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_update, Opt_journal_dev,
1340         Opt_journal_checksum, Opt_journal_async_commit,
1341         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1342         Opt_data_err_abort, Opt_data_err_ignore,
1343         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1344         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1345         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1346         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1347         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1348         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1349         Opt_inode_readahead_blks, Opt_journal_ioprio,
1350         Opt_dioread_nolock, Opt_dioread_lock,
1351         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1352 };
1353
1354 static const match_table_t tokens = {
1355         {Opt_bsd_df, "bsddf"},
1356         {Opt_minix_df, "minixdf"},
1357         {Opt_grpid, "grpid"},
1358         {Opt_grpid, "bsdgroups"},
1359         {Opt_nogrpid, "nogrpid"},
1360         {Opt_nogrpid, "sysvgroups"},
1361         {Opt_resgid, "resgid=%u"},
1362         {Opt_resuid, "resuid=%u"},
1363         {Opt_sb, "sb=%u"},
1364         {Opt_err_cont, "errors=continue"},
1365         {Opt_err_panic, "errors=panic"},
1366         {Opt_err_ro, "errors=remount-ro"},
1367         {Opt_nouid32, "nouid32"},
1368         {Opt_debug, "debug"},
1369         {Opt_oldalloc, "oldalloc"},
1370         {Opt_orlov, "orlov"},
1371         {Opt_user_xattr, "user_xattr"},
1372         {Opt_nouser_xattr, "nouser_xattr"},
1373         {Opt_acl, "acl"},
1374         {Opt_noacl, "noacl"},
1375         {Opt_noload, "noload"},
1376         {Opt_noload, "norecovery"},
1377         {Opt_nobh, "nobh"},
1378         {Opt_bh, "bh"},
1379         {Opt_commit, "commit=%u"},
1380         {Opt_min_batch_time, "min_batch_time=%u"},
1381         {Opt_max_batch_time, "max_batch_time=%u"},
1382         {Opt_journal_update, "journal=update"},
1383         {Opt_journal_dev, "journal_dev=%u"},
1384         {Opt_journal_checksum, "journal_checksum"},
1385         {Opt_journal_async_commit, "journal_async_commit"},
1386         {Opt_abort, "abort"},
1387         {Opt_data_journal, "data=journal"},
1388         {Opt_data_ordered, "data=ordered"},
1389         {Opt_data_writeback, "data=writeback"},
1390         {Opt_data_err_abort, "data_err=abort"},
1391         {Opt_data_err_ignore, "data_err=ignore"},
1392         {Opt_offusrjquota, "usrjquota="},
1393         {Opt_usrjquota, "usrjquota=%s"},
1394         {Opt_offgrpjquota, "grpjquota="},
1395         {Opt_grpjquota, "grpjquota=%s"},
1396         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1397         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1398         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1399         {Opt_grpquota, "grpquota"},
1400         {Opt_noquota, "noquota"},
1401         {Opt_quota, "quota"},
1402         {Opt_usrquota, "usrquota"},
1403         {Opt_barrier, "barrier=%u"},
1404         {Opt_barrier, "barrier"},
1405         {Opt_nobarrier, "nobarrier"},
1406         {Opt_i_version, "i_version"},
1407         {Opt_stripe, "stripe=%u"},
1408         {Opt_resize, "resize"},
1409         {Opt_delalloc, "delalloc"},
1410         {Opt_nodelalloc, "nodelalloc"},
1411         {Opt_mblk_io_submit, "mblk_io_submit"},
1412         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1413         {Opt_block_validity, "block_validity"},
1414         {Opt_noblock_validity, "noblock_validity"},
1415         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1416         {Opt_journal_ioprio, "journal_ioprio=%u"},
1417         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1418         {Opt_auto_da_alloc, "auto_da_alloc"},
1419         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1420         {Opt_dioread_nolock, "dioread_nolock"},
1421         {Opt_dioread_lock, "dioread_lock"},
1422         {Opt_discard, "discard"},
1423         {Opt_nodiscard, "nodiscard"},
1424         {Opt_init_itable, "init_itable=%u"},
1425         {Opt_init_itable, "init_itable"},
1426         {Opt_noinit_itable, "noinit_itable"},
1427         {Opt_err, NULL},
1428 };
1429
1430 static ext4_fsblk_t get_sb_block(void **data)
1431 {
1432         ext4_fsblk_t    sb_block;
1433         char            *options = (char *) *data;
1434
1435         if (!options || strncmp(options, "sb=", 3) != 0)
1436                 return 1;       /* Default location */
1437
1438         options += 3;
1439         /* TODO: use simple_strtoll with >32bit ext4 */
1440         sb_block = simple_strtoul(options, &options, 0);
1441         if (*options && *options != ',') {
1442                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1443                        (char *) *data);
1444                 return 1;
1445         }
1446         if (*options == ',')
1447                 options++;
1448         *data = (void *) options;
1449
1450         return sb_block;
1451 }
1452
1453 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1454 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1455         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1456
1457 #ifdef CONFIG_QUOTA
1458 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1459 {
1460         struct ext4_sb_info *sbi = EXT4_SB(sb);
1461         char *qname;
1462
1463         if (sb_any_quota_loaded(sb) &&
1464                 !sbi->s_qf_names[qtype]) {
1465                 ext4_msg(sb, KERN_ERR,
1466                         "Cannot change journaled "
1467                         "quota options when quota turned on");
1468                 return 0;
1469         }
1470         qname = match_strdup(args);
1471         if (!qname) {
1472                 ext4_msg(sb, KERN_ERR,
1473                         "Not enough memory for storing quotafile name");
1474                 return 0;
1475         }
1476         if (sbi->s_qf_names[qtype] &&
1477                 strcmp(sbi->s_qf_names[qtype], qname)) {
1478                 ext4_msg(sb, KERN_ERR,
1479                         "%s quota file already specified", QTYPE2NAME(qtype));
1480                 kfree(qname);
1481                 return 0;
1482         }
1483         sbi->s_qf_names[qtype] = qname;
1484         if (strchr(sbi->s_qf_names[qtype], '/')) {
1485                 ext4_msg(sb, KERN_ERR,
1486                         "quotafile must be on filesystem root");
1487                 kfree(sbi->s_qf_names[qtype]);
1488                 sbi->s_qf_names[qtype] = NULL;
1489                 return 0;
1490         }
1491         set_opt(sb, QUOTA);
1492         return 1;
1493 }
1494
1495 static int clear_qf_name(struct super_block *sb, int qtype)
1496 {
1497
1498         struct ext4_sb_info *sbi = EXT4_SB(sb);
1499
1500         if (sb_any_quota_loaded(sb) &&
1501                 sbi->s_qf_names[qtype]) {
1502                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1503                         " when quota turned on");
1504                 return 0;
1505         }
1506         /*
1507          * The space will be released later when all options are confirmed
1508          * to be correct
1509          */
1510         sbi->s_qf_names[qtype] = NULL;
1511         return 1;
1512 }
1513 #endif
1514
1515 static int parse_options(char *options, struct super_block *sb,
1516                          unsigned long *journal_devnum,
1517                          unsigned int *journal_ioprio,
1518                          ext4_fsblk_t *n_blocks_count, int is_remount)
1519 {
1520         struct ext4_sb_info *sbi = EXT4_SB(sb);
1521         char *p;
1522         substring_t args[MAX_OPT_ARGS];
1523         int data_opt = 0;
1524         int option;
1525 #ifdef CONFIG_QUOTA
1526         int qfmt;
1527 #endif
1528
1529         if (!options)
1530                 return 1;
1531
1532         while ((p = strsep(&options, ",")) != NULL) {
1533                 int token;
1534                 if (!*p)
1535                         continue;
1536
1537                 /*
1538                  * Initialize args struct so we know whether arg was
1539                  * found; some options take optional arguments.
1540                  */
1541                 args[0].to = args[0].from = NULL;
1542                 token = match_token(p, tokens, args);
1543                 switch (token) {
1544                 case Opt_bsd_df:
1545                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1546                         clear_opt(sb, MINIX_DF);
1547                         break;
1548                 case Opt_minix_df:
1549                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1550                         set_opt(sb, MINIX_DF);
1551
1552                         break;
1553                 case Opt_grpid:
1554                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1555                         set_opt(sb, GRPID);
1556
1557                         break;
1558                 case Opt_nogrpid:
1559                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1560                         clear_opt(sb, GRPID);
1561
1562                         break;
1563                 case Opt_resuid:
1564                         if (match_int(&args[0], &option))
1565                                 return 0;
1566                         sbi->s_resuid = option;
1567                         break;
1568                 case Opt_resgid:
1569                         if (match_int(&args[0], &option))
1570                                 return 0;
1571                         sbi->s_resgid = option;
1572                         break;
1573                 case Opt_sb:
1574                         /* handled by get_sb_block() instead of here */
1575                         /* *sb_block = match_int(&args[0]); */
1576                         break;
1577                 case Opt_err_panic:
1578                         clear_opt(sb, ERRORS_CONT);
1579                         clear_opt(sb, ERRORS_RO);
1580                         set_opt(sb, ERRORS_PANIC);
1581                         break;
1582                 case Opt_err_ro:
1583                         clear_opt(sb, ERRORS_CONT);
1584                         clear_opt(sb, ERRORS_PANIC);
1585                         set_opt(sb, ERRORS_RO);
1586                         break;
1587                 case Opt_err_cont:
1588                         clear_opt(sb, ERRORS_RO);
1589                         clear_opt(sb, ERRORS_PANIC);
1590                         set_opt(sb, ERRORS_CONT);
1591                         break;
1592                 case Opt_nouid32:
1593                         set_opt(sb, NO_UID32);
1594                         break;
1595                 case Opt_debug:
1596                         set_opt(sb, DEBUG);
1597                         break;
1598                 case Opt_oldalloc:
1599                         ext4_msg(sb, KERN_WARNING,
1600                                  "Ignoring deprecated oldalloc option");
1601                         break;
1602                 case Opt_orlov:
1603                         ext4_msg(sb, KERN_WARNING,
1604                                  "Ignoring deprecated orlov option");
1605                         break;
1606 #ifdef CONFIG_EXT4_FS_XATTR
1607                 case Opt_user_xattr:
1608                         set_opt(sb, XATTR_USER);
1609                         break;
1610                 case Opt_nouser_xattr:
1611                         clear_opt(sb, XATTR_USER);
1612                         break;
1613 #else
1614                 case Opt_user_xattr:
1615                 case Opt_nouser_xattr:
1616                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1617                         break;
1618 #endif
1619 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1620                 case Opt_acl:
1621                         set_opt(sb, POSIX_ACL);
1622                         break;
1623                 case Opt_noacl:
1624                         clear_opt(sb, POSIX_ACL);
1625                         break;
1626 #else
1627                 case Opt_acl:
1628                 case Opt_noacl:
1629                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1630                         break;
1631 #endif
1632                 case Opt_journal_update:
1633                         /* @@@ FIXME */
1634                         /* Eventually we will want to be able to create
1635                            a journal file here.  For now, only allow the
1636                            user to specify an existing inode to be the
1637                            journal file. */
1638                         if (is_remount) {
1639                                 ext4_msg(sb, KERN_ERR,
1640                                          "Cannot specify journal on remount");
1641                                 return 0;
1642                         }
1643                         set_opt(sb, UPDATE_JOURNAL);
1644                         break;
1645                 case Opt_journal_dev:
1646                         if (is_remount) {
1647                                 ext4_msg(sb, KERN_ERR,
1648                                         "Cannot specify journal on remount");
1649                                 return 0;
1650                         }
1651                         if (match_int(&args[0], &option))
1652                                 return 0;
1653                         *journal_devnum = option;
1654                         break;
1655                 case Opt_journal_checksum:
1656                         set_opt(sb, JOURNAL_CHECKSUM);
1657                         break;
1658                 case Opt_journal_async_commit:
1659                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1660                         set_opt(sb, JOURNAL_CHECKSUM);
1661                         break;
1662                 case Opt_noload:
1663                         set_opt(sb, NOLOAD);
1664                         break;
1665                 case Opt_commit:
1666                         if (match_int(&args[0], &option))
1667                                 return 0;
1668                         if (option < 0)
1669                                 return 0;
1670                         if (option == 0)
1671                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1672                         sbi->s_commit_interval = HZ * option;
1673                         break;
1674                 case Opt_max_batch_time:
1675                         if (match_int(&args[0], &option))
1676                                 return 0;
1677                         if (option < 0)
1678                                 return 0;
1679                         if (option == 0)
1680                                 option = EXT4_DEF_MAX_BATCH_TIME;
1681                         sbi->s_max_batch_time = option;
1682                         break;
1683                 case Opt_min_batch_time:
1684                         if (match_int(&args[0], &option))
1685                                 return 0;
1686                         if (option < 0)
1687                                 return 0;
1688                         sbi->s_min_batch_time = option;
1689                         break;
1690                 case Opt_data_journal:
1691                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1692                         goto datacheck;
1693                 case Opt_data_ordered:
1694                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1695                         goto datacheck;
1696                 case Opt_data_writeback:
1697                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1698                 datacheck:
1699                         if (is_remount) {
1700                                 if (!sbi->s_journal)
1701                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1702                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1703                                         ext4_msg(sb, KERN_ERR,
1704                                                 "Cannot change data mode on remount");
1705                                         return 0;
1706                                 }
1707                         } else {
1708                                 clear_opt(sb, DATA_FLAGS);
1709                                 sbi->s_mount_opt |= data_opt;
1710                         }
1711                         break;
1712                 case Opt_data_err_abort:
1713                         set_opt(sb, DATA_ERR_ABORT);
1714                         break;
1715                 case Opt_data_err_ignore:
1716                         clear_opt(sb, DATA_ERR_ABORT);
1717                         break;
1718 #ifdef CONFIG_QUOTA
1719                 case Opt_usrjquota:
1720                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1721                                 return 0;
1722                         break;
1723                 case Opt_grpjquota:
1724                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1725                                 return 0;
1726                         break;
1727                 case Opt_offusrjquota:
1728                         if (!clear_qf_name(sb, USRQUOTA))
1729                                 return 0;
1730                         break;
1731                 case Opt_offgrpjquota:
1732                         if (!clear_qf_name(sb, GRPQUOTA))
1733                                 return 0;
1734                         break;
1735
1736                 case Opt_jqfmt_vfsold:
1737                         qfmt = QFMT_VFS_OLD;
1738                         goto set_qf_format;
1739                 case Opt_jqfmt_vfsv0:
1740                         qfmt = QFMT_VFS_V0;
1741                         goto set_qf_format;
1742                 case Opt_jqfmt_vfsv1:
1743                         qfmt = QFMT_VFS_V1;
1744 set_qf_format:
1745                         if (sb_any_quota_loaded(sb) &&
1746                             sbi->s_jquota_fmt != qfmt) {
1747                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1748                                         "journaled quota options when "
1749                                         "quota turned on");
1750                                 return 0;
1751                         }
1752                         sbi->s_jquota_fmt = qfmt;
1753                         break;
1754                 case Opt_quota:
1755                 case Opt_usrquota:
1756                         set_opt(sb, QUOTA);
1757                         set_opt(sb, USRQUOTA);
1758                         break;
1759                 case Opt_grpquota:
1760                         set_opt(sb, QUOTA);
1761                         set_opt(sb, GRPQUOTA);
1762                         break;
1763                 case Opt_noquota:
1764                         if (sb_any_quota_loaded(sb)) {
1765                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1766                                         "options when quota turned on");
1767                                 return 0;
1768                         }
1769                         clear_opt(sb, QUOTA);
1770                         clear_opt(sb, USRQUOTA);
1771                         clear_opt(sb, GRPQUOTA);
1772                         break;
1773 #else
1774                 case Opt_quota:
1775                 case Opt_usrquota:
1776                 case Opt_grpquota:
1777                         ext4_msg(sb, KERN_ERR,
1778                                 "quota options not supported");
1779                         break;
1780                 case Opt_usrjquota:
1781                 case Opt_grpjquota:
