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         dquot_drop(inode);
 765         ext4_discard_preallocations(inode);
 766         if (EXT4_JOURNAL(inode))
 767                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 768                                        &EXT4_I(inode)->jinode);
 769 }
 770
 771 static inline void ext4_show_quota_options(struct seq_file *seq,
 772                                            struct super_block *sb)
 773 {
 774 #if defined(CONFIG_QUOTA)
 775         struct ext4_sb_info *sbi = EXT4_SB(sb);
 776
 777         if (sbi->s_jquota_fmt) {
 778                 char *fmtname = "";
 779
 780                 switch (sbi->s_jquota_fmt) {
 781                 case QFMT_VFS_OLD:
 782                         fmtname = "vfsold";
 783                         break;
 784                 case QFMT_VFS_V0:
 785                         fmtname = "vfsv0";
 786                         break;
 787                 case QFMT_VFS_V1:
 788                         fmtname = "vfsv1";
 789                         break;
 790                 }
 791                 seq_printf(seq, ",jqfmt=%s", fmtname);
 792         }
 793
 794         if (sbi->s_qf_names[USRQUOTA])
 795                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 796
 797         if (sbi->s_qf_names[GRPQUOTA])
 798                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 799
 800         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 801                 seq_puts(seq, ",usrquota");
 802
 803         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 804                 seq_puts(seq, ",grpquota");
 805 #endif
 806 }
 807
 808 /*
 809  * Show an option if
 810  *  - it's set to a non-default value OR
 811  *  - if the per-sb default is different from the global default
 812  */
 813 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 814 {
 815         int def_errors;
 816         unsigned long def_mount_opts;
 817         struct super_block *sb = vfs->mnt_sb;
 818         struct ext4_sb_info *sbi = EXT4_SB(sb);
 819         struct ext4_super_block *es = sbi->s_es;
 820
 821         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 822         def_errors     = le16_to_cpu(es->s_errors);
 823
 824         if (sbi->s_sb_block != 1)
 825                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 826         if (test_opt(sb, MINIX_DF))
 827                 seq_puts(seq, ",minixdf");
 828         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 829                 seq_puts(seq, ",grpid");
 830         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 831                 seq_puts(seq, ",nogrpid");
 832         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 833             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 834                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 835         }
 836         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 837             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 838                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 839         }
 840         if (test_opt(sb, ERRORS_RO)) {
 841                 if (def_errors == EXT4_ERRORS_PANIC ||
 842                     def_errors == EXT4_ERRORS_CONTINUE) {
 843                         seq_puts(seq, ",errors=remount-ro");
 844                 }
 845         }
 846         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 847                 seq_puts(seq, ",errors=continue");
 848         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 849                 seq_puts(seq, ",errors=panic");
 850         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 851                 seq_puts(seq, ",nouid32");
 852         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 853                 seq_puts(seq, ",debug");
 854         if (test_opt(sb, OLDALLOC))
 855                 seq_puts(seq, ",oldalloc");
 856 #ifdef CONFIG_EXT4_FS_XATTR
 857         if (test_opt(sb, XATTR_USER) &&
 858                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 859                 seq_puts(seq, ",user_xattr");
 860         if (!test_opt(sb, XATTR_USER) &&
 861             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 862                 seq_puts(seq, ",nouser_xattr");
 863         }
 864 #endif
 865 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 866         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 867                 seq_puts(seq, ",acl");
 868         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 869                 seq_puts(seq, ",noacl");
 870 #endif
 871         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 872                 seq_printf(seq, ",commit=%u",
 873                            (unsigned) (sbi->s_commit_interval / HZ));
 874         }
 875         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 876                 seq_printf(seq, ",min_batch_time=%u",
 877                            (unsigned) sbi->s_min_batch_time);
 878         }
 879         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 880                 seq_printf(seq, ",max_batch_time=%u",
 881                            (unsigned) sbi->s_min_batch_time);
 882         }
 883
 884         /*
 885          * We're changing the default of barrier mount option, so
 886          * let's always display its mount state so it's clear what its
 887          * status is.
 888          */
 889         seq_puts(seq, ",barrier=");
 890         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 891         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 892                 seq_puts(seq, ",journal_async_commit");
 893         if (test_opt(sb, NOBH))
 894                 seq_puts(seq, ",nobh");
 895         if (test_opt(sb, I_VERSION))
 896                 seq_puts(seq, ",i_version");
 897         if (!test_opt(sb, DELALLOC))
 898                 seq_puts(seq, ",nodelalloc");
 899
 900
 901         if (sbi->s_stripe)
 902                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 903         /*
 904          * journal mode get enabled in different ways
 905          * So just print the value even if we didn't specify it
 906          */
 907         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 908                 seq_puts(seq, ",data=journal");
 909         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 910                 seq_puts(seq, ",data=ordered");
 911         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 912                 seq_puts(seq, ",data=writeback");
 913
 914         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 915                 seq_printf(seq, ",inode_readahead_blks=%u",
 916                            sbi->s_inode_readahead_blks);
 917
 918         if (test_opt(sb, DATA_ERR_ABORT))
 919                 seq_puts(seq, ",data_err=abort");
 920
 921         if (test_opt(sb, NO_AUTO_DA_ALLOC))
 922                 seq_puts(seq, ",noauto_da_alloc");
 923
 924         if (test_opt(sb, DISCARD))
 925                 seq_puts(seq, ",discard");
 926
 927         if (test_opt(sb, NOLOAD))
 928                 seq_puts(seq, ",norecovery");
 929
 930         ext4_show_quota_options(seq, sb);
 931
 932         return 0;
 933 }
 934
 935 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 936                                         u64 ino, u32 generation)
 937 {
 938         struct inode *inode;
 939
 940         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 941                 return ERR_PTR(-ESTALE);
 942         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 943                 return ERR_PTR(-ESTALE);
 944
 945         /* iget isn't really right if the inode is currently unallocated!!
 946          *
 947          * ext4_read_inode will return a bad_inode if the inode had been
 948          * deleted, so we should be safe.
 949          *
 950          * Currently we don't know the generation for parent directory, so
 951          * a generation of 0 means "accept any"
 952          */
 953         inode = ext4_iget(sb, ino);
 954         if (IS_ERR(inode))
 955                 return ERR_CAST(inode);
 956         if (generation && inode->i_generation != generation) {
 957                 iput(inode);
 958                 return ERR_PTR(-ESTALE);
 959         }
 960
 961         return inode;
 962 }
 963
 964 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 965                                         int fh_len, int fh_type)
 966 {
 967         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 968                                     ext4_nfs_get_inode);
 969 }
 970
 971 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 972                                         int fh_len, int fh_type)
 973 {
 974         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 975                                     ext4_nfs_get_inode);
 976 }
 977
 978 /*
 979  * Try to release metadata pages (indirect blocks, directories) which are
 980  * mapped via the block device.  Since these pages could have journal heads
 981  * which would prevent try_to_free_buffers() from freeing them, we must use
 982  * jbd2 layer's try_to_free_buffers() function to release them.
 983  */
 984 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 985                                  gfp_t wait)
 986 {
 987         journal_t *journal = EXT4_SB(sb)->s_journal;
 988
 989         WARN_ON(PageChecked(page));
 990         if (!page_has_buffers(page))
 991                 return 0;
 992         if (journal)
 993                 return jbd2_journal_try_to_free_buffers(journal, page,
 994                                                         wait & ~__GFP_WAIT);
 995         return try_to_free_buffers(page);
 996 }
 997
 998 #ifdef CONFIG_QUOTA
 999 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1000 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1001
1002 static int ext4_write_dquot(struct dquot *dquot);
1003 static int ext4_acquire_dquot(struct dquot *dquot);
1004 static int ext4_release_dquot(struct dquot *dquot);
1005 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1006 static int ext4_write_info(struct super_block *sb, int type);
1007 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1008                                 char *path, int remount);
1009 static int ext4_quota_on_mount(struct super_block *sb, int type);
1010 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1011                                size_t len, loff_t off);
1012 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1013                                 const char *data, size_t len, loff_t off);
1014
1015 static const struct dquot_operations ext4_quota_operations = {
1016 #ifdef CONFIG_QUOTA
1017         .get_reserved_space = ext4_get_reserved_space,
1018 #endif
1019         .write_dquot    = ext4_write_dquot,
1020         .acquire_dquot  = ext4_acquire_dquot,
1021         .release_dquot  = ext4_release_dquot,
1022         .mark_dirty     = ext4_mark_dquot_dirty,
1023         .write_info     = ext4_write_info,
1024         .alloc_dquot    = dquot_alloc,
1025         .destroy_dquot  = dquot_destroy,
1026 };
1027
1028 static const struct quotactl_ops ext4_qctl_operations = {
1029         .quota_on       = ext4_quota_on,
1030         .quota_off      = vfs_quota_off,
1031         .quota_sync     = vfs_quota_sync,
1032         .get_info       = vfs_get_dqinfo,
1033         .set_info       = vfs_set_dqinfo,
1034         .get_dqblk      = vfs_get_dqblk,
1035         .set_dqblk      = vfs_set_dqblk
1036 };
1037 #endif
1038
1039 static const struct super_operations ext4_sops = {
1040         .alloc_inode    = ext4_alloc_inode,
1041         .destroy_inode  = ext4_destroy_inode,
1042         .write_inode    = ext4_write_inode,
1043         .dirty_inode    = ext4_dirty_inode,
1044         .delete_inode   = ext4_delete_inode,
1045         .put_super      = ext4_put_super,
1046         .sync_fs        = ext4_sync_fs,
1047         .freeze_fs      = ext4_freeze,
1048         .unfreeze_fs    = ext4_unfreeze,
1049         .statfs         = ext4_statfs,
1050         .remount_fs     = ext4_remount,
1051         .clear_inode    = ext4_clear_inode,
1052         .show_options   = ext4_show_options,
1053 #ifdef CONFIG_QUOTA
1054         .quota_read     = ext4_quota_read,
1055         .quota_write    = ext4_quota_write,
1056 #endif
1057         .bdev_try_to_free_page = bdev_try_to_free_page,
1058 };
1059
1060 static const struct super_operations ext4_nojournal_sops = {
1061         .alloc_inode    = ext4_alloc_inode,
1062         .destroy_inode  = ext4_destroy_inode,
1063         .write_inode    = ext4_write_inode,
1064         .dirty_inode    = ext4_dirty_inode,
1065         .delete_inode   = ext4_delete_inode,
1066         .write_super    = ext4_write_super,
1067         .put_super      = ext4_put_super,
1068         .statfs         = ext4_statfs,
1069         .remount_fs     = ext4_remount,
1070         .clear_inode    = ext4_clear_inode,
1071         .show_options   = ext4_show_options,
1072 #ifdef CONFIG_QUOTA
1073         .quota_read     = ext4_quota_read,
1074         .quota_write    = ext4_quota_write,
1075 #endif
1076         .bdev_try_to_free_page = bdev_try_to_free_page,
1077 };
1078
1079 static const struct export_operations ext4_export_ops = {
1080         .fh_to_dentry = ext4_fh_to_dentry,
1081         .fh_to_parent = ext4_fh_to_parent,
1082         .get_parent = ext4_get_parent,
1083 };
1084
1085 enum {
1086         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1087         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1088         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1089         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1090         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1091         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1092         Opt_journal_update, Opt_journal_dev,
1093         Opt_journal_checksum, Opt_journal_async_commit,
1094         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1095         Opt_data_err_abort, Opt_data_err_ignore,
1096         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1097         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1098         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1099         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1100         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1101         Opt_block_validity, Opt_noblock_validity,
1102         Opt_inode_readahead_blks, Opt_journal_ioprio,
1103         Opt_discard, Opt_nodiscard,
1104 };
1105
1106 static const match_table_t tokens = {
1107         {Opt_bsd_df, "bsddf"},
1108         {Opt_minix_df, "minixdf"},
1109         {Opt_grpid, "grpid"},
1110         {Opt_grpid, "bsdgroups"},
1111         {Opt_nogrpid, "nogrpid"},
1112         {Opt_nogrpid, "sysvgroups"},
1113         {Opt_resgid, "resgid=%u"},
1114         {Opt_resuid, "resuid=%u"},
1115         {Opt_sb, "sb=%u"},
1116         {Opt_err_cont, "errors=continue"},
1117         {Opt_err_panic, "errors=panic"},
1118         {Opt_err_ro, "errors=remount-ro"},
1119         {Opt_nouid32, "nouid32"},
1120         {Opt_debug, "debug"},
1121         {Opt_oldalloc, "oldalloc"},
1122         {Opt_orlov, "orlov"},
1123         {Opt_user_xattr, "user_xattr"},
1124         {Opt_nouser_xattr, "nouser_xattr"},
1125         {Opt_acl, "acl"},
1126         {Opt_noacl, "noacl"},
1127         {Opt_noload, "noload"},
1128         {Opt_noload, "norecovery"},
1129         {Opt_nobh, "nobh"},
1130         {Opt_bh, "bh"},
1131         {Opt_commit, "commit=%u"},
1132         {Opt_min_batch_time, "min_batch_time=%u"},
1133         {Opt_max_batch_time, "max_batch_time=%u"},
1134         {Opt_journal_update, "journal=update"},
1135         {Opt_journal_dev, "journal_dev=%u"},
1136         {Opt_journal_checksum, "journal_checksum"},
1137         {Opt_journal_async_commit, "journal_async_commit"},
1138         {Opt_abort, "abort"},
1139         {Opt_data_journal, "data=journal"},
1140         {Opt_data_ordered, "data=ordered"},
1141         {Opt_data_writeback, "data=writeback"},
1142         {Opt_data_err_abort, "data_err=abort"},
1143         {Opt_data_err_ignore, "data_err=ignore"},
1144         {Opt_offusrjquota, "usrjquota="},
1145         {Opt_usrjquota, "usrjquota=%s"},
1146         {Opt_offgrpjquota, "grpjquota="},
1147         {Opt_grpjquota, "grpjquota=%s"},
1148         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1149         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1150         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1151         {Opt_grpquota, "grpquota"},
1152         {Opt_noquota, "noquota"},
1153         {Opt_quota, "quota"},
1154         {Opt_usrquota, "usrquota"},
1155         {Opt_barrier, "barrier=%u"},
1156         {Opt_barrier, "barrier"},
1157         {Opt_nobarrier, "nobarrier"},
1158         {Opt_i_version, "i_version"},
1159         {Opt_stripe, "stripe=%u"},
1160         {Opt_resize, "resize"},
1161         {Opt_delalloc, "delalloc"},
1162         {Opt_nodelalloc, "nodelalloc"},
1163         {Opt_block_validity, "block_validity"},
1164         {Opt_noblock_validity, "noblock_validity"},
1165         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1166         {Opt_journal_ioprio, "journal_ioprio=%u"},
1167         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1168         {Opt_auto_da_alloc, "auto_da_alloc"},
1169         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1170         {Opt_discard, "discard"},
1171         {Opt_nodiscard, "nodiscard"},
1172         {Opt_err, NULL},
1173 };
1174
1175 static ext4_fsblk_t get_sb_block(void **data)
1176 {
1177         ext4_fsblk_t    sb_block;
1178         char            *options = (char *) *data;
1179
1180         if (!options || strncmp(options, "sb=", 3) != 0)
1181                 return 1;       /* Default location */
1182
1183         options += 3;
1184         /* TODO: use simple_strtoll with >32bit ext4 */
1185         sb_block = simple_strtoul(options, &options, 0);
1186         if (*options && *options != ',') {
1187                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1188                        (char *) *data);
1189                 return 1;
1190         }
1191         if (*options == ',')
1192                 options++;
1193         *data = (void *) options;
1194
1195         return sb_block;
1196 }
1197
