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 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
  72                        const char *dev_name, void *data, struct vfsmount *mnt);
  73
  74 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  75 static struct file_system_type ext3_fs_type = {
  76         .owner          = THIS_MODULE,
  77         .name           = "ext3",
  78         .get_sb         = ext4_get_sb,
  79         .kill_sb        = kill_block_super,
  80         .fs_flags       = FS_REQUIRES_DEV,
  81 };
  82 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  83 #else
  84 #define IS_EXT3_SB(sb) (0)
  85 #endif
  86
  87 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  88                                struct ext4_group_desc *bg)
  89 {
  90         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  91                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  92                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  93 }
  94
  95 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  96                                struct ext4_group_desc *bg)
  97 {
  98         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  99                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 100                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 101 }
 102
 103 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 104                               struct ext4_group_desc *bg)
 105 {
 106         return le32_to_cpu(bg->bg_inode_table_lo) |
 107                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 108                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 109 }
 110
 111 __u32 ext4_free_blks_count(struct super_block *sb,
 112                               struct ext4_group_desc *bg)
 113 {
 114         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 115                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 116                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 117 }
 118
 119 __u32 ext4_free_inodes_count(struct super_block *sb,
 120                               struct ext4_group_desc *bg)
 121 {
 122         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 123                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 124                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 125 }
 126
 127 __u32 ext4_used_dirs_count(struct super_block *sb,
 128                               struct ext4_group_desc *bg)
 129 {
 130         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 131                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 132                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 133 }
 134
 135 __u32 ext4_itable_unused_count(struct super_block *sb,
 136                               struct ext4_group_desc *bg)
 137 {
 138         return le16_to_cpu(bg->bg_itable_unused_lo) |
 139                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 140                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 141 }
 142
 143 void ext4_block_bitmap_set(struct super_block *sb,
 144                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 145 {
 146         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 147         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 148                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 149 }
 150
 151 void ext4_inode_bitmap_set(struct super_block *sb,
 152                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 153 {
 154         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 155         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 156                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 157 }
 158
 159 void ext4_inode_table_set(struct super_block *sb,
 160                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 161 {
 162         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 163         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 164                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 165 }
 166
 167 void ext4_free_blks_set(struct super_block *sb,
 168                           struct ext4_group_desc *bg, __u32 count)
 169 {
 170         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 171         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 172                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 173 }
 174
 175 void ext4_free_inodes_set(struct super_block *sb,
 176                           struct ext4_group_desc *bg, __u32 count)
 177 {
 178         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 179         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 180                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 181 }
 182
 183 void ext4_used_dirs_set(struct super_block *sb,
 184                           struct ext4_group_desc *bg, __u32 count)
 185 {
 186         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 187         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 188                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 189 }
 190
 191 void ext4_itable_unused_set(struct super_block *sb,
 192                           struct ext4_group_desc *bg, __u32 count)
 193 {
 194         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 195         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 196                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 197 }
 198
 199
 200 /* Just increment the non-pointer handle value */
 201 static handle_t *ext4_get_nojournal(void)
 202 {
 203         handle_t *handle = current->journal_info;
 204         unsigned long ref_cnt = (unsigned long)handle;
 205
 206         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 207
 208         ref_cnt++;
 209         handle = (handle_t *)ref_cnt;
 210
 211         current->journal_info = handle;
 212         return handle;
 213 }
 214
 215
 216 /* Decrement the non-pointer handle value */
 217 static void ext4_put_nojournal(handle_t *handle)
 218 {
 219         unsigned long ref_cnt = (unsigned long)handle;
 220
 221         BUG_ON(ref_cnt == 0);
 222
 223         ref_cnt--;
 224         handle = (handle_t *)ref_cnt;
 225
 226         current->journal_info = handle;
 227 }
 228
 229 /*
 230  * Wrappers for jbd2_journal_start/end.
 231  *
 232  * The only special thing we need to do here is to make sure that all
 233  * journal_end calls result in the superblock being marked dirty, so
 234  * that sync() will call the filesystem's write_super callback if
 235  * appropriate.
 236  */
 237 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 238 {
 239         journal_t *journal;
 240
 241         if (sb->s_flags & MS_RDONLY)
 242                 return ERR_PTR(-EROFS);
 243
 244         vfs_check_frozen(sb, SB_FREEZE_WRITE);
 245         /* Special case here: if the journal has aborted behind our
 246          * backs (eg. EIO in the commit thread), then we still need to
 247          * take the FS itself readonly cleanly. */
 248         journal = EXT4_SB(sb)->s_journal;
 249         if (journal) {
 250                 if (is_journal_aborted(journal)) {
 251                         ext4_abort(sb, "Detected aborted journal");
 252                         return ERR_PTR(-EROFS);
 253                 }
 254                 return jbd2_journal_start(journal, nblocks);
 255         }
 256         return ext4_get_nojournal();
 257 }
 258
 259 /*
 260  * The only special thing we need to do here is to make sure that all
 261  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 262  * that sync() will call the filesystem's write_super callback if
 263  * appropriate.
 264  */
 265 int __ext4_journal_stop(const char *where, handle_t *handle)
 266 {
 267         struct super_block *sb;
 268         int err;
 269         int rc;
 270
 271         if (!ext4_handle_valid(handle)) {
 272                 ext4_put_nojournal(handle);
 273                 return 0;
 274         }
 275         sb = handle->h_transaction->t_journal->j_private;
 276         err = handle->h_err;
 277         rc = jbd2_journal_stop(handle);
 278
 279         if (!err)
 280                 err = rc;
 281         if (err)
 282                 __ext4_std_error(sb, where, err);
 283         return err;
 284 }
 285
 286 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 287                 struct buffer_head *bh, handle_t *handle, int err)
 288 {
 289         char nbuf[16];
 290         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 291
 292         BUG_ON(!ext4_handle_valid(handle));
 293
 294         if (bh)
 295                 BUFFER_TRACE(bh, "abort");
 296
 297         if (!handle->h_err)
 298                 handle->h_err = err;
 299
 300         if (is_handle_aborted(handle))
 301                 return;
 302
 303         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 304                caller, errstr, err_fn);
 305
 306         jbd2_journal_abort_handle(handle);
 307 }
 308
 309 /* Deal with the reporting of failure conditions on a filesystem such as
 310  * inconsistencies detected or read IO failures.
 311  *
 312  * On ext2, we can store the error state of the filesystem in the
 313  * superblock.  That is not possible on ext4, because we may have other
 314  * write ordering constraints on the superblock which prevent us from
 315  * writing it out straight away; and given that the journal is about to
 316  * be aborted, we can't rely on the current, or future, transactions to
 317  * write out the superblock safely.
 318  *
 319  * We'll just use the jbd2_journal_abort() error code to record an error in
 320  * the journal instead.  On recovery, the journal will complain about
 321  * that error until we've noted it down and cleared it.
 322  */
 323
 324 static void ext4_handle_error(struct super_block *sb)
 325 {
 326         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 327
 328         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 329         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 330
 331         if (sb->s_flags & MS_RDONLY)
 332                 return;
 333
 334         if (!test_opt(sb, ERRORS_CONT)) {
 335                 journal_t *journal = EXT4_SB(sb)->s_journal;
 336
 337                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 338                 if (journal)
 339                         jbd2_journal_abort(journal, -EIO);
 340         }
 341         if (test_opt(sb, ERRORS_RO)) {
 342                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 343                 sb->s_flags |= MS_RDONLY;
 344         }
 345         ext4_commit_super(sb, 1);
 346         if (test_opt(sb, ERRORS_PANIC))
 347                 panic("EXT4-fs (device %s): panic forced after error\n",
 348                         sb->s_id);
 349 }
 350
 351 void __ext4_error(struct super_block *sb, const char *function,
 352                 const char *fmt, ...)
 353 {
 354         va_list args;
 355
 356         va_start(args, fmt);
 357         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 358         vprintk(fmt, args);
 359         printk("\n");
 360         va_end(args);
 361
 362         ext4_handle_error(sb);
 363 }
 364
 365 void ext4_error_inode(const char *function, struct inode *inode,
 366                       const char *fmt, ...)
 367 {
 368         va_list args;
 369
 370         va_start(args, fmt);
 371         printk(KERN_CRIT "EXT4-fs error (device %s): %s: inode #%lu: (comm %s) ",
 372                inode->i_sb->s_id, function, inode->i_ino, current->comm);
 373         vprintk(fmt, args);
 374         printk("\n");
 375         va_end(args);
 376
 377         ext4_handle_error(inode->i_sb);
 378 }
 379
 380 void ext4_error_file(const char *function, struct file *file,
 381                      const char *fmt, ...)
 382 {
 383         va_list args;
 384         struct inode *inode = file->f_dentry->d_inode;
 385         char pathname[80], *path;
 386
 387         va_start(args, fmt);
 388         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 389         if (!path)
 390                 path = "(unknown)";
 391         printk(KERN_CRIT
 392                "EXT4-fs error (device %s): %s: inode #%lu (comm %s path %s): ",
 393                inode->i_sb->s_id, function, inode->i_ino, current->comm, path);
 394         vprintk(fmt, args);
 395         printk("\n");
 396         va_end(args);
 397
 398         ext4_handle_error(inode->i_sb);
 399 }
 400
 401 static const char *ext4_decode_error(struct super_block *sb, int errno,
 402                                      char nbuf[16])
 403 {
 404         char *errstr = NULL;
 405
 406         switch (errno) {
 407         case -EIO:
 408                 errstr = "IO failure";
 409                 break;
 410         case -ENOMEM:
 411                 errstr = "Out of memory";
 412                 break;
 413         case -EROFS:
 414                 if (!sb || (EXT4_SB(sb)->s_journal &&
 415                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 416                         errstr = "Journal has aborted";
 417                 else
 418                         errstr = "Readonly filesystem";
 419                 break;
 420         default:
 421                 /* If the caller passed in an extra buffer for unknown
 422                  * errors, textualise them now.  Else we just return
 423                  * NULL. */
 424                 if (nbuf) {
 425                         /* Check for truncated error codes... */
 426                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 427                                 errstr = nbuf;
 428                 }
 429                 break;
 430         }
 431
 432         return errstr;
 433 }
 434
 435 /* __ext4_std_error decodes expected errors from journaling functions
 436  * automatically and invokes the appropriate error response.  */
 437
 438 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 439 {
 440         char nbuf[16];
 441         const char *errstr;
 442
 443         /* Special case: if the error is EROFS, and we're not already
 444          * inside a transaction, then there's really no point in logging
 445          * an error. */
 446         if (errno == -EROFS && journal_current_handle() == NULL &&
 447             (sb->s_flags & MS_RDONLY))
 448                 return;
 449
 450         errstr = ext4_decode_error(sb, errno, nbuf);
 451         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 452                sb->s_id, function, errstr);
 453
 454         ext4_handle_error(sb);
 455 }
 456
 457 /*
 458  * ext4_abort is a much stronger failure handler than ext4_error.  The
 459  * abort function may be used to deal with unrecoverable failures such
 460  * as journal IO errors or ENOMEM at a critical moment in log management.
 461  *
 462  * We unconditionally force the filesystem into an ABORT|READONLY state,
 463  * unless the error response on the fs has been set to panic in which
 464  * case we take the easy way out and panic immediately.
 465  */
 466
 467 void __ext4_abort(struct super_block *sb, const char *function,
 468                   const char *fmt, ...)
 469 {
 470         va_list args;
 471
 472         va_start(args, fmt);
 473         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 474         vprintk(fmt, args);
 475         printk("\n");
 476         va_end(args);
 477
 478         if (test_opt(sb, ERRORS_PANIC))
 479                 panic("EXT4-fs panic from previous error\n");
 480
 481         if (sb->s_flags & MS_RDONLY)
 482                 return;
 483
 484         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 485         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 486         sb->s_flags |= MS_RDONLY;
 487         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 488         if (EXT4_SB(sb)->s_journal)
 489                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 490 }
 491
 492 void ext4_msg (struct super_block * sb, const char *prefix,
 493                    const char *fmt, ...)
 494 {
 495         va_list args;
 496
 497         va_start(args, fmt);
 498         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 499         vprintk(fmt, args);
 500         printk("\n");
 501         va_end(args);
 502 }
 503
 504 void __ext4_warning(struct super_block *sb, const char *function,
 505                   const char *fmt, ...)
 506 {
 507         va_list args;
 508
 509         va_start(args, fmt);
 510         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 511                sb->s_id, function);
 512         vprintk(fmt, args);
 513         printk("\n");
 514         va_end(args);
 515 }
 516
 517 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
 518                            const char *function, const char *fmt, ...)
 519 __releases(bitlock)
 520 __acquires(bitlock)
 521 {
 522         va_list args;
 523         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 524
 525         va_start(args, fmt);
 526         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 527         vprintk(fmt, args);
 528         printk("\n");
 529         va_end(args);
 530
 531         if (test_opt(sb, ERRORS_CONT)) {
 532                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 533                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 534                 ext4_commit_super(sb, 0);
 535                 return;
 536         }
 537         ext4_unlock_group(sb, grp);
 538         ext4_handle_error(sb);
 539         /*
 540          * We only get here in the ERRORS_RO case; relocking the group
 541          * may be dangerous, but nothing bad will happen since the
 542          * filesystem will have already been marked read/only and the
 543          * journal has been aborted.  We return 1 as a hint to callers
 544          * who might what to use the return value from
 545          * ext4_grp_locked_error() to distinguish beween the
 546          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 547          * aggressively from the ext4 function in question, with a
 548          * more appropriate error code.
 549          */
 550         ext4_lock_group(sb, grp);
 551         return;
 552 }
 553
 554 void ext4_update_dynamic_rev(struct super_block *sb)
 555 {
 556         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 557
 558         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 559                 return;
 560
 561         ext4_warning(sb,
 562                      "updating to rev %d because of new feature flag, "
 563                      "running e2fsck is recommended",
 564                      EXT4_DYNAMIC_REV);
 565
 566         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 567         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 568         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 569         /* leave es->s_feature_*compat flags alone */
 570         /* es->s_uuid will be set by e2fsck if empty */
 571
 572         /*
 573          * The rest of the superblock fields should be zero, and if not it
 574          * means they are likely already in use, so leave them alone.  We
 575          * can leave it up to e2fsck to clean up any inconsistencies there.