1782                 case Opt_offusrjquota:
1783                 case Opt_offgrpjquota:
1784                 case Opt_jqfmt_vfsold:
1785                 case Opt_jqfmt_vfsv0:
1786                 case Opt_jqfmt_vfsv1:
1787                         ext4_msg(sb, KERN_ERR,
1788                                 "journaled quota options not supported");
1789                         break;
1790                 case Opt_noquota:
1791                         break;
1792 #endif
1793                 case Opt_abort:
1794                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1795                         break;
1796                 case Opt_nobarrier:
1797                         clear_opt(sb, BARRIER);
1798                         break;
1799                 case Opt_barrier:
1800                         if (args[0].from) {
1801                                 if (match_int(&args[0], &option))
1802                                         return 0;
1803                         } else
1804                                 option = 1;     /* No argument, default to 1 */
1805                         if (option)
1806                                 set_opt(sb, BARRIER);
1807                         else
1808                                 clear_opt(sb, BARRIER);
1809                         break;
1810                 case Opt_ignore:
1811                         break;
1812                 case Opt_resize:
1813                         if (!is_remount) {
1814                                 ext4_msg(sb, KERN_ERR,
1815                                         "resize option only available "
1816                                         "for remount");
1817                                 return 0;
1818                         }
1819                         if (match_int(&args[0], &option) != 0)
1820                                 return 0;
1821                         *n_blocks_count = option;
1822                         break;
1823                 case Opt_nobh:
1824                         ext4_msg(sb, KERN_WARNING,
1825                                  "Ignoring deprecated nobh option");
1826                         break;
1827                 case Opt_bh:
1828                         ext4_msg(sb, KERN_WARNING,
1829                                  "Ignoring deprecated bh option");
1830                         break;
1831                 case Opt_i_version:
1832                         set_opt(sb, I_VERSION);
1833                         sb->s_flags |= MS_I_VERSION;
1834                         break;
1835                 case Opt_nodelalloc:
1836                         clear_opt(sb, DELALLOC);
1837                         clear_opt2(sb, EXPLICIT_DELALLOC);
1838                         break;
1839                 case Opt_mblk_io_submit:
1840                         set_opt(sb, MBLK_IO_SUBMIT);
1841                         break;
1842                 case Opt_nomblk_io_submit:
1843                         clear_opt(sb, MBLK_IO_SUBMIT);
1844                         break;
1845                 case Opt_stripe:
1846                         if (match_int(&args[0], &option))
1847                                 return 0;
1848                         if (option < 0)
1849                                 return 0;
1850                         sbi->s_stripe = option;
1851                         break;
1852                 case Opt_delalloc:
1853                         set_opt(sb, DELALLOC);
1854                         set_opt2(sb, EXPLICIT_DELALLOC);
1855                         break;
1856                 case Opt_block_validity:
1857                         set_opt(sb, BLOCK_VALIDITY);
1858                         break;
1859                 case Opt_noblock_validity:
1860                         clear_opt(sb, BLOCK_VALIDITY);
1861                         break;
1862                 case Opt_inode_readahead_blks:
1863                         if (match_int(&args[0], &option))
1864                                 return 0;
1865                         if (option < 0 || option > (1 << 30))
1866                                 return 0;
1867                         if (option && !is_power_of_2(option)) {
1868                                 ext4_msg(sb, KERN_ERR,
1869                                          "EXT4-fs: inode_readahead_blks"
1870                                          " must be a power of 2");
1871                                 return 0;
1872                         }
1873                         sbi->s_inode_readahead_blks = option;
1874                         break;
1875                 case Opt_journal_ioprio:
1876                         if (match_int(&args[0], &option))
1877                                 return 0;
1878                         if (option < 0 || option > 7)
1879                                 break;
1880                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1881                                                             option);
1882                         break;
1883                 case Opt_noauto_da_alloc:
1884                         set_opt(sb, NO_AUTO_DA_ALLOC);
1885                         break;
1886                 case Opt_auto_da_alloc:
1887                         if (args[0].from) {
1888                                 if (match_int(&args[0], &option))
1889                                         return 0;
1890                         } else
1891                                 option = 1;     /* No argument, default to 1 */
1892                         if (option)
1893                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1894                         else
1895                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1896                         break;
1897                 case Opt_discard:
1898                         set_opt(sb, DISCARD);
1899                         break;
1900                 case Opt_nodiscard:
1901                         clear_opt(sb, DISCARD);
1902                         break;
1903                 case Opt_dioread_nolock:
1904                         set_opt(sb, DIOREAD_NOLOCK);
1905                         break;
1906                 case Opt_dioread_lock:
1907                         clear_opt(sb, DIOREAD_NOLOCK);
1908                         break;
1909                 case Opt_init_itable:
1910                         set_opt(sb, INIT_INODE_TABLE);
1911                         if (args[0].from) {
1912                                 if (match_int(&args[0], &option))
1913                                         return 0;
1914                         } else
1915                                 option = EXT4_DEF_LI_WAIT_MULT;
1916                         if (option < 0)
1917                                 return 0;
1918                         sbi->s_li_wait_mult = option;
1919                         break;
1920                 case Opt_noinit_itable:
1921                         clear_opt(sb, INIT_INODE_TABLE);
1922                         break;
1923                 default:
1924                         ext4_msg(sb, KERN_ERR,
1925                                "Unrecognized mount option \"%s\" "
1926                                "or missing value", p);
1927                         return 0;
1928                 }
1929         }
1930 #ifdef CONFIG_QUOTA
1931         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1932                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1933                         clear_opt(sb, USRQUOTA);
1934
1935                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1936                         clear_opt(sb, GRPQUOTA);
1937
1938                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1939                         ext4_msg(sb, KERN_ERR, "old and new quota "
1940                                         "format mixing");
1941                         return 0;
1942                 }
1943
1944                 if (!sbi->s_jquota_fmt) {
1945                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1946                                         "not specified");
1947                         return 0;
1948                 }
1949         } else {
1950                 if (sbi->s_jquota_fmt) {
1951                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1952                                         "specified with no journaling "
1953                                         "enabled");
1954                         return 0;
1955                 }
1956         }
1957 #endif
1958         return 1;
1959 }
1960
1961 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1962                             int read_only)
1963 {
1964         struct ext4_sb_info *sbi = EXT4_SB(sb);
1965         int res = 0;
1966
1967         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1968                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1969                          "forcing read-only mode");
1970                 res = MS_RDONLY;
1971         }
1972         if (read_only)
1973                 goto done;
1974         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1975                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1976                          "running e2fsck is recommended");
1977         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1978                 ext4_msg(sb, KERN_WARNING,
1979                          "warning: mounting fs with errors, "
1980                          "running e2fsck is recommended");
1981         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1982                  le16_to_cpu(es->s_mnt_count) >=
1983                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1984                 ext4_msg(sb, KERN_WARNING,
1985                          "warning: maximal mount count reached, "
1986                          "running e2fsck is recommended");
1987         else if (le32_to_cpu(es->s_checkinterval) &&
1988                 (le32_to_cpu(es->s_lastcheck) +
1989                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1990                 ext4_msg(sb, KERN_WARNING,
1991                          "warning: checktime reached, "
1992                          "running e2fsck is recommended");
1993         if (!sbi->s_journal)
1994                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1995         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1996                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1997         le16_add_cpu(&es->s_mnt_count, 1);
1998         es->s_mtime = cpu_to_le32(get_seconds());
1999         ext4_update_dynamic_rev(sb);
2000         if (sbi->s_journal)
2001                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2002
2003         ext4_commit_super(sb, 1);
2004 done:
2005         if (test_opt(sb, DEBUG))
2006                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2007                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2008                         sb->s_blocksize,
2009                         sbi->s_groups_count,
2010                         EXT4_BLOCKS_PER_GROUP(sb),
2011                         EXT4_INODES_PER_GROUP(sb),
2012                         sbi->s_mount_opt, sbi->s_mount_opt2);
2013
2014         cleancache_init_fs(sb);
2015         return res;
2016 }
2017
2018 static int ext4_fill_flex_info(struct super_block *sb)
2019 {
2020         struct ext4_sb_info *sbi = EXT4_SB(sb);
2021         struct ext4_group_desc *gdp = NULL;
2022         ext4_group_t flex_group_count;
2023         ext4_group_t flex_group;
2024         unsigned int groups_per_flex = 0;
2025         size_t size;
2026         int i;
2027
2028         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2029         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2030                 sbi->s_log_groups_per_flex = 0;
2031                 return 1;
2032         }
2033         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2034
2035         /* We allocate both existing and potentially added groups */
2036         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2037                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2038                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2039         size = flex_group_count * sizeof(struct flex_groups);
2040         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2041         if (sbi->s_flex_groups == NULL) {
2042                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2043                          flex_group_count);
2044                 goto failed;
2045         }
2046
2047         for (i = 0; i < sbi->s_groups_count; i++) {
2048                 gdp = ext4_get_group_desc(sb, i, NULL);
2049
2050                 flex_group = ext4_flex_group(sbi, i);
2051                 atomic_add(ext4_free_inodes_count(sb, gdp),
2052                            &sbi->s_flex_groups[flex_group].free_inodes);
2053                 atomic_add(ext4_free_group_clusters(sb, gdp),
2054                            &sbi->s_flex_groups[flex_group].free_clusters);
2055                 atomic_add(ext4_used_dirs_count(sb, gdp),
2056                            &sbi->s_flex_groups[flex_group].used_dirs);
2057         }
2058
2059         return 1;
2060 failed:
2061         return 0;
2062 }
2063
2064 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2065                             struct ext4_group_desc *gdp)
2066 {
2067         __u16 crc = 0;
2068
2069         if (sbi->s_es->s_feature_ro_compat &
2070             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2071                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2072                 __le32 le_group = cpu_to_le32(block_group);
2073
2074                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2075                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2076                 crc = crc16(crc, (__u8 *)gdp, offset);
2077                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2078                 /* for checksum of struct ext4_group_desc do the rest...*/
2079                 if ((sbi->s_es->s_feature_incompat &
2080                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2081                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2082                         crc = crc16(crc, (__u8 *)gdp + offset,
2083                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2084                                         offset);
2085         }
2086
2087         return cpu_to_le16(crc);
2088 }
2089
2090 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2091                                 struct ext4_group_desc *gdp)
2092 {
2093         if ((sbi->s_es->s_feature_ro_compat &
2094              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2095             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2096                 return 0;
2097
2098         return 1;
2099 }
2100
2101 /* Called at mount-time, super-block is locked */
2102 static int ext4_check_descriptors(struct super_block *sb,
2103                                   ext4_group_t *first_not_zeroed)
2104 {
2105         struct ext4_sb_info *sbi = EXT4_SB(sb);
2106         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2107         ext4_fsblk_t last_block;
2108         ext4_fsblk_t block_bitmap;
2109         ext4_fsblk_t inode_bitmap;
2110         ext4_fsblk_t inode_table;
2111         int flexbg_flag = 0;
2112         ext4_group_t i, grp = sbi->s_groups_count;
2113
2114         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2115                 flexbg_flag = 1;
2116
2117         ext4_debug("Checking group descriptors");
2118
2119         for (i = 0; i < sbi->s_groups_count; i++) {
2120                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2121
2122                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2123                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2124                 else
2125                         last_block = first_block +
2126                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2127
2128                 if ((grp == sbi->s_groups_count) &&
2129                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2130                         grp = i;
2131
2132                 block_bitmap = ext4_block_bitmap(sb, gdp);
2133                 if (block_bitmap < first_block || block_bitmap > last_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                "Block bitmap for group %u not in group "
2136                                "(block %llu)!", i, block_bitmap);
2137                         return 0;
2138                 }
2139                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2140                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2141                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2142                                "Inode bitmap for group %u not in group "
2143                                "(block %llu)!", i, inode_bitmap);
2144                         return 0;
2145                 }
2146                 inode_table = ext4_inode_table(sb, gdp);
2147                 if (inode_table < first_block ||
2148                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2149                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2150                                "Inode table for group %u not in group "
2151                                "(block %llu)!", i, inode_table);
2152                         return 0;
2153                 }
2154                 ext4_lock_group(sb, i);
2155                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2156                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2157                                  "Checksum for group %u failed (%u!=%u)",
2158                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2159                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2160                         if (!(sb->s_flags & MS_RDONLY)) {
2161                                 ext4_unlock_group(sb, i);
2162                                 return 0;
2163                         }
2164                 }
2165                 ext4_unlock_group(sb, i);
2166                 if (!flexbg_flag)
2167                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2168         }
2169         if (NULL != first_not_zeroed)
2170                 *first_not_zeroed = grp;
2171
2172         ext4_free_blocks_count_set(sbi->s_es,
2173                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2174         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2175         return 1;
2176 }
2177
2178 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2179  * the superblock) which were deleted from all directories, but held open by
2180  * a process at the time of a crash.  We walk the list and try to delete these
2181  * inodes at recovery time (only with a read-write filesystem).
2182  *
2183  * In order to keep the orphan inode chain consistent during traversal (in
2184  * case of crash during recovery), we link each inode into the superblock
2185  * orphan list_head and handle it the same way as an inode deletion during
2186  * normal operation (which journals the operations for us).
2187  *
2188  * We only do an iget() and an iput() on each inode, which is very safe if we
2189  * accidentally point at an in-use or already deleted inode.  The worst that
2190  * can happen in this case is that we get a "bit already cleared" message from
2191  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2192  * e2fsck was run on this filesystem, and it must have already done the orphan
2193  * inode cleanup for us, so we can safely abort without any further action.
