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"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait);
  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 int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  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 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 int ext4_superblock_csum_verify(struct super_block *sb,
 141                                 struct ext4_super_block *es)
 142 {
 143         if (!ext4_has_metadata_csum(sb))
 144                 return 1;
 145
 146         return es->s_checksum == ext4_superblock_csum(sb, es);
 147 }
 148
 149 void ext4_superblock_csum_set(struct super_block *sb)
 150 {
 151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 152
 153         if (!ext4_has_metadata_csum(sb))
 154                 return;
 155
 156         es->s_checksum = ext4_superblock_csum(sb, es);
 157 }
 158
 159 void *ext4_kvmalloc(size_t size, gfp_t flags)
 160 {
 161         void *ret;
 162
 163         ret = kmalloc(size, flags | __GFP_NOWARN);
 164         if (!ret)
 165                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 166         return ret;
 167 }
 168
 169 void *ext4_kvzalloc(size_t size, gfp_t flags)
 170 {
 171         void *ret;
 172
 173         ret = kzalloc(size, flags | __GFP_NOWARN);
 174         if (!ret)
 175                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 176         return ret;
 177 }
 178
 179 void ext4_kvfree(void *ptr)
 180 {
 181         if (is_vmalloc_addr(ptr))
 182                 vfree(ptr);
 183         else
 184                 kfree(ptr);
 185
 186 }
 187
 188 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 189                                struct ext4_group_desc *bg)
 190 {
 191         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 192                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 193                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 194 }
 195
 196 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 197                                struct ext4_group_desc *bg)
 198 {
 199         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 200                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 201                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 202 }
 203
 204 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 205                               struct ext4_group_desc *bg)
 206 {
 207         return le32_to_cpu(bg->bg_inode_table_lo) |
 208                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 209                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 210 }
 211
 212 __u32 ext4_free_group_clusters(struct super_block *sb,
 213                                struct ext4_group_desc *bg)
 214 {
 215         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 216                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 217                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 218 }
 219
 220 __u32 ext4_free_inodes_count(struct super_block *sb,
 221                               struct ext4_group_desc *bg)
 222 {
 223         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 224                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 225                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 226 }
 227
 228 __u32 ext4_used_dirs_count(struct super_block *sb,
 229                               struct ext4_group_desc *bg)
 230 {
 231         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 232                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 233                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 234 }
 235
 236 __u32 ext4_itable_unused_count(struct super_block *sb,
 237                               struct ext4_group_desc *bg)
 238 {
 239         return le16_to_cpu(bg->bg_itable_unused_lo) |
 240                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 241                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 242 }
 243
 244 void ext4_block_bitmap_set(struct super_block *sb,
 245                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 246 {
 247         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 248         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 249                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 250 }
 251
 252 void ext4_inode_bitmap_set(struct super_block *sb,
 253                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 254 {
 255         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 256         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 257                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 258 }
 259
 260 void ext4_inode_table_set(struct super_block *sb,
 261                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 262 {
 263         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 264         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 265                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 266 }
 267
 268 void ext4_free_group_clusters_set(struct super_block *sb,
 269                                   struct ext4_group_desc *bg, __u32 count)
 270 {
 271         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 272         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 273                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 274 }
 275
 276 void ext4_free_inodes_set(struct super_block *sb,
 277                           struct ext4_group_desc *bg, __u32 count)
 278 {
 279         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 280         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 281                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 282 }
 283
 284 void ext4_used_dirs_set(struct super_block *sb,
 285                           struct ext4_group_desc *bg, __u32 count)
 286 {
 287         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 288         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 289                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 290 }
 291
 292 void ext4_itable_unused_set(struct super_block *sb,
 293                           struct ext4_group_desc *bg, __u32 count)
 294 {
 295         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 296         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 297                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 298 }
 299
 300
 301 static void __save_error_info(struct super_block *sb, const char *func,
 302                             unsigned int line)
 303 {
 304         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 305
 306         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 307         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 308         es->s_last_error_time = cpu_to_le32(get_seconds());
 309         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 310         es->s_last_error_line = cpu_to_le32(line);
 311         if (!es->s_first_error_time) {
 312                 es->s_first_error_time = es->s_last_error_time;
 313                 strncpy(es->s_first_error_func, func,
 314                         sizeof(es->s_first_error_func));
 315                 es->s_first_error_line = cpu_to_le32(line);
 316                 es->s_first_error_ino = es->s_last_error_ino;
 317                 es->s_first_error_block = es->s_last_error_block;
 318         }
 319         /*
 320          * Start the daily error reporting function if it hasn't been
 321          * started already
 322          */
 323         if (!es->s_error_count)
 324                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 325         le32_add_cpu(&es->s_error_count, 1);
 326 }
 327
 328 static void save_error_info(struct super_block *sb, const char *func,
 329                             unsigned int line)
 330 {
 331         __save_error_info(sb, func, line);
 332         ext4_commit_super(sb, 1);
 333 }
 334
 335 /*
 336  * The del_gendisk() function uninitializes the disk-specific data
 337  * structures, including the bdi structure, without telling anyone
 338  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 339  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 340  * This is a kludge to prevent these oops until we can put in a proper
 341  * hook in del_gendisk() to inform the VFS and file system layers.
 342  */
 343 static int block_device_ejected(struct super_block *sb)
 344 {
 345         struct inode *bd_inode = sb->s_bdev->bd_inode;
 346         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 347
 348         return bdi->dev == NULL;
 349 }
 350
 351 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 352 {
 353         struct super_block              *sb = journal->j_private;
 354         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 355         int                             error = is_journal_aborted(journal);
 356         struct ext4_journal_cb_entry    *jce;
 357
 358         BUG_ON(txn->t_state == T_FINISHED);
 359         spin_lock(&sbi->s_md_lock);
 360         while (!list_empty(&txn->t_private_list)) {
 361                 jce = list_entry(txn->t_private_list.next,
 362                                  struct ext4_journal_cb_entry, jce_list);
 363                 list_del_init(&jce->jce_list);
 364                 spin_unlock(&sbi->s_md_lock);
 365                 jce->jce_func(sb, jce, error);
 366                 spin_lock(&sbi->s_md_lock);
 367         }
 368         spin_unlock(&sbi->s_md_lock);
 369 }
 370
 371 /* Deal with the reporting of failure conditions on a filesystem such as
 372  * inconsistencies detected or read IO failures.
 373  *
 374  * On ext2, we can store the error state of the filesystem in the
 375  * superblock.  That is not possible on ext4, because we may have other
 376  * write ordering constraints on the superblock which prevent us from
 377  * writing it out straight away; and given that the journal is about to
 378  * be aborted, we can't rely on the current, or future, transactions to
 379  * write out the superblock safely.
 380  *
 381  * We'll just use the jbd2_journal_abort() error code to record an error in
 382  * the journal instead.  On recovery, the journal will complain about
 383  * that error until we've noted it down and cleared it.
 384  */
 385
 386 static void ext4_handle_error(struct super_block *sb)
 387 {
 388         if (sb->s_flags & MS_RDONLY)
 389                 return;
 390
 391         if (!test_opt(sb, ERRORS_CONT)) {
 392                 journal_t *journal = EXT4_SB(sb)->s_journal;
 393
 394                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 395                 if (journal)
 396                         jbd2_journal_abort(journal, -EIO);
 397         }
 398         if (test_opt(sb, ERRORS_RO)) {
 399                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 400                 /*
 401                  * Make sure updated value of ->s_mount_flags will be visible
 402                  * before ->s_flags update
 403                  */
 404                 smp_wmb();
 405                 sb->s_flags |= MS_RDONLY;
 406         }
 407         if (test_opt(sb, ERRORS_PANIC))
 408                 panic("EXT4-fs (device %s): panic forced after error\n",
 409                         sb->s_id);
 410 }
 411
 412 #define ext4_error_ratelimit(sb)                                        \
 413                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 414                              "EXT4-fs error")
 415
 416 void __ext4_error(struct super_block *sb, const char *function,
 417                   unsigned int line, const char *fmt, ...)
 418 {
 419         struct va_format vaf;
 420         va_list args;
 421
 422         if (ext4_error_ratelimit(sb)) {
 423                 va_start(args, fmt);
 424                 vaf.fmt = fmt;
 425                 vaf.va = &args;
 426                 printk(KERN_CRIT
 427                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 428                        sb->s_id, function, line, current->comm, &vaf);
 429                 va_end(args);
 430         }
 431         save_error_info(sb, function, line);
 432         ext4_handle_error(sb);
 433 }
 434
 435 void __ext4_error_inode(struct inode *inode, const char *function,
 436                         unsigned int line, ext4_fsblk_t block,
 437                         const char *fmt, ...)
 438 {
 439         va_list args;
 440         struct va_format vaf;
 441         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 442
 443         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 444         es->s_last_error_block = cpu_to_le64(block);
 445         if (ext4_error_ratelimit(inode->i_sb)) {
 446                 va_start(args, fmt);
 447                 vaf.fmt = fmt;
 448                 vaf.va = &args;
 449                 if (block)
 450                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 451                                "inode #%lu: block %llu: comm %s: %pV\n",
 452                                inode->i_sb->s_id, function, line, inode->i_ino,
 453                                block, current->comm, &vaf);
 454                 else
 455                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 456                                "inode #%lu: comm %s: %pV\n",
 457                                inode->i_sb->s_id, function, line, inode->i_ino,
 458                                current->comm, &vaf);
 459                 va_end(args);
 460         }
 461         save_error_info(inode->i_sb, function, line);
 462         ext4_handle_error(inode->i_sb);
 463 }
 464
 465 void __ext4_error_file(struct file *file, const char *function,
 466                        unsigned int line, ext4_fsblk_t block,
 467                        const char *fmt, ...)
 468 {
 469         va_list args;
 470         struct va_format vaf;
 471         struct ext4_super_block *es;
 472         struct inode *inode = file_inode(file);
 473         char pathname[80], *path;
 474
 475         es = EXT4_SB(inode->i_sb)->s_es;
 476         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 477         if (ext4_error_ratelimit(inode->i_sb)) {
 478                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 479                 if (IS_ERR(path))
 480                         path = "(unknown)";
 481                 va_start(args, fmt);
 482                 vaf.fmt = fmt;
 483                 vaf.va = &args;
 484                 if (block)
 485                         printk(KERN_CRIT
 486                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 487                                "block %llu: comm %s: path %s: %pV\n",
 488                                inode->i_sb->s_id, function, line, inode->i_ino,
 489                                block, current->comm, path, &vaf);
 490                 else
 491                         printk(KERN_CRIT
 492                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 493                                "comm %s: path %s: %pV\n",
 494                                inode->i_sb->s_id, function, line, inode->i_ino,
 495                                current->comm, path, &vaf);
 496                 va_end(args);
 497         }
 498         save_error_info(inode->i_sb, function, line);
 499         ext4_handle_error(inode->i_sb);
 500 }
 501
 502 const char *ext4_decode_error(struct super_block *sb, int errno,
 503                               char nbuf[16])
 504 {
 505         char *errstr = NULL;
 506
 507         switch (errno) {
 508         case -EIO:
 509                 errstr = "IO failure";
 510                 break;
 511         case -ENOMEM:
 512                 errstr = "Out of memory";
 513                 break;
 514         case -EROFS:
 515                 if (!sb || (EXT4_SB(sb)->s_journal &&
 516                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 517                         errstr = "Journal has aborted";
 518                 else
 519                         errstr = "Readonly filesystem";
 520                 break;
 521         default:
 522                 /* If the caller passed in an extra buffer for unknown
 523                  * errors, textualise them now.  Else we just return
 524                  * NULL. */
 525                 if (nbuf) {
 526                         /* Check for truncated error codes... */
 527                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 528                                 errstr = nbuf;
 529                 }
 530                 break;
 531         }
 532
 533         return errstr;
 534 }
 535
 536 /* __ext4_std_error decodes expected errors from journaling functions
 537  * automatically and invokes the appropriate error response.  */
 538
 539 void __ext4_std_error(struct super_block *sb, const char *function,
 540                       unsigned int line, int errno)
 541 {
 542         char nbuf[16];
 543         const char *errstr;
 544
 545         /* Special case: if the error is EROFS, and we're not already
 546          * inside a transaction, then there's really no point in logging
 547          * an error. */
 548         if (errno == -EROFS && journal_current_handle() == NULL &&
 549             (sb->s_flags & MS_RDONLY))
 550                 return;
 551
 552         if (ext4_error_ratelimit(sb)) {
 553                 errstr = ext4_decode_error(sb, errno, nbuf);
 554                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 555                        sb->s_id, function, line, errstr);
 556         }
 557
 558         save_error_info(sb, function, line);
 559         ext4_handle_error(sb);
 560 }
 561
 562 /*
 563  * ext4_abort is a much stronger failure handler than ext4_error.  The
 564  * abort function may be used to deal with unrecoverable failures such
 565  * as journal IO errors or ENOMEM at a critical moment in log management.
 566  *
 567  * We unconditionally force the filesystem into an ABORT|READONLY state,
 568  * unless the error response on the fs has been set to panic in which
 569  * case we take the easy way out and panic immediately.
 570  */
 571
 572 void __ext4_abort(struct super_block *sb, const char *function,
 573                 unsigned int line, const char *fmt, ...)
 574 {
 575         va_list args;
 576
 577         save_error_info(sb, function, line);
 578         va_start(args, fmt);
 579         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 580                function, line);
 581         vprintk(fmt, args);
 582         printk("\n");
 583         va_end(args);
 584
 585         if ((sb->s_flags & MS_RDONLY) == 0) {
 586                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 587                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 588                 /*
 589                  * Make sure updated value of ->s_mount_flags will be visible
 590                  * before ->s_flags update
 591                  */
 592                 smp_wmb();
 593                 sb->s_flags |= MS_RDONLY;
 594                 if (EXT4_SB(sb)->s_journal)
 595                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 596                 save_error_info(sb, function, line);
 597         }
 598         if (test_opt(sb, ERRORS_PANIC))
 599                 panic("EXT4-fs panic from previous error\n");
 600 }
 601
 602 void __ext4_msg(struct super_block *sb,
 603                 const char *prefix, const char *fmt, ...)
 604 {
 605         struct va_format vaf;
 606         va_list args;
 607
 608         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 609                 return;
 610
 611         va_start(args, fmt);
 612         vaf.fmt = fmt;
 613         vaf.va = &args;
 614         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 615         va_end(args);
 616 }
 617
 618 void __ext4_warning(struct super_block *sb, const char *function,
 619                     unsigned int line, const char *fmt, ...)
 620 {
 621         struct va_format vaf;
 622         va_list args;
 623
 624         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 625                           "EXT4-fs warning"))
 626                 return;
 627
 628         va_start(args, fmt);
 629         vaf.fmt = fmt;
 630         vaf.va = &args;
 631         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 632                sb->s_id, function, line, &vaf);
 633         va_end(args);
 634 }
 635
 636 void __ext4_grp_locked_error(const char *function, unsigned int line,
 637                              struct super_block *sb, ext4_group_t grp,
 638                              unsigned long ino, ext4_fsblk_t block,
 639                              const char *fmt, ...)
 640 __releases(bitlock)
 641 __acquires(bitlock)
 642 {
 643         struct va_format vaf;
 644         va_list args;
 645         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 646
 647         es->s_last_error_ino = cpu_to_le32(ino);
 648         es->s_last_error_block = cpu_to_le64(block);
 649         __save_error_info(sb, function, line);
 650
 651         if (ext4_error_ratelimit(sb)) {
 652                 va_start(args, fmt);
 653                 vaf.fmt = fmt;
 654                 vaf.va = &args;
 655                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 656                        sb->s_id, function, line, grp);
 657                 if (ino)
 658                         printk(KERN_CONT "inode %lu: ", ino);
 659                 if (block)
 660                         printk(KERN_CONT "block %llu:",
 661                                (unsigned long long) block);
 662                 printk(KERN_CONT "%pV\n", &vaf);
 663                 va_end(args);
 664         }
 665
 666         if (test_opt(sb, ERRORS_CONT)) {
 667                 ext4_commit_super(sb, 0);
 668                 return;
 669         }
 670
 671         ext4_unlock_group(sb, grp);
 672         ext4_handle_error(sb);
 673         /*
 674          * We only get here in the ERRORS_RO case; relocking the group
 675          * may be dangerous, but nothing bad will happen since the
 676          * filesystem will have already been marked read/only and the
 677          * journal has been aborted.  We return 1 as a hint to callers
 678          * who might what to use the return value from
 679          * ext4_grp_locked_error() to distinguish between the
 680          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 681          * aggressively from the ext4 function in question, with a
 682          * more appropriate error code.
 683          */
 684         ext4_lock_group(sb, grp);
 685         return;
 686 }
 687
 688 void ext4_update_dynamic_rev(struct super_block *sb)
 689 {
 690         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 691
 692         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 693                 return;
 694
 695         ext4_warning(sb,
 696                      "updating to rev %d because of new feature flag, "
 697                      "running e2fsck is recommended",
 698                      EXT4_DYNAMIC_REV);
 699
 700         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 701         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 702         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 703         /* leave es->s_feature_*compat flags alone */
 704         /* es->s_uuid will be set by e2fsck if empty */
 705
 706         /*
 707          * The rest of the superblock fields should be zero, and if not it
 708          * means they are likely already in use, so leave them alone.  We
 709          * can leave it up to e2fsck to clean up any inconsistencies there.
 710          */
 711 }
 712
 713 /*
 714  * Open the external journal device
 715  */
 716 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 717 {
 718         struct block_device *bdev;
 719         char b[BDEVNAME_SIZE];
 720
 721         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 722         if (IS_ERR(bdev))
 723                 goto fail;
 724         return bdev;
 725
 726 fail:
 727         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 728                         __bdevname(dev, b), PTR_ERR(bdev));
 729         return NULL;
 730 }
 731
 732 /*
 733  * Release the journal device
 734  */
 735 static void ext4_blkdev_put(struct block_device *bdev)
 736 {
 737         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 738 }
 739
 740 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 741 {
 742         struct block_device *bdev;
 743         bdev = sbi->journal_bdev;
 744         if (bdev) {
 745                 ext4_blkdev_put(bdev);
 746                 sbi->journal_bdev = NULL;
 747         }
 748 }
 749
 750 static inline struct inode *orphan_list_entry(struct list_head *l)
 751 {
 752         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 753 }
 754
 755 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 756 {
 757         struct list_head *l;
 758
 759         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 760                  le32_to_cpu(sbi->s_es->s_last_orphan));
 761
 762         printk(KERN_ERR "sb_info orphan list:\n");
 763         list_for_each(l, &sbi->s_orphan) {
 764                 struct inode *inode = orphan_list_entry(l);
 765                 printk(KERN_ERR "  "
 766                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 767                        inode->i_sb->s_id, inode->i_ino, inode,
 768                        inode->i_mode, inode->i_nlink,
 769                        NEXT_ORPHAN(inode));
 770         }
 771 }
 772
 773 static void ext4_put_super(struct super_block *sb)
 774 {
 775         struct ext4_sb_info *sbi = EXT4_SB(sb);
 776         struct ext4_super_block *es = sbi->s_es;
 777         int i, err;
 778
 779         ext4_unregister_li_request(sb);
 780         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 781
 782         flush_workqueue(sbi->rsv_conversion_wq);
 783         destroy_workqueue(sbi->rsv_conversion_wq);
 784
 785         if (sbi->s_journal) {
 786                 err = jbd2_journal_destroy(sbi->s_journal);
 787                 sbi->s_journal = NULL;
 788                 if (err < 0)
 789                         ext4_abort(sb, "Couldn't clean up the journal");
 790         }
 791
 792         ext4_es_unregister_shrinker(sbi);
 793         del_timer_sync(&sbi->s_err_report);
 794         ext4_release_system_zone(sb);
 795         ext4_mb_release(sb);
 796         ext4_ext_release(sb);
 797         ext4_xattr_put_super(sb);
 798
 799         if (!(sb->s_flags & MS_RDONLY)) {
 800                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 801                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 802         }
 803         if (!(sb->s_flags & MS_RDONLY))
 804                 ext4_commit_super(sb, 1);
 805
 806         if (sbi->s_proc) {
 807                 remove_proc_entry("options", sbi->s_proc);
 808                 remove_proc_entry(sb->s_id, ext4_proc_root);
 809         }
 810         kobject_del(&sbi->s_kobj);
 811
 812         for (i = 0; i < sbi->s_gdb_count; i++)
 813                 brelse(sbi->s_group_desc[i]);
 814         ext4_kvfree(sbi->s_group_desc);
 815         ext4_kvfree(sbi->s_flex_groups);
 816         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 817         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 818         percpu_counter_destroy(&sbi->s_dirs_counter);
 819         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 820         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 821         brelse(sbi->s_sbh);
 822 #ifdef CONFIG_QUOTA
 823         for (i = 0; i < MAXQUOTAS; i++)
 824                 kfree(sbi->s_qf_names[i]);
 825 #endif
 826
 827         /* Debugging code just in case the in-memory inode orphan list
 828          * isn't empty.  The on-disk one can be non-empty if we've
 829          * detected an error and taken the fs readonly, but the
 830          * in-memory list had better be clean by this point. */
 831         if (!list_empty(&sbi->s_orphan))
 832                 dump_orphan_list(sb, sbi);
 833         J_ASSERT(list_empty(&sbi->s_orphan));
 834
 835         invalidate_bdev(sb->s_bdev);
 836         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 837                 /*
 838                  * Invalidate the journal device's buffers.  We don't want them
 839                  * floating about in memory - the physical journal device may
 840                  * hotswapped, and it breaks the `ro-after' testing code.
