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/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static void ext4_write_super(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 111 #else
 112 #define IS_EXT3_SB(sb) (0)
 113 #endif
 114
 115 void *ext4_kvmalloc(size_t size, gfp_t flags)
 116 {
 117         void *ret;
 118
 119         ret = kmalloc(size, flags);
 120         if (!ret)
 121                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 122         return ret;
 123 }
 124
 125 void *ext4_kvzalloc(size_t size, gfp_t flags)
 126 {
 127         void *ret;
 128
 129         ret = kzalloc(size, flags);
 130         if (!ret)
 131                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 132         return ret;
 133 }
 134
 135 void ext4_kvfree(void *ptr)
 136 {
 137         if (is_vmalloc_addr(ptr))
 138                 vfree(ptr);
 139         else
 140                 kfree(ptr);
 141
 142 }
 143
 144 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 145                                struct ext4_group_desc *bg)
 146 {
 147         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 150 }
 151
 152 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 153                                struct ext4_group_desc *bg)
 154 {
 155         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 156                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 157                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 158 }
 159
 160 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 161                               struct ext4_group_desc *bg)
 162 {
 163         return le32_to_cpu(bg->bg_inode_table_lo) |
 164                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 165                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 166 }
 167
 168 __u32 ext4_free_group_clusters(struct super_block *sb,
 169                                struct ext4_group_desc *bg)
 170 {
 171         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 172                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 173                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 174 }
 175
 176 __u32 ext4_free_inodes_count(struct super_block *sb,
 177                               struct ext4_group_desc *bg)
 178 {
 179         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 180                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 181                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 182 }
 183
 184 __u32 ext4_used_dirs_count(struct super_block *sb,
 185                               struct ext4_group_desc *bg)
 186 {
 187         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 188                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 189                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 190 }
 191
 192 __u32 ext4_itable_unused_count(struct super_block *sb,
 193                               struct ext4_group_desc *bg)
 194 {
 195         return le16_to_cpu(bg->bg_itable_unused_lo) |
 196                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 197                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 198 }
 199
 200 void ext4_block_bitmap_set(struct super_block *sb,
 201                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 202 {
 203         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 206 }
 207
 208 void ext4_inode_bitmap_set(struct super_block *sb,
 209                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 210 {
 211         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 212         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 213                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 214 }
 215
 216 void ext4_inode_table_set(struct super_block *sb,
 217                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 218 {
 219         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 220         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 221                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 222 }
 223
 224 void ext4_free_group_clusters_set(struct super_block *sb,
 225                                   struct ext4_group_desc *bg, __u32 count)
 226 {
 227         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 228         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 229                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 230 }
 231
 232 void ext4_free_inodes_set(struct super_block *sb,
 233                           struct ext4_group_desc *bg, __u32 count)
 234 {
 235         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 236         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 237                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 238 }
 239
 240 void ext4_used_dirs_set(struct super_block *sb,
 241                           struct ext4_group_desc *bg, __u32 count)
 242 {
 243         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 244         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 245                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 246 }
 247
 248 void ext4_itable_unused_set(struct super_block *sb,
 249                           struct ext4_group_desc *bg, __u32 count)
 250 {
 251         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 252         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 253                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 254 }
 255
 256
 257 /* Just increment the non-pointer handle value */
 258 static handle_t *ext4_get_nojournal(void)
 259 {
 260         handle_t *handle = current->journal_info;
 261         unsigned long ref_cnt = (unsigned long)handle;
 262
 263         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 264
 265         ref_cnt++;
 266         handle = (handle_t *)ref_cnt;
 267
 268         current->journal_info = handle;
 269         return handle;
 270 }
 271
 272
 273 /* Decrement the non-pointer handle value */
 274 static void ext4_put_nojournal(handle_t *handle)
 275 {
 276         unsigned long ref_cnt = (unsigned long)handle;
 277
 278         BUG_ON(ref_cnt == 0);
 279
 280         ref_cnt--;
 281         handle = (handle_t *)ref_cnt;
 282
 283         current->journal_info = handle;
 284 }
 285
 286 /*
 287  * Wrappers for jbd2_journal_start/end.
 288  *
 289  * The only special thing we need to do here is to make sure that all
 290  * journal_end calls result in the superblock being marked dirty, so
 291  * that sync() will call the filesystem's write_super callback if
 292  * appropriate.
 293  *
 294  * To avoid j_barrier hold in userspace when a user calls freeze(),
 295  * ext4 prevents a new handle from being started by s_frozen, which
 296  * is in an upper layer.
 297  */
 298 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 299 {
 300         journal_t *journal;
 301         handle_t  *handle;
 302
 303         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 304         if (sb->s_flags & MS_RDONLY)
 305                 return ERR_PTR(-EROFS);
 306
 307         journal = EXT4_SB(sb)->s_journal;
 308         handle = ext4_journal_current_handle();
 309
 310         /*
 311          * If a handle has been started, it should be allowed to
 312          * finish, otherwise deadlock could happen between freeze
 313          * and others(e.g. truncate) due to the restart of the
 314          * journal handle if the filesystem is forzen and active
 315          * handles are not stopped.
 316          */
 317         if (!handle)
 318                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 319
 320         if (!journal)
 321                 return ext4_get_nojournal();
 322         /*
 323          * Special case here: if the journal has aborted behind our
 324          * backs (eg. EIO in the commit thread), then we still need to
 325          * take the FS itself readonly cleanly.
 326          */
 327         if (is_journal_aborted(journal)) {
 328                 ext4_abort(sb, "Detected aborted journal");
 329                 return ERR_PTR(-EROFS);
 330         }
 331         return jbd2_journal_start(journal, nblocks);
 332 }
 333
 334 /*
 335  * The only special thing we need to do here is to make sure that all
 336  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 337  * that sync() will call the filesystem's write_super callback if
 338  * appropriate.
 339  */
 340 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 341 {
 342         struct super_block *sb;
 343         int err;
 344         int rc;
 345
 346         if (!ext4_handle_valid(handle)) {
 347                 ext4_put_nojournal(handle);
 348                 return 0;
 349         }
 350         sb = handle->h_transaction->t_journal->j_private;
 351         err = handle->h_err;
 352         rc = jbd2_journal_stop(handle);
 353
 354         if (!err)
 355                 err = rc;
 356         if (err)
 357                 __ext4_std_error(sb, where, line, err);
 358         return err;
 359 }
 360
 361 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 362                                const char *err_fn, struct buffer_head *bh,
 363                                handle_t *handle, int err)
 364 {
 365         char nbuf[16];
 366         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 367
 368         BUG_ON(!ext4_handle_valid(handle));
 369
 370         if (bh)
 371                 BUFFER_TRACE(bh, "abort");
 372
 373         if (!handle->h_err)
 374                 handle->h_err = err;
 375
 376         if (is_handle_aborted(handle))
 377                 return;
 378
 379         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 380                caller, line, errstr, err_fn);
 381
 382         jbd2_journal_abort_handle(handle);
 383 }
 384
 385 static void __save_error_info(struct super_block *sb, const char *func,
 386                             unsigned int line)
 387 {
 388         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 389
 390         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 391         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 392         es->s_last_error_time = cpu_to_le32(get_seconds());
 393         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 394         es->s_last_error_line = cpu_to_le32(line);
 395         if (!es->s_first_error_time) {
 396                 es->s_first_error_time = es->s_last_error_time;
 397                 strncpy(es->s_first_error_func, func,
 398                         sizeof(es->s_first_error_func));
 399                 es->s_first_error_line = cpu_to_le32(line);
 400                 es->s_first_error_ino = es->s_last_error_ino;
 401                 es->s_first_error_block = es->s_last_error_block;
 402         }
 403         /*
 404          * Start the daily error reporting function if it hasn't been
 405          * started already
 406          */
 407         if (!es->s_error_count)
 408                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 409         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 410 }
 411
 412 static void save_error_info(struct super_block *sb, const char *func,
 413                             unsigned int line)
 414 {
 415         __save_error_info(sb, func, line);
 416         ext4_commit_super(sb, 1);
 417 }
 418
 419 /*
 420  * The del_gendisk() function uninitializes the disk-specific data
 421  * structures, including the bdi structure, without telling anyone
 422  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 423  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 424  * This is a kludge to prevent these oops until we can put in a proper
 425  * hook in del_gendisk() to inform the VFS and file system layers.
 426  */
 427 static int block_device_ejected(struct super_block *sb)
 428 {
 429         struct inode *bd_inode = sb->s_bdev->bd_inode;
 430         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 431
 432         return bdi->dev == NULL;
 433 }
 434
 435 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 436 {
 437         struct super_block              *sb = journal->j_private;
 438         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 439         int                             error = is_journal_aborted(journal);
 440         struct ext4_journal_cb_entry    *jce, *tmp;
 441
 442         spin_lock(&sbi->s_md_lock);
 443         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 444                 list_del_init(&jce->jce_list);
 445                 spin_unlock(&sbi->s_md_lock);
 446                 jce->jce_func(sb, jce, error);
 447                 spin_lock(&sbi->s_md_lock);
 448         }
 449         spin_unlock(&sbi->s_md_lock);
 450 }
 451
 452 /* Deal with the reporting of failure conditions on a filesystem such as
 453  * inconsistencies detected or read IO failures.
 454  *
 455  * On ext2, we can store the error state of the filesystem in the
 456  * superblock.  That is not possible on ext4, because we may have other
 457  * write ordering constraints on the superblock which prevent us from
 458  * writing it out straight away; and given that the journal is about to
 459  * be aborted, we can't rely on the current, or future, transactions to
 460  * write out the superblock safely.
 461  *
 462  * We'll just use the jbd2_journal_abort() error code to record an error in
 463  * the journal instead.  On recovery, the journal will complain about
 464  * that error until we've noted it down and cleared it.
 465  */
 466
 467 static void ext4_handle_error(struct super_block *sb)
 468 {
 469         if (sb->s_flags & MS_RDONLY)
 470                 return;
 471
 472         if (!test_opt(sb, ERRORS_CONT)) {
 473                 journal_t *journal = EXT4_SB(sb)->s_journal;
 474
 475                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 476                 if (journal)
 477                         jbd2_journal_abort(journal, -EIO);
 478         }
 479         if (test_opt(sb, ERRORS_RO)) {
 480                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 481                 sb->s_flags |= MS_RDONLY;
 482         }
 483         if (test_opt(sb, ERRORS_PANIC))
 484                 panic("EXT4-fs (device %s): panic forced after error\n",
 485                         sb->s_id);
 486 }
 487
 488 void __ext4_error(struct super_block *sb, const char *function,
 489                   unsigned int line, const char *fmt, ...)
 490 {
 491         struct va_format vaf;
 492         va_list args;
 493
 494         va_start(args, fmt);
 495         vaf.fmt = fmt;
 496         vaf.va = &args;
 497         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 498                sb->s_id, function, line, current->comm, &vaf);
 499         va_end(args);
 500         save_error_info(sb, function, line);
 501
 502         ext4_handle_error(sb);
 503 }
 504
 505 void ext4_error_inode(struct inode *inode, const char *function,
 506                       unsigned int line, ext4_fsblk_t block,
 507                       const char *fmt, ...)
 508 {
 509         va_list args;
 510         struct va_format vaf;
 511         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 512
 513         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 514         es->s_last_error_block = cpu_to_le64(block);
 515         save_error_info(inode->i_sb, function, line);
 516         va_start(args, fmt);
 517         vaf.fmt = fmt;
 518         vaf.va = &args;
 519         if (block)
 520                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 521                        "inode #%lu: block %llu: comm %s: %pV\n",
 522                        inode->i_sb->s_id, function, line, inode->i_ino,
 523                        block, current->comm, &vaf);
 524         else
 525                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 526                        "inode #%lu: comm %s: %pV\n",
 527                        inode->i_sb->s_id, function, line, inode->i_ino,
 528                        current->comm, &vaf);
 529         va_end(args);
 530
 531         ext4_handle_error(inode->i_sb);
 532 }
 533
 534 void ext4_error_file(struct file *file, const char *function,
 535                      unsigned int line, ext4_fsblk_t block,
 536                      const char *fmt, ...)
 537 {
 538         va_list args;
 539         struct va_format vaf;
 540         struct ext4_super_block *es;
 541         struct inode *inode = file->f_dentry->d_inode;
 542         char pathname[80], *path;
 543
 544         es = EXT4_SB(inode->i_sb)->s_es;
 545         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 546         save_error_info(inode->i_sb, function, line);
 547         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 548         if (IS_ERR(path))
 549                 path = "(unknown)";
 550         va_start(args, fmt);
 551         vaf.fmt = fmt;
 552         vaf.va = &args;
 553         if (block)
 554                 printk(KERN_CRIT
 555                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 556                        "block %llu: comm %s: path %s: %pV\n",
 557                        inode->i_sb->s_id, function, line, inode->i_ino,
 558                        block, current->comm, path, &vaf);
 559         else
 560                 printk(KERN_CRIT
 561                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 562                        "comm %s: path %s: %pV\n",
 563                        inode->i_sb->s_id, function, line, inode->i_ino,
 564                        current->comm, path, &vaf);
 565         va_end(args);
 566
 567         ext4_handle_error(inode->i_sb);
 568 }
 569
 570 static const char *ext4_decode_error(struct super_block *sb, int errno,
 571                                      char nbuf[16])
 572 {
 573         char *errstr = NULL;
 574
 575         switch (errno) {
 576         case -EIO:
 577                 errstr = "IO failure";
 578                 break;
 579         case -ENOMEM:
 580                 errstr = "Out of memory";
 581                 break;
 582         case -EROFS:
 583                 if (!sb || (EXT4_SB(sb)->s_journal &&
 584                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 585                         errstr = "Journal has aborted";
 586                 else
 587                         errstr = "Readonly filesystem";
 588                 break;
 589         default:
 590                 /* If the caller passed in an extra buffer for unknown
 591                  * errors, textualise them now.  Else we just return
 592                  * NULL. */
 593                 if (nbuf) {
 594                         /* Check for truncated error codes... */
 595                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 596                                 errstr = nbuf;
 597                 }
 598                 break;
 599         }
 600
 601         return errstr;
 602 }
 603
 604 /* __ext4_std_error decodes expected errors from journaling functions
 605  * automatically and invokes the appropriate error response.  */
 606
 607 void __ext4_std_error(struct super_block *sb, const char *function,
 608                       unsigned int line, int errno)
 609 {
 610         char nbuf[16];
 611         const char *errstr;
 612
 613         /* Special case: if the error is EROFS, and we're not already
 614          * inside a transaction, then there's really no point in logging
 615          * an error. */
 616         if (errno == -EROFS && journal_current_handle() == NULL &&
 617             (sb->s_flags & MS_RDONLY))
 618                 return;
 619
 620         errstr = ext4_decode_error(sb, errno, nbuf);
 621         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 622                sb->s_id, function, line, errstr);
 623         save_error_info(sb, function, line);
 624
 625         ext4_handle_error(sb);
 626 }
 627
 628 /*
 629  * ext4_abort is a much stronger failure handler than ext4_error.  The
 630  * abort function may be used to deal with unrecoverable failures such
 631  * as journal IO errors or ENOMEM at a critical moment in log management.
 632  *
 633  * We unconditionally force the filesystem into an ABORT|READONLY state,
 634  * unless the error response on the fs has been set to panic in which
 635  * case we take the easy way out and panic immediately.
 636  */
 637
 638 void __ext4_abort(struct super_block *sb, const char *function,
 639                 unsigned int line, const char *fmt, ...)
 640 {
 641         va_list args;
 642
 643         save_error_info(sb, function, line);
 644         va_start(args, fmt);
 645         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 646                function, line);
 647         vprintk(fmt, args);
 648         printk("\n");
 649         va_end(args);
 650
 651         if ((sb->s_flags & MS_RDONLY) == 0) {
 652                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 653                 sb->s_flags |= MS_RDONLY;
 654                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 655                 if (EXT4_SB(sb)->s_journal)
 656                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 657                 save_error_info(sb, function, line);
 658         }
 659         if (test_opt(sb, ERRORS_PANIC))
 660                 panic("EXT4-fs panic from previous error\n");
 661 }
 662
 663 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 664 {
 665         struct va_format vaf;
 666         va_list args;
 667
 668         va_start(args, fmt);
 669         vaf.fmt = fmt;
 670         vaf.va = &args;
 671         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 672         va_end(args);
 673 }
 674
 675 void __ext4_warning(struct super_block *sb, const char *function,
 676                     unsigned int line, const char *fmt, ...)
 677 {
 678         struct va_format vaf;
 679         va_list args;
 680
 681         va_start(args, fmt);
 682         vaf.fmt = fmt;
 683         vaf.va = &args;
 684         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 685                sb->s_id, function, line, &vaf);
 686         va_end(args);
 687 }
 688
 689 void __ext4_grp_locked_error(const char *function, unsigned int line,
 690                              struct super_block *sb, ext4_group_t grp,
 691                              unsigned long ino, ext4_fsblk_t block,
 692                              const char *fmt, ...)
 693 __releases(bitlock)
 694 __acquires(bitlock)
 695 {
 696         struct va_format vaf;
 697         va_list args;
 698         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 699
 700         es->s_last_error_ino = cpu_to_le32(ino);
 701         es->s_last_error_block = cpu_to_le64(block);
 702         __save_error_info(sb, function, line);
 703
 704         va_start(args, fmt);
 705
 706         vaf.fmt = fmt;
 707         vaf.va = &args;
 708         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 709                sb->s_id, function, line, grp);
 710         if (ino)
 711                 printk(KERN_CONT "inode %lu: ", ino);
 712         if (block)
 713                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 714         printk(KERN_CONT "%pV\n", &vaf);
 715         va_end(args);
 716
 717         if (test_opt(sb, ERRORS_CONT)) {
 718                 ext4_commit_super(sb, 0);
 719                 return;
 720         }
 721
 722         ext4_unlock_group(sb, grp);
 723         ext4_handle_error(sb);
 724         /*
 725          * We only get here in the ERRORS_RO case; relocking the group
 726          * may be dangerous, but nothing bad will happen since the
 727          * filesystem will have already been marked read/only and the
 728          * journal has been aborted.  We return 1 as a hint to callers
 729          * who might what to use the return value from
 730          * ext4_grp_locked_error() to distinguish between the
 731          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 732          * aggressively from the ext4 function in question, with a
 733          * more appropriate error code.
 734          */
 735         ext4_lock_group(sb, grp);
 736         return;
 737 }
 738
 739 void ext4_update_dynamic_rev(struct super_block *sb)
 740 {
 741         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 742
 743         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 744                 return;
 745
 746         ext4_warning(sb,
 747                      "updating to rev %d because of new feature flag, "
 748                      "running e2fsck is recommended",
 749                      EXT4_DYNAMIC_REV);
 750
 751         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 752         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 753         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 754         /* leave es->s_feature_*compat flags alone */
 755         /* es->s_uuid will be set by e2fsck if empty */
 756
 757         /*
 758          * The rest of the superblock fields should be zero, and if not it
 759          * means they are likely already in use, so leave them alone.  We
 760          * can leave it up to e2fsck to clean up any inconsistencies there.