2194  */
2195 static void ext4_orphan_cleanup(struct super_block *sb,
2196                                 struct ext4_super_block *es)
2197 {
2198         unsigned int s_flags = sb->s_flags;
2199         int nr_orphans = 0, nr_truncates = 0;
2200 #ifdef CONFIG_QUOTA
2201         int i;
2202 #endif
2203         if (!es->s_last_orphan) {
2204                 jbd_debug(4, "no orphan inodes to clean up\n");
2205                 return;
2206         }
2207
2208         if (bdev_read_only(sb->s_bdev)) {
2209                 ext4_msg(sb, KERN_ERR, "write access "
2210                         "unavailable, skipping orphan cleanup");
2211                 return;
2212         }
2213
2214         /* Check if feature set would not allow a r/w mount */
2215         if (!ext4_feature_set_ok(sb, 0)) {
2216                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2217                          "unknown ROCOMPAT features");
2218                 return;
2219         }
2220
2221         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2222                 if (es->s_last_orphan)
2223                         jbd_debug(1, "Errors on filesystem, "
2224                                   "clearing orphan list.\n");
2225                 es->s_last_orphan = 0;
2226                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2227                 return;
2228         }
2229
2230         if (s_flags & MS_RDONLY) {
2231                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2232                 sb->s_flags &= ~MS_RDONLY;
2233         }
2234 #ifdef CONFIG_QUOTA
2235         /* Needed for iput() to work correctly and not trash data */
2236         sb->s_flags |= MS_ACTIVE;
2237         /* Turn on quotas so that they are updated correctly */
2238         for (i = 0; i < MAXQUOTAS; i++) {
2239                 if (EXT4_SB(sb)->s_qf_names[i]) {
2240                         int ret = ext4_quota_on_mount(sb, i);
2241                         if (ret < 0)
2242                                 ext4_msg(sb, KERN_ERR,
2243                                         "Cannot turn on journaled "
2244                                         "quota: error %d", ret);
2245                 }
2246         }
2247 #endif
2248
2249         while (es->s_last_orphan) {
2250                 struct inode *inode;
2251
2252                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2253                 if (IS_ERR(inode)) {
2254                         es->s_last_orphan = 0;
2255                         break;
2256                 }
2257
2258                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2259                 dquot_initialize(inode);
2260                 if (inode->i_nlink) {
2261                         ext4_msg(sb, KERN_DEBUG,
2262                                 "%s: truncating inode %lu to %lld bytes",
2263                                 __func__, inode->i_ino, inode->i_size);
2264                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2265                                   inode->i_ino, inode->i_size);
2266                         ext4_truncate(inode);
2267                         nr_truncates++;
2268                 } else {
2269                         ext4_msg(sb, KERN_DEBUG,
2270                                 "%s: deleting unreferenced inode %lu",
2271                                 __func__, inode->i_ino);
2272                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2273                                   inode->i_ino);
2274                         nr_orphans++;
2275                 }
2276                 iput(inode);  /* The delete magic happens here! */
2277         }
2278
2279 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2280
2281         if (nr_orphans)
2282                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2283                        PLURAL(nr_orphans));
2284         if (nr_truncates)
2285                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2286                        PLURAL(nr_truncates));
2287 #ifdef CONFIG_QUOTA
2288         /* Turn quotas off */
2289         for (i = 0; i < MAXQUOTAS; i++) {
2290                 if (sb_dqopt(sb)->files[i])
2291                         dquot_quota_off(sb, i);
2292         }
2293 #endif
2294         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2295 }
2296
2297 /*
2298  * Maximal extent format file size.
2299  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2300  * extent format containers, within a sector_t, and within i_blocks
2301  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2302  * so that won't be a limiting factor.
2303  *
2304  * However there is other limiting factor. We do store extents in the form
2305  * of starting block and length, hence the resulting length of the extent
2306  * covering maximum file size must fit into on-disk format containers as
2307  * well. Given that length is always by 1 unit bigger than max unit (because
2308  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2309  *
2310  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2311  */
2312 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2313 {
2314         loff_t res;
2315         loff_t upper_limit = MAX_LFS_FILESIZE;
2316
2317         /* small i_blocks in vfs inode? */
2318         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2319                 /*
2320                  * CONFIG_LBDAF is not enabled implies the inode
2321                  * i_block represent total blocks in 512 bytes
2322                  * 32 == size of vfs inode i_blocks * 8
2323                  */
2324                 upper_limit = (1LL << 32) - 1;
2325
2326                 /* total blocks in file system block size */
2327                 upper_limit >>= (blkbits - 9);
2328                 upper_limit <<= blkbits;
2329         }
2330
2331         /*
2332          * 32-bit extent-start container, ee_block. We lower the maxbytes
2333          * by one fs block, so ee_len can cover the extent of maximum file
2334          * size
2335          */
2336         res = (1LL << 32) - 1;
2337         res <<= blkbits;
2338
2339         /* Sanity check against vm- & vfs- imposed limits */
2340         if (res > upper_limit)
2341                 res = upper_limit;
2342
2343         return res;
2344 }
2345
2346 /*
2347  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2348  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2349  * We need to be 1 filesystem block less than the 2^48 sector limit.
2350  */
2351 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2352 {
2353         loff_t res = EXT4_NDIR_BLOCKS;
2354         int meta_blocks;
2355         loff_t upper_limit;
2356         /* This is calculated to be the largest file size for a dense, block
2357          * mapped file such that the file's total number of 512-byte sectors,
2358          * including data and all indirect blocks, does not exceed (2^48 - 1).
2359          *
2360          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2361          * number of 512-byte sectors of the file.
2362          */
2363
2364         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2365                 /*
2366                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2367                  * the inode i_block field represents total file blocks in
2368                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2369                  */
2370                 upper_limit = (1LL << 32) - 1;
2371
2372                 /* total blocks in file system block size */
2373                 upper_limit >>= (bits - 9);
2374
2375         } else {
2376                 /*
2377                  * We use 48 bit ext4_inode i_blocks
2378                  * With EXT4_HUGE_FILE_FL set the i_blocks
2379                  * represent total number of blocks in
2380                  * file system block size
2381                  */
2382                 upper_limit = (1LL << 48) - 1;
2383
2384         }
2385
2386         /* indirect blocks */
2387         meta_blocks = 1;
2388         /* double indirect blocks */
2389         meta_blocks += 1 + (1LL << (bits-2));
2390         /* tripple indirect blocks */
2391         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2392
2393         upper_limit -= meta_blocks;
2394         upper_limit <<= bits;
2395
2396         res += 1LL << (bits-2);
2397         res += 1LL << (2*(bits-2));
2398         res += 1LL << (3*(bits-2));
2399         res <<= bits;
2400         if (res > upper_limit)
2401                 res = upper_limit;
2402
2403         if (res > MAX_LFS_FILESIZE)
2404                 res = MAX_LFS_FILESIZE;
2405
2406         return res;
2407 }
2408
2409 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2410                                    ext4_fsblk_t logical_sb_block, int nr)
2411 {
2412         struct ext4_sb_info *sbi = EXT4_SB(sb);
2413         ext4_group_t bg, first_meta_bg;
2414         int has_super = 0;
2415
2416         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2417
2418         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2419             nr < first_meta_bg)
2420                 return logical_sb_block + nr + 1;
2421         bg = sbi->s_desc_per_block * nr;
2422         if (ext4_bg_has_super(sb, bg))
2423                 has_super = 1;
2424
2425         return (has_super + ext4_group_first_block_no(sb, bg));
2426 }
2427
2428 /**
2429  * ext4_get_stripe_size: Get the stripe size.
2430  * @sbi: In memory super block info
2431  *
2432  * If we have specified it via mount option, then
2433  * use the mount option value. If the value specified at mount time is
2434  * greater than the blocks per group use the super block value.
2435  * If the super block value is greater than blocks per group return 0.
2436  * Allocator needs it be less than blocks per group.
2437  *
2438  */
2439 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2440 {
2441         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2442         unsigned long stripe_width =
2443                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2444         int ret;
2445
2446         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2447                 ret = sbi->s_stripe;
2448         else if (stripe_width <= sbi->s_blocks_per_group)
2449                 ret = stripe_width;
2450         else if (stride <= sbi->s_blocks_per_group)
2451                 ret = stride;
2452         else
2453                 ret = 0;
2454
2455         /*
2456          * If the stripe width is 1, this makes no sense and
2457          * we set it to 0 to turn off stripe handling code.
2458          */
2459         if (ret <= 1)
2460                 ret = 0;
2461
2462         return ret;
2463 }
2464
2465 /* sysfs supprt */
2466
2467 struct ext4_attr {
2468         struct attribute attr;
2469         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2470         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2471                          const char *, size_t);
2472         int offset;
2473 };
2474
2475 static int parse_strtoul(const char *buf,
2476                 unsigned long max, unsigned long *value)
2477 {
2478         char *endp;
2479
2480         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2481         endp = skip_spaces(endp);
2482         if (*endp || *value > max)
2483                 return -EINVAL;
2484
2485         return 0;
2486 }
2487
2488 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2489                                               struct ext4_sb_info *sbi,
2490                                               char *buf)
2491 {
2492         return snprintf(buf, PAGE_SIZE, "%llu\n",
2493                 (s64) EXT4_C2B(sbi,
2494                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2495 }
2496
2497 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2498                                          struct ext4_sb_info *sbi, char *buf)
2499 {
2500         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2501
2502         if (!sb->s_bdev->bd_part)
2503                 return snprintf(buf, PAGE_SIZE, "0\n");
2504         return snprintf(buf, PAGE_SIZE, "%lu\n",
2505                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2506                          sbi->s_sectors_written_start) >> 1);
2507 }
2508
2509 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2510                                           struct ext4_sb_info *sbi, char *buf)
2511 {
2512         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2513
2514         if (!sb->s_bdev->bd_part)
2515                 return snprintf(buf, PAGE_SIZE, "0\n");
2516         return snprintf(buf, PAGE_SIZE, "%llu\n",
2517                         (unsigned long long)(sbi->s_kbytes_written +
2518                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2519                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2520 }
2521
2522 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2523                                       struct ext4_sb_info *sbi, char *buf)
2524 {
2525         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2526 }
2527
2528 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2529                                         struct ext4_sb_info *sbi, char *buf)
2530 {
2531         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2532 }
2533
2534 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2535                                           struct ext4_sb_info *sbi,
2536                                           const char *buf, size_t count)
2537 {
2538         unsigned long t;
2539
2540         if (parse_strtoul(buf, 0x40000000, &t))
2541                 return -EINVAL;
2542
2543         if (t && !is_power_of_2(t))
2544                 return -EINVAL;
2545
2546         sbi->s_inode_readahead_blks = t;
2547         return count;
2548 }
2549
2550 static ssize_t sbi_ui_show(struct ext4_attr *a,
2551                            struct ext4_sb_info *sbi, char *buf)
2552 {
2553         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2554
2555         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2556 }
2557
2558 static ssize_t sbi_ui_store(struct ext4_attr *a,
2559                             struct ext4_sb_info *sbi,
2560                             const char *buf, size_t count)
2561 {
2562         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2563         unsigned long t;
2564
2565         if (parse_strtoul(buf, 0xffffffff, &t))
2566                 return -EINVAL;
2567         *ui = t;
2568         return count;
2569 }
2570
2571 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2572 static struct ext4_attr ext4_attr_##_name = {                   \
2573         .attr = {.name = __stringify(_name), .mode = _mode },   \
2574         .show   = _show,                                        \
2575         .store  = _store,                                       \
2576         .offset = offsetof(struct ext4_sb_info, _elname),       \
2577 }
2578 #define EXT4_ATTR(name, mode, show, store) \
2579 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2580
2581 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2582 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2583 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2584 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2585         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2586 #define ATTR_LIST(name) &ext4_attr_##name.attr
2587
2588 EXT4_RO_ATTR(delayed_allocation_blocks);
2589 EXT4_RO_ATTR(session_write_kbytes);
2590 EXT4_RO_ATTR(lifetime_write_kbytes);
2591 EXT4_RO_ATTR(extent_cache_hits);
2592 EXT4_RO_ATTR(extent_cache_misses);
2593 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2594                  inode_readahead_blks_store, s_inode_readahead_blks);
2595 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2596 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2597 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2598 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2599 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2600 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2601 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2602 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2603
2604 static struct attribute *ext4_attrs[] = {
2605         ATTR_LIST(delayed_allocation_blocks),
2606         ATTR_LIST(session_write_kbytes),
2607         ATTR_LIST(lifetime_write_kbytes),
2608         ATTR_LIST(extent_cache_hits),
2609         ATTR_LIST(extent_cache_misses),
2610         ATTR_LIST(inode_readahead_blks),
2611         ATTR_LIST(inode_goal),
2612         ATTR_LIST(mb_stats),
2613         ATTR_LIST(mb_max_to_scan),
2614         ATTR_LIST(mb_min_to_scan),
2615         ATTR_LIST(mb_order2_req),
2616         ATTR_LIST(mb_stream_req),
2617         ATTR_LIST(mb_group_prealloc),
2618         ATTR_LIST(max_writeback_mb_bump),
2619         NULL,
2620 };
2621
2622 /* Features this copy of ext4 supports */
2623 EXT4_INFO_ATTR(lazy_itable_init);
2624 EXT4_INFO_ATTR(batched_discard);
2625
2626 static struct attribute *ext4_feat_attrs[] = {
2627         ATTR_LIST(lazy_itable_init),
2628         ATTR_LIST(batched_discard),
2629         NULL,
2630 };
2631
2632 static ssize_t ext4_attr_show(struct kobject *kobj,
2633                               struct attribute *attr, char *buf)
2634 {
2635         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2636                                                 s_kobj);
2637         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2638
2639         return a->show ? a->show(a, sbi, buf) : 0;
2640 }
2641
2642 static ssize_t ext4_attr_store(struct kobject *kobj,
2643                                struct attribute *attr,
2644                                const char *buf, size_t len)
2645 {
2646         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2647                                                 s_kobj);
2648         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2649
2650         return a->store ? a->store(a, sbi, buf, len) : 0;
2651 }
2652
2653 static void ext4_sb_release(struct kobject *kobj)
2654 {
2655         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2656                                                 s_kobj);
2657         complete(&sbi->s_kobj_unregister);
2658 }
2659
2660 static const struct sysfs_ops ext4_attr_ops = {
2661         .show   = ext4_attr_show,
2662         .store  = ext4_attr_store,
2663 };
2664
2665 static struct kobj_type ext4_ktype = {
2666         .default_attrs  = ext4_attrs,
2667         .sysfs_ops      = &ext4_attr_ops,
2668         .release        = ext4_sb_release,
2669 };
2670
2671 static void ext4_feat_release(struct kobject *kobj)
2672 {
2673         complete(&ext4_feat->f_kobj_unregister);
2674 }
2675
2676 static struct kobj_type ext4_feat_ktype = {
2677         .default_attrs  = ext4_feat_attrs,
2678         .sysfs_ops      = &ext4_attr_ops,
2679         .release        = ext4_feat_release,
2680 };
2681
2682 /*
2683  * Check whether this filesystem can be mounted based on
2684  * the features present and the RDONLY/RDWR mount requested.
2685  * Returns 1 if this filesystem can be mounted as requested,
2686  * 0 if it cannot be.