 841                  */
 842                 sync_blockdev(sbi->journal_bdev);
 843                 invalidate_bdev(sbi->journal_bdev);
 844                 ext4_blkdev_remove(sbi);
 845         }
 846         if (sbi->s_mmp_tsk)
 847                 kthread_stop(sbi->s_mmp_tsk);
 848         sb->s_fs_info = NULL;
 849         /*
 850          * Now that we are completely done shutting down the
 851          * superblock, we need to actually destroy the kobject.
 852          */
 853         kobject_put(&sbi->s_kobj);
 854         wait_for_completion(&sbi->s_kobj_unregister);
 855         if (sbi->s_chksum_driver)
 856                 crypto_free_shash(sbi->s_chksum_driver);
 857         kfree(sbi->s_blockgroup_lock);
 858         kfree(sbi);
 859 }
 860
 861 static struct kmem_cache *ext4_inode_cachep;
 862
 863 /*
 864  * Called inside transaction, so use GFP_NOFS
 865  */
 866 static struct inode *ext4_alloc_inode(struct super_block *sb)
 867 {
 868         struct ext4_inode_info *ei;
 869
 870         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 871         if (!ei)
 872                 return NULL;
 873
 874         ei->vfs_inode.i_version = 1;
 875         INIT_LIST_HEAD(&ei->i_prealloc_list);
 876         spin_lock_init(&ei->i_prealloc_lock);
 877         ext4_es_init_tree(&ei->i_es_tree);
 878         rwlock_init(&ei->i_es_lock);
 879         INIT_LIST_HEAD(&ei->i_es_lru);
 880         ei->i_es_lru_nr = 0;
 881         ei->i_touch_when = 0;
 882         ei->i_reserved_data_blocks = 0;
 883         ei->i_reserved_meta_blocks = 0;
 884         ei->i_allocated_meta_blocks = 0;
 885         ei->i_da_metadata_calc_len = 0;
 886         ei->i_da_metadata_calc_last_lblock = 0;
 887         spin_lock_init(&(ei->i_block_reservation_lock));
 888 #ifdef CONFIG_QUOTA
 889         ei->i_reserved_quota = 0;
 890 #endif
 891         ei->jinode = NULL;
 892         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 893         spin_lock_init(&ei->i_completed_io_lock);
 894         ei->i_sync_tid = 0;
 895         ei->i_datasync_tid = 0;
 896         atomic_set(&ei->i_ioend_count, 0);
 897         atomic_set(&ei->i_unwritten, 0);
 898         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 899
 900         return &ei->vfs_inode;
 901 }
 902
 903 static int ext4_drop_inode(struct inode *inode)
 904 {
 905         int drop = generic_drop_inode(inode);
 906
 907         trace_ext4_drop_inode(inode, drop);
 908         return drop;
 909 }
 910
 911 static void ext4_i_callback(struct rcu_head *head)
 912 {
 913         struct inode *inode = container_of(head, struct inode, i_rcu);
 914         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 915 }
 916
 917 static void ext4_destroy_inode(struct inode *inode)
 918 {
 919         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 920                 ext4_msg(inode->i_sb, KERN_ERR,
 921                          "Inode %lu (%p): orphan list check failed!",
 922                          inode->i_ino, EXT4_I(inode));
 923                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 924                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 925                                 true);
 926                 dump_stack();
 927         }
 928         call_rcu(&inode->i_rcu, ext4_i_callback);
 929 }
 930
 931 static void init_once(void *foo)
 932 {
 933         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 934
 935         INIT_LIST_HEAD(&ei->i_orphan);
 936         init_rwsem(&ei->xattr_sem);
 937         init_rwsem(&ei->i_data_sem);
 938         inode_init_once(&ei->vfs_inode);
 939 }
 940
 941 static int init_inodecache(void)
 942 {
 943         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 944                                              sizeof(struct ext4_inode_info),
 945                                              0, (SLAB_RECLAIM_ACCOUNT|
 946                                                 SLAB_MEM_SPREAD),
 947                                              init_once);
 948         if (ext4_inode_cachep == NULL)
 949                 return -ENOMEM;
 950         return 0;
 951 }
 952
 953 static void destroy_inodecache(void)
 954 {
 955         /*
 956          * Make sure all delayed rcu free inodes are flushed before we
 957          * destroy cache.
 958          */
 959         rcu_barrier();
 960         kmem_cache_destroy(ext4_inode_cachep);
 961 }
 962
 963 void ext4_clear_inode(struct inode *inode)
 964 {
 965         invalidate_inode_buffers(inode);
 966         clear_inode(inode);
 967         dquot_drop(inode);
 968         ext4_discard_preallocations(inode);
 969         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 970         ext4_es_lru_del(inode);
 971         if (EXT4_I(inode)->jinode) {
 972                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 973                                                EXT4_I(inode)->jinode);
 974                 jbd2_free_inode(EXT4_I(inode)->jinode);
 975                 EXT4_I(inode)->jinode = NULL;
 976         }
 977 }
 978
 979 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 980                                         u64 ino, u32 generation)
 981 {
 982         struct inode *inode;
 983
 984         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 985                 return ERR_PTR(-ESTALE);
 986         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 987                 return ERR_PTR(-ESTALE);
 988
 989         /* iget isn't really right if the inode is currently unallocated!!
 990          *
 991          * ext4_read_inode will return a bad_inode if the inode had been
 992          * deleted, so we should be safe.
 993          *
 994          * Currently we don't know the generation for parent directory, so
 995          * a generation of 0 means "accept any"
 996          */
 997         inode = ext4_iget_normal(sb, ino);
 998         if (IS_ERR(inode))
 999                 return ERR_CAST(inode);
1000         if (generation && inode->i_generation != generation) {
1001                 iput(inode);
1002                 return ERR_PTR(-ESTALE);
1003         }
1004
1005         return inode;
1006 }
1007
1008 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1009                                         int fh_len, int fh_type)
1010 {
1011         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1012                                     ext4_nfs_get_inode);
1013 }
1014
1015 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1016                                         int fh_len, int fh_type)
1017 {
1018         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1019                                     ext4_nfs_get_inode);
1020 }
1021
1022 /*
1023  * Try to release metadata pages (indirect blocks, directories) which are
1024  * mapped via the block device.  Since these pages could have journal heads
1025  * which would prevent try_to_free_buffers() from freeing them, we must use
1026  * jbd2 layer's try_to_free_buffers() function to release them.
1027  */
1028 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1029                                  gfp_t wait)
1030 {
1031         journal_t *journal = EXT4_SB(sb)->s_journal;
1032
1033         WARN_ON(PageChecked(page));
1034         if (!page_has_buffers(page))
1035                 return 0;
1036         if (journal)
1037                 return jbd2_journal_try_to_free_buffers(journal, page,
1038                                                         wait & ~__GFP_WAIT);
1039         return try_to_free_buffers(page);
1040 }
1041
1042 #ifdef CONFIG_QUOTA
1043 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1044 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1045
1046 static int ext4_write_dquot(struct dquot *dquot);
1047 static int ext4_acquire_dquot(struct dquot *dquot);
1048 static int ext4_release_dquot(struct dquot *dquot);
1049 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1050 static int ext4_write_info(struct super_block *sb, int type);
1051 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1052                          struct path *path);
1053 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1054                                  int format_id);
1055 static int ext4_quota_off(struct super_block *sb, int type);
1056 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1057 static int ext4_quota_on_mount(struct super_block *sb, int type);
1058 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1059                                size_t len, loff_t off);
1060 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1061                                 const char *data, size_t len, loff_t off);
1062 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1063                              unsigned int flags);
1064 static int ext4_enable_quotas(struct super_block *sb);
1065
1066 static const struct dquot_operations ext4_quota_operations = {
1067         .get_reserved_space = ext4_get_reserved_space,
1068         .write_dquot    = ext4_write_dquot,
1069         .acquire_dquot  = ext4_acquire_dquot,
1070         .release_dquot  = ext4_release_dquot,
1071         .mark_dirty     = ext4_mark_dquot_dirty,
1072         .write_info     = ext4_write_info,
1073         .alloc_dquot    = dquot_alloc,
1074         .destroy_dquot  = dquot_destroy,
1075 };
1076
1077 static const struct quotactl_ops ext4_qctl_operations = {
1078         .quota_on       = ext4_quota_on,
1079         .quota_off      = ext4_quota_off,
1080         .quota_sync     = dquot_quota_sync,
1081         .get_info       = dquot_get_dqinfo,
1082         .set_info       = dquot_set_dqinfo,
1083         .get_dqblk      = dquot_get_dqblk,
1084         .set_dqblk      = dquot_set_dqblk
1085 };
1086
1087 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1088         .quota_on_meta  = ext4_quota_on_sysfile,
1089         .quota_off      = ext4_quota_off_sysfile,
1090         .quota_sync     = dquot_quota_sync,
1091         .get_info       = dquot_get_dqinfo,
1092         .set_info       = dquot_set_dqinfo,
1093         .get_dqblk      = dquot_get_dqblk,
1094         .set_dqblk      = dquot_set_dqblk
1095 };
1096 #endif
1097
1098 static const struct super_operations ext4_sops = {
1099         .alloc_inode    = ext4_alloc_inode,
1100         .destroy_inode  = ext4_destroy_inode,
1101         .write_inode    = ext4_write_inode,
1102         .dirty_inode    = ext4_dirty_inode,
1103         .drop_inode     = ext4_drop_inode,
1104         .evict_inode    = ext4_evict_inode,
1105         .put_super      = ext4_put_super,
1106         .sync_fs        = ext4_sync_fs,
1107         .freeze_fs      = ext4_freeze,
1108         .unfreeze_fs    = ext4_unfreeze,
1109         .statfs         = ext4_statfs,
1110         .remount_fs     = ext4_remount,
1111         .show_options   = ext4_show_options,
1112 #ifdef CONFIG_QUOTA
1113         .quota_read     = ext4_quota_read,
1114         .quota_write    = ext4_quota_write,
1115 #endif
1116         .bdev_try_to_free_page = bdev_try_to_free_page,
1117 };
1118
1119 static const struct super_operations ext4_nojournal_sops = {
1120         .alloc_inode    = ext4_alloc_inode,
1121         .destroy_inode  = ext4_destroy_inode,
1122         .write_inode    = ext4_write_inode,
1123         .dirty_inode    = ext4_dirty_inode,
1124         .drop_inode     = ext4_drop_inode,
1125         .evict_inode    = ext4_evict_inode,
1126         .sync_fs        = ext4_sync_fs_nojournal,
1127         .put_super      = ext4_put_super,
1128         .statfs         = ext4_statfs,
1129         .remount_fs     = ext4_remount,
1130         .show_options   = ext4_show_options,
1131 #ifdef CONFIG_QUOTA
1132         .quota_read     = ext4_quota_read,
1133         .quota_write    = ext4_quota_write,
1134 #endif
1135         .bdev_try_to_free_page = bdev_try_to_free_page,
1136 };
1137
1138 static const struct export_operations ext4_export_ops = {
1139         .fh_to_dentry = ext4_fh_to_dentry,
1140         .fh_to_parent = ext4_fh_to_parent,
1141         .get_parent = ext4_get_parent,
1142 };
1143
1144 enum {
1145         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1146         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1147         Opt_nouid32, Opt_debug, Opt_removed,
1148         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1149         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1150         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1151         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1152         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1153         Opt_data_err_abort, Opt_data_err_ignore,
1154         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1155         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1156         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1157         Opt_usrquota, Opt_grpquota, Opt_i_version,
1158         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1159         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1160         Opt_inode_readahead_blks, Opt_journal_ioprio,
1161         Opt_dioread_nolock, Opt_dioread_lock,
1162         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1163         Opt_max_dir_size_kb,
1164 };
1165
1166 static const match_table_t tokens = {
1167         {Opt_bsd_df, "bsddf"},
1168         {Opt_minix_df, "minixdf"},
1169         {Opt_grpid, "grpid"},
1170         {Opt_grpid, "bsdgroups"},
1171         {Opt_nogrpid, "nogrpid"},
1172         {Opt_nogrpid, "sysvgroups"},
1173         {Opt_resgid, "resgid=%u"},
1174         {Opt_resuid, "resuid=%u"},
1175         {Opt_sb, "sb=%u"},
1176         {Opt_err_cont, "errors=continue"},
1177         {Opt_err_panic, "errors=panic"},
1178         {Opt_err_ro, "errors=remount-ro"},
1179         {Opt_nouid32, "nouid32"},
1180         {Opt_debug, "debug"},
1181         {Opt_removed, "oldalloc"},
1182         {Opt_removed, "orlov"},
1183         {Opt_user_xattr, "user_xattr"},
1184         {Opt_nouser_xattr, "nouser_xattr"},
1185         {Opt_acl, "acl"},
1186         {Opt_noacl, "noacl"},
1187         {Opt_noload, "norecovery"},
1188         {Opt_noload, "noload"},
1189         {Opt_removed, "nobh"},
1190         {Opt_removed, "bh"},
1191         {Opt_commit, "commit=%u"},
1192         {Opt_min_batch_time, "min_batch_time=%u"},
1193         {Opt_max_batch_time, "max_batch_time=%u"},
1194         {Opt_journal_dev, "journal_dev=%u"},
1195         {Opt_journal_path, "journal_path=%s"},
1196         {Opt_journal_checksum, "journal_checksum"},
1197         {Opt_journal_async_commit, "journal_async_commit"},
1198         {Opt_abort, "abort"},
1199         {Opt_data_journal, "data=journal"},
1200         {Opt_data_ordered, "data=ordered"},
1201         {Opt_data_writeback, "data=writeback"},
1202         {Opt_data_err_abort, "data_err=abort"},
1203         {Opt_data_err_ignore, "data_err=ignore"},
1204         {Opt_offusrjquota, "usrjquota="},
1205         {Opt_usrjquota, "usrjquota=%s"},
1206         {Opt_offgrpjquota, "grpjquota="},
1207         {Opt_grpjquota, "grpjquota=%s"},
1208         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1209         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1210         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1211         {Opt_grpquota, "grpquota"},
1212         {Opt_noquota, "noquota"},
1213         {Opt_quota, "quota"},
1214         {Opt_usrquota, "usrquota"},
1215         {Opt_barrier, "barrier=%u"},
1216         {Opt_barrier, "barrier"},
1217         {Opt_nobarrier, "nobarrier"},
1218         {Opt_i_version, "i_version"},
1219         {Opt_stripe, "stripe=%u"},
1220         {Opt_delalloc, "delalloc"},
1221         {Opt_nodelalloc, "nodelalloc"},
1222         {Opt_removed, "mblk_io_submit"},
1223         {Opt_removed, "nomblk_io_submit"},
1224         {Opt_block_validity, "block_validity"},
1225         {Opt_noblock_validity, "noblock_validity"},
1226         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1227         {Opt_journal_ioprio, "journal_ioprio=%u"},
1228         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1229         {Opt_auto_da_alloc, "auto_da_alloc"},
1230         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1231         {Opt_dioread_nolock, "dioread_nolock"},
1232         {Opt_dioread_lock, "dioread_lock"},
1233         {Opt_discard, "discard"},
1234         {Opt_nodiscard, "nodiscard"},
1235         {Opt_init_itable, "init_itable=%u"},
1236         {Opt_init_itable, "init_itable"},
1237         {Opt_noinit_itable, "noinit_itable"},
1238         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1239         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1240         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1241         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1242         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1243         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1244         {Opt_err, NULL},
1245 };
1246
1247 static ext4_fsblk_t get_sb_block(void **data)
1248 {
1249         ext4_fsblk_t    sb_block;
1250         char            *options = (char *) *data;
1251
1252         if (!options || strncmp(options, "sb=", 3) != 0)
1253                 return 1;       /* Default location */
1254
1255         options += 3;
1256         /* TODO: use simple_strtoll with >32bit ext4 */
1257         sb_block = simple_strtoul(options, &options, 0);
1258         if (*options && *options != ',') {
1259                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1260                        (char *) *data);
1261                 return 1;
1262         }
1263         if (*options == ',')
1264                 options++;
1265         *data = (void *) options;
1266
1267         return sb_block;
1268 }
1269
1270 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1271 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1272         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1273
1274 #ifdef CONFIG_QUOTA
1275 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1276 {
1277         struct ext4_sb_info *sbi = EXT4_SB(sb);
1278         char *qname;
1279         int ret = -1;
1280
1281         if (sb_any_quota_loaded(sb) &&
1282                 !sbi->s_qf_names[qtype]) {
1283                 ext4_msg(sb, KERN_ERR,
1284                         "Cannot change journaled "
1285                         "quota options when quota turned on");
1286                 return -1;
1287         }
1288         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1289                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1290                          "when QUOTA feature is enabled");
1291                 return -1;
1292         }
1293         qname = match_strdup(args);
1294         if (!qname) {
1295                 ext4_msg(sb, KERN_ERR,
1296                         "Not enough memory for storing quotafile name");
1297                 return -1;
1298         }
1299         if (sbi->s_qf_names[qtype]) {
1300                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1301                         ret = 1;
1302                 else
1303                         ext4_msg(sb, KERN_ERR,
1304                                  "%s quota file already specified",
1305                                  QTYPE2NAME(qtype));
1306                 goto errout;
1307         }
1308         if (strchr(qname, '/')) {
1309                 ext4_msg(sb, KERN_ERR,
1310                         "quotafile must be on filesystem root");
1311                 goto errout;
1312         }
1313         sbi->s_qf_names[qtype] = qname;
1314         set_opt(sb, QUOTA);
1315         return 1;
1316 errout:
1317         kfree(qname);
1318         return ret;
1319 }
1320
1321 static int clear_qf_name(struct super_block *sb, int qtype)
1322 {
1323
1324         struct ext4_sb_info *sbi = EXT4_SB(sb);
1325
1326         if (sb_any_quota_loaded(sb) &&
1327                 sbi->s_qf_names[qtype]) {
1328                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1329                         " when quota turned on");
1330                 return -1;
1331         }
1332         kfree(sbi->s_qf_names[qtype]);
1333         sbi->s_qf_names[qtype] = NULL;
1334         return 1;
1335 }
1336 #endif
1337
1338 #define MOPT_SET        0x0001
1339 #define MOPT_CLEAR      0x0002
1340 #define MOPT_NOSUPPORT  0x0004
1341 #define MOPT_EXPLICIT   0x0008
1342 #define MOPT_CLEAR_ERR  0x0010
1343 #define MOPT_GTE0       0x0020
1344 #ifdef CONFIG_QUOTA
1345 #define MOPT_Q          0
1346 #define MOPT_QFMT       0x0040
1347 #else
1348 #define MOPT_Q          MOPT_NOSUPPORT
1349 #define MOPT_QFMT       MOPT_NOSUPPORT
1350 #endif
1351 #define MOPT_DATAJ      0x0080
1352 #define MOPT_NO_EXT2    0x0100
1353 #define MOPT_NO_EXT3    0x0200
1354 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1355 #define MOPT_STRING     0x0400
1356
1357 static const struct mount_opts {
1358         int     token;
1359         int     mount_opt;
1360         int     flags;
1361 } ext4_mount_opts[] = {
1362         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1363         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1364         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1365         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1366         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1367         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1368         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1369          MOPT_EXT4_ONLY | MOPT_SET},
1370         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1371          MOPT_EXT4_ONLY | MOPT_CLEAR},
1372         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1373         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1374         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1375          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1376         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1377          MOPT_EXT4_ONLY | MOPT_CLEAR},
1378         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1379          MOPT_EXT4_ONLY | MOPT_SET},
1380         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1381                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1382          MOPT_EXT4_ONLY | MOPT_SET},
1383         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1384         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1385         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1386         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1387         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1388          MOPT_NO_EXT2 | MOPT_SET},
1389         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1390          MOPT_NO_EXT2 | MOPT_CLEAR},
1391         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1392         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1393         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1394         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1395         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1396         {Opt_commit, 0, MOPT_GTE0},
1397         {Opt_max_batch_time, 0, MOPT_GTE0},
1398         {Opt_min_batch_time, 0, MOPT_GTE0},
1399         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1400         {Opt_init_itable, 0, MOPT_GTE0},
1401         {Opt_stripe, 0, MOPT_GTE0},
1402         {Opt_resuid, 0, MOPT_GTE0},
1403         {Opt_resgid, 0, MOPT_GTE0},
1404         {Opt_journal_dev, 0, MOPT_GTE0},
1405         {Opt_journal_path, 0, MOPT_STRING},
1406         {Opt_journal_ioprio, 0, MOPT_GTE0},
1407         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1408         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1409         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1410          MOPT_NO_EXT2 | MOPT_DATAJ},
1411         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1412         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1413 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1414         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1415         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1416 #else
1417         {Opt_acl, 0, MOPT_NOSUPPORT},
1418         {Opt_noacl, 0, MOPT_NOSUPPORT},
1419 #endif
1420         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1421         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1422         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1423         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1424                                                         MOPT_SET | MOPT_Q},
1425         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1426                                                         MOPT_SET | MOPT_Q},
1427         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1428                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1429         {Opt_usrjquota, 0, MOPT_Q},
1430         {Opt_grpjquota, 0, MOPT_Q},
1431         {Opt_offusrjquota, 0, MOPT_Q},
1432         {Opt_offgrpjquota, 0, MOPT_Q},
1433         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1434         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1435         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1436         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1437         {Opt_err, 0, 0}
1438 };
1439
1440 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1441                             substring_t *args, unsigned long *journal_devnum,
1442                             unsigned int *journal_ioprio, int is_remount)
1443 {
1444         struct ext4_sb_info *sbi = EXT4_SB(sb);
1445         const struct mount_opts *m;
1446         kuid_t uid;
1447         kgid_t gid;
1448         int arg = 0;
1449
1450 #ifdef CONFIG_QUOTA
1451         if (token == Opt_usrjquota)
1452                 return set_qf_name(sb, USRQUOTA, &args[0]);
1453         else if (token == Opt_grpjquota)
1454                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1455         else if (token == Opt_offusrjquota)
1456                 return clear_qf_name(sb, USRQUOTA);
1457         else if (token == Opt_offgrpjquota)
1458                 return clear_qf_name(sb, GRPQUOTA);
1459 #endif
1460         switch (token) {
1461         case Opt_noacl:
1462         case Opt_nouser_xattr:
1463                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1464                 break;
1465         case Opt_sb:
1466                 return 1;       /* handled by get_sb_block() */
1467         case Opt_removed:
1468                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1469                 return 1;
1470         case Opt_abort:
1471                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1472                 return 1;
1473         case Opt_i_version:
1474                 sb->s_flags |= MS_I_VERSION;
1475                 return 1;
1476         }
1477
1478         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1479                 if (token == m->token)
1480                         break;
1481
1482         if (m->token == Opt_err) {
1483                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1484                          "or missing value", opt);
1485                 return -1;
1486         }
1487
1488         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1489                 ext4_msg(sb, KERN_ERR,
1490                          "Mount option \"%s\" incompatible with ext2", opt);
1491                 return -1;
1492         }
1493         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1494                 ext4_msg(sb, KERN_ERR,
1495                          "Mount option \"%s\" incompatible with ext3", opt);
1496                 return -1;
1497         }
1498
1499         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1500                 return -1;
1501         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1502                 return -1;
1503         if (m->flags & MOPT_EXPLICIT)
1504                 set_opt2(sb, EXPLICIT_DELALLOC);
1505         if (m->flags & MOPT_CLEAR_ERR)
1506                 clear_opt(sb, ERRORS_MASK);