 576          */
 577 }
 578
 579 /*
 580  * Open the external journal device
 581  */
 582 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 583 {
 584         struct block_device *bdev;
 585         char b[BDEVNAME_SIZE];
 586
 587         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 588         if (IS_ERR(bdev))
 589                 goto fail;
 590         return bdev;
 591
 592 fail:
 593         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 594                         __bdevname(dev, b), PTR_ERR(bdev));
 595         return NULL;
 596 }
 597
 598 /*
 599  * Release the journal device
 600  */
 601 static int ext4_blkdev_put(struct block_device *bdev)
 602 {
 603         bd_release(bdev);
 604         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 605 }
 606
 607 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 608 {
 609         struct block_device *bdev;
 610         int ret = -ENODEV;
 611
 612         bdev = sbi->journal_bdev;
 613         if (bdev) {
 614                 ret = ext4_blkdev_put(bdev);
 615                 sbi->journal_bdev = NULL;
 616         }
 617         return ret;
 618 }
 619
 620 static inline struct inode *orphan_list_entry(struct list_head *l)
 621 {
 622         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 623 }
 624
 625 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 626 {
 627         struct list_head *l;
 628
 629         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 630                  le32_to_cpu(sbi->s_es->s_last_orphan));
 631
 632         printk(KERN_ERR "sb_info orphan list:\n");
 633         list_for_each(l, &sbi->s_orphan) {
 634                 struct inode *inode = orphan_list_entry(l);
 635                 printk(KERN_ERR "  "
 636                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 637                        inode->i_sb->s_id, inode->i_ino, inode,
 638                        inode->i_mode, inode->i_nlink,
 639                        NEXT_ORPHAN(inode));
 640         }
 641 }
 642
 643 static void ext4_put_super(struct super_block *sb)
 644 {
 645         struct ext4_sb_info *sbi = EXT4_SB(sb);
 646         struct ext4_super_block *es = sbi->s_es;
 647         int i, err;
 648
 649         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 650
 651         flush_workqueue(sbi->dio_unwritten_wq);
 652         destroy_workqueue(sbi->dio_unwritten_wq);
 653
 654         lock_super(sb);
 655         lock_kernel();
 656         if (sb->s_dirt)
 657                 ext4_commit_super(sb, 1);
 658
 659         if (sbi->s_journal) {
 660                 err = jbd2_journal_destroy(sbi->s_journal);
 661                 sbi->s_journal = NULL;
 662                 if (err < 0)
 663                         ext4_abort(sb, "Couldn't clean up the journal");
 664         }
 665
 666         ext4_release_system_zone(sb);
 667         ext4_mb_release(sb);
 668         ext4_ext_release(sb);
 669         ext4_xattr_put_super(sb);
 670
 671         if (!(sb->s_flags & MS_RDONLY)) {
 672                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 673                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 674                 ext4_commit_super(sb, 1);
 675         }
 676         if (sbi->s_proc) {
 677                 remove_proc_entry(sb->s_id, ext4_proc_root);
 678         }
 679         kobject_del(&sbi->s_kobj);
 680
 681         for (i = 0; i < sbi->s_gdb_count; i++)
 682                 brelse(sbi->s_group_desc[i]);
 683         kfree(sbi->s_group_desc);
 684         if (is_vmalloc_addr(sbi->s_flex_groups))
 685                 vfree(sbi->s_flex_groups);
 686         else
 687                 kfree(sbi->s_flex_groups);
 688         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 689         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 690         percpu_counter_destroy(&sbi->s_dirs_counter);
 691         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 692         brelse(sbi->s_sbh);
 693 #ifdef CONFIG_QUOTA
 694         for (i = 0; i < MAXQUOTAS; i++)
 695                 kfree(sbi->s_qf_names[i]);
 696 #endif
 697
 698         /* Debugging code just in case the in-memory inode orphan list
 699          * isn't empty.  The on-disk one can be non-empty if we've
 700          * detected an error and taken the fs readonly, but the
 701          * in-memory list had better be clean by this point. */
 702         if (!list_empty(&sbi->s_orphan))
 703                 dump_orphan_list(sb, sbi);
 704         J_ASSERT(list_empty(&sbi->s_orphan));
 705
 706         invalidate_bdev(sb->s_bdev);
 707         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 708                 /*
 709                  * Invalidate the journal device's buffers.  We don't want them
 710                  * floating about in memory - the physical journal device may
 711                  * hotswapped, and it breaks the `ro-after' testing code.
 712                  */
 713                 sync_blockdev(sbi->journal_bdev);
 714                 invalidate_bdev(sbi->journal_bdev);
 715                 ext4_blkdev_remove(sbi);
 716         }
 717         sb->s_fs_info = NULL;
 718         /*
 719          * Now that we are completely done shutting down the
 720          * superblock, we need to actually destroy the kobject.
 721          */
 722         unlock_kernel();
 723         unlock_super(sb);
 724         kobject_put(&sbi->s_kobj);
 725         wait_for_completion(&sbi->s_kobj_unregister);
 726         kfree(sbi->s_blockgroup_lock);
 727         kfree(sbi);
 728 }
 729
 730 static struct kmem_cache *ext4_inode_cachep;
 731
 732 /*
 733  * Called inside transaction, so use GFP_NOFS
 734  */
 735 static struct inode *ext4_alloc_inode(struct super_block *sb)
 736 {
 737         struct ext4_inode_info *ei;
 738
 739         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 740         if (!ei)
 741                 return NULL;
 742
 743         ei->vfs_inode.i_version = 1;
 744         ei->vfs_inode.i_data.writeback_index = 0;
 745         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 746         INIT_LIST_HEAD(&ei->i_prealloc_list);
 747         spin_lock_init(&ei->i_prealloc_lock);
 748         /*
 749          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 750          * therefore it can be null here.  Don't check it, just initialize
 751          * jinode.
 752          */
 753         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 754         ei->i_reserved_data_blocks = 0;
 755         ei->i_reserved_meta_blocks = 0;
 756         ei->i_allocated_meta_blocks = 0;
 757         ei->i_da_metadata_calc_len = 0;
 758         ei->i_delalloc_reserved_flag = 0;
 759         spin_lock_init(&(ei->i_block_reservation_lock));
 760 #ifdef CONFIG_QUOTA
 761         ei->i_reserved_quota = 0;
 762 #endif
 763         INIT_LIST_HEAD(&ei->i_completed_io_list);
 764         spin_lock_init(&ei->i_completed_io_lock);
 765         ei->cur_aio_dio = NULL;
 766         ei->i_sync_tid = 0;
 767         ei->i_datasync_tid = 0;
 768
 769         return &ei->vfs_inode;
 770 }
 771
 772 static void ext4_destroy_inode(struct inode *inode)
 773 {
 774         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 775                 ext4_msg(inode->i_sb, KERN_ERR,
 776                          "Inode %lu (%p): orphan list check failed!",
 777                          inode->i_ino, EXT4_I(inode));
 778                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 779                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 780                                 true);
 781                 dump_stack();
 782         }
 783         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 784 }
 785
 786 static void init_once(void *foo)
 787 {
 788         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 789
 790         INIT_LIST_HEAD(&ei->i_orphan);
 791 #ifdef CONFIG_EXT4_FS_XATTR
 792         init_rwsem(&ei->xattr_sem);
 793 #endif
 794         init_rwsem(&ei->i_data_sem);
 795         inode_init_once(&ei->vfs_inode);
 796 }
 797
 798 static int init_inodecache(void)
 799 {
 800         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 801                                              sizeof(struct ext4_inode_info),
 802                                              0, (SLAB_RECLAIM_ACCOUNT|
 803                                                 SLAB_MEM_SPREAD),
 804                                              init_once);
 805         if (ext4_inode_cachep == NULL)
 806                 return -ENOMEM;
 807         return 0;
 808 }
 809
 810 static void destroy_inodecache(void)
 811 {
 812         kmem_cache_destroy(ext4_inode_cachep);
 813 }
 814
 815 static void ext4_clear_inode(struct inode *inode)
 816 {
 817         dquot_drop(inode);
 818         ext4_discard_preallocations(inode);
 819         if (EXT4_JOURNAL(inode))
 820                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 821                                        &EXT4_I(inode)->jinode);
 822 }
 823
 824 static inline void ext4_show_quota_options(struct seq_file *seq,
 825                                            struct super_block *sb)
 826 {
 827 #if defined(CONFIG_QUOTA)
 828         struct ext4_sb_info *sbi = EXT4_SB(sb);
 829
 830         if (sbi->s_jquota_fmt) {
 831                 char *fmtname = "";
 832
 833                 switch (sbi->s_jquota_fmt) {
 834                 case QFMT_VFS_OLD:
 835                         fmtname = "vfsold";
 836                         break;
 837                 case QFMT_VFS_V0:
 838                         fmtname = "vfsv0";
 839                         break;
 840                 case QFMT_VFS_V1:
 841                         fmtname = "vfsv1";
 842                         break;
 843                 }
 844                 seq_printf(seq, ",jqfmt=%s", fmtname);
 845         }
 846
 847         if (sbi->s_qf_names[USRQUOTA])
 848                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 849
 850         if (sbi->s_qf_names[GRPQUOTA])
 851                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 852
 853         if (test_opt(sb, USRQUOTA))
 854                 seq_puts(seq, ",usrquota");
 855
 856         if (test_opt(sb, GRPQUOTA))
 857                 seq_puts(seq, ",grpquota");
 858 #endif
 859 }
 860
 861 /*
 862  * Show an option if
 863  *  - it's set to a non-default value OR
 864  *  - if the per-sb default is different from the global default
 865  */
 866 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 867 {
 868         int def_errors;
 869         unsigned long def_mount_opts;
 870         struct super_block *sb = vfs->mnt_sb;
 871         struct ext4_sb_info *sbi = EXT4_SB(sb);
 872         struct ext4_super_block *es = sbi->s_es;
 873
 874         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 875         def_errors     = le16_to_cpu(es->s_errors);
 876
 877         if (sbi->s_sb_block != 1)
 878                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 879         if (test_opt(sb, MINIX_DF))
 880                 seq_puts(seq, ",minixdf");
 881         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 882                 seq_puts(seq, ",grpid");
 883         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 884                 seq_puts(seq, ",nogrpid");
 885         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 886             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 887                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 888         }
 889         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 890             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 891                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 892         }
 893         if (test_opt(sb, ERRORS_RO)) {
 894                 if (def_errors == EXT4_ERRORS_PANIC ||
 895                     def_errors == EXT4_ERRORS_CONTINUE) {
 896                         seq_puts(seq, ",errors=remount-ro");
 897                 }
 898         }
 899         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 900                 seq_puts(seq, ",errors=continue");
 901         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 902                 seq_puts(seq, ",errors=panic");
 903         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 904                 seq_puts(seq, ",nouid32");
 905         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 906                 seq_puts(seq, ",debug");
 907         if (test_opt(sb, OLDALLOC))
 908                 seq_puts(seq, ",oldalloc");
 909 #ifdef CONFIG_EXT4_FS_XATTR
 910         if (test_opt(sb, XATTR_USER) &&
 911                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 912                 seq_puts(seq, ",user_xattr");
 913         if (!test_opt(sb, XATTR_USER) &&
 914             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 915                 seq_puts(seq, ",nouser_xattr");
 916         }
 917 #endif
 918 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 919         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 920                 seq_puts(seq, ",acl");
 921         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 922                 seq_puts(seq, ",noacl");
 923 #endif
 924         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 925                 seq_printf(seq, ",commit=%u",
 926                            (unsigned) (sbi->s_commit_interval / HZ));
 927         }
 928         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 929                 seq_printf(seq, ",min_batch_time=%u",
 930                            (unsigned) sbi->s_min_batch_time);
 931         }
 932         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 933                 seq_printf(seq, ",max_batch_time=%u",
 934                            (unsigned) sbi->s_min_batch_time);
 935         }
 936
 937         /*
 938          * We're changing the default of barrier mount option, so
 939          * let's always display its mount state so it's clear what its
 940          * status is.
 941          */
 942         seq_puts(seq, ",barrier=");
 943         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 944         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 945                 seq_puts(seq, ",journal_async_commit");
 946         else if (test_opt(sb, JOURNAL_CHECKSUM))
 947                 seq_puts(seq, ",journal_checksum");
 948         if (test_opt(sb, I_VERSION))
 949                 seq_puts(seq, ",i_version");
 950         if (!test_opt(sb, DELALLOC))
 951                 seq_puts(seq, ",nodelalloc");
 952
 953
 954         if (sbi->s_stripe)
 955                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 956         /*
 957          * journal mode get enabled in different ways
 958          * So just print the value even if we didn't specify it
 959          */
 960         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 961                 seq_puts(seq, ",data=journal");
 962         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 963                 seq_puts(seq, ",data=ordered");
 964         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 965                 seq_puts(seq, ",data=writeback");
 966
 967         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 968                 seq_printf(seq, ",inode_readahead_blks=%u",
 969                            sbi->s_inode_readahead_blks);
 970
 971         if (test_opt(sb, DATA_ERR_ABORT))
 972                 seq_puts(seq, ",data_err=abort");
 973
 974         if (test_opt(sb, NO_AUTO_DA_ALLOC))
 975                 seq_puts(seq, ",noauto_da_alloc");
 976
 977         if (test_opt(sb, DISCARD))
 978                 seq_puts(seq, ",discard");
 979
 980         if (test_opt(sb, NOLOAD))
 981                 seq_puts(seq, ",norecovery");
 982
 983         if (test_opt(sb, DIOREAD_NOLOCK))
 984                 seq_puts(seq, ",dioread_nolock");
 985
 986         ext4_show_quota_options(seq, sb);
 987
 988         return 0;
 989 }
 990
 991 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 992                                         u64 ino, u32 generation)
 993 {
 994         struct inode *inode;
 995
 996         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 997                 return ERR_PTR(-ESTALE);
 998         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 999                 return ERR_PTR(-ESTALE);
1000
1001         /* iget isn't really right if the inode is currently unallocated!!
1002          *
1003          * ext4_read_inode will return a bad_inode if the inode had been
1004          * deleted, so we should be safe.
1005          *
1006          * Currently we don't know the generation for parent directory, so
1007          * a generation of 0 means "accept any"
1008          */
1009         inode = ext4_iget(sb, ino);
1010         if (IS_ERR(inode))
1011                 return ERR_CAST(inode);
1012         if (generation && inode->i_generation != generation) {
1013                 iput(inode);
1014                 return ERR_PTR(-ESTALE);
1015         }
1016
1017         return inode;
1018 }
1019
1020 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1021                                         int fh_len, int fh_type)
1022 {
1023         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1024                                     ext4_nfs_get_inode);
1025 }
1026
1027 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1028                                         int fh_len, int fh_type)
1029 {
1030         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1031                                     ext4_nfs_get_inode);
1032 }
1033
1034 /*
1035  * Try to release metadata pages (indirect blocks, directories) which are
1036  * mapped via the block device.  Since these pages could have journal heads
1037  * which would prevent try_to_free_buffers() from freeing them, we must use
1038  * jbd2 layer's try_to_free_buffers() function to release them.