2687  */
2688 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2689 {
2690         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2691                 ext4_msg(sb, KERN_ERR,
2692                         "Couldn't mount because of "
2693                         "unsupported optional features (%x)",
2694                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2695                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2696                 return 0;
2697         }
2698
2699         if (readonly)
2700                 return 1;
2701
2702         /* Check that feature set is OK for a read-write mount */
2703         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2704                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2705                          "unsupported optional features (%x)",
2706                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2707                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2708                 return 0;
2709         }
2710         /*
2711          * Large file size enabled file system can only be mounted
2712          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2713          */
2714         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2715                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2716                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2717                                  "cannot be mounted RDWR without "
2718                                  "CONFIG_LBDAF");
2719                         return 0;
2720                 }
2721         }
2722         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2723             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2724                 ext4_msg(sb, KERN_ERR,
2725                          "Can't support bigalloc feature without "
2726                          "extents feature\n");
2727                 return 0;
2728         }
2729         return 1;
2730 }
2731
2732 /*
2733  * This function is called once a day if we have errors logged
2734  * on the file system
2735  */
2736 static void print_daily_error_info(unsigned long arg)
2737 {
2738         struct super_block *sb = (struct super_block *) arg;
2739         struct ext4_sb_info *sbi;
2740         struct ext4_super_block *es;
2741
2742         sbi = EXT4_SB(sb);
2743         es = sbi->s_es;
2744
2745         if (es->s_error_count)
2746                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2747                          le32_to_cpu(es->s_error_count));
2748         if (es->s_first_error_time) {
2749                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2750                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2751                        (int) sizeof(es->s_first_error_func),
2752                        es->s_first_error_func,
2753                        le32_to_cpu(es->s_first_error_line));
2754                 if (es->s_first_error_ino)
2755                         printk(": inode %u",
2756                                le32_to_cpu(es->s_first_error_ino));
2757                 if (es->s_first_error_block)
2758                         printk(": block %llu", (unsigned long long)
2759                                le64_to_cpu(es->s_first_error_block));
2760                 printk("\n");
2761         }
2762         if (es->s_last_error_time) {
2763                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2764                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2765                        (int) sizeof(es->s_last_error_func),
2766                        es->s_last_error_func,
2767                        le32_to_cpu(es->s_last_error_line));
2768                 if (es->s_last_error_ino)
2769                         printk(": inode %u",
2770                                le32_to_cpu(es->s_last_error_ino));
2771                 if (es->s_last_error_block)
2772                         printk(": block %llu", (unsigned long long)
2773                                le64_to_cpu(es->s_last_error_block));
2774                 printk("\n");
2775         }
2776         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2777 }
2778
2779 /* Find next suitable group and run ext4_init_inode_table */
2780 static int ext4_run_li_request(struct ext4_li_request *elr)
2781 {
2782         struct ext4_group_desc *gdp = NULL;
2783         ext4_group_t group, ngroups;
2784         struct super_block *sb;
2785         unsigned long timeout = 0;
2786         int ret = 0;
2787
2788         sb = elr->lr_super;
2789         ngroups = EXT4_SB(sb)->s_groups_count;
2790
2791         for (group = elr->lr_next_group; group < ngroups; group++) {
2792                 gdp = ext4_get_group_desc(sb, group, NULL);
2793                 if (!gdp) {
2794                         ret = 1;
2795                         break;
2796                 }
2797
2798                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2799                         break;
2800         }
2801
2802         if (group == ngroups)
2803                 ret = 1;
2804
2805         if (!ret) {
2806                 timeout = jiffies;
2807                 ret = ext4_init_inode_table(sb, group,
2808                                             elr->lr_timeout ? 0 : 1);
2809                 if (elr->lr_timeout == 0) {
2810                         timeout = (jiffies - timeout) *
2811                                   elr->lr_sbi->s_li_wait_mult;
2812                         elr->lr_timeout = timeout;
2813                 }
2814                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2815                 elr->lr_next_group = group + 1;
2816         }
2817
2818         return ret;
2819 }
2820
2821 /*
2822  * Remove lr_request from the list_request and free the
2823  * request structure. Should be called with li_list_mtx held
2824  */
2825 static void ext4_remove_li_request(struct ext4_li_request *elr)
2826 {
2827         struct ext4_sb_info *sbi;
2828
2829         if (!elr)
2830                 return;
2831
2832         sbi = elr->lr_sbi;
2833
2834         list_del(&elr->lr_request);
2835         sbi->s_li_request = NULL;
2836         kfree(elr);
2837 }
2838
2839 static void ext4_unregister_li_request(struct super_block *sb)
2840 {
2841         mutex_lock(&ext4_li_mtx);
2842         if (!ext4_li_info) {
2843                 mutex_unlock(&ext4_li_mtx);
2844                 return;
2845         }
2846
2847         mutex_lock(&ext4_li_info->li_list_mtx);
2848         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2849         mutex_unlock(&ext4_li_info->li_list_mtx);
2850         mutex_unlock(&ext4_li_mtx);
2851 }
2852
2853 static struct task_struct *ext4_lazyinit_task;
2854
2855 /*
2856  * This is the function where ext4lazyinit thread lives. It walks
2857  * through the request list searching for next scheduled filesystem.
2858  * When such a fs is found, run the lazy initialization request
2859  * (ext4_rn_li_request) and keep track of the time spend in this
2860  * function. Based on that time we compute next schedule time of
2861  * the request. When walking through the list is complete, compute
2862  * next waking time and put itself into sleep.
2863  */
2864 static int ext4_lazyinit_thread(void *arg)
2865 {
2866         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2867         struct list_head *pos, *n;
2868         struct ext4_li_request *elr;
2869         unsigned long next_wakeup, cur;
2870
2871         BUG_ON(NULL == eli);
2872
2873 cont_thread:
2874         while (true) {
2875                 next_wakeup = MAX_JIFFY_OFFSET;
2876
2877                 mutex_lock(&eli->li_list_mtx);
2878                 if (list_empty(&eli->li_request_list)) {
2879                         mutex_unlock(&eli->li_list_mtx);
2880                         goto exit_thread;
2881                 }
2882
2883                 list_for_each_safe(pos, n, &eli->li_request_list) {
2884                         elr = list_entry(pos, struct ext4_li_request,
2885                                          lr_request);
2886
2887                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2888                                 if (ext4_run_li_request(elr) != 0) {
2889                                         /* error, remove the lazy_init job */
2890                                         ext4_remove_li_request(elr);
2891                                         continue;
2892                                 }
2893                         }
2894
2895                         if (time_before(elr->lr_next_sched, next_wakeup))
2896                                 next_wakeup = elr->lr_next_sched;
2897                 }
2898                 mutex_unlock(&eli->li_list_mtx);
2899
2900                 try_to_freeze();
2901
2902                 cur = jiffies;
2903                 if ((time_after_eq(cur, next_wakeup)) ||
2904                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2905                         cond_resched();
2906                         continue;
2907                 }
2908
2909                 schedule_timeout_interruptible(next_wakeup - cur);
2910
2911                 if (kthread_should_stop()) {
2912                         ext4_clear_request_list();
2913                         goto exit_thread;
2914                 }
2915         }
2916
2917 exit_thread:
2918         /*
2919          * It looks like the request list is empty, but we need
2920          * to check it under the li_list_mtx lock, to prevent any
2921          * additions into it, and of course we should lock ext4_li_mtx
2922          * to atomically free the list and ext4_li_info, because at
2923          * this point another ext4 filesystem could be registering
2924          * new one.
2925          */
2926         mutex_lock(&ext4_li_mtx);
2927         mutex_lock(&eli->li_list_mtx);
2928         if (!list_empty(&eli->li_request_list)) {
2929                 mutex_unlock(&eli->li_list_mtx);
2930                 mutex_unlock(&ext4_li_mtx);
2931                 goto cont_thread;
2932         }
2933         mutex_unlock(&eli->li_list_mtx);
2934         kfree(ext4_li_info);
2935         ext4_li_info = NULL;
2936         mutex_unlock(&ext4_li_mtx);
2937
2938         return 0;
2939 }
2940
2941 static void ext4_clear_request_list(void)
2942 {
2943         struct list_head *pos, *n;
2944         struct ext4_li_request *elr;
2945
2946         mutex_lock(&ext4_li_info->li_list_mtx);
2947         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2948                 elr = list_entry(pos, struct ext4_li_request,
2949                                  lr_request);
2950                 ext4_remove_li_request(elr);
2951         }
2952         mutex_unlock(&ext4_li_info->li_list_mtx);
2953 }
2954
2955 static int ext4_run_lazyinit_thread(void)
2956 {
2957         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2958                                          ext4_li_info, "ext4lazyinit");
2959         if (IS_ERR(ext4_lazyinit_task)) {
2960                 int err = PTR_ERR(ext4_lazyinit_task);
2961                 ext4_clear_request_list();
2962                 kfree(ext4_li_info);
2963                 ext4_li_info = NULL;
2964                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2965                                  "initialization thread\n",
2966                                  err);
2967                 return err;
2968         }
2969         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2970         return 0;
2971 }
2972
2973 /*
2974  * Check whether it make sense to run itable init. thread or not.
2975  * If there is at least one uninitialized inode table, return
2976  * corresponding group number, else the loop goes through all
2977  * groups and return total number of groups.
2978  */
2979 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2980 {
2981         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2982         struct ext4_group_desc *gdp = NULL;
2983
2984         for (group = 0; group < ngroups; group++) {
2985                 gdp = ext4_get_group_desc(sb, group, NULL);
2986                 if (!gdp)
2987                         continue;
2988
2989                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2990                         break;
2991         }
2992
2993         return group;
2994 }
2995
2996 static int ext4_li_info_new(void)
2997 {
2998         struct ext4_lazy_init *eli = NULL;
2999
3000         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3001         if (!eli)
3002                 return -ENOMEM;
3003
3004         INIT_LIST_HEAD(&eli->li_request_list);
3005         mutex_init(&eli->li_list_mtx);
3006
3007         eli->li_state |= EXT4_LAZYINIT_QUIT;
3008
3009         ext4_li_info = eli;
3010
3011         return 0;
3012 }
3013
3014 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3015                                             ext4_group_t start)
3016 {
3017         struct ext4_sb_info *sbi = EXT4_SB(sb);
3018         struct ext4_li_request *elr;
3019         unsigned long rnd;
3020
3021         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3022         if (!elr)
3023                 return NULL;
3024
3025         elr->lr_super = sb;
3026         elr->lr_sbi = sbi;
3027         elr->lr_next_group = start;
3028
3029         /*
3030          * Randomize first schedule time of the request to
3031          * spread the inode table initialization requests
3032          * better.
3033          */
3034         get_random_bytes(&rnd, sizeof(rnd));
3035         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3036                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3037
3038         return elr;
3039 }
3040
3041 static int ext4_register_li_request(struct super_block *sb,
3042                                     ext4_group_t first_not_zeroed)
3043 {
3044         struct ext4_sb_info *sbi = EXT4_SB(sb);
3045         struct ext4_li_request *elr;
3046         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3047         int ret = 0;
3048
3049         if (sbi->s_li_request != NULL) {
3050                 /*
3051                  * Reset timeout so it can be computed again, because
3052                  * s_li_wait_mult might have changed.
3053                  */
3054                 sbi->s_li_request->lr_timeout = 0;
3055                 return 0;
3056         }
3057
3058         if (first_not_zeroed == ngroups ||
3059             (sb->s_flags & MS_RDONLY) ||
3060             !test_opt(sb, INIT_INODE_TABLE))
3061                 return 0;
3062
3063         elr = ext4_li_request_new(sb, first_not_zeroed);
3064         if (!elr)
3065                 return -ENOMEM;
3066
3067         mutex_lock(&ext4_li_mtx);
3068
3069         if (NULL == ext4_li_info) {
3070                 ret = ext4_li_info_new();
3071                 if (ret)
3072                         goto out;
3073         }
3074
3075         mutex_lock(&ext4_li_info->li_list_mtx);
3076         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3077         mutex_unlock(&ext4_li_info->li_list_mtx);
3078
3079         sbi->s_li_request = elr;
3080         /*
3081          * set elr to NULL here since it has been inserted to
3082          * the request_list and the removal and free of it is
3083          * handled by ext4_clear_request_list from now on.
3084          */
3085         elr = NULL;
3086
3087         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3088                 ret = ext4_run_lazyinit_thread();
3089                 if (ret)
3090                         goto out;
3091         }
3092 out:
3093         mutex_unlock(&ext4_li_mtx);
3094         if (ret)
3095                 kfree(elr);
3096         return ret;
3097 }
3098
3099 /*
3100  * We do not need to lock anything since this is called on
3101  * module unload.
3102  */
3103 static void ext4_destroy_lazyinit_thread(void)
3104 {
3105         /*
3106          * If thread exited earlier
3107          * there's nothing to be done.
3108          */
3109         if (!ext4_li_info || !ext4_lazyinit_task)
3110                 return;
3111
3112         kthread_stop(ext4_lazyinit_task);
3113 }
3114
3115 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3116 {
3117         char *orig_data = kstrdup(data, GFP_KERNEL);
3118         struct buffer_head *bh;
3119         struct ext4_super_block *es = NULL;
3120         struct ext4_sb_info *sbi;
3121         ext4_fsblk_t block;
3122         ext4_fsblk_t sb_block = get_sb_block(&data);
3123         ext4_fsblk_t logical_sb_block;
3124         unsigned long offset = 0;
3125         unsigned long journal_devnum = 0;
3126         unsigned long def_mount_opts;
3127         struct inode *root;
3128         char *cp;
3129         const char *descr;
3130         int ret = -ENOMEM;
3131         int blocksize, clustersize;
3132         unsigned int db_count;
3133         unsigned int i;
3134         int needs_recovery, has_huge_files, has_bigalloc;
3135         __u64 blocks_count;
3136         int err;
3137         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3138         ext4_group_t first_not_zeroed;
3139
3140         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3141         if (!sbi)
3142                 goto out_free_orig;
3143
3144         sbi->s_blockgroup_lock =
3145                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3146         if (!sbi->s_blockgroup_lock) {
3147                 kfree(sbi);
3148                 goto out_free_orig;
3149         }
3150         sb->s_fs_info = sbi;
3151         sbi->s_mount_opt = 0;
3152         sbi->s_resuid = EXT4_DEF_RESUID;
3153         sbi->s_resgid = EXT4_DEF_RESGID;
3154         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3155         sbi->s_sb_block = sb_block;
3156         if (sb->s_bdev->bd_part)
3157                 sbi->s_sectors_written_start =
3158                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3159
3160         /* Cleanup superblock name */
3161         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3162                 *cp = '!';
3163
3164         ret = -EINVAL;
3165         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3166         if (!blocksize) {
3167                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3168                 goto out_fail;
3169         }
3170
3171         /*
3172          * The ext4 superblock will not be buffer aligned for other than 1kB
3173          * block sizes.  We need to calculate the offset from buffer start.