1507         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1508                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1509                          "options when quota turned on");
1510                 return -1;
1511         }
1512
1513         if (m->flags & MOPT_NOSUPPORT) {
1514                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1515         } else if (token == Opt_commit) {
1516                 if (arg == 0)
1517                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1518                 sbi->s_commit_interval = HZ * arg;
1519         } else if (token == Opt_max_batch_time) {
1520                 sbi->s_max_batch_time = arg;
1521         } else if (token == Opt_min_batch_time) {
1522                 sbi->s_min_batch_time = arg;
1523         } else if (token == Opt_inode_readahead_blks) {
1524                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1525                         ext4_msg(sb, KERN_ERR,
1526                                  "EXT4-fs: inode_readahead_blks must be "
1527                                  "0 or a power of 2 smaller than 2^31");
1528                         return -1;
1529                 }
1530                 sbi->s_inode_readahead_blks = arg;
1531         } else if (token == Opt_init_itable) {
1532                 set_opt(sb, INIT_INODE_TABLE);
1533                 if (!args->from)
1534                         arg = EXT4_DEF_LI_WAIT_MULT;
1535                 sbi->s_li_wait_mult = arg;
1536         } else if (token == Opt_max_dir_size_kb) {
1537                 sbi->s_max_dir_size_kb = arg;
1538         } else if (token == Opt_stripe) {
1539                 sbi->s_stripe = arg;
1540         } else if (token == Opt_resuid) {
1541                 uid = make_kuid(current_user_ns(), arg);
1542                 if (!uid_valid(uid)) {
1543                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1544                         return -1;
1545                 }
1546                 sbi->s_resuid = uid;
1547         } else if (token == Opt_resgid) {
1548                 gid = make_kgid(current_user_ns(), arg);
1549                 if (!gid_valid(gid)) {
1550                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1551                         return -1;
1552                 }
1553                 sbi->s_resgid = gid;
1554         } else if (token == Opt_journal_dev) {
1555                 if (is_remount) {
1556                         ext4_msg(sb, KERN_ERR,
1557                                  "Cannot specify journal on remount");
1558                         return -1;
1559                 }
1560                 *journal_devnum = arg;
1561         } else if (token == Opt_journal_path) {
1562                 char *journal_path;
1563                 struct inode *journal_inode;
1564                 struct path path;
1565                 int error;
1566
1567                 if (is_remount) {
1568                         ext4_msg(sb, KERN_ERR,
1569                                  "Cannot specify journal on remount");
1570                         return -1;
1571                 }
1572                 journal_path = match_strdup(&args[0]);
1573                 if (!journal_path) {
1574                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1575                                 "journal device string");
1576                         return -1;
1577                 }
1578
1579                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1580                 if (error) {
1581                         ext4_msg(sb, KERN_ERR, "error: could not find "
1582                                 "journal device path: error %d", error);
1583                         kfree(journal_path);
1584                         return -1;
1585                 }
1586
1587                 journal_inode = path.dentry->d_inode;
1588                 if (!S_ISBLK(journal_inode->i_mode)) {
1589                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1590                                 "is not a block device", journal_path);
1591                         path_put(&path);
1592                         kfree(journal_path);
1593                         return -1;
1594                 }
1595
1596                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1597                 path_put(&path);
1598                 kfree(journal_path);
1599         } else if (token == Opt_journal_ioprio) {
1600                 if (arg > 7) {
1601                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1602                                  " (must be 0-7)");
1603                         return -1;
1604                 }
1605                 *journal_ioprio =
1606                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1607         } else if (m->flags & MOPT_DATAJ) {
1608                 if (is_remount) {
1609                         if (!sbi->s_journal)
1610                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1611                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1612                                 ext4_msg(sb, KERN_ERR,
1613                                          "Cannot change data mode on remount");
1614                                 return -1;
1615                         }
1616                 } else {
1617                         clear_opt(sb, DATA_FLAGS);
1618                         sbi->s_mount_opt |= m->mount_opt;
1619                 }
1620 #ifdef CONFIG_QUOTA
1621         } else if (m->flags & MOPT_QFMT) {
1622                 if (sb_any_quota_loaded(sb) &&
1623                     sbi->s_jquota_fmt != m->mount_opt) {
1624                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1625                                  "quota options when quota turned on");
1626                         return -1;
1627                 }
1628                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1629                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1630                         ext4_msg(sb, KERN_ERR,
1631                                  "Cannot set journaled quota options "
1632                                  "when QUOTA feature is enabled");
1633                         return -1;
1634                 }
1635                 sbi->s_jquota_fmt = m->mount_opt;
1636 #endif
1637         } else {
1638                 if (!args->from)
1639                         arg = 1;
1640                 if (m->flags & MOPT_CLEAR)
1641                         arg = !arg;
1642                 else if (unlikely(!(m->flags & MOPT_SET))) {
1643                         ext4_msg(sb, KERN_WARNING,
1644                                  "buggy handling of option %s", opt);
1645                         WARN_ON(1);
1646                         return -1;
1647                 }
1648                 if (arg != 0)
1649                         sbi->s_mount_opt |= m->mount_opt;
1650                 else
1651                         sbi->s_mount_opt &= ~m->mount_opt;
1652         }
1653         return 1;
1654 }
1655
1656 static int parse_options(char *options, struct super_block *sb,
1657                          unsigned long *journal_devnum,
1658                          unsigned int *journal_ioprio,
1659                          int is_remount)
1660 {
1661         struct ext4_sb_info *sbi = EXT4_SB(sb);
1662         char *p;
1663         substring_t args[MAX_OPT_ARGS];
1664         int token;
1665
1666         if (!options)
1667                 return 1;
1668
1669         while ((p = strsep(&options, ",")) != NULL) {
1670                 if (!*p)
1671                         continue;
1672                 /*
1673                  * Initialize args struct so we know whether arg was
1674                  * found; some options take optional arguments.
1675                  */
1676                 args[0].to = args[0].from = NULL;
1677                 token = match_token(p, tokens, args);
1678                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1679                                      journal_ioprio, is_remount) < 0)
1680                         return 0;
1681         }
1682 #ifdef CONFIG_QUOTA
1683         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1684             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1685                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1686                          "feature is enabled");
1687                 return 0;
1688         }
1689         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1690                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1691                         clear_opt(sb, USRQUOTA);
1692
1693                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1694                         clear_opt(sb, GRPQUOTA);
1695
1696                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1697                         ext4_msg(sb, KERN_ERR, "old and new quota "
1698                                         "format mixing");
1699                         return 0;
1700                 }
1701
1702                 if (!sbi->s_jquota_fmt) {
1703                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1704                                         "not specified");
1705                         return 0;
1706                 }
1707         }
1708 #endif
1709         if (test_opt(sb, DIOREAD_NOLOCK)) {
1710                 int blocksize =
1711                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1712
1713                 if (blocksize < PAGE_CACHE_SIZE) {
1714                         ext4_msg(sb, KERN_ERR, "can't mount with "
1715                                  "dioread_nolock if block size != PAGE_SIZE");
1716                         return 0;
1717                 }
1718         }
1719         return 1;
1720 }
1721
1722 static inline void ext4_show_quota_options(struct seq_file *seq,
1723                                            struct super_block *sb)
1724 {
1725 #if defined(CONFIG_QUOTA)
1726         struct ext4_sb_info *sbi = EXT4_SB(sb);
1727
1728         if (sbi->s_jquota_fmt) {
1729                 char *fmtname = "";
1730
1731                 switch (sbi->s_jquota_fmt) {
1732                 case QFMT_VFS_OLD:
1733                         fmtname = "vfsold";
1734                         break;
1735                 case QFMT_VFS_V0:
1736                         fmtname = "vfsv0";
1737                         break;
1738                 case QFMT_VFS_V1:
1739                         fmtname = "vfsv1";
1740                         break;
1741                 }
1742                 seq_printf(seq, ",jqfmt=%s", fmtname);
1743         }
1744
1745         if (sbi->s_qf_names[USRQUOTA])
1746                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1747
1748         if (sbi->s_qf_names[GRPQUOTA])
1749                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1750 #endif
1751 }
1752
1753 static const char *token2str(int token)
1754 {
1755         const struct match_token *t;
1756
1757         for (t = tokens; t->token != Opt_err; t++)
1758                 if (t->token == token && !strchr(t->pattern, '='))
1759                         break;
1760         return t->pattern;
1761 }
1762
1763 /*
1764  * Show an option if
1765  *  - it's set to a non-default value OR
1766  *  - if the per-sb default is different from the global default
1767  */
1768 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1769                               int nodefs)
1770 {
1771         struct ext4_sb_info *sbi = EXT4_SB(sb);
1772         struct ext4_super_block *es = sbi->s_es;
1773         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1774         const struct mount_opts *m;
1775         char sep = nodefs ? '\n' : ',';
1776
1777 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1778 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1779
1780         if (sbi->s_sb_block != 1)
1781                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1782
1783         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1784                 int want_set = m->flags & MOPT_SET;
1785                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1786                     (m->flags & MOPT_CLEAR_ERR))
1787                         continue;
1788                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1789                         continue; /* skip if same as the default */
1790                 if ((want_set &&
1791                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1792                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1793                         continue; /* select Opt_noFoo vs Opt_Foo */
1794                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1795         }
1796
1797         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1798             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1799                 SEQ_OPTS_PRINT("resuid=%u",
1800                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1801         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1802             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1803                 SEQ_OPTS_PRINT("resgid=%u",
1804                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1805         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1806         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1807                 SEQ_OPTS_PUTS("errors=remount-ro");
1808         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1809                 SEQ_OPTS_PUTS("errors=continue");
1810         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1811                 SEQ_OPTS_PUTS("errors=panic");
1812         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1813                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1814         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1815                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1816         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1817                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1818         if (sb->s_flags & MS_I_VERSION)
1819                 SEQ_OPTS_PUTS("i_version");
1820         if (nodefs || sbi->s_stripe)
1821                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1822         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1823                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1824                         SEQ_OPTS_PUTS("data=journal");
1825                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1826                         SEQ_OPTS_PUTS("data=ordered");
1827                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1828                         SEQ_OPTS_PUTS("data=writeback");
1829         }
1830         if (nodefs ||
1831             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1832                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1833                                sbi->s_inode_readahead_blks);
1834
1835         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1836                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1837                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1838         if (nodefs || sbi->s_max_dir_size_kb)
1839                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1840
1841         ext4_show_quota_options(seq, sb);
1842         return 0;
1843 }
1844
1845 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1846 {
1847         return _ext4_show_options(seq, root->d_sb, 0);
1848 }
1849
1850 static int options_seq_show(struct seq_file *seq, void *offset)
1851 {
1852         struct super_block *sb = seq->private;
1853         int rc;
1854
1855         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1856         rc = _ext4_show_options(seq, sb, 1);
1857         seq_puts(seq, "\n");
1858         return rc;
1859 }
1860
1861 static int options_open_fs(struct inode *inode, struct file *file)
1862 {
1863         return single_open(file, options_seq_show, PDE_DATA(inode));
1864 }
1865
1866 static const struct file_operations ext4_seq_options_fops = {
1867         .owner = THIS_MODULE,
1868         .open = options_open_fs,
1869         .read = seq_read,
1870         .llseek = seq_lseek,
1871         .release = single_release,
1872 };
1873
1874 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1875                             int read_only)
1876 {
1877         struct ext4_sb_info *sbi = EXT4_SB(sb);
1878         int res = 0;
1879
1880         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1881                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1882                          "forcing read-only mode");
1883                 res = MS_RDONLY;
1884         }
1885         if (read_only)
1886                 goto done;
1887         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1888                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1889                          "running e2fsck is recommended");
1890         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "warning: mounting fs with errors, "
1893                          "running e2fsck is recommended");
1894         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1895                  le16_to_cpu(es->s_mnt_count) >=
1896                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1897                 ext4_msg(sb, KERN_WARNING,
1898                          "warning: maximal mount count reached, "
1899                          "running e2fsck is recommended");
1900         else if (le32_to_cpu(es->s_checkinterval) &&
1901                 (le32_to_cpu(es->s_lastcheck) +
1902                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1903                 ext4_msg(sb, KERN_WARNING,
1904                          "warning: checktime reached, "
1905                          "running e2fsck is recommended");
1906         if (!sbi->s_journal)
1907                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1908         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1909                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1910         le16_add_cpu(&es->s_mnt_count, 1);
1911         es->s_mtime = cpu_to_le32(get_seconds());
1912         ext4_update_dynamic_rev(sb);
1913         if (sbi->s_journal)
1914                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1915
1916         ext4_commit_super(sb, 1);
1917 done:
1918         if (test_opt(sb, DEBUG))
1919                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1920                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1921                         sb->s_blocksize,
1922                         sbi->s_groups_count,
1923                         EXT4_BLOCKS_PER_GROUP(sb),
1924                         EXT4_INODES_PER_GROUP(sb),
1925                         sbi->s_mount_opt, sbi->s_mount_opt2);
1926
1927         cleancache_init_fs(sb);
1928         return res;
1929 }
1930
1931 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1932 {
1933         struct ext4_sb_info *sbi = EXT4_SB(sb);
1934         struct flex_groups *new_groups;
1935         int size;
1936
1937         if (!sbi->s_log_groups_per_flex)
1938                 return 0;
1939
1940         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1941         if (size <= sbi->s_flex_groups_allocated)
1942                 return 0;
1943
1944         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1945         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1946         if (!new_groups) {
1947                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1948                          size / (int) sizeof(struct flex_groups));
1949                 return -ENOMEM;
1950         }
1951
1952         if (sbi->s_flex_groups) {
1953                 memcpy(new_groups, sbi->s_flex_groups,
1954                        (sbi->s_flex_groups_allocated *
1955                         sizeof(struct flex_groups)));
1956                 ext4_kvfree(sbi->s_flex_groups);
1957         }
1958         sbi->s_flex_groups = new_groups;
1959         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1960         return 0;
1961 }
1962
1963 static int ext4_fill_flex_info(struct super_block *sb)
1964 {
1965         struct ext4_sb_info *sbi = EXT4_SB(sb);
1966         struct ext4_group_desc *gdp = NULL;
1967         ext4_group_t flex_group;
1968         int i, err;
1969
1970         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1971         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1972                 sbi->s_log_groups_per_flex = 0;
1973                 return 1;
1974         }
1975
1976         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1977         if (err)
1978                 goto failed;
1979
1980         for (i = 0; i < sbi->s_groups_count; i++) {
1981                 gdp = ext4_get_group_desc(sb, i, NULL);
1982
1983                 flex_group = ext4_flex_group(sbi, i);
1984                 atomic_add(ext4_free_inodes_count(sb, gdp),
1985                            &sbi->s_flex_groups[flex_group].free_inodes);
1986                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1987                              &sbi->s_flex_groups[flex_group].free_clusters);
1988                 atomic_add(ext4_used_dirs_count(sb, gdp),
1989                            &sbi->s_flex_groups[flex_group].used_dirs);
1990         }
1991
1992         return 1;
1993 failed:
1994         return 0;
1995 }
1996
1997 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1998                                    struct ext4_group_desc *gdp)
1999 {
2000         int offset;
2001         __u16 crc = 0;
2002         __le32 le_group = cpu_to_le32(block_group);
2003
2004         if (ext4_has_metadata_csum(sbi->s_sb)) {
2005                 /* Use new metadata_csum algorithm */
2006                 __le16 save_csum;
2007                 __u32 csum32;
2008
2009                 save_csum = gdp->bg_checksum;
2010                 gdp->bg_checksum = 0;
2011                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2012                                      sizeof(le_group));
2013                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2014                                      sbi->s_desc_size);
2015                 gdp->bg_checksum = save_csum;
2016
2017                 crc = csum32 & 0xFFFF;
2018                 goto out;
2019         }
2020
2021         /* old crc16 code */
2022         if (!(sbi->s_es->s_feature_ro_compat &
2023               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2024                 return 0;
2025
2026         offset = offsetof(struct ext4_group_desc, bg_checksum);
2027
2028         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2029         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2030         crc = crc16(crc, (__u8 *)gdp, offset);
2031         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2032         /* for checksum of struct ext4_group_desc do the rest...*/
2033         if ((sbi->s_es->s_feature_incompat &
2034              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2035             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2036                 crc = crc16(crc, (__u8 *)gdp + offset,
2037                             le16_to_cpu(sbi->s_es->s_desc_size) -
2038                                 offset);
2039
2040 out:
2041         return cpu_to_le16(crc);
2042 }
2043
2044 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2045                                 struct ext4_group_desc *gdp)
2046 {
2047         if (ext4_has_group_desc_csum(sb) &&
2048             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2049                                                       block_group, gdp)))
2050                 return 0;
2051
2052         return 1;
2053 }
2054
2055 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2056                               struct ext4_group_desc *gdp)
2057 {
2058         if (!ext4_has_group_desc_csum(sb))
2059                 return;
2060         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2061 }
2062
2063 /* Called at mount-time, super-block is locked */
2064 static int ext4_check_descriptors(struct super_block *sb,
2065                                   ext4_group_t *first_not_zeroed)
2066 {
2067         struct ext4_sb_info *sbi = EXT4_SB(sb);
2068         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2069         ext4_fsblk_t last_block;
2070         ext4_fsblk_t block_bitmap;
2071         ext4_fsblk_t inode_bitmap;
2072         ext4_fsblk_t inode_table;
2073         int flexbg_flag = 0;
2074         ext4_group_t i, grp = sbi->s_groups_count;
2075
2076         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2077                 flexbg_flag = 1;
2078
2079         ext4_debug("Checking group descriptors");
2080
2081         for (i = 0; i < sbi->s_groups_count; i++) {
2082                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2083
2084                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2085                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2086                 else
2087                         last_block = first_block +
2088                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2089
2090                 if ((grp == sbi->s_groups_count) &&
2091                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2092                         grp = i;
2093
2094                 block_bitmap = ext4_block_bitmap(sb, gdp);
2095                 if (block_bitmap < first_block || block_bitmap > last_block) {
2096                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2097                                "Block bitmap for group %u not in group "
2098                                "(block %llu)!", i, block_bitmap);
2099                         return 0;
2100                 }
2101                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2102                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2103                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2104                                "Inode bitmap for group %u not in group "
2105                                "(block %llu)!", i, inode_bitmap);
2106                         return 0;
2107                 }
2108                 inode_table = ext4_inode_table(sb, gdp);
2109                 if (inode_table < first_block ||
2110                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2111                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2112                                "Inode table for group %u not in group "
2113                                "(block %llu)!", i, inode_table);
2114                         return 0;
2115                 }
2116                 ext4_lock_group(sb, i);
2117                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2118                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2119                                  "Checksum for group %u failed (%u!=%u)",
2120                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2121                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2122                         if (!(sb->s_flags & MS_RDONLY)) {
2123                                 ext4_unlock_group(sb, i);
2124                                 return 0;
2125                         }
2126                 }
2127                 ext4_unlock_group(sb, i);
2128                 if (!flexbg_flag)
2129                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2130         }
2131         if (NULL != first_not_zeroed)
2132                 *first_not_zeroed = grp;
2133
2134         ext4_free_blocks_count_set(sbi->s_es,
2135                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2136         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2137         return 1;
2138 }
2139
2140 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2141  * the superblock) which were deleted from all directories, but held open by
2142  * a process at the time of a crash.  We walk the list and try to delete these
2143  * inodes at recovery time (only with a read-write filesystem).
