fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 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 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 MODULE_ALIAS_FS("ext2");
  97 MODULE_ALIAS("ext2");
  98 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  99 #else
 100 #define IS_EXT2_SB(sb) (0)
 101 #endif
 102
 103
 104 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 105 static struct file_system_type ext3_fs_type = {
 106         .owner          = THIS_MODULE,
 107         .name           = "ext3",
 108         .mount          = ext4_mount,
 109         .kill_sb        = kill_block_super,
 110         .fs_flags       = FS_REQUIRES_DEV,
 111 };
 112 MODULE_ALIAS_FS("ext3");
 113 MODULE_ALIAS("ext3");
 114 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 115 #else
 116 #define IS_EXT3_SB(sb) (0)
 117 #endif
 118
 119 static int ext4_verify_csum_type(struct super_block *sb,
 120                                  struct ext4_super_block *es)
 121 {
 122         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 123                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 124                 return 1;
 125
 126         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 127 }
 128
 129 static __le32 ext4_superblock_csum(struct super_block *sb,
 130                                    struct ext4_super_block *es)
 131 {
 132         struct ext4_sb_info *sbi = EXT4_SB(sb);
 133         int offset = offsetof(struct ext4_super_block, s_checksum);
 134         __u32 csum;
 135
 136         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 137
 138         return cpu_to_le32(csum);
 139 }
 140
 141 static int ext4_superblock_csum_verify(struct super_block *sb,
 142                                        struct ext4_super_block *es)
 143 {
 144         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 145                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 146                 return 1;
 147
 148         return es->s_checksum == ext4_superblock_csum(sb, es);
 149 }
 150
 151 void ext4_superblock_csum_set(struct super_block *sb)
 152 {
 153         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 154
 155         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 156                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 157                 return;
 158
 159         es->s_checksum = ext4_superblock_csum(sb, es);
 160 }
 161
 162 void *ext4_kvmalloc(size_t size, gfp_t flags)
 163 {
 164         void *ret;
 165
 166         ret = kmalloc(size, flags | __GFP_NOWARN);
 167         if (!ret)
 168                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 169         return ret;
 170 }
 171
 172 void *ext4_kvzalloc(size_t size, gfp_t flags)
 173 {
 174         void *ret;
 175
 176         ret = kzalloc(size, flags | __GFP_NOWARN);
 177         if (!ret)
 178                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 179         return ret;
 180 }
 181
 182 void ext4_kvfree(void *ptr)
 183 {
 184         if (is_vmalloc_addr(ptr))
 185                 vfree(ptr);
 186         else
 187                 kfree(ptr);
 188
 189 }
 190
 191 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 192                                struct ext4_group_desc *bg)
 193 {
 194         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 197 }
 198
 199 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 200                                struct ext4_group_desc *bg)
 201 {
 202         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 203                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 204                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 205 }
 206
 207 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 208                               struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_inode_table_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 213 }
 214
 215 __u32 ext4_free_group_clusters(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 221 }
 222
 223 __u32 ext4_free_inodes_count(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 229 }
 230
 231 __u32 ext4_used_dirs_count(struct super_block *sb,
 232                               struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_itable_unused_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_itable_unused_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 245 }
 246
 247 void ext4_block_bitmap_set(struct super_block *sb,
 248                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 249 {
 250         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 253 }
 254
 255 void ext4_inode_bitmap_set(struct super_block *sb,
 256                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 257 {
 258         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 259         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 260                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 261 }
 262
 263 void ext4_inode_table_set(struct super_block *sb,
 264                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_free_group_clusters_set(struct super_block *sb,
 272                                   struct ext4_group_desc *bg, __u32 count)
 273 {
 274         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 277 }
 278
 279 void ext4_free_inodes_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, __u32 count)
 281 {
 282         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 285 }
 286
 287 void ext4_used_dirs_set(struct super_block *sb,
 288                           struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_itable_unused_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303
 304 static void __save_error_info(struct super_block *sb, const char *func,
 305                             unsigned int line)
 306 {
 307         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 308
 309         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 310         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 311         es->s_last_error_time = cpu_to_le32(get_seconds());
 312         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 313         es->s_last_error_line = cpu_to_le32(line);
 314         if (!es->s_first_error_time) {
 315                 es->s_first_error_time = es->s_last_error_time;
 316                 strncpy(es->s_first_error_func, func,
 317                         sizeof(es->s_first_error_func));
 318                 es->s_first_error_line = cpu_to_le32(line);
 319                 es->s_first_error_ino = es->s_last_error_ino;
 320                 es->s_first_error_block = es->s_last_error_block;
 321         }
 322         /*
 323          * Start the daily error reporting function if it hasn't been
 324          * started already
 325          */
 326         if (!es->s_error_count)
 327                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 328         le32_add_cpu(&es->s_error_count, 1);
 329 }
 330
 331 static void save_error_info(struct super_block *sb, const char *func,
 332                             unsigned int line)
 333 {
 334         __save_error_info(sb, func, line);
 335         ext4_commit_super(sb, 1);
 336 }
 337
 338 /*
 339  * The del_gendisk() function uninitializes the disk-specific data
 340  * structures, including the bdi structure, without telling anyone
 341  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 342  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 343  * This is a kludge to prevent these oops until we can put in a proper
 344  * hook in del_gendisk() to inform the VFS and file system layers.
 345  */
 346 static int block_device_ejected(struct super_block *sb)
 347 {
 348         struct inode *bd_inode = sb->s_bdev->bd_inode;
 349         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 350
 351         return bdi->dev == NULL;
 352 }
 353
 354 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 355 {
 356         struct super_block              *sb = journal->j_private;
 357         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 358         int                             error = is_journal_aborted(journal);
 359         struct ext4_journal_cb_entry    *jce;
 360
 361         BUG_ON(txn->t_state == T_FINISHED);
 362         spin_lock(&sbi->s_md_lock);
 363         while (!list_empty(&txn->t_private_list)) {
 364                 jce = list_entry(txn->t_private_list.next,
 365                                  struct ext4_journal_cb_entry, jce_list);
 366                 list_del_init(&jce->jce_list);
 367                 spin_unlock(&sbi->s_md_lock);
 368                 jce->jce_func(sb, jce, error);
 369                 spin_lock(&sbi->s_md_lock);
 370         }
 371         spin_unlock(&sbi->s_md_lock);
 372 }
 373
 374 /* Deal with the reporting of failure conditions on a filesystem such as
 375  * inconsistencies detected or read IO failures.
 376  *
 377  * On ext2, we can store the error state of the filesystem in the
 378  * superblock.  That is not possible on ext4, because we may have other
 379  * write ordering constraints on the superblock which prevent us from
 380  * writing it out straight away; and given that the journal is about to
 381  * be aborted, we can't rely on the current, or future, transactions to
 382  * write out the superblock safely.
 383  *
 384  * We'll just use the jbd2_journal_abort() error code to record an error in
 385  * the journal instead.  On recovery, the journal will complain about
 386  * that error until we've noted it down and cleared it.