 761          */
 762 }
 763
 764 /*
 765  * Open the external journal device
 766  */
 767 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 768 {
 769         struct block_device *bdev;
 770         char b[BDEVNAME_SIZE];
 771
 772         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 773         if (IS_ERR(bdev))
 774                 goto fail;
 775         return bdev;
 776
 777 fail:
 778         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 779                         __bdevname(dev, b), PTR_ERR(bdev));
 780         return NULL;
 781 }
 782
 783 /*
 784  * Release the journal device
 785  */
 786 static int ext4_blkdev_put(struct block_device *bdev)
 787 {
 788         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 789 }
 790
 791 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 792 {
 793         struct block_device *bdev;
 794         int ret = -ENODEV;
 795
 796         bdev = sbi->journal_bdev;
 797         if (bdev) {
 798                 ret = ext4_blkdev_put(bdev);
 799                 sbi->journal_bdev = NULL;
 800         }
 801         return ret;
 802 }
 803
 804 static inline struct inode *orphan_list_entry(struct list_head *l)
 805 {
 806         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 807 }
 808
 809 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 810 {
 811         struct list_head *l;
 812
 813         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 814                  le32_to_cpu(sbi->s_es->s_last_orphan));
 815
 816         printk(KERN_ERR "sb_info orphan list:\n");
 817         list_for_each(l, &sbi->s_orphan) {
 818                 struct inode *inode = orphan_list_entry(l);
 819                 printk(KERN_ERR "  "
 820                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 821                        inode->i_sb->s_id, inode->i_ino, inode,
 822                        inode->i_mode, inode->i_nlink,
 823                        NEXT_ORPHAN(inode));
 824         }
 825 }
 826
 827 static void ext4_put_super(struct super_block *sb)
 828 {
 829         struct ext4_sb_info *sbi = EXT4_SB(sb);
 830         struct ext4_super_block *es = sbi->s_es;
 831         int i, err;
 832
 833         ext4_unregister_li_request(sb);
 834         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 835
 836         flush_workqueue(sbi->dio_unwritten_wq);
 837         destroy_workqueue(sbi->dio_unwritten_wq);
 838
 839         lock_super(sb);
 840         if (sbi->s_journal) {
 841                 err = jbd2_journal_destroy(sbi->s_journal);
 842                 sbi->s_journal = NULL;
 843                 if (err < 0)
 844                         ext4_abort(sb, "Couldn't clean up the journal");
 845         }
 846
 847         del_timer(&sbi->s_err_report);
 848         ext4_release_system_zone(sb);
 849         ext4_mb_release(sb);
 850         ext4_ext_release(sb);
 851         ext4_xattr_put_super(sb);
 852
 853         if (!(sb->s_flags & MS_RDONLY)) {
 854                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 855                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 856         }
 857         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
 858                 ext4_commit_super(sb, 1);
 859
 860         if (sbi->s_proc) {
 861                 remove_proc_entry("options", sbi->s_proc);
 862                 remove_proc_entry(sb->s_id, ext4_proc_root);
 863         }
 864         kobject_del(&sbi->s_kobj);
 865
 866         for (i = 0; i < sbi->s_gdb_count; i++)
 867                 brelse(sbi->s_group_desc[i]);
 868         ext4_kvfree(sbi->s_group_desc);
 869         ext4_kvfree(sbi->s_flex_groups);
 870         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 871         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 872         percpu_counter_destroy(&sbi->s_dirs_counter);
 873         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 874         brelse(sbi->s_sbh);
 875 #ifdef CONFIG_QUOTA
 876         for (i = 0; i < MAXQUOTAS; i++)
 877                 kfree(sbi->s_qf_names[i]);
 878 #endif
 879
 880         /* Debugging code just in case the in-memory inode orphan list
 881          * isn't empty.  The on-disk one can be non-empty if we've
 882          * detected an error and taken the fs readonly, but the
 883          * in-memory list had better be clean by this point. */
 884         if (!list_empty(&sbi->s_orphan))
 885                 dump_orphan_list(sb, sbi);
 886         J_ASSERT(list_empty(&sbi->s_orphan));
 887
 888         invalidate_bdev(sb->s_bdev);
 889         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 890                 /*
 891                  * Invalidate the journal device's buffers.  We don't want them
 892                  * floating about in memory - the physical journal device may
 893                  * hotswapped, and it breaks the `ro-after' testing code.
 894                  */
 895                 sync_blockdev(sbi->journal_bdev);
 896                 invalidate_bdev(sbi->journal_bdev);
 897                 ext4_blkdev_remove(sbi);
 898         }
 899         if (sbi->s_mmp_tsk)
 900                 kthread_stop(sbi->s_mmp_tsk);
 901         sb->s_fs_info = NULL;
 902         /*
 903          * Now that we are completely done shutting down the
 904          * superblock, we need to actually destroy the kobject.
 905          */
 906         unlock_super(sb);
 907         kobject_put(&sbi->s_kobj);
 908         wait_for_completion(&sbi->s_kobj_unregister);
 909         kfree(sbi->s_blockgroup_lock);
 910         kfree(sbi);
 911 }
 912
 913 static struct kmem_cache *ext4_inode_cachep;
 914
 915 /*
 916  * Called inside transaction, so use GFP_NOFS
 917  */
 918 static struct inode *ext4_alloc_inode(struct super_block *sb)
 919 {
 920         struct ext4_inode_info *ei;
 921
 922         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 923         if (!ei)
 924                 return NULL;
 925
 926         ei->vfs_inode.i_version = 1;
 927         ei->vfs_inode.i_data.writeback_index = 0;
 928         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 929         INIT_LIST_HEAD(&ei->i_prealloc_list);
 930         spin_lock_init(&ei->i_prealloc_lock);
 931         ei->i_reserved_data_blocks = 0;
 932         ei->i_reserved_meta_blocks = 0;
 933         ei->i_allocated_meta_blocks = 0;
 934         ei->i_da_metadata_calc_len = 0;
 935         ei->i_da_metadata_calc_last_lblock = 0;
 936         spin_lock_init(&(ei->i_block_reservation_lock));
 937 #ifdef CONFIG_QUOTA
 938         ei->i_reserved_quota = 0;
 939 #endif
 940         ei->jinode = NULL;
 941         INIT_LIST_HEAD(&ei->i_completed_io_list);
 942         spin_lock_init(&ei->i_completed_io_lock);
 943         ei->cur_aio_dio = NULL;
 944         ei->i_sync_tid = 0;
 945         ei->i_datasync_tid = 0;
 946         atomic_set(&ei->i_ioend_count, 0);
 947         atomic_set(&ei->i_aiodio_unwritten, 0);
 948
 949         return &ei->vfs_inode;
 950 }
 951
 952 static int ext4_drop_inode(struct inode *inode)
 953 {
 954         int drop = generic_drop_inode(inode);
 955
 956         trace_ext4_drop_inode(inode, drop);
 957         return drop;
 958 }
 959
 960 static void ext4_i_callback(struct rcu_head *head)
 961 {
 962         struct inode *inode = container_of(head, struct inode, i_rcu);
 963         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 964 }
 965
 966 static void ext4_destroy_inode(struct inode *inode)
 967 {
 968         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 969                 ext4_msg(inode->i_sb, KERN_ERR,
 970                          "Inode %lu (%p): orphan list check failed!",
 971                          inode->i_ino, EXT4_I(inode));
 972                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 973                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 974                                 true);
 975                 dump_stack();
 976         }
 977         call_rcu(&inode->i_rcu, ext4_i_callback);
 978 }
 979
 980 static void init_once(void *foo)
 981 {
 982         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 983
 984         INIT_LIST_HEAD(&ei->i_orphan);
 985 #ifdef CONFIG_EXT4_FS_XATTR
 986         init_rwsem(&ei->xattr_sem);
 987 #endif
 988         init_rwsem(&ei->i_data_sem);
 989         inode_init_once(&ei->vfs_inode);
 990 }
 991
 992 static int init_inodecache(void)
 993 {
 994         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 995                                              sizeof(struct ext4_inode_info),
 996                                              0, (SLAB_RECLAIM_ACCOUNT|
 997                                                 SLAB_MEM_SPREAD),
 998                                              init_once);
 999         if (ext4_inode_cachep == NULL)
1000                 return -ENOMEM;
1001         return 0;
1002 }
1003
1004 static void destroy_inodecache(void)
1005 {
1006         kmem_cache_destroy(ext4_inode_cachep);
1007 }
1008
1009 void ext4_clear_inode(struct inode *inode)
1010 {
1011         invalidate_inode_buffers(inode);
1012         end_writeback(inode);
1013         dquot_drop(inode);
1014         ext4_discard_preallocations(inode);
1015         if (EXT4_I(inode)->jinode) {
1016                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1017                                                EXT4_I(inode)->jinode);
1018                 jbd2_free_inode(EXT4_I(inode)->jinode);
1019                 EXT4_I(inode)->jinode = NULL;
1020         }
1021 }
1022
1023 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1024                                         u64 ino, u32 generation)
1025 {
1026         struct inode *inode;
1027
1028         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1029                 return ERR_PTR(-ESTALE);
1030         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1031                 return ERR_PTR(-ESTALE);
1032
1033         /* iget isn't really right if the inode is currently unallocated!!
1034          *
1035          * ext4_read_inode will return a bad_inode if the inode had been
1036          * deleted, so we should be safe.
1037          *
1038          * Currently we don't know the generation for parent directory, so
1039          * a generation of 0 means "accept any"
1040          */
1041         inode = ext4_iget(sb, ino);
1042         if (IS_ERR(inode))
1043                 return ERR_CAST(inode);
1044         if (generation && inode->i_generation != generation) {
1045                 iput(inode);
1046                 return ERR_PTR(-ESTALE);
1047         }
1048
1049         return inode;
1050 }
1051
1052 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1053                                         int fh_len, int fh_type)
1054 {
1055         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1056                                     ext4_nfs_get_inode);
1057 }
1058
1059 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1060                                         int fh_len, int fh_type)
1061 {
1062         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1063                                     ext4_nfs_get_inode);
1064 }
1065
1066 /*
1067  * Try to release metadata pages (indirect blocks, directories) which are
1068  * mapped via the block device.  Since these pages could have journal heads
1069  * which would prevent try_to_free_buffers() from freeing them, we must use
1070  * jbd2 layer's try_to_free_buffers() function to release them.
1071  */
1072 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1073                                  gfp_t wait)
1074 {
1075         journal_t *journal = EXT4_SB(sb)->s_journal;
1076
1077         WARN_ON(PageChecked(page));
1078         if (!page_has_buffers(page))
1079                 return 0;
1080         if (journal)
1081                 return jbd2_journal_try_to_free_buffers(journal, page,
1082                                                         wait & ~__GFP_WAIT);
1083         return try_to_free_buffers(page);
1084 }
1085
1086 #ifdef CONFIG_QUOTA
1087 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1088 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1089
1090 static int ext4_write_dquot(struct dquot *dquot);
1091 static int ext4_acquire_dquot(struct dquot *dquot);
1092 static int ext4_release_dquot(struct dquot *dquot);
1093 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1094 static int ext4_write_info(struct super_block *sb, int type);
1095 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1096                          struct path *path);
1097 static int ext4_quota_off(struct super_block *sb, int type);
1098 static int ext4_quota_on_mount(struct super_block *sb, int type);
1099 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1100                                size_t len, loff_t off);
1101 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1102                                 const char *data, size_t len, loff_t off);
1103
1104 static const struct dquot_operations ext4_quota_operations = {
1105         .get_reserved_space = ext4_get_reserved_space,
1106         .write_dquot    = ext4_write_dquot,
1107         .acquire_dquot  = ext4_acquire_dquot,
1108         .release_dquot  = ext4_release_dquot,
1109         .mark_dirty     = ext4_mark_dquot_dirty,
1110         .write_info     = ext4_write_info,
1111         .alloc_dquot    = dquot_alloc,
1112         .destroy_dquot  = dquot_destroy,
1113 };
1114
1115 static const struct quotactl_ops ext4_qctl_operations = {
1116         .quota_on       = ext4_quota_on,
1117         .quota_off      = ext4_quota_off,
1118         .quota_sync     = dquot_quota_sync,
1119         .get_info       = dquot_get_dqinfo,
1120         .set_info       = dquot_set_dqinfo,
1121         .get_dqblk      = dquot_get_dqblk,
1122         .set_dqblk      = dquot_set_dqblk
1123 };
1124 #endif
1125
1126 static const struct super_operations ext4_sops = {
1127         .alloc_inode    = ext4_alloc_inode,
1128         .destroy_inode  = ext4_destroy_inode,
1129         .write_inode    = ext4_write_inode,
1130         .dirty_inode    = ext4_dirty_inode,
1131         .drop_inode     = ext4_drop_inode,
1132         .evict_inode    = ext4_evict_inode,
1133         .put_super      = ext4_put_super,
1134         .sync_fs        = ext4_sync_fs,
1135         .freeze_fs      = ext4_freeze,
1136         .unfreeze_fs    = ext4_unfreeze,
1137         .statfs         = ext4_statfs,
1138         .remount_fs     = ext4_remount,
1139         .show_options   = ext4_show_options,
1140 #ifdef CONFIG_QUOTA
1141         .quota_read     = ext4_quota_read,
1142         .quota_write    = ext4_quota_write,
1143 #endif
1144         .bdev_try_to_free_page = bdev_try_to_free_page,
1145 };
1146
1147 static const struct super_operations ext4_nojournal_sops = {
1148         .alloc_inode    = ext4_alloc_inode,
1149         .destroy_inode  = ext4_destroy_inode,
1150         .write_inode    = ext4_write_inode,
1151         .dirty_inode    = ext4_dirty_inode,
1152         .drop_inode     = ext4_drop_inode,
1153         .evict_inode    = ext4_evict_inode,
1154         .write_super    = ext4_write_super,
1155         .put_super      = ext4_put_super,
1156         .statfs         = ext4_statfs,
1157         .remount_fs     = ext4_remount,
1158         .show_options   = ext4_show_options,
1159 #ifdef CONFIG_QUOTA
1160         .quota_read     = ext4_quota_read,
1161         .quota_write    = ext4_quota_write,
1162 #endif
1163         .bdev_try_to_free_page = bdev_try_to_free_page,
1164 };
1165
1166 static const struct export_operations ext4_export_ops = {
1167         .fh_to_dentry = ext4_fh_to_dentry,
1168         .fh_to_parent = ext4_fh_to_parent,
1169         .get_parent = ext4_get_parent,
1170 };
1171
1172 enum {
1173         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1174         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1175         Opt_nouid32, Opt_debug, Opt_removed,
1176         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1177         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1178         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1179         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1180         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1181         Opt_data_err_abort, Opt_data_err_ignore,
1182         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1183         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1184         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1185         Opt_usrquota, Opt_grpquota, Opt_i_version,
1186         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1187         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1188         Opt_inode_readahead_blks, Opt_journal_ioprio,
1189         Opt_dioread_nolock, Opt_dioread_lock,
1190         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1191 };
1192
1193 static const match_table_t tokens = {
1194         {Opt_bsd_df, "bsddf"},
1195         {Opt_minix_df, "minixdf"},
1196         {Opt_grpid, "grpid"},
1197         {Opt_grpid, "bsdgroups"},
1198         {Opt_nogrpid, "nogrpid"},
1199         {Opt_nogrpid, "sysvgroups"},
1200         {Opt_resgid, "resgid=%u"},
1201         {Opt_resuid, "resuid=%u"},
1202         {Opt_sb, "sb=%u"},
1203         {Opt_err_cont, "errors=continue"},
1204         {Opt_err_panic, "errors=panic"},
1205         {Opt_err_ro, "errors=remount-ro"},
1206         {Opt_nouid32, "nouid32"},
1207         {Opt_debug, "debug"},
1208         {Opt_removed, "oldalloc"},
1209         {Opt_removed, "orlov"},
1210         {Opt_user_xattr, "user_xattr"},
1211         {Opt_nouser_xattr, "nouser_xattr"},
1212         {Opt_acl, "acl"},
1213         {Opt_noacl, "noacl"},
1214         {Opt_noload, "norecovery"},
1215         {Opt_noload, "noload"},
1216         {Opt_removed, "nobh"},
1217         {Opt_removed, "bh"},
1218         {Opt_commit, "commit=%u"},
1219         {Opt_min_batch_time, "min_batch_time=%u"},
1220         {Opt_max_batch_time, "max_batch_time=%u"},
1221         {Opt_journal_dev, "journal_dev=%u"},
1222         {Opt_journal_checksum, "journal_checksum"},
1223         {Opt_journal_async_commit, "journal_async_commit"},
1224         {Opt_abort, "abort"},
1225         {Opt_data_journal, "data=journal"},
1226         {Opt_data_ordered, "data=ordered"},
1227         {Opt_data_writeback, "data=writeback"},
1228         {Opt_data_err_abort, "data_err=abort"},
1229         {Opt_data_err_ignore, "data_err=ignore"},
1230         {Opt_offusrjquota, "usrjquota="},
1231         {Opt_usrjquota, "usrjquota=%s"},
1232         {Opt_offgrpjquota, "grpjquota="},
1233         {Opt_grpjquota, "grpjquota=%s"},
1234         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1235         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1236         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1237         {Opt_grpquota, "grpquota"},
1238         {Opt_noquota, "noquota"},
1239         {Opt_quota, "quota"},
1240         {Opt_usrquota, "usrquota"},
1241         {Opt_barrier, "barrier=%u"},
1242         {Opt_barrier, "barrier"},
1243         {Opt_nobarrier, "nobarrier"},
1244         {Opt_i_version, "i_version"},
1245         {Opt_stripe, "stripe=%u"},
1246         {Opt_delalloc, "delalloc"},
1247         {Opt_nodelalloc, "nodelalloc"},
1248         {Opt_mblk_io_submit, "mblk_io_submit"},
1249         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1250         {Opt_block_validity, "block_validity"},
1251         {Opt_noblock_validity, "noblock_validity"},
1252         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1253         {Opt_journal_ioprio, "journal_ioprio=%u"},
1254         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1255         {Opt_auto_da_alloc, "auto_da_alloc"},
1256         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1257         {Opt_dioread_nolock, "dioread_nolock"},
1258         {Opt_dioread_lock, "dioread_lock"},
1259         {Opt_discard, "discard"},
1260         {Opt_nodiscard, "nodiscard"},
1261         {Opt_init_itable, "init_itable=%u"},
1262         {Opt_init_itable, "init_itable"},
1263         {Opt_noinit_itable, "noinit_itable"},
1264         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1265         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1266         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1267         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1268         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1269         {Opt_err, NULL},
1270 };
1271
1272 static ext4_fsblk_t get_sb_block(void **data)
1273 {
1274         ext4_fsblk_t    sb_block;
1275         char            *options = (char *) *data;
1276
1277         if (!options || strncmp(options, "sb=", 3) != 0)
1278                 return 1;       /* Default location */
1279
1280         options += 3;
1281         /* TODO: use simple_strtoll with >32bit ext4 */
1282         sb_block = simple_strtoul(options, &options, 0);
1283         if (*options && *options != ',') {
1284                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1285                        (char *) *data);
1286                 return 1;
1287         }
1288         if (*options == ',')
1289                 options++;
1290         *data = (void *) options;
1291
1292         return sb_block;
1293 }
1294
1295 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1296 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1297         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1298
1299 #ifdef CONFIG_QUOTA
1300 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1301 {
1302         struct ext4_sb_info *sbi = EXT4_SB(sb);
1303         char *qname;
1304
1305         if (sb_any_quota_loaded(sb) &&
1306                 !sbi->s_qf_names[qtype]) {
1307                 ext4_msg(sb, KERN_ERR,
1308                         "Cannot change journaled "
1309                         "quota options when quota turned on");
1310                 return -1;
1311         }
1312         qname = match_strdup(args);
1313         if (!qname) {
1314                 ext4_msg(sb, KERN_ERR,
1315                         "Not enough memory for storing quotafile name");
1316                 return -1;
1317         }
1318         if (sbi->s_qf_names[qtype] &&
1319                 strcmp(sbi->s_qf_names[qtype], qname)) {
1320                 ext4_msg(sb, KERN_ERR,
1321                         "%s quota file already specified", QTYPE2NAME(qtype));
1322                 kfree(qname);
1323                 return -1;
1324         }
1325         sbi->s_qf_names[qtype] = qname;
1326         if (strchr(sbi->s_qf_names[qtype], '/')) {
1327                 ext4_msg(sb, KERN_ERR,
1328                         "quotafile must be on filesystem root");
1329                 kfree(sbi->s_qf_names[qtype]);
1330                 sbi->s_qf_names[qtype] = NULL;
1331                 return -1;
1332         }
1333         set_opt(sb, QUOTA);
1334         return 1;
1335 }
1336
1337 static int clear_qf_name(struct super_block *sb, int qtype)
1338 {
1339
1340         struct ext4_sb_info *sbi = EXT4_SB(sb);
1341
1342         if (sb_any_quota_loaded(sb) &&
1343                 sbi->s_qf_names[qtype]) {
1344                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1345                         " when quota turned on");
1346                 return -1;
1347         }
1348         /*
1349          * The space will be released later when all options are confirmed
1350          * to be correct
1351          */
1352         sbi->s_qf_names[qtype] = NULL;
1353         return 1;
1354 }
1355 #endif
1356
1357 #define MOPT_SET        0x0001
1358 #define MOPT_CLEAR      0x0002
1359 #define MOPT_NOSUPPORT  0x0004
1360 #define MOPT_EXPLICIT   0x0008
1361 #define MOPT_CLEAR_ERR  0x0010
1362 #define MOPT_GTE0       0x0020
1363 #ifdef CONFIG_QUOTA
1364 #define MOPT_Q          0
1365 #define MOPT_QFMT       0x0040
1366 #else
1367 #define MOPT_Q          MOPT_NOSUPPORT
1368 #define MOPT_QFMT       MOPT_NOSUPPORT
1369 #endif
1370 #define MOPT_DATAJ      0x0080
1371
1372 static const struct mount_opts {
1373         int     token;
1374         int     mount_opt;
1375         int     flags;
1376 } ext4_mount_opts[] = {
1377         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1378         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1379         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1380         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1381         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1382         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1383         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1384         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1385         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1386         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1387         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1388         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1389         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1390         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1391         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1392         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1393                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1394         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1395         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1397         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1398         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1399         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1400         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1401         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1402         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1403         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1404         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1405         {Opt_commit, 0, MOPT_GTE0},
1406         {Opt_max_batch_time, 0, MOPT_GTE0},
1407         {Opt_min_batch_time, 0, MOPT_GTE0},
1408         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1409         {Opt_init_itable, 0, MOPT_GTE0},
1410         {Opt_stripe, 0, MOPT_GTE0},
1411         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1412         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1413         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1414 #ifdef CONFIG_EXT4_FS_XATTR
1415         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1416         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1417 #else
1418         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1419         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1420 #endif
1421 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1422         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1423         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1424 #else
1425         {Opt_acl, 0, MOPT_NOSUPPORT},
1426         {Opt_noacl, 0, MOPT_NOSUPPORT},
1427 #endif
1428         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1429         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1430         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1431         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1432                                                         MOPT_SET | MOPT_Q},
1433         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1434                                                         MOPT_SET | MOPT_Q},