2144  *
2145  * In order to keep the orphan inode chain consistent during traversal (in
2146  * case of crash during recovery), we link each inode into the superblock
2147  * orphan list_head and handle it the same way as an inode deletion during
2148  * normal operation (which journals the operations for us).
2149  *
2150  * We only do an iget() and an iput() on each inode, which is very safe if we
2151  * accidentally point at an in-use or already deleted inode.  The worst that
2152  * can happen in this case is that we get a "bit already cleared" message from
2153  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2154  * e2fsck was run on this filesystem, and it must have already done the orphan
2155  * inode cleanup for us, so we can safely abort without any further action.
2156  */
2157 static void ext4_orphan_cleanup(struct super_block *sb,
2158                                 struct ext4_super_block *es)
2159 {
2160         unsigned int s_flags = sb->s_flags;
2161         int nr_orphans = 0, nr_truncates = 0;
2162 #ifdef CONFIG_QUOTA
2163         int i;
2164 #endif
2165         if (!es->s_last_orphan) {
2166                 jbd_debug(4, "no orphan inodes to clean up\n");
2167                 return;
2168         }
2169
2170         if (bdev_read_only(sb->s_bdev)) {
2171                 ext4_msg(sb, KERN_ERR, "write access "
2172                         "unavailable, skipping orphan cleanup");
2173                 return;
2174         }
2175
2176         /* Check if feature set would not allow a r/w mount */
2177         if (!ext4_feature_set_ok(sb, 0)) {
2178                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2179                          "unknown ROCOMPAT features");
2180                 return;
2181         }
2182
2183         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2184                 /* don't clear list on RO mount w/ errors */
2185                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2186                         jbd_debug(1, "Errors on filesystem, "
2187                                   "clearing orphan list.\n");
2188                         es->s_last_orphan = 0;
2189                 }
2190                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2191                 return;
2192         }
2193
2194         if (s_flags & MS_RDONLY) {
2195                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2196                 sb->s_flags &= ~MS_RDONLY;
2197         }
2198 #ifdef CONFIG_QUOTA
2199         /* Needed for iput() to work correctly and not trash data */
2200         sb->s_flags |= MS_ACTIVE;
2201         /* Turn on quotas so that they are updated correctly */
2202         for (i = 0; i < MAXQUOTAS; i++) {
2203                 if (EXT4_SB(sb)->s_qf_names[i]) {
2204                         int ret = ext4_quota_on_mount(sb, i);
2205                         if (ret < 0)
2206                                 ext4_msg(sb, KERN_ERR,
2207                                         "Cannot turn on journaled "
2208                                         "quota: error %d", ret);
2209                 }
2210         }
2211 #endif
2212
2213         while (es->s_last_orphan) {
2214                 struct inode *inode;
2215
2216                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2217                 if (IS_ERR(inode)) {
2218                         es->s_last_orphan = 0;
2219                         break;
2220                 }
2221
2222                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2223                 dquot_initialize(inode);
2224                 if (inode->i_nlink) {
2225                         if (test_opt(sb, DEBUG))
2226                                 ext4_msg(sb, KERN_DEBUG,
2227                                         "%s: truncating inode %lu to %lld bytes",
2228                                         __func__, inode->i_ino, inode->i_size);
2229                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2230                                   inode->i_ino, inode->i_size);
2231                         mutex_lock(&inode->i_mutex);
2232                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2233                         ext4_truncate(inode);
2234                         mutex_unlock(&inode->i_mutex);
2235                         nr_truncates++;
2236                 } else {
2237                         if (test_opt(sb, DEBUG))
2238                                 ext4_msg(sb, KERN_DEBUG,
2239                                         "%s: deleting unreferenced inode %lu",
2240                                         __func__, inode->i_ino);
2241                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2242                                   inode->i_ino);
2243                         nr_orphans++;
2244                 }
2245                 iput(inode);  /* The delete magic happens here! */
2246         }
2247
2248 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2249
2250         if (nr_orphans)
2251                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2252                        PLURAL(nr_orphans));
2253         if (nr_truncates)
2254                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2255                        PLURAL(nr_truncates));
2256 #ifdef CONFIG_QUOTA
2257         /* Turn quotas off */
2258         for (i = 0; i < MAXQUOTAS; i++) {
2259                 if (sb_dqopt(sb)->files[i])
2260                         dquot_quota_off(sb, i);
2261         }
2262 #endif
2263         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2264 }
2265
2266 /*
2267  * Maximal extent format file size.
2268  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2269  * extent format containers, within a sector_t, and within i_blocks
2270  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2271  * so that won't be a limiting factor.
2272  *
2273  * However there is other limiting factor. We do store extents in the form
2274  * of starting block and length, hence the resulting length of the extent
2275  * covering maximum file size must fit into on-disk format containers as
2276  * well. Given that length is always by 1 unit bigger than max unit (because
2277  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2278  *
2279  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2280  */
2281 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2282 {
2283         loff_t res;
2284         loff_t upper_limit = MAX_LFS_FILESIZE;
2285
2286         /* small i_blocks in vfs inode? */
2287         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2288                 /*
2289                  * CONFIG_LBDAF is not enabled implies the inode
2290                  * i_block represent total blocks in 512 bytes
2291                  * 32 == size of vfs inode i_blocks * 8
2292                  */
2293                 upper_limit = (1LL << 32) - 1;
2294
2295                 /* total blocks in file system block size */
2296                 upper_limit >>= (blkbits - 9);
2297                 upper_limit <<= blkbits;
2298         }
2299
2300         /*
2301          * 32-bit extent-start container, ee_block. We lower the maxbytes
2302          * by one fs block, so ee_len can cover the extent of maximum file
2303          * size
2304          */
2305         res = (1LL << 32) - 1;
2306         res <<= blkbits;
2307
2308         /* Sanity check against vm- & vfs- imposed limits */
2309         if (res > upper_limit)
2310                 res = upper_limit;
2311
2312         return res;
2313 }
2314
2315 /*
2316  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2317  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2318  * We need to be 1 filesystem block less than the 2^48 sector limit.
2319  */
2320 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2321 {
2322         loff_t res = EXT4_NDIR_BLOCKS;
2323         int meta_blocks;
2324         loff_t upper_limit;
2325         /* This is calculated to be the largest file size for a dense, block
2326          * mapped file such that the file's total number of 512-byte sectors,
2327          * including data and all indirect blocks, does not exceed (2^48 - 1).
2328          *
2329          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2330          * number of 512-byte sectors of the file.
2331          */
2332
2333         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2334                 /*
2335                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2336                  * the inode i_block field represents total file blocks in
2337                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2338                  */
2339                 upper_limit = (1LL << 32) - 1;
2340
2341                 /* total blocks in file system block size */
2342                 upper_limit >>= (bits - 9);
2343
2344         } else {
2345                 /*
2346                  * We use 48 bit ext4_inode i_blocks
2347                  * With EXT4_HUGE_FILE_FL set the i_blocks
2348                  * represent total number of blocks in
2349                  * file system block size
2350                  */
2351                 upper_limit = (1LL << 48) - 1;
2352
2353         }
2354
2355         /* indirect blocks */
2356         meta_blocks = 1;
2357         /* double indirect blocks */
2358         meta_blocks += 1 + (1LL << (bits-2));
2359         /* tripple indirect blocks */
2360         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2361
2362         upper_limit -= meta_blocks;
2363         upper_limit <<= bits;
2364
2365         res += 1LL << (bits-2);
2366         res += 1LL << (2*(bits-2));
2367         res += 1LL << (3*(bits-2));
2368         res <<= bits;
2369         if (res > upper_limit)
2370                 res = upper_limit;
2371
2372         if (res > MAX_LFS_FILESIZE)
2373                 res = MAX_LFS_FILESIZE;
2374
2375         return res;
2376 }
2377
2378 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2379                                    ext4_fsblk_t logical_sb_block, int nr)
2380 {
2381         struct ext4_sb_info *sbi = EXT4_SB(sb);
2382         ext4_group_t bg, first_meta_bg;
2383         int has_super = 0;
2384
2385         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2386
2387         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2388             nr < first_meta_bg)
2389                 return logical_sb_block + nr + 1;
2390         bg = sbi->s_desc_per_block * nr;
2391         if (ext4_bg_has_super(sb, bg))
2392                 has_super = 1;
2393
2394         return (has_super + ext4_group_first_block_no(sb, bg));
2395 }
2396
2397 /**
2398  * ext4_get_stripe_size: Get the stripe size.
2399  * @sbi: In memory super block info
2400  *
2401  * If we have specified it via mount option, then
2402  * use the mount option value. If the value specified at mount time is
2403  * greater than the blocks per group use the super block value.
2404  * If the super block value is greater than blocks per group return 0.
2405  * Allocator needs it be less than blocks per group.
2406  *
2407  */
2408 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2409 {
2410         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2411         unsigned long stripe_width =
2412                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2413         int ret;
2414
2415         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2416                 ret = sbi->s_stripe;
2417         else if (stripe_width <= sbi->s_blocks_per_group)
2418                 ret = stripe_width;
2419         else if (stride <= sbi->s_blocks_per_group)
2420                 ret = stride;
2421         else
2422                 ret = 0;
2423
2424         /*
2425          * If the stripe width is 1, this makes no sense and
2426          * we set it to 0 to turn off stripe handling code.
2427          */
2428         if (ret <= 1)
2429                 ret = 0;
2430
2431         return ret;
2432 }
2433
2434 /* sysfs supprt */
2435
2436 struct ext4_attr {
2437         struct attribute attr;
2438         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2439         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2440                          const char *, size_t);
2441         union {
2442                 int offset;
2443                 int deprecated_val;
2444         } u;
2445 };
2446
2447 static int parse_strtoull(const char *buf,
2448                 unsigned long long max, unsigned long long *value)
2449 {
2450         int ret;
2451
2452         ret = kstrtoull(skip_spaces(buf), 0, value);
2453         if (!ret && *value > max)
2454                 ret = -EINVAL;
2455         return ret;
2456 }
2457
2458 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2459                                               struct ext4_sb_info *sbi,
2460                                               char *buf)
2461 {
2462         return snprintf(buf, PAGE_SIZE, "%llu\n",
2463                 (s64) EXT4_C2B(sbi,
2464                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2465 }
2466
2467 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2468                                          struct ext4_sb_info *sbi, char *buf)
2469 {
2470         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2471
2472         if (!sb->s_bdev->bd_part)
2473                 return snprintf(buf, PAGE_SIZE, "0\n");
2474         return snprintf(buf, PAGE_SIZE, "%lu\n",
2475                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2476                          sbi->s_sectors_written_start) >> 1);
2477 }
2478
2479 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2480                                           struct ext4_sb_info *sbi, char *buf)
2481 {
2482         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2483
2484         if (!sb->s_bdev->bd_part)
2485                 return snprintf(buf, PAGE_SIZE, "0\n");
2486         return snprintf(buf, PAGE_SIZE, "%llu\n",
2487                         (unsigned long long)(sbi->s_kbytes_written +
2488                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2489                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2490 }
2491
2492 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2493                                           struct ext4_sb_info *sbi,
2494                                           const char *buf, size_t count)
2495 {
2496         unsigned long t;
2497         int ret;
2498
2499         ret = kstrtoul(skip_spaces(buf), 0, &t);
2500         if (ret)
2501                 return ret;
2502
2503         if (t && (!is_power_of_2(t) || t > 0x40000000))
2504                 return -EINVAL;
2505
2506         sbi->s_inode_readahead_blks = t;
2507         return count;
2508 }
2509
2510 static ssize_t sbi_ui_show(struct ext4_attr *a,
2511                            struct ext4_sb_info *sbi, char *buf)
2512 {
2513         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2514
2515         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2516 }
2517
2518 static ssize_t sbi_ui_store(struct ext4_attr *a,
2519                             struct ext4_sb_info *sbi,
2520                             const char *buf, size_t count)
2521 {
2522         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2523         unsigned long t;
2524         int ret;
2525
2526         ret = kstrtoul(skip_spaces(buf), 0, &t);
2527         if (ret)
2528                 return ret;
2529         *ui = t;
2530         return count;
2531 }
2532
2533 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2534                                   struct ext4_sb_info *sbi, char *buf)
2535 {
2536         return snprintf(buf, PAGE_SIZE, "%llu\n",
2537                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2538 }
2539
2540 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2541                                    struct ext4_sb_info *sbi,
2542                                    const char *buf, size_t count)
2543 {
2544         unsigned long long val;
2545         int ret;
2546
2547         if (parse_strtoull(buf, -1ULL, &val))
2548                 return -EINVAL;
2549         ret = ext4_reserve_clusters(sbi, val);
2550
2551         return ret ? ret : count;
2552 }
2553
2554 static ssize_t trigger_test_error(struct ext4_attr *a,
2555                                   struct ext4_sb_info *sbi,
2556                                   const char *buf, size_t count)
2557 {
2558         int len = count;
2559
2560         if (!capable(CAP_SYS_ADMIN))
2561                 return -EPERM;
2562
2563         if (len && buf[len-1] == '\n')
2564                 len--;
2565
2566         if (len)
2567                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2568         return count;
2569 }
2570
2571 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2572                                    struct ext4_sb_info *sbi, char *buf)
2573 {
2574         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2575 }
2576
2577 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2578 static struct ext4_attr ext4_attr_##_name = {                   \
2579         .attr = {.name = __stringify(_name), .mode = _mode },   \
2580         .show   = _show,                                        \
2581         .store  = _store,                                       \
2582         .u = {                                                  \
2583                 .offset = offsetof(struct ext4_sb_info, _elname),\
2584         },                                                      \
2585 }
2586 #define EXT4_ATTR(name, mode, show, store) \
2587 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2588
2589 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2590 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2591 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2592 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2593         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2594 #define ATTR_LIST(name) &ext4_attr_##name.attr
2595 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2596 static struct ext4_attr ext4_attr_##_name = {                   \
2597         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2598         .show   = sbi_deprecated_show,                          \
2599         .u = {                                                  \
2600                 .deprecated_val = _val,                         \
2601         },                                                      \
2602 }
2603
2604 EXT4_RO_ATTR(delayed_allocation_blocks);
2605 EXT4_RO_ATTR(session_write_kbytes);
2606 EXT4_RO_ATTR(lifetime_write_kbytes);
2607 EXT4_RW_ATTR(reserved_clusters);
2608 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2609                  inode_readahead_blks_store, s_inode_readahead_blks);
2610 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2611 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2612 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2613 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2614 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2615 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2616 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2617 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2618 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2619 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2620 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2621 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2622 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2623 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2624 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2625 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2626
2627 static struct attribute *ext4_attrs[] = {
2628         ATTR_LIST(delayed_allocation_blocks),
2629         ATTR_LIST(session_write_kbytes),
2630         ATTR_LIST(lifetime_write_kbytes),
2631         ATTR_LIST(reserved_clusters),
2632         ATTR_LIST(inode_readahead_blks),
2633         ATTR_LIST(inode_goal),
2634         ATTR_LIST(mb_stats),
2635         ATTR_LIST(mb_max_to_scan),
2636         ATTR_LIST(mb_min_to_scan),
2637         ATTR_LIST(mb_order2_req),
2638         ATTR_LIST(mb_stream_req),
2639         ATTR_LIST(mb_group_prealloc),
2640         ATTR_LIST(max_writeback_mb_bump),
2641         ATTR_LIST(extent_max_zeroout_kb),
2642         ATTR_LIST(trigger_fs_error),
2643         ATTR_LIST(err_ratelimit_interval_ms),
2644         ATTR_LIST(err_ratelimit_burst),
2645         ATTR_LIST(warning_ratelimit_interval_ms),
2646         ATTR_LIST(warning_ratelimit_burst),
2647         ATTR_LIST(msg_ratelimit_interval_ms),
2648         ATTR_LIST(msg_ratelimit_burst),
2649         NULL,
2650 };
2651
2652 /* Features this copy of ext4 supports */
2653 EXT4_INFO_ATTR(lazy_itable_init);
2654 EXT4_INFO_ATTR(batched_discard);
2655 EXT4_INFO_ATTR(meta_bg_resize);
2656
2657 static struct attribute *ext4_feat_attrs[] = {
2658         ATTR_LIST(lazy_itable_init),
2659         ATTR_LIST(batched_discard),
2660         ATTR_LIST(meta_bg_resize),
2661         NULL,
2662 };
2663
2664 static ssize_t ext4_attr_show(struct kobject *kobj,
2665                               struct attribute *attr, char *buf)
2666 {
2667         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2668                                                 s_kobj);
2669         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2670
2671         return a->show ? a->show(a, sbi, buf) : 0;
2672 }
2673
2674 static ssize_t ext4_attr_store(struct kobject *kobj,
2675                                struct attribute *attr,
2676                                const char *buf, size_t len)
2677 {
2678         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2679                                                 s_kobj);
2680         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2681
2682         return a->store ? a->store(a, sbi, buf, len) : 0;
2683 }
2684
2685 static void ext4_sb_release(struct kobject *kobj)
2686 {
2687         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2688                                                 s_kobj);
2689         complete(&sbi->s_kobj_unregister);
2690 }
2691
2692 static const struct sysfs_ops ext4_attr_ops = {
2693         .show   = ext4_attr_show,
2694         .store  = ext4_attr_store,
2695 };
2696
2697 static struct kobj_type ext4_ktype = {
2698         .default_attrs  = ext4_attrs,
2699         .sysfs_ops      = &ext4_attr_ops,
2700         .release        = ext4_sb_release,
2701 };
2702
2703 static void ext4_feat_release(struct kobject *kobj)
2704 {
2705         complete(&ext4_feat->f_kobj_unregister);
2706 }
2707
2708 static struct kobj_type ext4_feat_ktype = {
2709         .default_attrs  = ext4_feat_attrs,
2710         .sysfs_ops      = &ext4_attr_ops,
2711         .release        = ext4_feat_release,
2712 };
2713
2714 /*
2715  * Check whether this filesystem can be mounted based on
2716  * the features present and the RDONLY/RDWR mount requested.
2717  * Returns 1 if this filesystem can be mounted as requested,
2718  * 0 if it cannot be.