 387  */
 388
 389 static void ext4_handle_error(struct super_block *sb)
 390 {
 391         if (sb->s_flags & MS_RDONLY)
 392                 return;
 393
 394         if (!test_opt(sb, ERRORS_CONT)) {
 395                 journal_t *journal = EXT4_SB(sb)->s_journal;
 396
 397                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 398                 if (journal)
 399                         jbd2_journal_abort(journal, -EIO);
 400         }
 401         if (test_opt(sb, ERRORS_RO)) {
 402                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 403                 /*
 404                  * Make sure updated value of ->s_mount_flags will be visible
 405                  * before ->s_flags update
 406                  */
 407                 smp_wmb();
 408                 sb->s_flags |= MS_RDONLY;
 409         }
 410         if (test_opt(sb, ERRORS_PANIC))
 411                 panic("EXT4-fs (device %s): panic forced after error\n",
 412                         sb->s_id);
 413 }
 414
 415 #define ext4_error_ratelimit(sb)                                        \
 416                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 417                              "EXT4-fs error")
 418
 419 void __ext4_error(struct super_block *sb, const char *function,
 420                   unsigned int line, const char *fmt, ...)
 421 {
 422         struct va_format vaf;
 423         va_list args;
 424
 425         if (ext4_error_ratelimit(sb)) {
 426                 va_start(args, fmt);
 427                 vaf.fmt = fmt;
 428                 vaf.va = &args;
 429                 printk(KERN_CRIT
 430                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 431                        sb->s_id, function, line, current->comm, &vaf);
 432                 va_end(args);
 433         }
 434         save_error_info(sb, function, line);
 435         ext4_handle_error(sb);
 436 }
 437
 438 void __ext4_error_inode(struct inode *inode, const char *function,
 439                         unsigned int line, ext4_fsblk_t block,
 440                         const char *fmt, ...)
 441 {
 442         va_list args;
 443         struct va_format vaf;
 444         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 445
 446         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 447         es->s_last_error_block = cpu_to_le64(block);
 448         if (ext4_error_ratelimit(inode->i_sb)) {
 449                 va_start(args, fmt);
 450                 vaf.fmt = fmt;
 451                 vaf.va = &args;
 452                 if (block)
 453                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 454                                "inode #%lu: block %llu: comm %s: %pV\n",
 455                                inode->i_sb->s_id, function, line, inode->i_ino,
 456                                block, current->comm, &vaf);
 457                 else
 458                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 459                                "inode #%lu: comm %s: %pV\n",
 460                                inode->i_sb->s_id, function, line, inode->i_ino,
 461                                current->comm, &vaf);
 462                 va_end(args);
 463         }
 464         save_error_info(inode->i_sb, function, line);
 465         ext4_handle_error(inode->i_sb);
 466 }
 467
 468 void __ext4_error_file(struct file *file, const char *function,
 469                        unsigned int line, ext4_fsblk_t block,
 470                        const char *fmt, ...)
 471 {
 472         va_list args;
 473         struct va_format vaf;
 474         struct ext4_super_block *es;
 475         struct inode *inode = file_inode(file);
 476         char pathname[80], *path;
 477
 478         es = EXT4_SB(inode->i_sb)->s_es;
 479         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 480         if (ext4_error_ratelimit(inode->i_sb)) {
 481                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 482                 if (IS_ERR(path))
 483                         path = "(unknown)";
 484                 va_start(args, fmt);
 485                 vaf.fmt = fmt;
 486                 vaf.va = &args;
 487                 if (block)
 488                         printk(KERN_CRIT
 489                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 490                                "block %llu: comm %s: path %s: %pV\n",
 491                                inode->i_sb->s_id, function, line, inode->i_ino,
 492                                block, current->comm, path, &vaf);
 493                 else
 494                         printk(KERN_CRIT
 495                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 496                                "comm %s: path %s: %pV\n",
 497                                inode->i_sb->s_id, function, line, inode->i_ino,
 498                                current->comm, path, &vaf);
 499                 va_end(args);
 500         }
 501         save_error_info(inode->i_sb, function, line);
 502         ext4_handle_error(inode->i_sb);
 503 }
 504
 505 const char *ext4_decode_error(struct super_block *sb, int errno,
 506                               char nbuf[16])
 507 {
 508         char *errstr = NULL;
 509
 510         switch (errno) {
 511         case -EIO:
 512                 errstr = "IO failure";
 513                 break;
 514         case -ENOMEM:
 515                 errstr = "Out of memory";
 516                 break;
 517         case -EROFS:
 518                 if (!sb || (EXT4_SB(sb)->s_journal &&
 519                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 520                         errstr = "Journal has aborted";
 521                 else
 522                         errstr = "Readonly filesystem";
 523                 break;
 524         default:
 525                 /* If the caller passed in an extra buffer for unknown
 526                  * errors, textualise them now.  Else we just return
 527                  * NULL. */
 528                 if (nbuf) {
 529                         /* Check for truncated error codes... */
 530                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 531                                 errstr = nbuf;
 532                 }
 533                 break;
 534         }
 535
 536         return errstr;
 537 }
 538
 539 /* __ext4_std_error decodes expected errors from journaling functions
 540  * automatically and invokes the appropriate error response.  */
 541
 542 void __ext4_std_error(struct super_block *sb, const char *function,
 543                       unsigned int line, int errno)
 544 {
 545         char nbuf[16];
 546         const char *errstr;
 547
 548         /* Special case: if the error is EROFS, and we're not already
 549          * inside a transaction, then there's really no point in logging
 550          * an error. */
 551         if (errno == -EROFS && journal_current_handle() == NULL &&
 552             (sb->s_flags & MS_RDONLY))
 553                 return;
 554
 555         if (ext4_error_ratelimit(sb)) {
 556                 errstr = ext4_decode_error(sb, errno, nbuf);
 557                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 558                        sb->s_id, function, line, errstr);
 559         }
 560
 561         save_error_info(sb, function, line);
 562         ext4_handle_error(sb);
 563 }
 564
 565 /*
 566  * ext4_abort is a much stronger failure handler than ext4_error.  The
 567  * abort function may be used to deal with unrecoverable failures such
 568  * as journal IO errors or ENOMEM at a critical moment in log management.
 569  *
 570  * We unconditionally force the filesystem into an ABORT|READONLY state,
 571  * unless the error response on the fs has been set to panic in which
 572  * case we take the easy way out and panic immediately.
 573  */
 574
 575 void __ext4_abort(struct super_block *sb, const char *function,
 576                 unsigned int line, const char *fmt, ...)
 577 {
 578         va_list args;
 579
 580         save_error_info(sb, function, line);
 581         va_start(args, fmt);
 582         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 583                function, line);
 584         vprintk(fmt, args);
 585         printk("\n");
 586         va_end(args);
 587
 588         if ((sb->s_flags & MS_RDONLY) == 0) {
 589                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 590                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 591                 /*
 592                  * Make sure updated value of ->s_mount_flags will be visible
 593                  * before ->s_flags update
 594                  */
 595                 smp_wmb();
 596                 sb->s_flags |= MS_RDONLY;
 597                 if (EXT4_SB(sb)->s_journal)
 598                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 599                 save_error_info(sb, function, line);
 600         }
 601         if (test_opt(sb, ERRORS_PANIC))
 602                 panic("EXT4-fs panic from previous error\n");
 603 }
 604
 605 void __ext4_msg(struct super_block *sb,
 606                 const char *prefix, const char *fmt, ...)