1435         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1436                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1437         {Opt_usrjquota, 0, MOPT_Q},
1438         {Opt_grpjquota, 0, MOPT_Q},
1439         {Opt_offusrjquota, 0, MOPT_Q},
1440         {Opt_offgrpjquota, 0, MOPT_Q},
1441         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1442         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1443         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1444         {Opt_err, 0, 0}
1445 };
1446
1447 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1448                             substring_t *args, unsigned long *journal_devnum,
1449                             unsigned int *journal_ioprio, int is_remount)
1450 {
1451         struct ext4_sb_info *sbi = EXT4_SB(sb);
1452         const struct mount_opts *m;
1453         int arg = 0;
1454
1455 #ifdef CONFIG_QUOTA
1456         if (token == Opt_usrjquota)
1457                 return set_qf_name(sb, USRQUOTA, &args[0]);
1458         else if (token == Opt_grpjquota)
1459                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1460         else if (token == Opt_offusrjquota)
1461                 return clear_qf_name(sb, USRQUOTA);
1462         else if (token == Opt_offgrpjquota)
1463                 return clear_qf_name(sb, GRPQUOTA);
1464 #endif
1465         if (args->from && match_int(args, &arg))
1466                 return -1;
1467         switch (token) {
1468         case Opt_noacl:
1469         case Opt_nouser_xattr:
1470                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1471                 break;
1472         case Opt_sb:
1473                 return 1;       /* handled by get_sb_block() */
1474         case Opt_removed:
1475                 ext4_msg(sb, KERN_WARNING,
1476                          "Ignoring removed %s option", opt);
1477                 return 1;
1478         case Opt_resuid:
1479                 sbi->s_resuid = arg;
1480                 return 1;
1481         case Opt_resgid:
1482                 sbi->s_resgid = arg;
1483                 return 1;
1484         case Opt_abort:
1485                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1486                 return 1;
1487         case Opt_i_version:
1488                 sb->s_flags |= MS_I_VERSION;
1489                 return 1;
1490         case Opt_journal_dev:
1491                 if (is_remount) {
1492                         ext4_msg(sb, KERN_ERR,
1493                                  "Cannot specify journal on remount");
1494                         return -1;
1495                 }
1496                 *journal_devnum = arg;
1497                 return 1;
1498         case Opt_journal_ioprio:
1499                 if (arg < 0 || arg > 7)
1500                         return -1;
1501                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1502                 return 1;
1503         }
1504
1505         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1506                 if (token != m->token)
1507                         continue;
1508                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1509                         return -1;
1510                 if (m->flags & MOPT_EXPLICIT)
1511                         set_opt2(sb, EXPLICIT_DELALLOC);
1512                 if (m->flags & MOPT_CLEAR_ERR)
1513                         clear_opt(sb, ERRORS_MASK);
1514                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1515                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1516                                  "options when quota turned on");
1517                         return -1;
1518                 }
1519
1520                 if (m->flags & MOPT_NOSUPPORT) {
1521                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1522                 } else if (token == Opt_commit) {
1523                         if (arg == 0)
1524                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1525                         sbi->s_commit_interval = HZ * arg;
1526                 } else if (token == Opt_max_batch_time) {
1527                         if (arg == 0)
1528                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1529                         sbi->s_max_batch_time = arg;
1530                 } else if (token == Opt_min_batch_time) {
1531                         sbi->s_min_batch_time = arg;
1532                 } else if (token == Opt_inode_readahead_blks) {
1533                         if (arg > (1 << 30))
1534                                 return -1;
1535                         if (arg && !is_power_of_2(arg)) {
1536                                 ext4_msg(sb, KERN_ERR,
1537                                          "EXT4-fs: inode_readahead_blks"
1538                                          " must be a power of 2");
1539                                 return -1;
1540                         }
1541                         sbi->s_inode_readahead_blks = arg;
1542                 } else if (token == Opt_init_itable) {
1543                         set_opt(sb, INIT_INODE_TABLE);
1544                         if (!args->from)
1545                                 arg = EXT4_DEF_LI_WAIT_MULT;
1546                         sbi->s_li_wait_mult = arg;
1547                 } else if (token == Opt_stripe) {
1548                         sbi->s_stripe = arg;
1549                 } else if (m->flags & MOPT_DATAJ) {
1550                         if (is_remount) {
1551                                 if (!sbi->s_journal)
1552                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1553                                 else if (test_opt(sb, DATA_FLAGS) !=
1554                                          m->mount_opt) {
1555                                         ext4_msg(sb, KERN_ERR,
1556                                          "Cannot change data mode on remount");
1557                                         return -1;
1558                                 }
1559                         } else {
1560                                 clear_opt(sb, DATA_FLAGS);
1561                                 sbi->s_mount_opt |= m->mount_opt;
1562                         }
1563 #ifdef CONFIG_QUOTA
1564                 } else if (m->flags & MOPT_QFMT) {
1565                         if (sb_any_quota_loaded(sb) &&
1566                             sbi->s_jquota_fmt != m->mount_opt) {
1567                                 ext4_msg(sb, KERN_ERR, "Cannot "
1568                                          "change journaled quota options "
1569                                          "when quota turned on");
1570                                 return -1;
1571                         }
1572                         sbi->s_jquota_fmt = m->mount_opt;
1573 #endif
1574                 } else {
1575                         if (!args->from)
1576                                 arg = 1;
1577                         if (m->flags & MOPT_CLEAR)
1578                                 arg = !arg;
1579                         else if (unlikely(!(m->flags & MOPT_SET))) {
1580                                 ext4_msg(sb, KERN_WARNING,
1581                                          "buggy handling of option %s", opt);
1582                                 WARN_ON(1);
1583                                 return -1;
1584                         }
1585                         if (arg != 0)
1586                                 sbi->s_mount_opt |= m->mount_opt;
1587                         else
1588                                 sbi->s_mount_opt &= ~m->mount_opt;
1589                 }
1590                 return 1;
1591         }
1592         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1593                  "or missing value", opt);
1594         return -1;
1595 }
1596
1597 static int parse_options(char *options, struct super_block *sb,
1598                          unsigned long *journal_devnum,
1599                          unsigned int *journal_ioprio,
1600                          int is_remount)
1601 {
1602         struct ext4_sb_info *sbi = EXT4_SB(sb);
1603         char *p;
1604         substring_t args[MAX_OPT_ARGS];
1605         int token;
1606
1607         if (!options)
1608                 return 1;
1609
1610         while ((p = strsep(&options, ",")) != NULL) {
1611                 if (!*p)
1612                         continue;
1613                 /*
1614                  * Initialize args struct so we know whether arg was
1615                  * found; some options take optional arguments.
1616                  */
1617                 args[0].to = args[0].from = 0;
1618                 token = match_token(p, tokens, args);
1619                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1620                                      journal_ioprio, is_remount) < 0)
1621                         return 0;
1622         }
1623 #ifdef CONFIG_QUOTA
1624         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1625                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1626                         clear_opt(sb, USRQUOTA);
1627
1628                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1629                         clear_opt(sb, GRPQUOTA);
1630
1631                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1632                         ext4_msg(sb, KERN_ERR, "old and new quota "
1633                                         "format mixing");
1634                         return 0;
1635                 }
1636
1637                 if (!sbi->s_jquota_fmt) {
1638                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1639                                         "not specified");
1640                         return 0;
1641                 }
1642         } else {
1643                 if (sbi->s_jquota_fmt) {
1644                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1645                                         "specified with no journaling "
1646                                         "enabled");
1647                         return 0;
1648                 }
1649         }
1650 #endif
1651         if (test_opt(sb, DIOREAD_NOLOCK)) {
1652                 int blocksize =
1653                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1654
1655                 if (blocksize < PAGE_CACHE_SIZE) {
1656                         ext4_msg(sb, KERN_ERR, "can't mount with "
1657                                  "dioread_nolock if block size != PAGE_SIZE");
1658                         return 0;
1659                 }
1660         }
1661         return 1;
1662 }
1663
1664 static inline void ext4_show_quota_options(struct seq_file *seq,
1665                                            struct super_block *sb)
1666 {
1667 #if defined(CONFIG_QUOTA)
1668         struct ext4_sb_info *sbi = EXT4_SB(sb);
1669
1670         if (sbi->s_jquota_fmt) {
1671                 char *fmtname = "";
1672
1673                 switch (sbi->s_jquota_fmt) {
1674                 case QFMT_VFS_OLD:
1675                         fmtname = "vfsold";
1676                         break;
1677                 case QFMT_VFS_V0:
1678                         fmtname = "vfsv0";
1679                         break;
1680                 case QFMT_VFS_V1:
1681                         fmtname = "vfsv1";
1682                         break;
1683                 }
1684                 seq_printf(seq, ",jqfmt=%s", fmtname);
1685         }
1686
1687         if (sbi->s_qf_names[USRQUOTA])
1688                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1689
1690         if (sbi->s_qf_names[GRPQUOTA])
1691                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1692
1693         if (test_opt(sb, USRQUOTA))
1694                 seq_puts(seq, ",usrquota");
1695
1696         if (test_opt(sb, GRPQUOTA))
1697                 seq_puts(seq, ",grpquota");
1698 #endif
1699 }
1700
1701 static const char *token2str(int token)
1702 {
1703         const struct match_token *t;
1704
1705         for (t = tokens; t->token != Opt_err; t++)
1706                 if (t->token == token && !strchr(t->pattern, '='))
1707                         break;
1708         return t->pattern;
1709 }
1710
1711 /*
1712  * Show an option if
1713  *  - it's set to a non-default value OR
1714  *  - if the per-sb default is different from the global default
1715  */
1716 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1717                               int nodefs)
1718 {
1719         struct ext4_sb_info *sbi = EXT4_SB(sb);
1720         struct ext4_super_block *es = sbi->s_es;
1721         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1722         const struct mount_opts *m;
1723         char sep = nodefs ? '\n' : ',';
1724
1725 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1726 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1727
1728         if (sbi->s_sb_block != 1)
1729                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1730
1731         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1732                 int want_set = m->flags & MOPT_SET;
1733                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1734                     (m->flags & MOPT_CLEAR_ERR))
1735                         continue;
1736                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1737                         continue; /* skip if same as the default */
1738                 if ((want_set &&
1739                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1740                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1741                         continue; /* select Opt_noFoo vs Opt_Foo */
1742                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1743         }
1744
1745         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1746             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1747                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1748         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1749             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1750                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1751         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1752         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1753                 SEQ_OPTS_PUTS("errors=remount-ro");
1754         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1755                 SEQ_OPTS_PUTS("errors=continue");
1756         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1757                 SEQ_OPTS_PUTS("errors=panic");
1758         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1759                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1760         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1761                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1762         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1763                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1764         if (sb->s_flags & MS_I_VERSION)
1765                 SEQ_OPTS_PUTS("i_version");
1766         if (nodefs || sbi->s_stripe)
1767                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1768         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1769                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1770                         SEQ_OPTS_PUTS("data=journal");
1771                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1772                         SEQ_OPTS_PUTS("data=ordered");
1773                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1774                         SEQ_OPTS_PUTS("data=writeback");
1775         }
1776         if (nodefs ||
1777             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1778                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1779                                sbi->s_inode_readahead_blks);
1780
1781         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1782                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1783                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1784
1785         ext4_show_quota_options(seq, sb);
1786         return 0;
1787 }
1788
1789 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1790 {
1791         return _ext4_show_options(seq, root->d_sb, 0);
1792 }
1793
1794 static int options_seq_show(struct seq_file *seq, void *offset)
1795 {
1796         struct super_block *sb = seq->private;
1797         int rc;
1798
1799         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1800         rc = _ext4_show_options(seq, sb, 1);
1801         seq_puts(seq, "\n");
1802         return rc;
1803 }
1804
1805 static int options_open_fs(struct inode *inode, struct file *file)
1806 {
1807         return single_open(file, options_seq_show, PDE(inode)->data);
1808 }
1809
1810 static const struct file_operations ext4_seq_options_fops = {
1811         .owner = THIS_MODULE,
1812         .open = options_open_fs,
1813         .read = seq_read,
1814         .llseek = seq_lseek,
1815         .release = single_release,
1816 };
1817
1818 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1819                             int read_only)
1820 {
1821         struct ext4_sb_info *sbi = EXT4_SB(sb);
1822         int res = 0;
1823
1824         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1825                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1826                          "forcing read-only mode");
1827                 res = MS_RDONLY;
1828         }
1829         if (read_only)
1830                 goto done;
1831         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1832                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1833                          "running e2fsck is recommended");
1834         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1835                 ext4_msg(sb, KERN_WARNING,
1836                          "warning: mounting fs with errors, "
1837                          "running e2fsck is recommended");
1838         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1839                  le16_to_cpu(es->s_mnt_count) >=
1840                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1841                 ext4_msg(sb, KERN_WARNING,
1842                          "warning: maximal mount count reached, "
1843                          "running e2fsck is recommended");
1844         else if (le32_to_cpu(es->s_checkinterval) &&
1845                 (le32_to_cpu(es->s_lastcheck) +
1846                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1847                 ext4_msg(sb, KERN_WARNING,
1848                          "warning: checktime reached, "
1849                          "running e2fsck is recommended");
1850         if (!sbi->s_journal)
1851                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1852         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1853                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1854         le16_add_cpu(&es->s_mnt_count, 1);
1855         es->s_mtime = cpu_to_le32(get_seconds());
1856         ext4_update_dynamic_rev(sb);
1857         if (sbi->s_journal)
1858                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1859
1860         ext4_commit_super(sb, 1);
1861 done:
1862         if (test_opt(sb, DEBUG))
1863                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1864                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1865                         sb->s_blocksize,
1866                         sbi->s_groups_count,
1867                         EXT4_BLOCKS_PER_GROUP(sb),
1868                         EXT4_INODES_PER_GROUP(sb),
1869                         sbi->s_mount_opt, sbi->s_mount_opt2);
1870
1871         cleancache_init_fs(sb);
1872         return res;
1873 }
1874
1875 static int ext4_fill_flex_info(struct super_block *sb)
1876 {
1877         struct ext4_sb_info *sbi = EXT4_SB(sb);
1878         struct ext4_group_desc *gdp = NULL;
1879         ext4_group_t flex_group_count;
1880         ext4_group_t flex_group;
1881         unsigned int groups_per_flex = 0;
1882         size_t size;
1883         int i;
1884
1885         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1886         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1887                 sbi->s_log_groups_per_flex = 0;
1888                 return 1;
1889         }
1890         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1891
1892         /* We allocate both existing and potentially added groups */
1893         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1894                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1895                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1896         size = flex_group_count * sizeof(struct flex_groups);
1897         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1898         if (sbi->s_flex_groups == NULL) {
1899                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1900                          flex_group_count);
1901                 goto failed;
1902         }
1903
1904         for (i = 0; i < sbi->s_groups_count; i++) {
1905                 gdp = ext4_get_group_desc(sb, i, NULL);
1906
1907                 flex_group = ext4_flex_group(sbi, i);
1908                 atomic_add(ext4_free_inodes_count(sb, gdp),
1909                            &sbi->s_flex_groups[flex_group].free_inodes);
1910                 atomic_add(ext4_free_group_clusters(sb, gdp),
1911                            &sbi->s_flex_groups[flex_group].free_clusters);
1912                 atomic_add(ext4_used_dirs_count(sb, gdp),
1913                            &sbi->s_flex_groups[flex_group].used_dirs);
1914         }
1915
1916         return 1;
1917 failed:
1918         return 0;
1919 }
1920
1921 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1922                             struct ext4_group_desc *gdp)
1923 {
1924         __u16 crc = 0;
1925
1926         if (sbi->s_es->s_feature_ro_compat &
1927             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1928                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1929                 __le32 le_group = cpu_to_le32(block_group);
1930
1931                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1932                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1933                 crc = crc16(crc, (__u8 *)gdp, offset);
1934                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1935                 /* for checksum of struct ext4_group_desc do the rest...*/
1936                 if ((sbi->s_es->s_feature_incompat &
1937                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1938                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1939                         crc = crc16(crc, (__u8 *)gdp + offset,
1940                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1941                                         offset);
1942         }
1943
1944         return cpu_to_le16(crc);
1945 }
1946
1947 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1948                                 struct ext4_group_desc *gdp)
1949 {
1950         if ((sbi->s_es->s_feature_ro_compat &
1951              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1952             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1953                 return 0;
1954
1955         return 1;
1956 }
1957
1958 /* Called at mount-time, super-block is locked */
1959 static int ext4_check_descriptors(struct super_block *sb,
1960                                   ext4_group_t *first_not_zeroed)
1961 {
1962         struct ext4_sb_info *sbi = EXT4_SB(sb);
1963         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1964         ext4_fsblk_t last_block;
1965         ext4_fsblk_t block_bitmap;
1966         ext4_fsblk_t inode_bitmap;
1967         ext4_fsblk_t inode_table;
1968         int flexbg_flag = 0;
1969         ext4_group_t i, grp = sbi->s_groups_count;
1970
1971         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1972                 flexbg_flag = 1;
1973
1974         ext4_debug("Checking group descriptors");
1975
1976         for (i = 0; i < sbi->s_groups_count; i++) {
1977                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1978
1979                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1980                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1981                 else
1982                         last_block = first_block +
1983                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1984
1985                 if ((grp == sbi->s_groups_count) &&
1986                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1987                         grp = i;
1988
1989                 block_bitmap = ext4_block_bitmap(sb, gdp);
1990                 if (block_bitmap < first_block || block_bitmap > last_block) {
1991                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1992                                "Block bitmap for group %u not in group "
1993                                "(block %llu)!", i, block_bitmap);
1994                         return 0;
1995                 }
1996                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1997                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1998                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1999                                "Inode bitmap for group %u not in group "
2000                                "(block %llu)!", i, inode_bitmap);
2001                         return 0;
2002                 }
2003                 inode_table = ext4_inode_table(sb, gdp);
2004                 if (inode_table < first_block ||
2005                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2006                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2007                                "Inode table for group %u not in group "
2008                                "(block %llu)!", i, inode_table);
2009                         return 0;
2010                 }
2011                 ext4_lock_group(sb, i);
2012                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2013                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2014                                  "Checksum for group %u failed (%u!=%u)",
2015                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2016                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2017                         if (!(sb->s_flags & MS_RDONLY)) {
2018                                 ext4_unlock_group(sb, i);
2019                                 return 0;
2020                         }
2021                 }
2022                 ext4_unlock_group(sb, i);
2023                 if (!flexbg_flag)
2024                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2025         }
2026         if (NULL != first_not_zeroed)
2027                 *first_not_zeroed = grp;
2028
2029         ext4_free_blocks_count_set(sbi->s_es,
2030                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2031         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2032         return 1;
2033 }
2034
2035 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2036  * the superblock) which were deleted from all directories, but held open by
2037  * a process at the time of a crash.  We walk the list and try to delete these
2038  * inodes at recovery time (only with a read-write filesystem).