2719  */
2720 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2721 {
2722         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2723                 ext4_msg(sb, KERN_ERR,
2724                         "Couldn't mount because of "
2725                         "unsupported optional features (%x)",
2726                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2727                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2728                 return 0;
2729         }
2730
2731         if (readonly)
2732                 return 1;
2733
2734         /* Check that feature set is OK for a read-write mount */
2735         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2736                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2737                          "unsupported optional features (%x)",
2738                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2739                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2740                 return 0;
2741         }
2742         /*
2743          * Large file size enabled file system can only be mounted
2744          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2745          */
2746         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2747                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2748                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2749                                  "cannot be mounted RDWR without "
2750                                  "CONFIG_LBDAF");
2751                         return 0;
2752                 }
2753         }
2754         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2755             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2756                 ext4_msg(sb, KERN_ERR,
2757                          "Can't support bigalloc feature without "
2758                          "extents feature\n");
2759                 return 0;
2760         }
2761
2762 #ifndef CONFIG_QUOTA
2763         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2764             !readonly) {
2765                 ext4_msg(sb, KERN_ERR,
2766                          "Filesystem with quota feature cannot be mounted RDWR "
2767                          "without CONFIG_QUOTA");
2768                 return 0;
2769         }
2770 #endif  /* CONFIG_QUOTA */
2771         return 1;
2772 }
2773
2774 /*
2775  * This function is called once a day if we have errors logged
2776  * on the file system
2777  */
2778 static void print_daily_error_info(unsigned long arg)
2779 {
2780         struct super_block *sb = (struct super_block *) arg;
2781         struct ext4_sb_info *sbi;
2782         struct ext4_super_block *es;
2783
2784         sbi = EXT4_SB(sb);
2785         es = sbi->s_es;
2786
2787         if (es->s_error_count)
2788                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2789                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2790                          le32_to_cpu(es->s_error_count));
2791         if (es->s_first_error_time) {
2792                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2793                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2794                        (int) sizeof(es->s_first_error_func),
2795                        es->s_first_error_func,
2796                        le32_to_cpu(es->s_first_error_line));
2797                 if (es->s_first_error_ino)
2798                         printk(": inode %u",
2799                                le32_to_cpu(es->s_first_error_ino));
2800                 if (es->s_first_error_block)
2801                         printk(": block %llu", (unsigned long long)
2802                                le64_to_cpu(es->s_first_error_block));
2803                 printk("\n");
2804         }
2805         if (es->s_last_error_time) {
2806                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2807                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2808                        (int) sizeof(es->s_last_error_func),
2809                        es->s_last_error_func,
2810                        le32_to_cpu(es->s_last_error_line));
2811                 if (es->s_last_error_ino)
2812                         printk(": inode %u",
2813                                le32_to_cpu(es->s_last_error_ino));
2814                 if (es->s_last_error_block)
2815                         printk(": block %llu", (unsigned long long)
2816                                le64_to_cpu(es->s_last_error_block));
2817                 printk("\n");
2818         }
2819         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2820 }
2821
2822 /* Find next suitable group and run ext4_init_inode_table */
2823 static int ext4_run_li_request(struct ext4_li_request *elr)
2824 {
2825         struct ext4_group_desc *gdp = NULL;
2826         ext4_group_t group, ngroups;
2827         struct super_block *sb;
2828         unsigned long timeout = 0;
2829         int ret = 0;
2830
2831         sb = elr->lr_super;
2832         ngroups = EXT4_SB(sb)->s_groups_count;
2833
2834         sb_start_write(sb);
2835         for (group = elr->lr_next_group; group < ngroups; group++) {
2836                 gdp = ext4_get_group_desc(sb, group, NULL);
2837                 if (!gdp) {
2838                         ret = 1;
2839                         break;
2840                 }
2841
2842                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2843                         break;
2844         }
2845
2846         if (group >= ngroups)
2847                 ret = 1;
2848
2849         if (!ret) {
2850                 timeout = jiffies;
2851                 ret = ext4_init_inode_table(sb, group,
2852                                             elr->lr_timeout ? 0 : 1);
2853                 if (elr->lr_timeout == 0) {
2854                         timeout = (jiffies - timeout) *
2855                                   elr->lr_sbi->s_li_wait_mult;
2856                         elr->lr_timeout = timeout;
2857                 }
2858                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2859                 elr->lr_next_group = group + 1;
2860         }
2861         sb_end_write(sb);
2862
2863         return ret;
2864 }
2865
2866 /*
2867  * Remove lr_request from the list_request and free the
2868  * request structure. Should be called with li_list_mtx held
2869  */
2870 static void ext4_remove_li_request(struct ext4_li_request *elr)
2871 {
2872         struct ext4_sb_info *sbi;
2873
2874         if (!elr)
2875                 return;
2876
2877         sbi = elr->lr_sbi;
2878
2879         list_del(&elr->lr_request);
2880         sbi->s_li_request = NULL;
2881         kfree(elr);
2882 }
2883
2884 static void ext4_unregister_li_request(struct super_block *sb)
2885 {
2886         mutex_lock(&ext4_li_mtx);
2887         if (!ext4_li_info) {
2888                 mutex_unlock(&ext4_li_mtx);
2889                 return;
2890         }
2891
2892         mutex_lock(&ext4_li_info->li_list_mtx);
2893         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2894         mutex_unlock(&ext4_li_info->li_list_mtx);
2895         mutex_unlock(&ext4_li_mtx);
2896 }
2897
2898 static struct task_struct *ext4_lazyinit_task;
2899
2900 /*
2901  * This is the function where ext4lazyinit thread lives. It walks
2902  * through the request list searching for next scheduled filesystem.
2903  * When such a fs is found, run the lazy initialization request
2904  * (ext4_rn_li_request) and keep track of the time spend in this
2905  * function. Based on that time we compute next schedule time of
2906  * the request. When walking through the list is complete, compute
2907  * next waking time and put itself into sleep.
2908  */
2909 static int ext4_lazyinit_thread(void *arg)
2910 {
2911         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2912         struct list_head *pos, *n;
2913         struct ext4_li_request *elr;
2914         unsigned long next_wakeup, cur;
2915
2916         BUG_ON(NULL == eli);
2917
2918 cont_thread:
2919         while (true) {
2920                 next_wakeup = MAX_JIFFY_OFFSET;
2921
2922                 mutex_lock(&eli->li_list_mtx);
2923                 if (list_empty(&eli->li_request_list)) {
2924                         mutex_unlock(&eli->li_list_mtx);
2925                         goto exit_thread;
2926                 }
2927
2928                 list_for_each_safe(pos, n, &eli->li_request_list) {
2929                         elr = list_entry(pos, struct ext4_li_request,
2930                                          lr_request);
2931
2932                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2933                                 if (ext4_run_li_request(elr) != 0) {
2934                                         /* error, remove the lazy_init job */
2935                                         ext4_remove_li_request(elr);
2936                                         continue;
2937                                 }
2938                         }
2939
2940                         if (time_before(elr->lr_next_sched, next_wakeup))
2941                                 next_wakeup = elr->lr_next_sched;
2942                 }
2943                 mutex_unlock(&eli->li_list_mtx);
2944
2945                 try_to_freeze();
2946
2947                 cur = jiffies;
2948                 if ((time_after_eq(cur, next_wakeup)) ||
2949                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2950                         cond_resched();
2951                         continue;
2952                 }
2953
2954                 schedule_timeout_interruptible(next_wakeup - cur);
2955
2956                 if (kthread_should_stop()) {
2957                         ext4_clear_request_list();
2958                         goto exit_thread;
2959                 }
2960         }
2961
2962 exit_thread:
2963         /*
2964          * It looks like the request list is empty, but we need
2965          * to check it under the li_list_mtx lock, to prevent any
2966          * additions into it, and of course we should lock ext4_li_mtx
2967          * to atomically free the list and ext4_li_info, because at
2968          * this point another ext4 filesystem could be registering
2969          * new one.
2970          */
2971         mutex_lock(&ext4_li_mtx);
2972         mutex_lock(&eli->li_list_mtx);
2973         if (!list_empty(&eli->li_request_list)) {
2974                 mutex_unlock(&eli->li_list_mtx);
2975                 mutex_unlock(&ext4_li_mtx);
2976                 goto cont_thread;
2977         }
2978         mutex_unlock(&eli->li_list_mtx);
2979         kfree(ext4_li_info);
2980         ext4_li_info = NULL;
2981         mutex_unlock(&ext4_li_mtx);
2982
2983         return 0;
2984 }
2985
2986 static void ext4_clear_request_list(void)
2987 {
2988         struct list_head *pos, *n;
2989         struct ext4_li_request *elr;
2990
2991         mutex_lock(&ext4_li_info->li_list_mtx);
2992         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2993                 elr = list_entry(pos, struct ext4_li_request,
2994                                  lr_request);
2995                 ext4_remove_li_request(elr);
2996         }
2997         mutex_unlock(&ext4_li_info->li_list_mtx);
2998 }
2999
3000 static int ext4_run_lazyinit_thread(void)
3001 {
3002         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3003                                          ext4_li_info, "ext4lazyinit");
3004         if (IS_ERR(ext4_lazyinit_task)) {
3005                 int err = PTR_ERR(ext4_lazyinit_task);
3006                 ext4_clear_request_list();
3007                 kfree(ext4_li_info);
3008                 ext4_li_info = NULL;
3009                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3010                                  "initialization thread\n",
3011                                  err);
3012                 return err;
3013         }
3014         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3015         return 0;
3016 }
3017
3018 /*
3019  * Check whether it make sense to run itable init. thread or not.
3020  * If there is at least one uninitialized inode table, return
3021  * corresponding group number, else the loop goes through all
3022  * groups and return total number of groups.
3023  */
3024 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3025 {
3026         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3027         struct ext4_group_desc *gdp = NULL;
3028
3029         for (group = 0; group < ngroups; group++) {
3030                 gdp = ext4_get_group_desc(sb, group, NULL);
3031                 if (!gdp)
3032                         continue;
3033
3034                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3035                         break;
3036         }
3037
3038         return group;
3039 }
3040
3041 static int ext4_li_info_new(void)
3042 {
3043         struct ext4_lazy_init *eli = NULL;
3044
3045         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3046         if (!eli)
3047                 return -ENOMEM;
3048
3049         INIT_LIST_HEAD(&eli->li_request_list);
3050         mutex_init(&eli->li_list_mtx);
3051
3052         eli->li_state |= EXT4_LAZYINIT_QUIT;
3053
3054         ext4_li_info = eli;
3055
3056         return 0;
3057 }
3058
3059 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3060                                             ext4_group_t start)
3061 {
3062         struct ext4_sb_info *sbi = EXT4_SB(sb);
3063         struct ext4_li_request *elr;
3064
3065         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3066         if (!elr)
3067                 return NULL;
3068
3069         elr->lr_super = sb;
3070         elr->lr_sbi = sbi;
3071         elr->lr_next_group = start;
3072
3073         /*
3074          * Randomize first schedule time of the request to
3075          * spread the inode table initialization requests
3076          * better.
3077          */
3078         elr->lr_next_sched = jiffies + (prandom_u32() %
3079                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3080         return elr;
3081 }
3082
3083 int ext4_register_li_request(struct super_block *sb,
3084                              ext4_group_t first_not_zeroed)
3085 {
3086         struct ext4_sb_info *sbi = EXT4_SB(sb);
3087         struct ext4_li_request *elr = NULL;
3088         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3089         int ret = 0;
3090
3091         mutex_lock(&ext4_li_mtx);
3092         if (sbi->s_li_request != NULL) {
3093                 /*
3094                  * Reset timeout so it can be computed again, because
3095                  * s_li_wait_mult might have changed.
3096                  */
3097                 sbi->s_li_request->lr_timeout = 0;
3098                 goto out;
3099         }
3100
3101         if (first_not_zeroed == ngroups ||
3102             (sb->s_flags & MS_RDONLY) ||
3103             !test_opt(sb, INIT_INODE_TABLE))
3104                 goto out;
3105
3106         elr = ext4_li_request_new(sb, first_not_zeroed);
3107         if (!elr) {
3108                 ret = -ENOMEM;
3109                 goto out;
3110         }
3111
3112         if (NULL == ext4_li_info) {
3113                 ret = ext4_li_info_new();
3114                 if (ret)
3115                         goto out;
3116         }
3117
3118         mutex_lock(&ext4_li_info->li_list_mtx);
3119         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3120         mutex_unlock(&ext4_li_info->li_list_mtx);
3121
3122         sbi->s_li_request = elr;
3123         /*
3124          * set elr to NULL here since it has been inserted to
3125          * the request_list and the removal and free of it is
3126          * handled by ext4_clear_request_list from now on.
3127          */
3128         elr = NULL;
3129
3130         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3131                 ret = ext4_run_lazyinit_thread();
3132                 if (ret)
3133                         goto out;
3134         }
3135 out:
3136         mutex_unlock(&ext4_li_mtx);
3137         if (ret)
3138                 kfree(elr);
3139         return ret;
3140 }
3141
3142 /*
3143  * We do not need to lock anything since this is called on
3144  * module unload.
3145  */
3146 static void ext4_destroy_lazyinit_thread(void)
3147 {
3148         /*
3149          * If thread exited earlier
3150          * there's nothing to be done.
3151          */
3152         if (!ext4_li_info || !ext4_lazyinit_task)
3153                 return;
3154
3155         kthread_stop(ext4_lazyinit_task);
3156 }
3157
3158 static int set_journal_csum_feature_set(struct super_block *sb)
3159 {
3160         int ret = 1;
3161         int compat, incompat;
3162         struct ext4_sb_info *sbi = EXT4_SB(sb);
3163
3164         if (ext4_has_metadata_csum(sb)) {
3165                 /* journal checksum v3 */
3166                 compat = 0;
3167                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3168         } else {
3169                 /* journal checksum v1 */
3170                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3171                 incompat = 0;
3172         }
3173
3174         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3175                 ret = jbd2_journal_set_features(sbi->s_journal,
3176                                 compat, 0,
3177                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3178                                 incompat);
3179         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3180                 ret = jbd2_journal_set_features(sbi->s_journal,
3181                                 compat, 0,
3182                                 incompat);
3183                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3184                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3185         } else {
3186                 jbd2_journal_clear_features(sbi->s_journal,
3187                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3188                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3189                                 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3190                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3191         }
3192
3193         return ret;
3194 }
3195
3196 /*
3197  * Note: calculating the overhead so we can be compatible with
3198  * historical BSD practice is quite difficult in the face of
3199  * clusters/bigalloc.  This is because multiple metadata blocks from
3200  * different block group can end up in the same allocation cluster.
3201  * Calculating the exact overhead in the face of clustered allocation
3202  * requires either O(all block bitmaps) in memory or O(number of block
3203  * groups**2) in time.  We will still calculate the superblock for
3204  * older file systems --- and if we come across with a bigalloc file
3205  * system with zero in s_overhead_clusters the estimate will be close to
3206  * correct especially for very large cluster sizes --- but for newer
3207  * file systems, it's better to calculate this figure once at mkfs
3208  * time, and store it in the superblock.  If the superblock value is
3209  * present (even for non-bigalloc file systems), we will use it.
3210  */
3211 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3212                           char *buf)
3213 {
3214         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3215         struct ext4_group_desc  *gdp;
3216         ext4_fsblk_t            first_block, last_block, b;
3217         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3218         int                     s, j, count = 0;
3219
3220         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3221                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3222                         sbi->s_itb_per_group + 2);
3223
3224         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3225                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3226         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3227         for (i = 0; i < ngroups; i++) {
3228                 gdp = ext4_get_group_desc(sb, i, NULL);
3229                 b = ext4_block_bitmap(sb, gdp);
3230                 if (b >= first_block && b <= last_block) {
3231                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3232                         count++;
3233                 }
3234                 b = ext4_inode_bitmap(sb, gdp);
3235                 if (b >= first_block && b <= last_block) {
3236                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3237                         count++;
3238                 }
3239                 b = ext4_inode_table(sb, gdp);
3240                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3241                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3242                                 int c = EXT4_B2C(sbi, b - first_block);
3243                                 ext4_set_bit(c, buf);
3244                                 count++;
3245                         }
3246                 if (i != grp)
3247                         continue;
3248                 s = 0;
3249                 if (ext4_bg_has_super(sb, grp)) {
3250                         ext4_set_bit(s++, buf);
3251                         count++;
3252                 }
3253                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3254                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3255                         count++;
3256                 }
3257         }
3258         if (!count)
3259                 return 0;
3260         return EXT4_CLUSTERS_PER_GROUP(sb) -
3261                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3262 }
3263
3264 /*
3265  * Compute the overhead and stash it in sbi->s_overhead
3266  */
3267 int ext4_calculate_overhead(struct super_block *sb)
3268 {
3269         struct ext4_sb_info *sbi = EXT4_SB(sb);
3270         struct ext4_super_block *es = sbi->s_es;
3271         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3272         ext4_fsblk_t overhead = 0;
3273         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3274
3275         if (!buf)
3276                 return -ENOMEM;
3277
3278         /*
3279          * Compute the overhead (FS structures).  This is constant
3280          * for a given filesystem unless the number of block groups
3281          * changes so we cache the previous value until it does.
3282          */
3283
3284         /*
3285          * All of the blocks before first_data_block are overhead
3286          */
3287         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3288
3289         /*
3290          * Add the overhead found in each block group
3291          */
3292         for (i = 0; i < ngroups; i++) {
3293                 int blks;
3294
3295                 blks = count_overhead(sb, i, buf);
3296                 overhead += blks;
3297                 if (blks)
3298                         memset(buf, 0, PAGE_SIZE);
3299                 cond_resched();
3300         }
3301         /* Add the journal blocks as well */
3302         if (sbi->s_journal)
3303                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3304
3305         sbi->s_overhead = overhead;
3306         smp_wmb();
3307         free_page((unsigned long) buf);
3308         return 0;
3309 }
3310
3311
3312 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3313 {
3314         ext4_fsblk_t resv_clusters;
3315
3316         /*
3317          * There's no need to reserve anything when we aren't using extents.
3318          * The space estimates are exact, there are no unwritten extents,
3319          * hole punching doesn't need new metadata... This is needed especially
3320          * to keep ext2/3 backward compatibility.
3321          */
3322         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3323                 return 0;
3324         /*
3325          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3326          * This should cover the situations where we can not afford to run
3327          * out of space like for example punch hole, or converting
3328          * uninitialized extents in delalloc path. In most cases such
3329          * allocation would require 1, or 2 blocks, higher numbers are
3330          * very rare.
3331          */
3332         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3333                         EXT4_SB(sb)->s_cluster_bits;
3334
3335         do_div(resv_clusters, 50);
3336         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3337
3338         return resv_clusters;
3339 }
3340
3341
3342 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3343 {
3344         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3345                                 sbi->s_cluster_bits;
3346
3347         if (count >= clusters)
3348                 return -EINVAL;
3349
3350         atomic64_set(&sbi->s_resv_clusters, count);
3351         return 0;
3352 }
3353
3354 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3355 {
3356         char *orig_data = kstrdup(data, GFP_KERNEL);
3357         struct buffer_head *bh;
3358         struct ext4_super_block *es = NULL;
3359         struct ext4_sb_info *sbi;
3360         ext4_fsblk_t block;
3361         ext4_fsblk_t sb_block = get_sb_block(&data);
3362         ext4_fsblk_t logical_sb_block;
3363         unsigned long offset = 0;
3364         unsigned long journal_devnum = 0;
3365         unsigned long def_mount_opts;
3366         struct inode *root;
3367         char *cp;
3368         const char *descr;
3369         int ret = -ENOMEM;
3370         int blocksize, clustersize;
3371         unsigned int db_count;
3372         unsigned int i;
3373         int needs_recovery, has_huge_files, has_bigalloc;
3374         __u64 blocks_count;
3375         int err = 0;
3376         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3377         ext4_group_t first_not_zeroed;
3378
3379         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3380         if (!sbi)
3381                 goto out_free_orig;
3382
3383         sbi->s_blockgroup_lock =
3384                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3385         if (!sbi->s_blockgroup_lock) {
3386                 kfree(sbi);
3387                 goto out_free_orig;
3388         }
3389         sb->s_fs_info = sbi;
3390         sbi->s_sb = sb;
3391         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3392         sbi->s_sb_block = sb_block;
3393         if (sb->s_bdev->bd_part)
3394                 sbi->s_sectors_written_start =
3395                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3396
3397         /* Cleanup superblock name */
3398         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3399                 *cp = '!';
3400
3401         /* -EINVAL is default */
3402         ret = -EINVAL;
3403         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3404         if (!blocksize) {
3405                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3406                 goto out_fail;
3407         }
3408
3409         /*
3410          * The ext4 superblock will not be buffer aligned for other than 1kB
3411          * block sizes.  We need to calculate the offset from buffer start.
3412          */
3413         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3414                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3415                 offset = do_div(logical_sb_block, blocksize);
3416         } else {
3417                 logical_sb_block = sb_block;
3418         }
3419
3420         if (!(bh = sb_bread(sb, logical_sb_block))) {
3421                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3422                 goto out_fail;
3423         }
3424         /*
3425          * Note: s_es must be initialized as soon as possible because
3426          *       some ext4 macro-instructions depend on its value
3427          */
3428         es = (struct ext4_super_block *) (bh->b_data + offset);
3429         sbi->s_es = es;
3430         sb->s_magic = le16_to_cpu(es->s_magic);
3431         if (sb->s_magic != EXT4_SUPER_MAGIC)
3432                 goto cantfind_ext4;
3433         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3434
3435         /* Warn if metadata_csum and gdt_csum are both set. */
3436         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3437                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3438             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3439                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3440                              "redundant flags; please run fsck.");
3441
3442         /* Check for a known checksum algorithm */
3443         if (!ext4_verify_csum_type(sb, es)) {
3444                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3445                          "unknown checksum algorithm.");
3446                 silent = 1;
3447                 goto cantfind_ext4;
3448         }
3449
3450         /* Load the checksum driver */
3451         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3452                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3453                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3454                 if (IS_ERR(sbi->s_chksum_driver)) {
3455                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3456                         ret = PTR_ERR(sbi->s_chksum_driver);
3457                         sbi->s_chksum_driver = NULL;
3458                         goto failed_mount;
3459                 }
3460         }
3461
3462         /* Check superblock checksum */
3463         if (!ext4_superblock_csum_verify(sb, es)) {
3464                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3465                          "invalid superblock checksum.  Run e2fsck?");
3466                 silent = 1;
3467                 goto cantfind_ext4;
3468         }
3469
3470         /* Precompute checksum seed for all metadata */
3471         if (ext4_has_metadata_csum(sb))
3472                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3473                                                sizeof(es->s_uuid));
3474
3475         /* Set defaults before we parse the mount options */
3476         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3477         set_opt(sb, INIT_INODE_TABLE);
3478         if (def_mount_opts & EXT4_DEFM_DEBUG)
3479                 set_opt(sb, DEBUG);
3480         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3481                 set_opt(sb, GRPID);
3482         if (def_mount_opts & EXT4_DEFM_UID16)
3483                 set_opt(sb, NO_UID32);
3484         /* xattr user namespace & acls are now defaulted on */
3485         set_opt(sb, XATTR_USER);
3486 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3487         set_opt(sb, POSIX_ACL);
3488 #endif
3489         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3490         if (ext4_has_metadata_csum(sb))
3491                 set_opt(sb, JOURNAL_CHECKSUM);
3492
3493         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3494                 set_opt(sb, JOURNAL_DATA);
3495         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3496                 set_opt(sb, ORDERED_DATA);
3497         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3498                 set_opt(sb, WRITEBACK_DATA);
3499
3500         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3501                 set_opt(sb, ERRORS_PANIC);
3502         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3503                 set_opt(sb, ERRORS_CONT);
3504         else
3505                 set_opt(sb, ERRORS_RO);
3506         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3507                 set_opt(sb, BLOCK_VALIDITY);
3508         if (def_mount_opts & EXT4_DEFM_DISCARD)
3509                 set_opt(sb, DISCARD);
3510
3511         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3512         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3513         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3514         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3515         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3516
3517         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3518                 set_opt(sb, BARRIER);
3519
3520         /*
3521          * enable delayed allocation by default
3522          * Use -o nodelalloc to turn it off
3523          */
3524         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3525             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3526                 set_opt(sb, DELALLOC);
3527
3528         /*
3529          * set default s_li_wait_mult for lazyinit, for the case there is
3530          * no mount option specified.
