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_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 435 {
 436         struct super_block              *sb = journal->j_private;
 437         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 438         int                             error = is_journal_aborted(journal);
 439         struct ext4_journal_cb_entry    *jce, *tmp;
 440
 441         spin_lock(&sbi->s_md_lock);
 442         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 443                 list_del_init(&jce->jce_list);
 444                 spin_unlock(&sbi->s_md_lock);
 445                 jce->jce_func(sb, jce, error);
 446                 spin_lock(&sbi->s_md_lock);
 447         }
 448         spin_unlock(&sbi->s_md_lock);
 449 }
 450
 451 /* Deal with the reporting of failure conditions on a filesystem such as
 452  * inconsistencies detected or read IO failures.
 453  *
 454  * On ext2, we can store the error state of the filesystem in the
 455  * superblock.  That is not possible on ext4, because we may have other
 456  * write ordering constraints on the superblock which prevent us from
 457  * writing it out straight away; and given that the journal is about to
 458  * be aborted, we can't rely on the current, or future, transactions to
 459  * write out the superblock safely.
 460  *
 461  * We'll just use the jbd2_journal_abort() error code to record an error in
 462  * the journal instead.  On recovery, the journal will complain about
 463  * that error until we've noted it down and cleared it.
 464  */
 465
 466 static void ext4_handle_error(struct super_block *sb)
 467 {
 468         if (sb->s_flags & MS_RDONLY)
 469                 return;
 470
 471         if (!test_opt(sb, ERRORS_CONT)) {
 472                 journal_t *journal = EXT4_SB(sb)->s_journal;
 473
 474                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 475                 if (journal)
 476                         jbd2_journal_abort(journal, -EIO);
 477         }
 478         if (test_opt(sb, ERRORS_RO)) {
 479                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 480                 sb->s_flags |= MS_RDONLY;
 481         }
 482         if (test_opt(sb, ERRORS_PANIC))
 483                 panic("EXT4-fs (device %s): panic forced after error\n",
 484                         sb->s_id);
 485 }
 486
 487 void __ext4_error(struct super_block *sb, const char *function,
 488                   unsigned int line, const char *fmt, ...)
 489 {
 490         struct va_format vaf;
 491         va_list args;
 492
 493         va_start(args, fmt);
 494         vaf.fmt = fmt;
 495         vaf.va = &args;
 496         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 497                sb->s_id, function, line, current->comm, &vaf);
 498         va_end(args);
 499
 500         ext4_handle_error(sb);
 501 }
 502
 503 void ext4_error_inode(struct inode *inode, const char *function,
 504                       unsigned int line, ext4_fsblk_t block,
 505                       const char *fmt, ...)
 506 {
 507         va_list args;
 508         struct va_format vaf;
 509         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 510
 511         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 512         es->s_last_error_block = cpu_to_le64(block);
 513         save_error_info(inode->i_sb, function, line);
 514         va_start(args, fmt);
 515         vaf.fmt = fmt;
 516         vaf.va = &args;
 517         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 518                inode->i_sb->s_id, function, line, inode->i_ino);
 519         if (block)
 520                 printk(KERN_CONT "block %llu: ", block);
 521         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 522         va_end(args);
 523
 524         ext4_handle_error(inode->i_sb);
 525 }
 526
 527 void ext4_error_file(struct file *file, const char *function,
 528                      unsigned int line, ext4_fsblk_t block,
 529                      const char *fmt, ...)
 530 {
 531         va_list args;
 532         struct va_format vaf;
 533         struct ext4_super_block *es;
 534         struct inode *inode = file->f_dentry->d_inode;
 535         char pathname[80], *path;
 536
 537         es = EXT4_SB(inode->i_sb)->s_es;
 538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 539         save_error_info(inode->i_sb, function, line);
 540         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 541         if (IS_ERR(path))
 542                 path = "(unknown)";
 543         printk(KERN_CRIT
 544                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 545                inode->i_sb->s_id, function, line, inode->i_ino);
 546         if (block)
 547                 printk(KERN_CONT "block %llu: ", block);
 548         va_start(args, fmt);
 549         vaf.fmt = fmt;
 550         vaf.va = &args;
 551         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 552         va_end(args);
 553
 554         ext4_handle_error(inode->i_sb);
 555 }
 556
 557 static const char *ext4_decode_error(struct super_block *sb, int errno,
 558                                      char nbuf[16])
 559 {
 560         char *errstr = NULL;
 561
 562         switch (errno) {
 563         case -EIO:
 564                 errstr = "IO failure";
 565                 break;
 566         case -ENOMEM:
 567                 errstr = "Out of memory";
 568                 break;
 569         case -EROFS:
 570                 if (!sb || (EXT4_SB(sb)->s_journal &&
 571                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 572                         errstr = "Journal has aborted";
 573                 else
 574                         errstr = "Readonly filesystem";
 575                 break;
 576         default:
 577                 /* If the caller passed in an extra buffer for unknown
 578                  * errors, textualise them now.  Else we just return
 579                  * NULL. */
 580                 if (nbuf) {
 581                         /* Check for truncated error codes... */
 582                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 583                                 errstr = nbuf;
 584                 }
 585                 break;
 586         }
 587
 588         return errstr;
 589 }
 590
 591 /* __ext4_std_error decodes expected errors from journaling functions
 592  * automatically and invokes the appropriate error response.  */
 593
 594 void __ext4_std_error(struct super_block *sb, const char *function,
 595                       unsigned int line, int errno)
 596 {
 597         char nbuf[16];
 598         const char *errstr;
 599
 600         /* Special case: if the error is EROFS, and we're not already
 601          * inside a transaction, then there's really no point in logging
 602          * an error. */
 603         if (errno == -EROFS && journal_current_handle() == NULL &&
 604             (sb->s_flags & MS_RDONLY))
 605                 return;
 606
 607         errstr = ext4_decode_error(sb, errno, nbuf);
 608         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 609                sb->s_id, function, line, errstr);
 610         save_error_info(sb, function, line);
 611
 612         ext4_handle_error(sb);
 613 }
 614
 615 /*
 616  * ext4_abort is a much stronger failure handler than ext4_error.  The
 617  * abort function may be used to deal with unrecoverable failures such
 618  * as journal IO errors or ENOMEM at a critical moment in log management.
 619  *
 620  * We unconditionally force the filesystem into an ABORT|READONLY state,
 621  * unless the error response on the fs has been set to panic in which
 622  * case we take the easy way out and panic immediately.
 623  */
 624
 625 void __ext4_abort(struct super_block *sb, const char *function,
 626                 unsigned int line, const char *fmt, ...)
 627 {
 628         va_list args;
 629
 630         save_error_info(sb, function, line);
 631         va_start(args, fmt);
 632         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 633                function, line);
 634         vprintk(fmt, args);
 635         printk("\n");
 636         va_end(args);
 637
 638         if ((sb->s_flags & MS_RDONLY) == 0) {
 639                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 640                 sb->s_flags |= MS_RDONLY;
 641                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 642                 if (EXT4_SB(sb)->s_journal)
 643                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 644                 save_error_info(sb, function, line);
 645         }
 646         if (test_opt(sb, ERRORS_PANIC))
 647                 panic("EXT4-fs panic from previous error\n");
 648 }
 649
 650 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         va_start(args, fmt);
 656         vaf.fmt = fmt;
 657         vaf.va = &args;
 658         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 659         va_end(args);
 660 }
 661
 662 void __ext4_warning(struct super_block *sb, const char *function,
 663                     unsigned int line, 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(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 672                sb->s_id, function, line, &vaf);
 673         va_end(args);
 674 }
 675
 676 void __ext4_grp_locked_error(const char *function, unsigned int line,
 677                              struct super_block *sb, ext4_group_t grp,
 678                              unsigned long ino, ext4_fsblk_t block,
 679                              const char *fmt, ...)
 680 __releases(bitlock)
 681 __acquires(bitlock)
 682 {
 683         struct va_format vaf;
 684         va_list args;
 685         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 686
 687         es->s_last_error_ino = cpu_to_le32(ino);
 688         es->s_last_error_block = cpu_to_le64(block);
 689         __save_error_info(sb, function, line);
 690
 691         va_start(args, fmt);
 692
 693         vaf.fmt = fmt;
 694         vaf.va = &args;
 695         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 696                sb->s_id, function, line, grp);
 697         if (ino)
 698                 printk(KERN_CONT "inode %lu: ", ino);
 699         if (block)
 700                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 701         printk(KERN_CONT "%pV\n", &vaf);
 702         va_end(args);
 703
 704         if (test_opt(sb, ERRORS_CONT)) {
 705                 ext4_commit_super(sb, 0);
 706                 return;
 707         }
 708
 709         ext4_unlock_group(sb, grp);
 710         ext4_handle_error(sb);
 711         /*
 712          * We only get here in the ERRORS_RO case; relocking the group
 713          * may be dangerous, but nothing bad will happen since the
 714          * filesystem will have already been marked read/only and the
 715          * journal has been aborted.  We return 1 as a hint to callers
 716          * who might what to use the return value from
 717          * ext4_grp_locked_error() to distinguish between the
 718          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 719          * aggressively from the ext4 function in question, with a
 720          * more appropriate error code.
 721          */
 722         ext4_lock_group(sb, grp);
 723         return;
 724 }
 725
 726 void ext4_update_dynamic_rev(struct super_block *sb)
 727 {
 728         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 729
 730         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 731                 return;
 732
 733         ext4_warning(sb,
 734                      "updating to rev %d because of new feature flag, "
 735                      "running e2fsck is recommended",
 736                      EXT4_DYNAMIC_REV);
 737
 738         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 739         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 740         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 741         /* leave es->s_feature_*compat flags alone */
 742         /* es->s_uuid will be set by e2fsck if empty */
 743
 744         /*
 745          * The rest of the superblock fields should be zero, and if not it
 746          * means they are likely already in use, so leave them alone.  We
 747          * can leave it up to e2fsck to clean up any inconsistencies there.
 748          */
 749 }
 750
 751 /*
 752  * Open the external journal device
 753  */
 754 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 755 {
 756         struct block_device *bdev;
 757         char b[BDEVNAME_SIZE];
 758
 759         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 760         if (IS_ERR(bdev))
 761                 goto fail;
 762         return bdev;
 763
 764 fail:
 765         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 766                         __bdevname(dev, b), PTR_ERR(bdev));
 767         return NULL;
 768 }
 769
 770 /*
 771  * Release the journal device
 772  */
 773 static int ext4_blkdev_put(struct block_device *bdev)
 774 {
 775         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 776 }
 777
 778 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 779 {
 780         struct block_device *bdev;
 781         int ret = -ENODEV;
 782
 783         bdev = sbi->journal_bdev;
 784         if (bdev) {
 785                 ret = ext4_blkdev_put(bdev);
 786                 sbi->journal_bdev = NULL;
 787         }
 788         return ret;
 789 }
 790
 791 static inline struct inode *orphan_list_entry(struct list_head *l)
 792 {
 793         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 794 }
 795
 796 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 797 {
 798         struct list_head *l;
 799
 800         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 801                  le32_to_cpu(sbi->s_es->s_last_orphan));
 802
 803         printk(KERN_ERR "sb_info orphan list:\n");
 804         list_for_each(l, &sbi->s_orphan) {
 805                 struct inode *inode = orphan_list_entry(l);
 806                 printk(KERN_ERR "  "
 807                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 808                        inode->i_sb->s_id, inode->i_ino, inode,
 809                        inode->i_mode, inode->i_nlink,
 810                        NEXT_ORPHAN(inode));
 811         }
 812 }
 813
 814 static void ext4_put_super(struct super_block *sb)
 815 {
 816         struct ext4_sb_info *sbi = EXT4_SB(sb);
 817         struct ext4_super_block *es = sbi->s_es;
 818         int i, err;
 819
 820         ext4_unregister_li_request(sb);
 821         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 822
 823         flush_workqueue(sbi->dio_unwritten_wq);
 824         destroy_workqueue(sbi->dio_unwritten_wq);
 825
 826         lock_super(sb);
 827         if (sb->s_dirt)
 828                 ext4_commit_super(sb, 1);
 829
 830         if (sbi->s_journal) {
 831                 err = jbd2_journal_destroy(sbi->s_journal);
 832                 sbi->s_journal = NULL;
 833                 if (err < 0)
 834                         ext4_abort(sb, "Couldn't clean up the journal");
 835         }
 836
 837         del_timer(&sbi->s_err_report);
 838         ext4_release_system_zone(sb);
 839         ext4_mb_release(sb);
 840         ext4_ext_release(sb);
 841         ext4_xattr_put_super(sb);
 842
 843         if (!(sb->s_flags & MS_RDONLY)) {
 844                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 845                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 846                 ext4_commit_super(sb, 1);
 847         }
 848         if (sbi->s_proc) {
 849                 remove_proc_entry(sb->s_id, ext4_proc_root);
 850         }
 851         kobject_del(&sbi->s_kobj);
 852
 853         for (i = 0; i < sbi->s_gdb_count; i++)
 854                 brelse(sbi->s_group_desc[i]);
 855         ext4_kvfree(sbi->s_group_desc);
 856         ext4_kvfree(sbi->s_flex_groups);
 857         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 858         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 859         percpu_counter_destroy(&sbi->s_dirs_counter);
 860         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 861         brelse(sbi->s_sbh);
 862 #ifdef CONFIG_QUOTA
 863         for (i = 0; i < MAXQUOTAS; i++)
 864                 kfree(sbi->s_qf_names[i]);
 865 #endif
 866
 867         /* Debugging code just in case the in-memory inode orphan list
 868          * isn't empty.  The on-disk one can be non-empty if we've
 869          * detected an error and taken the fs readonly, but the
 870          * in-memory list had better be clean by this point. */
 871         if (!list_empty(&sbi->s_orphan))
 872                 dump_orphan_list(sb, sbi);
 873         J_ASSERT(list_empty(&sbi->s_orphan));
 874
 875         invalidate_bdev(sb->s_bdev);
 876         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 877                 /*
 878                  * Invalidate the journal device's buffers.  We don't want them
 879                  * floating about in memory - the physical journal device may
 880                  * hotswapped, and it breaks the `ro-after' testing code.
 881                  */
 882                 sync_blockdev(sbi->journal_bdev);
 883                 invalidate_bdev(sbi->journal_bdev);
 884                 ext4_blkdev_remove(sbi);
 885         }
 886         if (sbi->s_mmp_tsk)
 887                 kthread_stop(sbi->s_mmp_tsk);
 888         sb->s_fs_info = NULL;
 889         /*
 890          * Now that we are completely done shutting down the
 891          * superblock, we need to actually destroy the kobject.
 892          */
 893         unlock_super(sb);
 894         kobject_put(&sbi->s_kobj);
 895         wait_for_completion(&sbi->s_kobj_unregister);
 896         kfree(sbi->s_blockgroup_lock);
 897         kfree(sbi);
 898 }
 899
 900 static struct kmem_cache *ext4_inode_cachep;
 901
 902 /*
 903  * Called inside transaction, so use GFP_NOFS
 904  */
 905 static struct inode *ext4_alloc_inode(struct super_block *sb)
 906 {
 907         struct ext4_inode_info *ei;
 908
 909         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 910         if (!ei)
 911                 return NULL;
 912
 913         ei->vfs_inode.i_version = 1;
 914         ei->vfs_inode.i_data.writeback_index = 0;
 915         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 916         INIT_LIST_HEAD(&ei->i_prealloc_list);
 917         spin_lock_init(&ei->i_prealloc_lock);
 918         ei->i_reserved_data_blocks = 0;
 919         ei->i_reserved_meta_blocks = 0;
 920         ei->i_allocated_meta_blocks = 0;
 921         ei->i_da_metadata_calc_len = 0;
 922         spin_lock_init(&(ei->i_block_reservation_lock));
 923 #ifdef CONFIG_QUOTA
 924         ei->i_reserved_quota = 0;
 925 #endif
 926         ei->jinode = NULL;
 927         INIT_LIST_HEAD(&ei->i_completed_io_list);
 928         spin_lock_init(&ei->i_completed_io_lock);
 929         ei->cur_aio_dio = NULL;
 930         ei->i_sync_tid = 0;
 931         ei->i_datasync_tid = 0;
 932         atomic_set(&ei->i_ioend_count, 0);
 933         atomic_set(&ei->i_aiodio_unwritten, 0);
 934
 935         return &ei->vfs_inode;
 936 }
 937
 938 static int ext4_drop_inode(struct inode *inode)
 939 {
 940         int drop = generic_drop_inode(inode);
 941
 942         trace_ext4_drop_inode(inode, drop);
 943         return drop;
 944 }
 945
 946 static void ext4_i_callback(struct rcu_head *head)
 947 {
 948         struct inode *inode = container_of(head, struct inode, i_rcu);
 949         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 950 }
 951
 952 static void ext4_destroy_inode(struct inode *inode)
 953 {
 954         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 955                 ext4_msg(inode->i_sb, KERN_ERR,
 956                          "Inode %lu (%p): orphan list check failed!",
 957                          inode->i_ino, EXT4_I(inode));
 958                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 959                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 960                                 true);
 961                 dump_stack();
 962         }
 963         call_rcu(&inode->i_rcu, ext4_i_callback);
 964 }
 965
 966 static void init_once(void *foo)
 967 {
 968         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 969
 970         INIT_LIST_HEAD(&ei->i_orphan);
 971 #ifdef CONFIG_EXT4_FS_XATTR
 972         init_rwsem(&ei->xattr_sem);
 973 #endif
 974         init_rwsem(&ei->i_data_sem);
 975         inode_init_once(&ei->vfs_inode);
 976 }
 977
 978 static int init_inodecache(void)
 979 {
 980         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 981                                              sizeof(struct ext4_inode_info),
 982                                              0, (SLAB_RECLAIM_ACCOUNT|
 983                                                 SLAB_MEM_SPREAD),
 984                                              init_once);
 985         if (ext4_inode_cachep == NULL)
 986                 return -ENOMEM;
 987         return 0;
 988 }
 989
 990 static void destroy_inodecache(void)
 991 {
 992         kmem_cache_destroy(ext4_inode_cachep);
 993 }
 994
 995 void ext4_clear_inode(struct inode *inode)
 996 {
 997         invalidate_inode_buffers(inode);
 998         end_writeback(inode);
 999         dquot_drop(inode);
1000         ext4_discard_preallocations(inode);
1001         if (EXT4_I(inode)->jinode) {
1002                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1003                                                EXT4_I(inode)->jinode);
1004                 jbd2_free_inode(EXT4_I(inode)->jinode);
1005                 EXT4_I(inode)->jinode = NULL;
1006         }
1007 }
1008
1009 static inline void ext4_show_quota_options(struct seq_file *seq,
1010                                            struct super_block *sb)
1011 {
1012 #if defined(CONFIG_QUOTA)
1013         struct ext4_sb_info *sbi = EXT4_SB(sb);
1014
1015         if (sbi->s_jquota_fmt) {
1016                 char *fmtname = "";
1017
1018                 switch (sbi->s_jquota_fmt) {
1019                 case QFMT_VFS_OLD:
1020                         fmtname = "vfsold";
1021                         break;
1022                 case QFMT_VFS_V0:
1023                         fmtname = "vfsv0";
1024                         break;
1025                 case QFMT_VFS_V1:
1026                         fmtname = "vfsv1";
1027                         break;
1028                 }
1029                 seq_printf(seq, ",jqfmt=%s", fmtname);
1030         }
1031
1032         if (sbi->s_qf_names[USRQUOTA])
1033                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1034
1035         if (sbi->s_qf_names[GRPQUOTA])
1036                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1037
1038         if (test_opt(sb, USRQUOTA))
1039                 seq_puts(seq, ",usrquota");
1040
1041         if (test_opt(sb, GRPQUOTA))
1042                 seq_puts(seq, ",grpquota");
1043 #endif
1044 }
1045
1046 /*
1047  * Show an option if
1048  *  - it's set to a non-default value OR
1049  *  - if the per-sb default is different from the global default
1050  */
1051 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1052 {
1053         int def_errors;
1054         unsigned long def_mount_opts;
1055         struct super_block *sb = root->d_sb;
1056         struct ext4_sb_info *sbi = EXT4_SB(sb);
1057         struct ext4_super_block *es = sbi->s_es;
1058
1059         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1060         def_errors     = le16_to_cpu(es->s_errors);
1061
1062         if (sbi->s_sb_block != 1)
1063                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1064         if (test_opt(sb, MINIX_DF))
1065                 seq_puts(seq, ",minixdf");
1066         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1067                 seq_puts(seq, ",grpid");
1068         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1069                 seq_puts(seq, ",nogrpid");
1070         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1071             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1072                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1073         }
1074         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1075             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1076                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1077         }
1078         if (test_opt(sb, ERRORS_RO)) {
1079                 if (def_errors == EXT4_ERRORS_PANIC ||
1080                     def_errors == EXT4_ERRORS_CONTINUE) {
1081                         seq_puts(seq, ",errors=remount-ro");
1082                 }
1083         }
1084         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1085                 seq_puts(seq, ",errors=continue");
1086         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1087                 seq_puts(seq, ",errors=panic");
1088         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1089                 seq_puts(seq, ",nouid32");
1090         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1091                 seq_puts(seq, ",debug");
1092 #ifdef CONFIG_EXT4_FS_XATTR
1093         if (test_opt(sb, XATTR_USER))
1094                 seq_puts(seq, ",user_xattr");
1095         if (!test_opt(sb, XATTR_USER))
1096                 seq_puts(seq, ",nouser_xattr");
1097 #endif
1098 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1099         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1100                 seq_puts(seq, ",acl");
1101         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1102                 seq_puts(seq, ",noacl");
1103 #endif
1104         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1105                 seq_printf(seq, ",commit=%u",
1106                            (unsigned) (sbi->s_commit_interval / HZ));
1107         }
1108         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1109                 seq_printf(seq, ",min_batch_time=%u",
1110                            (unsigned) sbi->s_min_batch_time);
1111         }
1112         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1113                 seq_printf(seq, ",max_batch_time=%u",
1114                            (unsigned) sbi->s_max_batch_time);
1115         }
1116
1117         /*
1118          * We're changing the default of barrier mount option, so
1119          * let's always display its mount state so it's clear what its
1120          * status is.