3174          */
3175         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3176                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3177                 offset = do_div(logical_sb_block, blocksize);
3178         } else {
3179                 logical_sb_block = sb_block;
3180         }
3181
3182         if (!(bh = sb_bread(sb, logical_sb_block))) {
3183                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3184                 goto out_fail;
3185         }
3186         /*
3187          * Note: s_es must be initialized as soon as possible because
3188          *       some ext4 macro-instructions depend on its value
3189          */
3190         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3191         sbi->s_es = es;
3192         sb->s_magic = le16_to_cpu(es->s_magic);
3193         if (sb->s_magic != EXT4_SUPER_MAGIC)
3194                 goto cantfind_ext4;
3195         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3196
3197         /* Set defaults before we parse the mount options */
3198         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3199         set_opt(sb, INIT_INODE_TABLE);
3200         if (def_mount_opts & EXT4_DEFM_DEBUG)
3201                 set_opt(sb, DEBUG);
3202         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3203                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3204                         "2.6.38");
3205                 set_opt(sb, GRPID);
3206         }
3207         if (def_mount_opts & EXT4_DEFM_UID16)
3208                 set_opt(sb, NO_UID32);
3209         /* xattr user namespace & acls are now defaulted on */
3210 #ifdef CONFIG_EXT4_FS_XATTR
3211         set_opt(sb, XATTR_USER);
3212 #endif
3213 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3214         set_opt(sb, POSIX_ACL);
3215 #endif
3216         set_opt(sb, MBLK_IO_SUBMIT);
3217         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3218                 set_opt(sb, JOURNAL_DATA);
3219         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3220                 set_opt(sb, ORDERED_DATA);
3221         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3222                 set_opt(sb, WRITEBACK_DATA);
3223
3224         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3225                 set_opt(sb, ERRORS_PANIC);
3226         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3227                 set_opt(sb, ERRORS_CONT);
3228         else
3229                 set_opt(sb, ERRORS_RO);
3230         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3231                 set_opt(sb, BLOCK_VALIDITY);
3232         if (def_mount_opts & EXT4_DEFM_DISCARD)
3233                 set_opt(sb, DISCARD);
3234
3235         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3236         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3237         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3238         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3239         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3240
3241         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3242                 set_opt(sb, BARRIER);
3243
3244         /*
3245          * enable delayed allocation by default
3246          * Use -o nodelalloc to turn it off
3247          */
3248         if (!IS_EXT3_SB(sb) &&
3249             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3250                 set_opt(sb, DELALLOC);
3251
3252         /*
3253          * set default s_li_wait_mult for lazyinit, for the case there is
3254          * no mount option specified.
3255          */
3256         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3257
3258         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3259                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3260                 ext4_msg(sb, KERN_WARNING,
3261                          "failed to parse options in superblock: %s",
3262                          sbi->s_es->s_mount_opts);
3263         }
3264         if (!parse_options((char *) data, sb, &journal_devnum,
3265                            &journal_ioprio, NULL, 0))
3266                 goto failed_mount;
3267
3268         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3269                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3270                             "with data=journal disables delayed "
3271                             "allocation and O_DIRECT support!\n");
3272                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3273                         ext4_msg(sb, KERN_ERR, "can't mount with "
3274                                  "both data=journal and delalloc");
3275                         goto failed_mount;
3276                 }
3277                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3278                         ext4_msg(sb, KERN_ERR, "can't mount with "
3279                                  "both data=journal and delalloc");
3280                         goto failed_mount;
3281                 }
3282                 if (test_opt(sb, DELALLOC))
3283                         clear_opt(sb, DELALLOC);
3284         }
3285
3286         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3287         if (test_opt(sb, DIOREAD_NOLOCK)) {
3288                 if (blocksize < PAGE_SIZE) {
3289                         ext4_msg(sb, KERN_ERR, "can't mount with "
3290                                  "dioread_nolock if block size != PAGE_SIZE");
3291                         goto failed_mount;
3292                 }
3293         }
3294
3295         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3296                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3297
3298         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3299             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3300              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3301              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3302                 ext4_msg(sb, KERN_WARNING,
3303                        "feature flags set on rev 0 fs, "
3304                        "running e2fsck is recommended");
3305
3306         if (IS_EXT2_SB(sb)) {
3307                 if (ext2_feature_set_ok(sb))
3308                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3309                                  "using the ext4 subsystem");
3310                 else {
3311                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3312                                  "to feature incompatibilities");
3313                         goto failed_mount;
3314                 }
3315         }
3316
3317         if (IS_EXT3_SB(sb)) {
3318                 if (ext3_feature_set_ok(sb))
3319                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3320                                  "using the ext4 subsystem");
3321                 else {
3322                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3323                                  "to feature incompatibilities");
3324                         goto failed_mount;
3325                 }
3326         }
3327
3328         /*
3329          * Check feature flags regardless of the revision level, since we
3330          * previously didn't change the revision level when setting the flags,
3331          * so there is a chance incompat flags are set on a rev 0 filesystem.
3332          */
3333         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3334                 goto failed_mount;
3335
3336         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3337             blocksize > EXT4_MAX_BLOCK_SIZE) {
3338                 ext4_msg(sb, KERN_ERR,
3339                        "Unsupported filesystem blocksize %d", blocksize);
3340                 goto failed_mount;
3341         }
3342
3343         if (sb->s_blocksize != blocksize) {
3344                 /* Validate the filesystem blocksize */
3345                 if (!sb_set_blocksize(sb, blocksize)) {
3346                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3347                                         blocksize);
3348                         goto failed_mount;
3349                 }
3350
3351                 brelse(bh);
3352                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3353                 offset = do_div(logical_sb_block, blocksize);
3354                 bh = sb_bread(sb, logical_sb_block);
3355                 if (!bh) {
3356                         ext4_msg(sb, KERN_ERR,
3357                                "Can't read superblock on 2nd try");
3358                         goto failed_mount;
3359                 }
3360                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3361                 sbi->s_es = es;
3362                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3363                         ext4_msg(sb, KERN_ERR,
3364                                "Magic mismatch, very weird!");
3365                         goto failed_mount;
3366                 }
3367         }
3368
3369         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3370                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3371         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3372                                                       has_huge_files);
3373         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3374
3375         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3376                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3377                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3378         } else {
3379                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3380                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3381                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3382                     (!is_power_of_2(sbi->s_inode_size)) ||
3383                     (sbi->s_inode_size > blocksize)) {
3384                         ext4_msg(sb, KERN_ERR,
3385                                "unsupported inode size: %d",
3386                                sbi->s_inode_size);
3387                         goto failed_mount;
3388                 }
3389                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3390                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3391         }
3392
3393         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3394         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3395                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3396                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3397                     !is_power_of_2(sbi->s_desc_size)) {
3398                         ext4_msg(sb, KERN_ERR,
3399                                "unsupported descriptor size %lu",
3400                                sbi->s_desc_size);
3401                         goto failed_mount;
3402                 }
3403         } else
3404                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3405
3406         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3407         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3408         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3409                 goto cantfind_ext4;
3410
3411         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3412         if (sbi->s_inodes_per_block == 0)
3413                 goto cantfind_ext4;
3414         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3415                                         sbi->s_inodes_per_block;
3416         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3417         sbi->s_sbh = bh;
3418         sbi->s_mount_state = le16_to_cpu(es->s_state);
3419         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3420         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3421
3422         for (i = 0; i < 4; i++)
3423                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3424         sbi->s_def_hash_version = es->s_def_hash_version;
3425         i = le32_to_cpu(es->s_flags);
3426         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3427                 sbi->s_hash_unsigned = 3;
3428         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3429 #ifdef __CHAR_UNSIGNED__
3430                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3431                 sbi->s_hash_unsigned = 3;
3432 #else
3433                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3434 #endif
3435                 sb->s_dirt = 1;
3436         }
3437
3438         /* Handle clustersize */
3439         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3440         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3441                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3442         if (has_bigalloc) {
3443                 if (clustersize < blocksize) {
3444                         ext4_msg(sb, KERN_ERR,
3445                                  "cluster size (%d) smaller than "
3446                                  "block size (%d)", clustersize, blocksize);
3447                         goto failed_mount;
3448                 }
3449                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3450                         le32_to_cpu(es->s_log_block_size);
3451                 sbi->s_clusters_per_group =
3452                         le32_to_cpu(es->s_clusters_per_group);
3453                 if (sbi->s_clusters_per_group > blocksize * 8) {
3454                         ext4_msg(sb, KERN_ERR,
3455                                  "#clusters per group too big: %lu",
3456                                  sbi->s_clusters_per_group);
3457                         goto failed_mount;
3458                 }
3459                 if (sbi->s_blocks_per_group !=
3460                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3461                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3462                                  "clusters per group (%lu) inconsistent",
3463                                  sbi->s_blocks_per_group,
3464                                  sbi->s_clusters_per_group);
3465                         goto failed_mount;
3466                 }
3467         } else {
3468                 if (clustersize != blocksize) {
3469                         ext4_warning(sb, "fragment/cluster size (%d) != "
3470                                      "block size (%d)", clustersize,
3471                                      blocksize);
3472                         clustersize = blocksize;
3473                 }
3474                 if (sbi->s_blocks_per_group > blocksize * 8) {
3475                         ext4_msg(sb, KERN_ERR,
3476                                  "#blocks per group too big: %lu",
3477                                  sbi->s_blocks_per_group);
3478                         goto failed_mount;
3479                 }
3480                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3481                 sbi->s_cluster_bits = 0;
3482         }
3483         sbi->s_cluster_ratio = clustersize / blocksize;
3484
3485         if (sbi->s_inodes_per_group > blocksize * 8) {
3486                 ext4_msg(sb, KERN_ERR,
3487                        "#inodes per group too big: %lu",
3488                        sbi->s_inodes_per_group);
3489                 goto failed_mount;
3490         }
3491
3492         /*
3493          * Test whether we have more sectors than will fit in sector_t,
3494          * and whether the max offset is addressable by the page cache.
3495          */
3496         err = generic_check_addressable(sb->s_blocksize_bits,
3497                                         ext4_blocks_count(es));
3498         if (err) {
3499                 ext4_msg(sb, KERN_ERR, "filesystem"
3500                          " too large to mount safely on this system");
3501                 if (sizeof(sector_t) < 8)
3502                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3503                 ret = err;
3504                 goto failed_mount;
3505         }
3506
3507         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3508                 goto cantfind_ext4;
3509
3510         /* check blocks count against device size */
3511         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3512         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3513                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3514                        "exceeds size of device (%llu blocks)",
3515                        ext4_blocks_count(es), blocks_count);
3516                 goto failed_mount;
3517         }
3518
3519         /*
3520          * It makes no sense for the first data block to be beyond the end
3521          * of the filesystem.
3522          */
3523         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3524                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3525                          "block %u is beyond end of filesystem (%llu)",
3526                          le32_to_cpu(es->s_first_data_block),
3527                          ext4_blocks_count(es));
3528                 goto failed_mount;
3529         }
3530         blocks_count = (ext4_blocks_count(es) -
3531                         le32_to_cpu(es->s_first_data_block) +
3532                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3533         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3534         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3535                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3536                        "(block count %llu, first data block %u, "
3537                        "blocks per group %lu)", sbi->s_groups_count,
3538                        ext4_blocks_count(es),
3539                        le32_to_cpu(es->s_first_data_block),
3540                        EXT4_BLOCKS_PER_GROUP(sb));
3541                 goto failed_mount;
3542         }
3543         sbi->s_groups_count = blocks_count;
3544         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3545                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3546         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3547                    EXT4_DESC_PER_BLOCK(sb);
3548         sbi->s_group_desc = ext4_kvmalloc(db_count *
3549                                           sizeof(struct buffer_head *),
3550                                           GFP_KERNEL);
3551         if (sbi->s_group_desc == NULL) {
3552                 ext4_msg(sb, KERN_ERR, "not enough memory");
3553                 goto failed_mount;
3554         }
3555
3556         if (ext4_proc_root)
3557                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3558
3559         bgl_lock_init(sbi->s_blockgroup_lock);
3560
3561         for (i = 0; i < db_count; i++) {
3562                 block = descriptor_loc(sb, logical_sb_block, i);
3563                 sbi->s_group_desc[i] = sb_bread(sb, block);
3564                 if (!sbi->s_group_desc[i]) {
3565                         ext4_msg(sb, KERN_ERR,
3566                                "can't read group descriptor %d", i);
3567                         db_count = i;
3568                         goto failed_mount2;
3569                 }
3570         }
3571         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3572                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3573                 goto failed_mount2;
3574         }
3575         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3576                 if (!ext4_fill_flex_info(sb)) {
3577                         ext4_msg(sb, KERN_ERR,
3578                                "unable to initialize "
3579                                "flex_bg meta info!");
3580                         goto failed_mount2;
3581                 }
3582
3583         sbi->s_gdb_count = db_count;
3584         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3585         spin_lock_init(&sbi->s_next_gen_lock);
3586
3587         init_timer(&sbi->s_err_report);
3588         sbi->s_err_report.function = print_daily_error_info;
3589         sbi->s_err_report.data = (unsigned long) sb;
3590
3591         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3592                         ext4_count_free_clusters(sb));
3593         if (!err) {
3594                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3595                                 ext4_count_free_inodes(sb));
3596         }
3597         if (!err) {
3598                 err = percpu_counter_init(&sbi->s_dirs_counter,
3599                                 ext4_count_dirs(sb));
3600         }
3601         if (!err) {
3602                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3603         }
3604         if (err) {
3605                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3606                 goto failed_mount3;
3607         }
3608
3609         sbi->s_stripe = ext4_get_stripe_size(sbi);
3610         sbi->s_max_writeback_mb_bump = 128;
3611
3612         /*
3613          * set up enough so that it can read an inode
3614          */
3615         if (!test_opt(sb, NOLOAD) &&
3616             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3617                 sb->s_op = &ext4_sops;
3618         else
3619                 sb->s_op = &ext4_nojournal_sops;
3620         sb->s_export_op = &ext4_export_ops;
3621         sb->s_xattr = ext4_xattr_handlers;
3622 #ifdef CONFIG_QUOTA
3623         sb->s_qcop = &ext4_qctl_operations;
3624         sb->dq_op = &ext4_quota_operations;
3625 #endif
3626         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3627
3628         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3629         mutex_init(&sbi->s_orphan_lock);
3630         sbi->s_resize_flags = 0;
3631
3632         sb->s_root = NULL;
3633
3634         needs_recovery = (es->s_last_orphan != 0 ||
3635                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3636                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3637
3638         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3639             !(sb->s_flags & MS_RDONLY))
3640                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3641                         goto failed_mount3;
3642
3643         /*
3644          * The first inode we look at is the journal inode.  Don't try
3645          * root first: it may be modified in the journal!