 607 {
 608         struct va_format vaf;
 609         va_list args;
 610
 611         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 612                 return;
 613
 614         va_start(args, fmt);
 615         vaf.fmt = fmt;
 616         vaf.va = &args;
 617         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 618         va_end(args);
 619 }
 620
 621 void __ext4_warning(struct super_block *sb, const char *function,
 622                     unsigned int line, const char *fmt, ...)
 623 {
 624         struct va_format vaf;
 625         va_list args;
 626
 627         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 628                           "EXT4-fs warning"))
 629                 return;
 630
 631         va_start(args, fmt);
 632         vaf.fmt = fmt;
 633         vaf.va = &args;
 634         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 635                sb->s_id, function, line, &vaf);
 636         va_end(args);
 637 }
 638
 639 void __ext4_grp_locked_error(const char *function, unsigned int line,
 640                              struct super_block *sb, ext4_group_t grp,
 641                              unsigned long ino, ext4_fsblk_t block,
 642                              const char *fmt, ...)
 643 __releases(bitlock)
 644 __acquires(bitlock)
 645 {
 646         struct va_format vaf;
 647         va_list args;
 648         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 649
 650         es->s_last_error_ino = cpu_to_le32(ino);
 651         es->s_last_error_block = cpu_to_le64(block);
 652         __save_error_info(sb, function, line);
 653
 654         if (ext4_error_ratelimit(sb)) {
 655                 va_start(args, fmt);
 656                 vaf.fmt = fmt;
 657                 vaf.va = &args;
 658                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 659                        sb->s_id, function, line, grp);
 660                 if (ino)
 661                         printk(KERN_CONT "inode %lu: ", ino);
 662                 if (block)
 663                         printk(KERN_CONT "block %llu:",
 664                                (unsigned long long) block);
 665                 printk(KERN_CONT "%pV\n", &vaf);
 666                 va_end(args);
 667         }
 668
 669         if (test_opt(sb, ERRORS_CONT)) {
 670                 ext4_commit_super(sb, 0);
 671                 return;
 672         }
 673
 674         ext4_unlock_group(sb, grp);
 675         ext4_handle_error(sb);
 676         /*
 677          * We only get here in the ERRORS_RO case; relocking the group
 678          * may be dangerous, but nothing bad will happen since the
 679          * filesystem will have already been marked read/only and the
 680          * journal has been aborted.  We return 1 as a hint to callers
 681          * who might what to use the return value from
 682          * ext4_grp_locked_error() to distinguish between the
 683          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 684          * aggressively from the ext4 function in question, with a
 685          * more appropriate error code.
 686          */
 687         ext4_lock_group(sb, grp);
 688         return;
 689 }
 690
 691 void ext4_update_dynamic_rev(struct super_block *sb)
 692 {
 693         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 694
 695         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 696                 return;
 697
 698         ext4_warning(sb,
 699                      "updating to rev %d because of new feature flag, "
 700                      "running e2fsck is recommended",
 701                      EXT4_DYNAMIC_REV);
 702
 703         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 704         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 705         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 706         /* leave es->s_feature_*compat flags alone */
 707         /* es->s_uuid will be set by e2fsck if empty */
 708
 709         /*
 710          * The rest of the superblock fields should be zero, and if not it
 711          * means they are likely already in use, so leave them alone.  We
 712          * can leave it up to e2fsck to clean up any inconsistencies there.
 713          */
 714 }
 715
 716 /*
 717  * Open the external journal device
 718  */
 719 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 720 {
 721         struct block_device *bdev;
 722         char b[BDEVNAME_SIZE];
 723
 724         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 725         if (IS_ERR(bdev))
 726                 goto fail;
 727         return bdev;
 728
 729 fail:
 730         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 731                         __bdevname(dev, b), PTR_ERR(bdev));
 732         return NULL;
 733 }
 734
 735 /*
 736  * Release the journal device
 737  */
 738 static void ext4_blkdev_put(struct block_device *bdev)
 739 {
 740         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 741 }
 742
 743 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 744 {
 745         struct block_device *bdev;
 746         bdev = sbi->journal_bdev;
 747         if (bdev) {
 748                 ext4_blkdev_put(bdev);
 749                 sbi->journal_bdev = NULL;
 750         }
 751 }
 752
 753 static inline struct inode *orphan_list_entry(struct list_head *l)
 754 {
 755         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 756 }
 757
 758 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 759 {
 760         struct list_head *l;
 761
 762         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 763                  le32_to_cpu(sbi->s_es->s_last_orphan));
 764
 765         printk(KERN_ERR "sb_info orphan list:\n");
 766         list_for_each(l, &sbi->s_orphan) {
 767                 struct inode *inode = orphan_list_entry(l);
 768                 printk(KERN_ERR "  "
 769                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 770                        inode->i_sb->s_id, inode->i_ino, inode,
 771                        inode->i_mode, inode->i_nlink,
 772                        NEXT_ORPHAN(inode));
 773         }
 774 }
 775
 776 static void ext4_put_super(struct super_block *sb)
 777 {
 778         struct ext4_sb_info *sbi = EXT4_SB(sb);
 779         struct ext4_super_block *es = sbi->s_es;
 780         int i, err;
 781
 782         ext4_unregister_li_request(sb);
 783         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 784
 785         flush_workqueue(sbi->rsv_conversion_wq);
 786         destroy_workqueue(sbi->rsv_conversion_wq);
 787
 788         if (sbi->s_journal) {
 789                 err = jbd2_journal_destroy(sbi->s_journal);
 790                 sbi->s_journal = NULL;
 791                 if (err < 0)
 792                         ext4_abort(sb, "Couldn't clean up the journal");
 793         }
 794
 795         ext4_es_unregister_shrinker(sbi);
 796         del_timer_sync(&sbi->s_err_report);
 797         ext4_release_system_zone(sb);
 798         ext4_mb_release(sb);
 799         ext4_ext_release(sb);
 800         ext4_xattr_put_super(sb);
 801
 802         if (!(sb->s_flags & MS_RDONLY)) {
 803                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 804                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 805         }
 806         if (!(sb->s_flags & MS_RDONLY))
 807                 ext4_commit_super(sb, 1);
 808
 809         if (sbi->s_proc) {
 810                 remove_proc_entry("options", sbi->s_proc);
 811                 remove_proc_entry(sb->s_id, ext4_proc_root);
 812         }
 813         kobject_del(&sbi->s_kobj);
 814
 815         for (i = 0; i < sbi->s_gdb_count; i++)
 816                 brelse(sbi->s_group_desc[i]);
 817         ext4_kvfree(sbi->s_group_desc);
 818         ext4_kvfree(sbi->s_flex_groups);
 819         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 820         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 821         percpu_counter_destroy(&sbi->s_dirs_counter);
 822         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 823         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 824         brelse(sbi->s_sbh);
 825 #ifdef CONFIG_QUOTA
 826         for (i = 0; i < MAXQUOTAS; i++)
 827                 kfree(sbi->s_qf_names[i]);
 828 #endif
 829
 830         /* Debugging code just in case the in-memory inode orphan list
 831          * isn't empty.  The on-disk one can be non-empty if we've
 832          * detected an error and taken the fs readonly, but the
 833          * in-memory list had better be clean by this point. */
 834         if (!list_empty(&sbi->s_orphan))
 835                 dump_orphan_list(sb, sbi);
 836         J_ASSERT(list_empty(&sbi->s_orphan));
 837
 838         invalidate_bdev(sb->s_bdev);
 839         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 840                 /*
 841                  * Invalidate the journal device's buffers.  We don't want them
 842                  * floating about in memory - the physical journal device may
 843                  * hotswapped, and it breaks the `ro-after' testing code.