1198 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1199
1200 static int parse_options(char *options, struct super_block *sb,
1201                          unsigned long *journal_devnum,
1202                          unsigned int *journal_ioprio,
1203                          ext4_fsblk_t *n_blocks_count, int is_remount)
1204 {
1205         struct ext4_sb_info *sbi = EXT4_SB(sb);
1206         char *p;
1207         substring_t args[MAX_OPT_ARGS];
1208         int data_opt = 0;
1209         int option;
1210 #ifdef CONFIG_QUOTA
1211         int qtype, qfmt;
1212         char *qname;
1213 #endif
1214
1215         if (!options)
1216                 return 1;
1217
1218         while ((p = strsep(&options, ",")) != NULL) {
1219                 int token;
1220                 if (!*p)
1221                         continue;
1222
1223                 token = match_token(p, tokens, args);
1224                 switch (token) {
1225                 case Opt_bsd_df:
1226                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1227                         break;
1228                 case Opt_minix_df:
1229                         set_opt(sbi->s_mount_opt, MINIX_DF);
1230                         break;
1231                 case Opt_grpid:
1232                         set_opt(sbi->s_mount_opt, GRPID);
1233                         break;
1234                 case Opt_nogrpid:
1235                         clear_opt(sbi->s_mount_opt, GRPID);
1236                         break;
1237                 case Opt_resuid:
1238                         if (match_int(&args[0], &option))
1239                                 return 0;
1240                         sbi->s_resuid = option;
1241                         break;
1242                 case Opt_resgid:
1243                         if (match_int(&args[0], &option))
1244                                 return 0;
1245                         sbi->s_resgid = option;
1246                         break;
1247                 case Opt_sb:
1248                         /* handled by get_sb_block() instead of here */
1249                         /* *sb_block = match_int(&args[0]); */
1250                         break;
1251                 case Opt_err_panic:
1252                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1253                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1254                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1255                         break;
1256                 case Opt_err_ro:
1257                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1258                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1259                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1260                         break;
1261                 case Opt_err_cont:
1262                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1263                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1264                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1265                         break;
1266                 case Opt_nouid32:
1267                         set_opt(sbi->s_mount_opt, NO_UID32);
1268                         break;
1269                 case Opt_debug:
1270                         set_opt(sbi->s_mount_opt, DEBUG);
1271                         break;
1272                 case Opt_oldalloc:
1273                         set_opt(sbi->s_mount_opt, OLDALLOC);
1274                         break;
1275                 case Opt_orlov:
1276                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1277                         break;
1278 #ifdef CONFIG_EXT4_FS_XATTR
1279                 case Opt_user_xattr:
1280                         set_opt(sbi->s_mount_opt, XATTR_USER);
1281                         break;
1282                 case Opt_nouser_xattr:
1283                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1284                         break;
1285 #else
1286                 case Opt_user_xattr:
1287                 case Opt_nouser_xattr:
1288                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1289                         break;
1290 #endif
1291 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1292                 case Opt_acl:
1293                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1294                         break;
1295                 case Opt_noacl:
1296                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1297                         break;
1298 #else
1299                 case Opt_acl:
1300                 case Opt_noacl:
1301                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1302                         break;
1303 #endif
1304                 case Opt_journal_update:
1305                         /* @@@ FIXME */
1306                         /* Eventually we will want to be able to create
1307                            a journal file here.  For now, only allow the
1308                            user to specify an existing inode to be the
1309                            journal file. */
1310                         if (is_remount) {
1311                                 ext4_msg(sb, KERN_ERR,
1312                                          "Cannot specify journal on remount");
1313                                 return 0;
1314                         }
1315                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1316                         break;
1317                 case Opt_journal_dev:
1318                         if (is_remount) {
1319                                 ext4_msg(sb, KERN_ERR,
1320                                         "Cannot specify journal on remount");
1321                                 return 0;
1322                         }
1323                         if (match_int(&args[0], &option))
1324                                 return 0;
1325                         *journal_devnum = option;
1326                         break;
1327                 case Opt_journal_checksum:
1328                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1329                         break;
1330                 case Opt_journal_async_commit:
1331                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1332                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1333                         break;
1334                 case Opt_noload:
1335                         set_opt(sbi->s_mount_opt, NOLOAD);
1336                         break;
1337                 case Opt_commit:
1338                         if (match_int(&args[0], &option))
1339                                 return 0;
1340                         if (option < 0)
1341                                 return 0;
1342                         if (option == 0)
1343                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1344                         sbi->s_commit_interval = HZ * option;
1345                         break;
1346                 case Opt_max_batch_time:
1347                         if (match_int(&args[0], &option))
1348                                 return 0;
1349                         if (option < 0)
1350                                 return 0;
1351                         if (option == 0)
1352                                 option = EXT4_DEF_MAX_BATCH_TIME;
1353                         sbi->s_max_batch_time = option;
1354                         break;
1355                 case Opt_min_batch_time:
1356                         if (match_int(&args[0], &option))
1357                                 return 0;
1358                         if (option < 0)
1359                                 return 0;
1360                         sbi->s_min_batch_time = option;
1361                         break;
1362                 case Opt_data_journal:
1363                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1364                         goto datacheck;
1365                 case Opt_data_ordered:
1366                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1367                         goto datacheck;
1368                 case Opt_data_writeback:
1369                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1370                 datacheck:
1371                         if (is_remount) {
1372                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1373                                                 != data_opt) {
1374                                         ext4_msg(sb, KERN_ERR,
1375                                                 "Cannot change data mode on remount");
1376                                         return 0;
1377                                 }
1378                         } else {
1379                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1380                                 sbi->s_mount_opt |= data_opt;
1381                         }
1382                         break;
1383                 case Opt_data_err_abort:
1384                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1385                         break;
1386                 case Opt_data_err_ignore:
1387                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1388                         break;
1389 #ifdef CONFIG_QUOTA
1390                 case Opt_usrjquota:
1391                         qtype = USRQUOTA;
1392                         goto set_qf_name;
1393                 case Opt_grpjquota:
1394                         qtype = GRPQUOTA;
1395 set_qf_name:
1396                         if (sb_any_quota_loaded(sb) &&
1397                             !sbi->s_qf_names[qtype]) {
1398                                 ext4_msg(sb, KERN_ERR,
1399                                        "Cannot change journaled "
1400                                        "quota options when quota turned on");
1401                                 return 0;
1402                         }
1403                         qname = match_strdup(&args[0]);
1404                         if (!qname) {
1405                                 ext4_msg(sb, KERN_ERR,
1406                                         "Not enough memory for "
1407                                         "storing quotafile name");
1408                                 return 0;
1409                         }
1410                         if (sbi->s_qf_names[qtype] &&
1411                             strcmp(sbi->s_qf_names[qtype], qname)) {
1412                                 ext4_msg(sb, KERN_ERR,
1413                                         "%s quota file already "
1414                                         "specified", QTYPE2NAME(qtype));
1415                                 kfree(qname);
1416                                 return 0;
1417                         }
1418                         sbi->s_qf_names[qtype] = qname;
1419                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1420                                 ext4_msg(sb, KERN_ERR,
1421                                         "quotafile must be on "
1422                                         "filesystem root");
1423                                 kfree(sbi->s_qf_names[qtype]);
1424                                 sbi->s_qf_names[qtype] = NULL;
1425                                 return 0;
1426                         }
1427                         set_opt(sbi->s_mount_opt, QUOTA);
1428                         break;
1429                 case Opt_offusrjquota:
1430                         qtype = USRQUOTA;
1431                         goto clear_qf_name;
1432                 case Opt_offgrpjquota:
1433                         qtype = GRPQUOTA;
1434 clear_qf_name:
1435                         if (sb_any_quota_loaded(sb) &&
1436                             sbi->s_qf_names[qtype]) {
1437                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1438                                         "journaled quota options when "
1439                                         "quota turned on");
1440                                 return 0;
1441                         }
1442                         /*
1443                          * The space will be released later when all options
1444                          * are confirmed to be correct
1445                          */
1446                         sbi->s_qf_names[qtype] = NULL;
1447                         break;
1448                 case Opt_jqfmt_vfsold:
1449                         qfmt = QFMT_VFS_OLD;
1450                         goto set_qf_format;
1451                 case Opt_jqfmt_vfsv0:
1452                         qfmt = QFMT_VFS_V0;
1453                         goto set_qf_format;
1454                 case Opt_jqfmt_vfsv1:
1455                         qfmt = QFMT_VFS_V1;
1456 set_qf_format:
1457                         if (sb_any_quota_loaded(sb) &&
1458                             sbi->s_jquota_fmt != qfmt) {
1459                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1460                                         "journaled quota options when "
1461                                         "quota turned on");
1462                                 return 0;
1463                         }
1464                         sbi->s_jquota_fmt = qfmt;
1465                         break;
1466                 case Opt_quota:
1467                 case Opt_usrquota:
1468                         set_opt(sbi->s_mount_opt, QUOTA);
1469                         set_opt(sbi->s_mount_opt, USRQUOTA);
1470                         break;
1471                 case Opt_grpquota:
1472                         set_opt(sbi->s_mount_opt, QUOTA);
1473                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1474                         break;
1475                 case Opt_noquota:
1476                         if (sb_any_quota_loaded(sb)) {
1477                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1478                                         "options when quota turned on");
1479                                 return 0;
1480                         }
1481                         clear_opt(sbi->s_mount_opt, QUOTA);
1482                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1483                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1484                         break;
1485 #else
1486                 case Opt_quota:
1487                 case Opt_usrquota:
1488                 case Opt_grpquota:
1489                         ext4_msg(sb, KERN_ERR,
1490                                 "quota options not supported");
1491                         break;
1492                 case Opt_usrjquota:
1493                 case Opt_grpjquota:
1494                 case Opt_offusrjquota:
1495                 case Opt_offgrpjquota:
1496                 case Opt_jqfmt_vfsold:
1497                 case Opt_jqfmt_vfsv0:
1498                 case Opt_jqfmt_vfsv1:
1499                         ext4_msg(sb, KERN_ERR,
1500                                 "journaled quota options not supported");
1501                         break;
1502                 case Opt_noquota:
1503                         break;
1504 #endif
1505                 case Opt_abort:
1506                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1507                         break;
1508                 case Opt_nobarrier:
1509                         clear_opt(sbi->s_mount_opt, BARRIER);
1510                         break;
1511                 case Opt_barrier:
1512                         if (match_int(&args[0], &option)) {
1513                                 set_opt(sbi->s_mount_opt, BARRIER);
1514                                 break;
1515                         }
1516                         if (option)
1517                                 set_opt(sbi->s_mount_opt, BARRIER);
1518                         else
1519                                 clear_opt(sbi->s_mount_opt, BARRIER);
1520                         break;
1521                 case Opt_ignore:
1522                         break;
1523                 case Opt_resize:
1524                         if (!is_remount) {
1525                                 ext4_msg(sb, KERN_ERR,
1526                                         "resize option only available "
1527                                         "for remount");
1528                                 return 0;
1529                         }
1530                         if (match_int(&args[0], &option) != 0)
1531                                 return 0;
1532                         *n_blocks_count = option;
1533                         break;
1534                 case Opt_nobh:
1535                         set_opt(sbi->s_mount_opt, NOBH);
1536                         break;
1537                 case Opt_bh:
1538                         clear_opt(sbi->s_mount_opt, NOBH);
1539                         break;
1540                 case Opt_i_version:
1541                         set_opt(sbi->s_mount_opt, I_VERSION);
1542                         sb->s_flags |= MS_I_VERSION;
1543                         break;
1544                 case Opt_nodelalloc:
1545                         clear_opt(sbi->s_mount_opt, DELALLOC);
1546                         break;
1547                 case Opt_stripe:
1548                         if (match_int(&args[0], &option))
1549                                 return 0;
1550                         if (option < 0)
1551                                 return 0;
1552                         sbi->s_stripe = option;
1553                         break;
1554                 case Opt_delalloc:
1555                         set_opt(sbi->s_mount_opt, DELALLOC);
1556                         break;
1557                 case Opt_block_validity:
1558                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1559                         break;
1560                 case Opt_noblock_validity:
1561                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1562                         break;
1563                 case Opt_inode_readahead_blks:
1564                         if (match_int(&args[0], &option))
1565                                 return 0;
1566                         if (option < 0 || option > (1 << 30))
1567                                 return 0;
1568                         if (!is_power_of_2(option)) {
1569                                 ext4_msg(sb, KERN_ERR,
1570                                          "EXT4-fs: inode_readahead_blks"
1571                                          " must be a power of 2");
1572                                 return 0;
