_calc_len = 0;
 706         ei->i_delalloc_reserved_flag = 0;
 707         spin_lock_init(&(ei->i_block_reservation_lock));
 708 #ifdef CONFIG_QUOTA
 709         ei->i_reserved_quota = 0;
 710 #endif
 711         INIT_LIST_HEAD(&ei->i_aio_dio_complete_list);
 712         ei->cur_aio_dio = NULL;
 713         ei->i_sync_tid = 0;
 714         ei->i_datasync_tid = 0;
 715
 716         return &ei->vfs_inode;
 717 }
 718
 719 static void ext4_destroy_inode(struct inode *inode)
 720 {
 721         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 722                 ext4_msg(inode->i_sb, KERN_ERR,
 723                          "Inode %lu (%p): orphan list check failed!",
 724                          inode->i_ino, EXT4_I(inode));
 725                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 726                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 727                                 true);
 728                 dump_stack();
 729         }
 730         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 731 }
 732
 733 static void init_once(void *foo)
 734 {
 735         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 736
 737         INIT_LIST_HEAD(&ei->i_orphan);
 738 #ifdef CONFIG_EXT4_FS_XATTR
 739         init_rwsem(&ei->xattr_sem);
 740 #endif
 741         init_rwsem(&ei->i_data_sem);
 742         inode_init_once(&ei->vfs_inode);
 743 }
 744
 745 static int init_inodecache(void)
 746 {
 747         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 748                                              sizeof(struct ext4_inode_info),
 749                                              0, (SLAB_RECLAIM_ACCOUNT|
 750                                                 SLAB_MEM_SPREAD),
 751                                              init_once);
 752         if (ext4_inode_cachep == NULL)
 753                 return -ENOMEM;
 754         return 0;
 755 }
 756
 757 static void destroy_inodecache(void)
 758 {
 759         kmem_cache_destroy(ext4_inode_cachep);
 760 }
 761
 762 static void ext4_clear_inode(struct inode *inode)
 763 {
 764         ext4_discard_preallocations(inode);
 765         if (EXT4_JOURNAL(inode))
 766                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 767                                        &EXT4_I(inode)->jinode);
 768 }
 769
 770 static inline void ext4_show_quota_options(struct seq_file *seq,
 771                                            struct super_block *sb)
 772 {
 773 #if defined(CONFIG_QUOTA)
 774         struct ext4_sb_info *sbi = EXT4_SB(sb);
 775
 776         if (sbi->s_jquota_fmt) {
 777                 char *fmtname = "";
 778
 779                 switch (sbi->s_jquota_fmt) {
 780                 case QFMT_VFS_OLD:
 781                         fmtname = "vfsold";
 782                         break;
 783                 case QFMT_VFS_V0:
 784                         fmtname = "vfsv0";
 785                         break;
 786                 case QFMT_VFS_V1:
 787                         fmtname = "vfsv1";
 788                         break;
 789                 }
 790                 seq_printf(seq, ",jqfmt=%s", fmtname);
 791         }
 792
 793         if (sbi->s_qf_names[USRQUOTA])
 794                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 795
 796         if (sbi->s_qf_names[GRPQUOTA])
 797                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 798
 799         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 800                 seq_puts(seq, ",usrquota");
 801
 802         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 803                 seq_puts(seq, ",grpquota");
 804 #endif
 805 }
 806
 807 /*
 808  * Show an option if
 809  *  - it's set to a non-default value OR
 810  *  - if the per-sb default is different from the global default
 811  */
 812 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 813 {
 814         int def_errors;
 815         unsigned long def_mount_opts;
 816         struct super_block *sb = vfs->mnt_sb;
 817         struct ext4_sb_info *sbi = EXT4_SB(sb);
 818         struct ext4_super_block *es = sbi->s_es;
 819
 820         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 821         def_errors     = le16_to_cpu(es->s_errors);
 822
 823         if (sbi->s_sb_block != 1)
 824                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 825         if (test_opt(sb, MINIX_DF))
 826                 seq_puts(seq, ",minixdf");
 827         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 828                 seq_puts(seq, ",grpid");
 829         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 830                 seq_puts(seq, ",nogrpid");
 831         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 832             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 833                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 834         }
 835         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 836             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 837                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 838         }
 839         if (test_opt(sb, ERRORS_RO)) {
 840                 if (def_errors == EXT4_ERRORS_PANIC ||
 841                     def_errors == EXT4_ERRORS_CONTINUE) {
 842                         seq_puts(seq, ",errors=remount-ro");
 843                 }
 844         }
 845         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 846                 seq_puts(seq, ",errors=continue");
 847         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 848                 seq_puts(seq, ",errors=panic");
 849         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 850                 seq_puts(seq, ",nouid32");
 851         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 852                 seq_puts(seq, ",debug");
 853         if (test_opt(sb, OLDALLOC))
 854                 seq_puts(seq, ",oldalloc");
 855 #ifdef CONFIG_EXT4_FS_XATTR
 856         if (test_opt(sb, XATTR_USER) &&
 857                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 858                 seq_puts(seq, ",user_xattr");
 859         if (!test_opt(sb, XATTR_USER) &&
 860             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 861                 seq_puts(seq, ",nouser_xattr");
 862         }
 863 #endif
 864 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 865         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 866                 seq_puts(seq, ",acl");
 867         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 868                 seq_puts(seq, ",noacl");
 869 #endif
 870         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 871                 seq_printf(seq, ",commit=%u",
 872                            (unsigned) (sbi->s_commit_interval / HZ));
 873         }
 874         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 875                 seq_printf(seq, ",min_batch_time=%u",
 876                            (unsigned) sbi->s_min_batch_time);
 877         }
 878         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 879                 seq_printf(seq, ",max_batch_time=%u",
 880                            (unsigned) sbi->s_min_batch_time);
 881         }
 882
 883         /*
 884          * We're changing the default of barrier mount option, so
 885          * let's always display its mount state so it's clear what its
 886          * status is.
 887          */
 888         seq_puts(seq, ",barrier=");
 889         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 890         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 891                 seq_puts(seq, ",journal_async_commit");
 892         if (test_opt(sb, NOBH))
 893                 seq_puts(seq, ",nobh");
 894         if (test_opt(sb, I_VERSION))
 895                 seq_puts(seq, ",i_version");
 896         if (!test_opt(sb, DELALLOC))
 897                 seq_puts(seq, ",nodelalloc");
 898
 899
 900         if (sbi->s_stripe)
 901                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 902         /*
 903          * journal mode get enabled in different ways
 904          * So just print the value even if we didn't specify it
 905          */
 906         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 907                 seq_puts(seq, ",data=journal");
 908         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 909                 seq_puts(seq, ",data=ordered");
 910         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 911                 seq_puts(seq, ",data=writeback");
 912
 913         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 914                 seq_printf(seq, ",inode_readahead_blks=%u",
 915                            sbi->s_inode_readahead_blks);
 916
 917         if (test_opt(sb, DATA_ERR_ABORT))
 918                 seq_puts(seq, ",data_err=abort");
 919
 920         if (test_opt(sb, NO_AUTO_DA_ALLOC))
 921                 seq_puts(seq, ",noauto_da_alloc");
 922
 923         if (test_opt(sb, DISCARD))
 924                 seq_puts(seq, ",discard");
 925
 926         if (test_opt(sb, NOLOAD))
 927                 seq_puts(seq, ",norecovery");
 928
 929         ext4_show_quota_options(seq, sb);
 930
 931         return 0;
 932 }
 933
 934 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 935                                         u64 ino, u32 generation)
 936 {
 937         struct inode *inode;
 938
 939         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 940                 return ERR_PTR(-ESTALE);
 941         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 942                 return ERR_PTR(-ESTALE);
 943
 944         /* iget isn't really right if the inode is currently unallocated!!
 945          *
 946          * ext4_read_inode will return a bad_inode if the inode had been
 947          * deleted, so we should be safe.
 948          *
 949          * Currently we don't know the generation for parent directory, so
 950          * a generation of 0 means "accept any"
 951          */
 952         inode = ext4_iget(sb, ino);
 953         if (IS_ERR(inode))
 954                 return ERR_CAST(inode);
 955         if (generation && inode->i_generation != generation) {
 956                 iput(inode);
 957                 return ERR_PTR(-ESTALE);
 958         }
 959
 960         return inode;
 961 }
 962
 963 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 964                                         int fh_len, int fh_type)
 965 {
 966         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 967                                     ext4_nfs_get_inode);
 968 }
 969
 970 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 971                                         int fh_len, int fh_type)
 972 {
 973         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 974                                     ext4_nfs_get_inode);
 975 }
 976
 977 /*
 978  * Try to release metadata pages (indirect blocks, directories) which are
 979  * mapped via the block device.  Since these pages could have journal heads
 980  * which would prevent try_to_free_buffers() from freeing them, we must use
 981  * jbd2 layer's try_to_free_buffers() function to release them.
 982  */
 983 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 984                                  gfp_t wait)
 985 {
 986         journal_t *journal = EXT4_SB(sb)->s_journal;
 987
 988         WARN_ON(PageChecked(page));
 989         if (!page_has_buffers(page))
 990                 return 0;
 991         if (journal)
 992                 return jbd2_journal_try_to_free_buffers(journal, page,
 993                                                         wait & ~__GFP_WAIT);
 994         return try_to_free_buffers(page);
 995 }
 996
 997 #ifdef CONFIG_QUOTA
 998 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
 999 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1000
1001 static int ext4_write_dquot(struct dquot *dquot);
1002 static int ext4_acquire_dquot(struct dquot *dquot);
1003 static int ext4_release_dquot(struct dquot *dquot);
1004 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1005 static int ext4_write_info(struct super_block *sb, int type);
1006 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1007                                 char *path, int remount);
1008 static int ext4_quota_on_mount(struct super_block *sb, int type);
1009 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1010                                size_t len, loff_t off);
1011 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1012                                 const char *data, size_t len, loff_t off);
1013
1014 static const struct dquot_operations ext4_quota_operations = {
1015         .initialize     = dquot_initialize,
1016         .drop           = dquot_drop,
1017         .alloc_space    = dquot_alloc_space,
1018         .reserve_space  = dquot_reserve_space,
1019         .claim_space    = dquot_claim_space,
1020         .release_rsv    = dquot_release_reserved_space,
1021 #ifdef CONFIG_QUOTA
1022         .get_reserved_space = ext4_get_reserved_space,
1023 #endif
1024         .alloc_inode    = dquot_alloc_inode,
1025         .free_space     = dquot_free_space,
1026         .free_inode     = dquot_free_inode,
1027         .transfer       = dquot_transfer,
1028         .write_dquot    = ext4_write_dquot,
1029         .acquire_dquot  = ext4_acquire_dquot,
1030         .release_dquot  = ext4_release_dquot,
1031         .mark_dirty     = ext4_mark_dquot_dirty,
1032         .write_info     = ext4_write_info,
1033         .alloc_dquot    = dquot_alloc,
1034         .destroy_dquot  = dquot_destroy,
1035 };
1036
1037 static const struct quotactl_ops ext4_qctl_operations = {
1038         .quota_on       = ext4_quota_on,
1039         .quota_off      = vfs_quota_off,
1040         .quota_sync     = vfs_quota_sync,
1041         .get_info       = vfs_get_dqinfo,
1042         .set_info       = vfs_set_dqinfo,
1043         .get_dqblk      = vfs_get_dqblk,
1044         .set_dqblk      = vfs_set_dqblk
1045 };
1046 #endif
1047
1048 static const struct super_operations ext4_sops = {
1049         .alloc_inode    = ext4_alloc_inode,
1050         .destroy_inode  = ext4_destroy_inode,
1051         .write_inode    = ext4_write_inode,
1052         .dirty_inode    = ext4_dirty_inode,
1053         .delete_inode   = ext4_delete_inode,
1054         .put_super      = ext4_put_super,
1055         .sync_fs        = ext4_sync_fs,
1056         .freeze_fs      = ext4_freeze,
1057         .unfreeze_fs    = ext4_unfreeze,
1058         .statfs         = ext4_statfs,
1059         .remount_fs     = ext4_remount,
1060         .clear_inode    = ext4_clear_inode,
1061         .show_options   = ext4_show_options,
1062 #ifdef CONFIG_QUOTA
1063         .quota_read     = ext4_quota_read,
1064         .quota_write    = ext4_quota_write,
1065 #endif
1066         .bdev_try_to_free_page = bdev_try_to_free_page,
1067 };
1068
1069 static const struct super_operations ext4_nojournal_sops = {
1070         .alloc_inode    = ext4_alloc_inode,
1071         .destroy_inode  = ext4_destroy_inode,
1072         .write_inode    = ext4_write_inode,
1073         .dirty_inode    = ext4_dirty_inode,
1074         .delete_inode   = ext4_delete_inode,
1075         .write_super    = ext4_write_super,
1076         .put_super      = ext4_put_super,
1077         .statfs         = ext4_statfs,
1078         .remount_fs     = ext4_remount,
1079         .clear_inode    = ext4_clear_inode,
1080         .show_options   = ext4_show_options,
1081 #ifdef CONFIG_QUOTA
1082         .quota_read     = ext4_quota_read,
1083         .quota_write    = ext4_quota_write,
1084 #endif
1085         .bdev_try_to_free_page = bdev_try_to_free_page,
1086 };
1087
1088 static const struct export_operations ext4_export_ops = {
1089         .fh_to_dentry = ext4_fh_to_dentry,
1090         .fh_to_parent = ext4_fh_to_parent,
1091         .get_parent = ext4_get_parent,
1092 };
1093
1094 enum {
1095         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1096         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1097         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1098         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1099         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1100         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1101         Opt_journal_update, Opt_journal_dev,
1102         Opt_journal_checksum, Opt_journal_async_commit,
1103         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1104         Opt_data_err_abort, Opt_data_err_ignore,
1105         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1106         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1107         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1108         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1109         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1110         Opt_block_validity, Opt_noblock_validity,
1111         Opt_inode_readahead_blks, Opt_journal_ioprio,
1112         Opt_discard, Opt_nodiscard,
1113 };
1114
1115 static const match_table_t tokens = {
1116         {Opt_bsd_df, "bsddf"},
1117         {Opt_minix_df, "minixdf"},
1118         {Opt_grpid, "grpid"},
1119         {Opt_grpid, "bsdgroups"},
1120         {Opt_nogrpid, "nogrpid"},
1121         {Opt_nogrpid, "sysvgroups"},
1122         {Opt_resgid, "resgid=%u"},
1123         {Opt_resuid, "resuid=%u"},
1124         {Opt_sb, "sb=%u"},
1125         {Opt_err_cont, "errors=continue"},
1126         {Opt_err_panic, "errors=panic"},
1127         {Opt_err_ro, "errors=remount-ro"},
1128         {Opt_nouid32, "nouid32"},
1129         {Opt_debug, "debug"},
1130         {Opt_oldalloc, "oldalloc"},
1131         {Opt_orlov, "orlov"},
1132         {Opt_user_xattr, "user_xattr"},
1133         {Opt_nouser_xattr, "nouser_xattr"},
1134         {Opt_acl, "acl"},
1135         {Opt_noacl, "noacl"},
1136         {Opt_noload, "noload"},
1137         {Opt_noload, "norecovery"},
1138         {Opt_nobh, "nobh"},
1139         {Opt_bh, "bh"},