 844                  */
 845                 sync_blockdev(sbi->journal_bdev);
 846                 invalidate_bdev(sbi->journal_bdev);
 847                 ext4_blkdev_remove(sbi);
 848         }
 849         if (sbi->s_mb_cache) {
 850                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 851                 sbi->s_mb_cache = NULL;
 852         }
 853         if (sbi->s_mmp_tsk)
 854                 kthread_stop(sbi->s_mmp_tsk);
 855         sb->s_fs_info = NULL;
 856         /*
 857          * Now that we are completely done shutting down the
 858          * superblock, we need to actually destroy the kobject.
 859          */
 860         kobject_put(&sbi->s_kobj);
 861         wait_for_completion(&sbi->s_kobj_unregister);
 862         if (sbi->s_chksum_driver)
 863                 crypto_free_shash(sbi->s_chksum_driver);
 864         kfree(sbi->s_blockgroup_lock);
 865         kfree(sbi);
 866 }
 867
 868 static struct kmem_cache *ext4_inode_cachep;
 869
 870 /*
 871  * Called inside transaction, so use GFP_NOFS
 872  */
 873 static struct inode *ext4_alloc_inode(struct super_block *sb)
 874 {
 875         struct ext4_inode_info *ei;
 876
 877         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 878         if (!ei)
 879                 return NULL;
 880
 881         ei->vfs_inode.i_version = 1;
 882         spin_lock_init(&ei->i_raw_lock);
 883         INIT_LIST_HEAD(&ei->i_prealloc_list);
 884         spin_lock_init(&ei->i_prealloc_lock);
 885         ext4_es_init_tree(&ei->i_es_tree);
 886         rwlock_init(&ei->i_es_lock);
 887         INIT_LIST_HEAD(&ei->i_es_lru);
 888         ei->i_es_lru_nr = 0;
 889         ei->i_touch_when = 0;
 890         ei->i_reserved_data_blocks = 0;
 891         ei->i_reserved_meta_blocks = 0;
 892         ei->i_allocated_meta_blocks = 0;
 893         ei->i_da_metadata_calc_len = 0;
 894         ei->i_da_metadata_calc_last_lblock = 0;
 895         spin_lock_init(&(ei->i_block_reservation_lock));
 896 #ifdef CONFIG_QUOTA
 897         ei->i_reserved_quota = 0;
 898 #endif
 899         ei->jinode = NULL;
 900         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 901         spin_lock_init(&ei->i_completed_io_lock);
 902         ei->i_sync_tid = 0;
 903         ei->i_datasync_tid = 0;
 904         atomic_set(&ei->i_ioend_count, 0);
 905         atomic_set(&ei->i_unwritten, 0);
 906         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 907
 908         return &ei->vfs_inode;
 909 }
 910
 911 static int ext4_drop_inode(struct inode *inode)
 912 {
 913         int drop = generic_drop_inode(inode);
 914
 915         trace_ext4_drop_inode(inode, drop);
 916         return drop;
 917 }
 918
 919 static void ext4_i_callback(struct rcu_head *head)
 920 {
 921         struct inode *inode = container_of(head, struct inode, i_rcu);
 922         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 923 }
 924
 925 static void ext4_destroy_inode(struct inode *inode)
 926 {
 927         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 928                 ext4_msg(inode->i_sb, KERN_ERR,
 929                          "Inode %lu (%p): orphan list check failed!",
 930                          inode->i_ino, EXT4_I(inode));
 931                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 932                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 933                                 true);
 934                 dump_stack();
 935         }
 936         call_rcu(&inode->i_rcu, ext4_i_callback);
 937 }
 938
 939 static void init_once(void *foo)
 940 {
 941         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 942
 943         INIT_LIST_HEAD(&ei->i_orphan);
 944         init_rwsem(&ei->xattr_sem);
 945         init_rwsem(&ei->i_data_sem);
 946         inode_init_once(&ei->vfs_inode);
 947 }
 948
 949 static int __init init_inodecache(void)
 950 {
 951         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 952                                              sizeof(struct ext4_inode_info),
 953                                              0, (SLAB_RECLAIM_ACCOUNT|
 954                                                 SLAB_MEM_SPREAD),
 955                                              init_once);
 956         if (ext4_inode_cachep == NULL)
 957                 return -ENOMEM;
 958         return 0;
 959 }
 960
 961 static void destroy_inodecache(void)
 962 {
 963         /*
 964          * Make sure all delayed rcu free inodes are flushed before we
 965          * destroy cache.
 966          */
 967         rcu_barrier();
 968         kmem_cache_destroy(ext4_inode_cachep);
 969 }
 970
 971 void ext4_clear_inode(struct inode *inode)
 972 {
 973         invalidate_inode_buffers(inode);
 974         clear_inode(inode);
 975         dquot_drop(inode);
 976         ext4_discard_preallocations(inode);
 977         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 978         ext4_es_lru_del(inode);
 979         if (EXT4_I(inode)->jinode) {
 980                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 981                                                EXT4_I(inode)->jinode);
 982                 jbd2_free_inode(EXT4_I(inode)->jinode);
 983                 EXT4_I(inode)->jinode = NULL;
 984         }
 985 }
 986
 987 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 988                                         u64 ino, u32 generation)
 989 {
 990         struct inode *inode;
 991
 992         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 993                 return ERR_PTR(-ESTALE);
 994         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 995                 return ERR_PTR(-ESTALE);
 996
 997         /* iget isn't really right if the inode is currently unallocated!!
 998          *
 999          * ext4_read_inode will return a bad_inode if the inode had been
1000          * deleted, so we should be safe.