1573                         }
1574                         sbi->s_inode_readahead_blks = option;
1575                         break;
1576                 case Opt_journal_ioprio:
1577                         if (match_int(&args[0], &option))
1578                                 return 0;
1579                         if (option < 0 || option > 7)
1580                                 break;
1581                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1582                                                             option);
1583                         break;
1584                 case Opt_noauto_da_alloc:
1585                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1586                         break;
1587                 case Opt_auto_da_alloc:
1588                         if (match_int(&args[0], &option)) {
1589                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1590                                 break;
1591                         }
1592                         if (option)
1593                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1594                         else
1595                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1596                         break;
1597                 case Opt_discard:
1598                         set_opt(sbi->s_mount_opt, DISCARD);
1599                         break;
1600                 case Opt_nodiscard:
1601                         clear_opt(sbi->s_mount_opt, DISCARD);
1602                         break;
1603                 default:
1604                         ext4_msg(sb, KERN_ERR,
1605                                "Unrecognized mount option \"%s\" "
1606                                "or missing value", p);
1607                         return 0;
1608                 }
1609         }
1610 #ifdef CONFIG_QUOTA
1611         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1612                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1613                      sbi->s_qf_names[USRQUOTA])
1614                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1615
1616                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1617                      sbi->s_qf_names[GRPQUOTA])
1618                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1619
1620                 if ((sbi->s_qf_names[USRQUOTA] &&
1621                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1622                     (sbi->s_qf_names[GRPQUOTA] &&
1623                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1624                         ext4_msg(sb, KERN_ERR, "old and new quota "
1625                                         "format mixing");
1626                         return 0;
1627                 }
1628
1629                 if (!sbi->s_jquota_fmt) {
1630                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1631                                         "not specified");
1632                         return 0;
1633                 }
1634         } else {
1635                 if (sbi->s_jquota_fmt) {
1636                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1637                                         "specified with no journaling "
1638                                         "enabled");
1639                         return 0;
1640                 }
1641         }
1642 #endif
1643         return 1;
1644 }
1645
1646 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1647                             int read_only)
1648 {
1649         struct ext4_sb_info *sbi = EXT4_SB(sb);
1650         int res = 0;
1651
1652         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1653                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1654                          "forcing read-only mode");
1655                 res = MS_RDONLY;
1656         }
1657         if (read_only)
1658                 return res;
1659         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1660                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1661                          "running e2fsck is recommended");
1662         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1663                 ext4_msg(sb, KERN_WARNING,
1664                          "warning: mounting fs with errors, "
1665                          "running e2fsck is recommended");
1666         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1667                  le16_to_cpu(es->s_mnt_count) >=
1668                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1669                 ext4_msg(sb, KERN_WARNING,
1670                          "warning: maximal mount count reached, "
1671                          "running e2fsck is recommended");
1672         else if (le32_to_cpu(es->s_checkinterval) &&
1673                 (le32_to_cpu(es->s_lastcheck) +
1674                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1675                 ext4_msg(sb, KERN_WARNING,
1676                          "warning: checktime reached, "
1677                          "running e2fsck is recommended");
1678         if (!sbi->s_journal)
1679                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1680         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1681                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1682         le16_add_cpu(&es->s_mnt_count, 1);
1683         es->s_mtime = cpu_to_le32(get_seconds());
1684         ext4_update_dynamic_rev(sb);
1685         if (sbi->s_journal)
1686                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1687
1688         ext4_commit_super(sb, 1);
1689         if (test_opt(sb, DEBUG))
1690                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1691                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1692                         sb->s_blocksize,
1693                         sbi->s_groups_count,
1694                         EXT4_BLOCKS_PER_GROUP(sb),
1695                         EXT4_INODES_PER_GROUP(sb),
1696                         sbi->s_mount_opt);
1697
1698         return res;
1699 }
1700
1701 static int ext4_fill_flex_info(struct super_block *sb)
1702 {
1703         struct ext4_sb_info *sbi = EXT4_SB(sb);
1704         struct ext4_group_desc *gdp = NULL;
1705         ext4_group_t flex_group_count;
1706         ext4_group_t flex_group;
1707         int groups_per_flex = 0;
1708         size_t size;
1709         int i;
1710
1711         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1712         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1713
1714         if (groups_per_flex < 2) {
1715                 sbi->s_log_groups_per_flex = 0;
1716                 return 1;
1717         }
1718
1719         /* We allocate both existing and potentially added groups */
1720         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1721                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1722                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1723         size = flex_group_count * sizeof(struct flex_groups);
1724         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1725         if (sbi->s_flex_groups == NULL) {
1726                 sbi->s_flex_groups = vmalloc(size);
1727                 if (sbi->s_flex_groups)
1728                         memset(sbi->s_flex_groups, 0, size);
1729         }
1730         if (sbi->s_flex_groups == NULL) {
1731                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1732                                 "%u flex groups", flex_group_count);
1733                 goto failed;
1734         }
1735
1736         for (i = 0; i < sbi->s_groups_count; i++) {
1737                 gdp = ext4_get_group_desc(sb, i, NULL);
1738
1739                 flex_group = ext4_flex_group(sbi, i);
1740                 atomic_add(ext4_free_inodes_count(sb, gdp),
1741                            &sbi->s_flex_groups[flex_group].free_inodes);
1742                 atomic_add(ext4_free_blks_count(sb, gdp),
1743                            &sbi->s_flex_groups[flex_group].free_blocks);
1744                 atomic_add(ext4_used_dirs_count(sb, gdp),
1745                            &sbi->s_flex_groups[flex_group].used_dirs);
1746         }
1747
1748         return 1;
1749 failed:
1750         return 0;
1751 }
1752
1753 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1754                             struct ext4_group_desc *gdp)
1755 {
1756         __u16 crc = 0;
1757
1758         if (sbi->s_es->s_feature_ro_compat &
1759             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1760                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1761                 __le32 le_group = cpu_to_le32(block_group);
1762
1763                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1764                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1765                 crc = crc16(crc, (__u8 *)gdp, offset);
1766                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1767                 /* for checksum of struct ext4_group_desc do the rest...*/
1768                 if ((sbi->s_es->s_feature_incompat &
1769                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1770                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1771                         crc = crc16(crc, (__u8 *)gdp + offset,
1772                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1773                                         offset);
1774         }
1775
1776         return cpu_to_le16(crc);
1777 }
1778
1779 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1780                                 struct ext4_group_desc *gdp)
1781 {
1782         if ((sbi->s_es->s_feature_ro_compat &
1783              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1784             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1785                 return 0;
1786
1787         return 1;
1788 }
1789
1790 /* Called at mount-time, super-block is locked */
1791 static int ext4_check_descriptors(struct super_block *sb)
1792 {
1793         struct ext4_sb_info *sbi = EXT4_SB(sb);
1794         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1795         ext4_fsblk_t last_block;
1796         ext4_fsblk_t block_bitmap;
1797         ext4_fsblk_t inode_bitmap;
1798         ext4_fsblk_t inode_table;
1799         int flexbg_flag = 0;
1800         ext4_group_t i;
1801
1802         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1803                 flexbg_flag = 1;
1804
1805         ext4_debug("Checking group descriptors");
1806
1807         for (i = 0; i < sbi->s_groups_count; i++) {
1808                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1809
1810                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1811                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1812                 else
1813                         last_block = first_block +
1814                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1815
1816                 block_bitmap = ext4_block_bitmap(sb, gdp);
1817                 if (block_bitmap < first_block || block_bitmap > last_block) {
1818                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1819                                "Block bitmap for group %u not in group "
1820                                "(block %llu)!", i, block_bitmap);
1821                         return 0;
1822                 }
1823                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1824                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1825                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1826                                "Inode bitmap for group %u not in group "
1827                                "(block %llu)!", i, inode_bitmap);
1828                         return 0;
1829                 }
1830                 inode_table = ext4_inode_table(sb, gdp);
1831                 if (inode_table < first_block ||
1832                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1833                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1834                                "Inode table for group %u not in group "
1835                                "(block %llu)!", i, inode_table);
1836                         return 0;
1837                 }
1838                 ext4_lock_group(sb, i);
1839                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1840                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1841                                  "Checksum for group %u failed (%u!=%u)",
1842                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1843                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1844                         if (!(sb->s_flags & MS_RDONLY)) {
1845                                 ext4_unlock_group(sb, i);
1846                                 return 0;
1847                         }
1848                 }
1849                 ext4_unlock_group(sb, i);
1850                 if (!flexbg_flag)
1851                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1852         }
1853
1854         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1855         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1856         return 1;
1857 }
1858
1859 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1860  * the superblock) which were deleted from all directories, but held open by
1861  * a process at the time of a crash.  We walk the list and try to delete these
1862  * inodes at recovery time (only with a read-write filesystem).
1863  *
1864  * In order to keep the orphan inode chain consistent during traversal (in
1865  * case of crash during recovery), we link each inode into the superblock
1866  * orphan list_head and handle it the same way as an inode deletion during
1867  * normal operation (which journals the operations for us).
1868  *
1869  * We only do an iget() and an iput() on each inode, which is very safe if we
1870  * accidentally point at an in-use or already deleted inode.  The worst that
1871  * can happen in this case is that we get a "bit already cleared" message from
1872  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1873  * e2fsck was run on this filesystem, and it must have already done the orphan
1874  * inode cleanup for us, so we can safely abort without any further action.
1875  */
1876 static void ext4_orphan_cleanup(struct super_block *sb,
1877                                 struct ext4_super_block *es)
1878 {
1879         unsigned int s_flags = sb->s_flags;
1880         int nr_orphans = 0, nr_truncates = 0;
1881 #ifdef CONFIG_QUOTA
1882         int i;
1883 #endif
1884         if (!es->s_last_orphan) {
1885                 jbd_debug(4, "no orphan inodes to clean up\n");
1886                 return;
1887         }
1888
1889         if (bdev_read_only(sb->s_bdev)) {
1890                 ext4_msg(sb, KERN_ERR, "write access "
1891                         "unavailable, skipping orphan cleanup");
1892                 return;
1893         }
1894
1895         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1896                 if (es->s_last_orphan)
1897                         jbd_debug(1, "Errors on filesystem, "
1898                                   "clearing orphan list.\n");
1899                 es->s_last_orphan = 0;
1900                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1901                 return;
1902         }
1903
1904         if (s_flags & MS_RDONLY) {
1905                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1906                 sb->s_flags &= ~MS_RDONLY;
1907         }
1908 #ifdef CONFIG_QUOTA
1909         /* Needed for iput() to work correctly and not trash data */
1910         sb->s_flags |= MS_ACTIVE;
1911         /* Turn on quotas so that they are updated correctly */
1912         for (i = 0; i < MAXQUOTAS; i++) {
1913                 if (EXT4_SB(sb)->s_qf_names[i]) {
1914                         int ret = ext4_quota_on_mount(sb, i);
1915                         if (ret < 0)
1916                                 ext4_msg(sb, KERN_ERR,
1917                                         "Cannot turn on journaled "
1918                                         "quota: error %d", ret);
1919                 }
1920         }
1921 #endif
1922
1923         while (es->s_last_orphan) {
1924                 struct inode *inode;
1925
1926                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1927                 if (IS_ERR(inode)) {
1928                         es->s_last_orphan = 0;
1929                         break;
1930                 }
1931
1932                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1933                 dquot_initialize(inode);
1934                 if (inode->i_nlink) {
1935                         ext4_msg(sb, KERN_DEBUG,
1936                                 "%s: truncating inode %lu to %lld bytes",
1937                                 __func__, inode->i_ino, inode->i_size);
1938                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1939                                   inode->i_ino, inode->i_size);
1940                         ext4_truncate(inode);
1941                         nr_truncates++;
1942                 } else {
1943                         ext4_msg(sb, KERN_DEBUG,
1944                                 "%s: deleting unreferenced inode %lu",
1945                                 __func__, inode->i_ino);
1946                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1947                                   inode->i_ino);
1948                         nr_orphans++;
1949                 }
1950                 iput(inode);  /* The delete magic happens here! */
1951         }
1952
1953 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1954
1955         if (nr_orphans)
1956                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1957                        PLURAL(nr_orphans));
1958         if (nr_truncates)
1959                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1960                        PLURAL(nr_truncates));
1961 #ifdef CONFIG_QUOTA
1962         /* Turn quotas off */
1963         for (i = 0; i < MAXQUOTAS; i++) {
1964                 if (sb_dqopt(sb)->files[i])
1965                         vfs_quota_off(sb, i, 0);
1966         }
1967 #endif
1968         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1969 }
1970
1971 /*
1972  * Maximal extent format file size.