1140         {Opt_commit, "commit=%u"},
1141         {Opt_min_batch_time, "min_batch_time=%u"},
1142         {Opt_max_batch_time, "max_batch_time=%u"},
1143         {Opt_journal_update, "journal=update"},
1144         {Opt_journal_dev, "journal_dev=%u"},
1145         {Opt_journal_checksum, "journal_checksum"},
1146         {Opt_journal_async_commit, "journal_async_commit"},
1147         {Opt_abort, "abort"},
1148         {Opt_data_journal, "data=journal"},
1149         {Opt_data_ordered, "data=ordered"},
1150         {Opt_data_writeback, "data=writeback"},
1151         {Opt_data_err_abort, "data_err=abort"},
1152         {Opt_data_err_ignore, "data_err=ignore"},
1153         {Opt_offusrjquota, "usrjquota="},
1154         {Opt_usrjquota, "usrjquota=%s"},
1155         {Opt_offgrpjquota, "grpjquota="},
1156         {Opt_grpjquota, "grpjquota=%s"},
1157         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1158         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1159         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1160         {Opt_grpquota, "grpquota"},
1161         {Opt_noquota, "noquota"},
1162         {Opt_quota, "quota"},
1163         {Opt_usrquota, "usrquota"},
1164         {Opt_barrier, "barrier=%u"},
1165         {Opt_barrier, "barrier"},
1166         {Opt_nobarrier, "nobarrier"},
1167         {Opt_i_version, "i_version"},
1168         {Opt_stripe, "stripe=%u"},
1169         {Opt_resize, "resize"},
1170         {Opt_delalloc, "delalloc"},
1171         {Opt_nodelalloc, "nodelalloc"},
1172         {Opt_block_validity, "block_validity"},
1173         {Opt_noblock_validity, "noblock_validity"},
1174         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1175         {Opt_journal_ioprio, "journal_ioprio=%u"},
1176         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1177         {Opt_auto_da_alloc, "auto_da_alloc"},
1178         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1179         {Opt_discard, "discard"},
1180         {Opt_nodiscard, "nodiscard"},
1181         {Opt_err, NULL},
1182 };
1183
1184 static ext4_fsblk_t get_sb_block(void **data)
1185 {
1186         ext4_fsblk_t    sb_block;
1187         char            *options = (char *) *data;
1188
1189         if (!options || strncmp(options, "sb=", 3) != 0)
1190                 return 1;       /* Default location */
1191
1192         options += 3;
1193         /* TODO: use simple_strtoll with >32bit ext4 */
1194         sb_block = simple_strtoul(options, &options, 0);
1195         if (*options && *options != ',') {
1196                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1197                        (char *) *data);
1198                 return 1;
1199         }
1200         if (*options == ',')
1201                 options++;
1202         *data = (void *) options;
1203
1204         return sb_block;
1205 }
1206
1207 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1208
1209 static int parse_options(char *options, struct super_block *sb,
1210                          unsigned long *journal_devnum,
1211                          unsigned int *journal_ioprio,
1212                          ext4_fsblk_t *n_blocks_count, int is_remount)
1213 {
1214         struct ext4_sb_info *sbi = EXT4_SB(sb);
1215         char *p;
1216         substring_t args[MAX_OPT_ARGS];
1217         int data_opt = 0;
1218         int option;
1219 #ifdef CONFIG_QUOTA
1220         int qtype, qfmt;
1221         char *qname;
1222 #endif
1223
1224         if (!options)
1225                 return 1;
1226
1227         while ((p = strsep(&options, ",")) != NULL) {
1228                 int token;
1229                 if (!*p)
1230                         continue;
1231
1232                 /*
1233                  * Initialize args struct so we know whether arg was
1234                  * found; some options take optional arguments.
1235                  */
1236                 args[0].to = args[0].from = 0;
1237                 token = match_token(p, tokens, args);
1238                 switch (token) {
1239                 case Opt_bsd_df:
1240                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1241                         break;
1242                 case Opt_minix_df:
1243                         set_opt(sbi->s_mount_opt, MINIX_DF);
1244                         break;
1245                 case Opt_grpid:
1246                         set_opt(sbi->s_mount_opt, GRPID);
1247                         break;
1248                 case Opt_nogrpid:
1249                         clear_opt(sbi->s_mount_opt, GRPID);
1250                         break;
1251                 case Opt_resuid:
1252                         if (match_int(&args[0], &option))
1253                                 return 0;
1254                         sbi->s_resuid = option;
1255                         break;
1256                 case Opt_resgid:
1257                         if (match_int(&args[0], &option))
1258                                 return 0;
1259                         sbi->s_resgid = option;
1260                         break;
1261                 case Opt_sb:
1262                         /* handled by get_sb_block() instead of here */
1263                         /* *sb_block = match_int(&args[0]); */
1264                         break;
1265                 case Opt_err_panic:
1266                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1267                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1268                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1269                         break;
1270                 case Opt_err_ro:
1271                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1272                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1273                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1274                         break;
1275                 case Opt_err_cont:
1276                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1277                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1278                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1279                         break;
1280                 case Opt_nouid32:
1281                         set_opt(sbi->s_mount_opt, NO_UID32);
1282                         break;
1283                 case Opt_debug:
1284                         set_opt(sbi->s_mount_opt, DEBUG);
1285                         break;
1286                 case Opt_oldalloc:
1287                         set_opt(sbi->s_mount_opt, OLDALLOC);
1288                         break;
1289                 case Opt_orlov:
1290                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1291                         break;
1292 #ifdef CONFIG_EXT4_FS_XATTR
1293                 case Opt_user_xattr:
1294                         set_opt(sbi->s_mount_opt, XATTR_USER);
1295                         break;
1296                 case Opt_nouser_xattr:
1297                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1298                         break;
1299 #else
1300                 case Opt_user_xattr:
1301                 case Opt_nouser_xattr:
1302                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1303                         break;
1304 #endif
1305 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1306                 case Opt_acl:
1307                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1308                         break;
1309                 case Opt_noacl:
1310                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1311                         break;
1312 #else
1313                 case Opt_acl:
1314                 case Opt_noacl:
1315                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1316                         break;
1317 #endif
1318                 case Opt_journal_update:
1319                         /* @@@ FIXME */
1320                         /* Eventually we will want to be able to create
1321                            a journal file here.  For now, only allow the
1322                            user to specify an existing inode to be the
1323                            journal file. */
1324                         if (is_remount) {
1325                                 ext4_msg(sb, KERN_ERR,
1326                                          "Cannot specify journal on remount");
1327                                 return 0;
1328                         }
1329                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1330                         break;
1331                 case Opt_journal_dev:
1332                         if (is_remount) {
1333                                 ext4_msg(sb, KERN_ERR,
1334                                         "Cannot specify journal on remount");
1335                                 return 0;
1336                         }
1337                         if (match_int(&args[0], &option))
1338                                 return 0;
1339                         *journal_devnum = option;
1340                         break;
1341                 case Opt_journal_checksum:
1342                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1343                         break;
1344                 case Opt_journal_async_commit:
1345                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1346                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1347                         break;
1348                 case Opt_noload:
1349                         set_opt(sbi->s_mount_opt, NOLOAD);
1350                         break;
1351                 case Opt_commit:
1352                         if (match_int(&args[0], &option))
1353                                 return 0;
1354                         if (option < 0)
1355                                 return 0;
1356                         if (option == 0)
1357                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1358                         sbi->s_commit_interval = HZ * option;
1359                         break;
1360                 case Opt_max_batch_time:
1361                         if (match_int(&args[0], &option))
1362                                 return 0;
1363                         if (option < 0)
1364                                 return 0;
1365                         if (option == 0)
1366                                 option = EXT4_DEF_MAX_BATCH_TIME;
1367                         sbi->s_max_batch_time = option;
1368                         break;
1369                 case Opt_min_batch_time:
1370                         if (match_int(&args[0], &option))
1371                                 return 0;
1372                         if (option < 0)
1373                                 return 0;
1374                         sbi->s_min_batch_time = option;
1375                         break;
1376                 case Opt_data_journal:
1377                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1378                         goto datacheck;
1379                 case Opt_data_ordered:
1380                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1381                         goto datacheck;
1382                 case Opt_data_writeback:
1383                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1384                 datacheck:
1385                         if (is_remount) {
1386                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1387                                                 != data_opt) {
1388                                         ext4_msg(sb, KERN_ERR,
1389                                                 "Cannot change data mode on remount");
1390                                         return 0;
1391                                 }
1392                         } else {
1393                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1394                                 sbi->s_mount_opt |= data_opt;
1395                         }
1396                         break;
1397                 case Opt_data_err_abort:
1398                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1399                         break;
1400                 case Opt_data_err_ignore:
1401                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1402                         break;
1403 #ifdef CONFIG_QUOTA
1404                 case Opt_usrjquota:
1405                         qtype = USRQUOTA;
1406                         goto set_qf_name;
1407                 case Opt_grpjquota:
1408                         qtype = GRPQUOTA;
1409 set_qf_name:
1410                         if (sb_any_quota_loaded(sb) &&
1411                             !sbi->s_qf_names[qtype]) {
1412                                 ext4_msg(sb, KERN_ERR,
1413                                        "Cannot change journaled "
1414                                        "quota options when quota turned on");
1415                                 return 0;
1416                         }
1417                         qname = match_strdup(&args[0]);
1418                         if (!qname) {
1419                                 ext4_msg(sb, KERN_ERR,
1420                                         "Not enough memory for "
1421                                         "storing quotafile name");
1422                                 return 0;
1423                         }
1424                         if (sbi->s_qf_names[qtype] &&
1425                             strcmp(sbi->s_qf_names[qtype], qname)) {
1426                                 ext4_msg(sb, KERN_ERR,
1427                                         "%s quota file already "
1428                                         "specified", QTYPE2NAME(qtype));
1429                                 kfree(qname);
1430                                 return 0;
1431                         }
1432                         sbi->s_qf_names[qtype] = qname;
1433                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1434                                 ext4_msg(sb, KERN_ERR,
1435                                         "quotafile must be on "
1436                                         "filesystem root");
1437                                 kfree(sbi->s_qf_names[qtype]);
1438                                 sbi->s_qf_names[qtype] = NULL;
1439                                 return 0;
1440                         }
1441                         set_opt(sbi->s_mount_opt, QUOTA);
1442                         break;
1443                 case Opt_offusrjquota:
1444                         qtype = USRQUOTA;
1445                         goto clear_qf_name;
1446                 case Opt_offgrpjquota:
1447                         qtype = GRPQUOTA;
1448 clear_qf_name:
1449                         if (sb_any_quota_loaded(sb) &&
1450                             sbi->s_qf_names[qtype]) {
1451                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1452                                         "journaled quota options when "
1453                                         "quota turned on");
1454                                 return 0;
1455                         }
1456                         /*
1457                          * The space will be released later when all options
1458                          * are confirmed to be correct
1459                          */
1460                         sbi->s_qf_names[qtype] = NULL;
1461                         break;
1462                 case Opt_jqfmt_vfsold:
1463                         qfmt = QFMT_VFS_OLD;
1464                         goto set_qf_format;
1465                 case Opt_jqfmt_vfsv0:
1466                         qfmt = QFMT_VFS_V0;
1467                         goto set_qf_format;
1468                 case Opt_jqfmt_vfsv1:
1469                         qfmt = QFMT_VFS_V1;
1470 set_qf_format:
1471                         if (sb_any_quota_loaded(sb) &&
1472                             sbi->s_jquota_fmt != qfmt) {
1473                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1474                                         "journaled quota options when "
1475                                         "quota turned on");
1476                                 return 0;
1477                         }
1478                         sbi->s_jquota_fmt = qfmt;
1479                         break;
1480                 case Opt_quota:
1481                 case Opt_usrquota:
1482                         set_opt(sbi->s_mount_opt, QUOTA);
1483                         set_opt(sbi->s_mount_opt, USRQUOTA);
1484                         break;
1485                 case Opt_grpquota:
1486                         set_opt(sbi->s_mount_opt, QUOTA);
1487                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1488                         break;
1489                 case Opt_noquota:
1490                         if (sb_any_quota_loaded(sb)) {
1491                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1492                                         "options when quota turned on");
1493                                 return 0;
1494                         }
1495                         clear_opt(sbi->s_mount_opt, QUOTA);
1496                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1497                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1498                         break;
1499 #else
1500                 case Opt_quota:
1501                 case Opt_usrquota:
1502                 case Opt_grpquota:
1503                         ext4_msg(sb, KERN_ERR,
1504                                 "quota options not supported");
1505                         break;
1506                 case Opt_usrjquota:
1507                 case Opt_grpjquota:
1508                 case Opt_offusrjquota:
1509                 case Opt_offgrpjquota:
1510                 case Opt_jqfmt_vfsold:
1511                 case Opt_jqfmt_vfsv0:
1512                 case Opt_jqfmt_vfsv1:
1513                         ext4_msg(sb, KERN_ERR,
1514                                 "journaled quota options not supported");