1039  */
1040 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1041                                  gfp_t wait)
1042 {
1043         journal_t *journal = EXT4_SB(sb)->s_journal;
1044
1045         WARN_ON(PageChecked(page));
1046         if (!page_has_buffers(page))
1047                 return 0;
1048         if (journal)
1049                 return jbd2_journal_try_to_free_buffers(journal, page,
1050                                                         wait & ~__GFP_WAIT);
1051         return try_to_free_buffers(page);
1052 }
1053
1054 #ifdef CONFIG_QUOTA
1055 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1056 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1057
1058 static int ext4_write_dquot(struct dquot *dquot);
1059 static int ext4_acquire_dquot(struct dquot *dquot);
1060 static int ext4_release_dquot(struct dquot *dquot);
1061 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1062 static int ext4_write_info(struct super_block *sb, int type);
1063 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1064                                 char *path);
1065 static int ext4_quota_on_mount(struct super_block *sb, int type);
1066 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1067                                size_t len, loff_t off);
1068 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1069                                 const char *data, size_t len, loff_t off);
1070
1071 static const struct dquot_operations ext4_quota_operations = {
1072 #ifdef CONFIG_QUOTA
1073         .get_reserved_space = ext4_get_reserved_space,
1074 #endif
1075         .write_dquot    = ext4_write_dquot,
1076         .acquire_dquot  = ext4_acquire_dquot,
1077         .release_dquot  = ext4_release_dquot,
1078         .mark_dirty     = ext4_mark_dquot_dirty,
1079         .write_info     = ext4_write_info,
1080         .alloc_dquot    = dquot_alloc,
1081         .destroy_dquot  = dquot_destroy,
1082 };
1083
1084 static const struct quotactl_ops ext4_qctl_operations = {
1085         .quota_on       = ext4_quota_on,
1086         .quota_off      = dquot_quota_off,
1087         .quota_sync     = dquot_quota_sync,
1088         .get_info       = dquot_get_dqinfo,
1089         .set_info       = dquot_set_dqinfo,
1090         .get_dqblk      = dquot_get_dqblk,
1091         .set_dqblk      = dquot_set_dqblk
1092 };
1093 #endif
1094
1095 static const struct super_operations ext4_sops = {
1096         .alloc_inode    = ext4_alloc_inode,
1097         .destroy_inode  = ext4_destroy_inode,
1098         .write_inode    = ext4_write_inode,
1099         .dirty_inode    = ext4_dirty_inode,
1100         .delete_inode   = ext4_delete_inode,
1101         .put_super      = ext4_put_super,
1102         .sync_fs        = ext4_sync_fs,
1103         .freeze_fs      = ext4_freeze,
1104         .unfreeze_fs    = ext4_unfreeze,
1105         .statfs         = ext4_statfs,
1106         .remount_fs     = ext4_remount,
1107         .clear_inode    = ext4_clear_inode,
1108         .show_options   = ext4_show_options,
1109 #ifdef CONFIG_QUOTA
1110         .quota_read     = ext4_quota_read,
1111         .quota_write    = ext4_quota_write,
1112 #endif
1113         .bdev_try_to_free_page = bdev_try_to_free_page,
1114 };
1115
1116 static const struct super_operations ext4_nojournal_sops = {
1117         .alloc_inode    = ext4_alloc_inode,
1118         .destroy_inode  = ext4_destroy_inode,
1119         .write_inode    = ext4_write_inode,
1120         .dirty_inode    = ext4_dirty_inode,
1121         .delete_inode   = ext4_delete_inode,
1122         .write_super    = ext4_write_super,
1123         .put_super      = ext4_put_super,
1124         .statfs         = ext4_statfs,
1125         .remount_fs     = ext4_remount,
1126         .clear_inode    = ext4_clear_inode,
1127         .show_options   = ext4_show_options,
1128 #ifdef CONFIG_QUOTA
1129         .quota_read     = ext4_quota_read,
1130         .quota_write    = ext4_quota_write,
1131 #endif
1132         .bdev_try_to_free_page = bdev_try_to_free_page,
1133 };
1134
1135 static const struct export_operations ext4_export_ops = {
1136         .fh_to_dentry = ext4_fh_to_dentry,
1137         .fh_to_parent = ext4_fh_to_parent,
1138         .get_parent = ext4_get_parent,
1139 };
1140
1141 enum {
1142         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1143         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1144         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1145         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1146         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1147         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1148         Opt_journal_update, Opt_journal_dev,
1149         Opt_journal_checksum, Opt_journal_async_commit,
1150         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1151         Opt_data_err_abort, Opt_data_err_ignore,
1152         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1153         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1154         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1155         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1156         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1157         Opt_block_validity, Opt_noblock_validity,
1158         Opt_inode_readahead_blks, Opt_journal_ioprio,
1159         Opt_dioread_nolock, Opt_dioread_lock,
1160         Opt_discard, Opt_nodiscard,
1161 };
1162
1163 static const match_table_t tokens = {
1164         {Opt_bsd_df, "bsddf"},
1165         {Opt_minix_df, "minixdf"},
1166         {Opt_grpid, "grpid"},
1167         {Opt_grpid, "bsdgroups"},
1168         {Opt_nogrpid, "nogrpid"},
1169         {Opt_nogrpid, "sysvgroups"},
1170         {Opt_resgid, "resgid=%u"},
1171         {Opt_resuid, "resuid=%u"},
1172         {Opt_sb, "sb=%u"},
1173         {Opt_err_cont, "errors=continue"},
1174         {Opt_err_panic, "errors=panic"},
1175         {Opt_err_ro, "errors=remount-ro"},
1176         {Opt_nouid32, "nouid32"},
1177         {Opt_debug, "debug"},
1178         {Opt_oldalloc, "oldalloc"},
1179         {Opt_orlov, "orlov"},
1180         {Opt_user_xattr, "user_xattr"},
1181         {Opt_nouser_xattr, "nouser_xattr"},
1182         {Opt_acl, "acl"},
1183         {Opt_noacl, "noacl"},
1184         {Opt_noload, "noload"},
1185         {Opt_noload, "norecovery"},
1186         {Opt_nobh, "nobh"},
1187         {Opt_bh, "bh"},
1188         {Opt_commit, "commit=%u"},
1189         {Opt_min_batch_time, "min_batch_time=%u"},
1190         {Opt_max_batch_time, "max_batch_time=%u"},
1191         {Opt_journal_update, "journal=update"},
1192         {Opt_journal_dev, "journal_dev=%u"},
1193         {Opt_journal_checksum, "journal_checksum"},
1194         {Opt_journal_async_commit, "journal_async_commit"},
1195         {Opt_abort, "abort"},
1196         {Opt_data_journal, "data=journal"},
1197         {Opt_data_ordered, "data=ordered"},
1198         {Opt_data_writeback, "data=writeback"},
1199         {Opt_data_err_abort, "data_err=abort"},
1200         {Opt_data_err_ignore, "data_err=ignore"},
1201         {Opt_offusrjquota, "usrjquota="},
1202         {Opt_usrjquota, "usrjquota=%s"},
1203         {Opt_offgrpjquota, "grpjquota="},
1204         {Opt_grpjquota, "grpjquota=%s"},
1205         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1206         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1207         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1208         {Opt_grpquota, "grpquota"},
1209         {Opt_noquota, "noquota"},
1210         {Opt_quota, "quota"},
1211         {Opt_usrquota, "usrquota"},
1212         {Opt_barrier, "barrier=%u"},
1213         {Opt_barrier, "barrier"},
1214         {Opt_nobarrier, "nobarrier"},
1215         {Opt_i_version, "i_version"},
1216         {Opt_stripe, "stripe=%u"},
1217         {Opt_resize, "resize"},
1218         {Opt_delalloc, "delalloc"},
1219         {Opt_nodelalloc, "nodelalloc"},
1220         {Opt_block_validity, "block_validity"},
1221         {Opt_noblock_validity, "noblock_validity"},
1222         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1223         {Opt_journal_ioprio, "journal_ioprio=%u"},
1224         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1225         {Opt_auto_da_alloc, "auto_da_alloc"},
1226         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1227         {Opt_dioread_nolock, "dioread_nolock"},
1228         {Opt_dioread_lock, "dioread_lock"},
1229         {Opt_discard, "discard"},
1230         {Opt_nodiscard, "nodiscard"},
1231         {Opt_err, NULL},
1232 };
1233
1234 static ext4_fsblk_t get_sb_block(void **data)
1235 {
1236         ext4_fsblk_t    sb_block;
1237         char            *options = (char *) *data;
1238
1239         if (!options || strncmp(options, "sb=", 3) != 0)
1240                 return 1;       /* Default location */
1241
1242         options += 3;
1243         /* TODO: use simple_strtoll with >32bit ext4 */
1244         sb_block = simple_strtoul(options, &options, 0);
1245         if (*options && *options != ',') {
1246                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1247                        (char *) *data);
1248                 return 1;
1249         }
1250         if (*options == ',')
1251                 options++;
1252         *data = (void *) options;
1253
1254         return sb_block;
1255 }
1256
1257 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1258 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1259         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1260
1261 #ifdef CONFIG_QUOTA
1262 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1263 {
1264         struct ext4_sb_info *sbi = EXT4_SB(sb);
1265         char *qname;
1266
1267         if (sb_any_quota_loaded(sb) &&
1268                 !sbi->s_qf_names[qtype]) {
1269                 ext4_msg(sb, KERN_ERR,
1270                         "Cannot change journaled "
1271                         "quota options when quota turned on");
1272                 return 0;
1273         }
1274         qname = match_strdup(args);
1275         if (!qname) {
1276                 ext4_msg(sb, KERN_ERR,
1277                         "Not enough memory for storing quotafile name");
1278                 return 0;
1279         }
1280         if (sbi->s_qf_names[qtype] &&
1281                 strcmp(sbi->s_qf_names[qtype], qname)) {
1282                 ext4_msg(sb, KERN_ERR,
1283                         "%s quota file already specified", QTYPE2NAME(qtype));
1284                 kfree(qname);
1285                 return 0;
1286         }
1287         sbi->s_qf_names[qtype] = qname;
1288         if (strchr(sbi->s_qf_names[qtype], '/')) {
1289                 ext4_msg(sb, KERN_ERR,
1290                         "quotafile must be on filesystem root");
1291                 kfree(sbi->s_qf_names[qtype]);
1292                 sbi->s_qf_names[qtype] = NULL;
1293                 return 0;
1294         }
1295         set_opt(sbi->s_mount_opt, QUOTA);
1296         return 1;
1297 }
1298
1299 static int clear_qf_name(struct super_block *sb, int qtype)
1300 {
1301
1302         struct ext4_sb_info *sbi = EXT4_SB(sb);
1303
1304         if (sb_any_quota_loaded(sb) &&
1305                 sbi->s_qf_names[qtype]) {
1306                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1307                         " when quota turned on");
1308                 return 0;
1309         }
1310         /*
1311          * The space will be released later when all options are confirmed
1312          * to be correct
1313          */
1314         sbi->s_qf_names[qtype] = NULL;
1315         return 1;
1316 }
1317 #endif
1318
1319 static int parse_options(char *options, struct super_block *sb,
1320                          unsigned long *journal_devnum,
1321                          unsigned int *journal_ioprio,
1322                          ext4_fsblk_t *n_blocks_count, int is_remount)
1323 {
1324         struct ext4_sb_info *sbi = EXT4_SB(sb);
1325         char *p;
1326         substring_t args[MAX_OPT_ARGS];
1327         int data_opt = 0;
1328         int option;
1329 #ifdef CONFIG_QUOTA
1330         int qfmt;
1331 #endif
1332
1333         if (!options)
1334                 return 1;
1335
1336         while ((p = strsep(&options, ",")) != NULL) {
1337                 int token;
1338                 if (!*p)
1339                         continue;
1340
1341                 /*
1342                  * Initialize args struct so we know whether arg was
1343                  * found; some options take optional arguments.
1344                  */
1345                 args[0].to = args[0].from = 0;
1346                 token = match_token(p, tokens, args);
1347                 switch (token) {
1348                 case Opt_bsd_df:
1349                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1350                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1351                         break;
1352                 case Opt_minix_df:
1353                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1354                         set_opt(sbi->s_mount_opt, MINIX_DF);
1355
1356                         break;
1357                 case Opt_grpid:
1358                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1359                         set_opt(sbi->s_mount_opt, GRPID);
1360
1361                         break;
1362                 case Opt_nogrpid:
1363                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1364                         clear_opt(sbi->s_mount_opt, GRPID);
1365
1366                         break;
1367                 case Opt_resuid:
1368                         if (match_int(&args[0], &option))
1369                                 return 0;
1370                         sbi->s_resuid = option;
1371                         break;
1372                 case Opt_resgid:
1373                         if (match_int(&args[0], &option))
1374                                 return 0;
1375                         sbi->s_resgid = option;
1376                         break;
1377                 case Opt_sb:
1378                         /* handled by get_sb_block() instead of here */
1379                         /* *sb_block = match_int(&args[0]); */
1380                         break;
1381                 case Opt_err_panic:
1382                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1383                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1384                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1385                         break;
1386                 case Opt_err_ro:
1387                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1388                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1389                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1390                         break;
1391                 case Opt_err_cont:
1392                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1393                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1394                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1395                         break;
1396                 case Opt_nouid32:
1397                         set_opt(sbi->s_mount_opt, NO_UID32);
1398                         break;
1399                 case Opt_debug:
1400                         set_opt(sbi->s_mount_opt, DEBUG);
1401                         break;
1402                 case Opt_oldalloc:
1403                         set_opt(sbi->s_mount_opt, OLDALLOC);
1404                         break;
1405                 case Opt_orlov:
1406                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1407                         break;
1408 #ifdef CONFIG_EXT4_FS_XATTR
1409                 case Opt_user_xattr:
1410                         set_opt(sbi->s_mount_opt, XATTR_USER);
1411                         break;
1412                 case Opt_nouser_xattr:
1413                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1414                         break;
1415 #else
1416                 case Opt_user_xattr:
1417                 case Opt_nouser_xattr:
1418                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1419                         break;
1420 #endif
1421 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1422                 case Opt_acl:
1423                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1424                         break;
1425                 case Opt_noacl:
1426                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1427                         break;
1428 #else
1429                 case Opt_acl:
1430                 case Opt_noacl:
1431                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1432                         break;
1433 #endif
1434                 case Opt_journal_update:
1435                         /* @@@ FIXME */
1436                         /* Eventually we will want to be able to create
1437                            a journal file here.  For now, only allow the
1438                            user to specify an existing inode to be the
1439                            journal file. */
1440                         if (is_remount) {
1441                                 ext4_msg(sb, KERN_ERR,
1442                                          "Cannot specify journal on remount");
1443                                 return 0;
1444                         }
1445                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1446                         break;
1447                 case Opt_journal_dev:
1448                         if (is_remount) {
1449                                 ext4_msg(sb, KERN_ERR,
1450                                         "Cannot specify journal on remount");
1451                                 return 0;
1452                         }
1453                         if (match_int(&args[0], &option))
1454                                 return 0;
1455                         *journal_devnum = option;
1456                         break;
1457                 case Opt_journal_checksum:
1458                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1459                         break;
1460                 case Opt_journal_async_commit:
1461                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1462                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1463                         break;
1464                 case Opt_noload:
1465                         set_opt(sbi->s_mount_opt, NOLOAD);
1466                         break;
1467                 case Opt_commit:
1468                         if (match_int(&args[0], &option))
1469                                 return 0;
1470                         if (option < 0)
1471                                 return 0;
1472                         if (option == 0)
1473                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1474                         sbi->s_commit_interval = HZ * option;
1475                         break;
1476                 case Opt_max_batch_time:
1477                         if (match_int(&args[0], &option))
1478                                 return 0;
1479                         if (option < 0)
1480                                 return 0;
1481                         if (option == 0)
1482                                 option = EXT4_DEF_MAX_BATCH_TIME;
1483                         sbi->s_max_batch_time = option;
1484                         break;
1485                 case Opt_min_batch_time:
1486                         if (match_int(&args[0], &option))
1487                                 return 0;
1488                         if (option < 0)
1489                                 return 0;
1490                         sbi->s_min_batch_time = option;
1491                         break;
1492                 case Opt_data_journal:
1493                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1494                         goto datacheck;
1495                 case Opt_data_ordered:
1496                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1497                         goto datacheck;
1498                 case Opt_data_writeback:
1499                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1500                 datacheck:
1501                         if (is_remount) {
1502                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1503                                         ext4_msg(sb, KERN_ERR,
1504                                                 "Cannot change data mode on remount");
1505                                         return 0;
1506                                 }
1507                         } else {
1508                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1509                                 sbi->s_mount_opt |= data_opt;
1510                         }
1511                         break;
1512                 case Opt_data_err_abort:
1513                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1514                         break;
1515                 case Opt_data_err_ignore:
1516                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1517                         break;
1518 #ifdef CONFIG_QUOTA
1519                 case Opt_usrjquota:
1520                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1521                                 return 0;
1522                         break;
1523                 case Opt_grpjquota:
1524                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1525                                 return 0;
1526                         break;
1527                 case Opt_offusrjquota:
1528                         if (!clear_qf_name(sb, USRQUOTA))
1529                                 return 0;
1530                         break;
1531                 case Opt_offgrpjquota:
1532                         if (!clear_qf_name(sb, GRPQUOTA))
1533                                 return 0;
1534                         break;
1535
1536                 case Opt_jqfmt_vfsold:
1537                         qfmt = QFMT_VFS_OLD;
1538                         goto set_qf_format;
1539                 case Opt_jqfmt_vfsv0:
1540                         qfmt = QFMT_VFS_V0;
1541                         goto set_qf_format;
1542                 case Opt_jqfmt_vfsv1:
1543                         qfmt = QFMT_VFS_V1;
1544 set_qf_format:
1545                         if (sb_any_quota_loaded(sb) &&
1546                             sbi->s_jquota_fmt != qfmt) {
1547                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1548                                         "journaled quota options when "
1549                                         "quota turned on");
1550                                 return 0;
1551                         }
1552                         sbi->s_jquota_fmt = qfmt;
1553                         break;
1554                 case Opt_quota:
1555                 case Opt_usrquota:
1556                         set_opt(sbi->s_mount_opt, QUOTA);
1557                         set_opt(sbi->s_mount_opt, USRQUOTA);
1558                         break;
1559                 case Opt_grpquota:
1560                         set_opt(sbi->s_mount_opt, QUOTA);
1561                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1562                         break;
1563                 case Opt_noquota:
1564                         if (sb_any_quota_loaded(sb)) {