1121          */
1122         seq_puts(seq, ",barrier=");
1123         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1124         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1125                 seq_puts(seq, ",journal_async_commit");
1126         else if (test_opt(sb, JOURNAL_CHECKSUM))
1127                 seq_puts(seq, ",journal_checksum");
1128         if (test_opt(sb, I_VERSION))
1129                 seq_puts(seq, ",i_version");
1130         if (!test_opt(sb, DELALLOC) &&
1131             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1132                 seq_puts(seq, ",nodelalloc");
1133
1134         if (!test_opt(sb, MBLK_IO_SUBMIT))
1135                 seq_puts(seq, ",nomblk_io_submit");
1136         if (sbi->s_stripe)
1137                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1138         /*
1139          * journal mode get enabled in different ways
1140          * So just print the value even if we didn't specify it
1141          */
1142         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1143                 seq_puts(seq, ",data=journal");
1144         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1145                 seq_puts(seq, ",data=ordered");
1146         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1147                 seq_puts(seq, ",data=writeback");
1148
1149         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1150                 seq_printf(seq, ",inode_readahead_blks=%u",
1151                            sbi->s_inode_readahead_blks);
1152
1153         if (test_opt(sb, DATA_ERR_ABORT))
1154                 seq_puts(seq, ",data_err=abort");
1155
1156         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1157                 seq_puts(seq, ",noauto_da_alloc");
1158
1159         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1160                 seq_puts(seq, ",discard");
1161
1162         if (test_opt(sb, NOLOAD))
1163                 seq_puts(seq, ",norecovery");
1164
1165         if (test_opt(sb, DIOREAD_NOLOCK))
1166                 seq_puts(seq, ",dioread_nolock");
1167
1168         if (test_opt(sb, BLOCK_VALIDITY) &&
1169             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1170                 seq_puts(seq, ",block_validity");
1171
1172         if (!test_opt(sb, INIT_INODE_TABLE))
1173                 seq_puts(seq, ",noinit_itable");
1174         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1175                 seq_printf(seq, ",init_itable=%u",
1176                            (unsigned) sbi->s_li_wait_mult);
1177
1178         ext4_show_quota_options(seq, sb);
1179
1180         return 0;
1181 }
1182
1183 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1184                                         u64 ino, u32 generation)
1185 {
1186         struct inode *inode;
1187
1188         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1189                 return ERR_PTR(-ESTALE);
1190         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1191                 return ERR_PTR(-ESTALE);
1192
1193         /* iget isn't really right if the inode is currently unallocated!!
1194          *
1195          * ext4_read_inode will return a bad_inode if the inode had been
1196          * deleted, so we should be safe.
1197          *
1198          * Currently we don't know the generation for parent directory, so
1199          * a generation of 0 means "accept any"
1200          */
1201         inode = ext4_iget(sb, ino);
1202         if (IS_ERR(inode))
1203                 return ERR_CAST(inode);
1204         if (generation && inode->i_generation != generation) {
1205                 iput(inode);
1206                 return ERR_PTR(-ESTALE);
1207         }
1208
1209         return inode;
1210 }
1211
1212 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1213                                         int fh_len, int fh_type)
1214 {
1215         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1216                                     ext4_nfs_get_inode);
1217 }
1218
1219 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1220                                         int fh_len, int fh_type)
1221 {
1222         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1223                                     ext4_nfs_get_inode);
1224 }
1225
1226 /*
1227  * Try to release metadata pages (indirect blocks, directories) which are
1228  * mapped via the block device.  Since these pages could have journal heads
1229  * which would prevent try_to_free_buffers() from freeing them, we must use
1230  * jbd2 layer's try_to_free_buffers() function to release them.
1231  */
1232 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1233                                  gfp_t wait)
1234 {
1235         journal_t *journal = EXT4_SB(sb)->s_journal;
1236
1237         WARN_ON(PageChecked(page));
1238         if (!page_has_buffers(page))
1239                 return 0;
1240         if (journal)
1241                 return jbd2_journal_try_to_free_buffers(journal, page,
1242                                                         wait & ~__GFP_WAIT);
1243         return try_to_free_buffers(page);
1244 }
1245
1246 #ifdef CONFIG_QUOTA
1247 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1248 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1249
1250 static int ext4_write_dquot(struct dquot *dquot);
1251 static int ext4_acquire_dquot(struct dquot *dquot);
1252 static int ext4_release_dquot(struct dquot *dquot);
1253 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1254 static int ext4_write_info(struct super_block *sb, int type);
1255 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1256                          struct path *path);
1257 static int ext4_quota_off(struct super_block *sb, int type);
1258 static int ext4_quota_on_mount(struct super_block *sb, int type);
1259 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1260                                size_t len, loff_t off);
1261 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1262                                 const char *data, size_t len, loff_t off);
1263
1264 static const struct dquot_operations ext4_quota_operations = {
1265         .get_reserved_space = ext4_get_reserved_space,
1266         .write_dquot    = ext4_write_dquot,
1267         .acquire_dquot  = ext4_acquire_dquot,
1268         .release_dquot  = ext4_release_dquot,
1269         .mark_dirty     = ext4_mark_dquot_dirty,
1270         .write_info     = ext4_write_info,
1271         .alloc_dquot    = dquot_alloc,
1272         .destroy_dquot  = dquot_destroy,
1273 };
1274
1275 static const struct quotactl_ops ext4_qctl_operations = {
1276         .quota_on       = ext4_quota_on,
1277         .quota_off      = ext4_quota_off,
1278         .quota_sync     = dquot_quota_sync,
1279         .get_info       = dquot_get_dqinfo,
1280         .set_info       = dquot_set_dqinfo,
1281         .get_dqblk      = dquot_get_dqblk,
1282         .set_dqblk      = dquot_set_dqblk
1283 };
1284 #endif
1285
1286 static const struct super_operations ext4_sops = {
1287         .alloc_inode    = ext4_alloc_inode,
1288         .destroy_inode  = ext4_destroy_inode,
1289         .write_inode    = ext4_write_inode,
1290         .dirty_inode    = ext4_dirty_inode,
1291         .drop_inode     = ext4_drop_inode,
1292         .evict_inode    = ext4_evict_inode,
1293         .put_super      = ext4_put_super,
1294         .sync_fs        = ext4_sync_fs,
1295         .freeze_fs      = ext4_freeze,
1296         .unfreeze_fs    = ext4_unfreeze,
1297         .statfs         = ext4_statfs,
1298         .remount_fs     = ext4_remount,
1299         .show_options   = ext4_show_options,
1300 #ifdef CONFIG_QUOTA
1301         .quota_read     = ext4_quota_read,
1302         .quota_write    = ext4_quota_write,
1303 #endif
1304         .bdev_try_to_free_page = bdev_try_to_free_page,
1305 };
1306
1307 static const struct super_operations ext4_nojournal_sops = {
1308         .alloc_inode    = ext4_alloc_inode,
1309         .destroy_inode  = ext4_destroy_inode,
1310         .write_inode    = ext4_write_inode,
1311         .dirty_inode    = ext4_dirty_inode,
1312         .drop_inode     = ext4_drop_inode,
1313         .evict_inode    = ext4_evict_inode,
1314         .write_super    = ext4_write_super,
1315         .put_super      = ext4_put_super,
1316         .statfs         = ext4_statfs,
1317         .remount_fs     = ext4_remount,
1318         .show_options   = ext4_show_options,
1319 #ifdef CONFIG_QUOTA
1320         .quota_read     = ext4_quota_read,
1321         .quota_write    = ext4_quota_write,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1339         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1348         Opt_inode_readahead_blks, Opt_journal_ioprio,
1349         Opt_dioread_nolock, Opt_dioread_lock,
1350         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_oldalloc, "oldalloc"},
1369         {Opt_orlov, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "noload"},
1375         {Opt_noload, "norecovery"},
1376         {Opt_nobh, "nobh"},
1377         {Opt_bh, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_checksum, "journal_checksum"},
1383         {Opt_journal_async_commit, "journal_async_commit"},
1384         {Opt_abort, "abort"},
1385         {Opt_data_journal, "data=journal"},
1386         {Opt_data_ordered, "data=ordered"},
1387         {Opt_data_writeback, "data=writeback"},
1388         {Opt_data_err_abort, "data_err=abort"},
1389         {Opt_data_err_ignore, "data_err=ignore"},
1390         {Opt_offusrjquota, "usrjquota="},
1391         {Opt_usrjquota, "usrjquota=%s"},
1392         {Opt_offgrpjquota, "grpjquota="},
1393         {Opt_grpjquota, "grpjquota=%s"},
1394         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1395         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1396         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1397         {Opt_grpquota, "grpquota"},
1398         {Opt_noquota, "noquota"},
1399         {Opt_quota, "quota"},
1400         {Opt_usrquota, "usrquota"},
1401         {Opt_barrier, "barrier=%u"},
1402         {Opt_barrier, "barrier"},
1403         {Opt_nobarrier, "nobarrier"},
1404         {Opt_i_version, "i_version"},
1405         {Opt_stripe, "stripe=%u"},
1406         {Opt_resize, "resize"},
1407         {Opt_delalloc, "delalloc"},
1408         {Opt_nodelalloc, "nodelalloc"},
1409         {Opt_mblk_io_submit, "mblk_io_submit"},
1410         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1411         {Opt_block_validity, "block_validity"},
1412         {Opt_noblock_validity, "noblock_validity"},
1413         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1414         {Opt_journal_ioprio, "journal_ioprio=%u"},
1415         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1416         {Opt_auto_da_alloc, "auto_da_alloc"},
1417         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1418         {Opt_dioread_nolock, "dioread_nolock"},
1419         {Opt_dioread_lock, "dioread_lock"},
1420         {Opt_discard, "discard"},
1421         {Opt_nodiscard, "nodiscard"},
1422         {Opt_init_itable, "init_itable=%u"},
1423         {Opt_init_itable, "init_itable"},
1424         {Opt_noinit_itable, "noinit_itable"},
1425         {Opt_err, NULL},
1426 };
1427
1428 static ext4_fsblk_t get_sb_block(void **data)
1429 {
1430         ext4_fsblk_t    sb_block;
1431         char            *options = (char *) *data;
1432
1433         if (!options || strncmp(options, "sb=", 3) != 0)
1434                 return 1;       /* Default location */
1435
1436         options += 3;
1437         /* TODO: use simple_strtoll with >32bit ext4 */
1438         sb_block = simple_strtoul(options, &options, 0);
1439         if (*options && *options != ',') {
1440                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1441                        (char *) *data);
1442                 return 1;
1443         }
1444         if (*options == ',')
1445                 options++;
1446         *data = (void *) options;
1447
1448         return sb_block;
1449 }
1450
1451 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1452 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1453         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1454
1455 #ifdef CONFIG_QUOTA
1456 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1457 {
1458         struct ext4_sb_info *sbi = EXT4_SB(sb);
1459         char *qname;
1460
1461         if (sb_any_quota_loaded(sb) &&
1462                 !sbi->s_qf_names[qtype]) {
1463                 ext4_msg(sb, KERN_ERR,
1464                         "Cannot change journaled "
1465                         "quota options when quota turned on");
1466                 return 0;
1467         }
1468         qname = match_strdup(args);
1469         if (!qname) {
1470                 ext4_msg(sb, KERN_ERR,
1471                         "Not enough memory for storing quotafile name");
1472                 return 0;
1473         }
1474         if (sbi->s_qf_names[qtype] &&
1475                 strcmp(sbi->s_qf_names[qtype], qname)) {
1476                 ext4_msg(sb, KERN_ERR,
1477                         "%s quota file already specified", QTYPE2NAME(qtype));
1478                 kfree(qname);
1479                 return 0;
1480         }
1481         sbi->s_qf_names[qtype] = qname;
1482         if (strchr(sbi->s_qf_names[qtype], '/')) {
1483                 ext4_msg(sb, KERN_ERR,
1484                         "quotafile must be on filesystem root");
1485                 kfree(sbi->s_qf_names[qtype]);
1486                 sbi->s_qf_names[qtype] = NULL;
1487                 return 0;
1488         }
1489         set_opt(sb, QUOTA);
1490         return 1;
1491 }
1492
1493 static int clear_qf_name(struct super_block *sb, int qtype)
1494 {
1495
1496         struct ext4_sb_info *sbi = EXT4_SB(sb);
1497
1498         if (sb_any_quota_loaded(sb) &&
1499                 sbi->s_qf_names[qtype]) {
1500                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1501                         " when quota turned on");
1502                 return 0;
1503         }
1504         /*
1505          * The space will be released later when all options are confirmed
1506          * to be correct
1507          */
1508         sbi->s_qf_names[qtype] = NULL;
1509         return 1;
1510 }
1511 #endif
1512
1513 static int parse_options(char *options, struct super_block *sb,
1514                          unsigned long *journal_devnum,
1515                          unsigned int *journal_ioprio,
1516                          ext4_fsblk_t *n_blocks_count, int is_remount)
1517 {
1518         struct ext4_sb_info *sbi = EXT4_SB(sb);
1519         char *p;
1520         substring_t args[MAX_OPT_ARGS];
1521         int data_opt = 0;
1522         int option;
1523 #ifdef CONFIG_QUOTA
1524         int qfmt;
1525 #endif
1526
1527         if (!options)
1528                 return 1;
1529
1530         while ((p = strsep(&options, ",")) != NULL) {
1531                 int token;
1532                 if (!*p)
1533                         continue;
1534
1535                 /*
1536                  * Initialize args struct so we know whether arg was
1537                  * found; some options take optional arguments.