2039  *
2040  * In order to keep the orphan inode chain consistent during traversal (in
2041  * case of crash during recovery), we link each inode into the superblock
2042  * orphan list_head and handle it the same way as an inode deletion during
2043  * normal operation (which journals the operations for us).
2044  *
2045  * We only do an iget() and an iput() on each inode, which is very safe if we
2046  * accidentally point at an in-use or already deleted inode.  The worst that
2047  * can happen in this case is that we get a "bit already cleared" message from
2048  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2049  * e2fsck was run on this filesystem, and it must have already done the orphan
2050  * inode cleanup for us, so we can safely abort without any further action.
2051  */
2052 static void ext4_orphan_cleanup(struct super_block *sb,
2053                                 struct ext4_super_block *es)
2054 {
2055         unsigned int s_flags = sb->s_flags;
2056         int nr_orphans = 0, nr_truncates = 0;
2057 #ifdef CONFIG_QUOTA
2058         int i;
2059 #endif
2060         if (!es->s_last_orphan) {
2061                 jbd_debug(4, "no orphan inodes to clean up\n");
2062                 return;
2063         }
2064
2065         if (bdev_read_only(sb->s_bdev)) {
2066                 ext4_msg(sb, KERN_ERR, "write access "
2067                         "unavailable, skipping orphan cleanup");
2068                 return;
2069         }
2070
2071         /* Check if feature set would not allow a r/w mount */
2072         if (!ext4_feature_set_ok(sb, 0)) {
2073                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2074                          "unknown ROCOMPAT features");
2075                 return;
2076         }
2077
2078         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2079                 if (es->s_last_orphan)
2080                         jbd_debug(1, "Errors on filesystem, "
2081                                   "clearing orphan list.\n");
2082                 es->s_last_orphan = 0;
2083                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2084                 return;
2085         }
2086
2087         if (s_flags & MS_RDONLY) {
2088                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2089                 sb->s_flags &= ~MS_RDONLY;
2090         }
2091 #ifdef CONFIG_QUOTA
2092         /* Needed for iput() to work correctly and not trash data */
2093         sb->s_flags |= MS_ACTIVE;
2094         /* Turn on quotas so that they are updated correctly */
2095         for (i = 0; i < MAXQUOTAS; i++) {
2096                 if (EXT4_SB(sb)->s_qf_names[i]) {
2097                         int ret = ext4_quota_on_mount(sb, i);
2098                         if (ret < 0)
2099                                 ext4_msg(sb, KERN_ERR,
2100                                         "Cannot turn on journaled "
2101                                         "quota: error %d", ret);
2102                 }
2103         }
2104 #endif
2105
2106         while (es->s_last_orphan) {
2107                 struct inode *inode;
2108
2109                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2110                 if (IS_ERR(inode)) {
2111                         es->s_last_orphan = 0;
2112                         break;
2113                 }
2114
2115                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2116                 dquot_initialize(inode);
2117                 if (inode->i_nlink) {
2118                         ext4_msg(sb, KERN_DEBUG,
2119                                 "%s: truncating inode %lu to %lld bytes",
2120                                 __func__, inode->i_ino, inode->i_size);
2121                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2122                                   inode->i_ino, inode->i_size);
2123                         mutex_lock(&inode->i_mutex);
2124                         ext4_truncate(inode);
2125                         mutex_unlock(&inode->i_mutex);
2126                         nr_truncates++;
2127                 } else {
2128                         ext4_msg(sb, KERN_DEBUG,
2129                                 "%s: deleting unreferenced inode %lu",
2130                                 __func__, inode->i_ino);
2131                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2132                                   inode->i_ino);
2133                         nr_orphans++;
2134                 }
2135                 iput(inode);  /* The delete magic happens here! */
2136         }
2137
2138 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2139
2140         if (nr_orphans)
2141                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2142                        PLURAL(nr_orphans));
2143         if (nr_truncates)
2144                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2145                        PLURAL(nr_truncates));
2146 #ifdef CONFIG_QUOTA
2147         /* Turn quotas off */
2148         for (i = 0; i < MAXQUOTAS; i++) {
2149                 if (sb_dqopt(sb)->files[i])
2150                         dquot_quota_off(sb, i);
2151         }
2152 #endif
2153         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2154 }
2155
2156 /*
2157  * Maximal extent format file size.
2158  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2159  * extent format containers, within a sector_t, and within i_blocks
2160  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2161  * so that won't be a limiting factor.
2162  *
2163  * However there is other limiting factor. We do store extents in the form
2164  * of starting block and length, hence the resulting length of the extent
2165  * covering maximum file size must fit into on-disk format containers as
2166  * well. Given that length is always by 1 unit bigger than max unit (because
2167  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2168  *
2169  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2170  */
2171 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2172 {
2173         loff_t res;
2174         loff_t upper_limit = MAX_LFS_FILESIZE;
2175
2176         /* small i_blocks in vfs inode? */
2177         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2178                 /*
2179                  * CONFIG_LBDAF is not enabled implies the inode
2180                  * i_block represent total blocks in 512 bytes
2181                  * 32 == size of vfs inode i_blocks * 8
2182                  */
2183                 upper_limit = (1LL << 32) - 1;
2184
2185                 /* total blocks in file system block size */
2186                 upper_limit >>= (blkbits - 9);
2187                 upper_limit <<= blkbits;
2188         }
2189
2190         /*
2191          * 32-bit extent-start container, ee_block. We lower the maxbytes
2192          * by one fs block, so ee_len can cover the extent of maximum file
2193          * size
2194          */
2195         res = (1LL << 32) - 1;
2196         res <<= blkbits;
2197
2198         /* Sanity check against vm- & vfs- imposed limits */
2199         if (res > upper_limit)
2200                 res = upper_limit;
2201
2202         return res;
2203 }
2204
2205 /*
2206  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2207  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2208  * We need to be 1 filesystem block less than the 2^48 sector limit.
2209  */
2210 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2211 {
2212         loff_t res = EXT4_NDIR_BLOCKS;
2213         int meta_blocks;
2214         loff_t upper_limit;
2215         /* This is calculated to be the largest file size for a dense, block
2216          * mapped file such that the file's total number of 512-byte sectors,
2217          * including data and all indirect blocks, does not exceed (2^48 - 1).
2218          *
2219          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2220          * number of 512-byte sectors of the file.
2221          */
2222
2223         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2224                 /*
2225                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2226                  * the inode i_block field represents total file blocks in
2227                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2228                  */
2229                 upper_limit = (1LL << 32) - 1;
2230
2231                 /* total blocks in file system block size */
2232                 upper_limit >>= (bits - 9);
2233
2234         } else {
2235                 /*
2236                  * We use 48 bit ext4_inode i_blocks
2237                  * With EXT4_HUGE_FILE_FL set the i_blocks
2238                  * represent total number of blocks in
2239                  * file system block size
2240                  */
2241                 upper_limit = (1LL << 48) - 1;
2242
2243         }
2244
2245         /* indirect blocks */
2246         meta_blocks = 1;
2247         /* double indirect blocks */
2248         meta_blocks += 1 + (1LL << (bits-2));
2249         /* tripple indirect blocks */
2250         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2251
2252         upper_limit -= meta_blocks;
2253         upper_limit <<= bits;
2254
2255         res += 1LL << (bits-2);
2256         res += 1LL << (2*(bits-2));
2257         res += 1LL << (3*(bits-2));
2258         res <<= bits;
2259         if (res > upper_limit)
2260                 res = upper_limit;
2261
2262         if (res > MAX_LFS_FILESIZE)
2263                 res = MAX_LFS_FILESIZE;
2264
2265         return res;
2266 }
2267
2268 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2269                                    ext4_fsblk_t logical_sb_block, int nr)
2270 {
2271         struct ext4_sb_info *sbi = EXT4_SB(sb);
2272         ext4_group_t bg, first_meta_bg;
2273         int has_super = 0;
2274
2275         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2276
2277         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2278             nr < first_meta_bg)
2279                 return logical_sb_block + nr + 1;
2280         bg = sbi->s_desc_per_block * nr;
2281         if (ext4_bg_has_super(sb, bg))
2282                 has_super = 1;
2283
2284         return (has_super + ext4_group_first_block_no(sb, bg));
2285 }
2286
2287 /**
2288  * ext4_get_stripe_size: Get the stripe size.
2289  * @sbi: In memory super block info
2290  *
2291  * If we have specified it via mount option, then
2292  * use the mount option value. If the value specified at mount time is
2293  * greater than the blocks per group use the super block value.
2294  * If the super block value is greater than blocks per group return 0.
2295  * Allocator needs it be less than blocks per group.
2296  *
2297  */
2298 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2299 {
2300         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2301         unsigned long stripe_width =
2302                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2303         int ret;
2304
2305         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2306                 ret = sbi->s_stripe;
2307         else if (stripe_width <= sbi->s_blocks_per_group)
2308                 ret = stripe_width;
2309         else if (stride <= sbi->s_blocks_per_group)
2310                 ret = stride;
2311         else
2312                 ret = 0;
2313
2314         /*
2315          * If the stripe width is 1, this makes no sense and
2316          * we set it to 0 to turn off stripe handling code.
2317          */
2318         if (ret <= 1)
2319                 ret = 0;
2320
2321         return ret;
2322 }
2323
2324 /* sysfs supprt */
2325
2326 struct ext4_attr {
2327         struct attribute attr;
2328         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2329         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2330                          const char *, size_t);
2331         int offset;
2332 };
2333
2334 static int parse_strtoul(const char *buf,
2335                 unsigned long max, unsigned long *value)
2336 {
2337         char *endp;
2338
2339         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2340         endp = skip_spaces(endp);
2341         if (*endp || *value > max)
2342                 return -EINVAL;
2343
2344         return 0;
2345 }
2346
2347 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2348                                               struct ext4_sb_info *sbi,
2349                                               char *buf)
2350 {
2351         return snprintf(buf, PAGE_SIZE, "%llu\n",
2352                 (s64) EXT4_C2B(sbi,
2353                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2354 }
2355
2356 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2357                                          struct ext4_sb_info *sbi, char *buf)
2358 {
2359         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2360
2361         if (!sb->s_bdev->bd_part)
2362                 return snprintf(buf, PAGE_SIZE, "0\n");
2363         return snprintf(buf, PAGE_SIZE, "%lu\n",
2364                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2365                          sbi->s_sectors_written_start) >> 1);
2366 }
2367
2368 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2369                                           struct ext4_sb_info *sbi, char *buf)
2370 {
2371         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2372
2373         if (!sb->s_bdev->bd_part)
2374                 return snprintf(buf, PAGE_SIZE, "0\n");
2375         return snprintf(buf, PAGE_SIZE, "%llu\n",
2376                         (unsigned long long)(sbi->s_kbytes_written +
2377                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2378                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2379 }
2380
2381 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2382                                           struct ext4_sb_info *sbi,
2383                                           const char *buf, size_t count)
2384 {
2385         unsigned long t;
2386
2387         if (parse_strtoul(buf, 0x40000000, &t))
2388                 return -EINVAL;
2389
2390         if (t && !is_power_of_2(t))
2391                 return -EINVAL;
2392
2393         sbi->s_inode_readahead_blks = t;
2394         return count;
2395 }
2396
2397 static ssize_t sbi_ui_show(struct ext4_attr *a,
2398                            struct ext4_sb_info *sbi, char *buf)
2399 {
2400         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2401
2402         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2403 }
2404
2405 static ssize_t sbi_ui_store(struct ext4_attr *a,
2406                             struct ext4_sb_info *sbi,
2407                             const char *buf, size_t count)
2408 {
2409         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2410         unsigned long t;
2411
2412         if (parse_strtoul(buf, 0xffffffff, &t))
2413                 return -EINVAL;
2414         *ui = t;
2415         return count;
2416 }
2417
2418 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2419 static struct ext4_attr ext4_attr_##_name = {                   \
2420         .attr = {.name = __stringify(_name), .mode = _mode },   \
2421         .show   = _show,                                        \
2422         .store  = _store,                                       \
2423         .offset = offsetof(struct ext4_sb_info, _elname),       \
2424 }
2425 #define EXT4_ATTR(name, mode, show, store) \
2426 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2427
2428 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2429 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2430 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2431 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2432         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2433 #define ATTR_LIST(name) &ext4_attr_##name.attr
2434
2435 EXT4_RO_ATTR(delayed_allocation_blocks);
2436 EXT4_RO_ATTR(session_write_kbytes);
2437 EXT4_RO_ATTR(lifetime_write_kbytes);
2438 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2439                  inode_readahead_blks_store, s_inode_readahead_blks);
2440 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2441 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2442 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2443 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2444 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2445 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2446 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2447 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2448
2449 static struct attribute *ext4_attrs[] = {
2450         ATTR_LIST(delayed_allocation_blocks),
2451         ATTR_LIST(session_write_kbytes),
2452         ATTR_LIST(lifetime_write_kbytes),
2453         ATTR_LIST(inode_readahead_blks),
2454         ATTR_LIST(inode_goal),
2455         ATTR_LIST(mb_stats),
2456         ATTR_LIST(mb_max_to_scan),
2457         ATTR_LIST(mb_min_to_scan),
2458         ATTR_LIST(mb_order2_req),
2459         ATTR_LIST(mb_stream_req),
2460         ATTR_LIST(mb_group_prealloc),
2461         ATTR_LIST(max_writeback_mb_bump),
2462         NULL,
2463 };
2464
2465 /* Features this copy of ext4 supports */
2466 EXT4_INFO_ATTR(lazy_itable_init);
2467 EXT4_INFO_ATTR(batched_discard);
2468
2469 static struct attribute *ext4_feat_attrs[] = {
2470         ATTR_LIST(lazy_itable_init),
2471         ATTR_LIST(batched_discard),
2472         NULL,
2473 };
2474
2475 static ssize_t ext4_attr_show(struct kobject *kobj,
2476                               struct attribute *attr, char *buf)
2477 {
2478         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2479                                                 s_kobj);
2480         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2481
2482         return a->show ? a->show(a, sbi, buf) : 0;
2483 }
2484
2485 static ssize_t ext4_attr_store(struct kobject *kobj,
2486                                struct attribute *attr,
2487                                const char *buf, size_t len)
2488 {
2489         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2490                                                 s_kobj);
2491         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2492
2493         return a->store ? a->store(a, sbi, buf, len) : 0;
2494 }
2495
2496 static void ext4_sb_release(struct kobject *kobj)
2497 {
2498         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2499                                                 s_kobj);
2500         complete(&sbi->s_kobj_unregister);
2501 }
2502
2503 static const struct sysfs_ops ext4_attr_ops = {
2504         .show   = ext4_attr_show,
2505         .store  = ext4_attr_store,
2506 };
2507
2508 static struct kobj_type ext4_ktype = {
2509         .default_attrs  = ext4_attrs,
2510         .sysfs_ops      = &ext4_attr_ops,
2511         .release        = ext4_sb_release,
2512 };
2513
2514 static void ext4_feat_release(struct kobject *kobj)
2515 {
2516         complete(&ext4_feat->f_kobj_unregister);
2517 }
2518
2519 static struct kobj_type ext4_feat_ktype = {
2520         .default_attrs  = ext4_feat_attrs,
2521         .sysfs_ops      = &ext4_attr_ops,
2522         .release        = ext4_feat_release,
2523 };
2524
2525 /*
2526  * Check whether this filesystem can be mounted based on
2527  * the features present and the RDONLY/RDWR mount requested.
2528  * Returns 1 if this filesystem can be mounted as requested,
2529  * 0 if it cannot be.