1565                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1566                                         "options when quota turned on");
1567                                 return 0;
1568                         }
1569                         clear_opt(sbi->s_mount_opt, QUOTA);
1570                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1571                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1572                         break;
1573 #else
1574                 case Opt_quota:
1575                 case Opt_usrquota:
1576                 case Opt_grpquota:
1577                         ext4_msg(sb, KERN_ERR,
1578                                 "quota options not supported");
1579                         break;
1580                 case Opt_usrjquota:
1581                 case Opt_grpjquota:
1582                 case Opt_offusrjquota:
1583                 case Opt_offgrpjquota:
1584                 case Opt_jqfmt_vfsold:
1585                 case Opt_jqfmt_vfsv0:
1586                 case Opt_jqfmt_vfsv1:
1587                         ext4_msg(sb, KERN_ERR,
1588                                 "journaled quota options not supported");
1589                         break;
1590                 case Opt_noquota:
1591                         break;
1592 #endif
1593                 case Opt_abort:
1594                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1595                         break;
1596                 case Opt_nobarrier:
1597                         clear_opt(sbi->s_mount_opt, BARRIER);
1598                         break;
1599                 case Opt_barrier:
1600                         if (args[0].from) {
1601                                 if (match_int(&args[0], &option))
1602                                         return 0;
1603                         } else
1604                                 option = 1;     /* No argument, default to 1 */
1605                         if (option)
1606                                 set_opt(sbi->s_mount_opt, BARRIER);
1607                         else
1608                                 clear_opt(sbi->s_mount_opt, BARRIER);
1609                         break;
1610                 case Opt_ignore:
1611                         break;
1612                 case Opt_resize:
1613                         if (!is_remount) {
1614                                 ext4_msg(sb, KERN_ERR,
1615                                         "resize option only available "
1616                                         "for remount");
1617                                 return 0;
1618                         }
1619                         if (match_int(&args[0], &option) != 0)
1620                                 return 0;
1621                         *n_blocks_count = option;
1622                         break;
1623                 case Opt_nobh:
1624                         ext4_msg(sb, KERN_WARNING,
1625                                  "Ignoring deprecated nobh option");
1626                         break;
1627                 case Opt_bh:
1628                         ext4_msg(sb, KERN_WARNING,
1629                                  "Ignoring deprecated bh option");
1630                         break;
1631                 case Opt_i_version:
1632                         set_opt(sbi->s_mount_opt, I_VERSION);
1633                         sb->s_flags |= MS_I_VERSION;
1634                         break;
1635                 case Opt_nodelalloc:
1636                         clear_opt(sbi->s_mount_opt, DELALLOC);
1637                         break;
1638                 case Opt_stripe:
1639                         if (match_int(&args[0], &option))
1640                                 return 0;
1641                         if (option < 0)
1642                                 return 0;
1643                         sbi->s_stripe = option;
1644                         break;
1645                 case Opt_delalloc:
1646                         set_opt(sbi->s_mount_opt, DELALLOC);
1647                         break;
1648                 case Opt_block_validity:
1649                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1650                         break;
1651                 case Opt_noblock_validity:
1652                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1653                         break;
1654                 case Opt_inode_readahead_blks:
1655                         if (match_int(&args[0], &option))
1656                                 return 0;
1657                         if (option < 0 || option > (1 << 30))
1658                                 return 0;
1659                         if (!is_power_of_2(option)) {
1660                                 ext4_msg(sb, KERN_ERR,
1661                                          "EXT4-fs: inode_readahead_blks"
1662                                          " must be a power of 2");
1663                                 return 0;
1664                         }
1665                         sbi->s_inode_readahead_blks = option;
1666                         break;
1667                 case Opt_journal_ioprio:
1668                         if (match_int(&args[0], &option))
1669                                 return 0;
1670                         if (option < 0 || option > 7)
1671                                 break;
1672                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1673                                                             option);
1674                         break;
1675                 case Opt_noauto_da_alloc:
1676                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1677                         break;
1678                 case Opt_auto_da_alloc:
1679                         if (args[0].from) {
1680                                 if (match_int(&args[0], &option))
1681                                         return 0;
1682                         } else
1683                                 option = 1;     /* No argument, default to 1 */
1684                         if (option)
1685                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1686                         else
1687                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1688                         break;
1689                 case Opt_discard:
1690                         set_opt(sbi->s_mount_opt, DISCARD);
1691                         break;
1692                 case Opt_nodiscard:
1693                         clear_opt(sbi->s_mount_opt, DISCARD);
1694                         break;
1695                 case Opt_dioread_nolock:
1696                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1697                         break;
1698                 case Opt_dioread_lock:
1699                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1700                         break;
1701                 default:
1702                         ext4_msg(sb, KERN_ERR,
1703                                "Unrecognized mount option \"%s\" "
1704                                "or missing value", p);
1705                         return 0;
1706                 }
1707         }
1708 #ifdef CONFIG_QUOTA
1709         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1710                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1711                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1712
1713                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1714                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1715
1716                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1717                         ext4_msg(sb, KERN_ERR, "old and new quota "
1718                                         "format mixing");
1719                         return 0;
1720                 }
1721
1722                 if (!sbi->s_jquota_fmt) {
1723                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1724                                         "not specified");
1725                         return 0;
1726                 }
1727         } else {
1728                 if (sbi->s_jquota_fmt) {
1729                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1730                                         "specified with no journaling "
1731                                         "enabled");
1732                         return 0;
1733                 }
1734         }
1735 #endif
1736         return 1;
1737 }
1738
1739 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1740                             int read_only)
1741 {
1742         struct ext4_sb_info *sbi = EXT4_SB(sb);
1743         int res = 0;
1744
1745         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1746                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1747                          "forcing read-only mode");
1748                 res = MS_RDONLY;
1749         }
1750         if (read_only)
1751                 return res;
1752         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1753                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1754                          "running e2fsck is recommended");
1755         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1756                 ext4_msg(sb, KERN_WARNING,
1757                          "warning: mounting fs with errors, "
1758                          "running e2fsck is recommended");
1759         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1760                  le16_to_cpu(es->s_mnt_count) >=
1761                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1762                 ext4_msg(sb, KERN_WARNING,
1763                          "warning: maximal mount count reached, "
1764                          "running e2fsck is recommended");
1765         else if (le32_to_cpu(es->s_checkinterval) &&
1766                 (le32_to_cpu(es->s_lastcheck) +
1767                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1768                 ext4_msg(sb, KERN_WARNING,
1769                          "warning: checktime reached, "
1770                          "running e2fsck is recommended");
1771         if (!sbi->s_journal)
1772                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1773         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1774                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1775         le16_add_cpu(&es->s_mnt_count, 1);
1776         es->s_mtime = cpu_to_le32(get_seconds());
1777         ext4_update_dynamic_rev(sb);
1778         if (sbi->s_journal)
1779                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1780
1781         ext4_commit_super(sb, 1);
1782         if (test_opt(sb, DEBUG))
1783                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1784                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1785                         sb->s_blocksize,
1786                         sbi->s_groups_count,
1787                         EXT4_BLOCKS_PER_GROUP(sb),
1788                         EXT4_INODES_PER_GROUP(sb),
1789                         sbi->s_mount_opt);
1790
1791         return res;
1792 }
1793
1794 static int ext4_fill_flex_info(struct super_block *sb)
1795 {
1796         struct ext4_sb_info *sbi = EXT4_SB(sb);
1797         struct ext4_group_desc *gdp = NULL;
1798         ext4_group_t flex_group_count;
1799         ext4_group_t flex_group;
1800         int groups_per_flex = 0;
1801         size_t size;
1802         int i;
1803
1804         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1805         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1806
1807         if (groups_per_flex < 2) {
1808                 sbi->s_log_groups_per_flex = 0;
1809                 return 1;
1810         }
1811
1812         /* We allocate both existing and potentially added groups */
1813         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1814                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1815                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1816         size = flex_group_count * sizeof(struct flex_groups);
1817         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1818         if (sbi->s_flex_groups == NULL) {
1819                 sbi->s_flex_groups = vmalloc(size);
1820                 if (sbi->s_flex_groups)
1821                         memset(sbi->s_flex_groups, 0, size);
1822         }
1823         if (sbi->s_flex_groups == NULL) {
1824                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1825                                 "%u flex groups", flex_group_count);
1826                 goto failed;
1827         }
1828
1829         for (i = 0; i < sbi->s_groups_count; i++) {
1830                 gdp = ext4_get_group_desc(sb, i, NULL);
1831
1832                 flex_group = ext4_flex_group(sbi, i);
1833                 atomic_add(ext4_free_inodes_count(sb, gdp),
1834                            &sbi->s_flex_groups[flex_group].free_inodes);
1835                 atomic_add(ext4_free_blks_count(sb, gdp),
1836                            &sbi->s_flex_groups[flex_group].free_blocks);
1837                 atomic_add(ext4_used_dirs_count(sb, gdp),
1838                            &sbi->s_flex_groups[flex_group].used_dirs);
1839         }
1840
1841         return 1;
1842 failed:
1843         return 0;
1844 }
1845
1846 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1847                             struct ext4_group_desc *gdp)
1848 {
1849         __u16 crc = 0;
1850
1851         if (sbi->s_es->s_feature_ro_compat &
1852             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1853                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1854                 __le32 le_group = cpu_to_le32(block_group);
1855
1856                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1857                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1858                 crc = crc16(crc, (__u8 *)gdp, offset);
1859                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1860                 /* for checksum of struct ext4_group_desc do the rest...*/
1861                 if ((sbi->s_es->s_feature_incompat &
1862                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1863                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1864                         crc = crc16(crc, (__u8 *)gdp + offset,
1865                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1866                                         offset);
1867         }
1868
1869         return cpu_to_le16(crc);
1870 }
1871
1872 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1873                                 struct ext4_group_desc *gdp)
1874 {
1875         if ((sbi->s_es->s_feature_ro_compat &
1876              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1877             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1878                 return 0;
1879
1880         return 1;
1881 }
1882
1883 /* Called at mount-time, super-block is locked */
1884 static int ext4_check_descriptors(struct super_block *sb)
1885 {
1886         struct ext4_sb_info *sbi = EXT4_SB(sb);
1887         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1888         ext4_fsblk_t last_block;
1889         ext4_fsblk_t block_bitmap;
1890         ext4_fsblk_t inode_bitmap;
1891         ext4_fsblk_t inode_table;
1892         int flexbg_flag = 0;
1893         ext4_group_t i;
1894
1895         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1896                 flexbg_flag = 1;
1897
1898         ext4_debug("Checking group descriptors");
1899
1900         for (i = 0; i < sbi->s_groups_count; i++) {
1901                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1902
1903                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1904                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1905                 else
1906                         last_block = first_block +
1907                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1908
1909                 block_bitmap = ext4_block_bitmap(sb, gdp);
1910                 if (block_bitmap < first_block || block_bitmap > last_block) {
1911                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1912                                "Block bitmap for group %u not in group "
1913                                "(block %llu)!", i, block_bitmap);
1914                         return 0;
1915                 }
1916                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1917                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1918                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1919                                "Inode bitmap for group %u not in group "
1920                                "(block %llu)!", i, inode_bitmap);
1921                         return 0;
1922                 }
1923                 inode_table = ext4_inode_table(sb, gdp);
1924                 if (inode_table < first_block ||
1925                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1926                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1927                                "Inode table for group %u not in group "
1928                                "(block %llu)!", i, inode_table);
1929                         return 0;
1930                 }
1931                 ext4_lock_group(sb, i);
1932                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1933                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1934                                  "Checksum for group %u failed (%u!=%u)",
1935                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1936                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1937                         if (!(sb->s_flags & MS_RDONLY)) {
1938                                 ext4_unlock_group(sb, i);
1939                                 return 0;
1940                         }
1941                 }
1942                 ext4_unlock_group(sb, i);
1943                 if (!flexbg_flag)
1944                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1945         }
1946
1947         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1948         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1949         return 1;
1950 }
1951
1952 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1953  * the superblock) which were deleted from all directories, but held open by
1954  * a process at the time of a crash.  We walk the list and try to delete these
1955  * inodes at recovery time (only with a read-write filesystem).
1956  *
1957  * In order to keep the orphan inode chain consistent during traversal (in
1958  * case of crash during recovery), we link each inode into the superblock
1959  * orphan list_head and handle it the same way as an inode deletion during
1960  * normal operation (which journals the operations for us).
1961  *
1962  * We only do an iget() and an iput() on each inode, which is very safe if we
1963  * accidentally point at an in-use or already deleted inode.  The worst that
1964  * can happen in this case is that we get a "bit already cleared" message from
1965  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1966  * e2fsck was run on this filesystem, and it must have already done the orphan
1967  * inode cleanup for us, so we can safely abort without any further action.
1968  */
1969 static void ext4_orphan_cleanup(struct super_block *sb,
1970                                 struct ext4_super_block *es)
1971 {
1972         unsigned int s_flags = sb->s_flags;
1973         int nr_orphans = 0, nr_truncates = 0;
1974 #ifdef CONFIG_QUOTA
1975         int i;
1976 #endif
1977         if (!es->s_last_orphan) {
1978                 jbd_debug(4, "no orphan inodes to clean up\n");
1979                 return;
1980         }
1981
1982         if (bdev_read_only(sb->s_bdev)) {
1983                 ext4_msg(sb, KERN_ERR, "write access "
1984                         "unavailable, skipping orphan cleanup");
1985                 return;
1986         }
1987
1988         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1989                 if (es->s_last_orphan)
1990                         jbd_debug(1, "Errors on filesystem, "
1991                                   "clearing orphan list.\n");
1992                 es->s_last_orphan = 0;
1993                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1994                 return;
1995         }
1996
1997         if (s_flags & MS_RDONLY) {
1998                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1999                 sb->s_flags &= ~MS_RDONLY;
2000         }
2001 #ifdef CONFIG_QUOTA
2002         /* Needed for iput() to work correctly and not trash data */
2003         sb->s_flags |= MS_ACTIVE;
2004         /* Turn on quotas so that they are updated correctly */
2005         for (i = 0; i < MAXQUOTAS; i++) {
2006                 if (EXT4_SB(sb)->s_qf_names[i]) {
2007                         int ret = ext4_quota_on_mount(sb, i);
2008                         if (ret < 0)
2009                                 ext4_msg(sb, KERN_ERR,
2010                                         "Cannot turn on journaled "
2011                                         "quota: error %d", ret);
2012                 }
2013         }
2014 #endif
2015
2016         while (es->s_last_orphan) {
2017                 struct inode *inode;
2018
2019                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2020                 if (IS_ERR(inode)) {
2021                         es->s_last_orphan = 0;
2022                         break;
2023                 }
2024
2025                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2026                 dquot_initialize(inode);
2027                 if (inode->i_nlink) {
2028                         ext4_msg(sb, KERN_DEBUG,
2029                                 "%s: truncating inode %lu to %lld bytes",
2030                                 __func__, inode->i_ino, inode->i_size);
2031                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2032                                   inode->i_ino, inode->i_size);
2033                         ext4_truncate(inode);
2034                         nr_truncates++;
2035                 } else {
2036                         ext4_msg(sb, KERN_DEBUG,
2037                                 "%s: deleting unreferenced inode %lu",
2038                                 __func__, inode->i_ino);
2039                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2040                                   inode->i_ino);
2041                         nr_orphans++;
2042                 }
2043                 iput(inode);  /* The delete magic happens here! */
2044         }
2045
2046 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2047
2048         if (nr_orphans)
2049                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2050                        PLURAL(nr_orphans));
2051         if (nr_truncates)
2052                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2053                        PLURAL(nr_truncates));
2054 #ifdef CONFIG_QUOTA
2055         /* Turn quotas off */
2056         for (i = 0; i < MAXQUOTAS; i++) {
2057                 if (sb_dqopt(sb)->files[i])
2058                         dquot_quota_off(sb, i);
2059         }
2060 #endif
2061         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2062 }
2063
2064 /*
2065  * Maximal extent format file size.