1515                         break;
1516                 case Opt_noquota:
1517                         break;
1518 #endif
1519                 case Opt_abort:
1520                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1521                         break;
1522                 case Opt_nobarrier:
1523                         clear_opt(sbi->s_mount_opt, BARRIER);
1524                         break;
1525                 case Opt_barrier:
1526                         if (args[0].from) {
1527                                 if (match_int(&args[0], &option))
1528                                         return 0;
1529                         } else
1530                                 option = 1;     /* No argument, default to 1 */
1531                         if (option)
1532                                 set_opt(sbi->s_mount_opt, BARRIER);
1533                         else
1534                                 clear_opt(sbi->s_mount_opt, BARRIER);
1535                         break;
1536                 case Opt_ignore:
1537                         break;
1538                 case Opt_resize:
1539                         if (!is_remount) {
1540                                 ext4_msg(sb, KERN_ERR,
1541                                         "resize option only available "
1542                                         "for remount");
1543                                 return 0;
1544                         }
1545                         if (match_int(&args[0], &option) != 0)
1546                                 return 0;
1547                         *n_blocks_count = option;
1548                         break;
1549                 case Opt_nobh:
1550                         set_opt(sbi->s_mount_opt, NOBH);
1551                         break;
1552                 case Opt_bh:
1553                         clear_opt(sbi->s_mount_opt, NOBH);
1554                         break;
1555                 case Opt_i_version:
1556                         set_opt(sbi->s_mount_opt, I_VERSION);
1557                         sb->s_flags |= MS_I_VERSION;
1558                         break;
1559                 case Opt_nodelalloc:
1560                         clear_opt(sbi->s_mount_opt, DELALLOC);
1561                         break;
1562                 case Opt_stripe:
1563                         if (match_int(&args[0], &option))
1564                                 return 0;
1565                         if (option < 0)
1566                                 return 0;
1567                         sbi->s_stripe = option;
1568                         break;
1569                 case Opt_delalloc:
1570                         set_opt(sbi->s_mount_opt, DELALLOC);
1571                         break;
1572                 case Opt_block_validity:
1573                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1574                         break;
1575                 case Opt_noblock_validity:
1576                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1577                         break;
1578                 case Opt_inode_readahead_blks:
1579                         if (match_int(&args[0], &option))
1580                                 return 0;
1581                         if (option < 0 || option > (1 << 30))
1582                                 return 0;
1583                         if (!is_power_of_2(option)) {
1584                                 ext4_msg(sb, KERN_ERR,
1585                                          "EXT4-fs: inode_readahead_blks"
1586                                          " must be a power of 2");
1587                                 return 0;
1588                         }
1589                         sbi->s_inode_readahead_blks = option;
1590                         break;
1591                 case Opt_journal_ioprio:
1592                         if (match_int(&args[0], &option))
1593                                 return 0;
1594                         if (option < 0 || option > 7)
1595                                 break;
1596                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1597                                                             option);
1598                         break;
1599                 case Opt_noauto_da_alloc:
1600                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1601                         break;
1602                 case Opt_auto_da_alloc:
1603                         if (args[0].from) {
1604                                 if (match_int(&args[0], &option))
1605                                         return 0;
1606                         } else
1607                                 option = 1;     /* No argument, default to 1 */
1608                         if (option)
1609                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1610                         else
1611                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1612                         break;
1613                 case Opt_discard:
1614                         set_opt(sbi->s_mount_opt, DISCARD);
1615                         break;
1616                 case Opt_nodiscard:
1617                         clear_opt(sbi->s_mount_opt, DISCARD);
1618                         break;
1619                 default:
1620                         ext4_msg(sb, KERN_ERR,
1621                                "Unrecognized mount option \"%s\" "
1622                                "or missing value", p);
1623                         return 0;
1624                 }
1625         }
1626 #ifdef CONFIG_QUOTA
1627         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1628                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1629                      sbi->s_qf_names[USRQUOTA])
1630                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1631
1632                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1633                      sbi->s_qf_names[GRPQUOTA])
1634                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1635
1636                 if ((sbi->s_qf_names[USRQUOTA] &&
1637                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1638                     (sbi->s_qf_names[GRPQUOTA] &&
1639                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1640                         ext4_msg(sb, KERN_ERR, "old and new quota "
1641                                         "format mixing");
1642                         return 0;
1643                 }
1644
1645                 if (!sbi->s_jquota_fmt) {
1646                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1647                                         "not specified");
1648                         return 0;
1649                 }
1650         } else {
1651                 if (sbi->s_jquota_fmt) {
1652                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1653                                         "specified with no journaling "
1654                                         "enabled");
1655                         return 0;
1656                 }
1657         }
1658 #endif
1659         return 1;
1660 }
1661
1662 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1663                             int read_only)
1664 {
1665         struct ext4_sb_info *sbi = EXT4_SB(sb);
1666         int res = 0;
1667
1668         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1669                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1670                          "forcing read-only mode");
1671                 res = MS_RDONLY;
1672         }
1673         if (read_only)
1674                 return res;
1675         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1676                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1677                          "running e2fsck is recommended");
1678         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1679                 ext4_msg(sb, KERN_WARNING,
1680                          "warning: mounting fs with errors, "
1681                          "running e2fsck is recommended");
1682         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1683                  le16_to_cpu(es->s_mnt_count) >=
1684                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1685                 ext4_msg(sb, KERN_WARNING,
1686                          "warning: maximal mount count reached, "
1687                          "running e2fsck is recommended");
1688         else if (le32_to_cpu(es->s_checkinterval) &&
1689                 (le32_to_cpu(es->s_lastcheck) +
1690                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1691                 ext4_msg(sb, KERN_WARNING,
1692                          "warning: checktime reached, "
1693                          "running e2fsck is recommended");
1694         if (!sbi->s_journal)
1695                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1696         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1697                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1698         le16_add_cpu(&es->s_mnt_count, 1);
1699         es->s_mtime = cpu_to_le32(get_seconds());
1700         ext4_update_dynamic_rev(sb);
1701         if (sbi->s_journal)
1702                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1703
1704         ext4_commit_super(sb, 1);
1705         if (test_opt(sb, DEBUG))
1706                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1707                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1708                         sb->s_blocksize,
1709                         sbi->s_groups_count,
1710                         EXT4_BLOCKS_PER_GROUP(sb),
1711                         EXT4_INODES_PER_GROUP(sb),
1712                         sbi->s_mount_opt);
1713
1714         return res;
1715 }
1716
1717 static int ext4_fill_flex_info(struct super_block *sb)
1718 {
1719         struct ext4_sb_info *sbi = EXT4_SB(sb);
1720         struct ext4_group_desc *gdp = NULL;
1721         ext4_group_t flex_group_count;
1722         ext4_group_t flex_group;
1723         int groups_per_flex = 0;
1724         size_t size;
1725         int i;
1726
1727         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1728         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1729
1730         if (groups_per_flex < 2) {
1731                 sbi->s_log_groups_per_flex = 0;
1732                 return 1;
1733         }
1734
1735         /* We allocate both existing and potentially added groups */
1736         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1737                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1738                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1739         size = flex_group_count * sizeof(struct flex_groups);
1740         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1741         if (sbi->s_flex_groups == NULL) {
1742                 sbi->s_flex_groups = vmalloc(size);
1743                 if (sbi->s_flex_groups)
1744                         memset(sbi->s_flex_groups, 0, size);
1745         }
1746         if (sbi->s_flex_groups == NULL) {
1747                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1748                                 "%u flex groups", flex_group_count);
1749                 goto failed;
1750         }
1751
1752         for (i = 0; i < sbi->s_groups_count; i++) {
1753                 gdp = ext4_get_group_desc(sb, i, NULL);
1754
1755                 flex_group = ext4_flex_group(sbi, i);
1756                 atomic_add(ext4_free_inodes_count(sb, gdp),
1757                            &sbi->s_flex_groups[flex_group].free_inodes);
1758                 atomic_add(ext4_free_blks_count(sb, gdp),
1759                            &sbi->s_flex_groups[flex_group].free_blocks);
1760                 atomic_add(ext4_used_dirs_count(sb, gdp),
1761                            &sbi->s_flex_groups[flex_group].used_dirs);
1762         }
1763
1764         return 1;
1765 failed:
1766         return 0;
1767 }
1768
1769 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1770                             struct ext4_group_desc *gdp)
1771 {
1772         __u16 crc = 0;
1773
1774         if (sbi->s_es->s_feature_ro_compat &
1775             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1776                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1777                 __le32 le_group = cpu_to_le32(block_group);
1778
1779                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1780                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1781                 crc = crc16(crc, (__u8 *)gdp, offset);
1782                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1783                 /* for checksum of struct ext4_group_desc do the rest...*/
1784                 if ((sbi->s_es->s_feature_incompat &
1785                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1786                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1787                         crc = crc16(crc, (__u8 *)gdp + offset,
1788                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1789                                         offset);
1790         }
1791
1792         return cpu_to_le16(crc);
1793 }
1794
1795 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1796                                 struct ext4_group_desc *gdp)
1797 {
1798         if ((sbi->s_es->s_feature_ro_compat &
1799              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1800             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1801                 return 0;
1802
1803         return 1;
1804 }
1805
1806 /* Called at mount-time, super-block is locked */
1807 static int ext4_check_descriptors(struct super_block *sb)
1808 {
1809         struct ext4_sb_info *sbi = EXT4_SB(sb);
1810         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1811         ext4_fsblk_t last_block;
1812         ext4_fsblk_t block_bitmap;
1813         ext4_fsblk_t inode_bitmap;
1814         ext4_fsblk_t inode_table;
1815         int flexbg_flag = 0;
1816         ext4_group_t i;
1817
1818         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1819                 flexbg_flag = 1;
1820
1821         ext4_debug("Checking group descriptors");
1822
1823         for (i = 0; i < sbi->s_groups_count; i++) {
1824                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1825
1826                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1827                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1828                 else
1829                         last_block = first_block +
1830                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1831
1832                 block_bitmap = ext4_block_bitmap(sb, gdp);
1833                 if (block_bitmap < first_block || block_bitmap > last_block) {
1834                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1835                                "Block bitmap for group %u not in group "
1836                                "(block %llu)!", i, block_bitmap);
1837                         return 0;
1838                 }
1839                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1840                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1841                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1842                                "Inode bitmap for group %u not in group "
1843                                "(block %llu)!", i, inode_bitmap);
1844                         return 0;
1845                 }
1846                 inode_table = ext4_inode_table(sb, gdp);
1847                 if (inode_table < first_block ||
1848                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1849                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1850                                "Inode table for group %u not in group "
1851                                "(block %llu)!", i, inode_table);
1852                         return 0;
1853                 }
1854                 ext4_lock_group(sb, i);
1855                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1856                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1857                                  "Checksum for group %u failed (%u!=%u)",
1858                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1859                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1860                         if (!(sb->s_flags & MS_RDONLY)) {
1861                                 ext4_unlock_group(sb, i);
1862                                 return 0;
1863                         }
1864                 }
1865                 ext4_unlock_group(sb, i);
1866                 if (!flexbg_flag)
1867                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1868         }
1869
1870         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1871         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1872         return 1;
1873 }
1874
1875 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1876  * the superblock) which were deleted from all directories, but held open by
1877  * a process at the time of a crash.  We walk the list and try to delete these
1878  * inodes at recovery time (only with a read-write filesystem).
1879  *
1880  * In order to keep the orphan inode chain consistent during traversal (in
1881  * case of crash during recovery), we link each inode into the superblock
1882  * orphan list_head and handle it the same way as an inode deletion during
1883  * normal operation (which journals the operations for us).
1884  *
1885  * We only do an iget() and an iput() on each inode, which is very safe if we
1886  * accidentally point at an in-use or already deleted inode.  The worst that
1887  * can happen in this case is that we get a "bit already cleared" message from
1888  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1889  * e2fsck was run on this filesystem, and it must have already done the orphan
1890  * inode cleanup for us, so we can safely abort without any further action.