3531          */
3532         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3533
3534         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3535                            &journal_devnum, &journal_ioprio, 0)) {
3536                 ext4_msg(sb, KERN_WARNING,
3537                          "failed to parse options in superblock: %s",
3538                          sbi->s_es->s_mount_opts);
3539         }
3540         sbi->s_def_mount_opt = sbi->s_mount_opt;
3541         if (!parse_options((char *) data, sb, &journal_devnum,
3542                            &journal_ioprio, 0))
3543                 goto failed_mount;
3544
3545         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3546                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3547                             "with data=journal disables delayed "
3548                             "allocation and O_DIRECT support!\n");
3549                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3550                         ext4_msg(sb, KERN_ERR, "can't mount with "
3551                                  "both data=journal and delalloc");
3552                         goto failed_mount;
3553                 }
3554                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3555                         ext4_msg(sb, KERN_ERR, "can't mount with "
3556                                  "both data=journal and dioread_nolock");
3557                         goto failed_mount;
3558                 }
3559                 if (test_opt(sb, DELALLOC))
3560                         clear_opt(sb, DELALLOC);
3561         }
3562
3563         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3564                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3565
3566         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3567             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3568              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3569              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3570                 ext4_msg(sb, KERN_WARNING,
3571                        "feature flags set on rev 0 fs, "
3572                        "running e2fsck is recommended");
3573
3574         if (IS_EXT2_SB(sb)) {
3575                 if (ext2_feature_set_ok(sb))
3576                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3577                                  "using the ext4 subsystem");
3578                 else {
3579                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3580                                  "to feature incompatibilities");
3581                         goto failed_mount;
3582                 }
3583         }
3584
3585         if (IS_EXT3_SB(sb)) {
3586                 if (ext3_feature_set_ok(sb))
3587                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3588                                  "using the ext4 subsystem");
3589                 else {
3590                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3591                                  "to feature incompatibilities");
3592                         goto failed_mount;
3593                 }
3594         }
3595
3596         /*
3597          * Check feature flags regardless of the revision level, since we
3598          * previously didn't change the revision level when setting the flags,
3599          * so there is a chance incompat flags are set on a rev 0 filesystem.
3600          */
3601         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3602                 goto failed_mount;
3603
3604         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3605         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3606             blocksize > EXT4_MAX_BLOCK_SIZE) {
3607                 ext4_msg(sb, KERN_ERR,
3608                        "Unsupported filesystem blocksize %d", blocksize);
3609                 goto failed_mount;
3610         }
3611
3612         if (sb->s_blocksize != blocksize) {
3613                 /* Validate the filesystem blocksize */
3614                 if (!sb_set_blocksize(sb, blocksize)) {
3615                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3616                                         blocksize);
3617                         goto failed_mount;
3618                 }
3619
3620                 brelse(bh);
3621                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3622                 offset = do_div(logical_sb_block, blocksize);
3623                 bh = sb_bread(sb, logical_sb_block);
3624                 if (!bh) {
3625                         ext4_msg(sb, KERN_ERR,
3626                                "Can't read superblock on 2nd try");
3627                         goto failed_mount;
3628                 }
3629                 es = (struct ext4_super_block *)(bh->b_data + offset);
3630                 sbi->s_es = es;
3631                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3632                         ext4_msg(sb, KERN_ERR,
3633                                "Magic mismatch, very weird!");
3634                         goto failed_mount;
3635                 }
3636         }
3637
3638         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3639                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3640         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3641                                                       has_huge_files);
3642         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3643
3644         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3645                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3646                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3647         } else {
3648                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3649                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3650                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3651                     (!is_power_of_2(sbi->s_inode_size)) ||
3652                     (sbi->s_inode_size > blocksize)) {
3653                         ext4_msg(sb, KERN_ERR,
3654                                "unsupported inode size: %d",
3655                                sbi->s_inode_size);
3656                         goto failed_mount;
3657                 }
3658                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3659                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3660         }
3661
3662         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3663         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3664                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3665                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3666                     !is_power_of_2(sbi->s_desc_size)) {
3667                         ext4_msg(sb, KERN_ERR,
3668                                "unsupported descriptor size %lu",
3669                                sbi->s_desc_size);
3670                         goto failed_mount;
3671                 }
3672         } else
3673                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3674
3675         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3676         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3677         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3678                 goto cantfind_ext4;
3679
3680         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3681         if (sbi->s_inodes_per_block == 0)
3682                 goto cantfind_ext4;
3683         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3684                                         sbi->s_inodes_per_block;
3685         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3686         sbi->s_sbh = bh;
3687         sbi->s_mount_state = le16_to_cpu(es->s_state);
3688         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3689         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3690
3691         for (i = 0; i < 4; i++)
3692                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3693         sbi->s_def_hash_version = es->s_def_hash_version;
3694         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3695                 i = le32_to_cpu(es->s_flags);
3696                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3697                         sbi->s_hash_unsigned = 3;
3698                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3699 #ifdef __CHAR_UNSIGNED__
3700                         if (!(sb->s_flags & MS_RDONLY))
3701                                 es->s_flags |=
3702                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3703                         sbi->s_hash_unsigned = 3;
3704 #else
3705                         if (!(sb->s_flags & MS_RDONLY))
3706                                 es->s_flags |=
3707                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3708 #endif
3709                 }
3710         }
3711
3712         /* Handle clustersize */
3713         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3714         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3715                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3716         if (has_bigalloc) {
3717                 if (clustersize < blocksize) {
3718                         ext4_msg(sb, KERN_ERR,
3719                                  "cluster size (%d) smaller than "
3720                                  "block size (%d)", clustersize, blocksize);
3721                         goto failed_mount;
3722                 }
3723                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3724                         le32_to_cpu(es->s_log_block_size);
3725                 sbi->s_clusters_per_group =
3726                         le32_to_cpu(es->s_clusters_per_group);
3727                 if (sbi->s_clusters_per_group > blocksize * 8) {
3728                         ext4_msg(sb, KERN_ERR,
3729                                  "#clusters per group too big: %lu",
3730                                  sbi->s_clusters_per_group);
3731                         goto failed_mount;
3732                 }
3733                 if (sbi->s_blocks_per_group !=
3734                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3735                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3736                                  "clusters per group (%lu) inconsistent",
3737                                  sbi->s_blocks_per_group,
3738                                  sbi->s_clusters_per_group);
3739                         goto failed_mount;
3740                 }
3741         } else {
3742                 if (clustersize != blocksize) {
3743                         ext4_warning(sb, "fragment/cluster size (%d) != "
3744                                      "block size (%d)", clustersize,
3745                                      blocksize);
3746                         clustersize = blocksize;
3747                 }
3748                 if (sbi->s_blocks_per_group > blocksize * 8) {
3749                         ext4_msg(sb, KERN_ERR,
3750                                  "#blocks per group too big: %lu",
3751                                  sbi->s_blocks_per_group);
3752                         goto failed_mount;
3753                 }
3754                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3755                 sbi->s_cluster_bits = 0;
3756         }
3757         sbi->s_cluster_ratio = clustersize / blocksize;
3758
3759         if (sbi->s_inodes_per_group > blocksize * 8) {
3760                 ext4_msg(sb, KERN_ERR,
3761                        "#inodes per group too big: %lu",
3762                        sbi->s_inodes_per_group);
3763                 goto failed_mount;
3764         }
3765
3766         /* Do we have standard group size of clustersize * 8 blocks ? */
3767         if (sbi->s_blocks_per_group == clustersize << 3)
3768                 set_opt2(sb, STD_GROUP_SIZE);
3769
3770         /*
3771          * Test whether we have more sectors than will fit in sector_t,
3772          * and whether the max offset is addressable by the page cache.
3773          */
3774         err = generic_check_addressable(sb->s_blocksize_bits,
3775                                         ext4_blocks_count(es));
3776         if (err) {
3777                 ext4_msg(sb, KERN_ERR, "filesystem"
3778                          " too large to mount safely on this system");
3779                 if (sizeof(sector_t) < 8)
3780                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3781                 goto failed_mount;
3782         }
3783
3784         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3785                 goto cantfind_ext4;
3786
3787         /* check blocks count against device size */
3788         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3789         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3790                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3791                        "exceeds size of device (%llu blocks)",
3792                        ext4_blocks_count(es), blocks_count);
3793                 goto failed_mount;
3794         }
3795
3796         /*
3797          * It makes no sense for the first data block to be beyond the end
3798          * of the filesystem.
3799          */
3800         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3801                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3802                          "block %u is beyond end of filesystem (%llu)",
3803                          le32_to_cpu(es->s_first_data_block),
3804                          ext4_blocks_count(es));
3805                 goto failed_mount;
3806         }
3807         blocks_count = (ext4_blocks_count(es) -
3808                         le32_to_cpu(es->s_first_data_block) +
3809                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3810         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3811         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3812                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3813                        "(block count %llu, first data block %u, "
3814                        "blocks per group %lu)", sbi->s_groups_count,
3815                        ext4_blocks_count(es),
3816                        le32_to_cpu(es->s_first_data_block),
3817                        EXT4_BLOCKS_PER_GROUP(sb));
3818                 goto failed_mount;
3819         }
3820         sbi->s_groups_count = blocks_count;
3821         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3822                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3823         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3824                    EXT4_DESC_PER_BLOCK(sb);
3825         sbi->s_group_desc = ext4_kvmalloc(db_count *
3826                                           sizeof(struct buffer_head *),
3827                                           GFP_KERNEL);
3828         if (sbi->s_group_desc == NULL) {
3829                 ext4_msg(sb, KERN_ERR, "not enough memory");
3830                 ret = -ENOMEM;
3831                 goto failed_mount;
3832         }
3833
3834         if (ext4_proc_root)
3835                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3836
3837         if (sbi->s_proc)
3838                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3839                                  &ext4_seq_options_fops, sb);
3840
3841         bgl_lock_init(sbi->s_blockgroup_lock);
3842
3843         for (i = 0; i < db_count; i++) {
3844                 block = descriptor_loc(sb, logical_sb_block, i);
3845                 sbi->s_group_desc[i] = sb_bread(sb, block);
3846                 if (!sbi->s_group_desc[i]) {
3847                         ext4_msg(sb, KERN_ERR,
3848                                "can't read group descriptor %d", i);
3849                         db_count = i;
3850                         goto failed_mount2;
3851                 }
3852         }
3853         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3854                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3855                 goto failed_mount2;
3856         }
3857         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3858                 if (!ext4_fill_flex_info(sb)) {
3859                         ext4_msg(sb, KERN_ERR,
3860                                "unable to initialize "
3861                                "flex_bg meta info!");
3862                         goto failed_mount2;
3863                 }
3864
3865         sbi->s_gdb_count = db_count;
3866         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3867         spin_lock_init(&sbi->s_next_gen_lock);
3868
3869         init_timer(&sbi->s_err_report);
3870         sbi->s_err_report.function = print_daily_error_info;
3871         sbi->s_err_report.data = (unsigned long) sb;
3872
3873         /* Register extent status tree shrinker */
3874         ext4_es_register_shrinker(sbi);
3875
3876         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3877                         ext4_count_free_clusters(sb));
3878         if (!err) {
3879                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3880                                 ext4_count_free_inodes(sb));
3881         }
3882         if (!err) {
3883                 err = percpu_counter_init(&sbi->s_dirs_counter,
3884                                 ext4_count_dirs(sb));
3885         }
3886         if (!err) {
3887                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3888         }
3889         if (!err) {
3890                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3891         }
3892         if (err) {
3893                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3894                 goto failed_mount3;
3895         }
3896
3897         sbi->s_stripe = ext4_get_stripe_size(sbi);
3898         sbi->s_extent_max_zeroout_kb = 32;
3899
3900         /*
3901          * set up enough so that it can read an inode
3902          */
3903         if (!test_opt(sb, NOLOAD) &&
3904             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3905                 sb->s_op = &ext4_sops;
3906         else
3907                 sb->s_op = &ext4_nojournal_sops;
3908         sb->s_export_op = &ext4_export_ops;
3909         sb->s_xattr = ext4_xattr_handlers;
3910 #ifdef CONFIG_QUOTA
3911         sb->dq_op = &ext4_quota_operations;
3912         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3913                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3914         else
3915                 sb->s_qcop = &ext4_qctl_operations;
3916 #endif
3917         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3918
3919         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3920         mutex_init(&sbi->s_orphan_lock);
3921
3922         sb->s_root = NULL;
3923
3924         needs_recovery = (es->s_last_orphan != 0 ||
3925                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3926                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3927
3928         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3929             !(sb->s_flags & MS_RDONLY))
3930                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3931                         goto failed_mount3;
3932
3933         /*
3934          * The first inode we look at is the journal inode.  Don't try
3935          * root first: it may be modified in the journal!
3936          */
3937         if (!test_opt(sb, NOLOAD) &&
3938             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3939                 if (ext4_load_journal(sb, es, journal_devnum))
3940                         goto failed_mount3;
3941         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3942               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3943                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3944                        "suppressed and not mounted read-only");
3945                 goto failed_mount_wq;
3946         } else {
3947                 clear_opt(sb, DATA_FLAGS);
3948                 sbi->s_journal = NULL;
3949                 needs_recovery = 0;
3950                 goto no_journal;
3951         }
3952
3953         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3954             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3955                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3956                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3957                 goto failed_mount_wq;
3958         }
3959
3960         if (!set_journal_csum_feature_set(sb)) {
3961                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3962                          "feature set");
3963                 goto failed_mount_wq;
3964         }
3965
3966         /* We have now updated the journal if required, so we can
3967          * validate the data journaling mode. */
3968         switch (test_opt(sb, DATA_FLAGS)) {
3969         case 0:
3970                 /* No mode set, assume a default based on the journal
3971                  * capabilities: ORDERED_DATA if the journal can
3972                  * cope, else JOURNAL_DATA
3973                  */
3974                 if (jbd2_journal_check_available_features
3975                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3976                         set_opt(sb, ORDERED_DATA);
3977                 else
3978                         set_opt(sb, JOURNAL_DATA);
3979                 break;
3980
3981         case EXT4_MOUNT_ORDERED_DATA:
3982         case EXT4_MOUNT_WRITEBACK_DATA:
3983                 if (!jbd2_journal_check_available_features
3984                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3985                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3986                                "requested data journaling mode");
3987                         goto failed_mount_wq;
3988                 }
3989         default:
3990                 break;
3991         }
3992         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3993
3994         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3995
3996         /*
3997          * The journal may have updated the bg summary counts, so we
3998          * need to update the global counters.
3999          */
4000         percpu_counter_set(&sbi->s_freeclusters_counter,
4001                            ext4_count_free_clusters(sb));
4002         percpu_counter_set(&sbi->s_freeinodes_counter,
4003                            ext4_count_free_inodes(sb));
4004         percpu_counter_set(&sbi->s_dirs_counter,
4005                            ext4_count_dirs(sb));
4006         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
4007
4008 no_journal:
4009         /*
4010          * Get the # of file system overhead blocks from the
4011          * superblock if present.
4012          */
4013         if (es->s_overhead_clusters)
4014                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4015         else {
4016                 err = ext4_calculate_overhead(sb);
4017                 if (err)
4018                         goto failed_mount_wq;
4019         }
4020
4021         /*
4022          * The maximum number of concurrent works can be high and
4023          * concurrency isn't really necessary.  Limit it to 1.
4024          */
4025         EXT4_SB(sb)->rsv_conversion_wq =
4026                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4027         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4028                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4029                 ret = -ENOMEM;
4030                 goto failed_mount4;
4031         }
4032
4033         /*
4034          * The jbd2_journal_load will have done any necessary log recovery,
4035          * so we can safely mount the rest of the filesystem now.
4036          */
4037
4038         root = ext4_iget(sb, EXT4_ROOT_INO);
4039         if (IS_ERR(root)) {
4040                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4041                 ret = PTR_ERR(root);
4042                 root = NULL;
4043                 goto failed_mount4;
4044         }
4045         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4046                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4047                 iput(root);
4048                 goto failed_mount4;
4049         }
4050         sb->s_root = d_make_root(root);
4051         if (!sb->s_root) {
4052                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4053                 ret = -ENOMEM;
4054                 goto failed_mount4;
4055         }
4056
4057         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4058                 sb->s_flags |= MS_RDONLY;
4059
4060         /* determine the minimum size of new large inodes, if present */
4061         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4062                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4063                                                      EXT4_GOOD_OLD_INODE_SIZE;
4064                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4065                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4066                         if (sbi->s_want_extra_isize <
4067                             le16_to_cpu(es->s_want_extra_isize))
4068                                 sbi->s_want_extra_isize =
4069                                         le16_to_cpu(es->s_want_extra_isize);
4070                         if (sbi->s_want_extra_isize <
4071                             le16_to_cpu(es->s_min_extra_isize))
4072                                 sbi->s_want_extra_isize =
4073                                         le16_to_cpu(es->s_min_extra_isize);
4074                 }
4075         }
4076         /* Check if enough inode space is available */
4077         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4078                                                         sbi->s_inode_size) {
4079                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4080                                                        EXT4_GOOD_OLD_INODE_SIZE;
4081                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4082                          "available");
4083         }
4084
4085         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4086         if (err) {
4087                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4088                          "reserved pool", ext4_calculate_resv_clusters(sb));
4089                 goto failed_mount4a;
4090         }
4091
4092         err = ext4_setup_system_zone(sb);
4093         if (err) {
4094                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4095                          "zone (%d)", err);
4096                 goto failed_mount4a;
4097         }
4098
4099         ext4_ext_init(sb);
4100         err = ext4_mb_init(sb);
4101         if (err) {
4102                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4103                          err);
4104                 goto failed_mount5;
4105         }
4106
4107         err = ext4_register_li_request(sb, first_not_zeroed);
4108         if (err)
4109                 goto failed_mount6;
4110
4111         sbi->s_kobj.kset = ext4_kset;
4112         init_completion(&sbi->s_kobj_unregister);
4113         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4114                                    "%s", sb->s_id);
4115         if (err)
4116                 goto failed_mount7;
4117
4118 #ifdef CONFIG_QUOTA
4119         /* Enable quota usage during mount. */
4120         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4121             !(sb->s_flags & MS_RDONLY)) {
4122                 err = ext4_enable_quotas(sb);
4123                 if (err)
4124                         goto failed_mount8;
4125         }
4126 #endif  /* CONFIG_QUOTA */
4127
4128         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4129         ext4_orphan_cleanup(sb, es);
4130         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4131         if (needs_recovery) {
4132                 ext4_msg(sb, KERN_INFO, "recovery complete");
4133                 ext4_mark_recovery_complete(sb, es);
4134         }
4135         if (EXT4_SB(sb)->s_journal) {
4136                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4137                         descr = " journalled data mode";
4138                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4139                         descr = " ordered data mode";
4140                 else
4141                         descr = " writeback data mode";
4142         } else
4143                 descr = "out journal";
4144
4145         if (test_opt(sb, DISCARD)) {
4146                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4147                 if (!blk_queue_discard(q))
4148                         ext4_msg(sb, KERN_WARNING,
4149                                  "mounting with \"discard\" option, but "
4150                                  "the device does not support discard");
4151         }
4152
4153         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4154                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4155                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4156
4157         if (es->s_error_count)
4158                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4159
4160         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4161         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4162         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4163         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4164
4165         kfree(orig_data);
4166         return 0;
4167
4168 cantfind_ext4:
4169         if (!silent)
4170                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4171         goto failed_mount;
4172
4173 #ifdef CONFIG_QUOTA
4174 failed_mount8:
4175         kobject_del(&sbi->s_kobj);
4176 #endif
4177 failed_mount7:
4178         ext4_unregister_li_request(sb);
4179 failed_mount6:
4180         ext4_mb_release(sb);
4181 failed_mount5:
4182         ext4_ext_release(sb);
4183         ext4_release_system_zone(sb);
4184 failed_mount4a:
4185         dput(sb->s_root);
4186         sb->s_root = NULL;
4187 failed_mount4:
4188         ext4_msg(sb, KERN_ERR, "mount failed");
4189         if (EXT4_SB(sb)->rsv_conversion_wq)
4190                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4191 failed_mount_wq:
4192         if (sbi->s_journal) {
4193                 jbd2_journal_destroy(sbi->s_journal);
4194                 sbi->s_journal = NULL;
4195         }
4196 failed_mount3:
4197         ext4_es_unregister_shrinker(sbi);
4198         del_timer_sync(&sbi->s_err_report);
4199         if (sbi->s_flex_groups)
4200                 ext4_kvfree(sbi->s_flex_groups);
4201         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4202         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4203         percpu_counter_destroy(&sbi->s_dirs_counter);
4204         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4205         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4206         if (sbi->s_mmp_tsk)
4207                 kthread_stop(sbi->s_mmp_tsk);
4208 failed_mount2:
4209         for (i = 0; i < db_count; i++)
4210                 brelse(sbi->s_group_desc[i]);
4211         ext4_kvfree(sbi->s_group_desc);
4212 failed_mount:
4213         if (sbi->s_chksum_driver)
4214                 crypto_free_shash(sbi->s_chksum_driver);
4215         if (sbi->s_proc) {
4216                 remove_proc_entry("options", sbi->s_proc);
4217                 remove_proc_entry(sb->s_id, ext4_proc_root);
4218         }
4219 #ifdef CONFIG_QUOTA
4220         for (i = 0; i < MAXQUOTAS; i++)
4221                 kfree(sbi->s_qf_names[i]);
4222 #endif
4223         ext4_blkdev_remove(sbi);
4224         brelse(bh);
4225 out_fail:
4226         sb->s_fs_info = NULL;
4227         kfree(sbi->s_blockgroup_lock);
4228         kfree(sbi);
4229 out_free_orig:
4230         kfree(orig_data);
4231         return err ? err : ret;
4232 }
4233
4234 /*
4235  * Setup any per-fs journal parameters now.  We'll do this both on
4236  * initial mount, once the journal has been initialised but before we've
4237  * done any recovery; and again on any subsequent remount.