3646          */
3647         if (!test_opt(sb, NOLOAD) &&
3648             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3649                 if (ext4_load_journal(sb, es, journal_devnum))
3650                         goto failed_mount3;
3651         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3652               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3653                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3654                        "suppressed and not mounted read-only");
3655                 goto failed_mount_wq;
3656         } else {
3657                 clear_opt(sb, DATA_FLAGS);
3658                 sbi->s_journal = NULL;
3659                 needs_recovery = 0;
3660                 goto no_journal;
3661         }
3662
3663         if (ext4_blocks_count(es) > 0xffffffffULL &&
3664             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3665                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3666                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3667                 goto failed_mount_wq;
3668         }
3669
3670         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3671                 jbd2_journal_set_features(sbi->s_journal,
3672                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3673                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3674         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3675                 jbd2_journal_set_features(sbi->s_journal,
3676                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3677                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3678                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3679         } else {
3680                 jbd2_journal_clear_features(sbi->s_journal,
3681                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3682                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3683         }
3684
3685         /* We have now updated the journal if required, so we can
3686          * validate the data journaling mode. */
3687         switch (test_opt(sb, DATA_FLAGS)) {
3688         case 0:
3689                 /* No mode set, assume a default based on the journal
3690                  * capabilities: ORDERED_DATA if the journal can
3691                  * cope, else JOURNAL_DATA
3692                  */
3693                 if (jbd2_journal_check_available_features
3694                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3695                         set_opt(sb, ORDERED_DATA);
3696                 else
3697                         set_opt(sb, JOURNAL_DATA);
3698                 break;
3699
3700         case EXT4_MOUNT_ORDERED_DATA:
3701         case EXT4_MOUNT_WRITEBACK_DATA:
3702                 if (!jbd2_journal_check_available_features
3703                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3704                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3705                                "requested data journaling mode");
3706                         goto failed_mount_wq;
3707                 }
3708         default:
3709                 break;
3710         }
3711         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3712
3713         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3714
3715         /*
3716          * The journal may have updated the bg summary counts, so we
3717          * need to update the global counters.
3718          */
3719         percpu_counter_set(&sbi->s_freeclusters_counter,
3720                            ext4_count_free_clusters(sb));
3721         percpu_counter_set(&sbi->s_freeinodes_counter,
3722                            ext4_count_free_inodes(sb));
3723         percpu_counter_set(&sbi->s_dirs_counter,
3724                            ext4_count_dirs(sb));
3725         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3726
3727 no_journal:
3728         /*
3729          * The maximum number of concurrent works can be high and
3730          * concurrency isn't really necessary.  Limit it to 1.
3731          */
3732         EXT4_SB(sb)->dio_unwritten_wq =
3733                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3734         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3735                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3736                 goto failed_mount_wq;
3737         }
3738
3739         /*
3740          * The jbd2_journal_load will have done any necessary log recovery,
3741          * so we can safely mount the rest of the filesystem now.
3742          */
3743
3744         root = ext4_iget(sb, EXT4_ROOT_INO);
3745         if (IS_ERR(root)) {
3746                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3747                 ret = PTR_ERR(root);
3748                 root = NULL;
3749                 goto failed_mount4;
3750         }
3751         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3752                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3753                 iput(root);
3754                 goto failed_mount4;
3755         }
3756         sb->s_root = d_alloc_root(root);
3757         if (!sb->s_root) {
3758                 iput(root);
3759                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3760                 ret = -ENOMEM;
3761                 goto failed_mount4;
3762         }
3763
3764         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3765
3766         /* determine the minimum size of new large inodes, if present */
3767         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3768                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3769                                                      EXT4_GOOD_OLD_INODE_SIZE;
3770                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3771                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3772                         if (sbi->s_want_extra_isize <
3773                             le16_to_cpu(es->s_want_extra_isize))
3774                                 sbi->s_want_extra_isize =
3775                                         le16_to_cpu(es->s_want_extra_isize);
3776                         if (sbi->s_want_extra_isize <
3777                             le16_to_cpu(es->s_min_extra_isize))
3778                                 sbi->s_want_extra_isize =
3779                                         le16_to_cpu(es->s_min_extra_isize);
3780                 }
3781         }
3782         /* Check if enough inode space is available */
3783         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3784                                                         sbi->s_inode_size) {
3785                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3786                                                        EXT4_GOOD_OLD_INODE_SIZE;
3787                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3788                          "available");
3789         }
3790
3791         err = ext4_setup_system_zone(sb);
3792         if (err) {
3793                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3794                          "zone (%d)", err);
3795                 goto failed_mount4a;
3796         }
3797
3798         ext4_ext_init(sb);
3799         err = ext4_mb_init(sb, needs_recovery);
3800         if (err) {
3801                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3802                          err);
3803                 goto failed_mount5;
3804         }
3805
3806         err = ext4_register_li_request(sb, first_not_zeroed);
3807         if (err)
3808                 goto failed_mount6;
3809
3810         sbi->s_kobj.kset = ext4_kset;
3811         init_completion(&sbi->s_kobj_unregister);
3812         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3813                                    "%s", sb->s_id);
3814         if (err)
3815                 goto failed_mount7;
3816
3817         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3818         ext4_orphan_cleanup(sb, es);
3819         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3820         if (needs_recovery) {
3821                 ext4_msg(sb, KERN_INFO, "recovery complete");
3822                 ext4_mark_recovery_complete(sb, es);
3823         }
3824         if (EXT4_SB(sb)->s_journal) {
3825                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3826                         descr = " journalled data mode";
3827                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3828                         descr = " ordered data mode";
3829                 else
3830                         descr = " writeback data mode";
3831         } else
3832                 descr = "out journal";
3833
3834         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3835                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3836                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3837
3838         if (es->s_error_count)
3839                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3840
3841         kfree(orig_data);
3842         return 0;
3843
3844 cantfind_ext4:
3845         if (!silent)
3846                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3847         goto failed_mount;
3848
3849 failed_mount7:
3850         ext4_unregister_li_request(sb);
3851 failed_mount6:
3852         ext4_mb_release(sb);
3853 failed_mount5:
3854         ext4_ext_release(sb);
3855         ext4_release_system_zone(sb);
3856 failed_mount4a:
3857         dput(sb->s_root);
3858         sb->s_root = NULL;
3859 failed_mount4:
3860         ext4_msg(sb, KERN_ERR, "mount failed");
3861         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3862 failed_mount_wq:
3863         if (sbi->s_journal) {
3864                 jbd2_journal_destroy(sbi->s_journal);
3865                 sbi->s_journal = NULL;
3866         }
3867 failed_mount3:
3868         del_timer(&sbi->s_err_report);
3869         if (sbi->s_flex_groups)
3870                 ext4_kvfree(sbi->s_flex_groups);
3871         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3872         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3873         percpu_counter_destroy(&sbi->s_dirs_counter);
3874         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3875         if (sbi->s_mmp_tsk)
3876                 kthread_stop(sbi->s_mmp_tsk);
3877 failed_mount2:
3878         for (i = 0; i < db_count; i++)
3879                 brelse(sbi->s_group_desc[i]);
3880         ext4_kvfree(sbi->s_group_desc);
3881 failed_mount:
3882         if (sbi->s_proc) {
3883                 remove_proc_entry(sb->s_id, ext4_proc_root);
3884         }
3885 #ifdef CONFIG_QUOTA
3886         for (i = 0; i < MAXQUOTAS; i++)
3887                 kfree(sbi->s_qf_names[i]);
3888 #endif
3889         ext4_blkdev_remove(sbi);
3890         brelse(bh);
3891 out_fail:
3892         sb->s_fs_info = NULL;
3893         kfree(sbi->s_blockgroup_lock);
3894         kfree(sbi);
3895 out_free_orig:
3896         kfree(orig_data);
3897         return ret;
3898 }
3899
3900 /*
3901  * Setup any per-fs journal parameters now.  We'll do this both on
3902  * initial mount, once the journal has been initialised but before we've
3903  * done any recovery; and again on any subsequent remount.
3904  */
3905 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3906 {
3907         struct ext4_sb_info *sbi = EXT4_SB(sb);
3908
3909         journal->j_commit_interval = sbi->s_commit_interval;
3910         journal->j_min_batch_time = sbi->s_min_batch_time;
3911         journal->j_max_batch_time = sbi->s_max_batch_time;
3912
3913         write_lock(&journal->j_state_lock);
3914         if (test_opt(sb, BARRIER))
3915                 journal->j_flags |= JBD2_BARRIER;
3916         else
3917                 journal->j_flags &= ~JBD2_BARRIER;
3918         if (test_opt(sb, DATA_ERR_ABORT))
3919                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3920         else
3921                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3922         write_unlock(&journal->j_state_lock);
3923 }
3924
3925 static journal_t *ext4_get_journal(struct super_block *sb,
3926                                    unsigned int journal_inum)
3927 {
3928         struct inode *journal_inode;
3929         journal_t *journal;
3930
3931         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3932
3933         /* First, test for the existence of a valid inode on disk.  Bad
3934          * things happen if we iget() an unused inode, as the subsequent
3935          * iput() will try to delete it. */
3936
3937         journal_inode = ext4_iget(sb, journal_inum);
3938         if (IS_ERR(journal_inode)) {
3939                 ext4_msg(sb, KERN_ERR, "no journal found");
3940                 return NULL;
3941         }
3942         if (!journal_inode->i_nlink) {
3943                 make_bad_inode(journal_inode);
3944                 iput(journal_inode);
3945                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3946                 return NULL;
3947         }
3948
3949         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3950                   journal_inode, journal_inode->i_size);
3951         if (!S_ISREG(journal_inode->i_mode)) {
3952                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3953                 iput(journal_inode);
3954                 return NULL;
3955         }
3956
3957         journal = jbd2_journal_init_inode(journal_inode);
3958         if (!journal) {
3959                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3960                 iput(journal_inode);
3961                 return NULL;
3962         }
3963         journal->j_private = sb;
3964         ext4_init_journal_params(sb, journal);
3965         return journal;
3966 }
3967
3968 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3969                                        dev_t j_dev)
3970 {
3971         struct buffer_head *bh;
3972         journal_t *journal;
3973         ext4_fsblk_t start;
3974         ext4_fsblk_t len;
3975         int hblock, blocksize;
3976         ext4_fsblk_t sb_block;
3977         unsigned long offset;
3978         struct ext4_super_block *es;
3979         struct block_device *bdev;
3980
3981         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3982
3983         bdev = ext4_blkdev_get(j_dev, sb);
3984         if (bdev == NULL)
3985                 return NULL;
3986
3987         blocksize = sb->s_blocksize;
3988         hblock = bdev_logical_block_size(bdev);
3989         if (blocksize < hblock) {
3990                 ext4_msg(sb, KERN_ERR,
3991                         "blocksize too small for journal device");
3992                 goto out_bdev;
3993         }
3994
3995         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3996         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3997         set_blocksize(bdev, blocksize);
3998         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3999                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4000                        "external journal");
4001                 goto out_bdev;
4002         }
4003
4004         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
4005         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4006             !(le32_to_cpu(es->s_feature_incompat) &
4007               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4008                 ext4_msg(sb, KERN_ERR, "external journal has "
4009                                         "bad superblock");
4010                 brelse(bh);
4011                 goto out_bdev;
4012         }
4013
4014         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4015                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4016                 brelse(bh);
4017                 goto out_bdev;
4018         }
4019
4020         len = ext4_blocks_count(es);
4021         start = sb_block + 1;
4022         brelse(bh);     /* we're done with the superblock */
4023
4024         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4025                                         start, len, blocksize);
4026         if (!journal) {
4027                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4028                 goto out_bdev;
4029         }
4030         journal->j_private = sb;
4031         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4032         wait_on_buffer(journal->j_sb_buffer);
4033         if (!buffer_uptodate(journal->j_sb_buffer)) {
4034                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4035                 goto out_journal;
4036         }
4037         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4038                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4039                                         "user (unsupported) - %d",
4040                         be32_to_cpu(journal->j_superblock->s_nr_users));
4041                 goto out_journal;
4042         }
4043         EXT4_SB(sb)->journal_bdev = bdev;
4044         ext4_init_journal_params(sb, journal);
4045         return journal;
4046
4047 out_journal:
4048         jbd2_journal_destroy(journal);
4049 out_bdev:
4050         ext4_blkdev_put(bdev);
4051         return NULL;
4052 }
4053
4054 static int ext4_load_journal(struct super_block *sb,
4055                              struct ext4_super_block *es,
4056                              unsigned long journal_devnum)
4057 {
4058         journal_t *journal;
4059         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4060         dev_t journal_dev;
4061         int err = 0;
4062         int really_read_only;
4063
4064         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4065
4066         if (journal_devnum &&
4067             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4068                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4069                         "numbers have changed");
4070                 journal_dev = new_decode_dev(journal_devnum);
4071         } else
4072                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4073
4074         really_read_only = bdev_read_only(sb->s_bdev);
4075
4076         /*
4077          * Are we loading a blank journal or performing recovery after a
4078          * crash?  For recovery, we need to check in advance whether we
4079          * can get read-write access to the device.
4080          */
4081         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4082                 if (sb->s_flags & MS_RDONLY) {
4083                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4084                                         "required on readonly filesystem");
4085                         if (really_read_only) {
4086                                 ext4_msg(sb, KERN_ERR, "write access "
4087                                         "unavailable, cannot proceed");
4088                                 return -EROFS;
4089                         }
4090                         ext4_msg(sb, KERN_INFO, "write access will "
4091                                "be enabled during recovery");
4092                 }
4093         }
4094
4095         if (journal_inum && journal_dev) {
4096                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4097                        "and inode journals!");
4098                 return -EINVAL;
4099         }
4100
4101         if (journal_inum) {
4102                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4103                         return -EINVAL;
4104         } else {
4105                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4106                         return -EINVAL;
4107         }
4108
4109         if (!(journal->j_flags & JBD2_BARRIER))
4110                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4111
4112         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4113                 err = jbd2_journal_update_format(journal);
4114                 if (err)  {
4115                         ext4_msg(sb, KERN_ERR, "error updating journal");
4116                         jbd2_journal_destroy(journal);
4117                         return err;
4118                 }
4119         }
4120
4121         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4122                 err = jbd2_journal_wipe(journal, !really_read_only);
4123         if (!err) {
4124                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4125                 if (save)
4126                         memcpy(save, ((char *) es) +
4127                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4128                 err = jbd2_journal_load(journal);
4129                 if (save)
4130                         memcpy(((char *) es) + EXT4_S_ERR_START,
4131                                save, EXT4_S_ERR_LEN);
4132                 kfree(save);
4133         }
4134
4135         if (err) {
4136                 ext4_msg(sb, KERN_ERR, "error loading journal");
4137                 jbd2_journal_destroy(journal);
4138                 return err;
4139         }
4140
4141         EXT4_SB(sb)->s_journal = journal;
4142         ext4_clear_journal_err(sb, es);
4143
4144         if (!really_read_only && journal_devnum &&
4145             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4146                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4147
4148                 /* Make sure we flush the recovery flag to disk. */
4149                 ext4_commit_super(sb, 1);
4150         }
4151
4152         return 0;
4153 }
4154
4155 static int ext4_commit_super(struct super_block *sb, int sync)
4156 {
4157         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4158         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4159         int error = 0;
4160
4161         if (!sbh || block_device_ejected(sb))
4162                 return error;
4163         if (buffer_write_io_error(sbh)) {
4164                 /*
4165                  * Oh, dear.  A previous attempt to write the
4166                  * superblock failed.  This could happen because the
4167                  * USB device was yanked out.  Or it could happen to
4168                  * be a transient write error and maybe the block will
4169                  * be remapped.  Nothing we can do but to retry the
4170                  * write and hope for the best.
4171                  */
4172                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4173                        "superblock detected");
4174                 clear_buffer_write_io_error(sbh);
4175                 set_buffer_uptodate(sbh);
4176         }
4177         /*
4178          * If the file system is mounted read-only, don't update the
4179          * superblock write time.  This avoids updating the superblock
4180          * write time when we are mounting the root file system
4181          * read/only but we need to replay the journal; at that point,
4182          * for people who are east of GMT and who make their clock
4183          * tick in localtime for Windows bug-for-bug compatibility,
4184          * the clock is set in the future, and this will cause e2fsck
4185          * to complain and force a full file system check.