1001          *
1002          * Currently we don't know the generation for parent directory, so
1003          * a generation of 0 means "accept any"
1004          */
1005         inode = ext4_iget(sb, ino);
1006         if (IS_ERR(inode))
1007                 return ERR_CAST(inode);
1008         if (generation && inode->i_generation != generation) {
1009                 iput(inode);
1010                 return ERR_PTR(-ESTALE);
1011         }
1012
1013         return inode;
1014 }
1015
1016 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1017                                         int fh_len, int fh_type)
1018 {
1019         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1020                                     ext4_nfs_get_inode);
1021 }
1022
1023 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1024                                         int fh_len, int fh_type)
1025 {
1026         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1027                                     ext4_nfs_get_inode);
1028 }
1029
1030 /*
1031  * Try to release metadata pages (indirect blocks, directories) which are
1032  * mapped via the block device.  Since these pages could have journal heads
1033  * which would prevent try_to_free_buffers() from freeing them, we must use
1034  * jbd2 layer's try_to_free_buffers() function to release them.
1035  */
1036 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1037                                  gfp_t wait)
1038 {
1039         journal_t *journal = EXT4_SB(sb)->s_journal;
1040
1041         WARN_ON(PageChecked(page));
1042         if (!page_has_buffers(page))
1043                 return 0;
1044         if (journal)
1045                 return jbd2_journal_try_to_free_buffers(journal, page,
1046                                                         wait & ~__GFP_WAIT);
1047         return try_to_free_buffers(page);
1048 }
1049
1050 #ifdef CONFIG_QUOTA
1051 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1052 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1053
1054 static int ext4_write_dquot(struct dquot *dquot);
1055 static int ext4_acquire_dquot(struct dquot *dquot);
1056 static int ext4_release_dquot(struct dquot *dquot);
1057 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1058 static int ext4_write_info(struct super_block *sb, int type);
1059 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1060                          struct path *path);
1061 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1062                                  int format_id);
1063 static int ext4_quota_off(struct super_block *sb, int type);
1064 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1065 static int ext4_quota_on_mount(struct super_block *sb, int type);
1066 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1067                                size_t len, loff_t off);
1068 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1069                                 const char *data, size_t len, loff_t off);
1070 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1071                              unsigned int flags);
1072 static int ext4_enable_quotas(struct super_block *sb);
1073
1074 static const struct dquot_operations ext4_quota_operations = {
1075         .get_reserved_space = ext4_get_reserved_space,
1076         .write_dquot    = ext4_write_dquot,
1077         .acquire_dquot  = ext4_acquire_dquot,
1078         .release_dquot  = ext4_release_dquot,
1079         .mark_dirty     = ext4_mark_dquot_dirty,
1080         .write_info     = ext4_write_info,
1081         .alloc_dquot    = dquot_alloc,
1082         .destroy_dquot  = dquot_destroy,
1083 };
1084
1085 static const struct quotactl_ops ext4_qctl_operations = {
1086         .quota_on       = ext4_quota_on,
1087         .quota_off      = ext4_quota_off,
1088         .quota_sync     = dquot_quota_sync,
1089         .get_info       = dquot_get_dqinfo,
1090         .set_info       = dquot_set_dqinfo,
1091         .get_dqblk      = dquot_get_dqblk,
1092         .set_dqblk      = dquot_set_dqblk
1093 };
1094
1095 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1096         .quota_on_meta  = ext4_quota_on_sysfile,
1097         .quota_off      = ext4_quota_off_sysfile,
1098         .quota_sync     = dquot_quota_sync,
1099         .get_info       = dquot_get_dqinfo,
1100         .set_info       = dquot_set_dqinfo,
1101         .get_dqblk      = dquot_get_dqblk,
1102         .set_dqblk      = dquot_set_dqblk
1103 };
1104 #endif
1105
1106 static const struct super_operations ext4_sops = {
1107         .alloc_inode    = ext4_alloc_inode,
1108         .destroy_inode  = ext4_destroy_inode,
1109         .write_inode    = ext4_write_inode,
1110         .dirty_inode    = ext4_dirty_inode,
1111         .drop_inode     = ext4_drop_inode,
1112         .evict_inode    = ext4_evict_inode,
1113         .put_super      = ext4_put_super,
1114         .sync_fs        = ext4_sync_fs,
1115         .freeze_fs      = ext4_freeze,
1116         .unfreeze_fs    = ext4_unfreeze,
1117         .statfs         = ext4_statfs,
1118         .remount_fs     = ext4_remount,
1119         .show_options   = ext4_show_options,
1120 #ifdef CONFIG_QUOTA
1121         .quota_read     = ext4_quota_read,
1122         .quota_write    = ext4_quota_write,
1123 #endif
1124         .bdev_try_to_free_page = bdev_try_to_free_page,
1125 };
1126
1127 static const struct super_operations ext4_nojournal_sops = {
1128         .alloc_inode    = ext4_alloc_inode,
1129         .destroy_inode  = ext4_destroy_inode,
1130         .write_inode    = ext4_write_inode,
1131         .dirty_inode    = ext4_dirty_inode,
1132         .drop_inode     = ext4_drop_inode,
1133         .evict_inode    = ext4_evict_inode,
1134         .sync_fs        = ext4_sync_fs_nojournal,
1135         .put_super      = ext4_put_super,
1136         .statfs         = ext4_statfs,
1137         .remount_fs     = ext4_remount,
1138         .show_options   = ext4_show_options,
1139 #ifdef CONFIG_QUOTA
1140         .quota_read     = ext4_quota_read,
1141         .quota_write    = ext4_quota_write,
1142 #endif
1143         .bdev_try_to_free_page = bdev_try_to_free_page,
1144 };
1145
1146 static const struct export_operations ext4_export_ops = {
1147         .fh_to_dentry = ext4_fh_to_dentry,
1148         .fh_to_parent = ext4_fh_to_parent,
1149         .get_parent = ext4_get_parent,
1150 };
1151
1152 enum {
1153         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1154         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1155         Opt_nouid32, Opt_debug, Opt_removed,
1156         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1157         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1158         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1159         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1160         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1161         Opt_data_err_abort, Opt_data_err_ignore,
1162         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1163         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1164         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1165         Opt_usrquota, Opt_grpquota, Opt_i_version,
1166         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1167         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1168         Opt_inode_readahead_blks, Opt_journal_ioprio,
1169         Opt_dioread_nolock, Opt_dioread_lock,
1170         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1171         Opt_max_dir_size_kb,
1172 };
1173
1174 static const match_table_t tokens = {
1175         {Opt_bsd_df, "bsddf"},
1176         {Opt_minix_df, "minixdf"},
1177         {Opt_grpid, "grpid"},
1178         {Opt_grpid, "bsdgroups"},
1179         {Opt_nogrpid, "nogrpid"},
1180         {Opt_nogrpid, "sysvgroups"},
1181         {Opt_resgid, "resgid=%u"},
1182         {Opt_resuid, "resuid=%u"},
1183         {Opt_sb, "sb=%u"},
1184         {Opt_err_cont, "errors=continue"},
1185         {Opt_err_panic, "errors=panic"},
1186         {Opt_err_ro, "errors=remount-ro"},
1187         {Opt_nouid32, "nouid32"},
1188         {Opt_debug, "debug"},
1189         {Opt_removed, "oldalloc"},
1190         {Opt_removed, "orlov"},
1191         {Opt_user_xattr, "user_xattr"},
1192         {Opt_nouser_xattr, "nouser_xattr"},
1193         {Opt_acl, "acl"},
1194         {Opt_noacl, "noacl"},
1195         {Opt_noload, "norecovery"},
1196         {Opt_noload, "noload"},
1197         {Opt_removed, "nobh"},
1198         {Opt_removed, "bh"},
1199         {Opt_commit, "commit=%u"},
1200         {Opt_min_batch_time, "min_batch_time=%u"},
1201         {Opt_max_batch_time, "max_batch_time=%u"},
1202         {Opt_journal_dev, "journal_dev=%u"},
1203         {Opt_journal_path, "journal_path=%s"},
1204         {Opt_journal_checksum, "journal_checksum"},
1205         {Opt_journal_async_commit, "journal_async_commit"},
1206         {Opt_abort, "abort"},