2066  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2067  * extent format containers, within a sector_t, and within i_blocks
2068  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2069  * so that won't be a limiting factor.
2070  *
2071  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2072  */
2073 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2074 {
2075         loff_t res;
2076         loff_t upper_limit = MAX_LFS_FILESIZE;
2077
2078         /* small i_blocks in vfs inode? */
2079         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2080                 /*
2081                  * CONFIG_LBDAF is not enabled implies the inode
2082                  * i_block represent total blocks in 512 bytes
2083                  * 32 == size of vfs inode i_blocks * 8
2084                  */
2085                 upper_limit = (1LL << 32) - 1;
2086
2087                 /* total blocks in file system block size */
2088                 upper_limit >>= (blkbits - 9);
2089                 upper_limit <<= blkbits;
2090         }
2091
2092         /* 32-bit extent-start container, ee_block */
2093         res = 1LL << 32;
2094         res <<= blkbits;
2095         res -= 1;
2096
2097         /* Sanity check against vm- & vfs- imposed limits */
2098         if (res > upper_limit)
2099                 res = upper_limit;
2100
2101         return res;
2102 }
2103
2104 /*
2105  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2106  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2107  * We need to be 1 filesystem block less than the 2^48 sector limit.
2108  */
2109 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2110 {
2111         loff_t res = EXT4_NDIR_BLOCKS;
2112         int meta_blocks;
2113         loff_t upper_limit;
2114         /* This is calculated to be the largest file size for a dense, block
2115          * mapped file such that the file's total number of 512-byte sectors,
2116          * including data and all indirect blocks, does not exceed (2^48 - 1).
2117          *
2118          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2119          * number of 512-byte sectors of the file.
2120          */
2121
2122         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2123                 /*
2124                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2125                  * the inode i_block field represents total file blocks in
2126                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2127                  */
2128                 upper_limit = (1LL << 32) - 1;
2129
2130                 /* total blocks in file system block size */
2131                 upper_limit >>= (bits - 9);
2132
2133         } else {
2134                 /*
2135                  * We use 48 bit ext4_inode i_blocks
2136                  * With EXT4_HUGE_FILE_FL set the i_blocks
2137                  * represent total number of blocks in
2138                  * file system block size
2139                  */
2140                 upper_limit = (1LL << 48) - 1;
2141
2142         }
2143
2144         /* indirect blocks */
2145         meta_blocks = 1;
2146         /* double indirect blocks */
2147         meta_blocks += 1 + (1LL << (bits-2));
2148         /* tripple indirect blocks */
2149         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2150
2151         upper_limit -= meta_blocks;
2152         upper_limit <<= bits;
2153
2154         res += 1LL << (bits-2);
2155         res += 1LL << (2*(bits-2));
2156         res += 1LL << (3*(bits-2));
2157         res <<= bits;
2158         if (res > upper_limit)
2159                 res = upper_limit;
2160
2161         if (res > MAX_LFS_FILESIZE)
2162                 res = MAX_LFS_FILESIZE;
2163
2164         return res;
2165 }
2166
2167 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2168                                    ext4_fsblk_t logical_sb_block, int nr)
2169 {
2170         struct ext4_sb_info *sbi = EXT4_SB(sb);
2171         ext4_group_t bg, first_meta_bg;
2172         int has_super = 0;
2173
2174         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2175
2176         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2177             nr < first_meta_bg)
2178                 return logical_sb_block + nr + 1;
2179         bg = sbi->s_desc_per_block * nr;
2180         if (ext4_bg_has_super(sb, bg))
2181                 has_super = 1;
2182
2183         return (has_super + ext4_group_first_block_no(sb, bg));
2184 }
2185
2186 /**
2187  * ext4_get_stripe_size: Get the stripe size.
2188  * @sbi: In memory super block info
2189  *
2190  * If we have specified it via mount option, then
2191  * use the mount option value. If the value specified at mount time is
2192  * greater than the blocks per group use the super block value.
2193  * If the super block value is greater than blocks per group return 0.
2194  * Allocator needs it be less than blocks per group.
2195  *
2196  */
2197 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2198 {
2199         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2200         unsigned long stripe_width =
2201                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2202
2203         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2204                 return sbi->s_stripe;
2205
2206         if (stripe_width <= sbi->s_blocks_per_group)
2207                 return stripe_width;
2208
2209         if (stride <= sbi->s_blocks_per_group)
2210                 return stride;
2211
2212         return 0;
2213 }
2214
2215 /* sysfs supprt */
2216
2217 struct ext4_attr {
2218         struct attribute attr;
2219         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2220         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2221                          const char *, size_t);
2222         int offset;
2223 };
2224
2225 static int parse_strtoul(const char *buf,
2226                 unsigned long max, unsigned long *value)
2227 {
2228         char *endp;
2229
2230         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2231         endp = skip_spaces(endp);
2232         if (*endp || *value > max)
2233                 return -EINVAL;
2234
2235         return 0;
2236 }
2237
2238 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2239                                               struct ext4_sb_info *sbi,
2240                                               char *buf)
2241 {
2242         return snprintf(buf, PAGE_SIZE, "%llu\n",
2243                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2244 }
2245
2246 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2247                                          struct ext4_sb_info *sbi, char *buf)
2248 {
2249         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2250
2251         return snprintf(buf, PAGE_SIZE, "%lu\n",
2252                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2253                          sbi->s_sectors_written_start) >> 1);
2254 }
2255
2256 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2257                                           struct ext4_sb_info *sbi, char *buf)
2258 {
2259         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2260
2261         return snprintf(buf, PAGE_SIZE, "%llu\n",
2262                         (unsigned long long)(sbi->s_kbytes_written +
2263                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2264                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2265 }
2266
2267 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2268                                           struct ext4_sb_info *sbi,
2269                                           const char *buf, size_t count)
2270 {
2271         unsigned long t;
2272
2273         if (parse_strtoul(buf, 0x40000000, &t))
2274                 return -EINVAL;
2275
2276         if (!is_power_of_2(t))
2277                 return -EINVAL;
2278
2279         sbi->s_inode_readahead_blks = t;
2280         return count;
2281 }
2282
2283 static ssize_t sbi_ui_show(struct ext4_attr *a,
2284                            struct ext4_sb_info *sbi, char *buf)
2285 {
2286         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2287
2288         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2289 }
2290
2291 static ssize_t sbi_ui_store(struct ext4_attr *a,
2292                             struct ext4_sb_info *sbi,
2293                             const char *buf, size_t count)
2294 {
2295         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2296         unsigned long t;
2297
2298         if (parse_strtoul(buf, 0xffffffff, &t))
2299                 return -EINVAL;
2300         *ui = t;
2301         return count;
2302 }
2303
2304 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2305 static struct ext4_attr ext4_attr_##_name = {                   \
2306         .attr = {.name = __stringify(_name), .mode = _mode },   \
2307         .show   = _show,                                        \
2308         .store  = _store,                                       \
2309         .offset = offsetof(struct ext4_sb_info, _elname),       \
2310 }
2311 #define EXT4_ATTR(name, mode, show, store) \
2312 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2313
2314 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2315 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2316 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2317         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2318 #define ATTR_LIST(name) &ext4_attr_##name.attr
2319
2320 EXT4_RO_ATTR(delayed_allocation_blocks);
2321 EXT4_RO_ATTR(session_write_kbytes);
2322 EXT4_RO_ATTR(lifetime_write_kbytes);
2323 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2324                  inode_readahead_blks_store, s_inode_readahead_blks);
2325 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2326 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2327 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2328 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2329 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2330 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2331 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2332 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2333
2334 static struct attribute *ext4_attrs[] = {
2335         ATTR_LIST(delayed_allocation_blocks),
2336         ATTR_LIST(session_write_kbytes),
2337         ATTR_LIST(lifetime_write_kbytes),
2338         ATTR_LIST(inode_readahead_blks),
2339         ATTR_LIST(inode_goal),
2340         ATTR_LIST(mb_stats),
2341         ATTR_LIST(mb_max_to_scan),
2342         ATTR_LIST(mb_min_to_scan),
2343         ATTR_LIST(mb_order2_req),
2344         ATTR_LIST(mb_stream_req),
2345         ATTR_LIST(mb_group_prealloc),
2346         ATTR_LIST(max_writeback_mb_bump),
2347         NULL,
2348 };
2349
2350 static ssize_t ext4_attr_show(struct kobject *kobj,
2351                               struct attribute *attr, char *buf)
2352 {
2353         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2354                                                 s_kobj);
2355         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2356
2357         return a->show ? a->show(a, sbi, buf) : 0;
2358 }
2359
2360 static ssize_t ext4_attr_store(struct kobject *kobj,
2361                                struct attribute *attr,
2362                                const char *buf, size_t len)
2363 {
2364         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2365                                                 s_kobj);
2366         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2367
2368         return a->store ? a->store(a, sbi, buf, len) : 0;
2369 }
2370
2371 static void ext4_sb_release(struct kobject *kobj)
2372 {
2373         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2374                                                 s_kobj);
2375         complete(&sbi->s_kobj_unregister);
2376 }
2377
2378
2379 static const struct sysfs_ops ext4_attr_ops = {
2380         .show   = ext4_attr_show,
2381         .store  = ext4_attr_store,
2382 };
2383
2384 static struct kobj_type ext4_ktype = {
2385         .default_attrs  = ext4_attrs,
2386         .sysfs_ops      = &ext4_attr_ops,
2387         .release        = ext4_sb_release,
2388 };
2389
2390 /*
2391  * Check whether this filesystem can be mounted based on
2392  * the features present and the RDONLY/RDWR mount requested.
2393  * Returns 1 if this filesystem can be mounted as requested,
2394  * 0 if it cannot be.
2395  */
2396 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2397 {
2398         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2399                 ext4_msg(sb, KERN_ERR,
2400                         "Couldn't mount because of "
2401                         "unsupported optional features (%x)",
2402                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2403                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2404                 return 0;
2405         }
2406
2407         if (readonly)
2408                 return 1;
2409
2410         /* Check that feature set is OK for a read-write mount */
2411         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2412                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2413                          "unsupported optional features (%x)",
2414                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2415                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2416                 return 0;
2417         }
2418         /*
2419          * Large file size enabled file system can only be mounted
2420          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2421          */
2422         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2423                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2424                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2425                                  "cannot be mounted RDWR without "
2426                                  "CONFIG_LBDAF");
2427                         return 0;
2428                 }
2429         }
2430         return 1;
2431 }
2432
2433 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2434                                 __releases(kernel_lock)
2435                                 __acquires(kernel_lock)
2436 {
2437         char *orig_data = kstrdup(data, GFP_KERNEL);
2438         struct buffer_head *bh;
2439         struct ext4_super_block *es = NULL;
2440         struct ext4_sb_info *sbi;
2441         ext4_fsblk_t block;
2442         ext4_fsblk_t sb_block = get_sb_block(&data);
2443         ext4_fsblk_t logical_sb_block;
2444         unsigned long offset = 0;
2445         unsigned long journal_devnum = 0;
2446         unsigned long def_mount_opts;
2447         struct inode *root;
2448         char *cp;
2449         const char *descr;
2450         int ret = -EINVAL;
2451         int blocksize;
2452         unsigned int db_count;
2453         unsigned int i;
2454         int needs_recovery, has_huge_files;
2455         __u64 blocks_count;
2456         int err;
2457         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2458
2459         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2460         if (!sbi)
2461                 return -ENOMEM;
2462
2463         sbi->s_blockgroup_lock =
2464                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2465         if (!sbi->s_blockgroup_lock) {
2466                 kfree(sbi);
2467                 return -ENOMEM;
2468         }
2469         sb->s_fs_info = sbi;
2470         sbi->s_mount_opt = 0;
2471         sbi->s_resuid = EXT4_DEF_RESUID;
2472         sbi->s_resgid = EXT4_DEF_RESGID;
2473         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2474         sbi->s_sb_block = sb_block;
2475         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2476                                                       sectors[1]);
2477
2478         unlock_kernel();
2479
2480         /* Cleanup superblock name */
2481         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2482                 *cp = '!';
2483
2484         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2485         if (!blocksize) {
2486                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2487                 goto out_fail;
2488         }
2489
2490         /*
2491          * The ext4 superblock will not be buffer aligned for other than 1kB
2492          * block sizes.  We need to calculate the offset from buffer start.
2493          */
2494         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2495                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2496                 offset = do_div(logical_sb_block, blocksize);
2497         } else {
2498                 logical_sb_block = sb_block;
2499         }
2500
2501         if (!(bh = sb_bread(sb, logical_sb_block))) {
2502                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2503                 goto out_fail;
2504         }
2505         /*
2506          * Note: s_es must be initialized as soon as possible because
2507          *       some ext4 macro-instructions depend on its value
2508          */
2509         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2510         sbi->s_es = es;
2511         sb->s_magic = le16_to_cpu(es->s_magic);
2512         if (sb->s_magic != EXT4_SUPER_MAGIC)
2513                 goto cantfind_ext4;
2514         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2515
2516         /* Set defaults before we parse the mount options */
2517         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2518         if (def_mount_opts & EXT4_DEFM_DEBUG)
2519                 set_opt(sbi->s_mount_opt, DEBUG);
2520         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
2521                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
2522                         "2.6.38");
2523                 set_opt(sbi->s_mount_opt, GRPID);
2524         }
2525         if (def_mount_opts & EXT4_DEFM_UID16)
2526                 set_opt(sbi->s_mount_opt, NO_UID32);
2527 #ifdef CONFIG_EXT4_FS_XATTR
2528         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2529                 set_opt(sbi->s_mount_opt, XATTR_USER);
2530 #endif
2531 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2532         if (def_mount_opts & EXT4_DEFM_ACL)
2533                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2534 #endif
2535         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2536                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2537         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2538                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
2539         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2540                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2541
2542         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2543                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2544         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2545                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2546         else
2547                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2548
2549         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2550         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2551         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2552         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2553         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2554
2555         set_opt(sbi->s_mount_opt, BARRIER);
2556
2557         /*
2558          * enable delayed allocation by default
2559          * Use -o nodelalloc to turn it off
2560          */
2561         if (!IS_EXT3_SB(sb))
2562                 set_opt(sbi->s_mount_opt, DELALLOC);
2563
2564         if (!parse_options((char *) data, sb, &journal_devnum,
2565                            &journal_ioprio, NULL, 0))
2566                 goto failed_mount;
2567
2568         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2569                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2570
2571         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2572             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2573              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2574              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2575                 ext4_msg(sb, KERN_WARNING,
2576                        "feature flags set on rev 0 fs, "
2577                        "running e2fsck is recommended");
2578
2579         /*
2580          * Check feature flags regardless of the revision level, since we
2581          * previously didn't change the revision level when setting the flags,
2582          * so there is a chance incompat flags are set on a rev 0 filesystem.