2530  */
2531 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2532 {
2533         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2534                 ext4_msg(sb, KERN_ERR,
2535                         "Couldn't mount because of "
2536                         "unsupported optional features (%x)",
2537                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2538                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2539                 return 0;
2540         }
2541
2542         if (readonly)
2543                 return 1;
2544
2545         /* Check that feature set is OK for a read-write mount */
2546         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2547                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2548                          "unsupported optional features (%x)",
2549                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2550                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2551                 return 0;
2552         }
2553         /*
2554          * Large file size enabled file system can only be mounted
2555          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2556          */
2557         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2558                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2559                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2560                                  "cannot be mounted RDWR without "
2561                                  "CONFIG_LBDAF");
2562                         return 0;
2563                 }
2564         }
2565         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2566             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2567                 ext4_msg(sb, KERN_ERR,
2568                          "Can't support bigalloc feature without "
2569                          "extents feature\n");
2570                 return 0;
2571         }
2572         return 1;
2573 }
2574
2575 /*
2576  * This function is called once a day if we have errors logged
2577  * on the file system
2578  */
2579 static void print_daily_error_info(unsigned long arg)
2580 {
2581         struct super_block *sb = (struct super_block *) arg;
2582         struct ext4_sb_info *sbi;
2583         struct ext4_super_block *es;
2584
2585         sbi = EXT4_SB(sb);
2586         es = sbi->s_es;
2587
2588         if (es->s_error_count)
2589                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2590                          le32_to_cpu(es->s_error_count));
2591         if (es->s_first_error_time) {
2592                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2593                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2594                        (int) sizeof(es->s_first_error_func),
2595                        es->s_first_error_func,
2596                        le32_to_cpu(es->s_first_error_line));
2597                 if (es->s_first_error_ino)
2598                         printk(": inode %u",
2599                                le32_to_cpu(es->s_first_error_ino));
2600                 if (es->s_first_error_block)
2601                         printk(": block %llu", (unsigned long long)
2602                                le64_to_cpu(es->s_first_error_block));
2603                 printk("\n");
2604         }
2605         if (es->s_last_error_time) {
2606                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2607                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2608                        (int) sizeof(es->s_last_error_func),
2609                        es->s_last_error_func,
2610                        le32_to_cpu(es->s_last_error_line));
2611                 if (es->s_last_error_ino)
2612                         printk(": inode %u",
2613                                le32_to_cpu(es->s_last_error_ino));
2614                 if (es->s_last_error_block)
2615                         printk(": block %llu", (unsigned long long)
2616                                le64_to_cpu(es->s_last_error_block));
2617                 printk("\n");
2618         }
2619         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2620 }
2621
2622 /* Find next suitable group and run ext4_init_inode_table */
2623 static int ext4_run_li_request(struct ext4_li_request *elr)
2624 {
2625         struct ext4_group_desc *gdp = NULL;
2626         ext4_group_t group, ngroups;
2627         struct super_block *sb;
2628         unsigned long timeout = 0;
2629         int ret = 0;
2630
2631         sb = elr->lr_super;
2632         ngroups = EXT4_SB(sb)->s_groups_count;
2633
2634         for (group = elr->lr_next_group; group < ngroups; group++) {
2635                 gdp = ext4_get_group_desc(sb, group, NULL);
2636                 if (!gdp) {
2637                         ret = 1;
2638                         break;
2639                 }
2640
2641                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2642                         break;
2643         }
2644
2645         if (group == ngroups)
2646                 ret = 1;
2647
2648         if (!ret) {
2649                 timeout = jiffies;
2650                 ret = ext4_init_inode_table(sb, group,
2651                                             elr->lr_timeout ? 0 : 1);
2652                 if (elr->lr_timeout == 0) {
2653                         timeout = (jiffies - timeout) *
2654                                   elr->lr_sbi->s_li_wait_mult;
2655                         elr->lr_timeout = timeout;
2656                 }
2657                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2658                 elr->lr_next_group = group + 1;
2659         }
2660
2661         return ret;
2662 }
2663
2664 /*
2665  * Remove lr_request from the list_request and free the
2666  * request structure. Should be called with li_list_mtx held
2667  */
2668 static void ext4_remove_li_request(struct ext4_li_request *elr)
2669 {
2670         struct ext4_sb_info *sbi;
2671
2672         if (!elr)
2673                 return;
2674
2675         sbi = elr->lr_sbi;
2676
2677         list_del(&elr->lr_request);
2678         sbi->s_li_request = NULL;
2679         kfree(elr);
2680 }
2681
2682 static void ext4_unregister_li_request(struct super_block *sb)
2683 {
2684         mutex_lock(&ext4_li_mtx);
2685         if (!ext4_li_info) {
2686                 mutex_unlock(&ext4_li_mtx);
2687                 return;
2688         }
2689
2690         mutex_lock(&ext4_li_info->li_list_mtx);
2691         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2692         mutex_unlock(&ext4_li_info->li_list_mtx);
2693         mutex_unlock(&ext4_li_mtx);
2694 }
2695
2696 static struct task_struct *ext4_lazyinit_task;
2697
2698 /*
2699  * This is the function where ext4lazyinit thread lives. It walks
2700  * through the request list searching for next scheduled filesystem.
2701  * When such a fs is found, run the lazy initialization request
2702  * (ext4_rn_li_request) and keep track of the time spend in this
2703  * function. Based on that time we compute next schedule time of
2704  * the request. When walking through the list is complete, compute
2705  * next waking time and put itself into sleep.
2706  */
2707 static int ext4_lazyinit_thread(void *arg)
2708 {
2709         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2710         struct list_head *pos, *n;
2711         struct ext4_li_request *elr;
2712         unsigned long next_wakeup, cur;
2713
2714         BUG_ON(NULL == eli);
2715
2716 cont_thread:
2717         while (true) {
2718                 next_wakeup = MAX_JIFFY_OFFSET;
2719
2720                 mutex_lock(&eli->li_list_mtx);
2721                 if (list_empty(&eli->li_request_list)) {
2722                         mutex_unlock(&eli->li_list_mtx);
2723                         goto exit_thread;
2724                 }
2725
2726                 list_for_each_safe(pos, n, &eli->li_request_list) {
2727                         elr = list_entry(pos, struct ext4_li_request,
2728                                          lr_request);
2729
2730                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2731                                 if (ext4_run_li_request(elr) != 0) {
2732                                         /* error, remove the lazy_init job */
2733                                         ext4_remove_li_request(elr);
2734                                         continue;
2735                                 }
2736                         }
2737
2738                         if (time_before(elr->lr_next_sched, next_wakeup))
2739                                 next_wakeup = elr->lr_next_sched;
2740                 }
2741                 mutex_unlock(&eli->li_list_mtx);
2742
2743                 try_to_freeze();
2744
2745                 cur = jiffies;
2746                 if ((time_after_eq(cur, next_wakeup)) ||
2747                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2748                         cond_resched();
2749                         continue;
2750                 }
2751
2752                 schedule_timeout_interruptible(next_wakeup - cur);
2753
2754                 if (kthread_should_stop()) {
2755                         ext4_clear_request_list();
2756                         goto exit_thread;
2757                 }
2758         }
2759
2760 exit_thread:
2761         /*
2762          * It looks like the request list is empty, but we need
2763          * to check it under the li_list_mtx lock, to prevent any
2764          * additions into it, and of course we should lock ext4_li_mtx
2765          * to atomically free the list and ext4_li_info, because at
2766          * this point another ext4 filesystem could be registering
2767          * new one.
2768          */
2769         mutex_lock(&ext4_li_mtx);
2770         mutex_lock(&eli->li_list_mtx);
2771         if (!list_empty(&eli->li_request_list)) {
2772                 mutex_unlock(&eli->li_list_mtx);
2773                 mutex_unlock(&ext4_li_mtx);
2774                 goto cont_thread;
2775         }
2776         mutex_unlock(&eli->li_list_mtx);
2777         kfree(ext4_li_info);
2778         ext4_li_info = NULL;
2779         mutex_unlock(&ext4_li_mtx);
2780
2781         return 0;
2782 }
2783
2784 static void ext4_clear_request_list(void)
2785 {
2786         struct list_head *pos, *n;
2787         struct ext4_li_request *elr;
2788
2789         mutex_lock(&ext4_li_info->li_list_mtx);
2790         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2791                 elr = list_entry(pos, struct ext4_li_request,
2792                                  lr_request);
2793                 ext4_remove_li_request(elr);
2794         }
2795         mutex_unlock(&ext4_li_info->li_list_mtx);
2796 }
2797
2798 static int ext4_run_lazyinit_thread(void)
2799 {
2800         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2801                                          ext4_li_info, "ext4lazyinit");
2802         if (IS_ERR(ext4_lazyinit_task)) {
2803                 int err = PTR_ERR(ext4_lazyinit_task);
2804                 ext4_clear_request_list();
2805                 kfree(ext4_li_info);
2806                 ext4_li_info = NULL;
2807                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2808                                  "initialization thread\n",
2809                                  err);
2810                 return err;
2811         }
2812         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2813         return 0;
2814 }
2815
2816 /*
2817  * Check whether it make sense to run itable init. thread or not.
2818  * If there is at least one uninitialized inode table, return
2819  * corresponding group number, else the loop goes through all
2820  * groups and return total number of groups.
2821  */
2822 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2823 {
2824         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2825         struct ext4_group_desc *gdp = NULL;
2826
2827         for (group = 0; group < ngroups; group++) {
2828                 gdp = ext4_get_group_desc(sb, group, NULL);
2829                 if (!gdp)
2830                         continue;
2831
2832                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2833                         break;
2834         }
2835
2836         return group;
2837 }
2838
2839 static int ext4_li_info_new(void)
2840 {
2841         struct ext4_lazy_init *eli = NULL;
2842
2843         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2844         if (!eli)
2845                 return -ENOMEM;
2846
2847         INIT_LIST_HEAD(&eli->li_request_list);
2848         mutex_init(&eli->li_list_mtx);
2849
2850         eli->li_state |= EXT4_LAZYINIT_QUIT;
2851
2852         ext4_li_info = eli;
2853
2854         return 0;
2855 }
2856
2857 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2858                                             ext4_group_t start)
2859 {
2860         struct ext4_sb_info *sbi = EXT4_SB(sb);
2861         struct ext4_li_request *elr;
2862         unsigned long rnd;
2863
2864         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2865         if (!elr)
2866                 return NULL;
2867
2868         elr->lr_super = sb;
2869         elr->lr_sbi = sbi;
2870         elr->lr_next_group = start;
2871
2872         /*
2873          * Randomize first schedule time of the request to
2874          * spread the inode table initialization requests
2875          * better.
2876          */
2877         get_random_bytes(&rnd, sizeof(rnd));
2878         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2879                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2880
2881         return elr;
2882 }
2883
2884 static int ext4_register_li_request(struct super_block *sb,
2885                                     ext4_group_t first_not_zeroed)
2886 {
2887         struct ext4_sb_info *sbi = EXT4_SB(sb);
2888         struct ext4_li_request *elr;
2889         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2890         int ret = 0;
2891
2892         if (sbi->s_li_request != NULL) {
2893                 /*
2894                  * Reset timeout so it can be computed again, because
2895                  * s_li_wait_mult might have changed.
2896                  */
2897                 sbi->s_li_request->lr_timeout = 0;
2898                 return 0;
2899         }
2900
2901         if (first_not_zeroed == ngroups ||
2902             (sb->s_flags & MS_RDONLY) ||
2903             !test_opt(sb, INIT_INODE_TABLE))
2904                 return 0;
2905
2906         elr = ext4_li_request_new(sb, first_not_zeroed);
2907         if (!elr)
2908                 return -ENOMEM;
2909
2910         mutex_lock(&ext4_li_mtx);
2911
2912         if (NULL == ext4_li_info) {
2913                 ret = ext4_li_info_new();
2914                 if (ret)
2915                         goto out;
2916         }
2917
2918         mutex_lock(&ext4_li_info->li_list_mtx);
2919         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2920         mutex_unlock(&ext4_li_info->li_list_mtx);
2921
2922         sbi->s_li_request = elr;
2923         /*
2924          * set elr to NULL here since it has been inserted to
2925          * the request_list and the removal and free of it is
2926          * handled by ext4_clear_request_list from now on.
2927          */
2928         elr = NULL;
2929
2930         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2931                 ret = ext4_run_lazyinit_thread();
2932                 if (ret)
2933                         goto out;
2934         }
2935 out:
2936         mutex_unlock(&ext4_li_mtx);
2937         if (ret)
2938                 kfree(elr);
2939         return ret;
2940 }
2941
2942 /*
2943  * We do not need to lock anything since this is called on
2944  * module unload.
2945  */
2946 static void ext4_destroy_lazyinit_thread(void)
2947 {
2948         /*
2949          * If thread exited earlier
2950          * there's nothing to be done.
2951          */
2952         if (!ext4_li_info || !ext4_lazyinit_task)
2953                 return;
2954
2955         kthread_stop(ext4_lazyinit_task);
2956 }
2957
2958 /*
2959  * Note: calculating the overhead so we can be compatible with
2960  * historical BSD practice is quite difficult in the face of
2961  * clusters/bigalloc.  This is because multiple metadata blocks from
2962  * different block group can end up in the same allocation cluster.
2963  * Calculating the exact overhead in the face of clustered allocation
2964  * requires either O(all block bitmaps) in memory or O(number of block
2965  * groups**2) in time.  We will still calculate the superblock for
2966  * older file systems --- and if we come across with a bigalloc file
2967  * system with zero in s_overhead_clusters the estimate will be close to
2968  * correct especially for very large cluster sizes --- but for newer
2969  * file systems, it's better to calculate this figure once at mkfs
2970  * time, and store it in the superblock.  If the superblock value is
2971  * present (even for non-bigalloc file systems), we will use it.
2972  */
2973 static int count_overhead(struct super_block *sb, ext4_group_t grp,
2974                           char *buf)
2975 {
2976         struct ext4_sb_info     *sbi = EXT4_SB(sb);
2977         struct ext4_group_desc  *gdp;
2978         ext4_fsblk_t            first_block, last_block, b;
2979         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
2980         int                     s, j, count = 0;
2981
2982         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
2983                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
2984                         sbi->s_itb_per_group + 2);
2985
2986         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
2987                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
2988         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
2989         for (i = 0; i < ngroups; i++) {
2990                 gdp = ext4_get_group_desc(sb, i, NULL);
2991                 b = ext4_block_bitmap(sb, gdp);
2992                 if (b >= first_block && b <= last_block) {
2993                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
2994                         count++;
2995                 }
2996                 b = ext4_inode_bitmap(sb, gdp);
2997                 if (b >= first_block && b <= last_block) {
2998                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
2999                         count++;
3000                 }
3001                 b = ext4_inode_table(sb, gdp);
3002                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3003                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3004                                 int c = EXT4_B2C(sbi, b - first_block);
3005                                 ext4_set_bit(c, buf);
3006                                 count++;
3007                         }
3008                 if (i != grp)
3009                         continue;
3010                 s = 0;
3011                 if (ext4_bg_has_super(sb, grp)) {
3012                         ext4_set_bit(s++, buf);
3013                         count++;
3014                 }
3015                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3016                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3017                         count++;
3018                 }
3019         }
3020         if (!count)
3021                 return 0;
3022         return EXT4_CLUSTERS_PER_GROUP(sb) -
3023                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3024 }
3025
3026 /*
3027  * Compute the overhead and stash it in sbi->s_overhead
3028  */
3029 int ext4_calculate_overhead(struct super_block *sb)
3030 {
3031         struct ext4_sb_info *sbi = EXT4_SB(sb);
3032         struct ext4_super_block *es = sbi->s_es;
3033         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3034         ext4_fsblk_t overhead = 0;
3035         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3036
3037         memset(buf, 0, PAGE_SIZE);
3038         if (!buf)
3039                 return -ENOMEM;
3040
3041         /*
3042          * Compute the overhead (FS structures).  This is constant
3043          * for a given filesystem unless the number of block groups
3044          * changes so we cache the previous value until it does.
3045          */
3046
3047         /*
3048          * All of the blocks before first_data_block are overhead
3049          */
3050         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3051
3052         /*
3053          * Add the overhead found in each block group
3054          */
3055         for (i = 0; i < ngroups; i++) {
3056                 int blks;
3057
3058                 blks = count_overhead(sb, i, buf);
3059                 overhead += blks;
3060                 if (blks)
3061                         memset(buf, 0, PAGE_SIZE);
3062                 cond_resched();
3063         }
3064         sbi->s_overhead = overhead;
3065         smp_wmb();
3066         free_page((unsigned long) buf);
3067         return 0;
3068 }
3069
3070 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3071 {
3072         char *orig_data = kstrdup(data, GFP_KERNEL);
3073         struct buffer_head *bh;
3074         struct ext4_super_block *es = NULL;
3075         struct ext4_sb_info *sbi;
3076         ext4_fsblk_t block;
3077         ext4_fsblk_t sb_block = get_sb_block(&data);
3078         ext4_fsblk_t logical_sb_block;
3079         unsigned long offset = 0;
3080         unsigned long journal_devnum = 0;
3081         unsigned long def_mount_opts;
3082         struct inode *root;
3083         char *cp;
3084         const char *descr;
3085         int ret = -ENOMEM;
3086         int blocksize, clustersize;
3087         unsigned int db_count;
3088         unsigned int i;
3089         int needs_recovery, has_huge_files, has_bigalloc;
3090         __u64 blocks_count;
3091         int err;
3092         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3093         ext4_group_t first_not_zeroed;
3094
3095         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3096         if (!sbi)
3097                 goto out_free_orig;
3098
3099         sbi->s_blockgroup_lock =
3100                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3101         if (!sbi->s_blockgroup_lock) {
3102                 kfree(sbi);
3103                 goto out_free_orig;
3104         }
3105         sb->s_fs_info = sbi;
3106         sbi->s_mount_opt = 0;
3107         sbi->s_resuid = EXT4_DEF_RESUID;
3108         sbi->s_resgid = EXT4_DEF_RESGID;
3109         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3110         sbi->s_sb_block = sb_block;
3111         if (sb->s_bdev->bd_part)
3112                 sbi->s_sectors_written_start =
3113                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3114
3115         /* Cleanup superblock name */
3116         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3117                 *cp = '!';
3118
3119         ret = -EINVAL;
3120         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3121         if (!blocksize) {
3122                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3123                 goto out_fail;
3124         }
3125
3126         /*
3127          * The ext4 superblock will not be buffer aligned for other than 1kB
3128          * block sizes.  We need to calculate the offset from buffer start.