1538                  */
1539                 args[0].to = args[0].from = NULL;
1540                 token = match_token(p, tokens, args);
1541                 switch (token) {
1542                 case Opt_bsd_df:
1543                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1544                         clear_opt(sb, MINIX_DF);
1545                         break;
1546                 case Opt_minix_df:
1547                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1548                         set_opt(sb, MINIX_DF);
1549
1550                         break;
1551                 case Opt_grpid:
1552                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1553                         set_opt(sb, GRPID);
1554
1555                         break;
1556                 case Opt_nogrpid:
1557                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1558                         clear_opt(sb, GRPID);
1559
1560                         break;
1561                 case Opt_resuid:
1562                         if (match_int(&args[0], &option))
1563                                 return 0;
1564                         sbi->s_resuid = option;
1565                         break;
1566                 case Opt_resgid:
1567                         if (match_int(&args[0], &option))
1568                                 return 0;
1569                         sbi->s_resgid = option;
1570                         break;
1571                 case Opt_sb:
1572                         /* handled by get_sb_block() instead of here */
1573                         /* *sb_block = match_int(&args[0]); */
1574                         break;
1575                 case Opt_err_panic:
1576                         clear_opt(sb, ERRORS_CONT);
1577                         clear_opt(sb, ERRORS_RO);
1578                         set_opt(sb, ERRORS_PANIC);
1579                         break;
1580                 case Opt_err_ro:
1581                         clear_opt(sb, ERRORS_CONT);
1582                         clear_opt(sb, ERRORS_PANIC);
1583                         set_opt(sb, ERRORS_RO);
1584                         break;
1585                 case Opt_err_cont:
1586                         clear_opt(sb, ERRORS_RO);
1587                         clear_opt(sb, ERRORS_PANIC);
1588                         set_opt(sb, ERRORS_CONT);
1589                         break;
1590                 case Opt_nouid32:
1591                         set_opt(sb, NO_UID32);
1592                         break;
1593                 case Opt_debug:
1594                         set_opt(sb, DEBUG);
1595                         break;
1596                 case Opt_oldalloc:
1597                         ext4_msg(sb, KERN_WARNING,
1598                                  "Ignoring deprecated oldalloc option");
1599                         break;
1600                 case Opt_orlov:
1601                         ext4_msg(sb, KERN_WARNING,
1602                                  "Ignoring deprecated orlov option");
1603                         break;
1604 #ifdef CONFIG_EXT4_FS_XATTR
1605                 case Opt_user_xattr:
1606                         set_opt(sb, XATTR_USER);
1607                         break;
1608                 case Opt_nouser_xattr:
1609                         clear_opt(sb, XATTR_USER);
1610                         break;
1611 #else
1612                 case Opt_user_xattr:
1613                 case Opt_nouser_xattr:
1614                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1615                         break;
1616 #endif
1617 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1618                 case Opt_acl:
1619                         set_opt(sb, POSIX_ACL);
1620                         break;
1621                 case Opt_noacl:
1622                         clear_opt(sb, POSIX_ACL);
1623                         break;
1624 #else
1625                 case Opt_acl:
1626                 case Opt_noacl:
1627                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1628                         break;
1629 #endif
1630                 case Opt_journal_dev:
1631                         if (is_remount) {
1632                                 ext4_msg(sb, KERN_ERR,
1633                                         "Cannot specify journal on remount");
1634                                 return 0;
1635                         }
1636                         if (match_int(&args[0], &option))
1637                                 return 0;
1638                         *journal_devnum = option;
1639                         break;
1640                 case Opt_journal_checksum:
1641                         set_opt(sb, JOURNAL_CHECKSUM);
1642                         break;
1643                 case Opt_journal_async_commit:
1644                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1645                         set_opt(sb, JOURNAL_CHECKSUM);
1646                         break;
1647                 case Opt_noload:
1648                         set_opt(sb, NOLOAD);
1649                         break;
1650                 case Opt_commit:
1651                         if (match_int(&args[0], &option))
1652                                 return 0;
1653                         if (option < 0)
1654                                 return 0;
1655                         if (option == 0)
1656                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1657                         sbi->s_commit_interval = HZ * option;
1658                         break;
1659                 case Opt_max_batch_time:
1660                         if (match_int(&args[0], &option))
1661                                 return 0;
1662                         if (option < 0)
1663                                 return 0;
1664                         if (option == 0)
1665                                 option = EXT4_DEF_MAX_BATCH_TIME;
1666                         sbi->s_max_batch_time = option;
1667                         break;
1668                 case Opt_min_batch_time:
1669                         if (match_int(&args[0], &option))
1670                                 return 0;
1671                         if (option < 0)
1672                                 return 0;
1673                         sbi->s_min_batch_time = option;
1674                         break;
1675                 case Opt_data_journal:
1676                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1677                         goto datacheck;
1678                 case Opt_data_ordered:
1679                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1680                         goto datacheck;
1681                 case Opt_data_writeback:
1682                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1683                 datacheck:
1684                         if (is_remount) {
1685                                 if (!sbi->s_journal)
1686                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1687                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1688                                         ext4_msg(sb, KERN_ERR,
1689                                                 "Cannot change data mode on remount");
1690                                         return 0;
1691                                 }
1692                         } else {
1693                                 clear_opt(sb, DATA_FLAGS);
1694                                 sbi->s_mount_opt |= data_opt;
1695                         }
1696                         break;
1697                 case Opt_data_err_abort:
1698                         set_opt(sb, DATA_ERR_ABORT);
1699                         break;
1700                 case Opt_data_err_ignore:
1701                         clear_opt(sb, DATA_ERR_ABORT);
1702                         break;
1703 #ifdef CONFIG_QUOTA
1704                 case Opt_usrjquota:
1705                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1706                                 return 0;
1707                         break;
1708                 case Opt_grpjquota:
1709                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1710                                 return 0;
1711                         break;
1712                 case Opt_offusrjquota:
1713                         if (!clear_qf_name(sb, USRQUOTA))
1714                                 return 0;
1715                         break;
1716                 case Opt_offgrpjquota:
1717                         if (!clear_qf_name(sb, GRPQUOTA))
1718                                 return 0;
1719                         break;
1720
1721                 case Opt_jqfmt_vfsold:
1722                         qfmt = QFMT_VFS_OLD;
1723                         goto set_qf_format;
1724                 case Opt_jqfmt_vfsv0:
1725                         qfmt = QFMT_VFS_V0;
1726                         goto set_qf_format;
1727                 case Opt_jqfmt_vfsv1:
1728                         qfmt = QFMT_VFS_V1;
1729 set_qf_format:
1730                         if (sb_any_quota_loaded(sb) &&
1731                             sbi->s_jquota_fmt != qfmt) {
1732                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1733                                         "journaled quota options when "
1734                                         "quota turned on");
1735                                 return 0;
1736                         }
1737                         sbi->s_jquota_fmt = qfmt;
1738                         break;
1739                 case Opt_quota:
1740                 case Opt_usrquota:
1741                         set_opt(sb, QUOTA);
1742                         set_opt(sb, USRQUOTA);
1743                         break;
1744                 case Opt_grpquota:
1745                         set_opt(sb, QUOTA);
1746                         set_opt(sb, GRPQUOTA);
1747                         break;
1748                 case Opt_noquota:
1749                         if (sb_any_quota_loaded(sb)) {
1750                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1751                                         "options when quota turned on");
1752                                 return 0;
1753                         }
1754                         clear_opt(sb, QUOTA);
1755                         clear_opt(sb, USRQUOTA);
1756                         clear_opt(sb, GRPQUOTA);
1757                         break;
1758 #else
1759                 case Opt_quota:
1760                 case Opt_usrquota:
1761                 case Opt_grpquota:
1762                         ext4_msg(sb, KERN_ERR,
1763                                 "quota options not supported");
1764                         break;
1765                 case Opt_usrjquota:
1766                 case Opt_grpjquota:
1767                 case Opt_offusrjquota:
1768                 case Opt_offgrpjquota:
1769                 case Opt_jqfmt_vfsold:
1770                 case Opt_jqfmt_vfsv0:
1771                 case Opt_jqfmt_vfsv1:
1772                         ext4_msg(sb, KERN_ERR,
1773                                 "journaled quota options not supported");
1774                         break;
1775                 case Opt_noquota:
1776                         break;
1777 #endif
1778                 case Opt_abort:
1779                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1780                         break;
1781                 case Opt_nobarrier:
1782                         clear_opt(sb, BARRIER);
1783                         break;
1784                 case Opt_barrier:
1785                         if (args[0].from) {
1786                                 if (match_int(&args[0], &option))
1787                                         return 0;
1788                         } else
1789                                 option = 1;     /* No argument, default to 1 */
1790                         if (option)
1791                                 set_opt(sb, BARRIER);
1792                         else
1793                                 clear_opt(sb, BARRIER);
1794                         break;
1795                 case Opt_ignore:
1796                         break;
1797                 case Opt_resize:
1798                         if (!is_remount) {
1799                                 ext4_msg(sb, KERN_ERR,
1800                                         "resize option only available "
1801                                         "for remount");
1802                                 return 0;
1803                         }
1804                         if (match_int(&args[0], &option) != 0)
1805                                 return 0;
1806                         *n_blocks_count = option;
1807                         break;
1808                 case Opt_nobh:
1809                         ext4_msg(sb, KERN_WARNING,
1810                                  "Ignoring deprecated nobh option");
1811                         break;
1812                 case Opt_bh:
1813                         ext4_msg(sb, KERN_WARNING,
1814                                  "Ignoring deprecated bh option");
1815                         break;
1816                 case Opt_i_version:
1817                         set_opt(sb, I_VERSION);
1818                         sb->s_flags |= MS_I_VERSION;
1819                         break;
1820                 case Opt_nodelalloc:
1821                         clear_opt(sb, DELALLOC);
1822                         clear_opt2(sb, EXPLICIT_DELALLOC);
1823                         break;
1824                 case Opt_mblk_io_submit:
1825                         set_opt(sb, MBLK_IO_SUBMIT);
1826                         break;
1827                 case Opt_nomblk_io_submit:
1828                         clear_opt(sb, MBLK_IO_SUBMIT);
1829                         break;
1830                 case Opt_stripe:
1831                         if (match_int(&args[0], &option))
1832                                 return 0;
1833                         if (option < 0)
1834                                 return 0;
1835                         sbi->s_stripe = option;
1836                         break;
1837                 case Opt_delalloc:
1838                         set_opt(sb, DELALLOC);
1839                         set_opt2(sb, EXPLICIT_DELALLOC);
1840                         break;
1841                 case Opt_block_validity:
1842                         set_opt(sb, BLOCK_VALIDITY);
1843                         break;
1844                 case Opt_noblock_validity:
1845                         clear_opt(sb, BLOCK_VALIDITY);
1846                         break;
1847                 case Opt_inode_readahead_blks:
1848                         if (match_int(&args[0], &option))
1849                                 return 0;
1850                         if (option < 0 || option > (1 << 30))
1851                                 return 0;
1852                         if (option && !is_power_of_2(option)) {
1853                                 ext4_msg(sb, KERN_ERR,
1854                                          "EXT4-fs: inode_readahead_blks"
1855                                          " must be a power of 2");
1856                                 return 0;
1857                         }
1858                         sbi->s_inode_readahead_blks = option;
1859                         break;
1860                 case Opt_journal_ioprio:
1861                         if (match_int(&args[0], &option))
1862                                 return 0;
1863                         if (option < 0 || option > 7)
1864                                 break;
1865                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1866                                                             option);
1867                         break;
1868                 case Opt_noauto_da_alloc:
1869                         set_opt(sb, NO_AUTO_DA_ALLOC);
1870                         break;
1871                 case Opt_auto_da_alloc:
1872                         if (args[0].from) {
1873                                 if (match_int(&args[0], &option))
1874                                         return 0;
1875                         } else
1876                                 option = 1;     /* No argument, default to 1 */
1877                         if (option)
1878                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1879                         else
1880                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1881                         break;
1882                 case Opt_discard:
1883                         set_opt(sb, DISCARD);
1884                         break;
1885                 case Opt_nodiscard:
1886                         clear_opt(sb, DISCARD);
1887                         break;
1888                 case Opt_dioread_nolock:
1889                         set_opt(sb, DIOREAD_NOLOCK);
1890                         break;
1891                 case Opt_dioread_lock:
1892                         clear_opt(sb, DIOREAD_NOLOCK);
1893                         break;
1894                 case Opt_init_itable:
1895                         set_opt(sb, INIT_INODE_TABLE);
1896                         if (args[0].from) {
1897                                 if (match_int(&args[0], &option))
1898                                         return 0;
1899                         } else
1900                                 option = EXT4_DEF_LI_WAIT_MULT;
1901                         if (option < 0)
1902                                 return 0;
1903                         sbi->s_li_wait_mult = option;
1904                         break;
1905                 case Opt_noinit_itable:
1906                         clear_opt(sb, INIT_INODE_TABLE);
1907                         break;
1908                 default:
1909                         ext4_msg(sb, KERN_ERR,
1910                                "Unrecognized mount option \"%s\" "
1911                                "or missing value", p);
1912                         return 0;
1913                 }
1914         }
1915 #ifdef CONFIG_QUOTA
1916         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1917                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1918                         clear_opt(sb, USRQUOTA);
1919
1920                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1921                         clear_opt(sb, GRPQUOTA);
1922
1923                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1924                         ext4_msg(sb, KERN_ERR, "old and new quota "
1925                                         "format mixing");
1926                         return 0;
1927                 }
1928
1929                 if (!sbi->s_jquota_fmt) {
1930                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1931                                         "not specified");
1932                         return 0;
1933                 }
1934         } else {
1935                 if (sbi->s_jquota_fmt) {
1936                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1937                                         "specified with no journaling "
1938                                         "enabled");
1939                         return 0;
1940                 }
1941         }
1942 #endif
1943         return 1;
1944 }
1945
1946 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1947                             int read_only)
1948 {
1949         struct ext4_sb_info *sbi = EXT4_SB(sb);
1950         int res = 0;
1951
1952         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1953                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1954                          "forcing read-only mode");
1955                 res = MS_RDONLY;
1956         }
1957         if (read_only)
1958                 goto done;
1959         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1960                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1961                          "running e2fsck is recommended");
1962         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1963                 ext4_msg(sb, KERN_WARNING,
1964                          "warning: mounting fs with errors, "
1965                          "running e2fsck is recommended");
1966         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1967                  le16_to_cpu(es->s_mnt_count) >=
1968                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1969                 ext4_msg(sb, KERN_WARNING,
1970                          "warning: maximal mount count reached, "
1971                          "running e2fsck is recommended");
1972         else if (le32_to_cpu(es->s_checkinterval) &&
1973                 (le32_to_cpu(es->s_lastcheck) +
1974                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1975                 ext4_msg(sb, KERN_WARNING,
1976                          "warning: checktime reached, "
1977                          "running e2fsck is recommended");
1978         if (!sbi->s_journal)
1979                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1980         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1981                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1982         le16_add_cpu(&es->s_mnt_count, 1);
1983         es->s_mtime = cpu_to_le32(get_seconds());
1984         ext4_update_dynamic_rev(sb);
1985         if (sbi->s_journal)
1986                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1987
1988         ext4_commit_super(sb, 1);
1989 done:
1990         if (test_opt(sb, DEBUG))
1991                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1992                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1993                         sb->s_blocksize,
1994                         sbi->s_groups_count,
1995                         EXT4_BLOCKS_PER_GROUP(sb),
1996                         EXT4_INODES_PER_GROUP(sb),
1997                         sbi->s_mount_opt, sbi->s_mount_opt2);
1998
1999         cleancache_init_fs(sb);
2000         return res;
2001 }
2002
2003 static int ext4_fill_flex_info(struct super_block *sb)
2004 {
2005         struct ext4_sb_info *sbi = EXT4_SB(sb);
2006         struct ext4_group_desc *gdp = NULL;
2007         ext4_group_t flex_group_count;
2008         ext4_group_t flex_group;
2009         unsigned int groups_per_flex = 0;
2010         size_t size;
2011         int i;
2012
2013         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2014         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2015                 sbi->s_log_groups_per_flex = 0;
2016                 return 1;
2017         }
2018         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2019
2020         /* We allocate both existing and potentially added groups */
2021         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2022                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2023                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2024         size = flex_group_count * sizeof(struct flex_groups);
2025         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2026         if (sbi->s_flex_groups == NULL) {
2027                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2028                          flex_group_count);
2029                 goto failed;
2030         }
2031
2032         for (i = 0; i < sbi->s_groups_count; i++) {
2033                 gdp = ext4_get_group_desc(sb, i, NULL);
2034
2035                 flex_group = ext4_flex_group(sbi, i);
2036                 atomic_add(ext4_free_inodes_count(sb, gdp),
2037                            &sbi->s_flex_groups[flex_group].free_inodes);
2038                 atomic_add(ext4_free_group_clusters(sb, gdp),
2039                            &sbi->s_flex_groups[flex_group].free_clusters);
2040                 atomic_add(ext4_used_dirs_count(sb, gdp),
2041                            &sbi->s_flex_groups[flex_group].used_dirs);
2042         }
2043
2044         return 1;
2045 failed:
2046         return 0;
2047 }
2048
2049 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2050                             struct ext4_group_desc *gdp)
2051 {
2052         __u16 crc = 0;
2053
2054         if (sbi->s_es->s_feature_ro_compat &
2055             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2056                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2057                 __le32 le_group = cpu_to_le32(block_group);
2058
2059                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2060                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2061                 crc = crc16(crc, (__u8 *)gdp, offset);
2062                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2063                 /* for checksum of struct ext4_group_desc do the rest...*/
2064                 if ((sbi->s_es->s_feature_incompat &
2065                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2066                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2067                         crc = crc16(crc, (__u8 *)gdp + offset,
2068                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2069                                         offset);
2070         }
2071
2072         return cpu_to_le16(crc);
2073 }
2074
2075 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2076                                 struct ext4_group_desc *gdp)
2077 {
2078         if ((sbi->s_es->s_feature_ro_compat &
2079              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2080             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2081                 return 0;
2082
2083         return 1;
2084 }
2085
2086 /* Called at mount-time, super-block is locked */
2087 static int ext4_check_descriptors(struct super_block *sb,
2088                                   ext4_group_t *first_not_zeroed)
2089 {
2090         struct ext4_sb_info *sbi = EXT4_SB(sb);
2091         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2092         ext4_fsblk_t last_block;
2093         ext4_fsblk_t block_bitmap;
2094         ext4_fsblk_t inode_bitmap;
2095         ext4_fsblk_t inode_table;
2096         int flexbg_flag = 0;
2097         ext4_group_t i, grp = sbi->s_groups_count;
2098
2099         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2100                 flexbg_flag = 1;
2101
2102         ext4_debug("Checking group descriptors");
2103
2104         for (i = 0; i < sbi->s_groups_count; i++) {
2105                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2106
2107                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2108                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2109                 else
2110                         last_block = first_block +
2111                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2112
2113                 if ((grp == sbi->s_groups_count) &&
2114                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2115                         grp = i;
2116
2117                 block_bitmap = ext4_block_bitmap(sb, gdp);
2118                 if (block_bitmap < first_block || block_bitmap > last_block) {
2119                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2120                                "Block bitmap for group %u not in group "
2121                                "(block %llu)!", i, block_bitmap);
2122                         return 0;
2123                 }
2124                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2125                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2126                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2127                                "Inode bitmap for group %u not in group "
2128                                "(block %llu)!", i, inode_bitmap);
2129                         return 0;
2130                 }
2131                 inode_table = ext4_inode_table(sb, gdp);
2132                 if (inode_table < first_block ||
2133                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                "Inode table for group %u not in group "
2136                                "(block %llu)!", i, inode_table);
2137                         return 0;
2138                 }
2139                 ext4_lock_group(sb, i);
2140                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2141                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2142                                  "Checksum for group %u failed (%u!=%u)",
2143                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2144                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2145                         if (!(sb->s_flags & MS_RDONLY)) {
2146                                 ext4_unlock_group(sb, i);
2147                                 return 0;
2148                         }
2149                 }
2150                 ext4_unlock_group(sb, i);
2151                 if (!flexbg_flag)
2152                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2153         }
2154         if (NULL != first_not_zeroed)
2155                 *first_not_zeroed = grp;
2156
2157         ext4_free_blocks_count_set(sbi->s_es,
2158                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2159         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2160         return 1;
2161 }
2162
2163 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2164  * the superblock) which were deleted from all directories, but held open by
2165  * a process at the time of a crash.  We walk the list and try to delete these
2166  * inodes at recovery time (only with a read-write filesystem).