4238  */
4239 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4240 {
4241         struct ext4_sb_info *sbi = EXT4_SB(sb);
4242
4243         journal->j_commit_interval = sbi->s_commit_interval;
4244         journal->j_min_batch_time = sbi->s_min_batch_time;
4245         journal->j_max_batch_time = sbi->s_max_batch_time;
4246
4247         write_lock(&journal->j_state_lock);
4248         if (test_opt(sb, BARRIER))
4249                 journal->j_flags |= JBD2_BARRIER;
4250         else
4251                 journal->j_flags &= ~JBD2_BARRIER;
4252         if (test_opt(sb, DATA_ERR_ABORT))
4253                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4254         else
4255                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4256         write_unlock(&journal->j_state_lock);
4257 }
4258
4259 static journal_t *ext4_get_journal(struct super_block *sb,
4260                                    unsigned int journal_inum)
4261 {
4262         struct inode *journal_inode;
4263         journal_t *journal;
4264
4265         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4266
4267         /* First, test for the existence of a valid inode on disk.  Bad
4268          * things happen if we iget() an unused inode, as the subsequent
4269          * iput() will try to delete it. */
4270
4271         journal_inode = ext4_iget(sb, journal_inum);
4272         if (IS_ERR(journal_inode)) {
4273                 ext4_msg(sb, KERN_ERR, "no journal found");
4274                 return NULL;
4275         }
4276         if (!journal_inode->i_nlink) {
4277                 make_bad_inode(journal_inode);
4278                 iput(journal_inode);
4279                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4280                 return NULL;
4281         }
4282
4283         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4284                   journal_inode, journal_inode->i_size);
4285         if (!S_ISREG(journal_inode->i_mode)) {
4286                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4287                 iput(journal_inode);
4288                 return NULL;
4289         }
4290
4291         journal = jbd2_journal_init_inode(journal_inode);
4292         if (!journal) {
4293                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4294                 iput(journal_inode);
4295                 return NULL;
4296         }
4297         journal->j_private = sb;
4298         ext4_init_journal_params(sb, journal);
4299         return journal;
4300 }
4301
4302 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4303                                        dev_t j_dev)
4304 {
4305         struct buffer_head *bh;
4306         journal_t *journal;
4307         ext4_fsblk_t start;
4308         ext4_fsblk_t len;
4309         int hblock, blocksize;
4310         ext4_fsblk_t sb_block;
4311         unsigned long offset;
4312         struct ext4_super_block *es;
4313         struct block_device *bdev;
4314
4315         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4316
4317         bdev = ext4_blkdev_get(j_dev, sb);
4318         if (bdev == NULL)
4319                 return NULL;
4320
4321         blocksize = sb->s_blocksize;
4322         hblock = bdev_logical_block_size(bdev);
4323         if (blocksize < hblock) {
4324                 ext4_msg(sb, KERN_ERR,
4325                         "blocksize too small for journal device");
4326                 goto out_bdev;
4327         }
4328
4329         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4330         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4331         set_blocksize(bdev, blocksize);
4332         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4333                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4334                        "external journal");
4335                 goto out_bdev;
4336         }
4337
4338         es = (struct ext4_super_block *) (bh->b_data + offset);
4339         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4340             !(le32_to_cpu(es->s_feature_incompat) &
4341               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4342                 ext4_msg(sb, KERN_ERR, "external journal has "
4343                                         "bad superblock");
4344                 brelse(bh);
4345                 goto out_bdev;
4346         }
4347
4348         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4349                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4350                 brelse(bh);
4351                 goto out_bdev;
4352         }
4353
4354         len = ext4_blocks_count(es);
4355         start = sb_block + 1;
4356         brelse(bh);     /* we're done with the superblock */
4357
4358         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4359                                         start, len, blocksize);
4360         if (!journal) {
4361                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4362                 goto out_bdev;
4363         }
4364         journal->j_private = sb;
4365         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4366         wait_on_buffer(journal->j_sb_buffer);
4367         if (!buffer_uptodate(journal->j_sb_buffer)) {
4368                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4369                 goto out_journal;
4370         }
4371         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4372                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4373                                         "user (unsupported) - %d",
4374                         be32_to_cpu(journal->j_superblock->s_nr_users));
4375                 goto out_journal;
4376         }
4377         EXT4_SB(sb)->journal_bdev = bdev;
4378         ext4_init_journal_params(sb, journal);
4379         return journal;
4380
4381 out_journal:
4382         jbd2_journal_destroy(journal);
4383 out_bdev:
4384         ext4_blkdev_put(bdev);
4385         return NULL;
4386 }
4387
4388 static int ext4_load_journal(struct super_block *sb,
4389                              struct ext4_super_block *es,
4390                              unsigned long journal_devnum)
4391 {
4392         journal_t *journal;
4393         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4394         dev_t journal_dev;
4395         int err = 0;
4396         int really_read_only;
4397
4398         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4399
4400         if (journal_devnum &&
4401             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4402                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4403                         "numbers have changed");
4404                 journal_dev = new_decode_dev(journal_devnum);
4405         } else
4406                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4407
4408         really_read_only = bdev_read_only(sb->s_bdev);
4409
4410         /*
4411          * Are we loading a blank journal or performing recovery after a
4412          * crash?  For recovery, we need to check in advance whether we
4413          * can get read-write access to the device.
4414          */
4415         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4416                 if (sb->s_flags & MS_RDONLY) {
4417                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4418                                         "required on readonly filesystem");
4419                         if (really_read_only) {
4420                                 ext4_msg(sb, KERN_ERR, "write access "
4421                                         "unavailable, cannot proceed");
4422                                 return -EROFS;
4423                         }
4424                         ext4_msg(sb, KERN_INFO, "write access will "
4425                                "be enabled during recovery");
4426                 }
4427         }
4428
4429         if (journal_inum && journal_dev) {
4430                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4431                        "and inode journals!");
4432                 return -EINVAL;
4433         }
4434
4435         if (journal_inum) {
4436                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4437                         return -EINVAL;
4438         } else {
4439                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4440                         return -EINVAL;
4441         }
4442
4443         if (!(journal->j_flags & JBD2_BARRIER))
4444                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4445
4446         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4447                 err = jbd2_journal_wipe(journal, !really_read_only);
4448         if (!err) {
4449                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4450                 if (save)
4451                         memcpy(save, ((char *) es) +
4452                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4453                 err = jbd2_journal_load(journal);
4454                 if (save)
4455                         memcpy(((char *) es) + EXT4_S_ERR_START,
4456                                save, EXT4_S_ERR_LEN);
4457                 kfree(save);
4458         }
4459
4460         if (err) {
4461                 ext4_msg(sb, KERN_ERR, "error loading journal");
4462                 jbd2_journal_destroy(journal);
4463                 return err;
4464         }
4465
4466         EXT4_SB(sb)->s_journal = journal;
4467         ext4_clear_journal_err(sb, es);
4468
4469         if (!really_read_only && journal_devnum &&
4470             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4471                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4472
4473                 /* Make sure we flush the recovery flag to disk. */
4474                 ext4_commit_super(sb, 1);
4475         }
4476
4477         return 0;
4478 }
4479
4480 static int ext4_commit_super(struct super_block *sb, int sync)
4481 {
4482         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4483         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4484         int error = 0;
4485
4486         if (!sbh || block_device_ejected(sb))
4487                 return error;
4488         if (buffer_write_io_error(sbh)) {
4489                 /*
4490                  * Oh, dear.  A previous attempt to write the
4491                  * superblock failed.  This could happen because the
4492                  * USB device was yanked out.  Or it could happen to
4493                  * be a transient write error and maybe the block will
4494                  * be remapped.  Nothing we can do but to retry the
4495                  * write and hope for the best.
4496                  */
4497                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4498                        "superblock detected");
4499                 clear_buffer_write_io_error(sbh);
4500                 set_buffer_uptodate(sbh);
4501         }
4502         /*
4503          * If the file system is mounted read-only, don't update the
4504          * superblock write time.  This avoids updating the superblock
4505          * write time when we are mounting the root file system
4506          * read/only but we need to replay the journal; at that point,
4507          * for people who are east of GMT and who make their clock
4508          * tick in localtime for Windows bug-for-bug compatibility,
4509          * the clock is set in the future, and this will cause e2fsck
4510          * to complain and force a full file system check.
4511          */
4512         if (!(sb->s_flags & MS_RDONLY))
4513                 es->s_wtime = cpu_to_le32(get_seconds());
4514         if (sb->s_bdev->bd_part)
4515                 es->s_kbytes_written =
4516                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4517                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4518                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4519         else
4520                 es->s_kbytes_written =
4521                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4522         ext4_free_blocks_count_set(es,
4523                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4524                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4525         es->s_free_inodes_count =
4526                 cpu_to_le32(percpu_counter_sum_positive(
4527                                 &EXT4_SB(sb)->s_freeinodes_counter));
4528         BUFFER_TRACE(sbh, "marking dirty");
4529         ext4_superblock_csum_set(sb);
4530         mark_buffer_dirty(sbh);
4531         if (sync) {
4532                 error = sync_dirty_buffer(sbh);
4533                 if (error)
4534                         return error;
4535
4536                 error = buffer_write_io_error(sbh);
4537                 if (error) {
4538                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4539                                "superblock");
4540                         clear_buffer_write_io_error(sbh);
4541                         set_buffer_uptodate(sbh);
4542                 }
4543         }
4544         return error;
4545 }
4546
4547 /*
4548  * Have we just finished recovery?  If so, and if we are mounting (or
4549  * remounting) the filesystem readonly, then we will end up with a
4550  * consistent fs on disk.  Record that fact.
4551  */
4552 static void ext4_mark_recovery_complete(struct super_block *sb,
4553                                         struct ext4_super_block *es)
4554 {
4555         journal_t *journal = EXT4_SB(sb)->s_journal;
4556
4557         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4558                 BUG_ON(journal != NULL);
4559                 return;
4560         }
4561         jbd2_journal_lock_updates(journal);
4562         if (jbd2_journal_flush(journal) < 0)
4563                 goto out;
4564
4565         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4566             sb->s_flags & MS_RDONLY) {
4567                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4568                 ext4_commit_super(sb, 1);
4569         }
4570
4571 out:
4572         jbd2_journal_unlock_updates(journal);
4573 }
4574
4575 /*
4576  * If we are mounting (or read-write remounting) a filesystem whose journal
4577  * has recorded an error from a previous lifetime, move that error to the
4578  * main filesystem now.
4579  */
4580 static void ext4_clear_journal_err(struct super_block *sb,
4581                                    struct ext4_super_block *es)
4582 {
4583         journal_t *journal;
4584         int j_errno;
4585         const char *errstr;
4586
4587         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4588
4589         journal = EXT4_SB(sb)->s_journal;
4590
4591         /*
4592          * Now check for any error status which may have been recorded in the
4593          * journal by a prior ext4_error() or ext4_abort()
4594          */
4595
4596         j_errno = jbd2_journal_errno(journal);
4597         if (j_errno) {
4598                 char nbuf[16];
4599
4600                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4601                 ext4_warning(sb, "Filesystem error recorded "
4602                              "from previous mount: %s", errstr);
4603                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4604
4605                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4606                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4607                 ext4_commit_super(sb, 1);
4608
4609                 jbd2_journal_clear_err(journal);
4610                 jbd2_journal_update_sb_errno(journal);
4611         }
4612 }
4613
4614 /*
4615  * Force the running and committing transactions to commit,
4616  * and wait on the commit.
4617  */
4618 int ext4_force_commit(struct super_block *sb)
4619 {
4620         journal_t *journal;
4621
4622         if (sb->s_flags & MS_RDONLY)
4623                 return 0;
4624
4625         journal = EXT4_SB(sb)->s_journal;
4626         return ext4_journal_force_commit(journal);
4627 }
4628
4629 static int ext4_sync_fs(struct super_block *sb, int wait)
4630 {
4631         int ret = 0;
4632         tid_t target;
4633         bool needs_barrier = false;
4634         struct ext4_sb_info *sbi = EXT4_SB(sb);
4635
4636         trace_ext4_sync_fs(sb, wait);
4637         flush_workqueue(sbi->rsv_conversion_wq);
4638         /*
4639          * Writeback quota in non-journalled quota case - journalled quota has
4640          * no dirty dquots
4641          */
4642         dquot_writeback_dquots(sb, -1);
4643         /*
4644          * Data writeback is possible w/o journal transaction, so barrier must
4645          * being sent at the end of the function. But we can skip it if
4646          * transaction_commit will do it for us.
4647          */
4648         target = jbd2_get_latest_transaction(sbi->s_journal);
4649         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4650             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4651                 needs_barrier = true;
4652
4653         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4654                 if (wait)
4655                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4656         }
4657         if (needs_barrier) {
4658                 int err;
4659                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4660                 if (!ret)
4661                         ret = err;
4662         }
4663
4664         return ret;
4665 }
4666
4667 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4668 {
4669         int ret = 0;
4670
4671         trace_ext4_sync_fs(sb, wait);
4672         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4673         dquot_writeback_dquots(sb, -1);
4674         if (wait && test_opt(sb, BARRIER))
4675                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4676
4677         return ret;
4678 }
4679
4680 /*
4681  * LVM calls this function before a (read-only) snapshot is created.  This
4682  * gives us a chance to flush the journal completely and mark the fs clean.
4683  *
4684  * Note that only this function cannot bring a filesystem to be in a clean
4685  * state independently. It relies on upper layer to stop all data & metadata
4686  * modifications.
4687  */
4688 static int ext4_freeze(struct super_block *sb)
4689 {
4690         int error = 0;
4691         journal_t *journal;
4692
4693         if (sb->s_flags & MS_RDONLY)
4694                 return 0;
4695
4696         journal = EXT4_SB(sb)->s_journal;
4697
4698         /* Now we set up the journal barrier. */
4699         jbd2_journal_lock_updates(journal);
4700
4701         /*
4702          * Don't clear the needs_recovery flag if we failed to flush
4703          * the journal.
4704          */
4705         error = jbd2_journal_flush(journal);
4706         if (error < 0)
4707                 goto out;
4708
4709         /* Journal blocked and flushed, clear needs_recovery flag. */
4710         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4711         error = ext4_commit_super(sb, 1);
4712 out:
4713         /* we rely on upper layer to stop further updates */
4714         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4715         return error;
4716 }
4717
4718 /*
4719  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4720  * flag here, even though the filesystem is not technically dirty yet.
4721  */
4722 static int ext4_unfreeze(struct super_block *sb)
4723 {
4724         if (sb->s_flags & MS_RDONLY)
4725                 return 0;
4726
4727         /* Reset the needs_recovery flag before the fs is unlocked. */
4728         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4729         ext4_commit_super(sb, 1);
4730         return 0;
4731 }
4732
4733 /*
4734  * Structure to save mount options for ext4_remount's benefit
4735  */
4736 struct ext4_mount_options {
4737         unsigned long s_mount_opt;
4738         unsigned long s_mount_opt2;
4739         kuid_t s_resuid;
4740         kgid_t s_resgid;
4741         unsigned long s_commit_interval;
4742         u32 s_min_batch_time, s_max_batch_time;
4743 #ifdef CONFIG_QUOTA
4744         int s_jquota_fmt;
4745         char *s_qf_names[MAXQUOTAS];
4746 #endif
4747 };
4748
4749 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4750 {
4751         struct ext4_super_block *es;
4752         struct ext4_sb_info *sbi = EXT4_SB(sb);
4753         unsigned long old_sb_flags;
4754         struct ext4_mount_options old_opts;
4755         int enable_quota = 0;
4756         ext4_group_t g;
4757         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4758         int err = 0;
4759 #ifdef CONFIG_QUOTA
4760         int i, j;
4761 #endif
4762         char *orig_data = kstrdup(data, GFP_KERNEL);
4763
4764         /* Store the original options */
4765         old_sb_flags = sb->s_flags;
4766         old_opts.s_mount_opt = sbi->s_mount_opt;
4767         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4768         old_opts.s_resuid = sbi->s_resuid;
4769         old_opts.s_resgid = sbi->s_resgid;
4770         old_opts.s_commit_interval = sbi->s_commit_interval;
4771         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4772         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4773 #ifdef CONFIG_QUOTA
4774         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4775         for (i = 0; i < MAXQUOTAS; i++)
4776                 if (sbi->s_qf_names[i]) {
4777                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4778                                                          GFP_KERNEL);
4779                         if (!old_opts.s_qf_names[i]) {
4780                                 for (j = 0; j < i; j++)
4781                                         kfree(old_opts.s_qf_names[j]);
4782                                 kfree(orig_data);
4783                                 return -ENOMEM;
4784                         }
4785                 } else
4786                         old_opts.s_qf_names[i] = NULL;
4787 #endif
4788         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4789                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4790
4791         /*
4792          * Allow the "check" option to be passed as a remount option.
4793          */
4794         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4795                 err = -EINVAL;
4796                 goto restore_opts;
4797         }
4798
4799         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4800                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4801                         ext4_msg(sb, KERN_ERR, "can't mount with "
4802                                  "both data=journal and delalloc");
4803                         err = -EINVAL;
4804                         goto restore_opts;
4805                 }
4806                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4807                         ext4_msg(sb, KERN_ERR, "can't mount with "
4808                                  "both data=journal and dioread_nolock");
4809                         err = -EINVAL;
4810                         goto restore_opts;
4811                 }
4812         }
4813
4814         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4815                 ext4_abort(sb, "Abort forced by user");
4816
4817         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4818                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4819
4820         es = sbi->s_es;
4821
4822         if (sbi->s_journal) {
4823                 ext4_init_journal_params(sb, sbi->s_journal);
4824                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4825         }
4826
4827         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4828                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4829                         err = -EROFS;
4830                         goto restore_opts;
4831                 }
4832
4833                 if (*flags & MS_RDONLY) {
4834                         err = dquot_suspend(sb, -1);
4835                         if (err < 0)
4836                                 goto restore_opts;
4837
4838                         /*
4839                          * First of all, the unconditional stuff we have to do
4840                          * to disable replay of the journal when we next remount
4841                          */
4842                         sb->s_flags |= MS_RDONLY;
4843
4844                         /*
4845                          * OK, test if we are remounting a valid rw partition
4846                          * readonly, and if so set the rdonly flag and then
4847                          * mark the partition as valid again.