1891  */
1892 static void ext4_orphan_cleanup(struct super_block *sb,
1893                                 struct ext4_super_block *es)
1894 {
1895         unsigned int s_flags = sb->s_flags;
1896         int nr_orphans = 0, nr_truncates = 0;
1897 #ifdef CONFIG_QUOTA
1898         int i;
1899 #endif
1900         if (!es->s_last_orphan) {
1901                 jbd_debug(4, "no orphan inodes to clean up\n");
1902                 return;
1903         }
1904
1905         if (bdev_read_only(sb->s_bdev)) {
1906                 ext4_msg(sb, KERN_ERR, "write access "
1907                         "unavailable, skipping orphan cleanup");
1908                 return;
1909         }
1910
1911         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1912                 if (es->s_last_orphan)
1913                         jbd_debug(1, "Errors on filesystem, "
1914                                   "clearing orphan list.\n");
1915                 es->s_last_orphan = 0;
1916                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1917                 return;
1918         }
1919
1920         if (s_flags & MS_RDONLY) {
1921                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1922                 sb->s_flags &= ~MS_RDONLY;
1923         }
1924 #ifdef CONFIG_QUOTA
1925         /* Needed for iput() to work correctly and not trash data */
1926         sb->s_flags |= MS_ACTIVE;
1927         /* Turn on quotas so that they are updated correctly */
1928         for (i = 0; i < MAXQUOTAS; i++) {
1929                 if (EXT4_SB(sb)->s_qf_names[i]) {
1930                         int ret = ext4_quota_on_mount(sb, i);
1931                         if (ret < 0)
1932                                 ext4_msg(sb, KERN_ERR,
1933                                         "Cannot turn on journaled "
1934                                         "quota: error %d", ret);
1935                 }
1936         }
1937 #endif
1938
1939         while (es->s_last_orphan) {
1940                 struct inode *inode;
1941
1942                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1943                 if (IS_ERR(inode)) {
1944                         es->s_last_orphan = 0;
1945                         break;
1946                 }
1947
1948                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1949                 vfs_dq_init(inode);
1950                 if (inode->i_nlink) {
1951                         ext4_msg(sb, KERN_DEBUG,
1952                                 "%s: truncating inode %lu to %lld bytes",
1953                                 __func__, inode->i_ino, inode->i_size);
1954                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1955                                   inode->i_ino, inode->i_size);
1956                         ext4_truncate(inode);
1957                         nr_truncates++;
1958                 } else {
1959                         ext4_msg(sb, KERN_DEBUG,
1960                                 "%s: deleting unreferenced inode %lu",
1961                                 __func__, inode->i_ino);
1962                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1963                                   inode->i_ino);
1964                         nr_orphans++;
1965                 }
1966                 iput(inode);  /* The delete magic happens here! */
1967         }
1968
1969 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1970
1971         if (nr_orphans)
1972                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1973                        PLURAL(nr_orphans));
1974         if (nr_truncates)
1975                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1976                        PLURAL(nr_truncates));
1977 #ifdef CONFIG_QUOTA
1978         /* Turn quotas off */
1979         for (i = 0; i < MAXQUOTAS; i++) {
1980                 if (sb_dqopt(sb)->files[i])
1981                         vfs_quota_off(sb, i, 0);
1982         }
1983 #endif
1984         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1985 }
1986
1987 /*
1988  * Maximal extent format file size.
1989  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1990  * extent format containers, within a sector_t, and within i_blocks
1991  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1992  * so that won't be a limiting factor.
1993  *
1994  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1995  */
1996 static loff_t ext4_max_size(int blkbits, int has_huge_files)
1997 {
1998         loff_t res;
1999         loff_t upper_limit = MAX_LFS_FILESIZE;
2000
2001         /* small i_blocks in vfs inode? */
2002         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2003                 /*
2004                  * CONFIG_LBDAF is not enabled implies the inode
2005                  * i_block represent total blocks in 512 bytes
2006                  * 32 == size of vfs inode i_blocks * 8
2007                  */
2008                 upper_limit = (1LL << 32) - 1;
2009
2010                 /* total blocks in file system block size */
2011                 upper_limit >>= (blkbits - 9);
2012                 upper_limit <<= blkbits;
2013         }
2014
2015         /* 32-bit extent-start container, ee_block */
2016         res = 1LL << 32;
2017         res <<= blkbits;
2018         res -= 1;
2019
2020         /* Sanity check against vm- & vfs- imposed limits */
2021         if (res > upper_limit)
2022                 res = upper_limit;
2023
2024         return res;
2025 }
2026
2027 /*
2028  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2029  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2030  * We need to be 1 filesystem block less than the 2^48 sector limit.
2031  */
2032 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2033 {
2034         loff_t res = EXT4_NDIR_BLOCKS;
2035         int meta_blocks;
2036         loff_t upper_limit;
2037         /* This is calculated to be the largest file size for a dense, block
2038          * mapped file such that the file's total number of 512-byte sectors,
2039          * including data and all indirect blocks, does not exceed (2^48 - 1).
2040          *
2041          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2042          * number of 512-byte sectors of the file.
2043          */
2044
2045         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2046                 /*
2047                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2048                  * the inode i_block field represents total file blocks in
2049                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2050                  */
2051                 upper_limit = (1LL << 32) - 1;
2052
2053                 /* total blocks in file system block size */
2054                 upper_limit >>= (bits - 9);
2055
2056         } else {
2057                 /*
2058                  * We use 48 bit ext4_inode i_blocks
2059                  * With EXT4_HUGE_FILE_FL set the i_blocks
2060                  * represent total number of blocks in
2061                  * file system block size
2062                  */
2063                 upper_limit = (1LL << 48) - 1;
2064
2065         }
2066
2067         /* indirect blocks */
2068         meta_blocks = 1;
2069         /* double indirect blocks */
2070         meta_blocks += 1 + (1LL << (bits-2));
2071         /* tripple indirect blocks */
2072         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2073
2074         upper_limit -= meta_blocks;
2075         upper_limit <<= bits;
2076
2077         res += 1LL << (bits-2);
2078         res += 1LL << (2*(bits-2));
2079         res += 1LL << (3*(bits-2));
2080         res <<= bits;
2081         if (res > upper_limit)
2082                 res = upper_limit;
2083
2084         if (res > MAX_LFS_FILESIZE)
2085                 res = MAX_LFS_FILESIZE;
2086
2087         return res;
2088 }
2089
2090 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2091                                    ext4_fsblk_t logical_sb_block, int nr)
2092 {
2093         struct ext4_sb_info *sbi = EXT4_SB(sb);
2094         ext4_group_t bg, first_meta_bg;
2095         int has_super = 0;
2096
2097         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2098
2099         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2100             nr < first_meta_bg)
2101                 return logical_sb_block + nr + 1;
2102         bg = sbi->s_desc_per_block * nr;
2103         if (ext4_bg_has_super(sb, bg))
2104                 has_super = 1;
2105
2106         return (has_super + ext4_group_first_block_no(sb, bg));
2107 }
2108
2109 /**
2110  * ext4_get_stripe_size: Get the stripe size.
2111  * @sbi: In memory super block info
2112  *
2113  * If we have specified it via mount option, then
2114  * use the mount option value. If the value specified at mount time is
2115  * greater than the blocks per group use the super block value.
2116  * If the super block value is greater than blocks per group return 0.
2117  * Allocator needs it be less than blocks per group.
2118  *
2119  */
2120 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2121 {
2122         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2123         unsigned long stripe_width =
2124                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2125
2126         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2127                 return sbi->s_stripe;
2128
2129         if (stripe_width <= sbi->s_blocks_per_group)
2130                 return stripe_width;
2131
2132         if (stride <= sbi->s_blocks_per_group)
2133                 return stride;
2134
2135         return 0;
2136 }
2137
2138 /* sysfs supprt */
2139
2140 struct ext4_attr {
2141         struct attribute attr;
2142         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2143         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2144                          const char *, size_t);
2145         int offset;
2146 };
2147
2148 static int parse_strtoul(const char *buf,
2149                 unsigned long max, unsigned long *value)
2150 {
2151         char *endp;
2152
2153         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2154         endp = skip_spaces(endp);
2155         if (*endp || *value > max)
2156                 return -EINVAL;
2157
2158         return 0;
2159 }
2160
2161 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2162                                               struct ext4_sb_info *sbi,
2163                                               char *buf)
2164 {
2165         return snprintf(buf, PAGE_SIZE, "%llu\n",
2166                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2167 }
2168
2169 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2170                                          struct ext4_sb_info *sbi, char *buf)
2171 {
2172         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2173
2174         return snprintf(buf, PAGE_SIZE, "%lu\n",
2175                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2176                          sbi->s_sectors_written_start) >> 1);
2177 }
2178
2179 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2180                                           struct ext4_sb_info *sbi, char *buf)
2181 {
2182         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2183
2184         return snprintf(buf, PAGE_SIZE, "%llu\n",
2185                         (unsigned long long)(sbi->s_kbytes_written +
2186                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2187                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2188 }
2189
2190 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2191                                           struct ext4_sb_info *sbi,
2192                                           const char *buf, size_t count)
2193 {
2194         unsigned long t;
2195
2196         if (parse_strtoul(buf, 0x40000000, &t))
2197                 return -EINVAL;
2198
2199         if (!is_power_of_2(t))
2200                 return -EINVAL;
2201
2202         sbi->s_inode_readahead_blks = t;
2203         return count;
2204 }
2205
2206 static ssize_t sbi_ui_show(struct ext4_attr *a,
2207                            struct ext4_sb_info *sbi, char *buf)
2208 {
2209         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2210
2211         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2212 }
2213
2214 static ssize_t sbi_ui_store(struct ext4_attr *a,
2215                             struct ext4_sb_info *sbi,
2216                             const char *buf, size_t count)
2217 {
2218         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2219         unsigned long t;
2220
2221         if (parse_strtoul(buf, 0xffffffff, &t))
2222                 return -EINVAL;
2223         *ui = t;
2224         return count;
2225 }
2226
2227 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2228 static struct ext4_attr ext4_attr_##_name = {                   \
2229         .attr = {.name = __stringify(_name), .mode = _mode },   \
2230         .show   = _show,                                        \
2231         .store  = _store,                                       \
2232         .offset = offsetof(struct ext4_sb_info, _elname),       \
2233 }
2234 #define EXT4_ATTR(name, mode, show, store) \
2235 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2236
2237 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2238 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2239 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2240         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2241 #define ATTR_LIST(name) &ext4_attr_##name.attr
2242
2243 EXT4_RO_ATTR(delayed_allocation_blocks);
2244 EXT4_RO_ATTR(session_write_kbytes);
2245 EXT4_RO_ATTR(lifetime_write_kbytes);
2246 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2247                  inode_readahead_blks_store, s_inode_readahead_blks);
2248 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2249 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2250 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2251 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2252 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2253 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2254 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2255 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2256
2257 static struct attribute *ext4_attrs[] = {
2258         ATTR_LIST(delayed_allocation_blocks),
2259         ATTR_LIST(session_write_kbytes),
2260         ATTR_LIST(lifetime_write_kbytes),
2261         ATTR_LIST(inode_readahead_blks),
2262         ATTR_LIST(inode_goal),
2263         ATTR_LIST(mb_stats),
2264         ATTR_LIST(mb_max_to_scan),
2265         ATTR_LIST(mb_min_to_scan),
2266         ATTR_LIST(mb_order2_req),
2267         ATTR_LIST(mb_stream_req),
2268         ATTR_LIST(mb_group_prealloc),
2269         ATTR_LIST(max_writeback_mb_bump),
2270         NULL,
2271 };
2272
2273 static ssize_t ext4_attr_show(struct kobject *kobj,
2274                               struct attribute *attr, char *buf)
2275 {
2276         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2277                                                 s_kobj);
2278         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2279
2280         return a->show ? a->show(a, sbi, buf) : 0;
2281 }
2282
2283 static ssize_t ext4_attr_store(struct kobject *kobj,
2284                                struct attribute *attr,
2285                                const char *buf, size_t len)
2286 {
2287         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2288                                                 s_kobj);
2289         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2290
2291         return a->store ? a->store(a, sbi, buf, len) : 0;
2292 }
2293
2294 static void ext4_sb_release(struct kobject *kobj)
2295 {
2296         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2297                                                 s_kobj);
2298         complete(&sbi->s_kobj_unregister);
2299 }
2300
2301
2302 static struct sysfs_ops ext4_attr_ops = {
2303         .show   = ext4_attr_show,
2304         .store  = ext4_attr_store,
2305 };
2306
2307 static struct kobj_type ext4_ktype = {
2308         .default_attrs  = ext4_attrs,
2309         .sysfs_ops      = &ext4_attr_ops,
2310         .release        = ext4_sb_release,
2311 };
2312
2313 /*
2314  * Check whether this filesystem can be mounted based on
2315  * the features present and the RDONLY/RDWR mount requested.
2316  * Returns 1 if this filesystem can be mounted as requested,
2317  * 0 if it cannot be.