2167  *
2168  * In order to keep the orphan inode chain consistent during traversal (in
2169  * case of crash during recovery), we link each inode into the superblock
2170  * orphan list_head and handle it the same way as an inode deletion during
2171  * normal operation (which journals the operations for us).
2172  *
2173  * We only do an iget() and an iput() on each inode, which is very safe if we
2174  * accidentally point at an in-use or already deleted inode.  The worst that
2175  * can happen in this case is that we get a "bit already cleared" message from
2176  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2177  * e2fsck was run on this filesystem, and it must have already done the orphan
2178  * inode cleanup for us, so we can safely abort without any further action.
2179  */
2180 static void ext4_orphan_cleanup(struct super_block *sb,
2181                                 struct ext4_super_block *es)
2182 {
2183         unsigned int s_flags = sb->s_flags;
2184         int nr_orphans = 0, nr_truncates = 0;
2185 #ifdef CONFIG_QUOTA
2186         int i;
2187 #endif
2188         if (!es->s_last_orphan) {
2189                 jbd_debug(4, "no orphan inodes to clean up\n");
2190                 return;
2191         }
2192
2193         if (bdev_read_only(sb->s_bdev)) {
2194                 ext4_msg(sb, KERN_ERR, "write access "
2195                         "unavailable, skipping orphan cleanup");
2196                 return;
2197         }
2198
2199         /* Check if feature set would not allow a r/w mount */
2200         if (!ext4_feature_set_ok(sb, 0)) {
2201                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2202                          "unknown ROCOMPAT features");
2203                 return;
2204         }
2205
2206         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2207                 if (es->s_last_orphan)
2208                         jbd_debug(1, "Errors on filesystem, "
2209                                   "clearing orphan list.\n");
2210                 es->s_last_orphan = 0;
2211                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2212                 return;
2213         }
2214
2215         if (s_flags & MS_RDONLY) {
2216                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2217                 sb->s_flags &= ~MS_RDONLY;
2218         }
2219 #ifdef CONFIG_QUOTA
2220         /* Needed for iput() to work correctly and not trash data */
2221         sb->s_flags |= MS_ACTIVE;
2222         /* Turn on quotas so that they are updated correctly */
2223         for (i = 0; i < MAXQUOTAS; i++) {
2224                 if (EXT4_SB(sb)->s_qf_names[i]) {
2225                         int ret = ext4_quota_on_mount(sb, i);
2226                         if (ret < 0)
2227                                 ext4_msg(sb, KERN_ERR,
2228                                         "Cannot turn on journaled "
2229                                         "quota: error %d", ret);
2230                 }
2231         }
2232 #endif
2233
2234         while (es->s_last_orphan) {
2235                 struct inode *inode;
2236
2237                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2238                 if (IS_ERR(inode)) {
2239                         es->s_last_orphan = 0;
2240                         break;
2241                 }
2242
2243                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2244                 dquot_initialize(inode);
2245                 if (inode->i_nlink) {
2246                         ext4_msg(sb, KERN_DEBUG,
2247                                 "%s: truncating inode %lu to %lld bytes",
2248                                 __func__, inode->i_ino, inode->i_size);
2249                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2250                                   inode->i_ino, inode->i_size);
2251                         ext4_truncate(inode);
2252                         nr_truncates++;
2253                 } else {
2254                         ext4_msg(sb, KERN_DEBUG,
2255                                 "%s: deleting unreferenced inode %lu",
2256                                 __func__, inode->i_ino);
2257                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2258                                   inode->i_ino);
2259                         nr_orphans++;
2260                 }
2261                 iput(inode);  /* The delete magic happens here! */
2262         }
2263
2264 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2265
2266         if (nr_orphans)
2267                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2268                        PLURAL(nr_orphans));
2269         if (nr_truncates)
2270                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2271                        PLURAL(nr_truncates));
2272 #ifdef CONFIG_QUOTA
2273         /* Turn quotas off */
2274         for (i = 0; i < MAXQUOTAS; i++) {
2275                 if (sb_dqopt(sb)->files[i])
2276                         dquot_quota_off(sb, i);
2277         }
2278 #endif
2279         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2280 }
2281
2282 /*
2283  * Maximal extent format file size.
2284  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2285  * extent format containers, within a sector_t, and within i_blocks
2286  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2287  * so that won't be a limiting factor.
2288  *
2289  * However there is other limiting factor. We do store extents in the form
2290  * of starting block and length, hence the resulting length of the extent
2291  * covering maximum file size must fit into on-disk format containers as
2292  * well. Given that length is always by 1 unit bigger than max unit (because
2293  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2294  *
2295  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2296  */
2297 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2298 {
2299         loff_t res;
2300         loff_t upper_limit = MAX_LFS_FILESIZE;
2301
2302         /* small i_blocks in vfs inode? */
2303         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2304                 /*
2305                  * CONFIG_LBDAF is not enabled implies the inode
2306                  * i_block represent total blocks in 512 bytes
2307                  * 32 == size of vfs inode i_blocks * 8
2308                  */
2309                 upper_limit = (1LL << 32) - 1;
2310
2311                 /* total blocks in file system block size */
2312                 upper_limit >>= (blkbits - 9);
2313                 upper_limit <<= blkbits;
2314         }
2315
2316         /*
2317          * 32-bit extent-start container, ee_block. We lower the maxbytes
2318          * by one fs block, so ee_len can cover the extent of maximum file
2319          * size
2320          */
2321         res = (1LL << 32) - 1;
2322         res <<= blkbits;
2323
2324         /* Sanity check against vm- & vfs- imposed limits */
2325         if (res > upper_limit)
2326                 res = upper_limit;
2327
2328         return res;
2329 }
2330
2331 /*
2332  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2333  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2334  * We need to be 1 filesystem block less than the 2^48 sector limit.
2335  */
2336 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2337 {
2338         loff_t res = EXT4_NDIR_BLOCKS;
2339         int meta_blocks;
2340         loff_t upper_limit;
2341         /* This is calculated to be the largest file size for a dense, block
2342          * mapped file such that the file's total number of 512-byte sectors,
2343          * including data and all indirect blocks, does not exceed (2^48 - 1).
2344          *
2345          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2346          * number of 512-byte sectors of the file.
2347          */
2348
2349         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2350                 /*
2351                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2352                  * the inode i_block field represents total file blocks in
2353                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2354                  */
2355                 upper_limit = (1LL << 32) - 1;
2356
2357                 /* total blocks in file system block size */
2358                 upper_limit >>= (bits - 9);
2359
2360         } else {
2361                 /*
2362                  * We use 48 bit ext4_inode i_blocks
2363                  * With EXT4_HUGE_FILE_FL set the i_blocks
2364                  * represent total number of blocks in
2365                  * file system block size
2366                  */
2367                 upper_limit = (1LL << 48) - 1;
2368
2369         }
2370
2371         /* indirect blocks */
2372         meta_blocks = 1;
2373         /* double indirect blocks */
2374         meta_blocks += 1 + (1LL << (bits-2));
2375         /* tripple indirect blocks */
2376         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2377
2378         upper_limit -= meta_blocks;
2379         upper_limit <<= bits;
2380
2381         res += 1LL << (bits-2);
2382         res += 1LL << (2*(bits-2));
2383         res += 1LL << (3*(bits-2));
2384         res <<= bits;
2385         if (res > upper_limit)
2386                 res = upper_limit;
2387
2388         if (res > MAX_LFS_FILESIZE)
2389                 res = MAX_LFS_FILESIZE;
2390
2391         return res;
2392 }
2393
2394 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2395                                    ext4_fsblk_t logical_sb_block, int nr)
2396 {
2397         struct ext4_sb_info *sbi = EXT4_SB(sb);
2398         ext4_group_t bg, first_meta_bg;
2399         int has_super = 0;
2400
2401         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2402
2403         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2404             nr < first_meta_bg)
2405                 return logical_sb_block + nr + 1;
2406         bg = sbi->s_desc_per_block * nr;
2407         if (ext4_bg_has_super(sb, bg))
2408                 has_super = 1;
2409
2410         return (has_super + ext4_group_first_block_no(sb, bg));
2411 }
2412
2413 /**
2414  * ext4_get_stripe_size: Get the stripe size.
2415  * @sbi: In memory super block info
2416  *
2417  * If we have specified it via mount option, then
2418  * use the mount option value. If the value specified at mount time is
2419  * greater than the blocks per group use the super block value.
2420  * If the super block value is greater than blocks per group return 0.
2421  * Allocator needs it be less than blocks per group.
2422  *
2423  */
2424 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2425 {
2426         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2427         unsigned long stripe_width =
2428                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2429         int ret;
2430
2431         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2432                 ret = sbi->s_stripe;
2433         else if (stripe_width <= sbi->s_blocks_per_group)
2434                 ret = stripe_width;
2435         else if (stride <= sbi->s_blocks_per_group)
2436                 ret = stride;
2437         else
2438                 ret = 0;
2439
2440         /*
2441          * If the stripe width is 1, this makes no sense and
2442          * we set it to 0 to turn off stripe handling code.
2443          */
2444         if (ret <= 1)
2445                 ret = 0;
2446
2447         return ret;
2448 }
2449
2450 /* sysfs supprt */
2451
2452 struct ext4_attr {
2453         struct attribute attr;
2454         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2455         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2456                          const char *, size_t);
2457         int offset;
2458 };
2459
2460 static int parse_strtoul(const char *buf,
2461                 unsigned long max, unsigned long *value)
2462 {
2463         char *endp;
2464
2465         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2466         endp = skip_spaces(endp);
2467         if (*endp || *value > max)
2468                 return -EINVAL;
2469
2470         return 0;
2471 }
2472
2473 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2474                                               struct ext4_sb_info *sbi,
2475                                               char *buf)
2476 {
2477         return snprintf(buf, PAGE_SIZE, "%llu\n",
2478                 (s64) EXT4_C2B(sbi,
2479                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2480 }
2481
2482 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2483                                          struct ext4_sb_info *sbi, char *buf)
2484 {
2485         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2486
2487         if (!sb->s_bdev->bd_part)
2488                 return snprintf(buf, PAGE_SIZE, "0\n");
2489         return snprintf(buf, PAGE_SIZE, "%lu\n",
2490                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2491                          sbi->s_sectors_written_start) >> 1);
2492 }
2493
2494 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2495                                           struct ext4_sb_info *sbi, char *buf)
2496 {
2497         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2498
2499         if (!sb->s_bdev->bd_part)
2500                 return snprintf(buf, PAGE_SIZE, "0\n");
2501         return snprintf(buf, PAGE_SIZE, "%llu\n",
2502                         (unsigned long long)(sbi->s_kbytes_written +
2503                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2504                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2505 }
2506
2507 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2508                                       struct ext4_sb_info *sbi, char *buf)
2509 {
2510         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2511 }
2512
2513 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2514                                         struct ext4_sb_info *sbi, char *buf)
2515 {
2516         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2517 }
2518
2519 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2520                                           struct ext4_sb_info *sbi,
2521                                           const char *buf, size_t count)
2522 {
2523         unsigned long t;
2524
2525         if (parse_strtoul(buf, 0x40000000, &t))
2526                 return -EINVAL;
2527
2528         if (t && !is_power_of_2(t))
2529                 return -EINVAL;
2530
2531         sbi->s_inode_readahead_blks = t;
2532         return count;
2533 }
2534
2535 static ssize_t sbi_ui_show(struct ext4_attr *a,
2536                            struct ext4_sb_info *sbi, char *buf)
2537 {
2538         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2539
2540         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2541 }
2542
2543 static ssize_t sbi_ui_store(struct ext4_attr *a,
2544                             struct ext4_sb_info *sbi,
2545                             const char *buf, size_t count)
2546 {
2547         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2548         unsigned long t;
2549
2550         if (parse_strtoul(buf, 0xffffffff, &t))
2551                 return -EINVAL;
2552         *ui = t;
2553         return count;
2554 }
2555
2556 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2557 static struct ext4_attr ext4_attr_##_name = {                   \
2558         .attr = {.name = __stringify(_name), .mode = _mode },   \
2559         .show   = _show,                                        \
2560         .store  = _store,                                       \
2561         .offset = offsetof(struct ext4_sb_info, _elname),       \
2562 }
2563 #define EXT4_ATTR(name, mode, show, store) \
2564 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2565
2566 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2567 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2568 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2569 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2570         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2571 #define ATTR_LIST(name) &ext4_attr_##name.attr
2572
2573 EXT4_RO_ATTR(delayed_allocation_blocks);
2574 EXT4_RO_ATTR(session_write_kbytes);
2575 EXT4_RO_ATTR(lifetime_write_kbytes);
2576 EXT4_RO_ATTR(extent_cache_hits);
2577 EXT4_RO_ATTR(extent_cache_misses);
2578 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2579                  inode_readahead_blks_store, s_inode_readahead_blks);
2580 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2581 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2582 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2583 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2584 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2585 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2586 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2587 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2588
2589 static struct attribute *ext4_attrs[] = {
2590         ATTR_LIST(delayed_allocation_blocks),
2591         ATTR_LIST(session_write_kbytes),
2592         ATTR_LIST(lifetime_write_kbytes),
2593         ATTR_LIST(extent_cache_hits),
2594         ATTR_LIST(extent_cache_misses),
2595         ATTR_LIST(inode_readahead_blks),
2596         ATTR_LIST(inode_goal),
2597         ATTR_LIST(mb_stats),
2598         ATTR_LIST(mb_max_to_scan),
2599         ATTR_LIST(mb_min_to_scan),
2600         ATTR_LIST(mb_order2_req),
2601         ATTR_LIST(mb_stream_req),
2602         ATTR_LIST(mb_group_prealloc),
2603         ATTR_LIST(max_writeback_mb_bump),
2604         NULL,
2605 };
2606
2607 /* Features this copy of ext4 supports */
2608 EXT4_INFO_ATTR(lazy_itable_init);
2609 EXT4_INFO_ATTR(batched_discard);
2610
2611 static struct attribute *ext4_feat_attrs[] = {
2612         ATTR_LIST(lazy_itable_init),
2613         ATTR_LIST(batched_discard),
2614         NULL,
2615 };
2616
2617 static ssize_t ext4_attr_show(struct kobject *kobj,
2618                               struct attribute *attr, char *buf)
2619 {
2620         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2621                                                 s_kobj);
2622         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2623
2624         return a->show ? a->show(a, sbi, buf) : 0;
2625 }
2626
2627 static ssize_t ext4_attr_store(struct kobject *kobj,
2628                                struct attribute *attr,
2629                                const char *buf, size_t len)
2630 {
2631         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2632                                                 s_kobj);
2633         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2634
2635         return a->store ? a->store(a, sbi, buf, len) : 0;
2636 }
2637
2638 static void ext4_sb_release(struct kobject *kobj)
2639 {
2640         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2641                                                 s_kobj);
2642         complete(&sbi->s_kobj_unregister);
2643 }
2644
2645 static const struct sysfs_ops ext4_attr_ops = {
2646         .show   = ext4_attr_show,
2647         .store  = ext4_attr_store,
2648 };
2649
2650 static struct kobj_type ext4_ktype = {
2651         .default_attrs  = ext4_attrs,
2652         .sysfs_ops      = &ext4_attr_ops,
2653         .release        = ext4_sb_release,
2654 };
2655
2656 static void ext4_feat_release(struct kobject *kobj)
2657 {
2658         complete(&ext4_feat->f_kobj_unregister);
2659 }
2660
2661 static struct kobj_type ext4_feat_ktype = {
2662         .default_attrs  = ext4_feat_attrs,
2663         .sysfs_ops      = &ext4_attr_ops,
2664         .release        = ext4_feat_release,
2665 };
2666
2667 /*
2668  * Check whether this filesystem can be mounted based on
2669  * the features present and the RDONLY/RDWR mount requested.
2670  * Returns 1 if this filesystem can be mounted as requested,
2671  * 0 if it cannot be.