1207         {Opt_data_journal, "data=journal"},
1208         {Opt_data_ordered, "data=ordered"},
1209         {Opt_data_writeback, "data=writeback"},
1210         {Opt_data_err_abort, "data_err=abort"},
1211         {Opt_data_err_ignore, "data_err=ignore"},
1212         {Opt_offusrjquota, "usrjquota="},
1213         {Opt_usrjquota, "usrjquota=%s"},
1214         {Opt_offgrpjquota, "grpjquota="},
1215         {Opt_grpjquota, "grpjquota=%s"},
1216         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1217         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1218         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1219         {Opt_grpquota, "grpquota"},
1220         {Opt_noquota, "noquota"},
1221         {Opt_quota, "quota"},
1222         {Opt_usrquota, "usrquota"},
1223         {Opt_barrier, "barrier=%u"},
1224         {Opt_barrier, "barrier"},
1225         {Opt_nobarrier, "nobarrier"},
1226         {Opt_i_version, "i_version"},
1227         {Opt_stripe, "stripe=%u"},
1228         {Opt_delalloc, "delalloc"},
1229         {Opt_nodelalloc, "nodelalloc"},
1230         {Opt_removed, "mblk_io_submit"},
1231         {Opt_removed, "nomblk_io_submit"},
1232         {Opt_block_validity, "block_validity"},
1233         {Opt_noblock_validity, "noblock_validity"},
1234         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1235         {Opt_journal_ioprio, "journal_ioprio=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1237         {Opt_auto_da_alloc, "auto_da_alloc"},
1238         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1239         {Opt_dioread_nolock, "dioread_nolock"},
1240         {Opt_dioread_lock, "dioread_lock"},
1241         {Opt_discard, "discard"},
1242         {Opt_nodiscard, "nodiscard"},
1243         {Opt_init_itable, "init_itable=%u"},
1244         {Opt_init_itable, "init_itable"},
1245         {Opt_noinit_itable, "noinit_itable"},
1246         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1247         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1248         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1249         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1250         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1251         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1252         {Opt_err, NULL},
1253 };
1254
1255 static ext4_fsblk_t get_sb_block(void **data)
1256 {
1257         ext4_fsblk_t    sb_block;
1258         char            *options = (char *) *data;
1259
1260         if (!options || strncmp(options, "sb=", 3) != 0)
1261                 return 1;       /* Default location */
1262
1263         options += 3;
1264         /* TODO: use simple_strtoll with >32bit ext4 */
1265         sb_block = simple_strtoul(options, &options, 0);
1266         if (*options && *options != ',') {
1267                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1268                        (char *) *data);
1269                 return 1;
1270         }
1271         if (*options == ',')
1272                 options++;
1273         *data = (void *) options;
1274
1275         return sb_block;
1276 }
1277
1278 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1279 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1280         "Contact [email protected] if you think we should keep it.\n";
1281
1282 #ifdef CONFIG_QUOTA
1283 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1284 {
1285         struct ext4_sb_info *sbi = EXT4_SB(sb);
1286         char *qname;
1287         int ret = -1;
1288
1289         if (sb_any_quota_loaded(sb) &&
1290                 !sbi->s_qf_names[qtype]) {
1291                 ext4_msg(sb, KERN_ERR,
1292                         "Cannot change journaled "
1293                         "quota options when quota turned on");
1294                 return -1;
1295         }
1296         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1297                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1298                          "when QUOTA feature is enabled");
1299                 return -1;
1300         }
1301         qname = match_strdup(args);
1302         if (!qname) {
1303                 ext4_msg(sb, KERN_ERR,
1304                         "Not enough memory for storing quotafile name");
1305                 return -1;
1306         }
1307         if (sbi->s_qf_names[qtype]) {
1308                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1309                         ret = 1;
1310                 else
1311                         ext4_msg(sb, KERN_ERR,
1312                                  "%s quota file already specified",
1313                                  QTYPE2NAME(qtype));
1314                 goto errout;
1315         }
1316         if (strchr(qname, '/')) {
1317                 ext4_msg(sb, KERN_ERR,
1318                         "quotafile must be on filesystem root");
1319                 goto errout;
1320         }
1321         sbi->s_qf_names[qtype] = qname;
1322         set_opt(sb, QUOTA);
1323         return 1;
1324 errout:
1325         kfree(qname);
1326         return ret;
1327 }
1328
1329 static int clear_qf_name(struct super_block *sb, int qtype)
1330 {
1331
1332         struct ext4_sb_info *sbi = EXT4_SB(sb);
1333
1334         if (sb_any_quota_loaded(sb) &&
1335                 sbi->s_qf_names[qtype]) {
1336                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1337                         " when quota turned on");
1338                 return -1;
1339         }
1340         kfree(sbi->s_qf_names[qtype]);
1341         sbi->s_qf_names[qtype] = NULL;
1342         return 1;
1343 }
1344 #endif
1345
1346 #define MOPT_SET        0x0001
1347 #define MOPT_CLEAR      0x0002
1348 #define MOPT_NOSUPPORT  0x0004
1349 #define MOPT_EXPLICIT   0x0008
1350 #define MOPT_CLEAR_ERR  0x0010
1351 #define MOPT_GTE0       0x0020
1352 #ifdef CONFIG_QUOTA
1353 #define MOPT_Q          0
1354 #define MOPT_QFMT       0x0040
1355 #else
1356 #define MOPT_Q          MOPT_NOSUPPORT
1357 #define MOPT_QFMT       MOPT_NOSUPPORT
1358 #endif
1359 #define MOPT_DATAJ      0x0080
1360 #define MOPT_NO_EXT2    0x0100
1361 #define MOPT_NO_EXT3    0x0200
1362 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1363 #define MOPT_STRING     0x0400
1364
1365 static const struct mount_opts {
1366         int     token;
1367         int     mount_opt;
1368         int     flags;
1369 } ext4_mount_opts[] = {
1370         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1371         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1372         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1373         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1374         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1375         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1376         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1377          MOPT_EXT4_ONLY | MOPT_SET},
1378         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1379          MOPT_EXT4_ONLY | MOPT_CLEAR},
1380         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1381         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1382         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1383          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1384         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1385          MOPT_EXT4_ONLY | MOPT_CLEAR},
1386         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1387          MOPT_EXT4_ONLY | MOPT_SET},
1388         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1389                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1390          MOPT_EXT4_ONLY | MOPT_SET},
1391         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1392         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1393         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1394         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1396          MOPT_NO_EXT2 | MOPT_SET},
1397         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1398          MOPT_NO_EXT2 | MOPT_CLEAR},
1399         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1400         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1401         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1402         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1403         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1404         {Opt_commit, 0, MOPT_GTE0},
1405         {Opt_max_batch_time, 0, MOPT_GTE0},
1406         {Opt_min_batch_time, 0, MOPT_GTE0},
1407         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1408         {Opt_init_itable, 0, MOPT_GTE0},
1409         {Opt_stripe, 0, MOPT_GTE0},
1410         {Opt_resuid, 0, MOPT_GTE0},
1411         {Opt_resgid, 0, MOPT_GTE0},
1412         {Opt_journal_dev, 0, MOPT_GTE0},
1413         {Opt_journal_path, 0, MOPT_STRING},
1414         {Opt_journal_ioprio, 0, MOPT_GTE0},