1973  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1974  * extent format containers, within a sector_t, and within i_blocks
1975  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1976  * so that won't be a limiting factor.
1977  *
1978  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1979  */
1980 static loff_t ext4_max_size(int blkbits, int has_huge_files)
1981 {
1982         loff_t res;
1983         loff_t upper_limit = MAX_LFS_FILESIZE;
1984
1985         /* small i_blocks in vfs inode? */
1986         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
1987                 /*
1988                  * CONFIG_LBDAF is not enabled implies the inode
1989                  * i_block represent total blocks in 512 bytes
1990                  * 32 == size of vfs inode i_blocks * 8
1991                  */
1992                 upper_limit = (1LL << 32) - 1;
1993
1994                 /* total blocks in file system block size */
1995                 upper_limit >>= (blkbits - 9);
1996                 upper_limit <<= blkbits;
1997         }
1998
1999         /* 32-bit extent-start container, ee_block */
2000         res = 1LL << 32;
2001         res <<= blkbits;
2002         res -= 1;
2003
2004         /* Sanity check against vm- & vfs- imposed limits */
2005         if (res > upper_limit)
2006                 res = upper_limit;
2007
2008         return res;
2009 }
2010
2011 /*
2012  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2013  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2014  * We need to be 1 filesystem block less than the 2^48 sector limit.
2015  */
2016 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2017 {
2018         loff_t res = EXT4_NDIR_BLOCKS;
2019         int meta_blocks;
2020         loff_t upper_limit;
2021         /* This is calculated to be the largest file size for a dense, block
2022          * mapped file such that the file's total number of 512-byte sectors,
2023          * including data and all indirect blocks, does not exceed (2^48 - 1).
2024          *
2025          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2026          * number of 512-byte sectors of the file.
2027          */
2028
2029         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2030                 /*
2031                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2032                  * the inode i_block field represents total file blocks in
2033                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2034                  */
2035                 upper_limit = (1LL << 32) - 1;
2036
2037                 /* total blocks in file system block size */
2038                 upper_limit >>= (bits - 9);
2039
2040         } else {
2041                 /*
2042                  * We use 48 bit ext4_inode i_blocks
2043                  * With EXT4_HUGE_FILE_FL set the i_blocks
2044                  * represent total number of blocks in
2045                  * file system block size
2046                  */
2047                 upper_limit = (1LL << 48) - 1;
2048
2049         }
2050
2051         /* indirect blocks */
2052         meta_blocks = 1;
2053         /* double indirect blocks */
2054         meta_blocks += 1 + (1LL << (bits-2));
2055         /* tripple indirect blocks */
2056         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2057
2058         upper_limit -= meta_blocks;
2059         upper_limit <<= bits;
2060
2061         res += 1LL << (bits-2);
2062         res += 1LL << (2*(bits-2));
2063         res += 1LL << (3*(bits-2));
2064         res <<= bits;
2065         if (res > upper_limit)
2066                 res = upper_limit;
2067
2068         if (res > MAX_LFS_FILESIZE)
2069                 res = MAX_LFS_FILESIZE;
2070
2071         return res;
2072 }
2073
2074 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2075                                    ext4_fsblk_t logical_sb_block, int nr)
2076 {
2077         struct ext4_sb_info *sbi = EXT4_SB(sb);
2078         ext4_group_t bg, first_meta_bg;
2079         int has_super = 0;
2080
2081         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2082
2083         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2084             nr < first_meta_bg)
2085                 return logical_sb_block + nr + 1;
2086         bg = sbi->s_desc_per_block * nr;
2087         if (ext4_bg_has_super(sb, bg))
2088                 has_super = 1;
2089
2090         return (has_super + ext4_group_first_block_no(sb, bg));
2091 }
2092
2093 /**
2094  * ext4_get_stripe_size: Get the stripe size.
2095  * @sbi: In memory super block info
2096  *
2097  * If we have specified it via mount option, then
2098  * use the mount option value. If the value specified at mount time is
2099  * greater than the blocks per group use the super block value.
2100  * If the super block value is greater than blocks per group return 0.
2101  * Allocator needs it be less than blocks per group.
2102  *
2103  */
2104 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2105 {
2106         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2107         unsigned long stripe_width =
2108                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2109
2110         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2111                 return sbi->s_stripe;
2112
2113         if (stripe_width <= sbi->s_blocks_per_group)
2114                 return stripe_width;
2115
2116         if (stride <= sbi->s_blocks_per_group)
2117                 return stride;
2118
2119         return 0;
2120 }
2121
2122 /* sysfs supprt */
2123
2124 struct ext4_attr {
2125         struct attribute attr;
2126         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2127         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2128                          const char *, size_t);
2129         int offset;
2130 };
2131
2132 static int parse_strtoul(const char *buf,
2133                 unsigned long max, unsigned long *value)
2134 {
2135         char *endp;
2136
2137         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2138         endp = skip_spaces(endp);
2139         if (*endp || *value > max)
2140                 return -EINVAL;
2141
2142         return 0;
2143 }
2144
2145 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2146                                               struct ext4_sb_info *sbi,
2147                                               char *buf)
2148 {
2149         return snprintf(buf, PAGE_SIZE, "%llu\n",
2150                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2151 }
2152
2153 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2154                                          struct ext4_sb_info *sbi, char *buf)
2155 {
2156         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2157
2158         return snprintf(buf, PAGE_SIZE, "%lu\n",
2159                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2160                          sbi->s_sectors_written_start) >> 1);
2161 }
2162
2163 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2164                                           struct ext4_sb_info *sbi, char *buf)
2165 {
2166         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2167
2168         return snprintf(buf, PAGE_SIZE, "%llu\n",
2169                         (unsigned long long)(sbi->s_kbytes_written +
2170                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2171                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2172 }
2173
2174 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2175                                           struct ext4_sb_info *sbi,
2176                                           const char *buf, size_t count)
2177 {
2178         unsigned long t;
2179
2180         if (parse_strtoul(buf, 0x40000000, &t))
2181                 return -EINVAL;
2182
2183         if (!is_power_of_2(t))
2184                 return -EINVAL;
2185
2186         sbi->s_inode_readahead_blks = t;
2187         return count;
2188 }
2189
2190 static ssize_t sbi_ui_show(struct ext4_attr *a,
2191                            struct ext4_sb_info *sbi, char *buf)
2192 {
2193         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2194
2195         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2196 }
2197
2198 static ssize_t sbi_ui_store(struct ext4_attr *a,
2199                             struct ext4_sb_info *sbi,
2200                             const char *buf, size_t count)
2201 {
2202         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2203         unsigned long t;
2204
2205         if (parse_strtoul(buf, 0xffffffff, &t))
2206                 return -EINVAL;
2207         *ui = t;
2208         return count;
2209 }
2210
2211 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2212 static struct ext4_attr ext4_attr_##_name = {                   \
2213         .attr = {.name = __stringify(_name), .mode = _mode },   \
2214         .show   = _show,                                        \
2215         .store  = _store,                                       \
2216         .offset = offsetof(struct ext4_sb_info, _elname),       \
2217 }
2218 #define EXT4_ATTR(name, mode, show, store) \
2219 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2220
2221 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2222 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2223 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2224         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2225 #define ATTR_LIST(name) &ext4_attr_##name.attr
2226
2227 EXT4_RO_ATTR(delayed_allocation_blocks);
2228 EXT4_RO_ATTR(session_write_kbytes);
2229 EXT4_RO_ATTR(lifetime_write_kbytes);
2230 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2231                  inode_readahead_blks_store, s_inode_readahead_blks);
2232 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2233 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2234 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2235 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2236 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2237 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2238 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2239 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2240
2241 static struct attribute *ext4_attrs[] = {
2242         ATTR_LIST(delayed_allocation_blocks),
2243         ATTR_LIST(session_write_kbytes),
2244         ATTR_LIST(lifetime_write_kbytes),
2245         ATTR_LIST(inode_readahead_blks),
2246         ATTR_LIST(inode_goal),
2247         ATTR_LIST(mb_stats),
2248         ATTR_LIST(mb_max_to_scan),
2249         ATTR_LIST(mb_min_to_scan),
2250         ATTR_LIST(mb_order2_req),
2251         ATTR_LIST(mb_stream_req),
2252         ATTR_LIST(mb_group_prealloc),
2253         ATTR_LIST(max_writeback_mb_bump),
2254         NULL,
2255 };
2256
2257 static ssize_t ext4_attr_show(struct kobject *kobj,
2258                               struct attribute *attr, char *buf)
2259 {
2260         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2261                                                 s_kobj);
2262         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2263
2264         return a->show ? a->show(a, sbi, buf) : 0;
2265 }
2266
2267 static ssize_t ext4_attr_store(struct kobject *kobj,
2268                                struct attribute *attr,
2269                                const char *buf, size_t len)
2270 {
2271         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2272                                                 s_kobj);
2273         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2274
2275         return a->store ? a->store(a, sbi, buf, len) : 0;
2276 }
2277
2278 static void ext4_sb_release(struct kobject *kobj)
2279 {
2280         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2281                                                 s_kobj);
2282         complete(&sbi->s_kobj_unregister);
2283 }
2284
2285
2286 static struct sysfs_ops ext4_attr_ops = {
2287         .show   = ext4_attr_show,
2288         .store  = ext4_attr_store,
2289 };
2290
2291 static struct kobj_type ext4_ktype = {
2292         .default_attrs  = ext4_attrs,
2293         .sysfs_ops      = &ext4_attr_ops,
2294         .release        = ext4_sb_release,
2295 };
2296
2297 /*
2298  * Check whether this filesystem can be mounted based on
2299  * the features present and the RDONLY/RDWR mount requested.
2300  * Returns 1 if this filesystem can be mounted as requested,
2301  * 0 if it cannot be.
2302  */
2303 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2304 {
2305         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2306                 ext4_msg(sb, KERN_ERR,
2307                         "Couldn't mount because of "
2308                         "unsupported optional features (%x)",
2309                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2310                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2311                 return 0;
2312         }
2313
2314         if (readonly)
2315                 return 1;
2316
2317         /* Check that feature set is OK for a read-write mount */
2318         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2319                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2320                          "unsupported optional features (%x)",
2321                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2322                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2323                 return 0;
2324         }
2325         /*
2326          * Large file size enabled file system can only be mounted
2327          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2328          */
2329         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2330                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2331                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2332                                  "cannot be mounted RDWR without "
2333                                  "CONFIG_LBDAF");
2334                         return 0;
2335                 }
2336         }
2337         return 1;
2338 }
2339
2340 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2341                                 __releases(kernel_lock)
2342                                 __acquires(kernel_lock)
2343 {
2344         struct buffer_head *bh;
2345         struct ext4_super_block *es = NULL;
2346         struct ext4_sb_info *sbi;
2347         ext4_fsblk_t block;
2348         ext4_fsblk_t sb_block = get_sb_block(&data);
2349         ext4_fsblk_t logical_sb_block;
2350         unsigned long offset = 0;
2351         unsigned long journal_devnum = 0;
2352         unsigned long def_mount_opts;
2353         struct inode *root;
2354         char *cp;
2355         const char *descr;
2356         int ret = -EINVAL;
2357         int blocksize;
2358         unsigned int db_count;
2359         unsigned int i;
2360         int needs_recovery, has_huge_files;
2361         __u64 blocks_count;
2362         int err;
2363         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2364
2365         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2366         if (!sbi)
2367                 return -ENOMEM;
2368
2369         sbi->s_blockgroup_lock =
2370                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2371         if (!sbi->s_blockgroup_lock) {
2372                 kfree(sbi);
2373                 return -ENOMEM;
2374         }
2375         sb->s_fs_info = sbi;
2376         sbi->s_mount_opt = 0;
2377         sbi->s_resuid = EXT4_DEF_RESUID;
2378         sbi->s_resgid = EXT4_DEF_RESGID;
2379         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2380         sbi->s_sb_block = sb_block;
2381         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2382                                                       sectors[1]);
2383
2384         unlock_kernel();
2385
2386         /* Cleanup superblock name */
2387         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2388                 *cp = '!';
2389
2390         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2391         if (!blocksize) {
2392                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2393                 goto out_fail;
2394         }
2395
2396         /*
2397          * The ext4 superblock will not be buffer aligned for other than 1kB
2398          * block sizes.  We need to calculate the offset from buffer start.