2672  */
2673 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2674 {
2675         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2676                 ext4_msg(sb, KERN_ERR,
2677                         "Couldn't mount because of "
2678                         "unsupported optional features (%x)",
2679                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2680                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2681                 return 0;
2682         }
2683
2684         if (readonly)
2685                 return 1;
2686
2687         /* Check that feature set is OK for a read-write mount */
2688         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2689                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2690                          "unsupported optional features (%x)",
2691                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2692                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2693                 return 0;
2694         }
2695         /*
2696          * Large file size enabled file system can only be mounted
2697          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2698          */
2699         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2700                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2701                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2702                                  "cannot be mounted RDWR without "
2703                                  "CONFIG_LBDAF");
2704                         return 0;
2705                 }
2706         }
2707         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2708             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2709                 ext4_msg(sb, KERN_ERR,
2710                          "Can't support bigalloc feature without "
2711                          "extents feature\n");
2712                 return 0;
2713         }
2714         return 1;
2715 }
2716
2717 /*
2718  * This function is called once a day if we have errors logged
2719  * on the file system
2720  */
2721 static void print_daily_error_info(unsigned long arg)
2722 {
2723         struct super_block *sb = (struct super_block *) arg;
2724         struct ext4_sb_info *sbi;
2725         struct ext4_super_block *es;
2726
2727         sbi = EXT4_SB(sb);
2728         es = sbi->s_es;
2729
2730         if (es->s_error_count)
2731                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2732                          le32_to_cpu(es->s_error_count));
2733         if (es->s_first_error_time) {
2734                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2735                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2736                        (int) sizeof(es->s_first_error_func),
2737                        es->s_first_error_func,
2738                        le32_to_cpu(es->s_first_error_line));
2739                 if (es->s_first_error_ino)
2740                         printk(": inode %u",
2741                                le32_to_cpu(es->s_first_error_ino));
2742                 if (es->s_first_error_block)
2743                         printk(": block %llu", (unsigned long long)
2744                                le64_to_cpu(es->s_first_error_block));
2745                 printk("\n");
2746         }
2747         if (es->s_last_error_time) {
2748                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2749                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2750                        (int) sizeof(es->s_last_error_func),
2751                        es->s_last_error_func,
2752                        le32_to_cpu(es->s_last_error_line));
2753                 if (es->s_last_error_ino)
2754                         printk(": inode %u",
2755                                le32_to_cpu(es->s_last_error_ino));
2756                 if (es->s_last_error_block)
2757                         printk(": block %llu", (unsigned long long)
2758                                le64_to_cpu(es->s_last_error_block));
2759                 printk("\n");
2760         }
2761         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2762 }
2763
2764 /* Find next suitable group and run ext4_init_inode_table */
2765 static int ext4_run_li_request(struct ext4_li_request *elr)
2766 {
2767         struct ext4_group_desc *gdp = NULL;
2768         ext4_group_t group, ngroups;
2769         struct super_block *sb;
2770         unsigned long timeout = 0;
2771         int ret = 0;
2772
2773         sb = elr->lr_super;
2774         ngroups = EXT4_SB(sb)->s_groups_count;
2775
2776         for (group = elr->lr_next_group; group < ngroups; group++) {
2777                 gdp = ext4_get_group_desc(sb, group, NULL);
2778                 if (!gdp) {
2779                         ret = 1;
2780                         break;
2781                 }
2782
2783                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2784                         break;
2785         }
2786
2787         if (group == ngroups)
2788                 ret = 1;
2789
2790         if (!ret) {
2791                 timeout = jiffies;
2792                 ret = ext4_init_inode_table(sb, group,
2793                                             elr->lr_timeout ? 0 : 1);
2794                 if (elr->lr_timeout == 0) {
2795                         timeout = (jiffies - timeout) *
2796                                   elr->lr_sbi->s_li_wait_mult;
2797                         elr->lr_timeout = timeout;
2798                 }
2799                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2800                 elr->lr_next_group = group + 1;
2801         }
2802
2803         return ret;
2804 }
2805
2806 /*
2807  * Remove lr_request from the list_request and free the
2808  * request structure. Should be called with li_list_mtx held
2809  */
2810 static void ext4_remove_li_request(struct ext4_li_request *elr)
2811 {
2812         struct ext4_sb_info *sbi;
2813
2814         if (!elr)
2815                 return;
2816
2817         sbi = elr->lr_sbi;
2818
2819         list_del(&elr->lr_request);
2820         sbi->s_li_request = NULL;
2821         kfree(elr);
2822 }
2823
2824 static void ext4_unregister_li_request(struct super_block *sb)
2825 {
2826         mutex_lock(&ext4_li_mtx);
2827         if (!ext4_li_info) {
2828                 mutex_unlock(&ext4_li_mtx);
2829                 return;
2830         }
2831
2832         mutex_lock(&ext4_li_info->li_list_mtx);
2833         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2834         mutex_unlock(&ext4_li_info->li_list_mtx);
2835         mutex_unlock(&ext4_li_mtx);
2836 }
2837
2838 static struct task_struct *ext4_lazyinit_task;
2839
2840 /*
2841  * This is the function where ext4lazyinit thread lives. It walks
2842  * through the request list searching for next scheduled filesystem.
2843  * When such a fs is found, run the lazy initialization request
2844  * (ext4_rn_li_request) and keep track of the time spend in this
2845  * function. Based on that time we compute next schedule time of
2846  * the request. When walking through the list is complete, compute
2847  * next waking time and put itself into sleep.
2848  */
2849 static int ext4_lazyinit_thread(void *arg)
2850 {
2851         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2852         struct list_head *pos, *n;
2853         struct ext4_li_request *elr;
2854         unsigned long next_wakeup, cur;
2855
2856         BUG_ON(NULL == eli);
2857
2858 cont_thread:
2859         while (true) {
2860                 next_wakeup = MAX_JIFFY_OFFSET;
2861
2862                 mutex_lock(&eli->li_list_mtx);
2863                 if (list_empty(&eli->li_request_list)) {
2864                         mutex_unlock(&eli->li_list_mtx);
2865                         goto exit_thread;
2866                 }
2867
2868                 list_for_each_safe(pos, n, &eli->li_request_list) {
2869                         elr = list_entry(pos, struct ext4_li_request,
2870                                          lr_request);
2871
2872                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2873                                 if (ext4_run_li_request(elr) != 0) {
2874                                         /* error, remove the lazy_init job */
2875                                         ext4_remove_li_request(elr);
2876                                         continue;
2877                                 }
2878                         }
2879
2880                         if (time_before(elr->lr_next_sched, next_wakeup))
2881                                 next_wakeup = elr->lr_next_sched;
2882                 }
2883                 mutex_unlock(&eli->li_list_mtx);
2884
2885                 try_to_freeze();
2886
2887                 cur = jiffies;
2888                 if ((time_after_eq(cur, next_wakeup)) ||
2889                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2890                         cond_resched();
2891                         continue;
2892                 }
2893
2894                 schedule_timeout_interruptible(next_wakeup - cur);
2895
2896                 if (kthread_should_stop()) {
2897                         ext4_clear_request_list();
2898                         goto exit_thread;
2899                 }
2900         }
2901
2902 exit_thread:
2903         /*
2904          * It looks like the request list is empty, but we need
2905          * to check it under the li_list_mtx lock, to prevent any
2906          * additions into it, and of course we should lock ext4_li_mtx
2907          * to atomically free the list and ext4_li_info, because at
2908          * this point another ext4 filesystem could be registering
2909          * new one.
2910          */
2911         mutex_lock(&ext4_li_mtx);
2912         mutex_lock(&eli->li_list_mtx);
2913         if (!list_empty(&eli->li_request_list)) {
2914                 mutex_unlock(&eli->li_list_mtx);
2915                 mutex_unlock(&ext4_li_mtx);
2916                 goto cont_thread;
2917         }
2918         mutex_unlock(&eli->li_list_mtx);
2919         kfree(ext4_li_info);
2920         ext4_li_info = NULL;
2921         mutex_unlock(&ext4_li_mtx);
2922
2923         return 0;
2924 }
2925
2926 static void ext4_clear_request_list(void)
2927 {
2928         struct list_head *pos, *n;
2929         struct ext4_li_request *elr;
2930
2931         mutex_lock(&ext4_li_info->li_list_mtx);
2932         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2933                 elr = list_entry(pos, struct ext4_li_request,
2934                                  lr_request);
2935                 ext4_remove_li_request(elr);
2936         }
2937         mutex_unlock(&ext4_li_info->li_list_mtx);
2938 }
2939
2940 static int ext4_run_lazyinit_thread(void)
2941 {
2942         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2943                                          ext4_li_info, "ext4lazyinit");
2944         if (IS_ERR(ext4_lazyinit_task)) {
2945                 int err = PTR_ERR(ext4_lazyinit_task);
2946                 ext4_clear_request_list();
2947                 kfree(ext4_li_info);
2948                 ext4_li_info = NULL;
2949                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2950                                  "initialization thread\n",
2951                                  err);
2952                 return err;
2953         }
2954         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2955         return 0;
2956 }
2957
2958 /*
2959  * Check whether it make sense to run itable init. thread or not.
2960  * If there is at least one uninitialized inode table, return
2961  * corresponding group number, else the loop goes through all
2962  * groups and return total number of groups.
2963  */
2964 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2965 {
2966         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2967         struct ext4_group_desc *gdp = NULL;
2968
2969         for (group = 0; group < ngroups; group++) {
2970                 gdp = ext4_get_group_desc(sb, group, NULL);
2971                 if (!gdp)
2972                         continue;
2973
2974                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2975                         break;
2976         }
2977
2978         return group;
2979 }
2980
2981 static int ext4_li_info_new(void)
2982 {
2983         struct ext4_lazy_init *eli = NULL;
2984
2985         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2986         if (!eli)
2987                 return -ENOMEM;
2988
2989         INIT_LIST_HEAD(&eli->li_request_list);
2990         mutex_init(&eli->li_list_mtx);
2991
2992         eli->li_state |= EXT4_LAZYINIT_QUIT;
2993
2994         ext4_li_info = eli;
2995
2996         return 0;
2997 }
2998
2999 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3000                                             ext4_group_t start)
3001 {
3002         struct ext4_sb_info *sbi = EXT4_SB(sb);
3003         struct ext4_li_request *elr;
3004         unsigned long rnd;
3005
3006         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3007         if (!elr)
3008                 return NULL;
3009
3010         elr->lr_super = sb;
3011         elr->lr_sbi = sbi;
3012         elr->lr_next_group = start;
3013
3014         /*
3015          * Randomize first schedule time of the request to
3016          * spread the inode table initialization requests
3017          * better.
3018          */
3019         get_random_bytes(&rnd, sizeof(rnd));
3020         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3021                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3022
3023         return elr;
3024 }
3025
3026 static int ext4_register_li_request(struct super_block *sb,
3027                                     ext4_group_t first_not_zeroed)
3028 {
3029         struct ext4_sb_info *sbi = EXT4_SB(sb);
3030         struct ext4_li_request *elr;
3031         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3032         int ret = 0;
3033
3034         if (sbi->s_li_request != NULL) {
3035                 /*
3036                  * Reset timeout so it can be computed again, because
3037                  * s_li_wait_mult might have changed.
3038                  */
3039                 sbi->s_li_request->lr_timeout = 0;
3040                 return 0;
3041         }
3042
3043         if (first_not_zeroed == ngroups ||
3044             (sb->s_flags & MS_RDONLY) ||
3045             !test_opt(sb, INIT_INODE_TABLE))
3046                 return 0;
3047
3048         elr = ext4_li_request_new(sb, first_not_zeroed);
3049         if (!elr)
3050                 return -ENOMEM;
3051
3052         mutex_lock(&ext4_li_mtx);
3053
3054         if (NULL == ext4_li_info) {
3055                 ret = ext4_li_info_new();
3056                 if (ret)
3057                         goto out;
3058         }
3059
3060         mutex_lock(&ext4_li_info->li_list_mtx);
3061         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3062         mutex_unlock(&ext4_li_info->li_list_mtx);
3063
3064         sbi->s_li_request = elr;
3065         /*
3066          * set elr to NULL here since it has been inserted to
3067          * the request_list and the removal and free of it is
3068          * handled by ext4_clear_request_list from now on.
3069          */
3070         elr = NULL;
3071
3072         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3073                 ret = ext4_run_lazyinit_thread();
3074                 if (ret)
3075                         goto out;
3076         }
3077 out:
3078         mutex_unlock(&ext4_li_mtx);
3079         if (ret)
3080                 kfree(elr);
3081         return ret;
3082 }
3083
3084 /*
3085  * We do not need to lock anything since this is called on
3086  * module unload.
3087  */
3088 static void ext4_destroy_lazyinit_thread(void)
3089 {
3090         /*
3091          * If thread exited earlier
3092          * there's nothing to be done.
3093          */
3094         if (!ext4_li_info || !ext4_lazyinit_task)
3095                 return;
3096
3097         kthread_stop(ext4_lazyinit_task);
3098 }
3099
3100 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3101 {
3102         char *orig_data = kstrdup(data, GFP_KERNEL);
3103         struct buffer_head *bh;
3104         struct ext4_super_block *es = NULL;
3105         struct ext4_sb_info *sbi;
3106         ext4_fsblk_t block;
3107         ext4_fsblk_t sb_block = get_sb_block(&data);
3108         ext4_fsblk_t logical_sb_block;
3109         unsigned long offset = 0;
3110         unsigned long journal_devnum = 0;
3111         unsigned long def_mount_opts;
3112         struct inode *root;
3113         char *cp;
3114         const char *descr;
3115         int ret = -ENOMEM;
3116         int blocksize, clustersize;
3117         unsigned int db_count;
3118         unsigned int i;
3119         int needs_recovery, has_huge_files, has_bigalloc;
3120         __u64 blocks_count;
3121         int err;
3122         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3123         ext4_group_t first_not_zeroed;
3124
3125         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3126         if (!sbi)
3127                 goto out_free_orig;
3128
3129         sbi->s_blockgroup_lock =
3130                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3131         if (!sbi->s_blockgroup_lock) {
3132                 kfree(sbi);
3133                 goto out_free_orig;
3134         }
3135         sb->s_fs_info = sbi;
3136         sbi->s_mount_opt = 0;
3137         sbi->s_resuid = EXT4_DEF_RESUID;
3138         sbi->s_resgid = EXT4_DEF_RESGID;
3139         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3140         sbi->s_sb_block = sb_block;
3141         if (sb->s_bdev->bd_part)
3142                 sbi->s_sectors_written_start =
3143                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3144
3145         /* Cleanup superblock name */
3146         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3147                 *cp = '!';
3148
3149         ret = -EINVAL;
3150         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3151         if (!blocksize) {
3152                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3153                 goto out_fail;
3154         }
3155
3156         /*
3157          * The ext4 superblock will not be buffer aligned for other than 1kB
3158          * block sizes.  We need to calculate the offset from buffer start.
3159          */
3160         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3161                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3162                 offset = do_div(logical_sb_block, blocksize);
3163         } else {
3164                 logical_sb_block = sb_block;
3165         }
3166
3167         if (!(bh = sb_bread(sb, logical_sb_block))) {
3168                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3169                 goto out_fail;
3170         }
3171         /*
3172          * Note: s_es must be initialized as soon as possible because
3173          *       some ext4 macro-instructions depend on its value
3174          */
3175         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3176         sbi->s_es = es;
3177         sb->s_magic = le16_to_cpu(es->s_magic);
3178         if (sb->s_magic != EXT4_SUPER_MAGIC)
3179                 goto cantfind_ext4;
3180         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3181
3182         /* Set defaults before we parse the mount options */
3183         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3184         set_opt(sb, INIT_INODE_TABLE);
3185         if (def_mount_opts & EXT4_DEFM_DEBUG)
3186                 set_opt(sb, DEBUG);
3187         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3188                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3189                         "2.6.38");
3190                 set_opt(sb, GRPID);
3191         }
3192         if (def_mount_opts & EXT4_DEFM_UID16)
3193                 set_opt(sb, NO_UID32);
3194         /* xattr user namespace & acls are now defaulted on */
3195 #ifdef CONFIG_EXT4_FS_XATTR
3196         set_opt(sb, XATTR_USER);
3197 #endif
3198 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3199         set_opt(sb, POSIX_ACL);
3200 #endif
3201         set_opt(sb, MBLK_IO_SUBMIT);
3202         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3203                 set_opt(sb, JOURNAL_DATA);
3204         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3205                 set_opt(sb, ORDERED_DATA);
3206         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3207                 set_opt(sb, WRITEBACK_DATA);
3208
3209         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3210                 set_opt(sb, ERRORS_PANIC);
3211         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3212                 set_opt(sb, ERRORS_CONT);
3213         else
3214                 set_opt(sb, ERRORS_RO);
3215         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3216                 set_opt(sb, BLOCK_VALIDITY);
3217         if (def_mount_opts & EXT4_DEFM_DISCARD)
3218                 set_opt(sb, DISCARD);
3219
3220         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3221         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3222         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3223         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3224         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3225
3226         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3227                 set_opt(sb, BARRIER);
3228
3229         /*
3230          * enable delayed allocation by default
3231          * Use -o nodelalloc to turn it off
3232          */
3233         if (!IS_EXT3_SB(sb) &&
3234             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3235                 set_opt(sb, DELALLOC);
3236
3237         /*
3238          * set default s_li_wait_mult for lazyinit, for the case there is
3239          * no mount option specified.
3240          */
3241         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3242
3243         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3244                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3245                 ext4_msg(sb, KERN_WARNING,
3246                          "failed to parse options in superblock: %s",
3247                          sbi->s_es->s_mount_opts);
3248         }
3249         if (!parse_options((char *) data, sb, &journal_devnum,
3250                            &journal_ioprio, NULL, 0))
3251                 goto failed_mount;
3252
3253         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3254                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3255                             "with data=journal disables delayed "
3256                             "allocation and O_DIRECT support!\n");
3257                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3258                         ext4_msg(sb, KERN_ERR, "can't mount with "
3259                                  "both data=journal and delalloc");
3260                         goto failed_mount;
3261                 }
3262                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3263                         ext4_msg(sb, KERN_ERR, "can't mount with "
3264                                  "both data=journal and delalloc");
3265                         goto failed_mount;
3266                 }
3267                 if (test_opt(sb, DELALLOC))
3268                         clear_opt(sb, DELALLOC);
3269         }
3270
3271         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3272         if (test_opt(sb, DIOREAD_NOLOCK)) {
3273                 if (blocksize < PAGE_SIZE) {
3274                         ext4_msg(sb, KERN_ERR, "can't mount with "
3275                                  "dioread_nolock if block size != PAGE_SIZE");
3276                         goto failed_mount;
3277                 }
3278         }
3279
3280         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3281                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3282
3283         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3284             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3285              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3286              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3287                 ext4_msg(sb, KERN_WARNING,
3288                        "feature flags set on rev 0 fs, "
3289                        "running e2fsck is recommended");
3290
3291         if (IS_EXT2_SB(sb)) {
3292                 if (ext2_feature_set_ok(sb))
3293                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3294                                  "using the ext4 subsystem");
3295                 else {
3296                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3297                                  "to feature incompatibilities");
3298                         goto failed_mount;
3299                 }
3300         }
3301
3302         if (IS_EXT3_SB(sb)) {
3303                 if (ext3_feature_set_ok(sb))
3304                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3305                                  "using the ext4 subsystem");
3306                 else {
3307                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3308                                  "to feature incompatibilities");
3309                         goto failed_mount;
3310                 }
3311         }
3312
3313         /*
3314          * Check feature flags regardless of the revision level, since we
3315          * previously didn't change the revision level when setting the flags,
3316          * so there is a chance incompat flags are set on a rev 0 filesystem.