4186          */
4187         if (!(sb->s_flags & MS_RDONLY))
4188                 es->s_wtime = cpu_to_le32(get_seconds());
4189         if (sb->s_bdev->bd_part)
4190                 es->s_kbytes_written =
4191                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4192                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4193                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4194         else
4195                 es->s_kbytes_written =
4196                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4197         ext4_free_blocks_count_set(es,
4198                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4199                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4200         es->s_free_inodes_count =
4201                 cpu_to_le32(percpu_counter_sum_positive(
4202                                 &EXT4_SB(sb)->s_freeinodes_counter));
4203         sb->s_dirt = 0;
4204         BUFFER_TRACE(sbh, "marking dirty");
4205         mark_buffer_dirty(sbh);
4206         if (sync) {
4207                 error = sync_dirty_buffer(sbh);
4208                 if (error)
4209                         return error;
4210
4211                 error = buffer_write_io_error(sbh);
4212                 if (error) {
4213                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4214                                "superblock");
4215                         clear_buffer_write_io_error(sbh);
4216                         set_buffer_uptodate(sbh);
4217                 }
4218         }
4219         return error;
4220 }
4221
4222 /*
4223  * Have we just finished recovery?  If so, and if we are mounting (or
4224  * remounting) the filesystem readonly, then we will end up with a
4225  * consistent fs on disk.  Record that fact.
4226  */
4227 static void ext4_mark_recovery_complete(struct super_block *sb,
4228                                         struct ext4_super_block *es)
4229 {
4230         journal_t *journal = EXT4_SB(sb)->s_journal;
4231
4232         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4233                 BUG_ON(journal != NULL);
4234                 return;
4235         }
4236         jbd2_journal_lock_updates(journal);
4237         if (jbd2_journal_flush(journal) < 0)
4238                 goto out;
4239
4240         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4241             sb->s_flags & MS_RDONLY) {
4242                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4243                 ext4_commit_super(sb, 1);
4244         }
4245
4246 out:
4247         jbd2_journal_unlock_updates(journal);
4248 }
4249
4250 /*
4251  * If we are mounting (or read-write remounting) a filesystem whose journal
4252  * has recorded an error from a previous lifetime, move that error to the
4253  * main filesystem now.
4254  */
4255 static void ext4_clear_journal_err(struct super_block *sb,
4256                                    struct ext4_super_block *es)
4257 {
4258         journal_t *journal;
4259         int j_errno;
4260         const char *errstr;
4261
4262         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4263
4264         journal = EXT4_SB(sb)->s_journal;
4265
4266         /*
4267          * Now check for any error status which may have been recorded in the
4268          * journal by a prior ext4_error() or ext4_abort()
4269          */
4270
4271         j_errno = jbd2_journal_errno(journal);
4272         if (j_errno) {
4273                 char nbuf[16];
4274
4275                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4276                 ext4_warning(sb, "Filesystem error recorded "
4277                              "from previous mount: %s", errstr);
4278                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4279
4280                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4281                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4282                 ext4_commit_super(sb, 1);
4283
4284                 jbd2_journal_clear_err(journal);
4285         }
4286 }
4287
4288 /*
4289  * Force the running and committing transactions to commit,
4290  * and wait on the commit.
4291  */
4292 int ext4_force_commit(struct super_block *sb)
4293 {
4294         journal_t *journal;
4295         int ret = 0;
4296
4297         if (sb->s_flags & MS_RDONLY)
4298                 return 0;
4299
4300         journal = EXT4_SB(sb)->s_journal;
4301         if (journal) {
4302                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4303                 ret = ext4_journal_force_commit(journal);
4304         }
4305
4306         return ret;
4307 }
4308
4309 static void ext4_write_super(struct super_block *sb)
4310 {
4311         lock_super(sb);
4312         ext4_commit_super(sb, 1);
4313         unlock_super(sb);
4314 }
4315
4316 static int ext4_sync_fs(struct super_block *sb, int wait)
4317 {
4318         int ret = 0;
4319         tid_t target;
4320         struct ext4_sb_info *sbi = EXT4_SB(sb);
4321
4322         trace_ext4_sync_fs(sb, wait);
4323         flush_workqueue(sbi->dio_unwritten_wq);
4324         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4325                 if (wait)
4326                         jbd2_log_wait_commit(sbi->s_journal, target);
4327         }
4328         return ret;
4329 }
4330
4331 /*
4332  * LVM calls this function before a (read-only) snapshot is created.  This
4333  * gives us a chance to flush the journal completely and mark the fs clean.
4334  *
4335  * Note that only this function cannot bring a filesystem to be in a clean
4336  * state independently, because ext4 prevents a new handle from being started
4337  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4338  * the upper layer.
4339  */
4340 static int ext4_freeze(struct super_block *sb)
4341 {
4342         int error = 0;
4343         journal_t *journal;
4344
4345         if (sb->s_flags & MS_RDONLY)
4346                 return 0;
4347
4348         journal = EXT4_SB(sb)->s_journal;
4349
4350         /* Now we set up the journal barrier. */
4351         jbd2_journal_lock_updates(journal);
4352
4353         /*
4354          * Don't clear the needs_recovery flag if we failed to flush
4355          * the journal.
4356          */
4357         error = jbd2_journal_flush(journal);
4358         if (error < 0)
4359                 goto out;
4360
4361         /* Journal blocked and flushed, clear needs_recovery flag. */
4362         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4363         error = ext4_commit_super(sb, 1);
4364 out:
4365         /* we rely on s_frozen to stop further updates */
4366         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4367         return error;
4368 }
4369
4370 /*
4371  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4372  * flag here, even though the filesystem is not technically dirty yet.
4373  */
4374 static int ext4_unfreeze(struct super_block *sb)
4375 {
4376         if (sb->s_flags & MS_RDONLY)
4377                 return 0;
4378
4379         lock_super(sb);
4380         /* Reset the needs_recovery flag before the fs is unlocked. */
4381         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4382         ext4_commit_super(sb, 1);
4383         unlock_super(sb);
4384         return 0;
4385 }
4386
4387 /*
4388  * Structure to save mount options for ext4_remount's benefit
4389  */
4390 struct ext4_mount_options {
4391         unsigned long s_mount_opt;
4392         unsigned long s_mount_opt2;
4393         uid_t s_resuid;
4394         gid_t s_resgid;
4395         unsigned long s_commit_interval;
4396         u32 s_min_batch_time, s_max_batch_time;
4397 #ifdef CONFIG_QUOTA
4398         int s_jquota_fmt;
4399         char *s_qf_names[MAXQUOTAS];
4400 #endif
4401 };
4402
4403 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4404 {
4405         struct ext4_super_block *es;
4406         struct ext4_sb_info *sbi = EXT4_SB(sb);
4407         ext4_fsblk_t n_blocks_count = 0;
4408         unsigned long old_sb_flags;
4409         struct ext4_mount_options old_opts;
4410         int enable_quota = 0;
4411         ext4_group_t g;
4412         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4413         int err = 0;
4414 #ifdef CONFIG_QUOTA
4415         int i;
4416 #endif
4417         char *orig_data = kstrdup(data, GFP_KERNEL);
4418
4419         /* Store the original options */
4420         lock_super(sb);
4421         old_sb_flags = sb->s_flags;
4422         old_opts.s_mount_opt = sbi->s_mount_opt;
4423         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4424         old_opts.s_resuid = sbi->s_resuid;
4425         old_opts.s_resgid = sbi->s_resgid;
4426         old_opts.s_commit_interval = sbi->s_commit_interval;
4427         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4428         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4429 #ifdef CONFIG_QUOTA
4430         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4431         for (i = 0; i < MAXQUOTAS; i++)
4432                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4433 #endif
4434         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4435                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4436
4437         /*
4438          * Allow the "check" option to be passed as a remount option.
4439          */
4440         if (!parse_options(data, sb, NULL, &journal_ioprio,
4441                            &n_blocks_count, 1)) {
4442                 err = -EINVAL;
4443                 goto restore_opts;
4444         }
4445
4446         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4447                 ext4_abort(sb, "Abort forced by user");
4448
4449         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4450                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4451
4452         es = sbi->s_es;
4453
4454         if (sbi->s_journal) {
4455                 ext4_init_journal_params(sb, sbi->s_journal);
4456                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4457         }
4458
4459         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4460                 n_blocks_count > ext4_blocks_count(es)) {
4461                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4462                         err = -EROFS;
4463                         goto restore_opts;
4464                 }
4465
4466                 if (*flags & MS_RDONLY) {
4467                         err = dquot_suspend(sb, -1);
4468                         if (err < 0)
4469                                 goto restore_opts;
4470
4471                         /*
4472                          * First of all, the unconditional stuff we have to do
4473                          * to disable replay of the journal when we next remount
4474                          */
4475                         sb->s_flags |= MS_RDONLY;
4476
4477                         /*
4478                          * OK, test if we are remounting a valid rw partition
4479                          * readonly, and if so set the rdonly flag and then
4480                          * mark the partition as valid again.
4481                          */
4482                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4483                             (sbi->s_mount_state & EXT4_VALID_FS))
4484                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4485
4486                         if (sbi->s_journal)
4487                                 ext4_mark_recovery_complete(sb, es);
4488                 } else {
4489                         /* Make sure we can mount this feature set readwrite */
4490                         if (!ext4_feature_set_ok(sb, 0)) {
4491                                 err = -EROFS;
4492                                 goto restore_opts;
4493                         }
4494                         /*
4495                          * Make sure the group descriptor checksums
4496                          * are sane.  If they aren't, refuse to remount r/w.
4497                          */
4498                         for (g = 0; g < sbi->s_groups_count; g++) {
4499                                 struct ext4_group_desc *gdp =
4500                                         ext4_get_group_desc(sb, g, NULL);
4501
4502                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4503                                         ext4_msg(sb, KERN_ERR,
4504                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4505                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4506                                                le16_to_cpu(gdp->bg_checksum));
4507                                         err = -EINVAL;
4508                                         goto restore_opts;
4509                                 }
4510                         }
4511
4512                         /*
4513                          * If we have an unprocessed orphan list hanging
4514                          * around from a previously readonly bdev mount,
4515                          * require a full umount/remount for now.
4516                          */
4517                         if (es->s_last_orphan) {
4518                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4519                                        "remount RDWR because of unprocessed "
4520                                        "orphan inode list.  Please "
4521                                        "umount/remount instead");
4522                                 err = -EINVAL;
4523                                 goto restore_opts;
4524                         }
4525
4526                         /*
4527                          * Mounting a RDONLY partition read-write, so reread
4528                          * and store the current valid flag.  (It may have
4529                          * been changed by e2fsck since we originally mounted
4530                          * the partition.)
4531                          */
4532                         if (sbi->s_journal)
4533                                 ext4_clear_journal_err(sb, es);
4534                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4535                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4536                                 goto restore_opts;
4537                         if (!ext4_setup_super(sb, es, 0))
4538                                 sb->s_flags &= ~MS_RDONLY;
4539                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4540                                                      EXT4_FEATURE_INCOMPAT_MMP))
4541                                 if (ext4_multi_mount_protect(sb,
4542                                                 le64_to_cpu(es->s_mmp_block))) {
4543                                         err = -EROFS;
4544                                         goto restore_opts;
4545                                 }
4546                         enable_quota = 1;
4547                 }
4548         }
4549
4550         /*
4551          * Reinitialize lazy itable initialization thread based on
4552          * current settings
4553          */
4554         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4555                 ext4_unregister_li_request(sb);
4556         else {
4557                 ext4_group_t first_not_zeroed;
4558                 first_not_zeroed = ext4_has_uninit_itable(sb);
4559                 ext4_register_li_request(sb, first_not_zeroed);
4560         }
4561
4562         ext4_setup_system_zone(sb);
4563         if (sbi->s_journal == NULL)
4564                 ext4_commit_super(sb, 1);
4565
4566 #ifdef CONFIG_QUOTA
4567         /* Release old quota file names */
4568         for (i = 0; i < MAXQUOTAS; i++)
4569                 if (old_opts.s_qf_names[i] &&
4570                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4571                         kfree(old_opts.s_qf_names[i]);
4572 #endif
4573         unlock_super(sb);
4574         if (enable_quota)
4575                 dquot_resume(sb, -1);
4576
4577         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4578         kfree(orig_data);
4579         return 0;
4580
4581 restore_opts:
4582         sb->s_flags = old_sb_flags;
4583         sbi->s_mount_opt = old_opts.s_mount_opt;
4584         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4585         sbi->s_resuid = old_opts.s_resuid;
4586         sbi->s_resgid = old_opts.s_resgid;
4587         sbi->s_commit_interval = old_opts.s_commit_interval;
4588         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4589         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4590 #ifdef CONFIG_QUOTA
4591         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4592         for (i = 0; i < MAXQUOTAS; i++) {
4593                 if (sbi->s_qf_names[i] &&
4594                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4595                         kfree(sbi->s_qf_names[i]);
4596                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4597         }
4598 #endif
4599         unlock_super(sb);
4600         kfree(orig_data);
4601         return err;
4602 }
4603
4604 /*
4605  * Note: calculating the overhead so we can be compatible with
4606  * historical BSD practice is quite difficult in the face of
4607  * clusters/bigalloc.  This is because multiple metadata blocks from
4608  * different block group can end up in the same allocation cluster.
4609  * Calculating the exact overhead in the face of clustered allocation
4610  * requires either O(all block bitmaps) in memory or O(number of block
4611  * groups**2) in time.  We will still calculate the superblock for
4612  * older file systems --- and if we come across with a bigalloc file
4613  * system with zero in s_overhead_clusters the estimate will be close to
4614  * correct especially for very large cluster sizes --- but for newer
4615  * file systems, it's better to calculate this figure once at mkfs
4616  * time, and store it in the superblock.  If the superblock value is
4617  * present (even for non-bigalloc file systems), we will use it.
4618  */
4619 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4620 {
4621         struct super_block *sb = dentry->d_sb;
4622         struct ext4_sb_info *sbi = EXT4_SB(sb);
4623         struct ext4_super_block *es = sbi->s_es;
4624         struct ext4_group_desc *gdp;
4625         u64 fsid;
4626         s64 bfree;
4627
4628         if (test_opt(sb, MINIX_DF)) {
4629                 sbi->s_overhead_last = 0;
4630         } else if (es->s_overhead_clusters) {
4631                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4632         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4633                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4634                 ext4_fsblk_t overhead = 0;
4635
4636                 /*
4637                  * Compute the overhead (FS structures).  This is constant
4638                  * for a given filesystem unless the number of block groups
4639                  * changes so we cache the previous value until it does.