2399          */
2400         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2401                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2402                 offset = do_div(logical_sb_block, blocksize);
2403         } else {
2404                 logical_sb_block = sb_block;
2405         }
2406
2407         if (!(bh = sb_bread(sb, logical_sb_block))) {
2408                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2409                 goto out_fail;
2410         }
2411         /*
2412          * Note: s_es must be initialized as soon as possible because
2413          *       some ext4 macro-instructions depend on its value
2414          */
2415         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2416         sbi->s_es = es;
2417         sb->s_magic = le16_to_cpu(es->s_magic);
2418         if (sb->s_magic != EXT4_SUPER_MAGIC)
2419                 goto cantfind_ext4;
2420         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2421
2422         /* Set defaults before we parse the mount options */
2423         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2424         if (def_mount_opts & EXT4_DEFM_DEBUG)
2425                 set_opt(sbi->s_mount_opt, DEBUG);
2426         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
2427                 set_opt(sbi->s_mount_opt, GRPID);
2428         if (def_mount_opts & EXT4_DEFM_UID16)
2429                 set_opt(sbi->s_mount_opt, NO_UID32);
2430 #ifdef CONFIG_EXT4_FS_XATTR
2431         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2432                 set_opt(sbi->s_mount_opt, XATTR_USER);
2433 #endif
2434 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2435         if (def_mount_opts & EXT4_DEFM_ACL)
2436                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2437 #endif
2438         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2439                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
2440         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2441                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
2442         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2443                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
2444
2445         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2446                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2447         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2448                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2449         else
2450                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2451
2452         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2453         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2454         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2455         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2456         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2457
2458         set_opt(sbi->s_mount_opt, BARRIER);
2459
2460         /*
2461          * enable delayed allocation by default
2462          * Use -o nodelalloc to turn it off
2463          */
2464         set_opt(sbi->s_mount_opt, DELALLOC);
2465
2466         if (!parse_options((char *) data, sb, &journal_devnum,
2467                            &journal_ioprio, NULL, 0))
2468                 goto failed_mount;
2469
2470         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2471                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2472
2473         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2474             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2475              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2476              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2477                 ext4_msg(sb, KERN_WARNING,
2478                        "feature flags set on rev 0 fs, "
2479                        "running e2fsck is recommended");
2480
2481         /*
2482          * Check feature flags regardless of the revision level, since we
2483          * previously didn't change the revision level when setting the flags,
2484          * so there is a chance incompat flags are set on a rev 0 filesystem.
2485          */
2486         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2487                 goto failed_mount;
2488
2489         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2490
2491         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2492             blocksize > EXT4_MAX_BLOCK_SIZE) {
2493                 ext4_msg(sb, KERN_ERR,
2494                        "Unsupported filesystem blocksize %d", blocksize);
2495                 goto failed_mount;
2496         }
2497
2498         if (sb->s_blocksize != blocksize) {
2499                 /* Validate the filesystem blocksize */
2500                 if (!sb_set_blocksize(sb, blocksize)) {
2501                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2502                                         blocksize);
2503                         goto failed_mount;
2504                 }
2505
2506                 brelse(bh);
2507                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2508                 offset = do_div(logical_sb_block, blocksize);
2509                 bh = sb_bread(sb, logical_sb_block);
2510                 if (!bh) {
2511                         ext4_msg(sb, KERN_ERR,
2512                                "Can't read superblock on 2nd try");
2513                         goto failed_mount;
2514                 }
2515                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2516                 sbi->s_es = es;
2517                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2518                         ext4_msg(sb, KERN_ERR,
2519                                "Magic mismatch, very weird!");
2520                         goto failed_mount;
2521                 }
2522         }
2523
2524         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2525                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2526         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2527                                                       has_huge_files);
2528         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2529
2530         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2531                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2532                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2533         } else {
2534                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2535                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2536                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2537                     (!is_power_of_2(sbi->s_inode_size)) ||
2538                     (sbi->s_inode_size > blocksize)) {
2539                         ext4_msg(sb, KERN_ERR,
2540                                "unsupported inode size: %d",
2541                                sbi->s_inode_size);
2542                         goto failed_mount;
2543                 }
2544                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2545                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2546         }
2547
2548         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2549         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2550                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2551                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2552                     !is_power_of_2(sbi->s_desc_size)) {
2553                         ext4_msg(sb, KERN_ERR,
2554                                "unsupported descriptor size %lu",
2555                                sbi->s_desc_size);
2556                         goto failed_mount;
2557                 }
2558         } else
2559                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2560
2561         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2562         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2563         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2564                 goto cantfind_ext4;
2565
2566         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2567         if (sbi->s_inodes_per_block == 0)
2568                 goto cantfind_ext4;
2569         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2570                                         sbi->s_inodes_per_block;
2571         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2572         sbi->s_sbh = bh;
2573         sbi->s_mount_state = le16_to_cpu(es->s_state);
2574         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2575         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2576
2577         for (i = 0; i < 4; i++)
2578                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2579         sbi->s_def_hash_version = es->s_def_hash_version;
2580         i = le32_to_cpu(es->s_flags);
2581         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2582                 sbi->s_hash_unsigned = 3;
2583         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2584 #ifdef __CHAR_UNSIGNED__
2585                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2586                 sbi->s_hash_unsigned = 3;
2587 #else
2588                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2589 #endif
2590                 sb->s_dirt = 1;
2591         }
2592
2593         if (sbi->s_blocks_per_group > blocksize * 8) {
2594                 ext4_msg(sb, KERN_ERR,
2595                        "#blocks per group too big: %lu",
2596                        sbi->s_blocks_per_group);
2597                 goto failed_mount;
2598         }
2599         if (sbi->s_inodes_per_group > blocksize * 8) {
2600                 ext4_msg(sb, KERN_ERR,
2601                        "#inodes per group too big: %lu",
2602                        sbi->s_inodes_per_group);
2603                 goto failed_mount;
2604         }
2605
2606         /*
2607          * Test whether we have more sectors than will fit in sector_t,
2608          * and whether the max offset is addressable by the page cache.
2609          */
2610         if ((ext4_blocks_count(es) >
2611              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2612             (ext4_blocks_count(es) >
2613              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2614                 ext4_msg(sb, KERN_ERR, "filesystem"
2615                          " too large to mount safely on this system");
2616                 if (sizeof(sector_t) < 8)
2617                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2618                 ret = -EFBIG;
2619                 goto failed_mount;
2620         }
2621
2622         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2623                 goto cantfind_ext4;
2624
2625         /* check blocks count against device size */
2626         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2627         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2628                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2629                        "exceeds size of device (%llu blocks)",
2630                        ext4_blocks_count(es), blocks_count);
2631                 goto failed_mount;
2632         }
2633
2634         /*
2635          * It makes no sense for the first data block to be beyond the end
2636          * of the filesystem.
2637          */
2638         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2639                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2640                          "block %u is beyond end of filesystem (%llu)",
2641                          le32_to_cpu(es->s_first_data_block),
2642                          ext4_blocks_count(es));
2643                 goto failed_mount;
2644         }
2645         blocks_count = (ext4_blocks_count(es) -
2646                         le32_to_cpu(es->s_first_data_block) +
2647                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2648         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2649         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2650                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2651                        "(block count %llu, first data block %u, "
2652                        "blocks per group %lu)", sbi->s_groups_count,
2653                        ext4_blocks_count(es),
2654                        le32_to_cpu(es->s_first_data_block),
2655                        EXT4_BLOCKS_PER_GROUP(sb));
2656                 goto failed_mount;
2657         }
2658         sbi->s_groups_count = blocks_count;
2659         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2660                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2661         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2662                    EXT4_DESC_PER_BLOCK(sb);
2663         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2664                                     GFP_KERNEL);
2665         if (sbi->s_group_desc == NULL) {
2666                 ext4_msg(sb, KERN_ERR, "not enough memory");
2667                 goto failed_mount;
2668         }
2669
2670 #ifdef CONFIG_PROC_FS
2671         if (ext4_proc_root)
2672                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2673 #endif
2674
2675         bgl_lock_init(sbi->s_blockgroup_lock);
2676
2677         for (i = 0; i < db_count; i++) {
2678                 block = descriptor_loc(sb, logical_sb_block, i);
2679                 sbi->s_group_desc[i] = sb_bread(sb, block);
2680                 if (!sbi->s_group_desc[i]) {
2681                         ext4_msg(sb, KERN_ERR,
2682                                "can't read group descriptor %d", i);
2683                         db_count = i;
2684                         goto failed_mount2;
2685                 }
2686         }
2687         if (!ext4_check_descriptors(sb)) {
2688                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2689                 goto failed_mount2;
2690         }
2691         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2692                 if (!ext4_fill_flex_info(sb)) {
2693                         ext4_msg(sb, KERN_ERR,
2694                                "unable to initialize "
2695                                "flex_bg meta info!");
2696                         goto failed_mount2;
2697                 }
2698
2699         sbi->s_gdb_count = db_count;
2700         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2701         spin_lock_init(&sbi->s_next_gen_lock);
2702
2703         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2704                         ext4_count_free_blocks(sb));
2705         if (!err) {
2706                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2707                                 ext4_count_free_inodes(sb));
2708         }
2709         if (!err) {
2710                 err = percpu_counter_init(&sbi->s_dirs_counter,
2711                                 ext4_count_dirs(sb));
2712         }
2713         if (!err) {
2714                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2715         }
2716         if (err) {
2717                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2718                 goto failed_mount3;
2719         }
2720
2721         sbi->s_stripe = ext4_get_stripe_size(sbi);
2722         sbi->s_max_writeback_mb_bump = 128;
2723
2724         /*
2725          * set up enough so that it can read an inode
2726          */
2727         if (!test_opt(sb, NOLOAD) &&
2728             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2729                 sb->s_op = &ext4_sops;
2730         else
2731                 sb->s_op = &ext4_nojournal_sops;
2732         sb->s_export_op = &ext4_export_ops;
2733         sb->s_xattr = ext4_xattr_handlers;
2734 #ifdef CONFIG_QUOTA
2735         sb->s_qcop = &ext4_qctl_operations;
2736         sb->dq_op = &ext4_quota_operations;
2737 #endif
2738         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2739         mutex_init(&sbi->s_orphan_lock);
2740         mutex_init(&sbi->s_resize_lock);
2741
2742         sb->s_root = NULL;
2743
2744         needs_recovery = (es->s_last_orphan != 0 ||
2745                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2746                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2747
2748         /*
2749          * The first inode we look at is the journal inode.  Don't try
2750          * root first: it may be modified in the journal!
2751          */
2752         if (!test_opt(sb, NOLOAD) &&
2753             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2754                 if (ext4_load_journal(sb, es, journal_devnum))
2755                         goto failed_mount3;
2756         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2757               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2758                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2759                        "suppressed and not mounted read-only");
2760                 goto failed_mount4;
2761         } else {
2762                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2763                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2764                 sbi->s_journal = NULL;
2765                 needs_recovery = 0;
2766                 goto no_journal;
2767         }
2768
2769         if (ext4_blocks_count(es) > 0xffffffffULL &&
2770             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2771                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2772                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2773                 goto failed_mount4;
2774         }
2775
2776         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2777                 jbd2_journal_set_features(sbi->s_journal,
2778                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2779                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2780         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2781                 jbd2_journal_set_features(sbi->s_journal,
2782                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2783                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2784                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2785         } else {
2786                 jbd2_journal_clear_features(sbi->s_journal,
2787                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2788                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2789         }
2790
2791         /* We have now updated the journal if required, so we can
2792          * validate the data journaling mode. */
2793         switch (test_opt(sb, DATA_FLAGS)) {
2794         case 0:
2795                 /* No mode set, assume a default based on the journal
2796                  * capabilities: ORDERED_DATA if the journal can
2797                  * cope, else JOURNAL_DATA
2798                  */
2799                 if (jbd2_journal_check_available_features
2800                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2801                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2802                 else
2803                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2804                 break;
2805
2806         case EXT4_MOUNT_ORDERED_DATA:
2807         case EXT4_MOUNT_WRITEBACK_DATA:
2808                 if (!jbd2_journal_check_available_features
2809                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2810                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2811                                "requested data journaling mode");
2812                         goto failed_mount4;
2813                 }
2814         default:
2815                 break;
2816         }
2817         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2818
2819 no_journal:
2820
2821         if (test_opt(sb, NOBH)) {
2822                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2823                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2824                                 "its supported only with writeback mode");
2825                         clear_opt(sbi->s_mount_opt, NOBH);
2826                 }
2827         }
2828         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2829         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2830                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2831                 goto failed_mount_wq;
2832         }
2833
2834         /*
2835          * The jbd2_journal_load will have done any necessary log recovery,
2836          * so we can safely mount the rest of the filesystem now.