3317          */
3318         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3319                 goto failed_mount;
3320
3321         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3322             blocksize > EXT4_MAX_BLOCK_SIZE) {
3323                 ext4_msg(sb, KERN_ERR,
3324                        "Unsupported filesystem blocksize %d", blocksize);
3325                 goto failed_mount;
3326         }
3327
3328         if (sb->s_blocksize != blocksize) {
3329                 /* Validate the filesystem blocksize */
3330                 if (!sb_set_blocksize(sb, blocksize)) {
3331                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3332                                         blocksize);
3333                         goto failed_mount;
3334                 }
3335
3336                 brelse(bh);
3337                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3338                 offset = do_div(logical_sb_block, blocksize);
3339                 bh = sb_bread(sb, logical_sb_block);
3340                 if (!bh) {
3341                         ext4_msg(sb, KERN_ERR,
3342                                "Can't read superblock on 2nd try");
3343                         goto failed_mount;
3344                 }
3345                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3346                 sbi->s_es = es;
3347                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3348                         ext4_msg(sb, KERN_ERR,
3349                                "Magic mismatch, very weird!");
3350                         goto failed_mount;
3351                 }
3352         }
3353
3354         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3355                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3356         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3357                                                       has_huge_files);
3358         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3359
3360         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3361                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3362                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3363         } else {
3364                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3365                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3366                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3367                     (!is_power_of_2(sbi->s_inode_size)) ||
3368                     (sbi->s_inode_size > blocksize)) {
3369                         ext4_msg(sb, KERN_ERR,
3370                                "unsupported inode size: %d",
3371                                sbi->s_inode_size);
3372                         goto failed_mount;
3373                 }
3374                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3375                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3376         }
3377
3378         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3379         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3380                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3381                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3382                     !is_power_of_2(sbi->s_desc_size)) {
3383                         ext4_msg(sb, KERN_ERR,
3384                                "unsupported descriptor size %lu",
3385                                sbi->s_desc_size);
3386                         goto failed_mount;
3387                 }
3388         } else
3389                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3390
3391         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3392         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3393         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3394                 goto cantfind_ext4;
3395
3396         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3397         if (sbi->s_inodes_per_block == 0)
3398                 goto cantfind_ext4;
3399         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3400                                         sbi->s_inodes_per_block;
3401         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3402         sbi->s_sbh = bh;
3403         sbi->s_mount_state = le16_to_cpu(es->s_state);
3404         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3405         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3406
3407         for (i = 0; i < 4; i++)
3408                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3409         sbi->s_def_hash_version = es->s_def_hash_version;
3410         i = le32_to_cpu(es->s_flags);
3411         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3412                 sbi->s_hash_unsigned = 3;
3413         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3414 #ifdef __CHAR_UNSIGNED__
3415                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3416                 sbi->s_hash_unsigned = 3;
3417 #else
3418                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3419 #endif
3420                 sb->s_dirt = 1;
3421         }
3422
3423         /* Handle clustersize */
3424         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3425         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3426                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3427         if (has_bigalloc) {
3428                 if (clustersize < blocksize) {
3429                         ext4_msg(sb, KERN_ERR,
3430                                  "cluster size (%d) smaller than "
3431                                  "block size (%d)", clustersize, blocksize);
3432                         goto failed_mount;
3433                 }
3434                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3435                         le32_to_cpu(es->s_log_block_size);
3436                 sbi->s_clusters_per_group =
3437                         le32_to_cpu(es->s_clusters_per_group);
3438                 if (sbi->s_clusters_per_group > blocksize * 8) {
3439                         ext4_msg(sb, KERN_ERR,
3440                                  "#clusters per group too big: %lu",
3441                                  sbi->s_clusters_per_group);
3442                         goto failed_mount;
3443                 }
3444                 if (sbi->s_blocks_per_group !=
3445                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3446                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3447                                  "clusters per group (%lu) inconsistent",
3448                                  sbi->s_blocks_per_group,
3449                                  sbi->s_clusters_per_group);
3450                         goto failed_mount;
3451                 }
3452         } else {
3453                 if (clustersize != blocksize) {
3454                         ext4_warning(sb, "fragment/cluster size (%d) != "
3455                                      "block size (%d)", clustersize,
3456                                      blocksize);
3457                         clustersize = blocksize;
3458                 }
3459                 if (sbi->s_blocks_per_group > blocksize * 8) {
3460                         ext4_msg(sb, KERN_ERR,
3461                                  "#blocks per group too big: %lu",
3462                                  sbi->s_blocks_per_group);
3463                         goto failed_mount;
3464                 }
3465                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3466                 sbi->s_cluster_bits = 0;
3467         }
3468         sbi->s_cluster_ratio = clustersize / blocksize;
3469
3470         if (sbi->s_inodes_per_group > blocksize * 8) {
3471                 ext4_msg(sb, KERN_ERR,
3472                        "#inodes per group too big: %lu",
3473                        sbi->s_inodes_per_group);
3474                 goto failed_mount;
3475         }
3476
3477         /*
3478          * Test whether we have more sectors than will fit in sector_t,
3479          * and whether the max offset is addressable by the page cache.
3480          */
3481         err = generic_check_addressable(sb->s_blocksize_bits,
3482                                         ext4_blocks_count(es));
3483         if (err) {
3484                 ext4_msg(sb, KERN_ERR, "filesystem"
3485                          " too large to mount safely on this system");
3486                 if (sizeof(sector_t) < 8)
3487                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3488                 ret = err;
3489                 goto failed_mount;
3490         }
3491
3492         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3493                 goto cantfind_ext4;
3494
3495         /* check blocks count against device size */
3496         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3497         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3498                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3499                        "exceeds size of device (%llu blocks)",
3500                        ext4_blocks_count(es), blocks_count);
3501                 goto failed_mount;
3502         }
3503
3504         /*
3505          * It makes no sense for the first data block to be beyond the end
3506          * of the filesystem.
3507          */
3508         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3509                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3510                          "block %u is beyond end of filesystem (%llu)",
3511                          le32_to_cpu(es->s_first_data_block),
3512                          ext4_blocks_count(es));
3513                 goto failed_mount;
3514         }
3515         blocks_count = (ext4_blocks_count(es) -
3516                         le32_to_cpu(es->s_first_data_block) +
3517                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3518         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3519         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3520                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3521                        "(block count %llu, first data block %u, "
3522                        "blocks per group %lu)", sbi->s_groups_count,
3523                        ext4_blocks_count(es),
3524                        le32_to_cpu(es->s_first_data_block),
3525                        EXT4_BLOCKS_PER_GROUP(sb));
3526                 goto failed_mount;
3527         }
3528         sbi->s_groups_count = blocks_count;
3529         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3530                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3531         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3532                    EXT4_DESC_PER_BLOCK(sb);
3533         sbi->s_group_desc = ext4_kvmalloc(db_count *
3534                                           sizeof(struct buffer_head *),
3535                                           GFP_KERNEL);
3536         if (sbi->s_group_desc == NULL) {
3537                 ext4_msg(sb, KERN_ERR, "not enough memory");
3538                 goto failed_mount;
3539         }
3540
3541         if (ext4_proc_root)
3542                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3543
3544         bgl_lock_init(sbi->s_blockgroup_lock);
3545
3546         for (i = 0; i < db_count; i++) {
3547                 block = descriptor_loc(sb, logical_sb_block, i);
3548                 sbi->s_group_desc[i] = sb_bread(sb, block);
3549                 if (!sbi->s_group_desc[i]) {
3550                         ext4_msg(sb, KERN_ERR,
3551                                "can't read group descriptor %d", i);
3552                         db_count = i;
3553                         goto failed_mount2;
3554                 }
3555         }
3556         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3557                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3558                 goto failed_mount2;
3559         }
3560         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3561                 if (!ext4_fill_flex_info(sb)) {
3562                         ext4_msg(sb, KERN_ERR,
3563                                "unable to initialize "
3564                                "flex_bg meta info!");
3565                         goto failed_mount2;
3566                 }
3567
3568         sbi->s_gdb_count = db_count;
3569         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3570         spin_lock_init(&sbi->s_next_gen_lock);
3571
3572         init_timer(&sbi->s_err_report);
3573         sbi->s_err_report.function = print_daily_error_info;
3574         sbi->s_err_report.data = (unsigned long) sb;
3575
3576         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3577                         ext4_count_free_clusters(sb));
3578         if (!err) {
3579                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3580                                 ext4_count_free_inodes(sb));
3581         }
3582         if (!err) {
3583                 err = percpu_counter_init(&sbi->s_dirs_counter,
3584                                 ext4_count_dirs(sb));
3585         }
3586         if (!err) {
3587                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3588         }
3589         if (err) {
3590                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3591                 goto failed_mount3;
3592         }
3593
3594         sbi->s_stripe = ext4_get_stripe_size(sbi);
3595         sbi->s_max_writeback_mb_bump = 128;
3596
3597         /*
3598          * set up enough so that it can read an inode
3599          */
3600         if (!test_opt(sb, NOLOAD) &&
3601             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3602                 sb->s_op = &ext4_sops;
3603         else
3604                 sb->s_op = &ext4_nojournal_sops;
3605         sb->s_export_op = &ext4_export_ops;
3606         sb->s_xattr = ext4_xattr_handlers;
3607 #ifdef CONFIG_QUOTA
3608         sb->s_qcop = &ext4_qctl_operations;
3609         sb->dq_op = &ext4_quota_operations;
3610 #endif
3611         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3612
3613         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3614         mutex_init(&sbi->s_orphan_lock);
3615         sbi->s_resize_flags = 0;
3616
3617         sb->s_root = NULL;
3618
3619         needs_recovery = (es->s_last_orphan != 0 ||
3620                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3621                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3622
3623         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3624             !(sb->s_flags & MS_RDONLY))
3625                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3626                         goto failed_mount3;
3627
3628         /*
3629          * The first inode we look at is the journal inode.  Don't try
3630          * root first: it may be modified in the journal!
3631          */
3632         if (!test_opt(sb, NOLOAD) &&
3633             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3634                 if (ext4_load_journal(sb, es, journal_devnum))
3635                         goto failed_mount3;
3636         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3637               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3638                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3639                        "suppressed and not mounted read-only");
3640                 goto failed_mount_wq;
3641         } else {
3642                 clear_opt(sb, DATA_FLAGS);
3643                 sbi->s_journal = NULL;
3644                 needs_recovery = 0;
3645                 goto no_journal;
3646         }
3647
3648         if (ext4_blocks_count(es) > 0xffffffffULL &&
3649             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3650                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3651                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3652                 goto failed_mount_wq;
3653         }
3654
3655         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3656                 jbd2_journal_set_features(sbi->s_journal,
3657                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3658                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3659         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3660                 jbd2_journal_set_features(sbi->s_journal,
3661                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3662                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3663                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3664         } else {
3665                 jbd2_journal_clear_features(sbi->s_journal,
3666                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3667                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3668         }
3669
3670         /* We have now updated the journal if required, so we can
3671          * validate the data journaling mode. */
3672         switch (test_opt(sb, DATA_FLAGS)) {
3673         case 0:
3674                 /* No mode set, assume a default based on the journal
3675                  * capabilities: ORDERED_DATA if the journal can
3676                  * cope, else JOURNAL_DATA
3677                  */
3678                 if (jbd2_journal_check_available_features
3679                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3680                         set_opt(sb, ORDERED_DATA);
3681                 else
3682                         set_opt(sb, JOURNAL_DATA);
3683                 break;
3684
3685         case EXT4_MOUNT_ORDERED_DATA:
3686         case EXT4_MOUNT_WRITEBACK_DATA:
3687                 if (!jbd2_journal_check_available_features
3688                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3689                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3690                                "requested data journaling mode");
3691                         goto failed_mount_wq;
3692                 }
3693         default:
3694                 break;
3695         }
3696         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3697
3698         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3699
3700         /*
3701          * The journal may have updated the bg summary counts, so we
3702          * need to update the global counters.
3703          */
3704         percpu_counter_set(&sbi->s_freeclusters_counter,
3705                            ext4_count_free_clusters(sb));
3706         percpu_counter_set(&sbi->s_freeinodes_counter,
3707                            ext4_count_free_inodes(sb));
3708         percpu_counter_set(&sbi->s_dirs_counter,
3709                            ext4_count_dirs(sb));
3710         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3711
3712 no_journal:
3713         /*
3714          * The maximum number of concurrent works can be high and
3715          * concurrency isn't really necessary.  Limit it to 1.
3716          */
3717         EXT4_SB(sb)->dio_unwritten_wq =
3718                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3719         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3720                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3721                 goto failed_mount_wq;
3722         }
3723
3724         /*
3725          * The jbd2_journal_load will have done any necessary log recovery,
3726          * so we can safely mount the rest of the filesystem now.
3727          */
3728
3729         root = ext4_iget(sb, EXT4_ROOT_INO);
3730         if (IS_ERR(root)) {
3731                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3732                 ret = PTR_ERR(root);
3733                 root = NULL;
3734                 goto failed_mount4;
3735         }
3736         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3737                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3738                 iput(root);
3739                 goto failed_mount4;
3740         }
3741         sb->s_root = d_alloc_root(root);
3742         if (!sb->s_root) {
3743                 iput(root);
3744                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3745                 ret = -ENOMEM;
3746                 goto failed_mount4;
3747         }
3748
3749         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3750
3751         /* determine the minimum size of new large inodes, if present */
3752         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3753                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3754                                                      EXT4_GOOD_OLD_INODE_SIZE;
3755                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3756                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3757                         if (sbi->s_want_extra_isize <
3758                             le16_to_cpu(es->s_want_extra_isize))
3759                                 sbi->s_want_extra_isize =
3760                                         le16_to_cpu(es->s_want_extra_isize);
3761                         if (sbi->s_want_extra_isize <
3762                             le16_to_cpu(es->s_min_extra_isize))
3763                                 sbi->s_want_extra_isize =
3764                                         le16_to_cpu(es->s_min_extra_isize);
3765                 }
3766         }
3767         /* Check if enough inode space is available */
3768         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3769                                                         sbi->s_inode_size) {
3770                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3771                                                        EXT4_GOOD_OLD_INODE_SIZE;
3772                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3773                          "available");
3774         }
3775
3776         err = ext4_setup_system_zone(sb);
3777         if (err) {
3778                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3779                          "zone (%d)", err);
3780                 goto failed_mount4a;
3781         }
3782
3783         ext4_ext_init(sb);
3784         err = ext4_mb_init(sb, needs_recovery);
3785         if (err) {
3786                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3787                          err);
3788                 goto failed_mount5;
3789         }
3790
3791         err = ext4_register_li_request(sb, first_not_zeroed);
3792         if (err)
3793                 goto failed_mount6;
3794
3795         sbi->s_kobj.kset = ext4_kset;
3796         init_completion(&sbi->s_kobj_unregister);
3797         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3798                                    "%s", sb->s_id);
3799         if (err)
3800                 goto failed_mount7;
3801
3802         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3803         ext4_orphan_cleanup(sb, es);
3804         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3805         if (needs_recovery) {
3806                 ext4_msg(sb, KERN_INFO, "recovery complete");
3807                 ext4_mark_recovery_complete(sb, es);
3808         }
3809         if (EXT4_SB(sb)->s_journal) {
3810                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3811                         descr = " journalled data mode";
3812                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3813                         descr = " ordered data mode";
3814                 else
3815                         descr = " writeback data mode";
3816         } else
3817                 descr = "out journal";
3818
3819         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3820                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3821                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3822
3823         if (es->s_error_count)
3824                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3825
3826         kfree(orig_data);
3827         return 0;
3828
3829 cantfind_ext4:
3830         if (!silent)
3831                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3832         goto failed_mount;
3833
3834 failed_mount7:
3835         ext4_unregister_li_request(sb);
3836 failed_mount6:
3837         ext4_mb_release(sb);
3838 failed_mount5:
3839         ext4_ext_release(sb);
3840         ext4_release_system_zone(sb);
3841 failed_mount4a:
3842         dput(sb->s_root);
3843         sb->s_root = NULL;
3844 failed_mount4:
3845         ext4_msg(sb, KERN_ERR, "mount failed");
3846         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3847 failed_mount_wq:
3848         if (sbi->s_journal) {
3849                 jbd2_journal_destroy(sbi->s_journal);
3850                 sbi->s_journal = NULL;
3851         }
3852 failed_mount3:
3853         del_timer(&sbi->s_err_report);
3854         if (sbi->s_flex_groups)
3855                 ext4_kvfree(sbi->s_flex_groups);
3856         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3857         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3858         percpu_counter_destroy(&sbi->s_dirs_counter);
3859         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3860         if (sbi->s_mmp_tsk)
3861                 kthread_stop(sbi->s_mmp_tsk);
3862 failed_mount2:
3863         for (i = 0; i < db_count; i++)
3864                 brelse(sbi->s_group_desc[i]);
3865         ext4_kvfree(sbi->s_group_desc);
3866 failed_mount:
3867         if (sbi->s_proc) {
3868                 remove_proc_entry(sb->s_id, ext4_proc_root);
3869         }
3870 #ifdef CONFIG_QUOTA
3871         for (i = 0; i < MAXQUOTAS; i++)
3872                 kfree(sbi->s_qf_names[i]);
3873 #endif
3874         ext4_blkdev_remove(sbi);
3875         brelse(bh);
3876 out_fail:
3877         sb->s_fs_info = NULL;
3878         kfree(sbi->s_blockgroup_lock);
3879         kfree(sbi);
3880 out_free_orig:
3881         kfree(orig_data);
3882         return ret;
3883 }
3884
3885 /*
3886  * Setup any per-fs journal parameters now.  We'll do this both on
3887  * initial mount, once the journal has been initialised but before we've
3888  * done any recovery; and again on any subsequent remount.