3129          */
3130         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3131                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3132                 offset = do_div(logical_sb_block, blocksize);
3133         } else {
3134                 logical_sb_block = sb_block;
3135         }
3136
3137         if (!(bh = sb_bread(sb, logical_sb_block))) {
3138                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3139                 goto out_fail;
3140         }
3141         /*
3142          * Note: s_es must be initialized as soon as possible because
3143          *       some ext4 macro-instructions depend on its value
3144          */
3145         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3146         sbi->s_es = es;
3147         sb->s_magic = le16_to_cpu(es->s_magic);
3148         if (sb->s_magic != EXT4_SUPER_MAGIC)
3149                 goto cantfind_ext4;
3150         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3151
3152         /* Set defaults before we parse the mount options */
3153         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3154         set_opt(sb, INIT_INODE_TABLE);
3155         if (def_mount_opts & EXT4_DEFM_DEBUG)
3156                 set_opt(sb, DEBUG);
3157         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3158                 set_opt(sb, GRPID);
3159         if (def_mount_opts & EXT4_DEFM_UID16)
3160                 set_opt(sb, NO_UID32);
3161         /* xattr user namespace & acls are now defaulted on */
3162 #ifdef CONFIG_EXT4_FS_XATTR
3163         set_opt(sb, XATTR_USER);
3164 #endif
3165 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3166         set_opt(sb, POSIX_ACL);
3167 #endif
3168         set_opt(sb, MBLK_IO_SUBMIT);
3169         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3170                 set_opt(sb, JOURNAL_DATA);
3171         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3172                 set_opt(sb, ORDERED_DATA);
3173         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3174                 set_opt(sb, WRITEBACK_DATA);
3175
3176         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3177                 set_opt(sb, ERRORS_PANIC);
3178         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3179                 set_opt(sb, ERRORS_CONT);
3180         else
3181                 set_opt(sb, ERRORS_RO);
3182         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3183                 set_opt(sb, BLOCK_VALIDITY);
3184         if (def_mount_opts & EXT4_DEFM_DISCARD)
3185                 set_opt(sb, DISCARD);
3186
3187         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3188         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3189         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3190         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3191         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3192
3193         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3194                 set_opt(sb, BARRIER);
3195
3196         /*
3197          * enable delayed allocation by default
3198          * Use -o nodelalloc to turn it off
3199          */
3200         if (!IS_EXT3_SB(sb) &&
3201             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3202                 set_opt(sb, DELALLOC);
3203
3204         /*
3205          * set default s_li_wait_mult for lazyinit, for the case there is
3206          * no mount option specified.
3207          */
3208         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3209
3210         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3211                            &journal_devnum, &journal_ioprio, 0)) {
3212                 ext4_msg(sb, KERN_WARNING,
3213                          "failed to parse options in superblock: %s",
3214                          sbi->s_es->s_mount_opts);
3215         }
3216         sbi->s_def_mount_opt = sbi->s_mount_opt;
3217         if (!parse_options((char *) data, sb, &journal_devnum,
3218                            &journal_ioprio, 0))
3219                 goto failed_mount;
3220
3221         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3222                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3223                             "with data=journal disables delayed "
3224                             "allocation and O_DIRECT support!\n");
3225                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3226                         ext4_msg(sb, KERN_ERR, "can't mount with "
3227                                  "both data=journal and delalloc");
3228                         goto failed_mount;
3229                 }
3230                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3231                         ext4_msg(sb, KERN_ERR, "can't mount with "
3232                                  "both data=journal and delalloc");
3233                         goto failed_mount;
3234                 }
3235                 if (test_opt(sb, DELALLOC))
3236                         clear_opt(sb, DELALLOC);
3237         }
3238
3239         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3240                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3241
3242         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3243             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3244              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3245              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3246                 ext4_msg(sb, KERN_WARNING,
3247                        "feature flags set on rev 0 fs, "
3248                        "running e2fsck is recommended");
3249
3250         if (IS_EXT2_SB(sb)) {
3251                 if (ext2_feature_set_ok(sb))
3252                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3253                                  "using the ext4 subsystem");
3254                 else {
3255                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3256                                  "to feature incompatibilities");
3257                         goto failed_mount;
3258                 }
3259         }
3260
3261         if (IS_EXT3_SB(sb)) {
3262                 if (ext3_feature_set_ok(sb))
3263                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3264                                  "using the ext4 subsystem");
3265                 else {
3266                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3267                                  "to feature incompatibilities");
3268                         goto failed_mount;
3269                 }
3270         }
3271
3272         /*
3273          * Check feature flags regardless of the revision level, since we
3274          * previously didn't change the revision level when setting the flags,
3275          * so there is a chance incompat flags are set on a rev 0 filesystem.
3276          */
3277         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3278                 goto failed_mount;
3279
3280         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3281         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3282             blocksize > EXT4_MAX_BLOCK_SIZE) {
3283                 ext4_msg(sb, KERN_ERR,
3284                        "Unsupported filesystem blocksize %d", blocksize);
3285                 goto failed_mount;
3286         }
3287
3288         if (sb->s_blocksize != blocksize) {
3289                 /* Validate the filesystem blocksize */
3290                 if (!sb_set_blocksize(sb, blocksize)) {
3291                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3292                                         blocksize);
3293                         goto failed_mount;
3294                 }
3295
3296                 brelse(bh);
3297                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3298                 offset = do_div(logical_sb_block, blocksize);
3299                 bh = sb_bread(sb, logical_sb_block);
3300                 if (!bh) {
3301                         ext4_msg(sb, KERN_ERR,
3302                                "Can't read superblock on 2nd try");
3303                         goto failed_mount;
3304                 }
3305                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3306                 sbi->s_es = es;
3307                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3308                         ext4_msg(sb, KERN_ERR,
3309                                "Magic mismatch, very weird!");
3310                         goto failed_mount;
3311                 }
3312         }
3313
3314         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3315                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3316         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3317                                                       has_huge_files);
3318         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3319
3320         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3321                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3322                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3323         } else {
3324                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3325                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3326                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3327                     (!is_power_of_2(sbi->s_inode_size)) ||
3328                     (sbi->s_inode_size > blocksize)) {
3329                         ext4_msg(sb, KERN_ERR,
3330                                "unsupported inode size: %d",
3331                                sbi->s_inode_size);
3332                         goto failed_mount;
3333                 }
3334                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3335                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3336         }
3337
3338         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3339         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3340                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3341                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3342                     !is_power_of_2(sbi->s_desc_size)) {
3343                         ext4_msg(sb, KERN_ERR,
3344                                "unsupported descriptor size %lu",
3345                                sbi->s_desc_size);
3346                         goto failed_mount;
3347                 }
3348         } else
3349                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3350
3351         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3352         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3353         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3354                 goto cantfind_ext4;
3355
3356         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3357         if (sbi->s_inodes_per_block == 0)
3358                 goto cantfind_ext4;
3359         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3360                                         sbi->s_inodes_per_block;
3361         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3362         sbi->s_sbh = bh;
3363         sbi->s_mount_state = le16_to_cpu(es->s_state);
3364         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3365         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3366
3367         for (i = 0; i < 4; i++)
3368                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3369         sbi->s_def_hash_version = es->s_def_hash_version;
3370         i = le32_to_cpu(es->s_flags);
3371         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3372                 sbi->s_hash_unsigned = 3;
3373         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3374 #ifdef __CHAR_UNSIGNED__
3375                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3376                 sbi->s_hash_unsigned = 3;
3377 #else
3378                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3379 #endif
3380         }
3381
3382         /* Handle clustersize */
3383         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3384         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3385                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3386         if (has_bigalloc) {
3387                 if (clustersize < blocksize) {
3388                         ext4_msg(sb, KERN_ERR,
3389                                  "cluster size (%d) smaller than "
3390                                  "block size (%d)", clustersize, blocksize);
3391                         goto failed_mount;
3392                 }
3393                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3394                         le32_to_cpu(es->s_log_block_size);
3395                 sbi->s_clusters_per_group =
3396                         le32_to_cpu(es->s_clusters_per_group);
3397                 if (sbi->s_clusters_per_group > blocksize * 8) {
3398                         ext4_msg(sb, KERN_ERR,
3399                                  "#clusters per group too big: %lu",
3400                                  sbi->s_clusters_per_group);
3401                         goto failed_mount;
3402                 }
3403                 if (sbi->s_blocks_per_group !=
3404                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3405                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3406                                  "clusters per group (%lu) inconsistent",
3407                                  sbi->s_blocks_per_group,
3408                                  sbi->s_clusters_per_group);
3409                         goto failed_mount;
3410                 }
3411         } else {
3412                 if (clustersize != blocksize) {
3413                         ext4_warning(sb, "fragment/cluster size (%d) != "
3414                                      "block size (%d)", clustersize,
3415                                      blocksize);
3416                         clustersize = blocksize;
3417                 }
3418                 if (sbi->s_blocks_per_group > blocksize * 8) {
3419                         ext4_msg(sb, KERN_ERR,
3420                                  "#blocks per group too big: %lu",
3421                                  sbi->s_blocks_per_group);
3422                         goto failed_mount;
3423                 }
3424                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3425                 sbi->s_cluster_bits = 0;
3426         }
3427         sbi->s_cluster_ratio = clustersize / blocksize;
3428
3429         if (sbi->s_inodes_per_group > blocksize * 8) {
3430                 ext4_msg(sb, KERN_ERR,
3431                        "#inodes per group too big: %lu",
3432                        sbi->s_inodes_per_group);
3433                 goto failed_mount;
3434         }
3435
3436         /*
3437          * Test whether we have more sectors than will fit in sector_t,
3438          * and whether the max offset is addressable by the page cache.
3439          */
3440         err = generic_check_addressable(sb->s_blocksize_bits,
3441                                         ext4_blocks_count(es));
3442         if (err) {
3443                 ext4_msg(sb, KERN_ERR, "filesystem"
3444                          " too large to mount safely on this system");
3445                 if (sizeof(sector_t) < 8)
3446                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3447                 ret = err;
3448                 goto failed_mount;
3449         }
3450
3451         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3452                 goto cantfind_ext4;
3453
3454         /* check blocks count against device size */
3455         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3456         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3457                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3458                        "exceeds size of device (%llu blocks)",
3459                        ext4_blocks_count(es), blocks_count);
3460                 goto failed_mount;
3461         }
3462
3463         /*
3464          * It makes no sense for the first data block to be beyond the end
3465          * of the filesystem.
3466          */
3467         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3468                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3469                          "block %u is beyond end of filesystem (%llu)",
3470                          le32_to_cpu(es->s_first_data_block),
3471                          ext4_blocks_count(es));
3472                 goto failed_mount;
3473         }
3474         blocks_count = (ext4_blocks_count(es) -
3475                         le32_to_cpu(es->s_first_data_block) +
3476                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3477         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3478         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3479                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3480                        "(block count %llu, first data block %u, "
3481                        "blocks per group %lu)", sbi->s_groups_count,
3482                        ext4_blocks_count(es),
3483                        le32_to_cpu(es->s_first_data_block),
3484                        EXT4_BLOCKS_PER_GROUP(sb));
3485                 goto failed_mount;
3486         }
3487         sbi->s_groups_count = blocks_count;
3488         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3489                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3490         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3491                    EXT4_DESC_PER_BLOCK(sb);
3492         sbi->s_group_desc = ext4_kvmalloc(db_count *
3493                                           sizeof(struct buffer_head *),
3494                                           GFP_KERNEL);
3495         if (sbi->s_group_desc == NULL) {
3496                 ext4_msg(sb, KERN_ERR, "not enough memory");
3497                 goto failed_mount;
3498         }
3499
3500         if (ext4_proc_root)
3501                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3502
3503         if (sbi->s_proc)
3504                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3505                                  &ext4_seq_options_fops, sb);
3506
3507         bgl_lock_init(sbi->s_blockgroup_lock);
3508
3509         for (i = 0; i < db_count; i++) {
3510                 block = descriptor_loc(sb, logical_sb_block, i);
3511                 sbi->s_group_desc[i] = sb_bread(sb, block);
3512                 if (!sbi->s_group_desc[i]) {
3513                         ext4_msg(sb, KERN_ERR,
3514                                "can't read group descriptor %d", i);
3515                         db_count = i;
3516                         goto failed_mount2;
3517                 }
3518         }
3519         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3520                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3521                 goto failed_mount2;
3522         }
3523         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3524                 if (!ext4_fill_flex_info(sb)) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "unable to initialize "
3527                                "flex_bg meta info!");
3528                         goto failed_mount2;
3529                 }
3530
3531         sbi->s_gdb_count = db_count;
3532         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3533         spin_lock_init(&sbi->s_next_gen_lock);
3534
3535         init_timer(&sbi->s_err_report);
3536         sbi->s_err_report.function = print_daily_error_info;
3537         sbi->s_err_report.data = (unsigned long) sb;
3538
3539         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3540                         ext4_count_free_clusters(sb));
3541         if (!err) {
3542                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3543                                 ext4_count_free_inodes(sb));
3544         }
3545         if (!err) {
3546                 err = percpu_counter_init(&sbi->s_dirs_counter,
3547                                 ext4_count_dirs(sb));
3548         }
3549         if (!err) {
3550                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3551         }
3552         if (err) {
3553                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3554                 goto failed_mount3;
3555         }
3556
3557         sbi->s_stripe = ext4_get_stripe_size(sbi);
3558         sbi->s_max_writeback_mb_bump = 128;
3559
3560         /*
3561          * set up enough so that it can read an inode
3562          */
3563         if (!test_opt(sb, NOLOAD) &&
3564             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3565                 sb->s_op = &ext4_sops;
3566         else
3567                 sb->s_op = &ext4_nojournal_sops;
3568         sb->s_export_op = &ext4_export_ops;
3569         sb->s_xattr = ext4_xattr_handlers;
3570 #ifdef CONFIG_QUOTA
3571         sb->s_qcop = &ext4_qctl_operations;
3572         sb->dq_op = &ext4_quota_operations;
3573 #endif
3574         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3575
3576         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3577         mutex_init(&sbi->s_orphan_lock);
3578         sbi->s_resize_flags = 0;
3579
3580         sb->s_root = NULL;
3581
3582         needs_recovery = (es->s_last_orphan != 0 ||
3583                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3584                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3585
3586         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3587             !(sb->s_flags & MS_RDONLY))
3588                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3589                         goto failed_mount3;
3590
3591         /*
3592          * The first inode we look at is the journal inode.  Don't try
3593          * root first: it may be modified in the journal!
3594          */
3595         if (!test_opt(sb, NOLOAD) &&
3596             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3597                 if (ext4_load_journal(sb, es, journal_devnum))
3598                         goto failed_mount3;
3599         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3600               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3601                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3602                        "suppressed and not mounted read-only");
3603                 goto failed_mount_wq;
3604         } else {
3605                 clear_opt(sb, DATA_FLAGS);
3606                 sbi->s_journal = NULL;
3607                 needs_recovery = 0;
3608                 goto no_journal;
3609         }
3610
3611         if (ext4_blocks_count(es) > 0xffffffffULL &&
3612             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3613                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3614                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3615                 goto failed_mount_wq;
3616         }
3617
3618         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3619                 jbd2_journal_set_features(sbi->s_journal,
3620                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3621                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3622         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3623                 jbd2_journal_set_features(sbi->s_journal,
3624                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3625                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3626                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3627         } else {
3628                 jbd2_journal_clear_features(sbi->s_journal,
3629                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3630                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3631         }
3632
3633         /* We have now updated the journal if required, so we can
3634          * validate the data journaling mode. */
3635         switch (test_opt(sb, DATA_FLAGS)) {
3636         case 0:
3637                 /* No mode set, assume a default based on the journal
3638                  * capabilities: ORDERED_DATA if the journal can
3639                  * cope, else JOURNAL_DATA
3640                  */
3641                 if (jbd2_journal_check_available_features
3642                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3643                         set_opt(sb, ORDERED_DATA);
3644                 else
3645                         set_opt(sb, JOURNAL_DATA);
3646                 break;
3647
3648         case EXT4_MOUNT_ORDERED_DATA:
3649         case EXT4_MOUNT_WRITEBACK_DATA:
3650                 if (!jbd2_journal_check_available_features
3651                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3652                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3653                                "requested data journaling mode");
3654                         goto failed_mount_wq;
3655                 }
3656         default:
3657                 break;
3658         }
3659         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3660
3661         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3662
3663         /*
3664          * The journal may have updated the bg summary counts, so we
3665          * need to update the global counters.
3666          */
3667         percpu_counter_set(&sbi->s_freeclusters_counter,
3668                            ext4_count_free_clusters(sb));
3669         percpu_counter_set(&sbi->s_freeinodes_counter,
3670                            ext4_count_free_inodes(sb));
3671         percpu_counter_set(&sbi->s_dirs_counter,
3672                            ext4_count_dirs(sb));
3673         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3674
3675 no_journal:
3676         /*
3677          * Get the # of file system overhead blocks from the
3678          * superblock if present.
3679          */
3680         if (es->s_overhead_clusters)
3681                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3682         else {
3683                 ret = ext4_calculate_overhead(sb);
3684                 if (ret)
3685                         goto failed_mount_wq;
3686         }
3687
3688         /*
3689          * The maximum number of concurrent works can be high and
3690          * concurrency isn't really necessary.  Limit it to 1.
3691          */
3692         EXT4_SB(sb)->dio_unwritten_wq =
3693                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3694         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3695                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3696                 goto failed_mount_wq;
3697         }
3698
3699         /*
3700          * The jbd2_journal_load will have done any necessary log recovery,
3701          * so we can safely mount the rest of the filesystem now.