4848                          */
4849                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4850                             (sbi->s_mount_state & EXT4_VALID_FS))
4851                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4852
4853                         if (sbi->s_journal)
4854                                 ext4_mark_recovery_complete(sb, es);
4855                 } else {
4856                         /* Make sure we can mount this feature set readwrite */
4857                         if (!ext4_feature_set_ok(sb, 0)) {
4858                                 err = -EROFS;
4859                                 goto restore_opts;
4860                         }
4861                         /*
4862                          * Make sure the group descriptor checksums
4863                          * are sane.  If they aren't, refuse to remount r/w.
4864                          */
4865                         for (g = 0; g < sbi->s_groups_count; g++) {
4866                                 struct ext4_group_desc *gdp =
4867                                         ext4_get_group_desc(sb, g, NULL);
4868
4869                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4870                                         ext4_msg(sb, KERN_ERR,
4871                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4872                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4873                                                le16_to_cpu(gdp->bg_checksum));
4874                                         err = -EINVAL;
4875                                         goto restore_opts;
4876                                 }
4877                         }
4878
4879                         /*
4880                          * If we have an unprocessed orphan list hanging
4881                          * around from a previously readonly bdev mount,
4882                          * require a full umount/remount for now.
4883                          */
4884                         if (es->s_last_orphan) {
4885                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4886                                        "remount RDWR because of unprocessed "
4887                                        "orphan inode list.  Please "
4888                                        "umount/remount instead");
4889                                 err = -EINVAL;
4890                                 goto restore_opts;
4891                         }
4892
4893                         /*
4894                          * Mounting a RDONLY partition read-write, so reread
4895                          * and store the current valid flag.  (It may have
4896                          * been changed by e2fsck since we originally mounted
4897                          * the partition.)
4898                          */
4899                         if (sbi->s_journal)
4900                                 ext4_clear_journal_err(sb, es);
4901                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4902                         if (!ext4_setup_super(sb, es, 0))
4903                                 sb->s_flags &= ~MS_RDONLY;
4904                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4905                                                      EXT4_FEATURE_INCOMPAT_MMP))
4906                                 if (ext4_multi_mount_protect(sb,
4907                                                 le64_to_cpu(es->s_mmp_block))) {
4908                                         err = -EROFS;
4909                                         goto restore_opts;
4910                                 }
4911                         enable_quota = 1;
4912                 }
4913         }
4914
4915         /*
4916          * Reinitialize lazy itable initialization thread based on
4917          * current settings
4918          */
4919         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4920                 ext4_unregister_li_request(sb);
4921         else {
4922                 ext4_group_t first_not_zeroed;
4923                 first_not_zeroed = ext4_has_uninit_itable(sb);
4924                 ext4_register_li_request(sb, first_not_zeroed);
4925         }
4926
4927         ext4_setup_system_zone(sb);
4928         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4929                 ext4_commit_super(sb, 1);
4930
4931 #ifdef CONFIG_QUOTA
4932         /* Release old quota file names */
4933         for (i = 0; i < MAXQUOTAS; i++)
4934                 kfree(old_opts.s_qf_names[i]);
4935         if (enable_quota) {
4936                 if (sb_any_quota_suspended(sb))
4937                         dquot_resume(sb, -1);
4938                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4939                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4940                         err = ext4_enable_quotas(sb);
4941                         if (err)
4942                                 goto restore_opts;
4943                 }
4944         }
4945 #endif
4946
4947         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4948         kfree(orig_data);
4949         return 0;
4950
4951 restore_opts:
4952         sb->s_flags = old_sb_flags;
4953         sbi->s_mount_opt = old_opts.s_mount_opt;
4954         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4955         sbi->s_resuid = old_opts.s_resuid;
4956         sbi->s_resgid = old_opts.s_resgid;
4957         sbi->s_commit_interval = old_opts.s_commit_interval;
4958         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4959         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4960 #ifdef CONFIG_QUOTA
4961         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4962         for (i = 0; i < MAXQUOTAS; i++) {
4963                 kfree(sbi->s_qf_names[i]);
4964                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4965         }
4966 #endif
4967         kfree(orig_data);
4968         return err;
4969 }
4970
4971 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4972 {
4973         struct super_block *sb = dentry->d_sb;
4974         struct ext4_sb_info *sbi = EXT4_SB(sb);
4975         struct ext4_super_block *es = sbi->s_es;
4976         ext4_fsblk_t overhead = 0, resv_blocks;
4977         u64 fsid;
4978         s64 bfree;
4979         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4980
4981         if (!test_opt(sb, MINIX_DF))
4982                 overhead = sbi->s_overhead;
4983
4984         buf->f_type = EXT4_SUPER_MAGIC;
4985         buf->f_bsize = sb->s_blocksize;
4986         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4987         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4988                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4989         /* prevent underflow in case that few free space is available */
4990         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4991         buf->f_bavail = buf->f_bfree -
4992                         (ext4_r_blocks_count(es) + resv_blocks);
4993         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4994                 buf->f_bavail = 0;
4995         buf->f_files = le32_to_cpu(es->s_inodes_count);
4996         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4997         buf->f_namelen = EXT4_NAME_LEN;
4998         fsid = le64_to_cpup((void *)es->s_uuid) ^
4999                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5000         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5001         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5002
5003         return 0;
5004 }
5005
5006 /* Helper function for writing quotas on sync - we need to start transaction
5007  * before quota file is locked for write. Otherwise the are possible deadlocks:
5008  * Process 1                         Process 2
5009  * ext4_create()                     quota_sync()
5010  *   jbd2_journal_start()                  write_dquot()
5011  *   dquot_initialize()                         down(dqio_mutex)
5012  *     down(dqio_mutex)                    jbd2_journal_start()
5013  *
5014  */
5015
5016 #ifdef CONFIG_QUOTA
5017
5018 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5019 {
5020         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5021 }
5022
5023 static int ext4_write_dquot(struct dquot *dquot)
5024 {
5025         int ret, err;
5026         handle_t *handle;
5027         struct inode *inode;
5028
5029         inode = dquot_to_inode(dquot);
5030         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5031                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5032         if (IS_ERR(handle))
5033                 return PTR_ERR(handle);
5034         ret = dquot_commit(dquot);
5035         err = ext4_journal_stop(handle);
5036         if (!ret)
5037                 ret = err;
5038         return ret;
5039 }
5040
5041 static int ext4_acquire_dquot(struct dquot *dquot)
5042 {
5043         int ret, err;
5044         handle_t *handle;
5045
5046         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5047                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5048         if (IS_ERR(handle))
5049                 return PTR_ERR(handle);
5050         ret = dquot_acquire(dquot);
5051         err = ext4_journal_stop(handle);
5052         if (!ret)
5053                 ret = err;
5054         return ret;
5055 }
5056
5057 static int ext4_release_dquot(struct dquot *dquot)
5058 {
5059         int ret, err;
5060         handle_t *handle;
5061
5062         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5063                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5064         if (IS_ERR(handle)) {
5065                 /* Release dquot anyway to avoid endless cycle in dqput() */
5066                 dquot_release(dquot);
5067                 return PTR_ERR(handle);
5068         }
5069         ret = dquot_release(dquot);
5070         err = ext4_journal_stop(handle);
5071         if (!ret)
5072                 ret = err;
5073         return ret;
5074 }
5075
5076 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5077 {
5078         struct super_block *sb = dquot->dq_sb;
5079         struct ext4_sb_info *sbi = EXT4_SB(sb);
5080
5081         /* Are we journaling quotas? */
5082         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5083             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5084                 dquot_mark_dquot_dirty(dquot);
5085                 return ext4_write_dquot(dquot);
5086         } else {
5087                 return dquot_mark_dquot_dirty(dquot);
5088         }
5089 }
5090
5091 static int ext4_write_info(struct super_block *sb, int type)
5092 {
5093         int ret, err;
5094         handle_t *handle;
5095
5096         /* Data block + inode block */
5097         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5098         if (IS_ERR(handle))
5099                 return PTR_ERR(handle);
5100         ret = dquot_commit_info(sb, type);
5101         err = ext4_journal_stop(handle);
5102         if (!ret)
5103                 ret = err;
5104         return ret;
5105 }
5106
5107 /*
5108  * Turn on quotas during mount time - we need to find
5109  * the quota file and such...
5110  */
5111 static int ext4_quota_on_mount(struct super_block *sb, int type)
5112 {
5113         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5114                                         EXT4_SB(sb)->s_jquota_fmt, type);
5115 }
5116
5117 /*
5118  * Standard function to be called on quota_on
5119  */
5120 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5121                          struct path *path)
5122 {
5123         int err;
5124
5125         if (!test_opt(sb, QUOTA))
5126                 return -EINVAL;
5127
5128         /* Quotafile not on the same filesystem? */
5129         if (path->dentry->d_sb != sb)
5130                 return -EXDEV;
5131         /* Journaling quota? */
5132         if (EXT4_SB(sb)->s_qf_names[type]) {
5133                 /* Quotafile not in fs root? */
5134                 if (path->dentry->d_parent != sb->s_root)
5135                         ext4_msg(sb, KERN_WARNING,
5136                                 "Quota file not on filesystem root. "
5137                                 "Journaled quota will not work");
5138         }
5139
5140         /*
5141          * When we journal data on quota file, we have to flush journal to see
5142          * all updates to the file when we bypass pagecache...
5143          */
5144         if (EXT4_SB(sb)->s_journal &&
5145             ext4_should_journal_data(path->dentry->d_inode)) {
5146                 /*
5147                  * We don't need to lock updates but journal_flush() could
5148                  * otherwise be livelocked...
5149                  */
5150                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5151                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5152                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5153                 if (err)
5154                         return err;
5155         }
5156
5157         return dquot_quota_on(sb, type, format_id, path);
5158 }
5159
5160 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5161                              unsigned int flags)
5162 {
5163         int err;
5164         struct inode *qf_inode;
5165         unsigned long qf_inums[MAXQUOTAS] = {
5166                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5167                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5168         };
5169
5170         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5171
5172         if (!qf_inums[type])
5173                 return -EPERM;
5174
5175         qf_inode = ext4_iget(sb, qf_inums[type]);
5176         if (IS_ERR(qf_inode)) {
5177                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5178                 return PTR_ERR(qf_inode);
5179         }
5180
5181         /* Don't account quota for quota files to avoid recursion */
5182         qf_inode->i_flags |= S_NOQUOTA;
5183         err = dquot_enable(qf_inode, type, format_id, flags);
5184         iput(qf_inode);
5185
5186         return err;
5187 }
5188
5189 /* Enable usage tracking for all quota types. */
5190 static int ext4_enable_quotas(struct super_block *sb)
5191 {
5192         int type, err = 0;
5193         unsigned long qf_inums[MAXQUOTAS] = {
5194                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5195                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5196         };
5197
5198         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5199         for (type = 0; type < MAXQUOTAS; type++) {
5200                 if (qf_inums[type]) {
5201                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5202                                                 DQUOT_USAGE_ENABLED);
5203                         if (err) {
5204                                 ext4_warning(sb,
5205                                         "Failed to enable quota tracking "
5206                                         "(type=%d, err=%d). Please run "
5207                                         "e2fsck to fix.", type, err);
5208                                 return err;
5209                         }
5210                 }
5211         }
5212         return 0;
5213 }
5214
5215 /*
5216  * quota_on function that is used when QUOTA feature is set.
5217  */
5218 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5219                                  int format_id)
5220 {
5221         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5222                 return -EINVAL;
5223
5224         /*
5225          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5226          */
5227         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5228 }
5229
5230 static int ext4_quota_off(struct super_block *sb, int type)
5231 {
5232         struct inode *inode = sb_dqopt(sb)->files[type];
5233         handle_t *handle;
5234
5235         /* Force all delayed allocation blocks to be allocated.
5236          * Caller already holds s_umount sem */
5237         if (test_opt(sb, DELALLOC))
5238                 sync_filesystem(sb);
5239
5240         if (!inode)
5241                 goto out;
5242
5243         /* Update modification times of quota files when userspace can
5244          * start looking at them */
5245         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5246         if (IS_ERR(handle))
5247                 goto out;
5248         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5249         ext4_mark_inode_dirty(handle, inode);
5250         ext4_journal_stop(handle);
5251
5252 out:
5253         return dquot_quota_off(sb, type);
5254 }
5255
5256 /*
5257  * quota_off function that is used when QUOTA feature is set.
5258  */
5259 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5260 {
5261         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5262                 return -EINVAL;
5263
5264         /* Disable only the limits. */
5265         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5266 }
5267
5268 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5269  * acquiring the locks... As quota files are never truncated and quota code
5270  * itself serializes the operations (and no one else should touch the files)
5271  * we don't have to be afraid of races */
5272 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5273                                size_t len, loff_t off)
5274 {
5275         struct inode *inode = sb_dqopt(sb)->files[type];
5276         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5277         int err = 0;
5278         int offset = off & (sb->s_blocksize - 1);
5279         int tocopy;
5280         size_t toread;
5281         struct buffer_head *bh;
5282         loff_t i_size = i_size_read(inode);
5283
5284         if (off > i_size)
5285                 return 0;
5286         if (off+len > i_size)
5287                 len = i_size-off;
5288         toread = len;
5289         while (toread > 0) {
5290                 tocopy = sb->s_blocksize - offset < toread ?
5291                                 sb->s_blocksize - offset : toread;
5292                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5293                 if (err)
5294                         return err;
5295                 if (!bh)        /* A hole? */
5296                         memset(data, 0, tocopy);
5297                 else
5298                         memcpy(data, bh->b_data+offset, tocopy);
5299                 brelse(bh);
5300                 offset = 0;
5301                 toread -= tocopy;
5302                 data += tocopy;
5303                 blk++;
5304         }
5305         return len;
5306 }
5307
5308 /* Write to quotafile (we know the transaction is already started and has
5309  * enough credits) */
5310 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5311                                 const char *data, size_t len, loff_t off)
5312 {
5313         struct inode *inode = sb_dqopt(sb)->files[type];
5314         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5315         int err = 0;
5316         int offset = off & (sb->s_blocksize - 1);
5317         struct buffer_head *bh;
5318         handle_t *handle = journal_current_handle();
5319
5320         if (EXT4_SB(sb)->s_journal && !handle) {
5321                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5322                         " cancelled because transaction is not started",
5323                         (unsigned long long)off, (unsigned long long)len);
5324                 return -EIO;
5325         }
5326         /*
5327          * Since we account only one data block in transaction credits,
5328          * then it is impossible to cross a block boundary.
5329          */
5330         if (sb->s_blocksize - offset < len) {
5331                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5332                         " cancelled because not block aligned",
5333                         (unsigned long long)off, (unsigned long long)len);
5334                 return -EIO;
5335         }
5336
5337         bh = ext4_bread(handle, inode, blk, 1, &err);
5338         if (!bh)
5339                 goto out;
5340         err = ext4_journal_get_write_access(handle, bh);
5341         if (err) {
5342                 brelse(bh);
5343                 goto out;
5344         }
5345         lock_buffer(bh);
5346         memcpy(bh->b_data+offset, data, len);
5347         flush_dcache_page(bh->b_page);
5348         unlock_buffer(bh);
5349         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5350         brelse(bh);
5351 out:
5352         if (err)
5353                 return err;
5354         if (inode->i_size < off + len) {
5355                 i_size_write(inode, off + len);
5356                 EXT4_I(inode)->i_disksize = inode->i_size;
5357                 ext4_mark_inode_dirty(handle, inode);
5358         }
5359         return len;
5360 }
5361
5362 #endif
5363
5364 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5365                        const char *dev_name, void *data)
5366 {
5367         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5368 }
5369
5370 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5371 static inline void register_as_ext2(void)
5372 {
5373         int err = register_filesystem(&ext2_fs_type);
5374         if (err)
5375                 printk(KERN_WARNING
5376                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5377 }
5378
5379 static inline void unregister_as_ext2(void)
5380 {
5381         unregister_filesystem(&ext2_fs_type);
5382 }
5383
5384 static inline int ext2_feature_set_ok(struct super_block *sb)
5385 {
5386         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5387                 return 0;
5388         if (sb->s_flags & MS_RDONLY)
5389                 return 1;
5390         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5391                 return 0;
5392         return 1;
5393 }
5394 #else
5395 static inline void register_as_ext2(void) { }
5396 static inline void unregister_as_ext2(void) { }
5397 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5398 #endif
5399
5400 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5401 static inline void register_as_ext3(void)
5402 {
5403         int err = register_filesystem(&ext3_fs_type);
5404         if (err)
5405                 printk(KERN_WARNING
5406                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5407 }
5408
5409 static inline void unregister_as_ext3(void)
5410 {
5411         unregister_filesystem(&ext3_fs_type);
5412 }
5413
5414 static inline int ext3_feature_set_ok(struct super_block *sb)
5415 {
5416         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5417                 return 0;
5418         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5419                 return 0;
5420         if (sb->s_flags & MS_RDONLY)
5421                 return 1;
5422         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5423                 return 0;
5424         return 1;
5425 }
5426 #else
5427 static inline void register_as_ext3(void) { }
5428 static inline void unregister_as_ext3(void) { }
5429 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5430 #endif
5431
5432 static struct file_system_type ext4_fs_type = {
5433         .owner          = THIS_MODULE,
5434         .name           = "ext4",
5435         .mount          = ext4_mount,
5436         .kill_sb        = kill_block_super,
5437         .fs_flags       = FS_REQUIRES_DEV,
5438 };
5439 MODULE_ALIAS_FS("ext4");
5440
5441 static int __init ext4_init_feat_adverts(void)
5442 {
5443         struct ext4_features *ef;
5444         int ret = -ENOMEM;
5445
5446         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5447         if (!ef)
5448                 goto out;
5449
5450         ef->f_kobj.kset = ext4_kset;
5451         init_completion(&ef->f_kobj_unregister);
5452         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5453                                    "features");
5454         if (ret) {
5455                 kfree(ef);
5456                 goto out;
5457         }
5458
5459         ext4_feat = ef;
5460         ret = 0;
5461 out:
5462         return ret;
5463 }
5464
5465 static void ext4_exit_feat_adverts(void)
5466 {
5467         kobject_put(&ext4_feat->f_kobj);
5468         wait_for_completion(&ext4_feat->f_kobj_unregister);
5469         kfree(ext4_feat);
5470 }
5471
5472 /* Shared across all ext4 file systems */
5473 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5474 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5475
5476 static int __init ext4_init_fs(void)
5477 {
5478         int i, err;
5479
5480         ext4_li_info = NULL;
5481         mutex_init(&ext4_li_mtx);
5482
5483         /* Build-time check for flags consistency */
5484         ext4_check_flag_values();
5485
5486         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5487                 mutex_init(&ext4__aio_mutex[i]);
5488                 init_waitqueue_head(&ext4__ioend_wq[i]);
5489         }
5490
5491         err = ext4_init_es();
5492         if (err)
5493                 return err;
5494
5495         err = ext4_init_pageio();
5496         if (err)
5497                 goto out7;
5498
5499         err = ext4_init_system_zone();
5500         if (err)
5501                 goto out6;
5502         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5503         if (!ext4_kset) {
5504                 err = -ENOMEM;
5505                 goto out5;
5506         }
5507         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5508
5509         err = ext4_init_feat_adverts();
5510         if (err)
5511                 goto out4;
5512
5513         err = ext4_init_mballoc();
5514         if (err)
5515                 goto out3;
5516
5517         err = ext4_init_xattr();
5518         if (err)
5519                 goto out2;
5520         err = init_inodecache();
5521         if (err)
5522                 goto out1;
5523         register_as_ext3();
5524         register_as_ext2();
5525         err = register_filesystem(&ext4_fs_type);
5526         if (err)
5527                 goto out;
5528
5529         return 0;
5530 out:
5531         unregister_as_ext2();
5532         unregister_as_ext3();
5533         destroy_inodecache();
5534 out1:
5535         ext4_exit_xattr();
5536 out2:
5537         ext4_exit_mballoc();
5538 out3:
5539         ext4_exit_feat_adverts();
5540 out4:
5541         if (ext4_proc_root)
5542                 remove_proc_entry("fs/ext4", NULL);
5543         kset_unregister(ext4_kset);
5544 out5:
5545         ext4_exit_system_zone();
5546 out6:
5547         ext4_exit_pageio();
5548 out7:
5549         ext4_exit_es();
5550
5551         return err;
5552 }
5553
5554 static void __exit ext4_exit_fs(void)
5555 {
5556         ext4_destroy_lazyinit_thread();
5557         unregister_as_ext2();
5558         unregister_as_ext3();
5559         unregister_filesystem(&ext4_fs_type);
5560         destroy_inodecache();
5561         ext4_exit_xattr();
5562         ext4_exit_mballoc();
5563         ext4_exit_feat_adverts();
5564         remove_proc_entry("fs/ext4", NULL);
5565         kset_unregister(ext4_kset);
5566         ext4_exit_system_zone();
5567         ext4_exit_pageio();
5568         ext4_exit_es();
5569 }
5570
5571 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5572 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5573 MODULE_LICENSE("GPL");
5574 module_init(ext4_init_fs)
5575 module_exit(ext4_exit_fs)