3889  */
3890 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3891 {
3892         struct ext4_sb_info *sbi = EXT4_SB(sb);
3893
3894         journal->j_commit_interval = sbi->s_commit_interval;
3895         journal->j_min_batch_time = sbi->s_min_batch_time;
3896         journal->j_max_batch_time = sbi->s_max_batch_time;
3897
3898         write_lock(&journal->j_state_lock);
3899         if (test_opt(sb, BARRIER))
3900                 journal->j_flags |= JBD2_BARRIER;
3901         else
3902                 journal->j_flags &= ~JBD2_BARRIER;
3903         if (test_opt(sb, DATA_ERR_ABORT))
3904                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3905         else
3906                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3907         write_unlock(&journal->j_state_lock);
3908 }
3909
3910 static journal_t *ext4_get_journal(struct super_block *sb,
3911                                    unsigned int journal_inum)
3912 {
3913         struct inode *journal_inode;
3914         journal_t *journal;
3915
3916         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3917
3918         /* First, test for the existence of a valid inode on disk.  Bad
3919          * things happen if we iget() an unused inode, as the subsequent
3920          * iput() will try to delete it. */
3921
3922         journal_inode = ext4_iget(sb, journal_inum);
3923         if (IS_ERR(journal_inode)) {
3924                 ext4_msg(sb, KERN_ERR, "no journal found");
3925                 return NULL;
3926         }
3927         if (!journal_inode->i_nlink) {
3928                 make_bad_inode(journal_inode);
3929                 iput(journal_inode);
3930                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3931                 return NULL;
3932         }
3933
3934         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3935                   journal_inode, journal_inode->i_size);
3936         if (!S_ISREG(journal_inode->i_mode)) {
3937                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3938                 iput(journal_inode);
3939                 return NULL;
3940         }
3941
3942         journal = jbd2_journal_init_inode(journal_inode);
3943         if (!journal) {
3944                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3945                 iput(journal_inode);
3946                 return NULL;
3947         }
3948         journal->j_private = sb;
3949         ext4_init_journal_params(sb, journal);
3950         return journal;
3951 }
3952
3953 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3954                                        dev_t j_dev)
3955 {
3956         struct buffer_head *bh;
3957         journal_t *journal;
3958         ext4_fsblk_t start;
3959         ext4_fsblk_t len;
3960         int hblock, blocksize;
3961         ext4_fsblk_t sb_block;
3962         unsigned long offset;
3963         struct ext4_super_block *es;
3964         struct block_device *bdev;
3965
3966         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3967
3968         bdev = ext4_blkdev_get(j_dev, sb);
3969         if (bdev == NULL)
3970                 return NULL;
3971
3972         blocksize = sb->s_blocksize;
3973         hblock = bdev_logical_block_size(bdev);
3974         if (blocksize < hblock) {
3975                 ext4_msg(sb, KERN_ERR,
3976                         "blocksize too small for journal device");
3977                 goto out_bdev;
3978         }
3979
3980         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3981         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3982         set_blocksize(bdev, blocksize);
3983         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3984                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3985                        "external journal");
3986                 goto out_bdev;
3987         }
3988
3989         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3990         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3991             !(le32_to_cpu(es->s_feature_incompat) &
3992               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3993                 ext4_msg(sb, KERN_ERR, "external journal has "
3994                                         "bad superblock");
3995                 brelse(bh);
3996                 goto out_bdev;
3997         }
3998
3999         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4000                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4001                 brelse(bh);
4002                 goto out_bdev;
4003         }
4004
4005         len = ext4_blocks_count(es);
4006         start = sb_block + 1;
4007         brelse(bh);     /* we're done with the superblock */
4008
4009         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4010                                         start, len, blocksize);
4011         if (!journal) {
4012                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4013                 goto out_bdev;
4014         }
4015         journal->j_private = sb;
4016         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4017         wait_on_buffer(journal->j_sb_buffer);
4018         if (!buffer_uptodate(journal->j_sb_buffer)) {
4019                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4020                 goto out_journal;
4021         }
4022         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4023                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4024                                         "user (unsupported) - %d",
4025                         be32_to_cpu(journal->j_superblock->s_nr_users));
4026                 goto out_journal;
4027         }
4028         EXT4_SB(sb)->journal_bdev = bdev;
4029         ext4_init_journal_params(sb, journal);
4030         return journal;
4031
4032 out_journal:
4033         jbd2_journal_destroy(journal);
4034 out_bdev:
4035         ext4_blkdev_put(bdev);
4036         return NULL;
4037 }
4038
4039 static int ext4_load_journal(struct super_block *sb,
4040                              struct ext4_super_block *es,
4041                              unsigned long journal_devnum)
4042 {
4043         journal_t *journal;
4044         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4045         dev_t journal_dev;
4046         int err = 0;
4047         int really_read_only;
4048
4049         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4050
4051         if (journal_devnum &&
4052             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4053                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4054                         "numbers have changed");
4055                 journal_dev = new_decode_dev(journal_devnum);
4056         } else
4057                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4058
4059         really_read_only = bdev_read_only(sb->s_bdev);
4060
4061         /*
4062          * Are we loading a blank journal or performing recovery after a
4063          * crash?  For recovery, we need to check in advance whether we
4064          * can get read-write access to the device.
4065          */
4066         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4067                 if (sb->s_flags & MS_RDONLY) {
4068                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4069                                         "required on readonly filesystem");
4070                         if (really_read_only) {
4071                                 ext4_msg(sb, KERN_ERR, "write access "
4072                                         "unavailable, cannot proceed");
4073                                 return -EROFS;
4074                         }
4075                         ext4_msg(sb, KERN_INFO, "write access will "
4076                                "be enabled during recovery");
4077                 }
4078         }
4079
4080         if (journal_inum && journal_dev) {
4081                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4082                        "and inode journals!");
4083                 return -EINVAL;
4084         }
4085
4086         if (journal_inum) {
4087                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4088                         return -EINVAL;
4089         } else {
4090                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4091                         return -EINVAL;
4092         }
4093
4094         if (!(journal->j_flags & JBD2_BARRIER))
4095                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4096
4097         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4098                 err = jbd2_journal_wipe(journal, !really_read_only);
4099         if (!err) {
4100                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4101                 if (save)
4102                         memcpy(save, ((char *) es) +
4103                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4104                 err = jbd2_journal_load(journal);
4105                 if (save)
4106                         memcpy(((char *) es) + EXT4_S_ERR_START,
4107                                save, EXT4_S_ERR_LEN);
4108                 kfree(save);
4109         }
4110
4111         if (err) {
4112                 ext4_msg(sb, KERN_ERR, "error loading journal");
4113                 jbd2_journal_destroy(journal);
4114                 return err;
4115         }
4116
4117         EXT4_SB(sb)->s_journal = journal;
4118         ext4_clear_journal_err(sb, es);
4119
4120         if (!really_read_only && journal_devnum &&
4121             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4122                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4123
4124                 /* Make sure we flush the recovery flag to disk. */
4125                 ext4_commit_super(sb, 1);
4126         }
4127
4128         return 0;
4129 }
4130
4131 static int ext4_commit_super(struct super_block *sb, int sync)
4132 {
4133         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4134         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4135         int error = 0;
4136
4137         if (!sbh || block_device_ejected(sb))
4138                 return error;
4139         if (buffer_write_io_error(sbh)) {
4140                 /*
4141                  * Oh, dear.  A previous attempt to write the
4142                  * superblock failed.  This could happen because the
4143                  * USB device was yanked out.  Or it could happen to
4144                  * be a transient write error and maybe the block will
4145                  * be remapped.  Nothing we can do but to retry the
4146                  * write and hope for the best.
4147                  */
4148                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4149                        "superblock detected");
4150                 clear_buffer_write_io_error(sbh);
4151                 set_buffer_uptodate(sbh);
4152         }
4153         /*
4154          * If the file system is mounted read-only, don't update the
4155          * superblock write time.  This avoids updating the superblock
4156          * write time when we are mounting the root file system
4157          * read/only but we need to replay the journal; at that point,
4158          * for people who are east of GMT and who make their clock
4159          * tick in localtime for Windows bug-for-bug compatibility,
4160          * the clock is set in the future, and this will cause e2fsck
4161          * to complain and force a full file system check.
4162          */
4163         if (!(sb->s_flags & MS_RDONLY))
4164                 es->s_wtime = cpu_to_le32(get_seconds());
4165         if (sb->s_bdev->bd_part)
4166                 es->s_kbytes_written =
4167                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4168                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4169                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4170         else
4171                 es->s_kbytes_written =
4172                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4173         ext4_free_blocks_count_set(es,
4174                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4175                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4176         es->s_free_inodes_count =
4177                 cpu_to_le32(percpu_counter_sum_positive(
4178                                 &EXT4_SB(sb)->s_freeinodes_counter));
4179         sb->s_dirt = 0;
4180         BUFFER_TRACE(sbh, "marking dirty");
4181         mark_buffer_dirty(sbh);
4182         if (sync) {
4183                 error = sync_dirty_buffer(sbh);
4184                 if (error)
4185                         return error;
4186
4187                 error = buffer_write_io_error(sbh);
4188                 if (error) {
4189                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4190                                "superblock");
4191                         clear_buffer_write_io_error(sbh);
4192                         set_buffer_uptodate(sbh);
4193                 }
4194         }
4195         return error;
4196 }
4197
4198 /*
4199  * Have we just finished recovery?  If so, and if we are mounting (or
4200  * remounting) the filesystem readonly, then we will end up with a
4201  * consistent fs on disk.  Record that fact.
4202  */
4203 static void ext4_mark_recovery_complete(struct super_block *sb,
4204                                         struct ext4_super_block *es)
4205 {
4206         journal_t *journal = EXT4_SB(sb)->s_journal;
4207
4208         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4209                 BUG_ON(journal != NULL);
4210                 return;
4211         }
4212         jbd2_journal_lock_updates(journal);
4213         if (jbd2_journal_flush(journal) < 0)
4214                 goto out;
4215
4216         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4217             sb->s_flags & MS_RDONLY) {
4218                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4219                 ext4_commit_super(sb, 1);
4220         }
4221
4222 out:
4223         jbd2_journal_unlock_updates(journal);
4224 }
4225
4226 /*
4227  * If we are mounting (or read-write remounting) a filesystem whose journal
4228  * has recorded an error from a previous lifetime, move that error to the
4229  * main filesystem now.
4230  */
4231 static void ext4_clear_journal_err(struct super_block *sb,
4232                                    struct ext4_super_block *es)
4233 {
4234         journal_t *journal;
4235         int j_errno;
4236         const char *errstr;
4237
4238         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4239
4240         journal = EXT4_SB(sb)->s_journal;
4241
4242         /*
4243          * Now check for any error status which may have been recorded in the
4244          * journal by a prior ext4_error() or ext4_abort()
4245          */
4246
4247         j_errno = jbd2_journal_errno(journal);
4248         if (j_errno) {
4249                 char nbuf[16];
4250
4251                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4252                 ext4_warning(sb, "Filesystem error recorded "
4253                              "from previous mount: %s", errstr);
4254                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4255
4256                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4257                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4258                 ext4_commit_super(sb, 1);
4259
4260                 jbd2_journal_clear_err(journal);
4261         }
4262 }
4263
4264 /*
4265  * Force the running and committing transactions to commit,
4266  * and wait on the commit.
4267  */
4268 int ext4_force_commit(struct super_block *sb)
4269 {
4270         journal_t *journal;
4271         int ret = 0;
4272
4273         if (sb->s_flags & MS_RDONLY)
4274                 return 0;
4275
4276         journal = EXT4_SB(sb)->s_journal;
4277         if (journal) {
4278                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4279                 ret = ext4_journal_force_commit(journal);
4280         }
4281
4282         return ret;
4283 }
4284
4285 static void ext4_write_super(struct super_block *sb)
4286 {
4287         lock_super(sb);
4288         ext4_commit_super(sb, 1);
4289         unlock_super(sb);
4290 }
4291
4292 static int ext4_sync_fs(struct super_block *sb, int wait)
4293 {
4294         int ret = 0;
4295         tid_t target;
4296         struct ext4_sb_info *sbi = EXT4_SB(sb);
4297
4298         trace_ext4_sync_fs(sb, wait);
4299         flush_workqueue(sbi->dio_unwritten_wq);
4300         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4301                 if (wait)
4302                         jbd2_log_wait_commit(sbi->s_journal, target);
4303         }
4304         return ret;
4305 }
4306
4307 /*
4308  * LVM calls this function before a (read-only) snapshot is created.  This
4309  * gives us a chance to flush the journal completely and mark the fs clean.
4310  *
4311  * Note that only this function cannot bring a filesystem to be in a clean
4312  * state independently, because ext4 prevents a new handle from being started
4313  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4314  * the upper layer.
4315  */
4316 static int ext4_freeze(struct super_block *sb)
4317 {
4318         int error = 0;
4319         journal_t *journal;
4320
4321         if (sb->s_flags & MS_RDONLY)
4322                 return 0;
4323
4324         journal = EXT4_SB(sb)->s_journal;
4325
4326         /* Now we set up the journal barrier. */
4327         jbd2_journal_lock_updates(journal);
4328
4329         /*
4330          * Don't clear the needs_recovery flag if we failed to flush
4331          * the journal.
4332          */
4333         error = jbd2_journal_flush(journal);
4334         if (error < 0)
4335                 goto out;
4336
4337         /* Journal blocked and flushed, clear needs_recovery flag. */
4338         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4339         error = ext4_commit_super(sb, 1);
4340 out:
4341         /* we rely on s_frozen to stop further updates */
4342         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4343         return error;
4344 }
4345
4346 /*
4347  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4348  * flag here, even though the filesystem is not technically dirty yet.
4349  */
4350 static int ext4_unfreeze(struct super_block *sb)
4351 {
4352         if (sb->s_flags & MS_RDONLY)
4353                 return 0;
4354
4355         lock_super(sb);
4356         /* Reset the needs_recovery flag before the fs is unlocked. */
4357         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4358         ext4_commit_super(sb, 1);
4359         unlock_super(sb);
4360         return 0;
4361 }
4362
4363 /*
4364  * Structure to save mount options for ext4_remount's benefit
4365  */
4366 struct ext4_mount_options {
4367         unsigned long s_mount_opt;
4368         unsigned long s_mount_opt2;
4369         uid_t s_resuid;
4370         gid_t s_resgid;
4371         unsigned long s_commit_interval;
4372         u32 s_min_batch_time, s_max_batch_time;
4373 #ifdef CONFIG_QUOTA
4374         int s_jquota_fmt;
4375         char *s_qf_names[MAXQUOTAS];
4376 #endif
4377 };
4378
4379 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4380 {
4381         struct ext4_super_block *es;
4382         struct ext4_sb_info *sbi = EXT4_SB(sb);
4383         ext4_fsblk_t n_blocks_count = 0;
4384         unsigned long old_sb_flags;
4385         struct ext4_mount_options old_opts;
4386         int enable_quota = 0;
4387         ext4_group_t g;
4388         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4389         int err = 0;
4390 #ifdef CONFIG_QUOTA
4391         int i;
4392 #endif
4393         char *orig_data = kstrdup(data, GFP_KERNEL);
4394
4395         /* Store the original options */
4396         lock_super(sb);
4397         old_sb_flags = sb->s_flags;
4398         old_opts.s_mount_opt = sbi->s_mount_opt;
4399         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4400         old_opts.s_resuid = sbi->s_resuid;
4401         old_opts.s_resgid = sbi->s_resgid;
4402         old_opts.s_commit_interval = sbi->s_commit_interval;
4403         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4404         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4405 #ifdef CONFIG_QUOTA
4406         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4407         for (i = 0; i < MAXQUOTAS; i++)
4408                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4409 #endif
4410         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4411                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4412
4413         /*
4414          * Allow the "check" option to be passed as a remount option.
4415          */
4416         if (!parse_options(data, sb, NULL, &journal_ioprio,
4417                            &n_blocks_count, 1)) {
4418                 err = -EINVAL;
4419                 goto restore_opts;
4420         }
4421
4422         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4423                 ext4_abort(sb, "Abort forced by user");
4424
4425         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4426                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4427
4428         es = sbi->s_es;
4429
4430         if (sbi->s_journal) {
4431                 ext4_init_journal_params(sb, sbi->s_journal);
4432                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4433         }
4434
4435         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4436                 n_blocks_count > ext4_blocks_count(es)) {
4437                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4438                         err = -EROFS;
4439                         goto restore_opts;
4440                 }
4441
4442                 if (*flags & MS_RDONLY) {
4443                         err = dquot_suspend(sb, -1);
4444                         if (err < 0)
4445                                 goto restore_opts;
4446
4447                         /*
4448                          * First of all, the unconditional stuff we have to do
4449                          * to disable replay of the journal when we next remount
4450                          */
4451                         sb->s_flags |= MS_RDONLY;
4452
4453                         /*
4454                          * OK, test if we are remounting a valid rw partition
4455                          * readonly, and if so set the rdonly flag and then
4456                          * mark the partition as valid again.
4457                          */
4458                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4459                             (sbi->s_mount_state & EXT4_VALID_FS))
4460                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4461
4462                         if (sbi->s_journal)
4463                                 ext4_mark_recovery_complete(sb, es);
4464                 } else {
4465                         /* Make sure we can mount this feature set readwrite */
4466                         if (!ext4_feature_set_ok(sb, 0)) {
4467                                 err = -EROFS;
4468                                 goto restore_opts;
4469                         }
4470                         /*
4471                          * Make sure the group descriptor checksums
4472                          * are sane.  If they aren't, refuse to remount r/w.
4473                          */
4474                         for (g = 0; g < sbi->s_groups_count; g++) {
4475                                 struct ext4_group_desc *gdp =
4476                                         ext4_get_group_desc(sb, g, NULL);
4477
4478                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4479                                         ext4_msg(sb, KERN_ERR,
4480                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4481                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4482                                                le16_to_cpu(gdp->bg_checksum));
4483                                         err = -EINVAL;
4484                                         goto restore_opts;
4485                                 }
4486                         }
4487
4488                         /*
4489                          * If we have an unprocessed orphan list hanging
4490                          * around from a previously readonly bdev mount,
4491                          * require a full umount/remount for now.
4492                          */
4493                         if (es->s_last_orphan) {
4494                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4495                                        "remount RDWR because of unprocessed "
4496                                        "orphan inode list.  Please "
4497                                        "umount/remount instead");
4498                                 err = -EINVAL;
4499                                 goto restore_opts;
4500                         }
4501
4502                         /*
4503                          * Mounting a RDONLY partition read-write, so reread
4504                          * and store the current valid flag.  (It may have
4505                          * been changed by e2fsck since we originally mounted
4506                          * the partition.)