2318  */
2319 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2320 {
2321         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2322                 ext4_msg(sb, KERN_ERR,
2323                         "Couldn't mount because of "
2324                         "unsupported optional features (%x)",
2325                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2326                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2327                 return 0;
2328         }
2329
2330         if (readonly)
2331                 return 1;
2332
2333         /* Check that feature set is OK for a read-write mount */
2334         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2335                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2336                          "unsupported optional features (%x)",
2337                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2338                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2339                 return 0;
2340         }
2341         /*
2342          * Large file size enabled file system can only be mounted
2343          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2344          */
2345         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2346                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2347                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2348                                  "cannot be mounted RDWR without "
2349                                  "CONFIG_LBDAF");
2350                         return 0;
2351                 }
2352         }
2353         return 1;
2354 }
2355
2356 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2357                                 __releases(kernel_lock)
2358                                 __acquires(kernel_lock)
2359 {
2360         struct buffer_head *bh;
2361         struct ext4_super_block *es = NULL;
2362         struct ext4_sb_info *sbi;
2363         ext4_fsblk_t block;
2364         ext4_fsblk_t sb_block = get_sb_block(&data);
2365         ext4_fsblk_t logical_sb_block;
2366         unsigned long offset = 0;
2367         unsigned long journal_devnum = 0;
2368         unsigned long def_mount_opts;
2369         struct inode *root;
2370         char *cp;
2371         const char *descr;
2372         int ret = -EINVAL;
2373         int blocksize;
2374         unsigned int db_count;
2375         unsigned int i;
2376         int needs_recovery, has_huge_files;
2377         __u64 blocks_count;
2378         int err;
2379         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2380
2381         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2382         if (!sbi)
2383                 return -ENOMEM;
2384
2385         sbi->s_blockgroup_lock =
2386                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2387         if (!sbi->s_blockgroup_lock) {
2388                 kfree(sbi);
2389                 return -ENOMEM;
2390         }
2391         sb->s_fs_info = sbi;
2392         sbi->s_mount_opt = 0;
2393         sbi->s_resuid = EXT4_DEF_RESUID;
2394         sbi->s_resgid = EXT4_DEF_RESGID;
2395         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2396         sbi->s_sb_block = sb_block;
2397         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2398                                                       sectors[1]);
2399
2400         unlock_kernel();
2401
2402         /* Cleanup superblock name */
2403         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2404                 *cp = '!';
2405
2406         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2407         if (!blocksize) {
2408                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2409                 goto out_fail;
2410         }
2411
2412         /*
2413          * The ext4 superblock will not be buffer aligned for other than 1kB
2414          * block sizes.  We need to calculate the offset from buffer start.
2415          */
2416         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2417                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2418                 offset = do_div(logical_sb_block, blocksize);
2419         } else {
2420                 logical_sb_block = sb_block;
2421         }
2422
2423         if (!(bh = sb_bread(sb, logical_sb_block))) {
2424                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2425                 goto out_fail;
2426         }
2427         /*
2428          * Note: s_es must be initialized as soon as possible because
2429          *       some ext4 macro-instructions depend on its value
2430          */
2431         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2432         sbi->s_es = es;
2433         sb->s_magic = le16_to_cpu(es->s_magic);
2434         if (sb->s_magic != EXT4_SUPER_MAGIC)
2435                 goto cantfind_ext4;
2436         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2437
2438         /* Set defaults before we parse the mount options */
2439         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2440         if (def_mount_opts & EXT4_DEFM_DEBUG)
2441                 set_opt(sbi->s_mount_opt, DEBUG);
2442         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
2443                 set_opt(sbi->s_mount_opt, GRPID);
2444         if (def_mount_opts & EXT4_DEFM_UID16)
2445                 set_opt(sbi->s_mount_opt, NO_UID32);
2446 #ifdef CONFIG_EXT4_FS_XATTR
2447         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2448                 set_opt(sbi->s_mount_opt, XATTR_USER);
2449 #endif
2450 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2451         if (def_mount_opts & EXT4_DEFM_ACL)
2452                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2453 #endif
2454         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2455                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
2456         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2457                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
2458         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2459                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
2460
2461         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2462                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2463         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2464                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2465         else
2466                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2467
2468         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2469         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2470         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2471         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2472         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2473
2474         set_opt(sbi->s_mount_opt, BARRIER);
2475
2476         /*
2477          * enable delayed allocation by default
2478          * Use -o nodelalloc to turn it off
2479          */
2480         set_opt(sbi->s_mount_opt, DELALLOC);
2481
2482         if (!parse_options((char *) data, sb, &journal_devnum,
2483                            &journal_ioprio, NULL, 0))
2484                 goto failed_mount;
2485
2486         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2487                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2488
2489         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2490             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2491              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2492              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2493                 ext4_msg(sb, KERN_WARNING,
2494                        "feature flags set on rev 0 fs, "
2495                        "running e2fsck is recommended");
2496
2497         /*
2498          * Check feature flags regardless of the revision level, since we
2499          * previously didn't change the revision level when setting the flags,
2500          * so there is a chance incompat flags are set on a rev 0 filesystem.
2501          */
2502         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2503                 goto failed_mount;
2504
2505         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2506
2507         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2508             blocksize > EXT4_MAX_BLOCK_SIZE) {
2509                 ext4_msg(sb, KERN_ERR,
2510                        "Unsupported filesystem blocksize %d", blocksize);
2511                 goto failed_mount;
2512         }
2513
2514         if (sb->s_blocksize != blocksize) {
2515                 /* Validate the filesystem blocksize */
2516                 if (!sb_set_blocksize(sb, blocksize)) {
2517                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2518                                         blocksize);
2519                         goto failed_mount;
2520                 }
2521
2522                 brelse(bh);
2523                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2524                 offset = do_div(logical_sb_block, blocksize);
2525                 bh = sb_bread(sb, logical_sb_block);
2526                 if (!bh) {
2527                         ext4_msg(sb, KERN_ERR,
2528                                "Can't read superblock on 2nd try");
2529                         goto failed_mount;
2530                 }
2531                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2532                 sbi->s_es = es;
2533                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2534                         ext4_msg(sb, KERN_ERR,
2535                                "Magic mismatch, very weird!");
2536                         goto failed_mount;
2537                 }
2538         }
2539
2540         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2541                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2542         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2543                                                       has_huge_files);
2544         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2545
2546         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2547                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2548                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2549         } else {
2550                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2551                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2552                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2553                     (!is_power_of_2(sbi->s_inode_size)) ||
2554                     (sbi->s_inode_size > blocksize)) {
2555                         ext4_msg(sb, KERN_ERR,
2556                                "unsupported inode size: %d",
2557                                sbi->s_inode_size);
2558                         goto failed_mount;
2559                 }
2560                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2561                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2562         }
2563
2564         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2565         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2566                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2567                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2568                     !is_power_of_2(sbi->s_desc_size)) {
2569                         ext4_msg(sb, KERN_ERR,
2570                                "unsupported descriptor size %lu",
2571                                sbi->s_desc_size);
2572                         goto failed_mount;
2573                 }
2574         } else
2575                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2576
2577         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2578         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2579         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2580                 goto cantfind_ext4;
2581
2582         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2583         if (sbi->s_inodes_per_block == 0)
2584                 goto cantfind_ext4;
2585         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2586                                         sbi->s_inodes_per_block;
2587         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2588         sbi->s_sbh = bh;
2589         sbi->s_mount_state = le16_to_cpu(es->s_state);
2590         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2591         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2592
2593         for (i = 0; i < 4; i++)
2594                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2595         sbi->s_def_hash_version = es->s_def_hash_version;
2596         i = le32_to_cpu(es->s_flags);
2597         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2598                 sbi->s_hash_unsigned = 3;
2599         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2600 #ifdef __CHAR_UNSIGNED__
2601                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2602                 sbi->s_hash_unsigned = 3;
2603 #else
2604                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2605 #endif
2606                 sb->s_dirt = 1;
2607         }
2608
2609         if (sbi->s_blocks_per_group > blocksize * 8) {
2610                 ext4_msg(sb, KERN_ERR,
2611                        "#blocks per group too big: %lu",
2612                        sbi->s_blocks_per_group);
2613                 goto failed_mount;
2614         }
2615         if (sbi->s_inodes_per_group > blocksize * 8) {
2616                 ext4_msg(sb, KERN_ERR,
2617                        "#inodes per group too big: %lu",
2618                        sbi->s_inodes_per_group);
2619                 goto failed_mount;
2620         }
2621
2622         /*
2623          * Test whether we have more sectors than will fit in sector_t,
2624          * and whether the max offset is addressable by the page cache.
2625          */
2626         if ((ext4_blocks_count(es) >
2627              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2628             (ext4_blocks_count(es) >
2629              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2630                 ext4_msg(sb, KERN_ERR, "filesystem"
2631                          " too large to mount safely on this system");
2632                 if (sizeof(sector_t) < 8)
2633                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2634                 ret = -EFBIG;
2635                 goto failed_mount;
2636         }
2637
2638         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2639                 goto cantfind_ext4;
2640
2641         /* check blocks count against device size */
2642         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2643         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2644                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2645                        "exceeds size of device (%llu blocks)",
2646                        ext4_blocks_count(es), blocks_count);
2647                 goto failed_mount;
2648         }
2649
2650         /*
2651          * It makes no sense for the first data block to be beyond the end
2652          * of the filesystem.
2653          */
2654         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2655                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2656                          "block %u is beyond end of filesystem (%llu)",
2657                          le32_to_cpu(es->s_first_data_block),
2658                          ext4_blocks_count(es));
2659                 goto failed_mount;
2660         }
2661         blocks_count = (ext4_blocks_count(es) -
2662                         le32_to_cpu(es->s_first_data_block) +
2663                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2664         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2665         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2666                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2667                        "(block count %llu, first data block %u, "
2668                        "blocks per group %lu)", sbi->s_groups_count,
2669                        ext4_blocks_count(es),
2670                        le32_to_cpu(es->s_first_data_block),
2671                        EXT4_BLOCKS_PER_GROUP(sb));
2672                 goto failed_mount;
2673         }
2674         sbi->s_groups_count = blocks_count;
2675         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2676                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2677         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2678                    EXT4_DESC_PER_BLOCK(sb);
2679         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2680                                     GFP_KERNEL);
2681         if (sbi->s_group_desc == NULL) {
2682                 ext4_msg(sb, KERN_ERR, "not enough memory");
2683                 goto failed_mount;
2684         }
2685
2686 #ifdef CONFIG_PROC_FS
2687         if (ext4_proc_root)
2688                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2689 #endif
2690
2691         bgl_lock_init(sbi->s_blockgroup_lock);
2692
2693         for (i = 0; i < db_count; i++) {
2694                 block = descriptor_loc(sb, logical_sb_block, i);
2695                 sbi->s_group_desc[i] = sb_bread(sb, block);
2696                 if (!sbi->s_group_desc[i]) {
2697                         ext4_msg(sb, KERN_ERR,
2698                                "can't read group descriptor %d", i);
2699                         db_count = i;
2700                         goto failed_mount2;
2701                 }
2702         }
2703         if (!ext4_check_descriptors(sb)) {
2704                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2705                 goto failed_mount2;
2706         }
2707         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2708                 if (!ext4_fill_flex_info(sb)) {
2709                         ext4_msg(sb, KERN_ERR,
2710                                "unable to initialize "
2711                                "flex_bg meta info!");
2712                         goto failed_mount2;
2713                 }
2714
2715         sbi->s_gdb_count = db_count;
2716         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2717         spin_lock_init(&sbi->s_next_gen_lock);
2718
2719         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2720                         ext4_count_free_blocks(sb));
2721         if (!err) {
2722                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2723                                 ext4_count_free_inodes(sb));
2724         }
2725         if (!err) {
2726                 err = percpu_counter_init(&sbi->s_dirs_counter,
2727                                 ext4_count_dirs(sb));
2728         }
2729         if (!err) {
2730                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2731         }
2732         if (err) {
2733                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2734                 goto failed_mount3;
2735         }
2736
2737         sbi->s_stripe = ext4_get_stripe_size(sbi);
2738         sbi->s_max_writeback_mb_bump = 128;
2739
2740         /*
2741          * set up enough so that it can read an inode
2742          */
2743         if (!test_opt(sb, NOLOAD) &&
2744             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2745                 sb->s_op = &ext4_sops;
2746         else
2747                 sb->s_op = &ext4_nojournal_sops;
2748         sb->s_export_op = &ext4_export_ops;
2749         sb->s_xattr = ext4_xattr_handlers;
2750 #ifdef CONFIG_QUOTA
2751         sb->s_qcop = &ext4_qctl_operations;
2752         sb->dq_op = &ext4_quota_operations;
2753 #endif
2754         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2755         mutex_init(&sbi->s_orphan_lock);
2756         mutex_init(&sbi->s_resize_lock);
2757
2758         sb->s_root = NULL;
2759
2760         needs_recovery = (es->s_last_orphan != 0 ||
2761                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2762                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2763
2764         /*
2765          * The first inode we look at is the journal inode.  Don't try
2766          * root first: it may be modified in the journal!