2583          */
2584         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2585                 goto failed_mount;
2586
2587         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2588
2589         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2590             blocksize > EXT4_MAX_BLOCK_SIZE) {
2591                 ext4_msg(sb, KERN_ERR,
2592                        "Unsupported filesystem blocksize %d", blocksize);
2593                 goto failed_mount;
2594         }
2595
2596         if (sb->s_blocksize != blocksize) {
2597                 /* Validate the filesystem blocksize */
2598                 if (!sb_set_blocksize(sb, blocksize)) {
2599                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2600                                         blocksize);
2601                         goto failed_mount;
2602                 }
2603
2604                 brelse(bh);
2605                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2606                 offset = do_div(logical_sb_block, blocksize);
2607                 bh = sb_bread(sb, logical_sb_block);
2608                 if (!bh) {
2609                         ext4_msg(sb, KERN_ERR,
2610                                "Can't read superblock on 2nd try");
2611                         goto failed_mount;
2612                 }
2613                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2614                 sbi->s_es = es;
2615                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2616                         ext4_msg(sb, KERN_ERR,
2617                                "Magic mismatch, very weird!");
2618                         goto failed_mount;
2619                 }
2620         }
2621
2622         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2623                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2624         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2625                                                       has_huge_files);
2626         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2627
2628         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2629                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2630                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2631         } else {
2632                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2633                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2634                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2635                     (!is_power_of_2(sbi->s_inode_size)) ||
2636                     (sbi->s_inode_size > blocksize)) {
2637                         ext4_msg(sb, KERN_ERR,
2638                                "unsupported inode size: %d",
2639                                sbi->s_inode_size);
2640                         goto failed_mount;
2641                 }
2642                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2643                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2644         }
2645
2646         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2647         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2648                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2649                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2650                     !is_power_of_2(sbi->s_desc_size)) {
2651                         ext4_msg(sb, KERN_ERR,
2652                                "unsupported descriptor size %lu",
2653                                sbi->s_desc_size);
2654                         goto failed_mount;
2655                 }
2656         } else
2657                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2658
2659         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2660         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2661         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2662                 goto cantfind_ext4;
2663
2664         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2665         if (sbi->s_inodes_per_block == 0)
2666                 goto cantfind_ext4;
2667         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2668                                         sbi->s_inodes_per_block;
2669         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2670         sbi->s_sbh = bh;
2671         sbi->s_mount_state = le16_to_cpu(es->s_state);
2672         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2673         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2674
2675         for (i = 0; i < 4; i++)
2676                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2677         sbi->s_def_hash_version = es->s_def_hash_version;
2678         i = le32_to_cpu(es->s_flags);
2679         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2680                 sbi->s_hash_unsigned = 3;
2681         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2682 #ifdef __CHAR_UNSIGNED__
2683                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2684                 sbi->s_hash_unsigned = 3;
2685 #else
2686                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2687 #endif
2688                 sb->s_dirt = 1;
2689         }
2690
2691         if (sbi->s_blocks_per_group > blocksize * 8) {
2692                 ext4_msg(sb, KERN_ERR,
2693                        "#blocks per group too big: %lu",
2694                        sbi->s_blocks_per_group);
2695                 goto failed_mount;
2696         }
2697         if (sbi->s_inodes_per_group > blocksize * 8) {
2698                 ext4_msg(sb, KERN_ERR,
2699                        "#inodes per group too big: %lu",
2700                        sbi->s_inodes_per_group);
2701                 goto failed_mount;
2702         }
2703
2704         /*
2705          * Test whether we have more sectors than will fit in sector_t,
2706          * and whether the max offset is addressable by the page cache.
2707          */
2708         if ((ext4_blocks_count(es) >
2709              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2710             (ext4_blocks_count(es) >
2711              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2712                 ext4_msg(sb, KERN_ERR, "filesystem"
2713                          " too large to mount safely on this system");
2714                 if (sizeof(sector_t) < 8)
2715                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2716                 ret = -EFBIG;
2717                 goto failed_mount;
2718         }
2719
2720         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2721                 goto cantfind_ext4;
2722
2723         /* check blocks count against device size */
2724         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2725         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2726                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2727                        "exceeds size of device (%llu blocks)",
2728                        ext4_blocks_count(es), blocks_count);
2729                 goto failed_mount;
2730         }
2731
2732         /*
2733          * It makes no sense for the first data block to be beyond the end
2734          * of the filesystem.
2735          */
2736         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2737                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2738                          "block %u is beyond end of filesystem (%llu)",
2739                          le32_to_cpu(es->s_first_data_block),
2740                          ext4_blocks_count(es));
2741                 goto failed_mount;
2742         }
2743         blocks_count = (ext4_blocks_count(es) -
2744                         le32_to_cpu(es->s_first_data_block) +
2745                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2746         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2747         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2748                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2749                        "(block count %llu, first data block %u, "
2750                        "blocks per group %lu)", sbi->s_groups_count,
2751                        ext4_blocks_count(es),
2752                        le32_to_cpu(es->s_first_data_block),
2753                        EXT4_BLOCKS_PER_GROUP(sb));
2754                 goto failed_mount;
2755         }
2756         sbi->s_groups_count = blocks_count;
2757         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2758                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2759         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2760                    EXT4_DESC_PER_BLOCK(sb);
2761         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2762                                     GFP_KERNEL);
2763         if (sbi->s_group_desc == NULL) {
2764                 ext4_msg(sb, KERN_ERR, "not enough memory");
2765                 goto failed_mount;
2766         }
2767
2768 #ifdef CONFIG_PROC_FS
2769         if (ext4_proc_root)
2770                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2771 #endif
2772
2773         bgl_lock_init(sbi->s_blockgroup_lock);
2774
2775         for (i = 0; i < db_count; i++) {
2776                 block = descriptor_loc(sb, logical_sb_block, i);
2777                 sbi->s_group_desc[i] = sb_bread(sb, block);
2778                 if (!sbi->s_group_desc[i]) {
2779                         ext4_msg(sb, KERN_ERR,
2780                                "can't read group descriptor %d", i);
2781                         db_count = i;
2782                         goto failed_mount2;
2783                 }
2784         }
2785         if (!ext4_check_descriptors(sb)) {
2786                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2787                 goto failed_mount2;
2788         }
2789         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2790                 if (!ext4_fill_flex_info(sb)) {
2791                         ext4_msg(sb, KERN_ERR,
2792                                "unable to initialize "
2793                                "flex_bg meta info!");
2794                         goto failed_mount2;
2795                 }
2796
2797         sbi->s_gdb_count = db_count;
2798         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2799         spin_lock_init(&sbi->s_next_gen_lock);
2800
2801         sbi->s_stripe = ext4_get_stripe_size(sbi);
2802         sbi->s_max_writeback_mb_bump = 128;
2803
2804         /*
2805          * set up enough so that it can read an inode
2806          */
2807         if (!test_opt(sb, NOLOAD) &&
2808             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2809                 sb->s_op = &ext4_sops;
2810         else
2811                 sb->s_op = &ext4_nojournal_sops;
2812         sb->s_export_op = &ext4_export_ops;
2813         sb->s_xattr = ext4_xattr_handlers;
2814 #ifdef CONFIG_QUOTA
2815         sb->s_qcop = &ext4_qctl_operations;
2816         sb->dq_op = &ext4_quota_operations;
2817 #endif
2818         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2819         mutex_init(&sbi->s_orphan_lock);
2820         mutex_init(&sbi->s_resize_lock);
2821
2822         sb->s_root = NULL;
2823
2824         needs_recovery = (es->s_last_orphan != 0 ||
2825                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2826                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2827
2828         /*
2829          * The first inode we look at is the journal inode.  Don't try
2830          * root first: it may be modified in the journal!
2831          */
2832         if (!test_opt(sb, NOLOAD) &&
2833             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2834                 if (ext4_load_journal(sb, es, journal_devnum))
2835                         goto failed_mount3;
2836         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2837               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2838                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2839                        "suppressed and not mounted read-only");
2840                 goto failed_mount_wq;
2841         } else {
2842                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2843                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2844                 sbi->s_journal = NULL;
2845                 needs_recovery = 0;
2846                 goto no_journal;
2847         }
2848
2849         if (ext4_blocks_count(es) > 0xffffffffULL &&
2850             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2851                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2852                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2853                 goto failed_mount_wq;
2854         }
2855
2856         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2857                 jbd2_journal_set_features(sbi->s_journal,
2858                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2859                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2860         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2861                 jbd2_journal_set_features(sbi->s_journal,
2862                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2863                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2864                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2865         } else {
2866                 jbd2_journal_clear_features(sbi->s_journal,
2867                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2868                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2869         }
2870
2871         /* We have now updated the journal if required, so we can
2872          * validate the data journaling mode. */
2873         switch (test_opt(sb, DATA_FLAGS)) {
2874         case 0:
2875                 /* No mode set, assume a default based on the journal
2876                  * capabilities: ORDERED_DATA if the journal can
2877                  * cope, else JOURNAL_DATA
2878                  */
2879                 if (jbd2_journal_check_available_features
2880                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2881                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2882                 else
2883                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2884                 break;
2885
2886         case EXT4_MOUNT_ORDERED_DATA:
2887         case EXT4_MOUNT_WRITEBACK_DATA:
2888                 if (!jbd2_journal_check_available_features
2889                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2890                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2891                                "requested data journaling mode");
2892                         goto failed_mount_wq;
2893                 }
2894         default:
2895                 break;
2896         }
2897         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2898
2899 no_journal:
2900         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2901                                   ext4_count_free_blocks(sb));
2902         if (!err)
2903                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2904                                           ext4_count_free_inodes(sb));
2905         if (!err)
2906                 err = percpu_counter_init(&sbi->s_dirs_counter,
2907                                           ext4_count_dirs(sb));
2908         if (!err)
2909                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2910         if (err) {
2911                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2912                 goto failed_mount_wq;
2913         }
2914
2915         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2916         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2917                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2918                 goto failed_mount_wq;
2919         }
2920
2921         /*
2922          * The jbd2_journal_load will have done any necessary log recovery,
2923          * so we can safely mount the rest of the filesystem now.
2924          */
2925
2926         root = ext4_iget(sb, EXT4_ROOT_INO);
2927         if (IS_ERR(root)) {
2928                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2929                 ret = PTR_ERR(root);
2930                 goto failed_mount4;
2931         }
2932         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2933                 iput(root);
2934                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2935                 goto failed_mount4;
2936         }
2937         sb->s_root = d_alloc_root(root);
2938         if (!sb->s_root) {
2939                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2940                 iput(root);
2941                 ret = -ENOMEM;
2942                 goto failed_mount4;
2943         }
2944
2945         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2946
2947         /* determine the minimum size of new large inodes, if present */
2948         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2949                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2950                                                      EXT4_GOOD_OLD_INODE_SIZE;
2951                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2952                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2953                         if (sbi->s_want_extra_isize <
2954                             le16_to_cpu(es->s_want_extra_isize))
2955                                 sbi->s_want_extra_isize =
2956                                         le16_to_cpu(es->s_want_extra_isize);
2957                         if (sbi->s_want_extra_isize <
2958                             le16_to_cpu(es->s_min_extra_isize))
2959                                 sbi->s_want_extra_isize =
2960                                         le16_to_cpu(es->s_min_extra_isize);
2961                 }
2962         }
2963         /* Check if enough inode space is available */
2964         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2965                                                         sbi->s_inode_size) {
2966                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2967                                                        EXT4_GOOD_OLD_INODE_SIZE;
2968                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2969                          "available");
2970         }
2971
2972         if (test_opt(sb, DELALLOC) &&
2973             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2974                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2975                          "requested data journaling mode");
2976                 clear_opt(sbi->s_mount_opt, DELALLOC);
2977         }
2978         if (test_opt(sb, DIOREAD_NOLOCK)) {
2979                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2980                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2981                                 "option - requested data journaling mode");
2982                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2983                 }
2984                 if (sb->s_blocksize < PAGE_SIZE) {
2985                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2986                                 "option - block size is too small");
2987                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2988                 }
2989         }
2990
2991         err = ext4_setup_system_zone(sb);
2992         if (err) {
2993                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
2994                          "zone (%d)", err);
2995                 goto failed_mount4;
2996         }
2997
2998         ext4_ext_init(sb);
2999         err = ext4_mb_init(sb, needs_recovery);
3000         if (err) {
3001                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
3002                          err);
3003                 goto failed_mount4;
3004         }
3005
3006         sbi->s_kobj.kset = ext4_kset;
3007         init_completion(&sbi->s_kobj_unregister);
3008         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3009                                    "%s", sb->s_id);
3010         if (err) {
3011                 ext4_mb_release(sb);
3012                 ext4_ext_release(sb);
3013                 goto failed_mount4;
3014         };
3015
3016         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3017         ext4_orphan_cleanup(sb, es);
3018         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3019         if (needs_recovery) {
3020                 ext4_msg(sb, KERN_INFO, "recovery complete");
3021                 ext4_mark_recovery_complete(sb, es);
3022         }
3023         if (EXT4_SB(sb)->s_journal) {
3024                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3025                         descr = " journalled data mode";
3026                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3027                         descr = " ordered data mode";
3028                 else
3029                         descr = " writeback data mode";
3030         } else
3031                 descr = "out journal";
3032
3033         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3034                 "Opts: %s", descr, orig_data);
3035
3036         lock_kernel();
3037         kfree(orig_data);
3038         return 0;
3039
3040 cantfind_ext4:
3041         if (!silent)
3042                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3043         goto failed_mount;
3044
3045 failed_mount4:
3046         ext4_msg(sb, KERN_ERR, "mount failed");
3047         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3048 failed_mount_wq:
3049         ext4_release_system_zone(sb);
3050         if (sbi->s_journal) {
3051                 jbd2_journal_destroy(sbi->s_journal);
3052                 sbi->s_journal = NULL;
3053         }
3054         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3055         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3056         percpu_counter_destroy(&sbi->s_dirs_counter);
3057         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3058 failed_mount3:
3059         if (sbi->s_flex_groups) {
3060                 if (is_vmalloc_addr(sbi->s_flex_groups))
3061                         vfree(sbi->s_flex_groups);
3062                 else
3063                         kfree(sbi->s_flex_groups);
3064         }
3065 failed_mount2:
3066         for (i = 0; i < db_count; i++)
3067                 brelse(sbi->s_group_desc[i]);
3068         kfree(sbi->s_group_desc);
3069 failed_mount:
3070         if (sbi->s_proc) {
3071                 remove_proc_entry(sb->s_id, ext4_proc_root);
3072         }
3073 #ifdef CONFIG_QUOTA
3074         for (i = 0; i < MAXQUOTAS; i++)
3075                 kfree(sbi->s_qf_names[i]);
3076 #endif
3077         ext4_blkdev_remove(sbi);
3078         brelse(bh);
3079 out_fail:
3080         sb->s_fs_info = NULL;
3081         kfree(sbi->s_blockgroup_lock);
3082         kfree(sbi);
3083         lock_kernel();
3084         kfree(orig_data);
3085         return ret;
3086 }
3087
3088 /*
3089  * Setup any per-fs journal parameters now.  We'll do this both on
3090  * initial mount, once the journal has been initialised but before we've
3091  * done any recovery; and again on any subsequent remount.