1415         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1416         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1417         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1418          MOPT_NO_EXT2 | MOPT_DATAJ},
1419         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1420         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1421 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1422         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1423         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1424 #else
1425         {Opt_acl, 0, MOPT_NOSUPPORT},
1426         {Opt_noacl, 0, MOPT_NOSUPPORT},
1427 #endif
1428         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1429         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1430         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1431         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1432                                                         MOPT_SET | MOPT_Q},
1433         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1434                                                         MOPT_SET | MOPT_Q},
1435         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1436                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1437         {Opt_usrjquota, 0, MOPT_Q},
1438         {Opt_grpjquota, 0, MOPT_Q},
1439         {Opt_offusrjquota, 0, MOPT_Q},
1440         {Opt_offgrpjquota, 0, MOPT_Q},
1441         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1442         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1443         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1444         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1445         {Opt_err, 0, 0}
1446 };
1447
1448 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1449                             substring_t *args, unsigned long *journal_devnum,
1450                             unsigned int *journal_ioprio, int is_remount)
1451 {
1452         struct ext4_sb_info *sbi = EXT4_SB(sb);
1453         const struct mount_opts *m;
1454         kuid_t uid;
1455         kgid_t gid;
1456         int arg = 0;
1457
1458 #ifdef CONFIG_QUOTA
1459         if (token == Opt_usrjquota)
1460                 return set_qf_name(sb, USRQUOTA, &args[0]);
1461         else if (token == Opt_grpjquota)
1462                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1463         else if (token == Opt_offusrjquota)
1464                 return clear_qf_name(sb, USRQUOTA);
1465         else if (token == Opt_offgrpjquota)
1466                 return clear_qf_name(sb, GRPQUOTA);
1467 #endif
1468         switch (token) {
1469         case Opt_noacl:
1470         case Opt_nouser_xattr:
1471                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1472                 break;
1473         case Opt_sb:
1474                 return 1;       /* handled by get_sb_block() */
1475         case Opt_removed:
1476                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1477                 return 1;
1478         case Opt_abort:
1479                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1480                 return 1;
1481         case Opt_i_version:
1482                 sb->s_flags |= MS_I_VERSION;
1483                 return 1;
1484         }
1485
1486         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1487                 if (token == m->token)
1488                         break;
1489
1490         if (m->token == Opt_err) {
1491                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1492                          "or missing value", opt);
1493                 return -1;
1494         }
1495
1496         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1497                 ext4_msg(sb, KERN_ERR,
1498                          "Mount option \"%s\" incompatible with ext2", opt);
1499                 return -1;
1500         }
1501         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1502                 ext4_msg(sb, KERN_ERR,
1503                          "Mount option \"%s\" incompatible with ext3", opt);
1504                 return -1;
1505         }
1506
1507         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1508                 return -1;
1509         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1510                 return -1;
1511         if (m->flags & MOPT_EXPLICIT)
1512                 set_opt2(sb, EXPLICIT_DELALLOC);
1513         if (m->flags & MOPT_CLEAR_ERR)
1514                 clear_opt(sb, ERRORS_MASK);
1515         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1516                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1517                          "options when quota turned on");
1518                 return -1;
1519         }
1520
1521         if (m->flags & MOPT_NOSUPPORT) {
1522                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1523         } else if (token == Opt_commit) {
1524                 if (arg == 0)
1525                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1526                 sbi->s_commit_interval = HZ * arg;
1527         } else if (token == Opt_max_batch_time) {
1528                 if (arg == 0)
1529                         arg = EXT4_DEF_MAX_BATCH_TIME;
1530                 sbi->s_max_batch_time = arg;
1531         } else if (token == Opt_min_batch_time) {
1532                 sbi->s_min_batch_time = arg;
1533         } else if (token == Opt_inode_readahead_blks) {
1534                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1535                         ext4_msg(sb, KERN_ERR,
1536                                  "EXT4-fs: inode_readahead_blks must be "
1537                                  "0 or a power of 2 smaller than 2^31");
1538                         return -1;
1539                 }
1540                 sbi->s_inode_readahead_blks = arg;
1541         } else if (token == Opt_init_itable) {
1542                 set_opt(sb, INIT_INODE_TABLE);
1543                 if (!args->from)
1544                         arg = EXT4_DEF_LI_WAIT_MULT;
1545                 sbi->s_li_wait_mult = arg;
1546         } else if (token == Opt_max_dir_size_kb) {
1547                 sbi->s_max_dir_size_kb = arg;
1548         } else if (token == Opt_stripe) {
1549                 sbi->s_stripe = arg;
1550         } else if (token == Opt_resuid) {
1551                 uid = make_kuid(current_user_ns(), arg);
1552                 if (!uid_valid(uid)) {
1553                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1554                         return -1;
1555                 }
1556                 sbi->s_resuid = uid;
1557         } else if (token == Opt_resgid) {
1558                 gid = make_kgid(current_user_ns(), arg);
1559                 if (!gid_valid(gid)) {
1560                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1561                         return -1;
1562                 }
1563                 sbi->s_resgid = gid;
1564         } else if (token == Opt_journal_dev) {
1565                 if (is_remount) {
1566                         ext4_msg(sb, KERN_ERR,
1567                                  "Cannot specify journal on remount");
1568                         return -1;
1569                 }
1570                 *journal_devnum = arg;
1571         } else if (token == Opt_journal_path) {
1572                 char *journal_path;
1573                 struct inode *journal_inode;
1574                 struct path path;
1575                 int error;
1576
1577                 if (is_remount) {
1578                         ext4_msg(sb, KERN_ERR,
1579                                  "Cannot specify journal on remount");
1580                         return -1;
1581                 }
1582                 journal_path = match_strdup(&args[0]);
1583                 if (!journal_path) {
1584                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1585                                 "journal device string");
1586                         return -1;
1587                 }
1588
1589                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1590                 if (error) {
1591                         ext4_msg(sb, KERN_ERR, "error: could not find "
1592                                 "journal device path: error %d", error);
1593                         kfree(journal_path);
1594                         return -1;
1595                 }
1596
1597                 journal_inode = path.dentry->d_inode;
1598                 if (!S_ISBLK(journal_inode->i_mode)) {
1599                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1600                                 "is not a block device", journal_path);
1601                         path_put(&path);
1602                         kfree(journal_path);
1603                         return -1;
1604                 }
1605
1606                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1607                 path_put(&path);
1608                 kfree(journal_path);