2767          */
2768         if (!test_opt(sb, NOLOAD) &&
2769             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2770                 if (ext4_load_journal(sb, es, journal_devnum))
2771                         goto failed_mount3;
2772         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2773               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2774                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2775                        "suppressed and not mounted read-only");
2776                 goto failed_mount4;
2777         } else {
2778                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2779                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2780                 sbi->s_journal = NULL;
2781                 needs_recovery = 0;
2782                 goto no_journal;
2783         }
2784
2785         if (ext4_blocks_count(es) > 0xffffffffULL &&
2786             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2787                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2788                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2789                 goto failed_mount4;
2790         }
2791
2792         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2793                 jbd2_journal_set_features(sbi->s_journal,
2794                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2795                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2796         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2797                 jbd2_journal_set_features(sbi->s_journal,
2798                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2799                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2800                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2801         } else {
2802                 jbd2_journal_clear_features(sbi->s_journal,
2803                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2804                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2805         }
2806
2807         /* We have now updated the journal if required, so we can
2808          * validate the data journaling mode. */
2809         switch (test_opt(sb, DATA_FLAGS)) {
2810         case 0:
2811                 /* No mode set, assume a default based on the journal
2812                  * capabilities: ORDERED_DATA if the journal can
2813                  * cope, else JOURNAL_DATA
2814                  */
2815                 if (jbd2_journal_check_available_features
2816                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2817                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2818                 else
2819                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2820                 break;
2821
2822         case EXT4_MOUNT_ORDERED_DATA:
2823         case EXT4_MOUNT_WRITEBACK_DATA:
2824                 if (!jbd2_journal_check_available_features
2825                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2826                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2827                                "requested data journaling mode");
2828                         goto failed_mount4;
2829                 }
2830         default:
2831                 break;
2832         }
2833         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2834
2835 no_journal:
2836
2837         if (test_opt(sb, NOBH)) {
2838                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2839                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2840                                 "its supported only with writeback mode");
2841                         clear_opt(sbi->s_mount_opt, NOBH);
2842                 }
2843         }
2844         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2845         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2846                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2847                 goto failed_mount_wq;
2848         }
2849
2850         /*
2851          * The jbd2_journal_load will have done any necessary log recovery,
2852          * so we can safely mount the rest of the filesystem now.
2853          */
2854
2855         root = ext4_iget(sb, EXT4_ROOT_INO);
2856         if (IS_ERR(root)) {
2857                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2858                 ret = PTR_ERR(root);
2859                 goto failed_mount4;
2860         }
2861         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2862                 iput(root);
2863                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2864                 goto failed_mount4;
2865         }
2866         sb->s_root = d_alloc_root(root);
2867         if (!sb->s_root) {
2868                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2869                 iput(root);
2870                 ret = -ENOMEM;
2871                 goto failed_mount4;
2872         }
2873
2874         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2875
2876         /* determine the minimum size of new large inodes, if present */
2877         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2878                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2879                                                      EXT4_GOOD_OLD_INODE_SIZE;
2880                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2881                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2882                         if (sbi->s_want_extra_isize <
2883                             le16_to_cpu(es->s_want_extra_isize))
2884                                 sbi->s_want_extra_isize =
2885                                         le16_to_cpu(es->s_want_extra_isize);
2886                         if (sbi->s_want_extra_isize <
2887                             le16_to_cpu(es->s_min_extra_isize))
2888                                 sbi->s_want_extra_isize =
2889                                         le16_to_cpu(es->s_min_extra_isize);
2890                 }
2891         }
2892         /* Check if enough inode space is available */
2893         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2894                                                         sbi->s_inode_size) {
2895                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2896                                                        EXT4_GOOD_OLD_INODE_SIZE;
2897                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2898                          "available");
2899         }
2900
2901         if (test_opt(sb, DELALLOC) &&
2902             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2903                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2904                          "requested data journaling mode");
2905                 clear_opt(sbi->s_mount_opt, DELALLOC);
2906         }
2907
2908         err = ext4_setup_system_zone(sb);
2909         if (err) {
2910                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
2911                          "zone (%d)\n", err);
2912                 goto failed_mount4;
2913         }
2914
2915         ext4_ext_init(sb);
2916         err = ext4_mb_init(sb, needs_recovery);
2917         if (err) {
2918                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
2919                          err);
2920                 goto failed_mount4;
2921         }
2922
2923         sbi->s_kobj.kset = ext4_kset;
2924         init_completion(&sbi->s_kobj_unregister);
2925         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
2926                                    "%s", sb->s_id);
2927         if (err) {
2928                 ext4_mb_release(sb);
2929                 ext4_ext_release(sb);
2930                 goto failed_mount4;
2931         };
2932
2933         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2934         ext4_orphan_cleanup(sb, es);
2935         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2936         if (needs_recovery) {
2937                 ext4_msg(sb, KERN_INFO, "recovery complete");
2938                 ext4_mark_recovery_complete(sb, es);
2939         }
2940         if (EXT4_SB(sb)->s_journal) {
2941                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2942                         descr = " journalled data mode";
2943                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2944                         descr = " ordered data mode";
2945                 else
2946                         descr = " writeback data mode";
2947         } else
2948                 descr = "out journal";
2949
2950         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s", descr);
2951
2952         lock_kernel();
2953         return 0;
2954
2955 cantfind_ext4:
2956         if (!silent)
2957                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
2958         goto failed_mount;
2959
2960 failed_mount4:
2961         ext4_msg(sb, KERN_ERR, "mount failed");
2962         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
2963 failed_mount_wq:
2964         ext4_release_system_zone(sb);
2965         if (sbi->s_journal) {
2966                 jbd2_journal_destroy(sbi->s_journal);
2967                 sbi->s_journal = NULL;
2968         }
2969 failed_mount3:
2970         if (sbi->s_flex_groups) {
2971                 if (is_vmalloc_addr(sbi->s_flex_groups))
2972                         vfree(sbi->s_flex_groups);
2973                 else
2974                         kfree(sbi->s_flex_groups);
2975         }
2976         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2977         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2978         percpu_counter_destroy(&sbi->s_dirs_counter);
2979         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2980 failed_mount2:
2981         for (i = 0; i < db_count; i++)
2982                 brelse(sbi->s_group_desc[i]);
2983         kfree(sbi->s_group_desc);
2984 failed_mount:
2985         if (sbi->s_proc) {
2986                 remove_proc_entry(sb->s_id, ext4_proc_root);
2987         }
2988 #ifdef CONFIG_QUOTA
2989         for (i = 0; i < MAXQUOTAS; i++)
2990                 kfree(sbi->s_qf_names[i]);
2991 #endif
2992         ext4_blkdev_remove(sbi);
2993         brelse(bh);
2994 out_fail:
2995         sb->s_fs_info = NULL;
2996         kfree(sbi->s_blockgroup_lock);
2997         kfree(sbi);
2998         lock_kernel();
2999         return ret;
3000 }
3001
3002 /*
3003  * Setup any per-fs journal parameters now.  We'll do this both on
3004  * initial mount, once the journal has been initialised but before we've
3005  * done any recovery; and again on any subsequent remount.
3006  */
3007 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3008 {
3009         struct ext4_sb_info *sbi = EXT4_SB(sb);
3010
3011         journal->j_commit_interval = sbi->s_commit_interval;
3012         journal->j_min_batch_time = sbi->s_min_batch_time;
3013         journal->j_max_batch_time = sbi->s_max_batch_time;
3014
3015         spin_lock(&journal->j_state_lock);
3016         if (test_opt(sb, BARRIER))
3017                 journal->j_flags |= JBD2_BARRIER;
3018         else
3019                 journal->j_flags &= ~JBD2_BARRIER;
3020         if (test_opt(sb, DATA_ERR_ABORT))
3021                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3022         else
3023                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3024         spin_unlock(&journal->j_state_lock);
3025 }
3026
3027 static journal_t *ext4_get_journal(struct super_block *sb,
3028                                    unsigned int journal_inum)
3029 {
3030         struct inode *journal_inode;
3031         journal_t *journal;
3032
3033         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3034
3035         /* First, test for the existence of a valid inode on disk.  Bad
3036          * things happen if we iget() an unused inode, as the subsequent
3037          * iput() will try to delete it. */
3038
3039         journal_inode = ext4_iget(sb, journal_inum);
3040         if (IS_ERR(journal_inode)) {
3041                 ext4_msg(sb, KERN_ERR, "no journal found");
3042                 return NULL;
3043         }
3044         if (!journal_inode->i_nlink) {
3045                 make_bad_inode(journal_inode);
3046                 iput(journal_inode);
3047                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3048                 return NULL;
3049         }
3050
3051         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3052                   journal_inode, journal_inode->i_size);
3053         if (!S_ISREG(journal_inode->i_mode)) {
3054                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3055                 iput(journal_inode);
3056                 return NULL;
3057         }
3058
3059         journal = jbd2_journal_init_inode(journal_inode);
3060         if (!journal) {
3061                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3062                 iput(journal_inode);
3063                 return NULL;
3064         }
3065         journal->j_private = sb;
3066         ext4_init_journal_params(sb, journal);
3067         return journal;
3068 }
3069
3070 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3071                                        dev_t j_dev)
3072 {
3073         struct buffer_head *bh;
3074         journal_t *journal;
3075         ext4_fsblk_t start;
3076         ext4_fsblk_t len;
3077         int hblock, blocksize;
3078         ext4_fsblk_t sb_block;
3079         unsigned long offset;
3080         struct ext4_super_block *es;
3081         struct block_device *bdev;
3082
3083         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3084
3085         bdev = ext4_blkdev_get(j_dev, sb);
3086         if (bdev == NULL)
3087                 return NULL;
3088
3089         if (bd_claim(bdev, sb)) {
3090                 ext4_msg(sb, KERN_ERR,
3091                         "failed to claim external journal device");
3092                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3093                 return NULL;
3094         }
3095
3096         blocksize = sb->s_blocksize;
3097         hblock = bdev_logical_block_size(bdev);
3098         if (blocksize < hblock) {
3099                 ext4_msg(sb, KERN_ERR,
3100                         "blocksize too small for journal device");
3101                 goto out_bdev;
3102         }
3103
3104         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3105         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3106         set_blocksize(bdev, blocksize);
3107         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3108                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3109                        "external journal");
3110                 goto out_bdev;
3111         }
3112
3113         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3114         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3115             !(le32_to_cpu(es->s_feature_incompat) &
3116               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3117                 ext4_msg(sb, KERN_ERR, "external journal has "
3118                                         "bad superblock");
3119                 brelse(bh);
3120                 goto out_bdev;
3121         }
3122
3123         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3124                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3125                 brelse(bh);
3126                 goto out_bdev;
3127         }
3128
3129         len = ext4_blocks_count(es);
3130         start = sb_block + 1;
3131         brelse(bh);     /* we're done with the superblock */
3132
3133         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3134                                         start, len, blocksize);
3135         if (!journal) {
3136                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3137                 goto out_bdev;
3138         }
3139         journal->j_private = sb;
3140         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3141         wait_on_buffer(journal->j_sb_buffer);
3142         if (!buffer_uptodate(journal->j_sb_buffer)) {
3143                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3144                 goto out_journal;
3145         }
3146         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3147                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3148                                         "user (unsupported) - %d",
3149                         be32_to_cpu(journal->j_superblock->s_nr_users));
3150                 goto out_journal;
3151         }
3152         EXT4_SB(sb)->journal_bdev = bdev;
3153         ext4_init_journal_params(sb, journal);
3154         return journal;
3155
3156 out_journal:
3157         jbd2_journal_destroy(journal);
3158 out_bdev:
3159         ext4_blkdev_put(bdev);
3160         return NULL;
3161 }
3162
3163 static int ext4_load_journal(struct super_block *sb,
3164                              struct ext4_super_block *es,
3165                              unsigned long journal_devnum)
3166 {
3167         journal_t *journal;
3168         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3169         dev_t journal_dev;
3170         int err = 0;
3171         int really_read_only;
3172
3173         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3174
3175         if (journal_devnum &&
3176             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3177                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3178                         "numbers have changed");
3179                 journal_dev = new_decode_dev(journal_devnum);
3180         } else
3181                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3182
3183         really_read_only = bdev_read_only(sb->s_bdev);
3184
3185         /*
3186          * Are we loading a blank journal or performing recovery after a
3187          * crash?  For recovery, we need to check in advance whether we
3188          * can get read-write access to the device.
3189          */
3190         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3191                 if (sb->s_flags & MS_RDONLY) {
3192                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3193                                         "required on readonly filesystem");
3194                         if (really_read_only) {
3195                                 ext4_msg(sb, KERN_ERR, "write access "
3196                                         "unavailable, cannot proceed");
3197                                 return -EROFS;
3198                         }
3199                         ext4_msg(sb, KERN_INFO, "write access will "
3200                                "be enabled during recovery");
3201                 }
3202         }
3203
3204         if (journal_inum && journal_dev) {
3205                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3206                        "and inode journals!");
3207                 return -EINVAL;
3208         }
3209
3210         if (journal_inum) {
3211                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3212                         return -EINVAL;
3213         } else {
3214                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3215                         return -EINVAL;
3216         }
3217
3218         if (!(journal->j_flags & JBD2_BARRIER))
3219                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3220
3221         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3222                 err = jbd2_journal_update_format(journal);
3223                 if (err)  {
3224                         ext4_msg(sb, KERN_ERR, "error updating journal");
3225                         jbd2_journal_destroy(journal);
3226                         return err;
3227                 }
3228         }
3229
3230         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3231                 err = jbd2_journal_wipe(journal, !really_read_only);
3232         if (!err)
3233                 err = jbd2_journal_load(journal);
3234
3235         if (err) {
3236                 ext4_msg(sb, KERN_ERR, "error loading journal");
3237                 jbd2_journal_destroy(journal);
3238                 return err;
3239         }
3240
3241         EXT4_SB(sb)->s_journal = journal;
3242         ext4_clear_journal_err(sb, es);
3243
3244         if (journal_devnum &&
3245             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3246                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3247
3248                 /* Make sure we flush the recovery flag to disk. */
3249                 ext4_commit_super(sb, 1);
3250         }
3251
3252         return 0;
3253 }
3254
3255 static int ext4_commit_super(struct super_block *sb, int sync)
3256 {
3257         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3258         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3259         int error = 0;
3260
3261         if (!sbh)
3262                 return error;
3263         if (buffer_write_io_error(sbh)) {
3264                 /*
3265                  * Oh, dear.  A previous attempt to write the
3266                  * superblock failed.  This could happen because the
3267                  * USB device was yanked out.  Or it could happen to
3268                  * be a transient write error and maybe the block will
3269                  * be remapped.  Nothing we can do but to retry the
3270                  * write and hope for the best.