3092  */
3093 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3094 {
3095         struct ext4_sb_info *sbi = EXT4_SB(sb);
3096
3097         journal->j_commit_interval = sbi->s_commit_interval;
3098         journal->j_min_batch_time = sbi->s_min_batch_time;
3099         journal->j_max_batch_time = sbi->s_max_batch_time;
3100
3101         spin_lock(&journal->j_state_lock);
3102         if (test_opt(sb, BARRIER))
3103                 journal->j_flags |= JBD2_BARRIER;
3104         else
3105                 journal->j_flags &= ~JBD2_BARRIER;
3106         if (test_opt(sb, DATA_ERR_ABORT))
3107                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3108         else
3109                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3110         spin_unlock(&journal->j_state_lock);
3111 }
3112
3113 static journal_t *ext4_get_journal(struct super_block *sb,
3114                                    unsigned int journal_inum)
3115 {
3116         struct inode *journal_inode;
3117         journal_t *journal;
3118
3119         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3120
3121         /* First, test for the existence of a valid inode on disk.  Bad
3122          * things happen if we iget() an unused inode, as the subsequent
3123          * iput() will try to delete it. */
3124
3125         journal_inode = ext4_iget(sb, journal_inum);
3126         if (IS_ERR(journal_inode)) {
3127                 ext4_msg(sb, KERN_ERR, "no journal found");
3128                 return NULL;
3129         }
3130         if (!journal_inode->i_nlink) {
3131                 make_bad_inode(journal_inode);
3132                 iput(journal_inode);
3133                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3134                 return NULL;
3135         }
3136
3137         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3138                   journal_inode, journal_inode->i_size);
3139         if (!S_ISREG(journal_inode->i_mode)) {
3140                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3141                 iput(journal_inode);
3142                 return NULL;
3143         }
3144
3145         journal = jbd2_journal_init_inode(journal_inode);
3146         if (!journal) {
3147                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3148                 iput(journal_inode);
3149                 return NULL;
3150         }
3151         journal->j_private = sb;
3152         ext4_init_journal_params(sb, journal);
3153         return journal;
3154 }
3155
3156 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3157                                        dev_t j_dev)
3158 {
3159         struct buffer_head *bh;
3160         journal_t *journal;
3161         ext4_fsblk_t start;
3162         ext4_fsblk_t len;
3163         int hblock, blocksize;
3164         ext4_fsblk_t sb_block;
3165         unsigned long offset;
3166         struct ext4_super_block *es;
3167         struct block_device *bdev;
3168
3169         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3170
3171         bdev = ext4_blkdev_get(j_dev, sb);
3172         if (bdev == NULL)
3173                 return NULL;
3174
3175         if (bd_claim(bdev, sb)) {
3176                 ext4_msg(sb, KERN_ERR,
3177                         "failed to claim external journal device");
3178                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3179                 return NULL;
3180         }
3181
3182         blocksize = sb->s_blocksize;
3183         hblock = bdev_logical_block_size(bdev);
3184         if (blocksize < hblock) {
3185                 ext4_msg(sb, KERN_ERR,
3186                         "blocksize too small for journal device");
3187                 goto out_bdev;
3188         }
3189
3190         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3191         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3192         set_blocksize(bdev, blocksize);
3193         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3194                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3195                        "external journal");
3196                 goto out_bdev;
3197         }
3198
3199         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3200         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3201             !(le32_to_cpu(es->s_feature_incompat) &
3202               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3203                 ext4_msg(sb, KERN_ERR, "external journal has "
3204                                         "bad superblock");
3205                 brelse(bh);
3206                 goto out_bdev;
3207         }
3208
3209         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3210                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3211                 brelse(bh);
3212                 goto out_bdev;
3213         }
3214
3215         len = ext4_blocks_count(es);
3216         start = sb_block + 1;
3217         brelse(bh);     /* we're done with the superblock */
3218
3219         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3220                                         start, len, blocksize);
3221         if (!journal) {
3222                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3223                 goto out_bdev;
3224         }
3225         journal->j_private = sb;
3226         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3227         wait_on_buffer(journal->j_sb_buffer);
3228         if (!buffer_uptodate(journal->j_sb_buffer)) {
3229                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3230                 goto out_journal;
3231         }
3232         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3233                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3234                                         "user (unsupported) - %d",
3235                         be32_to_cpu(journal->j_superblock->s_nr_users));
3236                 goto out_journal;
3237         }
3238         EXT4_SB(sb)->journal_bdev = bdev;
3239         ext4_init_journal_params(sb, journal);
3240         return journal;
3241
3242 out_journal:
3243         jbd2_journal_destroy(journal);
3244 out_bdev:
3245         ext4_blkdev_put(bdev);
3246         return NULL;
3247 }
3248
3249 static int ext4_load_journal(struct super_block *sb,
3250                              struct ext4_super_block *es,
3251                              unsigned long journal_devnum)
3252 {
3253         journal_t *journal;
3254         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3255         dev_t journal_dev;
3256         int err = 0;
3257         int really_read_only;
3258
3259         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3260
3261         if (journal_devnum &&
3262             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3263                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3264                         "numbers have changed");
3265                 journal_dev = new_decode_dev(journal_devnum);
3266         } else
3267                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3268
3269         really_read_only = bdev_read_only(sb->s_bdev);
3270
3271         /*
3272          * Are we loading a blank journal or performing recovery after a
3273          * crash?  For recovery, we need to check in advance whether we
3274          * can get read-write access to the device.
3275          */
3276         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3277                 if (sb->s_flags & MS_RDONLY) {
3278                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3279                                         "required on readonly filesystem");
3280                         if (really_read_only) {
3281                                 ext4_msg(sb, KERN_ERR, "write access "
3282                                         "unavailable, cannot proceed");
3283                                 return -EROFS;
3284                         }
3285                         ext4_msg(sb, KERN_INFO, "write access will "
3286                                "be enabled during recovery");
3287                 }
3288         }
3289
3290         if (journal_inum && journal_dev) {
3291                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3292                        "and inode journals!");
3293                 return -EINVAL;
3294         }
3295
3296         if (journal_inum) {
3297                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3298                         return -EINVAL;
3299         } else {
3300                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3301                         return -EINVAL;
3302         }
3303
3304         if (!(journal->j_flags & JBD2_BARRIER))
3305                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3306
3307         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3308                 err = jbd2_journal_update_format(journal);
3309                 if (err)  {
3310                         ext4_msg(sb, KERN_ERR, "error updating journal");
3311                         jbd2_journal_destroy(journal);
3312                         return err;
3313                 }
3314         }
3315
3316         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3317                 err = jbd2_journal_wipe(journal, !really_read_only);
3318         if (!err)
3319                 err = jbd2_journal_load(journal);
3320
3321         if (err) {
3322                 ext4_msg(sb, KERN_ERR, "error loading journal");
3323                 jbd2_journal_destroy(journal);
3324                 return err;
3325         }
3326
3327         EXT4_SB(sb)->s_journal = journal;
3328         ext4_clear_journal_err(sb, es);
3329
3330         if (journal_devnum &&
3331             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3332                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3333
3334                 /* Make sure we flush the recovery flag to disk. */
3335                 ext4_commit_super(sb, 1);
3336         }
3337
3338         return 0;
3339 }
3340
3341 static int ext4_commit_super(struct super_block *sb, int sync)
3342 {
3343         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3344         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3345         int error = 0;
3346
3347         if (!sbh)
3348                 return error;
3349         if (buffer_write_io_error(sbh)) {
3350                 /*
3351                  * Oh, dear.  A previous attempt to write the
3352                  * superblock failed.  This could happen because the
3353                  * USB device was yanked out.  Or it could happen to
3354                  * be a transient write error and maybe the block will
3355                  * be remapped.  Nothing we can do but to retry the
3356                  * write and hope for the best.
3357                  */
3358                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3359                        "superblock detected");
3360                 clear_buffer_write_io_error(sbh);
3361                 set_buffer_uptodate(sbh);
3362         }
3363         /*
3364          * If the file system is mounted read-only, don't update the
3365          * superblock write time.  This avoids updating the superblock
3366          * write time when we are mounting the root file system
3367          * read/only but we need to replay the journal; at that point,
3368          * for people who are east of GMT and who make their clock
3369          * tick in localtime for Windows bug-for-bug compatibility,
3370          * the clock is set in the future, and this will cause e2fsck
3371          * to complain and force a full file system check.
3372          */
3373         if (!(sb->s_flags & MS_RDONLY))
3374                 es->s_wtime = cpu_to_le32(get_seconds());
3375         es->s_kbytes_written =
3376                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3377                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3378                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3379         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3380                                         &EXT4_SB(sb)->s_freeblocks_counter));
3381         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3382                                         &EXT4_SB(sb)->s_freeinodes_counter));
3383         sb->s_dirt = 0;
3384         BUFFER_TRACE(sbh, "marking dirty");
3385         mark_buffer_dirty(sbh);
3386         if (sync) {
3387                 error = sync_dirty_buffer(sbh);
3388                 if (error)
3389                         return error;
3390
3391                 error = buffer_write_io_error(sbh);
3392                 if (error) {
3393                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3394                                "superblock");
3395                         clear_buffer_write_io_error(sbh);
3396                         set_buffer_uptodate(sbh);
3397                 }
3398         }
3399         return error;
3400 }
3401
3402 /*
3403  * Have we just finished recovery?  If so, and if we are mounting (or
3404  * remounting) the filesystem readonly, then we will end up with a
3405  * consistent fs on disk.  Record that fact.
3406  */
3407 static void ext4_mark_recovery_complete(struct super_block *sb,
3408                                         struct ext4_super_block *es)
3409 {
3410         journal_t *journal = EXT4_SB(sb)->s_journal;
3411
3412         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3413                 BUG_ON(journal != NULL);
3414                 return;
3415         }
3416         jbd2_journal_lock_updates(journal);
3417         if (jbd2_journal_flush(journal) < 0)
3418                 goto out;
3419
3420         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3421             sb->s_flags & MS_RDONLY) {
3422                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3423                 ext4_commit_super(sb, 1);
3424         }
3425
3426 out:
3427         jbd2_journal_unlock_updates(journal);
3428 }
3429
3430 /*
3431  * If we are mounting (or read-write remounting) a filesystem whose journal
3432  * has recorded an error from a previous lifetime, move that error to the
3433  * main filesystem now.
3434  */
3435 static void ext4_clear_journal_err(struct super_block *sb,
3436                                    struct ext4_super_block *es)
3437 {
3438         journal_t *journal;
3439         int j_errno;
3440         const char *errstr;
3441
3442         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3443
3444         journal = EXT4_SB(sb)->s_journal;
3445
3446         /*
3447          * Now check for any error status which may have been recorded in the
3448          * journal by a prior ext4_error() or ext4_abort()
3449          */
3450
3451         j_errno = jbd2_journal_errno(journal);
3452         if (j_errno) {
3453                 char nbuf[16];
3454
3455                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3456                 ext4_warning(sb, "Filesystem error recorded "
3457                              "from previous mount: %s", errstr);
3458                 ext4_warning(sb, "Marking fs in need of filesystem check.");
3459
3460                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3461                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3462                 ext4_commit_super(sb, 1);
3463
3464                 jbd2_journal_clear_err(journal);
3465         }
3466 }
3467
3468 /*
3469  * Force the running and committing transactions to commit,
3470  * and wait on the commit.
3471  */
3472 int ext4_force_commit(struct super_block *sb)
3473 {
3474         journal_t *journal;
3475         int ret = 0;
3476
3477         if (sb->s_flags & MS_RDONLY)
3478                 return 0;
3479
3480         journal = EXT4_SB(sb)->s_journal;
3481         if (journal) {
3482                 vfs_check_frozen(sb, SB_FREEZE_WRITE);
3483                 ret = ext4_journal_force_commit(journal);
3484         }
3485
3486         return ret;
3487 }
3488
3489 static void ext4_write_super(struct super_block *sb)
3490 {
3491         lock_super(sb);
3492         ext4_commit_super(sb, 1);
3493         unlock_super(sb);
3494 }
3495
3496 static int ext4_sync_fs(struct super_block *sb, int wait)
3497 {
3498         int ret = 0;
3499         tid_t target;
3500         struct ext4_sb_info *sbi = EXT4_SB(sb);
3501
3502         trace_ext4_sync_fs(sb, wait);
3503         flush_workqueue(sbi->dio_unwritten_wq);
3504         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3505                 if (wait)
3506                         jbd2_log_wait_commit(sbi->s_journal, target);
3507         }
3508         return ret;
3509 }
3510
3511 /*
3512  * LVM calls this function before a (read-only) snapshot is created.  This
3513  * gives us a chance to flush the journal completely and mark the fs clean.
3514  */
3515 static int ext4_freeze(struct super_block *sb)
3516 {
3517         int error = 0;
3518         journal_t *journal;
3519
3520         if (sb->s_flags & MS_RDONLY)
3521                 return 0;
3522
3523         journal = EXT4_SB(sb)->s_journal;
3524
3525         /* Now we set up the journal barrier. */
3526         jbd2_journal_lock_updates(journal);
3527
3528         /*
3529          * Don't clear the needs_recovery flag if we failed to flush
3530          * the journal.
3531          */
3532         error = jbd2_journal_flush(journal);
3533         if (error < 0)
3534                 goto out;
3535
3536         /* Journal blocked and flushed, clear needs_recovery flag. */
3537         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3538         error = ext4_commit_super(sb, 1);
3539 out:
3540         /* we rely on s_frozen to stop further updates */
3541         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3542         return error;
3543 }
3544
3545 /*
3546  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3547  * flag here, even though the filesystem is not technically dirty yet.
3548  */
3549 static int ext4_unfreeze(struct super_block *sb)
3550 {
3551         if (sb->s_flags & MS_RDONLY)
3552                 return 0;
3553
3554         lock_super(sb);
3555         /* Reset the needs_recovery flag before the fs is unlocked. */
3556         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3557         ext4_commit_super(sb, 1);
3558         unlock_super(sb);
3559         return 0;
3560 }
3561
3562 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3563 {
3564         struct ext4_super_block *es;
3565         struct ext4_sb_info *sbi = EXT4_SB(sb);
3566         ext4_fsblk_t n_blocks_count = 0;
3567         unsigned long old_sb_flags;
3568         struct ext4_mount_options old_opts;
3569         int enable_quota = 0;
3570         ext4_group_t g;
3571         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3572         int err;
3573 #ifdef CONFIG_QUOTA
3574         int i;
3575 #endif
3576         char *orig_data = kstrdup(data, GFP_KERNEL);
3577
3578         lock_kernel();
3579
3580         /* Store the original options */
3581         lock_super(sb);
3582         old_sb_flags = sb->s_flags;
3583         old_opts.s_mount_opt = sbi->s_mount_opt;
3584         old_opts.s_resuid = sbi->s_resuid;
3585         old_opts.s_resgid = sbi->s_resgid;
3586         old_opts.s_commit_interval = sbi->s_commit_interval;
3587         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3588         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3589 #ifdef CONFIG_QUOTA
3590         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3591         for (i = 0; i < MAXQUOTAS; i++)
3592                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3593 #endif
3594         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3595                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3596
3597         /*
3598          * Allow the "check" option to be passed as a remount option.
3599          */
3600         if (!parse_options(data, sb, NULL, &journal_ioprio,
3601                            &n_blocks_count, 1)) {
3602                 err = -EINVAL;
3603                 goto restore_opts;
3604         }
3605
3606         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3607                 ext4_abort(sb, "Abort forced by user");
3608
3609         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3610                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3611
3612         es = sbi->s_es;
3613
3614         if (sbi->s_journal) {
3615                 ext4_init_journal_params(sb, sbi->s_journal);
3616                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3617         }
3618
3619         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3620                 n_blocks_count > ext4_blocks_count(es)) {
3621                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3622                         err = -EROFS;
3623                         goto restore_opts;
3624                 }
3625
3626                 if (*flags & MS_RDONLY) {
3627                         err = dquot_suspend(sb, -1);
3628                         if (err < 0)
3629                                 goto restore_opts;
3630
3631                         /*
3632                          * First of all, the unconditional stuff we have to do
3633                          * to disable replay of the journal when we next remount
3634                          */
3635                         sb->s_flags |= MS_RDONLY;
3636
3637                         /*
3638                          * OK, test if we are remounting a valid rw partition
3639                          * readonly, and if so set the rdonly flag and then
3640                          * mark the partition as valid again.
3641                          */
3642                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3643                             (sbi->s_mount_state & EXT4_VALID_FS))
3644                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3645
3646                         if (sbi->s_journal)
3647                                 ext4_mark_recovery_complete(sb, es);
3648                 } else {
3649                         /* Make sure we can mount this feature set readwrite */
3650                         if (!ext4_feature_set_ok(sb, 0)) {
3651                                 err = -EROFS;
3652                                 goto restore_opts;
3653                         }
3654                         /*
3655                          * Make sure the group descriptor checksums
3656                          * are sane.  If they aren't, refuse to remount r/w.
3657                          */
3658                         for (g = 0; g < sbi->s_groups_count; g++) {
3659                                 struct ext4_group_desc *gdp =
3660                                         ext4_get_group_desc(sb, g, NULL);
3661
3662                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3663                                         ext4_msg(sb, KERN_ERR,
3664                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3665                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3666                                                le16_to_cpu(gdp->bg_checksum));
3667                                         err = -EINVAL;
3668                                         goto restore_opts;
3669                                 }
3670                         }
3671
3672                         /*
3673                          * If we have an unprocessed orphan list hanging
3674                          * around from a previously readonly bdev mount,
3675                          * require a full umount/remount for now.