2837          */
2838
2839         root = ext4_iget(sb, EXT4_ROOT_INO);
2840         if (IS_ERR(root)) {
2841                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2842                 ret = PTR_ERR(root);
2843                 goto failed_mount4;
2844         }
2845         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2846                 iput(root);
2847                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2848                 goto failed_mount4;
2849         }
2850         sb->s_root = d_alloc_root(root);
2851         if (!sb->s_root) {
2852                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2853                 iput(root);
2854                 ret = -ENOMEM;
2855                 goto failed_mount4;
2856         }
2857
2858         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2859
2860         /* determine the minimum size of new large inodes, if present */
2861         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2862                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2863                                                      EXT4_GOOD_OLD_INODE_SIZE;
2864                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2865                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2866                         if (sbi->s_want_extra_isize <
2867                             le16_to_cpu(es->s_want_extra_isize))
2868                                 sbi->s_want_extra_isize =
2869                                         le16_to_cpu(es->s_want_extra_isize);
2870                         if (sbi->s_want_extra_isize <
2871                             le16_to_cpu(es->s_min_extra_isize))
2872                                 sbi->s_want_extra_isize =
2873                                         le16_to_cpu(es->s_min_extra_isize);
2874                 }
2875         }
2876         /* Check if enough inode space is available */
2877         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2878                                                         sbi->s_inode_size) {
2879                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2880                                                        EXT4_GOOD_OLD_INODE_SIZE;
2881                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2882                          "available");
2883         }
2884
2885         if (test_opt(sb, DELALLOC) &&
2886             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2887                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2888                          "requested data journaling mode");
2889                 clear_opt(sbi->s_mount_opt, DELALLOC);
2890         }
2891
2892         err = ext4_setup_system_zone(sb);
2893         if (err) {
2894                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
2895                          "zone (%d)\n", err);
2896                 goto failed_mount4;
2897         }
2898
2899         ext4_ext_init(sb);
2900         err = ext4_mb_init(sb, needs_recovery);
2901         if (err) {
2902                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
2903                          err);
2904                 goto failed_mount4;
2905         }
2906
2907         sbi->s_kobj.kset = ext4_kset;
2908         init_completion(&sbi->s_kobj_unregister);
2909         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
2910                                    "%s", sb->s_id);
2911         if (err) {
2912                 ext4_mb_release(sb);
2913                 ext4_ext_release(sb);
2914                 goto failed_mount4;
2915         };
2916
2917         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2918         ext4_orphan_cleanup(sb, es);
2919         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2920         if (needs_recovery) {
2921                 ext4_msg(sb, KERN_INFO, "recovery complete");
2922                 ext4_mark_recovery_complete(sb, es);
2923         }
2924         if (EXT4_SB(sb)->s_journal) {
2925                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2926                         descr = " journalled data mode";
2927                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2928                         descr = " ordered data mode";
2929                 else
2930                         descr = " writeback data mode";
2931         } else
2932                 descr = "out journal";
2933
2934         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s", descr);
2935
2936         lock_kernel();
2937         return 0;
2938
2939 cantfind_ext4:
2940         if (!silent)
2941                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
2942         goto failed_mount;
2943
2944 failed_mount4:
2945         ext4_msg(sb, KERN_ERR, "mount failed");
2946         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
2947 failed_mount_wq:
2948         ext4_release_system_zone(sb);
2949         if (sbi->s_journal) {
2950                 jbd2_journal_destroy(sbi->s_journal);
2951                 sbi->s_journal = NULL;
2952         }
2953 failed_mount3:
2954         if (sbi->s_flex_groups) {
2955                 if (is_vmalloc_addr(sbi->s_flex_groups))
2956                         vfree(sbi->s_flex_groups);
2957                 else
2958                         kfree(sbi->s_flex_groups);
2959         }
2960         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2961         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2962         percpu_counter_destroy(&sbi->s_dirs_counter);
2963         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2964 failed_mount2:
2965         for (i = 0; i < db_count; i++)
2966                 brelse(sbi->s_group_desc[i]);
2967         kfree(sbi->s_group_desc);
2968 failed_mount:
2969         if (sbi->s_proc) {
2970                 remove_proc_entry(sb->s_id, ext4_proc_root);
2971         }
2972 #ifdef CONFIG_QUOTA
2973         for (i = 0; i < MAXQUOTAS; i++)
2974                 kfree(sbi->s_qf_names[i]);
2975 #endif
2976         ext4_blkdev_remove(sbi);
2977         brelse(bh);
2978 out_fail:
2979         sb->s_fs_info = NULL;
2980         kfree(sbi->s_blockgroup_lock);
2981         kfree(sbi);
2982         lock_kernel();
2983         return ret;
2984 }
2985
2986 /*
2987  * Setup any per-fs journal parameters now.  We'll do this both on
2988  * initial mount, once the journal has been initialised but before we've
2989  * done any recovery; and again on any subsequent remount.
2990  */
2991 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2992 {
2993         struct ext4_sb_info *sbi = EXT4_SB(sb);
2994
2995         journal->j_commit_interval = sbi->s_commit_interval;
2996         journal->j_min_batch_time = sbi->s_min_batch_time;
2997         journal->j_max_batch_time = sbi->s_max_batch_time;
2998
2999         spin_lock(&journal->j_state_lock);
3000         if (test_opt(sb, BARRIER))
3001                 journal->j_flags |= JBD2_BARRIER;
3002         else
3003                 journal->j_flags &= ~JBD2_BARRIER;
3004         if (test_opt(sb, DATA_ERR_ABORT))
3005                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3006         else
3007                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3008         spin_unlock(&journal->j_state_lock);
3009 }
3010
3011 static journal_t *ext4_get_journal(struct super_block *sb,
3012                                    unsigned int journal_inum)
3013 {
3014         struct inode *journal_inode;
3015         journal_t *journal;
3016
3017         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3018
3019         /* First, test for the existence of a valid inode on disk.  Bad
3020          * things happen if we iget() an unused inode, as the subsequent
3021          * iput() will try to delete it. */
3022
3023         journal_inode = ext4_iget(sb, journal_inum);
3024         if (IS_ERR(journal_inode)) {
3025                 ext4_msg(sb, KERN_ERR, "no journal found");
3026                 return NULL;
3027         }
3028         if (!journal_inode->i_nlink) {
3029                 make_bad_inode(journal_inode);
3030                 iput(journal_inode);
3031                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3032                 return NULL;
3033         }
3034
3035         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3036                   journal_inode, journal_inode->i_size);
3037         if (!S_ISREG(journal_inode->i_mode)) {
3038                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3039                 iput(journal_inode);
3040                 return NULL;
3041         }
3042
3043         journal = jbd2_journal_init_inode(journal_inode);
3044         if (!journal) {
3045                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3046                 iput(journal_inode);
3047                 return NULL;
3048         }
3049         journal->j_private = sb;
3050         ext4_init_journal_params(sb, journal);
3051         return journal;
3052 }
3053
3054 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3055                                        dev_t j_dev)
3056 {
3057         struct buffer_head *bh;
3058         journal_t *journal;
3059         ext4_fsblk_t start;
3060         ext4_fsblk_t len;
3061         int hblock, blocksize;
3062         ext4_fsblk_t sb_block;
3063         unsigned long offset;
3064         struct ext4_super_block *es;
3065         struct block_device *bdev;
3066
3067         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3068
3069         bdev = ext4_blkdev_get(j_dev, sb);
3070         if (bdev == NULL)
3071                 return NULL;
3072
3073         if (bd_claim(bdev, sb)) {
3074                 ext4_msg(sb, KERN_ERR,
3075                         "failed to claim external journal device");
3076                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3077                 return NULL;
3078         }
3079
3080         blocksize = sb->s_blocksize;
3081         hblock = bdev_logical_block_size(bdev);
3082         if (blocksize < hblock) {
3083                 ext4_msg(sb, KERN_ERR,
3084                         "blocksize too small for journal device");
3085                 goto out_bdev;
3086         }
3087
3088         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3089         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3090         set_blocksize(bdev, blocksize);
3091         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3092                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3093                        "external journal");
3094                 goto out_bdev;
3095         }
3096
3097         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3098         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3099             !(le32_to_cpu(es->s_feature_incompat) &
3100               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3101                 ext4_msg(sb, KERN_ERR, "external journal has "
3102                                         "bad superblock");
3103                 brelse(bh);
3104                 goto out_bdev;
3105         }
3106
3107         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3108                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3109                 brelse(bh);
3110                 goto out_bdev;
3111         }
3112
3113         len = ext4_blocks_count(es);
3114         start = sb_block + 1;
3115         brelse(bh);     /* we're done with the superblock */
3116
3117         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3118                                         start, len, blocksize);
3119         if (!journal) {
3120                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3121                 goto out_bdev;
3122         }
3123         journal->j_private = sb;
3124         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3125         wait_on_buffer(journal->j_sb_buffer);
3126         if (!buffer_uptodate(journal->j_sb_buffer)) {
3127                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3128                 goto out_journal;
3129         }
3130         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3131                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3132                                         "user (unsupported) - %d",
3133                         be32_to_cpu(journal->j_superblock->s_nr_users));
3134                 goto out_journal;
3135         }
3136         EXT4_SB(sb)->journal_bdev = bdev;
3137         ext4_init_journal_params(sb, journal);
3138         return journal;
3139
3140 out_journal:
3141         jbd2_journal_destroy(journal);
3142 out_bdev:
3143         ext4_blkdev_put(bdev);
3144         return NULL;
3145 }
3146
3147 static int ext4_load_journal(struct super_block *sb,
3148                              struct ext4_super_block *es,
3149                              unsigned long journal_devnum)
3150 {
3151         journal_t *journal;
3152         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3153         dev_t journal_dev;
3154         int err = 0;
3155         int really_read_only;
3156
3157         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3158
3159         if (journal_devnum &&
3160             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3161                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3162                         "numbers have changed");
3163                 journal_dev = new_decode_dev(journal_devnum);
3164         } else
3165                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3166
3167         really_read_only = bdev_read_only(sb->s_bdev);
3168
3169         /*
3170          * Are we loading a blank journal or performing recovery after a
3171          * crash?  For recovery, we need to check in advance whether we
3172          * can get read-write access to the device.
3173          */
3174         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3175                 if (sb->s_flags & MS_RDONLY) {
3176                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3177                                         "required on readonly filesystem");
3178                         if (really_read_only) {
3179                                 ext4_msg(sb, KERN_ERR, "write access "
3180                                         "unavailable, cannot proceed");
3181                                 return -EROFS;
3182                         }
3183                         ext4_msg(sb, KERN_INFO, "write access will "
3184                                "be enabled during recovery");
3185                 }
3186         }
3187
3188         if (journal_inum && journal_dev) {
3189                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3190                        "and inode journals!");
3191                 return -EINVAL;
3192         }
3193
3194         if (journal_inum) {
3195                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3196                         return -EINVAL;
3197         } else {
3198                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3199                         return -EINVAL;
3200         }
3201
3202         if (!(journal->j_flags & JBD2_BARRIER))
3203                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3204
3205         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3206                 err = jbd2_journal_update_format(journal);
3207                 if (err)  {
3208                         ext4_msg(sb, KERN_ERR, "error updating journal");
3209                         jbd2_journal_destroy(journal);
3210                         return err;
3211                 }
3212         }
3213
3214         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3215                 err = jbd2_journal_wipe(journal, !really_read_only);
3216         if (!err)
3217                 err = jbd2_journal_load(journal);
3218
3219         if (err) {
3220                 ext4_msg(sb, KERN_ERR, "error loading journal");
3221                 jbd2_journal_destroy(journal);
3222                 return err;
3223         }
3224
3225         EXT4_SB(sb)->s_journal = journal;
3226         ext4_clear_journal_err(sb, es);
3227
3228         if (journal_devnum &&
3229             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3230                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3231
3232                 /* Make sure we flush the recovery flag to disk. */
3233                 ext4_commit_super(sb, 1);
3234         }
3235
3236         return 0;
3237 }
3238
3239 static int ext4_commit_super(struct super_block *sb, int sync)
3240 {
3241         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3242         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3243         int error = 0;
3244
3245         if (!sbh)
3246                 return error;
3247         if (buffer_write_io_error(sbh)) {
3248                 /*
3249                  * Oh, dear.  A previous attempt to write the
3250                  * superblock failed.  This could happen because the
3251                  * USB device was yanked out.  Or it could happen to
3252                  * be a transient write error and maybe the block will
3253                  * be remapped.  Nothing we can do but to retry the
3254                  * write and hope for the best.