3702          */
3703
3704         root = ext4_iget(sb, EXT4_ROOT_INO);
3705         if (IS_ERR(root)) {
3706                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3707                 ret = PTR_ERR(root);
3708                 root = NULL;
3709                 goto failed_mount4;
3710         }
3711         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3712                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3713                 iput(root);
3714                 goto failed_mount4;
3715         }
3716         sb->s_root = d_make_root(root);
3717         if (!sb->s_root) {
3718                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3719                 ret = -ENOMEM;
3720                 goto failed_mount4;
3721         }
3722
3723         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3724                 sb->s_flags |= MS_RDONLY;
3725
3726         /* determine the minimum size of new large inodes, if present */
3727         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3728                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3729                                                      EXT4_GOOD_OLD_INODE_SIZE;
3730                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3731                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3732                         if (sbi->s_want_extra_isize <
3733                             le16_to_cpu(es->s_want_extra_isize))
3734                                 sbi->s_want_extra_isize =
3735                                         le16_to_cpu(es->s_want_extra_isize);
3736                         if (sbi->s_want_extra_isize <
3737                             le16_to_cpu(es->s_min_extra_isize))
3738                                 sbi->s_want_extra_isize =
3739                                         le16_to_cpu(es->s_min_extra_isize);
3740                 }
3741         }
3742         /* Check if enough inode space is available */
3743         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3744                                                         sbi->s_inode_size) {
3745                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3746                                                        EXT4_GOOD_OLD_INODE_SIZE;
3747                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3748                          "available");
3749         }
3750
3751         err = ext4_setup_system_zone(sb);
3752         if (err) {
3753                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3754                          "zone (%d)", err);
3755                 goto failed_mount4a;
3756         }
3757
3758         ext4_ext_init(sb);
3759         err = ext4_mb_init(sb, needs_recovery);
3760         if (err) {
3761                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3762                          err);
3763                 goto failed_mount5;
3764         }
3765
3766         err = ext4_register_li_request(sb, first_not_zeroed);
3767         if (err)
3768                 goto failed_mount6;
3769
3770         sbi->s_kobj.kset = ext4_kset;
3771         init_completion(&sbi->s_kobj_unregister);
3772         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3773                                    "%s", sb->s_id);
3774         if (err)
3775                 goto failed_mount7;
3776
3777         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3778         ext4_orphan_cleanup(sb, es);
3779         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3780         if (needs_recovery) {
3781                 ext4_msg(sb, KERN_INFO, "recovery complete");
3782                 ext4_mark_recovery_complete(sb, es);
3783         }
3784         if (EXT4_SB(sb)->s_journal) {
3785                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3786                         descr = " journalled data mode";
3787                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3788                         descr = " ordered data mode";
3789                 else
3790                         descr = " writeback data mode";
3791         } else
3792                 descr = "out journal";
3793
3794         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3795                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3796                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3797
3798         if (es->s_error_count)
3799                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3800
3801         kfree(orig_data);
3802         return 0;
3803
3804 cantfind_ext4:
3805         if (!silent)
3806                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3807         goto failed_mount;
3808
3809 failed_mount7:
3810         ext4_unregister_li_request(sb);
3811 failed_mount6:
3812         ext4_mb_release(sb);
3813 failed_mount5:
3814         ext4_ext_release(sb);
3815         ext4_release_system_zone(sb);
3816 failed_mount4a:
3817         dput(sb->s_root);
3818         sb->s_root = NULL;
3819 failed_mount4:
3820         ext4_msg(sb, KERN_ERR, "mount failed");
3821         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3822 failed_mount_wq:
3823         if (sbi->s_journal) {
3824                 jbd2_journal_destroy(sbi->s_journal);
3825                 sbi->s_journal = NULL;
3826         }
3827 failed_mount3:
3828         del_timer(&sbi->s_err_report);
3829         if (sbi->s_flex_groups)
3830                 ext4_kvfree(sbi->s_flex_groups);
3831         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3832         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3833         percpu_counter_destroy(&sbi->s_dirs_counter);
3834         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3835         if (sbi->s_mmp_tsk)
3836                 kthread_stop(sbi->s_mmp_tsk);
3837 failed_mount2:
3838         for (i = 0; i < db_count; i++)
3839                 brelse(sbi->s_group_desc[i]);
3840         ext4_kvfree(sbi->s_group_desc);
3841 failed_mount:
3842         if (sbi->s_proc) {
3843                 remove_proc_entry("options", sbi->s_proc);
3844                 remove_proc_entry(sb->s_id, ext4_proc_root);
3845         }
3846 #ifdef CONFIG_QUOTA
3847         for (i = 0; i < MAXQUOTAS; i++)
3848                 kfree(sbi->s_qf_names[i]);
3849 #endif
3850         ext4_blkdev_remove(sbi);
3851         brelse(bh);
3852 out_fail:
3853         sb->s_fs_info = NULL;
3854         kfree(sbi->s_blockgroup_lock);
3855         kfree(sbi);
3856 out_free_orig:
3857         kfree(orig_data);
3858         return ret;
3859 }
3860
3861 /*
3862  * Setup any per-fs journal parameters now.  We'll do this both on
3863  * initial mount, once the journal has been initialised but before we've
3864  * done any recovery; and again on any subsequent remount.
3865  */
3866 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3867 {
3868         struct ext4_sb_info *sbi = EXT4_SB(sb);
3869
3870         journal->j_commit_interval = sbi->s_commit_interval;
3871         journal->j_min_batch_time = sbi->s_min_batch_time;
3872         journal->j_max_batch_time = sbi->s_max_batch_time;
3873
3874         write_lock(&journal->j_state_lock);
3875         if (test_opt(sb, BARRIER))
3876                 journal->j_flags |= JBD2_BARRIER;
3877         else
3878                 journal->j_flags &= ~JBD2_BARRIER;
3879         if (test_opt(sb, DATA_ERR_ABORT))
3880                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3881         else
3882                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3883         write_unlock(&journal->j_state_lock);
3884 }
3885
3886 static journal_t *ext4_get_journal(struct super_block *sb,
3887                                    unsigned int journal_inum)
3888 {
3889         struct inode *journal_inode;
3890         journal_t *journal;
3891
3892         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3893
3894         /* First, test for the existence of a valid inode on disk.  Bad
3895          * things happen if we iget() an unused inode, as the subsequent
3896          * iput() will try to delete it. */
3897
3898         journal_inode = ext4_iget(sb, journal_inum);
3899         if (IS_ERR(journal_inode)) {
3900                 ext4_msg(sb, KERN_ERR, "no journal found");
3901                 return NULL;
3902         }
3903         if (!journal_inode->i_nlink) {
3904                 make_bad_inode(journal_inode);
3905                 iput(journal_inode);
3906                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3907                 return NULL;
3908         }
3909
3910         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3911                   journal_inode, journal_inode->i_size);
3912         if (!S_ISREG(journal_inode->i_mode)) {
3913                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3914                 iput(journal_inode);
3915                 return NULL;
3916         }
3917
3918         journal = jbd2_journal_init_inode(journal_inode);
3919         if (!journal) {
3920                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3921                 iput(journal_inode);
3922                 return NULL;
3923         }
3924         journal->j_private = sb;
3925         ext4_init_journal_params(sb, journal);
3926         return journal;
3927 }
3928
3929 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3930                                        dev_t j_dev)
3931 {
3932         struct buffer_head *bh;
3933         journal_t *journal;
3934         ext4_fsblk_t start;
3935         ext4_fsblk_t len;
3936         int hblock, blocksize;
3937         ext4_fsblk_t sb_block;
3938         unsigned long offset;
3939         struct ext4_super_block *es;
3940         struct block_device *bdev;
3941
3942         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3943
3944         bdev = ext4_blkdev_get(j_dev, sb);
3945         if (bdev == NULL)
3946                 return NULL;
3947
3948         blocksize = sb->s_blocksize;
3949         hblock = bdev_logical_block_size(bdev);
3950         if (blocksize < hblock) {
3951                 ext4_msg(sb, KERN_ERR,
3952                         "blocksize too small for journal device");
3953                 goto out_bdev;
3954         }
3955
3956         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3957         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3958         set_blocksize(bdev, blocksize);
3959         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3960                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3961                        "external journal");
3962                 goto out_bdev;
3963         }
3964
3965         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3966         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3967             !(le32_to_cpu(es->s_feature_incompat) &
3968               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3969                 ext4_msg(sb, KERN_ERR, "external journal has "
3970                                         "bad superblock");
3971                 brelse(bh);
3972                 goto out_bdev;
3973         }
3974
3975         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3976                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3977                 brelse(bh);
3978                 goto out_bdev;
3979         }
3980
3981         len = ext4_blocks_count(es);
3982         start = sb_block + 1;
3983         brelse(bh);     /* we're done with the superblock */
3984
3985         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3986                                         start, len, blocksize);
3987         if (!journal) {
3988                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3989                 goto out_bdev;
3990         }
3991         journal->j_private = sb;
3992         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3993         wait_on_buffer(journal->j_sb_buffer);
3994         if (!buffer_uptodate(journal->j_sb_buffer)) {
3995                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3996                 goto out_journal;
3997         }
3998         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3999                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4000                                         "user (unsupported) - %d",
4001                         be32_to_cpu(journal->j_superblock->s_nr_users));
4002                 goto out_journal;
4003         }
4004         EXT4_SB(sb)->journal_bdev = bdev;
4005         ext4_init_journal_params(sb, journal);
4006         return journal;
4007
4008 out_journal:
4009         jbd2_journal_destroy(journal);
4010 out_bdev:
4011         ext4_blkdev_put(bdev);
4012         return NULL;
4013 }
4014
4015 static int ext4_load_journal(struct super_block *sb,
4016                              struct ext4_super_block *es,
4017                              unsigned long journal_devnum)
4018 {
4019         journal_t *journal;
4020         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4021         dev_t journal_dev;
4022         int err = 0;
4023         int really_read_only;
4024
4025         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4026
4027         if (journal_devnum &&
4028             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4029                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4030                         "numbers have changed");
4031                 journal_dev = new_decode_dev(journal_devnum);
4032         } else
4033                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4034
4035         really_read_only = bdev_read_only(sb->s_bdev);
4036
4037         /*
4038          * Are we loading a blank journal or performing recovery after a
4039          * crash?  For recovery, we need to check in advance whether we
4040          * can get read-write access to the device.
4041          */
4042         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4043                 if (sb->s_flags & MS_RDONLY) {
4044                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4045                                         "required on readonly filesystem");
4046                         if (really_read_only) {
4047                                 ext4_msg(sb, KERN_ERR, "write access "
4048                                         "unavailable, cannot proceed");
4049                                 return -EROFS;
4050                         }
4051                         ext4_msg(sb, KERN_INFO, "write access will "
4052                                "be enabled during recovery");
4053                 }
4054         }
4055
4056         if (journal_inum && journal_dev) {
4057                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4058                        "and inode journals!");
4059                 return -EINVAL;
4060         }
4061
4062         if (journal_inum) {
4063                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4064                         return -EINVAL;
4065         } else {
4066                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4067                         return -EINVAL;
4068         }
4069
4070         if (!(journal->j_flags & JBD2_BARRIER))
4071                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4072
4073         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4074                 err = jbd2_journal_wipe(journal, !really_read_only);
4075         if (!err) {
4076                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4077                 if (save)
4078                         memcpy(save, ((char *) es) +
4079                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4080                 err = jbd2_journal_load(journal);
4081                 if (save)
4082                         memcpy(((char *) es) + EXT4_S_ERR_START,
4083                                save, EXT4_S_ERR_LEN);
4084                 kfree(save);
4085         }
4086
4087         if (err) {
4088                 ext4_msg(sb, KERN_ERR, "error loading journal");
4089                 jbd2_journal_destroy(journal);
4090                 return err;
4091         }
4092
4093         EXT4_SB(sb)->s_journal = journal;
4094         ext4_clear_journal_err(sb, es);
4095
4096         if (!really_read_only && journal_devnum &&
4097             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4098                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4099
4100                 /* Make sure we flush the recovery flag to disk. */
4101                 ext4_commit_super(sb, 1);
4102         }
4103
4104         return 0;
4105 }
4106
4107 static int ext4_commit_super(struct super_block *sb, int sync)
4108 {
4109         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4110         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4111         int error = 0;
4112
4113         if (!sbh || block_device_ejected(sb))
4114                 return error;
4115         if (buffer_write_io_error(sbh)) {
4116                 /*
4117                  * Oh, dear.  A previous attempt to write the
4118                  * superblock failed.  This could happen because the
4119                  * USB device was yanked out.  Or it could happen to
4120                  * be a transient write error and maybe the block will
4121                  * be remapped.  Nothing we can do but to retry the
4122                  * write and hope for the best.
4123                  */
4124                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4125                        "superblock detected");
4126                 clear_buffer_write_io_error(sbh);
4127                 set_buffer_uptodate(sbh);
4128         }
4129         /*
4130          * If the file system is mounted read-only, don't update the
4131          * superblock write time.  This avoids updating the superblock
4132          * write time when we are mounting the root file system
4133          * read/only but we need to replay the journal; at that point,
4134          * for people who are east of GMT and who make their clock
4135          * tick in localtime for Windows bug-for-bug compatibility,
4136          * the clock is set in the future, and this will cause e2fsck
4137          * to complain and force a full file system check.
4138          */
4139         if (!(sb->s_flags & MS_RDONLY))
4140                 es->s_wtime = cpu_to_le32(get_seconds());
4141         if (sb->s_bdev->bd_part)
4142                 es->s_kbytes_written =
4143                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4144                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4145                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4146         else
4147                 es->s_kbytes_written =
4148                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4149         ext4_free_blocks_count_set(es,
4150                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4151                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4152         es->s_free_inodes_count =
4153                 cpu_to_le32(percpu_counter_sum_positive(
4154                                 &EXT4_SB(sb)->s_freeinodes_counter));
4155         sb->s_dirt = 0;
4156         BUFFER_TRACE(sbh, "marking dirty");
4157         mark_buffer_dirty(sbh);
4158         if (sync) {
4159                 error = sync_dirty_buffer(sbh);
4160                 if (error)
4161                         return error;
4162
4163                 error = buffer_write_io_error(sbh);
4164                 if (error) {
4165                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4166                                "superblock");
4167                         clear_buffer_write_io_error(sbh);
4168                         set_buffer_uptodate(sbh);
4169                 }
4170         }
4171         return error;
4172 }
4173
4174 /*
4175  * Have we just finished recovery?  If so, and if we are mounting (or
4176  * remounting) the filesystem readonly, then we will end up with a
4177  * consistent fs on disk.  Record that fact.
4178  */
4179 static void ext4_mark_recovery_complete(struct super_block *sb,
4180                                         struct ext4_super_block *es)
4181 {
4182         journal_t *journal = EXT4_SB(sb)->s_journal;
4183
4184         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4185                 BUG_ON(journal != NULL);
4186                 return;
4187         }
4188         jbd2_journal_lock_updates(journal);
4189         if (jbd2_journal_flush(journal) < 0)
4190                 goto out;
4191
4192         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4193             sb->s_flags & MS_RDONLY) {
4194                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4195                 ext4_commit_super(sb, 1);
4196         }
4197
4198 out:
4199         jbd2_journal_unlock_updates(journal);
4200 }
4201
4202 /*
4203  * If we are mounting (or read-write remounting) a filesystem whose journal
4204  * has recorded an error from a previous lifetime, move that error to the
4205  * main filesystem now.
4206  */
4207 static void ext4_clear_journal_err(struct super_block *sb,
4208                                    struct ext4_super_block *es)
4209 {
4210         journal_t *journal;
4211         int j_errno;
4212         const char *errstr;
4213
4214         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4215
4216         journal = EXT4_SB(sb)->s_journal;
4217
4218         /*
4219          * Now check for any error status which may have been recorded in the
4220          * journal by a prior ext4_error() or ext4_abort()
4221          */
4222
4223         j_errno = jbd2_journal_errno(journal);
4224         if (j_errno) {
4225                 char nbuf[16];
4226
4227                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4228                 ext4_warning(sb, "Filesystem error recorded "
4229                              "from previous mount: %s", errstr);
4230                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4231
4232                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4233                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4234                 ext4_commit_super(sb, 1);
4235
4236                 jbd2_journal_clear_err(journal);
4237                 jbd2_journal_update_sb_errno(journal);
4238         }
4239 }
4240
4241 /*
4242  * Force the running and committing transactions to commit,
4243  * and wait on the commit.
4244  */
4245 int ext4_force_commit(struct super_block *sb)
4246 {
4247         journal_t *journal;
4248         int ret = 0;
4249
4250         if (sb->s_flags & MS_RDONLY)
4251                 return 0;
4252
4253         journal = EXT4_SB(sb)->s_journal;
4254         if (journal) {
4255                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4256                 ret = ext4_journal_force_commit(journal);
4257         }
4258
4259         return ret;
4260 }
4261
4262 static void ext4_write_super(struct super_block *sb)
4263 {
4264         lock_super(sb);
4265         ext4_commit_super(sb, 1);
4266         unlock_super(sb);
4267 }
4268
4269 static int ext4_sync_fs(struct super_block *sb, int wait)
4270 {
4271         int ret = 0;
4272         tid_t target;
4273         struct ext4_sb_info *sbi = EXT4_SB(sb);
4274
4275         trace_ext4_sync_fs(sb, wait);
4276         flush_workqueue(sbi->dio_unwritten_wq);
4277         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4278                 if (wait)
4279                         jbd2_log_wait_commit(sbi->s_journal, target);
4280         }
4281         return ret;
4282 }
4283
4284 /*
4285  * LVM calls this function before a (read-only) snapshot is created.  This
4286  * gives us a chance to flush the journal completely and mark the fs clean.
4287  *
4288  * Note that only this function cannot bring a filesystem to be in a clean
4289  * state independently, because ext4 prevents a new handle from being started
4290  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4291  * the upper layer.
4292  */
4293 static int ext4_freeze(struct super_block *sb)
4294 {
4295         int error = 0;
4296         journal_t *journal;
4297
4298         if (sb->s_flags & MS_RDONLY)
4299                 return 0;
4300
4301         journal = EXT4_SB(sb)->s_journal;
4302
4303         /* Now we set up the journal barrier. */
4304         jbd2_journal_lock_updates(journal);
4305
4306         /*
4307          * Don't clear the needs_recovery flag if we failed to flush
4308          * the journal.
4309          */
4310         error = jbd2_journal_flush(journal);
4311         if (error < 0)
4312                 goto out;
4313
4314         /* Journal blocked and flushed, clear needs_recovery flag. */
4315         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4316         error = ext4_commit_super(sb, 1);
4317 out:
4318         /* we rely on s_frozen to stop further updates */
4319         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4320         return error;
4321 }
4322
4323 /*
4324  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4325  * flag here, even though the filesystem is not technically dirty yet.
4326  */
4327 static int ext4_unfreeze(struct super_block *sb)
4328 {
4329         if (sb->s_flags & MS_RDONLY)
4330                 return 0;
4331
4332         lock_super(sb);
4333         /* Reset the needs_recovery flag before the fs is unlocked. */
4334         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4335         ext4_commit_super(sb, 1);
4336         unlock_super(sb);
4337         return 0;
4338 }
4339
4340 /*
4341  * Structure to save mount options for ext4_remount's benefit
4342  */
4343 struct ext4_mount_options {
4344         unsigned long s_mount_opt;
4345         unsigned long s_mount_opt2;
4346         uid_t s_resuid;
4347         gid_t s_resgid;
4348         unsigned long s_commit_interval;
4349         u32 s_min_batch_time, s_max_batch_time;
4350 #ifdef CONFIG_QUOTA
4351         int s_jquota_fmt;
4352         char *s_qf_names[MAXQUOTAS];
4353 #endif
4354 };
4355
4356 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4357 {
4358         struct ext4_super_block *es;
4359         struct ext4_sb_info *sbi = EXT4_SB(sb);
4360         unsigned long old_sb_flags;
4361         struct ext4_mount_options old_opts;
4362         int enable_quota = 0;
4363         ext4_group_t g;
4364         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4365         int err = 0;
4366 #ifdef CONFIG_QUOTA
4367         int i;
4368 #endif
4369         char *orig_data = kstrdup(data, GFP_KERNEL);
4370
4371         /* Store the original options */
4372         lock_super(sb);
4373         old_sb_flags = sb->s_flags;
4374         old_opts.s_mount_opt = sbi->s_mount_opt;
4375         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4376         old_opts.s_resuid = sbi->s_resuid;
4377         old_opts.s_resgid = sbi->s_resgid;
4378         old_opts.s_commit_interval = sbi->s_commit_interval;
4379         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4380         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4381 #ifdef CONFIG_QUOTA
4382         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4383         for (i = 0; i < MAXQUOTAS; i++)
4384                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4385 #endif
4386         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4387                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4388
4389         /*
4390          * Allow the "check" option to be passed as a remount option.
4391          */
4392         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4393                 err = -EINVAL;
4394                 goto restore_opts;
4395         }
4396
4397         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4398                 ext4_abort(sb, "Abort forced by user");
4399
4400         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4401                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4402
4403         es = sbi->s_es;
4404
4405         if (sbi->s_journal) {
4406                 ext4_init_journal_params(sb, sbi->s_journal);
4407                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4408         }
4409
4410         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4411                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4412                         err = -EROFS;
4413                         goto restore_opts;
4414                 }
4415
4416                 if (*flags & MS_RDONLY) {
4417                         err = dquot_suspend(sb, -1);
4418                         if (err < 0)
4419                                 goto restore_opts;
4420
4421                         /*
4422                          * First of all, the unconditional stuff we have to do
4423                          * to disable replay of the journal when we next remount
4424                          */
4425                         sb->s_flags |= MS_RDONLY;
4426
4427                         /*
4428                          * OK, test if we are remounting a valid rw partition
4429                          * readonly, and if so set the rdonly flag and then
4430                          * mark the partition as valid again.