3676                          */
3677                         if (es->s_last_orphan) {
3678                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3679                                        "remount RDWR because of unprocessed "
3680                                        "orphan inode list.  Please "
3681                                        "umount/remount instead");
3682                                 err = -EINVAL;
3683                                 goto restore_opts;
3684                         }
3685
3686                         /*
3687                          * Mounting a RDONLY partition read-write, so reread
3688                          * and store the current valid flag.  (It may have
3689                          * been changed by e2fsck since we originally mounted
3690                          * the partition.)
3691                          */
3692                         if (sbi->s_journal)
3693                                 ext4_clear_journal_err(sb, es);
3694                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3695                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3696                                 goto restore_opts;
3697                         if (!ext4_setup_super(sb, es, 0))
3698                                 sb->s_flags &= ~MS_RDONLY;
3699                         enable_quota = 1;
3700                 }
3701         }
3702         ext4_setup_system_zone(sb);
3703         if (sbi->s_journal == NULL)
3704                 ext4_commit_super(sb, 1);
3705
3706 #ifdef CONFIG_QUOTA
3707         /* Release old quota file names */
3708         for (i = 0; i < MAXQUOTAS; i++)
3709                 if (old_opts.s_qf_names[i] &&
3710                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3711                         kfree(old_opts.s_qf_names[i]);
3712 #endif
3713         unlock_super(sb);
3714         unlock_kernel();
3715         if (enable_quota)
3716                 dquot_resume(sb, -1);
3717
3718         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
3719         kfree(orig_data);
3720         return 0;
3721
3722 restore_opts:
3723         sb->s_flags = old_sb_flags;
3724         sbi->s_mount_opt = old_opts.s_mount_opt;
3725         sbi->s_resuid = old_opts.s_resuid;
3726         sbi->s_resgid = old_opts.s_resgid;
3727         sbi->s_commit_interval = old_opts.s_commit_interval;
3728         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3729         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3730 #ifdef CONFIG_QUOTA
3731         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3732         for (i = 0; i < MAXQUOTAS; i++) {
3733                 if (sbi->s_qf_names[i] &&
3734                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3735                         kfree(sbi->s_qf_names[i]);
3736                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3737         }
3738 #endif
3739         unlock_super(sb);
3740         unlock_kernel();
3741         kfree(orig_data);
3742         return err;
3743 }
3744
3745 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3746 {
3747         struct super_block *sb = dentry->d_sb;
3748         struct ext4_sb_info *sbi = EXT4_SB(sb);
3749         struct ext4_super_block *es = sbi->s_es;
3750         u64 fsid;
3751
3752         if (test_opt(sb, MINIX_DF)) {
3753                 sbi->s_overhead_last = 0;
3754         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3755                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3756                 ext4_fsblk_t overhead = 0;
3757
3758                 /*
3759                  * Compute the overhead (FS structures).  This is constant
3760                  * for a given filesystem unless the number of block groups
3761                  * changes so we cache the previous value until it does.
3762                  */
3763
3764                 /*
3765                  * All of the blocks before first_data_block are
3766                  * overhead
3767                  */
3768                 overhead = le32_to_cpu(es->s_first_data_block);
3769
3770                 /*
3771                  * Add the overhead attributed to the superblock and
3772                  * block group descriptors.  If the sparse superblocks
3773                  * feature is turned on, then not all groups have this.
3774                  */
3775                 for (i = 0; i < ngroups; i++) {
3776                         overhead += ext4_bg_has_super(sb, i) +
3777                                 ext4_bg_num_gdb(sb, i);
3778                         cond_resched();
3779                 }
3780
3781                 /*
3782                  * Every block group has an inode bitmap, a block
3783                  * bitmap, and an inode table.
3784                  */
3785                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3786                 sbi->s_overhead_last = overhead;
3787                 smp_wmb();
3788                 sbi->s_blocks_last = ext4_blocks_count(es);
3789         }
3790
3791         buf->f_type = EXT4_SUPER_MAGIC;
3792         buf->f_bsize = sb->s_blocksize;
3793         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3794         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3795                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3796         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3797         if (buf->f_bfree < ext4_r_blocks_count(es))
3798                 buf->f_bavail = 0;
3799         buf->f_files = le32_to_cpu(es->s_inodes_count);
3800         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3801         buf->f_namelen = EXT4_NAME_LEN;
3802         fsid = le64_to_cpup((void *)es->s_uuid) ^
3803                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3804         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3805         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3806
3807         return 0;
3808 }
3809
3810 /* Helper function for writing quotas on sync - we need to start transaction
3811  * before quota file is locked for write. Otherwise the are possible deadlocks:
3812  * Process 1                         Process 2
3813  * ext4_create()                     quota_sync()
3814  *   jbd2_journal_start()                  write_dquot()
3815  *   dquot_initialize()                         down(dqio_mutex)
3816  *     down(dqio_mutex)                    jbd2_journal_start()
3817  *
3818  */
3819
3820 #ifdef CONFIG_QUOTA
3821
3822 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3823 {
3824         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3825 }
3826
3827 static int ext4_write_dquot(struct dquot *dquot)
3828 {
3829         int ret, err;
3830         handle_t *handle;
3831         struct inode *inode;
3832
3833         inode = dquot_to_inode(dquot);
3834         handle = ext4_journal_start(inode,
3835                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3836         if (IS_ERR(handle))
3837                 return PTR_ERR(handle);
3838         ret = dquot_commit(dquot);
3839         err = ext4_journal_stop(handle);
3840         if (!ret)
3841                 ret = err;
3842         return ret;
3843 }
3844
3845 static int ext4_acquire_dquot(struct dquot *dquot)
3846 {
3847         int ret, err;
3848         handle_t *handle;
3849
3850         handle = ext4_journal_start(dquot_to_inode(dquot),
3851                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3852         if (IS_ERR(handle))
3853                 return PTR_ERR(handle);
3854         ret = dquot_acquire(dquot);
3855         err = ext4_journal_stop(handle);
3856         if (!ret)
3857                 ret = err;
3858         return ret;
3859 }
3860
3861 static int ext4_release_dquot(struct dquot *dquot)
3862 {
3863         int ret, err;
3864         handle_t *handle;
3865
3866         handle = ext4_journal_start(dquot_to_inode(dquot),
3867                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3868         if (IS_ERR(handle)) {
3869                 /* Release dquot anyway to avoid endless cycle in dqput() */
3870                 dquot_release(dquot);
3871                 return PTR_ERR(handle);
3872         }
3873         ret = dquot_release(dquot);
3874         err = ext4_journal_stop(handle);
3875         if (!ret)
3876                 ret = err;
3877         return ret;
3878 }
3879
3880 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3881 {
3882         /* Are we journaling quotas? */
3883         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3884             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3885                 dquot_mark_dquot_dirty(dquot);
3886                 return ext4_write_dquot(dquot);
3887         } else {
3888                 return dquot_mark_dquot_dirty(dquot);
3889         }
3890 }
3891
3892 static int ext4_write_info(struct super_block *sb, int type)
3893 {
3894         int ret, err;
3895         handle_t *handle;
3896
3897         /* Data block + inode block */
3898         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3899         if (IS_ERR(handle))
3900                 return PTR_ERR(handle);
3901         ret = dquot_commit_info(sb, type);
3902         err = ext4_journal_stop(handle);
3903         if (!ret)
3904                 ret = err;
3905         return ret;
3906 }
3907
3908 /*
3909  * Turn on quotas during mount time - we need to find
3910  * the quota file and such...
3911  */
3912 static int ext4_quota_on_mount(struct super_block *sb, int type)
3913 {
3914         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3915                                         EXT4_SB(sb)->s_jquota_fmt, type);
3916 }
3917
3918 /*
3919  * Standard function to be called on quota_on
3920  */
3921 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3922                          char *name)
3923 {
3924         int err;
3925         struct path path;
3926
3927         if (!test_opt(sb, QUOTA))
3928                 return -EINVAL;
3929
3930         err = kern_path(name, LOOKUP_FOLLOW, &path);
3931         if (err)
3932                 return err;
3933
3934         /* Quotafile not on the same filesystem? */
3935         if (path.mnt->mnt_sb != sb) {
3936                 path_put(&path);
3937                 return -EXDEV;
3938         }
3939         /* Journaling quota? */
3940         if (EXT4_SB(sb)->s_qf_names[type]) {
3941                 /* Quotafile not in fs root? */
3942                 if (path.dentry->d_parent != sb->s_root)
3943                         ext4_msg(sb, KERN_WARNING,
3944                                 "Quota file not on filesystem root. "
3945                                 "Journaled quota will not work");
3946         }
3947
3948         /*
3949          * When we journal data on quota file, we have to flush journal to see
3950          * all updates to the file when we bypass pagecache...
3951          */
3952         if (EXT4_SB(sb)->s_journal &&
3953             ext4_should_journal_data(path.dentry->d_inode)) {
3954                 /*
3955                  * We don't need to lock updates but journal_flush() could
3956                  * otherwise be livelocked...
3957                  */
3958                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3959                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3960                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3961                 if (err) {
3962                         path_put(&path);
3963                         return err;
3964                 }
3965         }
3966
3967         err = dquot_quota_on_path(sb, type, format_id, &path);
3968         path_put(&path);
3969         return err;
3970 }
3971
3972 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3973  * acquiring the locks... As quota files are never truncated and quota code
3974  * itself serializes the operations (and noone else should touch the files)
3975  * we don't have to be afraid of races */
3976 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3977                                size_t len, loff_t off)
3978 {
3979         struct inode *inode = sb_dqopt(sb)->files[type];
3980         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3981         int err = 0;
3982         int offset = off & (sb->s_blocksize - 1);
3983         int tocopy;
3984         size_t toread;
3985         struct buffer_head *bh;
3986         loff_t i_size = i_size_read(inode);
3987
3988         if (off > i_size)
3989                 return 0;
3990         if (off+len > i_size)
3991                 len = i_size-off;
3992         toread = len;
3993         while (toread > 0) {
3994                 tocopy = sb->s_blocksize - offset < toread ?
3995                                 sb->s_blocksize - offset : toread;
3996                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3997                 if (err)
3998                         return err;
3999                 if (!bh)        /* A hole? */
4000                         memset(data, 0, tocopy);
4001                 else
4002                         memcpy(data, bh->b_data+offset, tocopy);
4003                 brelse(bh);
4004                 offset = 0;
4005                 toread -= tocopy;
4006                 data += tocopy;
4007                 blk++;
4008         }
4009         return len;
4010 }
4011
4012 /* Write to quotafile (we know the transaction is already started and has
4013  * enough credits) */
4014 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4015                                 const char *data, size_t len, loff_t off)
4016 {
4017         struct inode *inode = sb_dqopt(sb)->files[type];
4018         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4019         int err = 0;
4020         int offset = off & (sb->s_blocksize - 1);
4021         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
4022         struct buffer_head *bh;
4023         handle_t *handle = journal_current_handle();
4024
4025         if (EXT4_SB(sb)->s_journal && !handle) {
4026                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4027                         " cancelled because transaction is not started",
4028                         (unsigned long long)off, (unsigned long long)len);
4029                 return -EIO;
4030         }
4031         /*
4032          * Since we account only one data block in transaction credits,
4033          * then it is impossible to cross a block boundary.
4034          */
4035         if (sb->s_blocksize - offset < len) {
4036                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4037                         " cancelled because not block aligned",
4038                         (unsigned long long)off, (unsigned long long)len);
4039                 return -EIO;
4040         }
4041
4042         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4043         bh = ext4_bread(handle, inode, blk, 1, &err);
4044         if (!bh)
4045                 goto out;
4046         if (journal_quota) {
4047                 err = ext4_journal_get_write_access(handle, bh);
4048                 if (err) {
4049                         brelse(bh);
4050                         goto out;
4051                 }
4052         }
4053         lock_buffer(bh);
4054         memcpy(bh->b_data+offset, data, len);
4055         flush_dcache_page(bh->b_page);
4056         unlock_buffer(bh);
4057         if (journal_quota)
4058                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
4059         else {
4060                 /* Always do at least ordered writes for quotas */
4061                 err = ext4_jbd2_file_inode(handle, inode);
4062                 mark_buffer_dirty(bh);
4063         }
4064         brelse(bh);
4065 out:
4066         if (err) {
4067                 mutex_unlock(&inode->i_mutex);
4068                 return err;
4069         }
4070         if (inode->i_size < off + len) {
4071                 i_size_write(inode, off + len);
4072                 EXT4_I(inode)->i_disksize = inode->i_size;
4073         }
4074         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4075         ext4_mark_inode_dirty(handle, inode);
4076         mutex_unlock(&inode->i_mutex);
4077         return len;
4078 }
4079
4080 #endif
4081
4082 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
4083                        const char *dev_name, void *data, struct vfsmount *mnt)
4084 {
4085         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
4086 }
4087
4088 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4089 static struct file_system_type ext2_fs_type = {
4090         .owner          = THIS_MODULE,
4091         .name           = "ext2",
4092         .get_sb         = ext4_get_sb,
4093         .kill_sb        = kill_block_super,
4094         .fs_flags       = FS_REQUIRES_DEV,
4095 };
4096
4097 static inline void register_as_ext2(void)
4098 {
4099         int err = register_filesystem(&ext2_fs_type);
4100         if (err)
4101                 printk(KERN_WARNING
4102                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4103 }
4104
4105 static inline void unregister_as_ext2(void)
4106 {
4107         unregister_filesystem(&ext2_fs_type);
4108 }
4109 MODULE_ALIAS("ext2");
4110 #else
4111 static inline void register_as_ext2(void) { }
4112 static inline void unregister_as_ext2(void) { }
4113 #endif
4114
4115 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4116 static inline void register_as_ext3(void)
4117 {
4118         int err = register_filesystem(&ext3_fs_type);
4119         if (err)
4120                 printk(KERN_WARNING
4121                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4122 }
4123
4124 static inline void unregister_as_ext3(void)
4125 {
4126         unregister_filesystem(&ext3_fs_type);
4127 }
4128 MODULE_ALIAS("ext3");
4129 #else
4130 static inline void register_as_ext3(void) { }
4131 static inline void unregister_as_ext3(void) { }
4132 #endif
4133
4134 static struct file_system_type ext4_fs_type = {
4135         .owner          = THIS_MODULE,
4136         .name           = "ext4",
4137         .get_sb         = ext4_get_sb,
4138         .kill_sb        = kill_block_super,
4139         .fs_flags       = FS_REQUIRES_DEV,
4140 };
4141
4142 static int __init init_ext4_fs(void)
4143 {
4144         int err;
4145
4146         ext4_check_flag_values();
4147         err = init_ext4_system_zone();
4148         if (err)
4149                 return err;
4150         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4151         if (!ext4_kset)
4152                 goto out4;
4153         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4154         err = init_ext4_mballoc();
4155         if (err)
4156                 goto out3;
4157
4158         err = init_ext4_xattr();
4159         if (err)
4160                 goto out2;
4161         err = init_inodecache();
4162         if (err)
4163                 goto out1;
4164         register_as_ext2();
4165         register_as_ext3();
4166         err = register_filesystem(&ext4_fs_type);
4167         if (err)
4168                 goto out;
4169         return 0;
4170 out:
4171         unregister_as_ext2();
4172         unregister_as_ext3();
4173         destroy_inodecache();
4174 out1:
4175         exit_ext4_xattr();
4176 out2:
4177         exit_ext4_mballoc();
4178 out3:
4179         remove_proc_entry("fs/ext4", NULL);
4180         kset_unregister(ext4_kset);
4181 out4:
4182         exit_ext4_system_zone();
4183         return err;
4184 }
4185
4186 static void __exit exit_ext4_fs(void)
4187 {
4188         unregister_as_ext2();
4189         unregister_as_ext3();
4190         unregister_filesystem(&ext4_fs_type);
4191         destroy_inodecache();
4192         exit_ext4_xattr();
4193         exit_ext4_mballoc();
4194         remove_proc_entry("fs/ext4", NULL);
4195         kset_unregister(ext4_kset);
4196         exit_ext4_system_zone();
4197 }
4198
4199 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4200 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4201 MODULE_LICENSE("GPL");
4202 module_init(init_ext4_fs)
4203 module_exit(exit_ext4_fs)