4507                          */
4508                         if (sbi->s_journal)
4509                                 ext4_clear_journal_err(sb, es);
4510                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4511                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4512                                 goto restore_opts;
4513                         if (!ext4_setup_super(sb, es, 0))
4514                                 sb->s_flags &= ~MS_RDONLY;
4515                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4516                                                      EXT4_FEATURE_INCOMPAT_MMP))
4517                                 if (ext4_multi_mount_protect(sb,
4518                                                 le64_to_cpu(es->s_mmp_block))) {
4519                                         err = -EROFS;
4520                                         goto restore_opts;
4521                                 }
4522                         enable_quota = 1;
4523                 }
4524         }
4525
4526         /*
4527          * Reinitialize lazy itable initialization thread based on
4528          * current settings
4529          */
4530         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4531                 ext4_unregister_li_request(sb);
4532         else {
4533                 ext4_group_t first_not_zeroed;
4534                 first_not_zeroed = ext4_has_uninit_itable(sb);
4535                 ext4_register_li_request(sb, first_not_zeroed);
4536         }
4537
4538         ext4_setup_system_zone(sb);
4539         if (sbi->s_journal == NULL)
4540                 ext4_commit_super(sb, 1);
4541
4542 #ifdef CONFIG_QUOTA
4543         /* Release old quota file names */
4544         for (i = 0; i < MAXQUOTAS; i++)
4545                 if (old_opts.s_qf_names[i] &&
4546                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4547                         kfree(old_opts.s_qf_names[i]);
4548 #endif
4549         unlock_super(sb);
4550         if (enable_quota)
4551                 dquot_resume(sb, -1);
4552
4553         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4554         kfree(orig_data);
4555         return 0;
4556
4557 restore_opts:
4558         sb->s_flags = old_sb_flags;
4559         sbi->s_mount_opt = old_opts.s_mount_opt;
4560         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4561         sbi->s_resuid = old_opts.s_resuid;
4562         sbi->s_resgid = old_opts.s_resgid;
4563         sbi->s_commit_interval = old_opts.s_commit_interval;
4564         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4565         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4566 #ifdef CONFIG_QUOTA
4567         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4568         for (i = 0; i < MAXQUOTAS; i++) {
4569                 if (sbi->s_qf_names[i] &&
4570                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4571                         kfree(sbi->s_qf_names[i]);
4572                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4573         }
4574 #endif
4575         unlock_super(sb);
4576         kfree(orig_data);
4577         return err;
4578 }
4579
4580 /*
4581  * Note: calculating the overhead so we can be compatible with
4582  * historical BSD practice is quite difficult in the face of
4583  * clusters/bigalloc.  This is because multiple metadata blocks from
4584  * different block group can end up in the same allocation cluster.
4585  * Calculating the exact overhead in the face of clustered allocation
4586  * requires either O(all block bitmaps) in memory or O(number of block
4587  * groups**2) in time.  We will still calculate the superblock for
4588  * older file systems --- and if we come across with a bigalloc file
4589  * system with zero in s_overhead_clusters the estimate will be close to
4590  * correct especially for very large cluster sizes --- but for newer
4591  * file systems, it's better to calculate this figure once at mkfs
4592  * time, and store it in the superblock.  If the superblock value is
4593  * present (even for non-bigalloc file systems), we will use it.
4594  */
4595 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4596 {
4597         struct super_block *sb = dentry->d_sb;
4598         struct ext4_sb_info *sbi = EXT4_SB(sb);
4599         struct ext4_super_block *es = sbi->s_es;
4600         struct ext4_group_desc *gdp;
4601         u64 fsid;
4602         s64 bfree;
4603
4604         if (test_opt(sb, MINIX_DF)) {
4605                 sbi->s_overhead_last = 0;
4606         } else if (es->s_overhead_clusters) {
4607                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4608         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4609                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4610                 ext4_fsblk_t overhead = 0;
4611
4612                 /*
4613                  * Compute the overhead (FS structures).  This is constant
4614                  * for a given filesystem unless the number of block groups
4615                  * changes so we cache the previous value until it does.
4616                  */
4617
4618                 /*
4619                  * All of the blocks before first_data_block are
4620                  * overhead
4621                  */
4622                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4623
4624                 /*
4625                  * Add the overhead found in each block group
4626                  */
4627                 for (i = 0; i < ngroups; i++) {
4628                         gdp = ext4_get_group_desc(sb, i, NULL);
4629                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4630                         cond_resched();
4631                 }
4632                 sbi->s_overhead_last = overhead;
4633                 smp_wmb();
4634                 sbi->s_blocks_last = ext4_blocks_count(es);
4635         }
4636
4637         buf->f_type = EXT4_SUPER_MAGIC;
4638         buf->f_bsize = sb->s_blocksize;
4639         buf->f_blocks = (ext4_blocks_count(es) -
4640                          EXT4_C2B(sbi, sbi->s_overhead_last));
4641         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4642                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4643         /* prevent underflow in case that few free space is available */
4644         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4645         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4646         if (buf->f_bfree < ext4_r_blocks_count(es))
4647                 buf->f_bavail = 0;
4648         buf->f_files = le32_to_cpu(es->s_inodes_count);
4649         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4650         buf->f_namelen = EXT4_NAME_LEN;
4651         fsid = le64_to_cpup((void *)es->s_uuid) ^
4652                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4653         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4654         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4655
4656         return 0;
4657 }
4658
4659 /* Helper function for writing quotas on sync - we need to start transaction
4660  * before quota file is locked for write. Otherwise the are possible deadlocks:
4661  * Process 1                         Process 2
4662  * ext4_create()                     quota_sync()
4663  *   jbd2_journal_start()                  write_dquot()
4664  *   dquot_initialize()                         down(dqio_mutex)
4665  *     down(dqio_mutex)                    jbd2_journal_start()
4666  *
4667  */
4668
4669 #ifdef CONFIG_QUOTA
4670
4671 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4672 {
4673         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4674 }
4675
4676 static int ext4_write_dquot(struct dquot *dquot)
4677 {
4678         int ret, err;
4679         handle_t *handle;
4680         struct inode *inode;
4681
4682         inode = dquot_to_inode(dquot);
4683         handle = ext4_journal_start(inode,
4684                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4685         if (IS_ERR(handle))
4686                 return PTR_ERR(handle);
4687         ret = dquot_commit(dquot);
4688         err = ext4_journal_stop(handle);
4689         if (!ret)
4690                 ret = err;
4691         return ret;
4692 }
4693
4694 static int ext4_acquire_dquot(struct dquot *dquot)
4695 {
4696         int ret, err;
4697         handle_t *handle;
4698
4699         handle = ext4_journal_start(dquot_to_inode(dquot),
4700                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4701         if (IS_ERR(handle))
4702                 return PTR_ERR(handle);
4703         ret = dquot_acquire(dquot);
4704         err = ext4_journal_stop(handle);
4705         if (!ret)
4706                 ret = err;
4707         return ret;
4708 }
4709
4710 static int ext4_release_dquot(struct dquot *dquot)
4711 {
4712         int ret, err;
4713         handle_t *handle;
4714
4715         handle = ext4_journal_start(dquot_to_inode(dquot),
4716                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4717         if (IS_ERR(handle)) {
4718                 /* Release dquot anyway to avoid endless cycle in dqput() */
4719                 dquot_release(dquot);
4720                 return PTR_ERR(handle);
4721         }
4722         ret = dquot_release(dquot);
4723         err = ext4_journal_stop(handle);
4724         if (!ret)
4725                 ret = err;
4726         return ret;
4727 }
4728
4729 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4730 {
4731         /* Are we journaling quotas? */
4732         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4733             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4734                 dquot_mark_dquot_dirty(dquot);
4735                 return ext4_write_dquot(dquot);
4736         } else {
4737                 return dquot_mark_dquot_dirty(dquot);
4738         }
4739 }
4740
4741 static int ext4_write_info(struct super_block *sb, int type)
4742 {
4743         int ret, err;
4744         handle_t *handle;
4745
4746         /* Data block + inode block */
4747         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4748         if (IS_ERR(handle))
4749                 return PTR_ERR(handle);
4750         ret = dquot_commit_info(sb, type);
4751         err = ext4_journal_stop(handle);
4752         if (!ret)
4753                 ret = err;
4754         return ret;
4755 }
4756
4757 /*
4758  * Turn on quotas during mount time - we need to find
4759  * the quota file and such...
4760  */
4761 static int ext4_quota_on_mount(struct super_block *sb, int type)
4762 {
4763         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4764                                         EXT4_SB(sb)->s_jquota_fmt, type);
4765 }
4766
4767 /*
4768  * Standard function to be called on quota_on
4769  */
4770 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4771                          struct path *path)
4772 {
4773         int err;
4774
4775         if (!test_opt(sb, QUOTA))
4776                 return -EINVAL;
4777
4778         /* Quotafile not on the same filesystem? */
4779         if (path->dentry->d_sb != sb)
4780                 return -EXDEV;
4781         /* Journaling quota? */
4782         if (EXT4_SB(sb)->s_qf_names[type]) {
4783                 /* Quotafile not in fs root? */
4784                 if (path->dentry->d_parent != sb->s_root)
4785                         ext4_msg(sb, KERN_WARNING,
4786                                 "Quota file not on filesystem root. "
4787                                 "Journaled quota will not work");
4788         }
4789
4790         /*
4791          * When we journal data on quota file, we have to flush journal to see
4792          * all updates to the file when we bypass pagecache...
4793          */
4794         if (EXT4_SB(sb)->s_journal &&
4795             ext4_should_journal_data(path->dentry->d_inode)) {
4796                 /*
4797                  * We don't need to lock updates but journal_flush() could
4798                  * otherwise be livelocked...
4799                  */
4800                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4801                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4802                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4803                 if (err)
4804                         return err;
4805         }
4806
4807         return dquot_quota_on(sb, type, format_id, path);
4808 }
4809
4810 static int ext4_quota_off(struct super_block *sb, int type)
4811 {
4812         struct inode *inode = sb_dqopt(sb)->files[type];
4813         handle_t *handle;
4814
4815         /* Force all delayed allocation blocks to be allocated.
4816          * Caller already holds s_umount sem */
4817         if (test_opt(sb, DELALLOC))
4818                 sync_filesystem(sb);
4819
4820         if (!inode)
4821                 goto out;
4822
4823         /* Update modification times of quota files when userspace can
4824          * start looking at them */
4825         handle = ext4_journal_start(inode, 1);
4826         if (IS_ERR(handle))
4827                 goto out;
4828         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4829         ext4_mark_inode_dirty(handle, inode);
4830         ext4_journal_stop(handle);
4831
4832 out:
4833         return dquot_quota_off(sb, type);
4834 }
4835
4836 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4837  * acquiring the locks... As quota files are never truncated and quota code
4838  * itself serializes the operations (and no one else should touch the files)
4839  * we don't have to be afraid of races */
4840 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4841                                size_t len, loff_t off)
4842 {
4843         struct inode *inode = sb_dqopt(sb)->files[type];
4844         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4845         int err = 0;
4846         int offset = off & (sb->s_blocksize - 1);
4847         int tocopy;
4848         size_t toread;
4849         struct buffer_head *bh;
4850         loff_t i_size = i_size_read(inode);
4851
4852         if (off > i_size)
4853                 return 0;
4854         if (off+len > i_size)
4855                 len = i_size-off;
4856         toread = len;
4857         while (toread > 0) {
4858                 tocopy = sb->s_blocksize - offset < toread ?
4859                                 sb->s_blocksize - offset : toread;
4860                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4861                 if (err)
4862                         return err;
4863                 if (!bh)        /* A hole? */
4864                         memset(data, 0, tocopy);
4865                 else
4866                         memcpy(data, bh->b_data+offset, tocopy);
4867                 brelse(bh);
4868                 offset = 0;
4869                 toread -= tocopy;
4870                 data += tocopy;
4871                 blk++;
4872         }
4873         return len;
4874 }
4875
4876 /* Write to quotafile (we know the transaction is already started and has
4877  * enough credits) */
4878 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4879                                 const char *data, size_t len, loff_t off)
4880 {
4881         struct inode *inode = sb_dqopt(sb)->files[type];
4882         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4883         int err = 0;
4884         int offset = off & (sb->s_blocksize - 1);
4885         struct buffer_head *bh;
4886         handle_t *handle = journal_current_handle();
4887
4888         if (EXT4_SB(sb)->s_journal && !handle) {
4889                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4890                         " cancelled because transaction is not started",
4891                         (unsigned long long)off, (unsigned long long)len);
4892                 return -EIO;
4893         }
4894         /*
4895          * Since we account only one data block in transaction credits,
4896          * then it is impossible to cross a block boundary.
4897          */
4898         if (sb->s_blocksize - offset < len) {
4899                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4900                         " cancelled because not block aligned",
4901                         (unsigned long long)off, (unsigned long long)len);
4902                 return -EIO;
4903         }
4904
4905         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4906         bh = ext4_bread(handle, inode, blk, 1, &err);
4907         if (!bh)
4908                 goto out;
4909         err = ext4_journal_get_write_access(handle, bh);
4910         if (err) {
4911                 brelse(bh);
4912                 goto out;
4913         }
4914         lock_buffer(bh);
4915         memcpy(bh->b_data+offset, data, len);
4916         flush_dcache_page(bh->b_page);
4917         unlock_buffer(bh);
4918         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4919         brelse(bh);
4920 out:
4921         if (err) {
4922                 mutex_unlock(&inode->i_mutex);
4923                 return err;
4924         }
4925         if (inode->i_size < off + len) {
4926                 i_size_write(inode, off + len);
4927                 EXT4_I(inode)->i_disksize = inode->i_size;
4928                 ext4_mark_inode_dirty(handle, inode);
4929         }
4930         mutex_unlock(&inode->i_mutex);
4931         return len;
4932 }
4933
4934 #endif
4935
4936 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4937                        const char *dev_name, void *data)
4938 {
4939         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4940 }
4941
4942 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4943 static inline void register_as_ext2(void)
4944 {
4945         int err = register_filesystem(&ext2_fs_type);
4946         if (err)
4947                 printk(KERN_WARNING
4948                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4949 }
4950
4951 static inline void unregister_as_ext2(void)
4952 {
4953         unregister_filesystem(&ext2_fs_type);
4954 }
4955
4956 static inline int ext2_feature_set_ok(struct super_block *sb)
4957 {
4958         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4959                 return 0;
4960         if (sb->s_flags & MS_RDONLY)
4961                 return 1;
4962         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4963                 return 0;
4964         return 1;
4965 }
4966 MODULE_ALIAS("ext2");
4967 #else
4968 static inline void register_as_ext2(void) { }
4969 static inline void unregister_as_ext2(void) { }
4970 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4971 #endif
4972
4973 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4974 static inline void register_as_ext3(void)
4975 {
4976         int err = register_filesystem(&ext3_fs_type);
4977         if (err)
4978                 printk(KERN_WARNING
4979                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4980 }
4981
4982 static inline void unregister_as_ext3(void)
4983 {
4984         unregister_filesystem(&ext3_fs_type);
4985 }
4986
4987 static inline int ext3_feature_set_ok(struct super_block *sb)
4988 {
4989         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4990                 return 0;
4991         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4992                 return 0;
4993         if (sb->s_flags & MS_RDONLY)
4994                 return 1;
4995         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4996                 return 0;
4997         return 1;
4998 }
4999 MODULE_ALIAS("ext3");
5000 #else
5001 static inline void register_as_ext3(void) { }
5002 static inline void unregister_as_ext3(void) { }
5003 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5004 #endif
5005
5006 static struct file_system_type ext4_fs_type = {
5007         .owner          = THIS_MODULE,
5008         .name           = "ext4",
5009         .mount          = ext4_mount,
5010         .kill_sb        = kill_block_super,
5011         .fs_flags       = FS_REQUIRES_DEV,
5012 };
5013
5014 static int __init ext4_init_feat_adverts(void)
5015 {
5016         struct ext4_features *ef;
5017         int ret = -ENOMEM;
5018
5019         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5020         if (!ef)
5021                 goto out;
5022
5023         ef->f_kobj.kset = ext4_kset;
5024         init_completion(&ef->f_kobj_unregister);
5025         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5026                                    "features");
5027         if (ret) {
5028                 kfree(ef);
5029                 goto out;
5030         }
5031
5032         ext4_feat = ef;
5033         ret = 0;
5034 out:
5035         return ret;
5036 }
5037
5038 static void ext4_exit_feat_adverts(void)
5039 {
5040         kobject_put(&ext4_feat->f_kobj);
5041         wait_for_completion(&ext4_feat->f_kobj_unregister);
5042         kfree(ext4_feat);
5043 }
5044
5045 /* Shared across all ext4 file systems */
5046 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5047 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5048
5049 static int __init ext4_init_fs(void)
5050 {
5051         int i, err;
5052
5053         ext4_check_flag_values();
5054
5055         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5056                 mutex_init(&ext4__aio_mutex[i]);
5057                 init_waitqueue_head(&ext4__ioend_wq[i]);
5058         }
5059
5060         err = ext4_init_pageio();
5061         if (err)
5062                 return err;
5063         err = ext4_init_system_zone();
5064         if (err)
5065                 goto out6;
5066         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5067         if (!ext4_kset)
5068                 goto out5;
5069         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5070
5071         err = ext4_init_feat_adverts();
5072         if (err)
5073                 goto out4;
5074
5075         err = ext4_init_mballoc();
5076         if (err)
5077                 goto out3;
5078
5079         err = ext4_init_xattr();
5080         if (err)
5081                 goto out2;
5082         err = init_inodecache();
5083         if (err)
5084                 goto out1;
5085         register_as_ext3();
5086         register_as_ext2();
5087         err = register_filesystem(&ext4_fs_type);
5088         if (err)
5089                 goto out;
5090
5091         ext4_li_info = NULL;
5092         mutex_init(&ext4_li_mtx);
5093         return 0;
5094 out:
5095         unregister_as_ext2();
5096         unregister_as_ext3();
5097         destroy_inodecache();
5098 out1:
5099         ext4_exit_xattr();
5100 out2:
5101         ext4_exit_mballoc();
5102 out3:
5103         ext4_exit_feat_adverts();
5104 out4:
5105         if (ext4_proc_root)
5106                 remove_proc_entry("fs/ext4", NULL);
5107         kset_unregister(ext4_kset);
5108 out5:
5109         ext4_exit_system_zone();
5110 out6:
5111         ext4_exit_pageio();
5112         return err;
5113 }
5114
5115 static void __exit ext4_exit_fs(void)
5116 {
5117         ext4_destroy_lazyinit_thread();
5118         unregister_as_ext2();
5119         unregister_as_ext3();
5120         unregister_filesystem(&ext4_fs_type);
5121         destroy_inodecache();
5122         ext4_exit_xattr();
5123         ext4_exit_mballoc();
5124         ext4_exit_feat_adverts();
5125         remove_proc_entry("fs/ext4", NULL);
5126         kset_unregister(ext4_kset);
5127         ext4_exit_system_zone();
5128         ext4_exit_pageio();
5129 }
5130
5131 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5132 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5133 MODULE_LICENSE("GPL");
5134 module_init(ext4_init_fs)
5135 module_exit(ext4_exit_fs)