3271                  */
3272                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3273                        "superblock detected");
3274                 clear_buffer_write_io_error(sbh);
3275                 set_buffer_uptodate(sbh);
3276         }
3277         /*
3278          * If the file system is mounted read-only, don't update the
3279          * superblock write time.  This avoids updating the superblock
3280          * write time when we are mounting the root file system
3281          * read/only but we need to replay the journal; at that point,
3282          * for people who are east of GMT and who make their clock
3283          * tick in localtime for Windows bug-for-bug compatibility,
3284          * the clock is set in the future, and this will cause e2fsck
3285          * to complain and force a full file system check.
3286          */
3287         if (!(sb->s_flags & MS_RDONLY))
3288                 es->s_wtime = cpu_to_le32(get_seconds());
3289         es->s_kbytes_written =
3290                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3291                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3292                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3293         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3294                                         &EXT4_SB(sb)->s_freeblocks_counter));
3295         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3296                                         &EXT4_SB(sb)->s_freeinodes_counter));
3297         sb->s_dirt = 0;
3298         BUFFER_TRACE(sbh, "marking dirty");
3299         mark_buffer_dirty(sbh);
3300         if (sync) {
3301                 error = sync_dirty_buffer(sbh);
3302                 if (error)
3303                         return error;
3304
3305                 error = buffer_write_io_error(sbh);
3306                 if (error) {
3307                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3308                                "superblock");
3309                         clear_buffer_write_io_error(sbh);
3310                         set_buffer_uptodate(sbh);
3311                 }
3312         }
3313         return error;
3314 }
3315
3316 /*
3317  * Have we just finished recovery?  If so, and if we are mounting (or
3318  * remounting) the filesystem readonly, then we will end up with a
3319  * consistent fs on disk.  Record that fact.
3320  */
3321 static void ext4_mark_recovery_complete(struct super_block *sb,
3322                                         struct ext4_super_block *es)
3323 {
3324         journal_t *journal = EXT4_SB(sb)->s_journal;
3325
3326         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3327                 BUG_ON(journal != NULL);
3328                 return;
3329         }
3330         jbd2_journal_lock_updates(journal);
3331         if (jbd2_journal_flush(journal) < 0)
3332                 goto out;
3333
3334         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3335             sb->s_flags & MS_RDONLY) {
3336                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3337                 ext4_commit_super(sb, 1);
3338         }
3339
3340 out:
3341         jbd2_journal_unlock_updates(journal);
3342 }
3343
3344 /*
3345  * If we are mounting (or read-write remounting) a filesystem whose journal
3346  * has recorded an error from a previous lifetime, move that error to the
3347  * main filesystem now.
3348  */
3349 static void ext4_clear_journal_err(struct super_block *sb,
3350                                    struct ext4_super_block *es)
3351 {
3352         journal_t *journal;
3353         int j_errno;
3354         const char *errstr;
3355
3356         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3357
3358         journal = EXT4_SB(sb)->s_journal;
3359
3360         /*
3361          * Now check for any error status which may have been recorded in the
3362          * journal by a prior ext4_error() or ext4_abort()
3363          */
3364
3365         j_errno = jbd2_journal_errno(journal);
3366         if (j_errno) {
3367                 char nbuf[16];
3368
3369                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3370                 ext4_warning(sb, __func__, "Filesystem error recorded "
3371                              "from previous mount: %s", errstr);
3372                 ext4_warning(sb, __func__, "Marking fs in need of "
3373                              "filesystem check.");
3374
3375                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3376                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3377                 ext4_commit_super(sb, 1);
3378
3379                 jbd2_journal_clear_err(journal);
3380         }
3381 }
3382
3383 /*
3384  * Force the running and committing transactions to commit,
3385  * and wait on the commit.
3386  */
3387 int ext4_force_commit(struct super_block *sb)
3388 {
3389         journal_t *journal;
3390         int ret = 0;
3391
3392         if (sb->s_flags & MS_RDONLY)
3393                 return 0;
3394
3395         journal = EXT4_SB(sb)->s_journal;
3396         if (journal)
3397                 ret = ext4_journal_force_commit(journal);
3398
3399         return ret;
3400 }
3401
3402 static void ext4_write_super(struct super_block *sb)
3403 {
3404         lock_super(sb);
3405         ext4_commit_super(sb, 1);
3406         unlock_super(sb);
3407 }
3408
3409 static int ext4_sync_fs(struct super_block *sb, int wait)
3410 {
3411         int ret = 0;
3412         tid_t target;
3413         struct ext4_sb_info *sbi = EXT4_SB(sb);
3414
3415         trace_ext4_sync_fs(sb, wait);
3416         flush_workqueue(sbi->dio_unwritten_wq);
3417         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3418                 if (wait)
3419                         jbd2_log_wait_commit(sbi->s_journal, target);
3420         }
3421         return ret;
3422 }
3423
3424 /*
3425  * LVM calls this function before a (read-only) snapshot is created.  This
3426  * gives us a chance to flush the journal completely and mark the fs clean.
3427  */
3428 static int ext4_freeze(struct super_block *sb)
3429 {
3430         int error = 0;
3431         journal_t *journal;
3432
3433         if (sb->s_flags & MS_RDONLY)
3434                 return 0;
3435
3436         journal = EXT4_SB(sb)->s_journal;
3437
3438         /* Now we set up the journal barrier. */
3439         jbd2_journal_lock_updates(journal);
3440
3441         /*
3442          * Don't clear the needs_recovery flag if we failed to flush
3443          * the journal.
3444          */
3445         error = jbd2_journal_flush(journal);
3446         if (error < 0) {
3447         out:
3448                 jbd2_journal_unlock_updates(journal);
3449                 return error;
3450         }
3451
3452         /* Journal blocked and flushed, clear needs_recovery flag. */
3453         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3454         error = ext4_commit_super(sb, 1);
3455         if (error)
3456                 goto out;
3457         return 0;
3458 }
3459
3460 /*
3461  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3462  * flag here, even though the filesystem is not technically dirty yet.
3463  */
3464 static int ext4_unfreeze(struct super_block *sb)
3465 {
3466         if (sb->s_flags & MS_RDONLY)
3467                 return 0;
3468
3469         lock_super(sb);
3470         /* Reset the needs_recovery flag before the fs is unlocked. */
3471         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3472         ext4_commit_super(sb, 1);
3473         unlock_super(sb);
3474         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3475         return 0;
3476 }
3477
3478 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3479 {
3480         struct ext4_super_block *es;
3481         struct ext4_sb_info *sbi = EXT4_SB(sb);
3482         ext4_fsblk_t n_blocks_count = 0;
3483         unsigned long old_sb_flags;
3484         struct ext4_mount_options old_opts;
3485         ext4_group_t g;
3486         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3487         int err;
3488 #ifdef CONFIG_QUOTA
3489         int i;
3490 #endif
3491
3492         lock_kernel();
3493
3494         /* Store the original options */
3495         lock_super(sb);
3496         old_sb_flags = sb->s_flags;
3497         old_opts.s_mount_opt = sbi->s_mount_opt;
3498         old_opts.s_resuid = sbi->s_resuid;
3499         old_opts.s_resgid = sbi->s_resgid;
3500         old_opts.s_commit_interval = sbi->s_commit_interval;
3501         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3502         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3503 #ifdef CONFIG_QUOTA
3504         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3505         for (i = 0; i < MAXQUOTAS; i++)
3506                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3507 #endif
3508         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3509                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3510
3511         /*
3512          * Allow the "check" option to be passed as a remount option.
3513          */
3514         if (!parse_options(data, sb, NULL, &journal_ioprio,
3515                            &n_blocks_count, 1)) {
3516                 err = -EINVAL;
3517                 goto restore_opts;
3518         }
3519
3520         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3521                 ext4_abort(sb, __func__, "Abort forced by user");
3522
3523         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3524                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3525
3526         es = sbi->s_es;
3527
3528         if (sbi->s_journal) {
3529                 ext4_init_journal_params(sb, sbi->s_journal);
3530                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3531         }
3532
3533         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3534                 n_blocks_count > ext4_blocks_count(es)) {
3535                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3536                         err = -EROFS;
3537                         goto restore_opts;
3538                 }
3539
3540                 if (*flags & MS_RDONLY) {
3541                         /*
3542                          * First of all, the unconditional stuff we have to do
3543                          * to disable replay of the journal when we next remount
3544                          */
3545                         sb->s_flags |= MS_RDONLY;
3546
3547                         /*
3548                          * OK, test if we are remounting a valid rw partition
3549                          * readonly, and if so set the rdonly flag and then
3550                          * mark the partition as valid again.
3551                          */
3552                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3553                             (sbi->s_mount_state & EXT4_VALID_FS))
3554                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3555
3556                         if (sbi->s_journal)
3557                                 ext4_mark_recovery_complete(sb, es);
3558                 } else {
3559                         /* Make sure we can mount this feature set readwrite */
3560                         if (!ext4_feature_set_ok(sb, 0)) {
3561                                 err = -EROFS;
3562                                 goto restore_opts;
3563                         }
3564                         /*
3565                          * Make sure the group descriptor checksums
3566                          * are sane.  If they aren't, refuse to remount r/w.
3567                          */
3568                         for (g = 0; g < sbi->s_groups_count; g++) {
3569                                 struct ext4_group_desc *gdp =
3570                                         ext4_get_group_desc(sb, g, NULL);
3571
3572                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3573                                         ext4_msg(sb, KERN_ERR,
3574                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3575                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3576                                                le16_to_cpu(gdp->bg_checksum));
3577                                         err = -EINVAL;
3578                                         goto restore_opts;
3579                                 }
3580                         }
3581
3582                         /*
3583                          * If we have an unprocessed orphan list hanging
3584                          * around from a previously readonly bdev mount,
3585                          * require a full umount/remount for now.
3586                          */
3587                         if (es->s_last_orphan) {
3588                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3589                                        "remount RDWR because of unprocessed "
3590                                        "orphan inode list.  Please "
3591                                        "umount/remount instead");
3592                                 err = -EINVAL;
3593                                 goto restore_opts;
3594                         }
3595
3596                         /*
3597                          * Mounting a RDONLY partition read-write, so reread
3598                          * and store the current valid flag.  (It may have
3599                          * been changed by e2fsck since we originally mounted
3600                          * the partition.)
3601                          */
3602                         if (sbi->s_journal)
3603                                 ext4_clear_journal_err(sb, es);
3604                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3605                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3606                                 goto restore_opts;
3607                         if (!ext4_setup_super(sb, es, 0))
3608                                 sb->s_flags &= ~MS_RDONLY;
3609                 }
3610         }
3611         ext4_setup_system_zone(sb);
3612         if (sbi->s_journal == NULL)
3613                 ext4_commit_super(sb, 1);
3614
3615 #ifdef CONFIG_QUOTA
3616         /* Release old quota file names */
3617         for (i = 0; i < MAXQUOTAS; i++)
3618                 if (old_opts.s_qf_names[i] &&
3619                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3620                         kfree(old_opts.s_qf_names[i]);
3621 #endif
3622         unlock_super(sb);
3623         unlock_kernel();
3624         return 0;
3625
3626 restore_opts:
3627         sb->s_flags = old_sb_flags;
3628         sbi->s_mount_opt = old_opts.s_mount_opt;
3629         sbi->s_resuid = old_opts.s_resuid;
3630         sbi->s_resgid = old_opts.s_resgid;
3631         sbi->s_commit_interval = old_opts.s_commit_interval;
3632         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3633         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3634 #ifdef CONFIG_QUOTA
3635         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3636         for (i = 0; i < MAXQUOTAS; i++) {
3637                 if (sbi->s_qf_names[i] &&
3638                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3639                         kfree(sbi->s_qf_names[i]);
3640                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3641         }
3642 #endif
3643         unlock_super(sb);
3644         unlock_kernel();
3645         return err;
3646 }
3647
3648 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3649 {
3650         struct super_block *sb = dentry->d_sb;
3651         struct ext4_sb_info *sbi = EXT4_SB(sb);
3652         struct ext4_super_block *es = sbi->s_es;
3653         u64 fsid;
3654
3655         if (test_opt(sb, MINIX_DF)) {
3656                 sbi->s_overhead_last = 0;
3657         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3658                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3659                 ext4_fsblk_t overhead = 0;
3660
3661                 /*
3662                  * Compute the overhead (FS structures).  This is constant
3663                  * for a given filesystem unless the number of block groups
3664                  * changes so we cache the previous value until it does.
3665                  */
3666
3667                 /*
3668                  * All of the blocks before first_data_block are
3669                  * overhead
3670                  */
3671                 overhead = le32_to_cpu(es->s_first_data_block);
3672
3673                 /*
3674                  * Add the overhead attributed to the superblock and
3675                  * block group descriptors.  If the sparse superblocks
3676                  * feature is turned on, then not all groups have this.
3677                  */
3678                 for (i = 0; i < ngroups; i++) {
3679                         overhead += ext4_bg_has_super(sb, i) +
3680                                 ext4_bg_num_gdb(sb, i);
3681                         cond_resched();
3682                 }
3683
3684                 /*
3685                  * Every block group has an inode bitmap, a block
3686                  * bitmap, and an inode table.