4431                          */
4432                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4433                             (sbi->s_mount_state & EXT4_VALID_FS))
4434                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4435
4436                         if (sbi->s_journal)
4437                                 ext4_mark_recovery_complete(sb, es);
4438                 } else {
4439                         /* Make sure we can mount this feature set readwrite */
4440                         if (!ext4_feature_set_ok(sb, 0)) {
4441                                 err = -EROFS;
4442                                 goto restore_opts;
4443                         }
4444                         /*
4445                          * Make sure the group descriptor checksums
4446                          * are sane.  If they aren't, refuse to remount r/w.
4447                          */
4448                         for (g = 0; g < sbi->s_groups_count; g++) {
4449                                 struct ext4_group_desc *gdp =
4450                                         ext4_get_group_desc(sb, g, NULL);
4451
4452                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4453                                         ext4_msg(sb, KERN_ERR,
4454                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4455                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4456                                                le16_to_cpu(gdp->bg_checksum));
4457                                         err = -EINVAL;
4458                                         goto restore_opts;
4459                                 }
4460                         }
4461
4462                         /*
4463                          * If we have an unprocessed orphan list hanging
4464                          * around from a previously readonly bdev mount,
4465                          * require a full umount/remount for now.
4466                          */
4467                         if (es->s_last_orphan) {
4468                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4469                                        "remount RDWR because of unprocessed "
4470                                        "orphan inode list.  Please "
4471                                        "umount/remount instead");
4472                                 err = -EINVAL;
4473                                 goto restore_opts;
4474                         }
4475
4476                         /*
4477                          * Mounting a RDONLY partition read-write, so reread
4478                          * and store the current valid flag.  (It may have
4479                          * been changed by e2fsck since we originally mounted
4480                          * the partition.)
4481                          */
4482                         if (sbi->s_journal)
4483                                 ext4_clear_journal_err(sb, es);
4484                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4485                         if (!ext4_setup_super(sb, es, 0))
4486                                 sb->s_flags &= ~MS_RDONLY;
4487                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4488                                                      EXT4_FEATURE_INCOMPAT_MMP))
4489                                 if (ext4_multi_mount_protect(sb,
4490                                                 le64_to_cpu(es->s_mmp_block))) {
4491                                         err = -EROFS;
4492                                         goto restore_opts;
4493                                 }
4494                         enable_quota = 1;
4495                 }
4496         }
4497
4498         /*
4499          * Reinitialize lazy itable initialization thread based on
4500          * current settings
4501          */
4502         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4503                 ext4_unregister_li_request(sb);
4504         else {
4505                 ext4_group_t first_not_zeroed;
4506                 first_not_zeroed = ext4_has_uninit_itable(sb);
4507                 ext4_register_li_request(sb, first_not_zeroed);
4508         }
4509
4510         ext4_setup_system_zone(sb);
4511         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4512                 ext4_commit_super(sb, 1);
4513
4514 #ifdef CONFIG_QUOTA
4515         /* Release old quota file names */
4516         for (i = 0; i < MAXQUOTAS; i++)
4517                 if (old_opts.s_qf_names[i] &&
4518                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4519                         kfree(old_opts.s_qf_names[i]);
4520 #endif
4521         unlock_super(sb);
4522         if (enable_quota)
4523                 dquot_resume(sb, -1);
4524
4525         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4526         kfree(orig_data);
4527         return 0;
4528
4529 restore_opts:
4530         sb->s_flags = old_sb_flags;
4531         sbi->s_mount_opt = old_opts.s_mount_opt;
4532         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4533         sbi->s_resuid = old_opts.s_resuid;
4534         sbi->s_resgid = old_opts.s_resgid;
4535         sbi->s_commit_interval = old_opts.s_commit_interval;
4536         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4537         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4538 #ifdef CONFIG_QUOTA
4539         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4540         for (i = 0; i < MAXQUOTAS; i++) {
4541                 if (sbi->s_qf_names[i] &&
4542                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4543                         kfree(sbi->s_qf_names[i]);
4544                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4545         }
4546 #endif
4547         unlock_super(sb);
4548         kfree(orig_data);
4549         return err;
4550 }
4551
4552 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4553 {
4554         struct super_block *sb = dentry->d_sb;
4555         struct ext4_sb_info *sbi = EXT4_SB(sb);
4556         struct ext4_super_block *es = sbi->s_es;
4557         ext4_fsblk_t overhead = 0;
4558         u64 fsid;
4559         s64 bfree;
4560
4561         if (!test_opt(sb, MINIX_DF))
4562                 overhead = sbi->s_overhead;
4563
4564         buf->f_type = EXT4_SUPER_MAGIC;
4565         buf->f_bsize = sb->s_blocksize;
4566         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4567         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4568                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4569         /* prevent underflow in case that few free space is available */
4570         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4571         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4572         if (buf->f_bfree < ext4_r_blocks_count(es))
4573                 buf->f_bavail = 0;
4574         buf->f_files = le32_to_cpu(es->s_inodes_count);
4575         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4576         buf->f_namelen = EXT4_NAME_LEN;
4577         fsid = le64_to_cpup((void *)es->s_uuid) ^
4578                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4579         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4580         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4581
4582         return 0;
4583 }
4584
4585 /* Helper function for writing quotas on sync - we need to start transaction
4586  * before quota file is locked for write. Otherwise the are possible deadlocks:
4587  * Process 1                         Process 2
4588  * ext4_create()                     quota_sync()
4589  *   jbd2_journal_start()                  write_dquot()
4590  *   dquot_initialize()                         down(dqio_mutex)
4591  *     down(dqio_mutex)                    jbd2_journal_start()
4592  *
4593  */
4594
4595 #ifdef CONFIG_QUOTA
4596
4597 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4598 {
4599         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4600 }
4601
4602 static int ext4_write_dquot(struct dquot *dquot)
4603 {
4604         int ret, err;
4605         handle_t *handle;
4606         struct inode *inode;
4607
4608         inode = dquot_to_inode(dquot);
4609         handle = ext4_journal_start(inode,
4610                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4611         if (IS_ERR(handle))
4612                 return PTR_ERR(handle);
4613         ret = dquot_commit(dquot);
4614         err = ext4_journal_stop(handle);
4615         if (!ret)
4616                 ret = err;
4617         return ret;
4618 }
4619
4620 static int ext4_acquire_dquot(struct dquot *dquot)
4621 {
4622         int ret, err;
4623         handle_t *handle;
4624
4625         handle = ext4_journal_start(dquot_to_inode(dquot),
4626                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4627         if (IS_ERR(handle))
4628                 return PTR_ERR(handle);
4629         ret = dquot_acquire(dquot);
4630         err = ext4_journal_stop(handle);
4631         if (!ret)
4632                 ret = err;
4633         return ret;
4634 }
4635
4636 static int ext4_release_dquot(struct dquot *dquot)
4637 {
4638         int ret, err;
4639         handle_t *handle;
4640
4641         handle = ext4_journal_start(dquot_to_inode(dquot),
4642                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4643         if (IS_ERR(handle)) {
4644                 /* Release dquot anyway to avoid endless cycle in dqput() */
4645                 dquot_release(dquot);
4646                 return PTR_ERR(handle);
4647         }
4648         ret = dquot_release(dquot);
4649         err = ext4_journal_stop(handle);
4650         if (!ret)
4651                 ret = err;
4652         return ret;
4653 }
4654
4655 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4656 {
4657         /* Are we journaling quotas? */
4658         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4659             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4660                 dquot_mark_dquot_dirty(dquot);
4661                 return ext4_write_dquot(dquot);
4662         } else {
4663                 return dquot_mark_dquot_dirty(dquot);
4664         }
4665 }
4666
4667 static int ext4_write_info(struct super_block *sb, int type)
4668 {
4669         int ret, err;
4670         handle_t *handle;
4671
4672         /* Data block + inode block */
4673         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4674         if (IS_ERR(handle))
4675                 return PTR_ERR(handle);
4676         ret = dquot_commit_info(sb, type);
4677         err = ext4_journal_stop(handle);
4678         if (!ret)
4679                 ret = err;
4680         return ret;
4681 }
4682
4683 /*
4684  * Turn on quotas during mount time - we need to find
4685  * the quota file and such...
4686  */
4687 static int ext4_quota_on_mount(struct super_block *sb, int type)
4688 {
4689         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4690                                         EXT4_SB(sb)->s_jquota_fmt, type);
4691 }
4692
4693 /*
4694  * Standard function to be called on quota_on
4695  */
4696 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4697                          struct path *path)
4698 {
4699         int err;
4700
4701         if (!test_opt(sb, QUOTA))
4702                 return -EINVAL;
4703
4704         /* Quotafile not on the same filesystem? */
4705         if (path->dentry->d_sb != sb)
4706                 return -EXDEV;
4707         /* Journaling quota? */
4708         if (EXT4_SB(sb)->s_qf_names[type]) {
4709                 /* Quotafile not in fs root? */
4710                 if (path->dentry->d_parent != sb->s_root)
4711                         ext4_msg(sb, KERN_WARNING,
4712                                 "Quota file not on filesystem root. "
4713                                 "Journaled quota will not work");
4714         }
4715
4716         /*
4717          * When we journal data on quota file, we have to flush journal to see
4718          * all updates to the file when we bypass pagecache...
4719          */
4720         if (EXT4_SB(sb)->s_journal &&
4721             ext4_should_journal_data(path->dentry->d_inode)) {
4722                 /*
4723                  * We don't need to lock updates but journal_flush() could
4724                  * otherwise be livelocked...
4725                  */
4726                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4727                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4728                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4729                 if (err)
4730                         return err;
4731         }
4732
4733         return dquot_quota_on(sb, type, format_id, path);
4734 }
4735
4736 static int ext4_quota_off(struct super_block *sb, int type)
4737 {
4738         struct inode *inode = sb_dqopt(sb)->files[type];
4739         handle_t *handle;
4740
4741         /* Force all delayed allocation blocks to be allocated.
4742          * Caller already holds s_umount sem */
4743         if (test_opt(sb, DELALLOC))
4744                 sync_filesystem(sb);
4745
4746         if (!inode)
4747                 goto out;
4748
4749         /* Update modification times of quota files when userspace can
4750          * start looking at them */
4751         handle = ext4_journal_start(inode, 1);
4752         if (IS_ERR(handle))
4753                 goto out;
4754         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4755         ext4_mark_inode_dirty(handle, inode);
4756         ext4_journal_stop(handle);
4757
4758 out:
4759         return dquot_quota_off(sb, type);
4760 }
4761
4762 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4763  * acquiring the locks... As quota files are never truncated and quota code
4764  * itself serializes the operations (and no one else should touch the files)
4765  * we don't have to be afraid of races */
4766 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4767                                size_t len, loff_t off)
4768 {
4769         struct inode *inode = sb_dqopt(sb)->files[type];
4770         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4771         int err = 0;
4772         int offset = off & (sb->s_blocksize - 1);
4773         int tocopy;
4774         size_t toread;
4775         struct buffer_head *bh;
4776         loff_t i_size = i_size_read(inode);
4777
4778         if (off > i_size)
4779                 return 0;
4780         if (off+len > i_size)
4781                 len = i_size-off;
4782         toread = len;
4783         while (toread > 0) {
4784                 tocopy = sb->s_blocksize - offset < toread ?
4785                                 sb->s_blocksize - offset : toread;
4786                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4787                 if (err)
4788                         return err;
4789                 if (!bh)        /* A hole? */
4790                         memset(data, 0, tocopy);
4791                 else
4792                         memcpy(data, bh->b_data+offset, tocopy);
4793                 brelse(bh);
4794                 offset = 0;
4795                 toread -= tocopy;
4796                 data += tocopy;
4797                 blk++;
4798         }
4799         return len;
4800 }
4801
4802 /* Write to quotafile (we know the transaction is already started and has
4803  * enough credits) */
4804 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4805                                 const char *data, size_t len, loff_t off)
4806 {
4807         struct inode *inode = sb_dqopt(sb)->files[type];
4808         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4809         int err = 0;
4810         int offset = off & (sb->s_blocksize - 1);
4811         struct buffer_head *bh;
4812         handle_t *handle = journal_current_handle();
4813
4814         if (EXT4_SB(sb)->s_journal && !handle) {
4815                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4816                         " cancelled because transaction is not started",
4817                         (unsigned long long)off, (unsigned long long)len);
4818                 return -EIO;
4819         }
4820         /*
4821          * Since we account only one data block in transaction credits,
4822          * then it is impossible to cross a block boundary.
4823          */
4824         if (sb->s_blocksize - offset < len) {
4825                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4826                         " cancelled because not block aligned",
4827                         (unsigned long long)off, (unsigned long long)len);
4828                 return -EIO;
4829         }
4830
4831         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4832         bh = ext4_bread(handle, inode, blk, 1, &err);
4833         if (!bh)
4834                 goto out;
4835         err = ext4_journal_get_write_access(handle, bh);
4836         if (err) {
4837                 brelse(bh);
4838                 goto out;
4839         }
4840         lock_buffer(bh);
4841         memcpy(bh->b_data+offset, data, len);
4842         flush_dcache_page(bh->b_page);
4843         unlock_buffer(bh);
4844         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4845         brelse(bh);
4846 out:
4847         if (err) {
4848                 mutex_unlock(&inode->i_mutex);
4849                 return err;
4850         }
4851         if (inode->i_size < off + len) {
4852                 i_size_write(inode, off + len);
4853                 EXT4_I(inode)->i_disksize = inode->i_size;
4854                 ext4_mark_inode_dirty(handle, inode);
4855         }
4856         mutex_unlock(&inode->i_mutex);
4857         return len;
4858 }
4859
4860 #endif
4861
4862 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4863                        const char *dev_name, void *data)
4864 {
4865         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4866 }
4867
4868 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4869 static inline void register_as_ext2(void)
4870 {
4871         int err = register_filesystem(&ext2_fs_type);
4872         if (err)
4873                 printk(KERN_WARNING
4874                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4875 }
4876
4877 static inline void unregister_as_ext2(void)
4878 {
4879         unregister_filesystem(&ext2_fs_type);
4880 }
4881
4882 static inline int ext2_feature_set_ok(struct super_block *sb)
4883 {
4884         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4885                 return 0;
4886         if (sb->s_flags & MS_RDONLY)
4887                 return 1;
4888         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4889                 return 0;
4890         return 1;
4891 }
4892 MODULE_ALIAS("ext2");
4893 #else
4894 static inline void register_as_ext2(void) { }
4895 static inline void unregister_as_ext2(void) { }
4896 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4897 #endif
4898
4899 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4900 static inline void register_as_ext3(void)
4901 {
4902         int err = register_filesystem(&ext3_fs_type);
4903         if (err)
4904                 printk(KERN_WARNING
4905                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4906 }
4907
4908 static inline void unregister_as_ext3(void)
4909 {
4910         unregister_filesystem(&ext3_fs_type);
4911 }
4912
4913 static inline int ext3_feature_set_ok(struct super_block *sb)
4914 {
4915         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4916                 return 0;
4917         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4918                 return 0;
4919         if (sb->s_flags & MS_RDONLY)
4920                 return 1;
4921         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4922                 return 0;
4923         return 1;
4924 }
4925 MODULE_ALIAS("ext3");
4926 #else
4927 static inline void register_as_ext3(void) { }
4928 static inline void unregister_as_ext3(void) { }
4929 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4930 #endif
4931
4932 static struct file_system_type ext4_fs_type = {
4933         .owner          = THIS_MODULE,
4934         .name           = "ext4",
4935         .mount          = ext4_mount,
4936         .kill_sb        = kill_block_super,
4937         .fs_flags       = FS_REQUIRES_DEV,
4938 };
4939
4940 static int __init ext4_init_feat_adverts(void)
4941 {
4942         struct ext4_features *ef;
4943         int ret = -ENOMEM;
4944
4945         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4946         if (!ef)
4947                 goto out;
4948
4949         ef->f_kobj.kset = ext4_kset;
4950         init_completion(&ef->f_kobj_unregister);
4951         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4952                                    "features");
4953         if (ret) {
4954                 kfree(ef);
4955                 goto out;
4956         }
4957
4958         ext4_feat = ef;
4959         ret = 0;
4960 out:
4961         return ret;
4962 }
4963
4964 static void ext4_exit_feat_adverts(void)
4965 {
4966         kobject_put(&ext4_feat->f_kobj);
4967         wait_for_completion(&ext4_feat->f_kobj_unregister);
4968         kfree(ext4_feat);
4969 }
4970
4971 /* Shared across all ext4 file systems */
4972 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4973 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4974
4975 static int __init ext4_init_fs(void)
4976 {
4977         int i, err;
4978
4979         ext4_li_info = NULL;
4980         mutex_init(&ext4_li_mtx);
4981
4982         ext4_check_flag_values();
4983
4984         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4985                 mutex_init(&ext4__aio_mutex[i]);
4986                 init_waitqueue_head(&ext4__ioend_wq[i]);
4987         }
4988
4989         err = ext4_init_pageio();
4990         if (err)
4991                 return err;
4992         err = ext4_init_system_zone();
4993         if (err)
4994                 goto out6;
4995         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4996         if (!ext4_kset)
4997                 goto out5;
4998         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4999
5000         err = ext4_init_feat_adverts();
5001         if (err)
5002                 goto out4;
5003
5004         err = ext4_init_mballoc();
5005         if (err)
5006                 goto out3;
5007
5008         err = ext4_init_xattr();
5009         if (err)
5010                 goto out2;
5011         err = init_inodecache();
5012         if (err)
5013                 goto out1;
5014         register_as_ext3();
5015         register_as_ext2();
5016         err = register_filesystem(&ext4_fs_type);
5017         if (err)
5018                 goto out;
5019
5020         return 0;
5021 out:
5022         unregister_as_ext2();
5023         unregister_as_ext3();
5024         destroy_inodecache();
5025 out1:
5026         ext4_exit_xattr();
5027 out2:
5028         ext4_exit_mballoc();
5029 out3:
5030         ext4_exit_feat_adverts();
5031 out4:
5032         if (ext4_proc_root)
5033                 remove_proc_entry("fs/ext4", NULL);
5034         kset_unregister(ext4_kset);
5035 out5:
5036         ext4_exit_system_zone();
5037 out6:
5038         ext4_exit_pageio();
5039         return err;
5040 }
5041
5042 static void __exit ext4_exit_fs(void)
5043 {
5044         ext4_destroy_lazyinit_thread();
5045         unregister_as_ext2();
5046         unregister_as_ext3();
5047         unregister_filesystem(&ext4_fs_type);
5048         destroy_inodecache();
5049         ext4_exit_xattr();
5050         ext4_exit_mballoc();
5051         ext4_exit_feat_adverts();
5052         remove_proc_entry("fs/ext4", NULL);
5053         kset_unregister(ext4_kset);
5054         ext4_exit_system_zone();
5055         ext4_exit_pageio();
5056 }
5057
5058 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5059 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5060 MODULE_LICENSE("GPL");
5061 module_init(ext4_init_fs)
5062 module_exit(ext4_exit_fs)