4640                  */
4641
4642                 /*
4643                  * All of the blocks before first_data_block are
4644                  * overhead
4645                  */
4646                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4647
4648                 /*
4649                  * Add the overhead found in each block group
4650                  */
4651                 for (i = 0; i < ngroups; i++) {
4652                         gdp = ext4_get_group_desc(sb, i, NULL);
4653                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4654                         cond_resched();
4655                 }
4656                 sbi->s_overhead_last = overhead;
4657                 smp_wmb();
4658                 sbi->s_blocks_last = ext4_blocks_count(es);
4659         }
4660
4661         buf->f_type = EXT4_SUPER_MAGIC;
4662         buf->f_bsize = sb->s_blocksize;
4663         buf->f_blocks = (ext4_blocks_count(es) -
4664                          EXT4_C2B(sbi, sbi->s_overhead_last));
4665         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4666                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4667         /* prevent underflow in case that few free space is available */
4668         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4669         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4670         if (buf->f_bfree < ext4_r_blocks_count(es))
4671                 buf->f_bavail = 0;
4672         buf->f_files = le32_to_cpu(es->s_inodes_count);
4673         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4674         buf->f_namelen = EXT4_NAME_LEN;
4675         fsid = le64_to_cpup((void *)es->s_uuid) ^
4676                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4677         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4678         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4679
4680         return 0;
4681 }
4682
4683 /* Helper function for writing quotas on sync - we need to start transaction
4684  * before quota file is locked for write. Otherwise the are possible deadlocks:
4685  * Process 1                         Process 2
4686  * ext4_create()                     quota_sync()
4687  *   jbd2_journal_start()                  write_dquot()
4688  *   dquot_initialize()                         down(dqio_mutex)
4689  *     down(dqio_mutex)                    jbd2_journal_start()
4690  *
4691  */
4692
4693 #ifdef CONFIG_QUOTA
4694
4695 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4696 {
4697         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4698 }
4699
4700 static int ext4_write_dquot(struct dquot *dquot)
4701 {
4702         int ret, err;
4703         handle_t *handle;
4704         struct inode *inode;
4705
4706         inode = dquot_to_inode(dquot);
4707         handle = ext4_journal_start(inode,
4708                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4709         if (IS_ERR(handle))
4710                 return PTR_ERR(handle);
4711         ret = dquot_commit(dquot);
4712         err = ext4_journal_stop(handle);
4713         if (!ret)
4714                 ret = err;
4715         return ret;
4716 }
4717
4718 static int ext4_acquire_dquot(struct dquot *dquot)
4719 {
4720         int ret, err;
4721         handle_t *handle;
4722
4723         handle = ext4_journal_start(dquot_to_inode(dquot),
4724                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4725         if (IS_ERR(handle))
4726                 return PTR_ERR(handle);
4727         ret = dquot_acquire(dquot);
4728         err = ext4_journal_stop(handle);
4729         if (!ret)
4730                 ret = err;
4731         return ret;
4732 }
4733
4734 static int ext4_release_dquot(struct dquot *dquot)
4735 {
4736         int ret, err;
4737         handle_t *handle;
4738
4739         handle = ext4_journal_start(dquot_to_inode(dquot),
4740                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4741         if (IS_ERR(handle)) {
4742                 /* Release dquot anyway to avoid endless cycle in dqput() */
4743                 dquot_release(dquot);
4744                 return PTR_ERR(handle);
4745         }
4746         ret = dquot_release(dquot);
4747         err = ext4_journal_stop(handle);
4748         if (!ret)
4749                 ret = err;
4750         return ret;
4751 }
4752
4753 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4754 {
4755         /* Are we journaling quotas? */
4756         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4757             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4758                 dquot_mark_dquot_dirty(dquot);
4759                 return ext4_write_dquot(dquot);
4760         } else {
4761                 return dquot_mark_dquot_dirty(dquot);
4762         }
4763 }
4764
4765 static int ext4_write_info(struct super_block *sb, int type)
4766 {
4767         int ret, err;
4768         handle_t *handle;
4769
4770         /* Data block + inode block */
4771         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4772         if (IS_ERR(handle))
4773                 return PTR_ERR(handle);
4774         ret = dquot_commit_info(sb, type);
4775         err = ext4_journal_stop(handle);
4776         if (!ret)
4777                 ret = err;
4778         return ret;
4779 }
4780
4781 /*
4782  * Turn on quotas during mount time - we need to find
4783  * the quota file and such...
4784  */
4785 static int ext4_quota_on_mount(struct super_block *sb, int type)
4786 {
4787         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4788                                         EXT4_SB(sb)->s_jquota_fmt, type);
4789 }
4790
4791 /*
4792  * Standard function to be called on quota_on
4793  */
4794 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4795                          struct path *path)
4796 {
4797         int err;
4798
4799         if (!test_opt(sb, QUOTA))
4800                 return -EINVAL;
4801
4802         /* Quotafile not on the same filesystem? */
4803         if (path->dentry->d_sb != sb)
4804                 return -EXDEV;
4805         /* Journaling quota? */
4806         if (EXT4_SB(sb)->s_qf_names[type]) {
4807                 /* Quotafile not in fs root? */
4808                 if (path->dentry->d_parent != sb->s_root)
4809                         ext4_msg(sb, KERN_WARNING,
4810                                 "Quota file not on filesystem root. "
4811                                 "Journaled quota will not work");
4812         }
4813
4814         /*
4815          * When we journal data on quota file, we have to flush journal to see
4816          * all updates to the file when we bypass pagecache...
4817          */
4818         if (EXT4_SB(sb)->s_journal &&
4819             ext4_should_journal_data(path->dentry->d_inode)) {
4820                 /*
4821                  * We don't need to lock updates but journal_flush() could
4822                  * otherwise be livelocked...
4823                  */
4824                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4825                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4826                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4827                 if (err)
4828                         return err;
4829         }
4830
4831         return dquot_quota_on(sb, type, format_id, path);
4832 }
4833
4834 static int ext4_quota_off(struct super_block *sb, int type)
4835 {
4836         struct inode *inode = sb_dqopt(sb)->files[type];
4837         handle_t *handle;
4838
4839         /* Force all delayed allocation blocks to be allocated.
4840          * Caller already holds s_umount sem */
4841         if (test_opt(sb, DELALLOC))
4842                 sync_filesystem(sb);
4843
4844         if (!inode)
4845                 goto out;
4846
4847         /* Update modification times of quota files when userspace can
4848          * start looking at them */
4849         handle = ext4_journal_start(inode, 1);
4850         if (IS_ERR(handle))
4851                 goto out;
4852         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4853         ext4_mark_inode_dirty(handle, inode);
4854         ext4_journal_stop(handle);
4855
4856 out:
4857         return dquot_quota_off(sb, type);
4858 }
4859
4860 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4861  * acquiring the locks... As quota files are never truncated and quota code
4862  * itself serializes the operations (and no one else should touch the files)
4863  * we don't have to be afraid of races */
4864 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4865                                size_t len, loff_t off)
4866 {
4867         struct inode *inode = sb_dqopt(sb)->files[type];
4868         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4869         int err = 0;
4870         int offset = off & (sb->s_blocksize - 1);
4871         int tocopy;
4872         size_t toread;
4873         struct buffer_head *bh;
4874         loff_t i_size = i_size_read(inode);
4875
4876         if (off > i_size)
4877                 return 0;
4878         if (off+len > i_size)
4879                 len = i_size-off;
4880         toread = len;
4881         while (toread > 0) {
4882                 tocopy = sb->s_blocksize - offset < toread ?
4883                                 sb->s_blocksize - offset : toread;
4884                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4885                 if (err)
4886                         return err;
4887                 if (!bh)        /* A hole? */
4888                         memset(data, 0, tocopy);
4889                 else
4890                         memcpy(data, bh->b_data+offset, tocopy);
4891                 brelse(bh);
4892                 offset = 0;
4893                 toread -= tocopy;
4894                 data += tocopy;
4895                 blk++;
4896         }
4897         return len;
4898 }
4899
4900 /* Write to quotafile (we know the transaction is already started and has
4901  * enough credits) */
4902 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4903                                 const char *data, size_t len, loff_t off)
4904 {
4905         struct inode *inode = sb_dqopt(sb)->files[type];
4906         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4907         int err = 0;
4908         int offset = off & (sb->s_blocksize - 1);
4909         struct buffer_head *bh;
4910         handle_t *handle = journal_current_handle();
4911
4912         if (EXT4_SB(sb)->s_journal && !handle) {
4913                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4914                         " cancelled because transaction is not started",
4915                         (unsigned long long)off, (unsigned long long)len);
4916                 return -EIO;
4917         }
4918         /*
4919          * Since we account only one data block in transaction credits,
4920          * then it is impossible to cross a block boundary.
4921          */
4922         if (sb->s_blocksize - offset < len) {
4923                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4924                         " cancelled because not block aligned",
4925                         (unsigned long long)off, (unsigned long long)len);
4926                 return -EIO;
4927         }
4928
4929         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4930         bh = ext4_bread(handle, inode, blk, 1, &err);
4931         if (!bh)
4932                 goto out;
4933         err = ext4_journal_get_write_access(handle, bh);
4934         if (err) {
4935                 brelse(bh);
4936                 goto out;
4937         }
4938         lock_buffer(bh);
4939         memcpy(bh->b_data+offset, data, len);
4940         flush_dcache_page(bh->b_page);
4941         unlock_buffer(bh);
4942         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4943         brelse(bh);
4944 out:
4945         if (err) {
4946                 mutex_unlock(&inode->i_mutex);
4947                 return err;
4948         }
4949         if (inode->i_size < off + len) {
4950                 i_size_write(inode, off + len);
4951                 EXT4_I(inode)->i_disksize = inode->i_size;
4952                 ext4_mark_inode_dirty(handle, inode);
4953         }
4954         mutex_unlock(&inode->i_mutex);
4955         return len;
4956 }
4957
4958 #endif
4959
4960 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4961                        const char *dev_name, void *data)
4962 {
4963         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4964 }
4965
4966 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4967 static inline void register_as_ext2(void)
4968 {
4969         int err = register_filesystem(&ext2_fs_type);
4970         if (err)
4971                 printk(KERN_WARNING
4972                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4973 }
4974
4975 static inline void unregister_as_ext2(void)
4976 {
4977         unregister_filesystem(&ext2_fs_type);
4978 }
4979
4980 static inline int ext2_feature_set_ok(struct super_block *sb)
4981 {
4982         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4983                 return 0;
4984         if (sb->s_flags & MS_RDONLY)
4985                 return 1;
4986         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4987                 return 0;
4988         return 1;
4989 }
4990 MODULE_ALIAS("ext2");
4991 #else
4992 static inline void register_as_ext2(void) { }
4993 static inline void unregister_as_ext2(void) { }
4994 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4995 #endif
4996
4997 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4998 static inline void register_as_ext3(void)
4999 {
5000         int err = register_filesystem(&ext3_fs_type);
5001         if (err)
5002                 printk(KERN_WARNING
5003                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5004 }
5005
5006 static inline void unregister_as_ext3(void)
5007 {
5008         unregister_filesystem(&ext3_fs_type);
5009 }
5010
5011 static inline int ext3_feature_set_ok(struct super_block *sb)
5012 {
5013         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5014                 return 0;
5015         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5016                 return 0;
5017         if (sb->s_flags & MS_RDONLY)
5018                 return 1;
5019         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5020                 return 0;
5021         return 1;
5022 }
5023 MODULE_ALIAS("ext3");
5024 #else
5025 static inline void register_as_ext3(void) { }
5026 static inline void unregister_as_ext3(void) { }
5027 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5028 #endif
5029
5030 static struct file_system_type ext4_fs_type = {
5031         .owner          = THIS_MODULE,
5032         .name           = "ext4",
5033         .mount          = ext4_mount,
5034         .kill_sb        = kill_block_super,
5035         .fs_flags       = FS_REQUIRES_DEV,
5036 };
5037
5038 static int __init ext4_init_feat_adverts(void)
5039 {
5040         struct ext4_features *ef;
5041         int ret = -ENOMEM;
5042
5043         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5044         if (!ef)
5045                 goto out;
5046
5047         ef->f_kobj.kset = ext4_kset;
5048         init_completion(&ef->f_kobj_unregister);
5049         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5050                                    "features");
5051         if (ret) {
5052                 kfree(ef);
5053                 goto out;
5054         }
5055
5056         ext4_feat = ef;
5057         ret = 0;
5058 out:
5059         return ret;
5060 }
5061
5062 static void ext4_exit_feat_adverts(void)
5063 {
5064         kobject_put(&ext4_feat->f_kobj);
5065         wait_for_completion(&ext4_feat->f_kobj_unregister);
5066         kfree(ext4_feat);
5067 }
5068
5069 /* Shared across all ext4 file systems */
5070 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5071 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5072
5073 static int __init ext4_init_fs(void)
5074 {
5075         int i, err;
5076
5077         ext4_check_flag_values();
5078
5079         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5080                 mutex_init(&ext4__aio_mutex[i]);
5081                 init_waitqueue_head(&ext4__ioend_wq[i]);
5082         }
5083
5084         err = ext4_init_pageio();
5085         if (err)
5086                 return err;
5087         err = ext4_init_system_zone();
5088         if (err)
5089                 goto out6;
5090         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5091         if (!ext4_kset)
5092                 goto out5;
5093         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5094
5095         err = ext4_init_feat_adverts();
5096         if (err)
5097                 goto out4;
5098
5099         err = ext4_init_mballoc();
5100         if (err)
5101                 goto out3;
5102
5103         err = ext4_init_xattr();
5104         if (err)
5105                 goto out2;
5106         err = init_inodecache();
5107         if (err)
5108                 goto out1;
5109         register_as_ext3();
5110         register_as_ext2();
5111         err = register_filesystem(&ext4_fs_type);
5112         if (err)
5113                 goto out;
5114
5115         ext4_li_info = NULL;
5116         mutex_init(&ext4_li_mtx);
5117         return 0;
5118 out:
5119         unregister_as_ext2();
5120         unregister_as_ext3();
5121         destroy_inodecache();
5122 out1:
5123         ext4_exit_xattr();
5124 out2:
5125         ext4_exit_mballoc();
5126 out3:
5127         ext4_exit_feat_adverts();
5128 out4:
5129         if (ext4_proc_root)
5130                 remove_proc_entry("fs/ext4", NULL);
5131         kset_unregister(ext4_kset);
5132 out5:
5133         ext4_exit_system_zone();
5134 out6:
5135         ext4_exit_pageio();
5136         return err;
5137 }
5138
5139 static void __exit ext4_exit_fs(void)
5140 {
5141         ext4_destroy_lazyinit_thread();
5142         unregister_as_ext2();
5143         unregister_as_ext3();
5144         unregister_filesystem(&ext4_fs_type);
5145         destroy_inodecache();
5146         ext4_exit_xattr();
5147         ext4_exit_mballoc();
5148         ext4_exit_feat_adverts();
5149         remove_proc_entry("fs/ext4", NULL);
5150         kset_unregister(ext4_kset);
5151         ext4_exit_system_zone();
5152         ext4_exit_pageio();
5153 }
5154
5155 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5156 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5157 MODULE_LICENSE("GPL");
5158 module_init(ext4_init_fs)
5159 module_exit(ext4_exit_fs)