3687                  */
3688                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3689                 sbi->s_overhead_last = overhead;
3690                 smp_wmb();
3691                 sbi->s_blocks_last = ext4_blocks_count(es);
3692         }
3693
3694         buf->f_type = EXT4_SUPER_MAGIC;
3695         buf->f_bsize = sb->s_blocksize;
3696         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3697         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3698                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3699         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3700         if (buf->f_bfree < ext4_r_blocks_count(es))
3701                 buf->f_bavail = 0;
3702         buf->f_files = le32_to_cpu(es->s_inodes_count);
3703         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3704         buf->f_namelen = EXT4_NAME_LEN;
3705         fsid = le64_to_cpup((void *)es->s_uuid) ^
3706                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3707         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3708         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3709
3710         return 0;
3711 }
3712
3713 /* Helper function for writing quotas on sync - we need to start transaction
3714  * before quota file is locked for write. Otherwise the are possible deadlocks:
3715  * Process 1                         Process 2
3716  * ext4_create()                     quota_sync()
3717  *   jbd2_journal_start()                  write_dquot()
3718  *   vfs_dq_init()                         down(dqio_mutex)
3719  *     down(dqio_mutex)                    jbd2_journal_start()
3720  *
3721  */
3722
3723 #ifdef CONFIG_QUOTA
3724
3725 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3726 {
3727         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3728 }
3729
3730 static int ext4_write_dquot(struct dquot *dquot)
3731 {
3732         int ret, err;
3733         handle_t *handle;
3734         struct inode *inode;
3735
3736         inode = dquot_to_inode(dquot);
3737         handle = ext4_journal_start(inode,
3738                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3739         if (IS_ERR(handle))
3740                 return PTR_ERR(handle);
3741         ret = dquot_commit(dquot);
3742         err = ext4_journal_stop(handle);
3743         if (!ret)
3744                 ret = err;
3745         return ret;
3746 }
3747
3748 static int ext4_acquire_dquot(struct dquot *dquot)
3749 {
3750         int ret, err;
3751         handle_t *handle;
3752
3753         handle = ext4_journal_start(dquot_to_inode(dquot),
3754                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3755         if (IS_ERR(handle))
3756                 return PTR_ERR(handle);
3757         ret = dquot_acquire(dquot);
3758         err = ext4_journal_stop(handle);
3759         if (!ret)
3760                 ret = err;
3761         return ret;
3762 }
3763
3764 static int ext4_release_dquot(struct dquot *dquot)
3765 {
3766         int ret, err;
3767         handle_t *handle;
3768
3769         handle = ext4_journal_start(dquot_to_inode(dquot),
3770                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3771         if (IS_ERR(handle)) {
3772                 /* Release dquot anyway to avoid endless cycle in dqput() */
3773                 dquot_release(dquot);
3774                 return PTR_ERR(handle);
3775         }
3776         ret = dquot_release(dquot);
3777         err = ext4_journal_stop(handle);
3778         if (!ret)
3779                 ret = err;
3780         return ret;
3781 }
3782
3783 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3784 {
3785         /* Are we journaling quotas? */
3786         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3787             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3788                 dquot_mark_dquot_dirty(dquot);
3789                 return ext4_write_dquot(dquot);
3790         } else {
3791                 return dquot_mark_dquot_dirty(dquot);
3792         }
3793 }
3794
3795 static int ext4_write_info(struct super_block *sb, int type)
3796 {
3797         int ret, err;
3798         handle_t *handle;
3799
3800         /* Data block + inode block */
3801         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3802         if (IS_ERR(handle))
3803                 return PTR_ERR(handle);
3804         ret = dquot_commit_info(sb, type);
3805         err = ext4_journal_stop(handle);
3806         if (!ret)
3807                 ret = err;
3808         return ret;
3809 }
3810
3811 /*
3812  * Turn on quotas during mount time - we need to find
3813  * the quota file and such...
3814  */
3815 static int ext4_quota_on_mount(struct super_block *sb, int type)
3816 {
3817         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3818                                   EXT4_SB(sb)->s_jquota_fmt, type);
3819 }
3820
3821 /*
3822  * Standard function to be called on quota_on
3823  */
3824 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3825                          char *name, int remount)
3826 {
3827         int err;
3828         struct path path;
3829
3830         if (!test_opt(sb, QUOTA))
3831                 return -EINVAL;
3832         /* When remounting, no checks are needed and in fact, name is NULL */
3833         if (remount)
3834                 return vfs_quota_on(sb, type, format_id, name, remount);
3835
3836         err = kern_path(name, LOOKUP_FOLLOW, &path);
3837         if (err)
3838                 return err;
3839
3840         /* Quotafile not on the same filesystem? */
3841         if (path.mnt->mnt_sb != sb) {
3842                 path_put(&path);
3843                 return -EXDEV;
3844         }
3845         /* Journaling quota? */
3846         if (EXT4_SB(sb)->s_qf_names[type]) {
3847                 /* Quotafile not in fs root? */
3848                 if (path.dentry->d_parent != sb->s_root)
3849                         ext4_msg(sb, KERN_WARNING,
3850                                 "Quota file not on filesystem root. "
3851                                 "Journaled quota will not work");
3852         }
3853
3854         /*
3855          * When we journal data on quota file, we have to flush journal to see
3856          * all updates to the file when we bypass pagecache...
3857          */
3858         if (EXT4_SB(sb)->s_journal &&
3859             ext4_should_journal_data(path.dentry->d_inode)) {
3860                 /*
3861                  * We don't need to lock updates but journal_flush() could
3862                  * otherwise be livelocked...
3863                  */
3864                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3865                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3866                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3867                 if (err) {
3868                         path_put(&path);
3869                         return err;
3870                 }
3871         }
3872
3873         err = vfs_quota_on_path(sb, type, format_id, &path);
3874         path_put(&path);
3875         return err;
3876 }
3877
3878 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3879  * acquiring the locks... As quota files are never truncated and quota code
3880  * itself serializes the operations (and noone else should touch the files)
3881  * we don't have to be afraid of races */
3882 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3883                                size_t len, loff_t off)
3884 {
3885         struct inode *inode = sb_dqopt(sb)->files[type];
3886         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3887         int err = 0;
3888         int offset = off & (sb->s_blocksize - 1);
3889         int tocopy;
3890         size_t toread;
3891         struct buffer_head *bh;
3892         loff_t i_size = i_size_read(inode);
3893
3894         if (off > i_size)
3895                 return 0;
3896         if (off+len > i_size)
3897                 len = i_size-off;
3898         toread = len;
3899         while (toread > 0) {
3900                 tocopy = sb->s_blocksize - offset < toread ?
3901                                 sb->s_blocksize - offset : toread;
3902                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3903                 if (err)
3904                         return err;
3905                 if (!bh)        /* A hole? */
3906                         memset(data, 0, tocopy);
3907                 else
3908                         memcpy(data, bh->b_data+offset, tocopy);
3909                 brelse(bh);
3910                 offset = 0;
3911                 toread -= tocopy;
3912                 data += tocopy;
3913                 blk++;
3914         }
3915         return len;
3916 }
3917
3918 /* Write to quotafile (we know the transaction is already started and has
3919  * enough credits) */
3920 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3921                                 const char *data, size_t len, loff_t off)
3922 {
3923         struct inode *inode = sb_dqopt(sb)->files[type];
3924         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3925         int err = 0;
3926         int offset = off & (sb->s_blocksize - 1);
3927         int tocopy;
3928         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3929         size_t towrite = len;
3930         struct buffer_head *bh;
3931         handle_t *handle = journal_current_handle();
3932
3933         if (EXT4_SB(sb)->s_journal && !handle) {
3934                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3935                         " cancelled because transaction is not started",
3936                         (unsigned long long)off, (unsigned long long)len);
3937                 return -EIO;
3938         }
3939         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3940         while (towrite > 0) {
3941                 tocopy = sb->s_blocksize - offset < towrite ?
3942                                 sb->s_blocksize - offset : towrite;
3943                 bh = ext4_bread(handle, inode, blk, 1, &err);
3944                 if (!bh)
3945                         goto out;
3946                 if (journal_quota) {
3947                         err = ext4_journal_get_write_access(handle, bh);
3948                         if (err) {
3949                                 brelse(bh);
3950                                 goto out;
3951                         }
3952                 }
3953                 lock_buffer(bh);
3954                 memcpy(bh->b_data+offset, data, tocopy);
3955                 flush_dcache_page(bh->b_page);
3956                 unlock_buffer(bh);
3957                 if (journal_quota)
3958                         err = ext4_handle_dirty_metadata(handle, NULL, bh);
3959                 else {
3960                         /* Always do at least ordered writes for quotas */
3961                         err = ext4_jbd2_file_inode(handle, inode);
3962                         mark_buffer_dirty(bh);
3963                 }
3964                 brelse(bh);
3965                 if (err)
3966                         goto out;
3967                 offset = 0;
3968                 towrite -= tocopy;
3969                 data += tocopy;
3970                 blk++;
3971         }
3972 out:
3973         if (len == towrite) {
3974                 mutex_unlock(&inode->i_mutex);
3975                 return err;
3976         }
3977         if (inode->i_size < off+len-towrite) {
3978                 i_size_write(inode, off+len-towrite);
3979                 EXT4_I(inode)->i_disksize = inode->i_size;
3980         }
3981         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3982         ext4_mark_inode_dirty(handle, inode);
3983         mutex_unlock(&inode->i_mutex);
3984         return len - towrite;
3985 }
3986
3987 #endif
3988
3989 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
3990                        const char *dev_name, void *data, struct vfsmount *mnt)
3991 {
3992         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
3993 }
3994
3995 #if !defined(CONTIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
3996 static struct file_system_type ext2_fs_type = {
3997         .owner          = THIS_MODULE,
3998         .name           = "ext2",
3999         .get_sb         = ext4_get_sb,
4000         .kill_sb        = kill_block_super,
4001         .fs_flags       = FS_REQUIRES_DEV,
4002 };
4003
4004 static inline void register_as_ext2(void)
4005 {
4006         int err = register_filesystem(&ext2_fs_type);
4007         if (err)
4008                 printk(KERN_WARNING
4009                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4010 }
4011
4012 static inline void unregister_as_ext2(void)
4013 {
4014         unregister_filesystem(&ext2_fs_type);
4015 }
4016 MODULE_ALIAS("ext2");
4017 #else
4018 static inline void register_as_ext2(void) { }
4019 static inline void unregister_as_ext2(void) { }
4020 #endif
4021
4022 #if !defined(CONTIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4023 static struct file_system_type ext3_fs_type = {
4024         .owner          = THIS_MODULE,
4025         .name           = "ext3",
4026         .get_sb         = ext4_get_sb,
4027         .kill_sb        = kill_block_super,
4028         .fs_flags       = FS_REQUIRES_DEV,
4029 };
4030
4031 static inline void register_as_ext3(void)
4032 {
4033         int err = register_filesystem(&ext3_fs_type);
4034         if (err)
4035                 printk(KERN_WARNING
4036                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4037 }
4038
4039 static inline void unregister_as_ext3(void)
4040 {
4041         unregister_filesystem(&ext3_fs_type);
4042 }
4043 MODULE_ALIAS("ext3");
4044 #else
4045 static inline void register_as_ext3(void) { }
4046 static inline void unregister_as_ext3(void) { }
4047 #endif
4048
4049 static struct file_system_type ext4_fs_type = {
4050         .owner          = THIS_MODULE,
4051         .name           = "ext4",
4052         .get_sb         = ext4_get_sb,
4053         .kill_sb        = kill_block_super,
4054         .fs_flags       = FS_REQUIRES_DEV,
4055 };
4056
4057 static int __init init_ext4_fs(void)
4058 {
4059         int err;
4060
4061         err = init_ext4_system_zone();
4062         if (err)
4063                 return err;
4064         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4065         if (!ext4_kset)
4066                 goto out4;
4067         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4068         err = init_ext4_mballoc();
4069         if (err)
4070                 goto out3;
4071
4072         err = init_ext4_xattr();
4073         if (err)
4074                 goto out2;
4075         err = init_inodecache();
4076         if (err)
4077                 goto out1;
4078         register_as_ext2();
4079         register_as_ext3();
4080         err = register_filesystem(&ext4_fs_type);
4081         if (err)
4082                 goto out;
4083         return 0;
4084 out:
4085         unregister_as_ext2();
4086         unregister_as_ext3();
4087         destroy_inodecache();
4088 out1:
4089         exit_ext4_xattr();
4090 out2:
4091         exit_ext4_mballoc();
4092 out3:
4093         remove_proc_entry("fs/ext4", NULL);
4094         kset_unregister(ext4_kset);
4095 out4:
4096         exit_ext4_system_zone();
4097         return err;
4098 }
4099
4100 static void __exit exit_ext4_fs(void)
4101 {
4102         unregister_as_ext2();
4103         unregister_as_ext3();
4104         unregister_filesystem(&ext4_fs_type);
4105         destroy_inodecache();
4106         exit_ext4_xattr();
4107         exit_ext4_mballoc();
4108         remove_proc_entry("fs/ext4", NULL);
4109         kset_unregister(ext4_kset);
4110         exit_ext4_system_zone();
4111 }
4112
4113 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4114 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4115 MODULE_LICENSE("GPL");
4116 module_init(init_ext4_fs)
4117 module_exit(exit_ext4_fs)