3255                  */
3256                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3257                        "superblock detected");
3258                 clear_buffer_write_io_error(sbh);
3259                 set_buffer_uptodate(sbh);
3260         }
3261         /*
3262          * If the file system is mounted read-only, don't update the
3263          * superblock write time.  This avoids updating the superblock
3264          * write time when we are mounting the root file system
3265          * read/only but we need to replay the journal; at that point,
3266          * for people who are east of GMT and who make their clock
3267          * tick in localtime for Windows bug-for-bug compatibility,
3268          * the clock is set in the future, and this will cause e2fsck
3269          * to complain and force a full file system check.
3270          */
3271         if (!(sb->s_flags & MS_RDONLY))
3272                 es->s_wtime = cpu_to_le32(get_seconds());
3273         es->s_kbytes_written =
3274                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3275                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3276                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3277         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3278                                         &EXT4_SB(sb)->s_freeblocks_counter));
3279         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3280                                         &EXT4_SB(sb)->s_freeinodes_counter));
3281         sb->s_dirt = 0;
3282         BUFFER_TRACE(sbh, "marking dirty");
3283         mark_buffer_dirty(sbh);
3284         if (sync) {
3285                 error = sync_dirty_buffer(sbh);
3286                 if (error)
3287                         return error;
3288
3289                 error = buffer_write_io_error(sbh);
3290                 if (error) {
3291                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3292                                "superblock");
3293                         clear_buffer_write_io_error(sbh);
3294                         set_buffer_uptodate(sbh);
3295                 }
3296         }
3297         return error;
3298 }
3299
3300 /*
3301  * Have we just finished recovery?  If so, and if we are mounting (or
3302  * remounting) the filesystem readonly, then we will end up with a
3303  * consistent fs on disk.  Record that fact.
3304  */
3305 static void ext4_mark_recovery_complete(struct super_block *sb,
3306                                         struct ext4_super_block *es)
3307 {
3308         journal_t *journal = EXT4_SB(sb)->s_journal;
3309
3310         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3311                 BUG_ON(journal != NULL);
3312                 return;
3313         }
3314         jbd2_journal_lock_updates(journal);
3315         if (jbd2_journal_flush(journal) < 0)
3316                 goto out;
3317
3318         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3319             sb->s_flags & MS_RDONLY) {
3320                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3321                 ext4_commit_super(sb, 1);
3322         }
3323
3324 out:
3325         jbd2_journal_unlock_updates(journal);
3326 }
3327
3328 /*
3329  * If we are mounting (or read-write remounting) a filesystem whose journal
3330  * has recorded an error from a previous lifetime, move that error to the
3331  * main filesystem now.
3332  */
3333 static void ext4_clear_journal_err(struct super_block *sb,
3334                                    struct ext4_super_block *es)
3335 {
3336         journal_t *journal;
3337         int j_errno;
3338         const char *errstr;
3339
3340         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3341
3342         journal = EXT4_SB(sb)->s_journal;
3343
3344         /*
3345          * Now check for any error status which may have been recorded in the
3346          * journal by a prior ext4_error() or ext4_abort()
3347          */
3348
3349         j_errno = jbd2_journal_errno(journal);
3350         if (j_errno) {
3351                 char nbuf[16];
3352
3353                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3354                 ext4_warning(sb, __func__, "Filesystem error recorded "
3355                              "from previous mount: %s", errstr);
3356                 ext4_warning(sb, __func__, "Marking fs in need of "
3357                              "filesystem check.");
3358
3359                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3360                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3361                 ext4_commit_super(sb, 1);
3362
3363                 jbd2_journal_clear_err(journal);
3364         }
3365 }
3366
3367 /*
3368  * Force the running and committing transactions to commit,
3369  * and wait on the commit.
3370  */
3371 int ext4_force_commit(struct super_block *sb)
3372 {
3373         journal_t *journal;
3374         int ret = 0;
3375
3376         if (sb->s_flags & MS_RDONLY)
3377                 return 0;
3378
3379         journal = EXT4_SB(sb)->s_journal;
3380         if (journal)
3381                 ret = ext4_journal_force_commit(journal);
3382
3383         return ret;
3384 }
3385
3386 static void ext4_write_super(struct super_block *sb)
3387 {
3388         lock_super(sb);
3389         ext4_commit_super(sb, 1);
3390         unlock_super(sb);
3391 }
3392
3393 static int ext4_sync_fs(struct super_block *sb, int wait)
3394 {
3395         int ret = 0;
3396         tid_t target;
3397         struct ext4_sb_info *sbi = EXT4_SB(sb);
3398
3399         trace_ext4_sync_fs(sb, wait);
3400         flush_workqueue(sbi->dio_unwritten_wq);
3401         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3402                 if (wait)
3403                         jbd2_log_wait_commit(sbi->s_journal, target);
3404         }
3405         return ret;
3406 }
3407
3408 /*
3409  * LVM calls this function before a (read-only) snapshot is created.  This
3410  * gives us a chance to flush the journal completely and mark the fs clean.
3411  */
3412 static int ext4_freeze(struct super_block *sb)
3413 {
3414         int error = 0;
3415         journal_t *journal;
3416
3417         if (sb->s_flags & MS_RDONLY)
3418                 return 0;
3419
3420         journal = EXT4_SB(sb)->s_journal;
3421
3422         /* Now we set up the journal barrier. */
3423         jbd2_journal_lock_updates(journal);
3424
3425         /*
3426          * Don't clear the needs_recovery flag if we failed to flush
3427          * the journal.
3428          */
3429         error = jbd2_journal_flush(journal);
3430         if (error < 0) {
3431         out:
3432                 jbd2_journal_unlock_updates(journal);
3433                 return error;
3434         }
3435
3436         /* Journal blocked and flushed, clear needs_recovery flag. */
3437         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3438         error = ext4_commit_super(sb, 1);
3439         if (error)
3440                 goto out;
3441         return 0;
3442 }
3443
3444 /*
3445  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3446  * flag here, even though the filesystem is not technically dirty yet.
3447  */
3448 static int ext4_unfreeze(struct super_block *sb)
3449 {
3450         if (sb->s_flags & MS_RDONLY)
3451                 return 0;
3452
3453         lock_super(sb);
3454         /* Reset the needs_recovery flag before the fs is unlocked. */
3455         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3456         ext4_commit_super(sb, 1);
3457         unlock_super(sb);
3458         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3459         return 0;
3460 }
3461
3462 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3463 {
3464         struct ext4_super_block *es;
3465         struct ext4_sb_info *sbi = EXT4_SB(sb);
3466         ext4_fsblk_t n_blocks_count = 0;
3467         unsigned long old_sb_flags;
3468         struct ext4_mount_options old_opts;
3469         ext4_group_t g;
3470         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3471         int err;
3472 #ifdef CONFIG_QUOTA
3473         int i;
3474 #endif
3475
3476         lock_kernel();
3477
3478         /* Store the original options */
3479         lock_super(sb);
3480         old_sb_flags = sb->s_flags;
3481         old_opts.s_mount_opt = sbi->s_mount_opt;
3482         old_opts.s_resuid = sbi->s_resuid;
3483         old_opts.s_resgid = sbi->s_resgid;
3484         old_opts.s_commit_interval = sbi->s_commit_interval;
3485         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3486         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3487 #ifdef CONFIG_QUOTA
3488         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3489         for (i = 0; i < MAXQUOTAS; i++)
3490                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3491 #endif
3492         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3493                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3494
3495         /*
3496          * Allow the "check" option to be passed as a remount option.
3497          */
3498         if (!parse_options(data, sb, NULL, &journal_ioprio,
3499                            &n_blocks_count, 1)) {
3500                 err = -EINVAL;
3501                 goto restore_opts;
3502         }
3503
3504         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3505                 ext4_abort(sb, __func__, "Abort forced by user");
3506
3507         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3508                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
3509
3510         es = sbi->s_es;
3511
3512         if (sbi->s_journal) {
3513                 ext4_init_journal_params(sb, sbi->s_journal);
3514                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3515         }
3516
3517         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3518                 n_blocks_count > ext4_blocks_count(es)) {
3519                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3520                         err = -EROFS;
3521                         goto restore_opts;
3522                 }
3523
3524                 if (*flags & MS_RDONLY) {
3525                         /*
3526                          * First of all, the unconditional stuff we have to do
3527                          * to disable replay of the journal when we next remount
3528                          */
3529                         sb->s_flags |= MS_RDONLY;
3530
3531                         /*
3532                          * OK, test if we are remounting a valid rw partition
3533                          * readonly, and if so set the rdonly flag and then
3534                          * mark the partition as valid again.
3535                          */
3536                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3537                             (sbi->s_mount_state & EXT4_VALID_FS))
3538                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3539
3540                         if (sbi->s_journal)
3541                                 ext4_mark_recovery_complete(sb, es);
3542                 } else {
3543                         /* Make sure we can mount this feature set readwrite */
3544                         if (!ext4_feature_set_ok(sb, 0)) {
3545                                 err = -EROFS;
3546                                 goto restore_opts;
3547                         }
3548                         /*
3549                          * Make sure the group descriptor checksums
3550                          * are sane.  If they aren't, refuse to remount r/w.
3551                          */
3552                         for (g = 0; g < sbi->s_groups_count; g++) {
3553                                 struct ext4_group_desc *gdp =
3554                                         ext4_get_group_desc(sb, g, NULL);
3555
3556                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3557                                         ext4_msg(sb, KERN_ERR,
3558                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3559                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3560                                                le16_to_cpu(gdp->bg_checksum));
3561                                         err = -EINVAL;
3562                                         goto restore_opts;
3563                                 }
3564                         }
3565
3566                         /*
3567                          * If we have an unprocessed orphan list hanging
3568                          * around from a previously readonly bdev mount,
3569                          * require a full umount/remount for now.
3570                          */
3571                         if (es->s_last_orphan) {
3572                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3573                                        "remount RDWR because of unprocessed "
3574                                        "orphan inode list.  Please "
3575                                        "umount/remount instead");
3576                                 err = -EINVAL;
3577                                 goto restore_opts;
3578                         }
3579
3580                         /*
3581                          * Mounting a RDONLY partition read-write, so reread
3582                          * and store the current valid flag.  (It may have
3583                          * been changed by e2fsck since we originally mounted
3584                          * the partition.)
3585                          */
3586                         if (sbi->s_journal)
3587                                 ext4_clear_journal_err(sb, es);
3588                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3589                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3590                                 goto restore_opts;
3591                         if (!ext4_setup_super(sb, es, 0))
3592                                 sb->s_flags &= ~MS_RDONLY;
3593                 }
3594         }
3595         ext4_setup_system_zone(sb);
3596         if (sbi->s_journal == NULL)
3597                 ext4_commit_super(sb, 1);
3598
3599 #ifdef CONFIG_QUOTA
3600         /* Release old quota file names */
3601         for (i = 0; i < MAXQUOTAS; i++)
3602                 if (old_opts.s_qf_names[i] &&
3603                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3604                         kfree(old_opts.s_qf_names[i]);
3605 #endif
3606         unlock_super(sb);
3607         unlock_kernel();
3608         return 0;
3609
3610 restore_opts:
3611         sb->s_flags = old_sb_flags;
3612         sbi->s_mount_opt = old_opts.s_mount_opt;
3613         sbi->s_resuid = old_opts.s_resuid;
3614         sbi->s_resgid = old_opts.s_resgid;
3615         sbi->s_commit_interval = old_opts.s_commit_interval;
3616         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3617         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3618 #ifdef CONFIG_QUOTA
3619         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3620         for (i = 0; i < MAXQUOTAS; i++) {
3621                 if (sbi->s_qf_names[i] &&
3622                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3623                         kfree(sbi->s_qf_names[i]);
3624                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3625         }
3626 #endif
3627         unlock_super(sb);
3628         unlock_kernel();
3629         return err;
3630 }
3631
3632 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3633 {
3634         struct super_block *sb = dentry->d_sb;
3635         struct ext4_sb_info *sbi = EXT4_SB(sb);
3636         struct ext4_super_block *es = sbi->s_es;
3637         u64 fsid;
3638
3639         if (test_opt(sb, MINIX_DF)) {
3640                 sbi->s_overhead_last = 0;
3641         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3642                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3643                 ext4_fsblk_t overhead = 0;
3644
3645                 /*
3646                  * Compute the overhead (FS structures).  This is constant
3647                  * for a given filesystem unless the number of block groups
3648                  * changes so we cache the previous value until it does.
3649                  */
3650
3651                 /*
3652                  * All of the blocks before first_data_block are
3653                  * overhead
3654                  */
3655                 overhead = le32_to_cpu(es->s_first_data_block);
3656
3657                 /*
3658                  * Add the overhead attributed to the superblock and
3659                  * block group descriptors.  If the sparse superblocks
3660                  * feature is turned on, then not all groups have this.
3661                  */
3662                 for (i = 0; i < ngroups; i++) {
3663                         overhead += ext4_bg_has_super(sb, i) +
3664                                 ext4_bg_num_gdb(sb, i);
3665                         cond_resched();
3666                 }
3667
3668                 /*
3669                  * Every block group has an inode bitmap, a block
3670                  * bitmap, and an inode table.
3671                  */
3672                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3673                 sbi->s_overhead_last = overhead;
3674                 smp_wmb();
3675                 sbi->s_blocks_last = ext4_blocks_count(es);
3676         }
3677
3678         buf->f_type = EXT4_SUPER_MAGIC;
3679         buf->f_bsize = sb->s_blocksize;
3680         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3681         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3682                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3683         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3684         if (buf->f_bfree < ext4_r_blocks_count(es))
3685                 buf->f_bavail = 0;
3686         buf->f_files = le32_to_cpu(es->s_inodes_count);
3687         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3688         buf->f_namelen = EXT4_NAME_LEN;
3689         fsid = le64_to_cpup((void *)es->s_uuid) ^
3690                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3691         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3692         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3693
3694         return 0;
3695 }
3696
3697 /* Helper function for writing quotas on sync - we need to start transaction
3698  * before quota file is locked for write. Otherwise the are possible deadlocks:
3699  * Process 1                         Process 2
3700  * ext4_create()                     quota_sync()
3701  *   jbd2_journal_start()                  write_dquot()
3702  *   dquot_initialize()                         down(dqio_mutex)
3703  *     down(dqio_mutex)                    jbd2_journal_start()
3704  *
3705  */
3706
3707 #ifdef CONFIG_QUOTA
3708
3709 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3710 {
3711         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3712 }
3713
3714 static int ext4_write_dquot(struct dquot *dquot)
3715 {
3716         int ret, err;
3717         handle_t *handle;
3718         struct inode *inode;
3719
3720         inode = dquot_to_inode(dquot);
3721         handle = ext4_journal_start(inode,
3722                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3723         if (IS_ERR(handle))
3724                 return PTR_ERR(handle);
3725         ret = dquot_commit(dquot);
3726         err = ext4_journal_stop(handle);
3727         if (!ret)
3728                 ret = err;
3729         return ret;
3730 }
3731
3732 static int ext4_acquire_dquot(struct dquot *dquot)
3733 {
3734         int ret, err;
3735         handle_t *handle;
3736
3737         handle = ext4_journal_start(dquot_to_inode(dquot),
3738                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3739         if (IS_ERR(handle))
3740                 return PTR_ERR(handle);
3741         ret = dquot_acquire(dquot);
3742         err = ext4_journal_stop(handle);
3743         if (!ret)
3744                 ret = err;
3745         return ret;
3746 }
3747
3748 static int ext4_release_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_DEL_BLOCKS(dquot->dq_sb));
3755         if (IS_ERR(handle)) {
3756                 /* Release dquot anyway to avoid endless cycle in dqput() */
3757                 dquot_release(dquot);
3758                 return PTR_ERR(handle);
3759         }
3760         ret = dquot_release(dquot);
3761         err = ext4_journal_stop(handle);
3762         if (!ret)
3763                 ret = err;
3764         return ret;
3765 }
3766
3767 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3768 {
3769         /* Are we journaling quotas? */
3770         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3771             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3772                 dquot_mark_dquot_dirty(dquot);
3773                 return ext4_write_dquot(dquot);
3774         } else {
3775                 return dquot_mark_dquot_dirty(dquot);
3776         }
3777 }
3778
3779 static int ext4_write_info(struct super_block *sb, int type)
3780 {
3781         int ret, err;
3782         handle_t *handle;
3783
3784         /* Data block + inode block */
3785         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3786         if (IS_ERR(handle))
3787                 return PTR_ERR(handle);
3788         ret = dquot_commit_info(sb, type);
3789         err = ext4_journal_stop(handle);
3790         if (!ret)
3791                 ret = err;
3792         return ret;
3793 }
3794
3795 /*
3796  * Turn on quotas during mount time - we need to find
3797  * the quota file and such...
3798  */
3799 static int ext4_quota_on_mount(struct super_block *sb, int type)
3800 {
3801         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3802                                   EXT4_SB(sb)->s_jquota_fmt, type);
3803 }
3804
3805 /*
3806  * Standard function to be called on quota_on
3807  */
3808 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3809                          char *name, int remount)
3810 {
3811         int err;
3812         struct path path;
3813
3814         if (!test_opt(sb, QUOTA))
3815                 return -EINVAL;
3816         /* When remounting, no checks are needed and in fact, name is NULL */
3817         if (remount)
3818                 return vfs_quota_on(sb, type, format_id, name, remount);
3819
3820         err = kern_path(name, LOOKUP_FOLLOW, &path);
3821         if (err)
3822                 return err;
3823
3824         /* Quotafile not on the same filesystem? */
3825         if (path.mnt->mnt_sb != sb) {
3826                 path_put(&path);
3827                 return -EXDEV;
3828         }
3829         /* Journaling quota? */
3830         if (EXT4_SB(sb)->s_qf_names[type]) {
3831                 /* Quotafile not in fs root? */
3832                 if (path.dentry->d_parent != sb->s_root)
3833                         ext4_msg(sb, KERN_WARNING,
3834                                 "Quota file not on filesystem root. "
3835                                 "Journaled quota will not work");
3836         }
3837
3838         /*
3839          * When we journal data on quota file, we have to flush journal to see
3840          * all updates to the file when we bypass pagecache...
3841          */
3842         if (EXT4_SB(sb)->s_journal &&
3843             ext4_should_journal_data(path.dentry->d_inode)) {
3844                 /*
3845                  * We don't need to lock updates but journal_flush() could
3846                  * otherwise be livelocked...
3847                  */
3848                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3849                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3850                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3851                 if (err) {
3852                         path_put(&path);
3853                         return err;
3854                 }
3855         }
3856
3857         err = vfs_quota_on_path(sb, type, format_id, &path);
3858         path_put(&path);
3859         return err;
3860 }
3861
3862 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3863  * acquiring the locks... As quota files are never truncated and quota code
3864  * itself serializes the operations (and noone else should touch the files)
3865  * we don't have to be afraid of races */
3866 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3867                                size_t len, loff_t off)
3868 {
3869         struct inode *inode = sb_dqopt(sb)->files[type];
3870         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3871         int err = 0;
3872         int offset = off & (sb->s_blocksize - 1);
3873         int tocopy;
3874         size_t toread;
3875         struct buffer_head *bh;
3876         loff_t i_size = i_size_read(inode);
3877
3878         if (off > i_size)
3879                 return 0;
3880         if (off+len > i_size)
3881                 len = i_size-off;
3882         toread = len;
3883         while (toread > 0) {
3884                 tocopy = sb->s_blocksize - offset < toread ?
3885                                 sb->s_blocksize - offset : toread;
3886                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3887                 if (err)
3888                         return err;
3889                 if (!bh)        /* A hole? */
3890                         memset(data, 0, tocopy);
3891                 else
3892                         memcpy(data, bh->b_data+offset, tocopy);
3893                 brelse(bh);
3894                 offset = 0;
3895                 toread -= tocopy;
3896                 data += tocopy;
3897                 blk++;
3898         }
3899         return len;
3900 }
3901
3902 /* Write to quotafile (we know the transaction is already started and has
3903  * enough credits) */
3904 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3905                                 const char *data, size_t len, loff_t off)
3906 {
3907         struct inode *inode = sb_dqopt(sb)->files[type];
3908         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3909         int err = 0;
3910         int offset = off & (sb->s_blocksize - 1);
3911         int tocopy;
3912         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3913         size_t towrite = len;
3914         struct buffer_head *bh;
3915         handle_t *handle = journal_current_handle();
3916
3917         if (EXT4_SB(sb)->s_journal && !handle) {
3918                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3919                         " cancelled because transaction is not started",
3920                         (unsigned long long)off, (unsigned long long)len);
3921                 return -EIO;
3922         }
3923         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3924         while (towrite > 0) {
3925                 tocopy = sb->s_blocksize - offset < towrite ?
3926                                 sb->s_blocksize - offset : towrite;
3927                 bh = ext4_bread(handle, inode, blk, 1, &err);
3928                 if (!bh)
3929                         goto out;
3930                 if (journal_quota) {
3931                         err = ext4_journal_get_write_access(handle, bh);
3932                         if (err) {
3933                                 brelse(bh);
3934                                 goto out;
3935                         }
3936                 }
3937                 lock_buffer(bh);
3938                 memcpy(bh->b_data+offset, data, tocopy);
3939                 flush_dcache_page(bh->b_page);
3940                 unlock_buffer(bh);
3941                 if (journal_quota)
3942                         err = ext4_handle_dirty_metadata(handle, NULL, bh);
3943                 else {
3944                         /* Always do at least ordered writes for quotas */
3945                         err = ext4_jbd2_file_inode(handle, inode);
3946                         mark_buffer_dirty(bh);
3947                 }
3948                 brelse(bh);
3949                 if (err)
3950                         goto out;
3951                 offset = 0;
3952                 towrite -= tocopy;
3953                 data += tocopy;
3954                 blk++;
3955         }
3956 out:
3957         if (len == towrite) {
3958                 mutex_unlock(&inode->i_mutex);
3959                 return err;
3960         }
3961         if (inode->i_size < off+len-towrite) {
3962                 i_size_write(inode, off+len-towrite);
3963                 EXT4_I(inode)->i_disksize = inode->i_size;
3964         }
3965         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3966         ext4_mark_inode_dirty(handle, inode);
3967         mutex_unlock(&inode->i_mutex);
3968         return len - towrite;
3969 }
3970
3971 #endif
3972
3973 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
3974                        const char *dev_name, void *data, struct vfsmount *mnt)
3975 {
3976         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
3977 }
3978
3979 #if !defined(CONTIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
3980 static struct file_system_type ext2_fs_type = {
3981         .owner          = THIS_MODULE,
3982         .name           = "ext2",
3983         .get_sb         = ext4_get_sb,
3984         .kill_sb        = kill_block_super,
3985         .fs_flags       = FS_REQUIRES_DEV,
3986 };
3987
3988 static inline void register_as_ext2(void)
3989 {
3990         int err = register_filesystem(&ext2_fs_type);
3991         if (err)
3992                 printk(KERN_WARNING
3993                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
3994 }
3995
3996 static inline void unregister_as_ext2(void)
3997 {
3998         unregister_filesystem(&ext2_fs_type);
3999 }
4000 MODULE_ALIAS("ext2");
4001 #else
4002 static inline void register_as_ext2(void) { }
4003 static inline void unregister_as_ext2(void) { }
4004 #endif
4005
4006 #if !defined(CONTIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4007 static struct file_system_type ext3_fs_type = {
4008         .owner          = THIS_MODULE,
4009         .name           = "ext3",
4010         .get_sb         = ext4_get_sb,
4011         .kill_sb        = kill_block_super,
4012         .fs_flags       = FS_REQUIRES_DEV,
4013 };
4014
4015 static inline void register_as_ext3(void)
4016 {
4017         int err = register_filesystem(&ext3_fs_type);
4018         if (err)
4019                 printk(KERN_WARNING
4020                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4021 }
4022
4023 static inline void unregister_as_ext3(void)
4024 {
4025         unregister_filesystem(&ext3_fs_type);
4026 }
4027 MODULE_ALIAS("ext3");
4028 #else
4029 static inline void register_as_ext3(void) { }
4030 static inline void unregister_as_ext3(void) { }
4031 #endif
4032
4033 static struct file_system_type ext4_fs_type = {
4034         .owner          = THIS_MODULE,
4035         .name           = "ext4",
4036         .get_sb         = ext4_get_sb,
4037         .kill_sb        = kill_block_super,
4038         .fs_flags       = FS_REQUIRES_DEV,
4039 };
4040
4041 static int __init init_ext4_fs(void)
4042 {
4043         int err;
4044
4045         err = init_ext4_system_zone();
4046         if (err)
4047                 return err;
4048         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4049         if (!ext4_kset)
4050                 goto out4;
4051         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4052         err = init_ext4_mballoc();
4053         if (err)
4054                 goto out3;
4055
4056         err = init_ext4_xattr();
4057         if (err)
4058                 goto out2;
4059         err = init_inodecache();
4060         if (err)
4061                 goto out1;
4062         register_as_ext2();
4063         register_as_ext3();
4064         err = register_filesystem(&ext4_fs_type);
4065         if (err)
4066                 goto out;
4067         return 0;
4068 out:
4069         unregister_as_ext2();
4070         unregister_as_ext3();
4071         destroy_inodecache();
4072 out1:
4073         exit_ext4_xattr();
4074 out2:
4075         exit_ext4_mballoc();
4076 out3:
4077         remove_proc_entry("fs/ext4", NULL);
4078         kset_unregister(ext4_kset);
4079 out4:
4080         exit_ext4_system_zone();
4081         return err;
4082 }
4083
4084 static void __exit exit_ext4_fs(void)
4085 {
4086         unregister_as_ext2();
4087         unregister_as_ext3();
4088         unregister_filesystem(&ext4_fs_type);
4089         destroy_inodecache();
4090         exit_ext4_xattr();
4091         exit_ext4_mballoc();
4092         remove_proc_entry("fs/ext4", NULL);
4093         kset_unregister(ext4_kset);
4094         exit_ext4_system_zone();
4095 }
4096
4097 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4098 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4099 MODULE_LICENSE("GPL");
4100 module_init(init_ext4_fs)
4101 module_exit(exit_ext4_fs)