1609         } else if (token == Opt_journal_ioprio) {
1610                 if (arg > 7) {
1611                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1612                                  " (must be 0-7)");
1613                         return -1;
1614                 }
1615                 *journal_ioprio =
1616                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1617         } else if (m->flags & MOPT_DATAJ) {
1618                 if (is_remount) {
1619                         if (!sbi->s_journal)
1620                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1621                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1622                                 ext4_msg(sb, KERN_ERR,
1623                                          "Cannot change data mode on remount");
1624                                 return -1;
1625                         }
1626                 } else {
1627                         clear_opt(sb, DATA_FLAGS);
1628                         sbi->s_mount_opt |= m->mount_opt;
1629                 }
1630 #ifdef CONFIG_QUOTA
1631         } else if (m->flags & MOPT_QFMT) {
1632                 if (sb_any_quota_loaded(sb) &&
1633                     sbi->s_jquota_fmt != m->mount_opt) {
1634                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1635                                  "quota options when quota turned on");
1636                         return -1;
1637                 }
1638                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1639                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1640                         ext4_msg(sb, KERN_ERR,
1641                                  "Cannot set journaled quota options "
1642                                  "when QUOTA feature is enabled");
1643                         return -1;
1644                 }
1645                 sbi->s_jquota_fmt = m->mount_opt;
1646 #endif
1647         } else {
1648                 if (!args->from)
1649                         arg = 1;
1650                 if (m->flags & MOPT_CLEAR)
1651                         arg = !arg;
1652                 else if (unlikely(!(m->flags & MOPT_SET))) {
1653                         ext4_msg(sb, KERN_WARNING,
1654                                  "buggy handling of option %s", opt);
1655                         WARN_ON(1);
1656                         return -1;
1657                 }
1658                 if (arg != 0)
1659                         sbi->s_mount_opt |= m->mount_opt;
1660                 else
1661                         sbi->s_mount_opt &= ~m->mount_opt;
1662         }
1663         return 1;
1664 }
1665
1666 static int parse_options(char *options, struct super_block *sb,
1667                          unsigned long *journal_devnum,
1668                          unsigned int *journal_ioprio,
1669                          int is_remount)
1670 {
1671         struct ext4_sb_info *sbi = EXT4_SB(sb);
1672         char *p;
1673         substring_t args[MAX_OPT_ARGS];
1674         int token;
1675
1676         if (!options)
1677                 return 1;
1678
1679         while ((p = strsep(&options, ",")) != NULL) {
1680                 if (!*p)
1681                         continue;
1682                 /*
1683                  * Initialize args struct so we know whether arg was
1684                  * found; some options take optional arguments.
1685                  */
1686                 args[0].to = args[0].from = NULL;
1687                 token = match_token(p, tokens, args);
1688                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1689                                      journal_ioprio, is_remount) < 0)
1690                         return 0;
1691         }
1692 #ifdef CONFIG_QUOTA
1693         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1694             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1695                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1696                          "feature is enabled");
1697                 return 0;
1698         }
1699         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1700                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1701                         clear_opt(sb, USRQUOTA);
1702
1703                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1704                         clear_opt(sb, GRPQUOTA);
1705
1706                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1707                         ext4_msg(sb, KERN_ERR, "old and new quota "
1708                                         "format mixing");
1709                         return 0;
1710                 }
1711
1712                 if (!sbi->s_jquota_fmt) {
1713                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1714                                         "not specified");
1715                         return 0;
1716                 }
1717         } else {
1718                 if (sbi->s_jquota_fmt) {
1719                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1720                                         "specified with no journaling "
1721                                         "enabled");
1722                         return 0;
1723                 }
1724         }
1725 #endif
1726         if (test_opt(sb, DIOREAD_NOLOCK)) {
1727                 int blocksize =
1728                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1729
1730                 if (blocksize < PAGE_CACHE_SIZE) {
1731                         ext4_msg(sb, KERN_ERR, "can't mount with "
1732                                  "dioread_nolock if block size != PAGE_SIZE");
1733                         return 0;
1734                 }
1735         }
1736         return 1;
1737 }
1738
1739 static inline void ext4_show_quota_options(struct seq_file *seq,
1740                                            struct super_block *sb)
1741 {
1742 #if defined(CONFIG_QUOTA)
1743         struct ext4_sb_info *sbi = EXT4_SB(sb);
1744
1745         if (sbi->s_jquota_fmt) {
1746                 char *fmtname = "";
1747
1748                 switch (sbi->s_jquota_fmt) {
1749                 case QFMT_VFS_OLD:
1750                         fmtname = "vfsold";
1751                         break;
1752                 case QFMT_VFS_V0:
1753                         fmtname = "vfsv0";
1754                         break;
1755                 case QFMT_VFS_V1:
1756                         fmtname = "vfsv1";
1757                         break;
1758                 }
1759                 seq_printf(seq, ",jqfmt=%s", fmtname);
1760         }
1761
1762         if (sbi->s_qf_names[USRQUOTA])
1763                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1764
1765         if (sbi->s_qf_names[GRPQUOTA])
1766                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1767 #endif
1768 }
1769
1770 static const char *token2str(int token)
1771 {
1772         const struct match_token *t;
1773
1774         for (t = tokens; t->token != Opt_err; t++)
1775                 if (t->token == token && !strchr(t->pattern, '='))
1776                         break;
1777         return t->pattern;
1778 }
1779
1780 /*
1781  * Show an option if
1782  *  - it's set to a non-default value OR
1783  *  - if the per-sb default is different from the global default
1784  */
1785 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1786                               int nodefs)
1787 {
1788         struct ext4_sb_info *sbi = EXT4_SB(sb);
1789         struct ext4_super_block *es = sbi->s_es;
1790         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1791         const struct mount_opts *m;
1792         char sep = nodefs ? '\n' : ',';
1793
1794 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1795 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1796
1797         if (sbi->s_sb_block != 1)
1798                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1799
1800         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1801                 int want_set = m->flags & MOPT_SET;
1802                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1803                     (m->flags & MOPT_CLEAR_ERR))
1804                         continue;
1805                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1806                         continue; /* skip if same as the default */
1807                 if ((want_set &&
1808                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1809                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1810                         continue; /* select Opt_noFoo vs Opt_Foo */
1811                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1812         }
1813
1814         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1815             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1816                 SEQ_OPTS_PRINT("resuid=%u",
1817                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1818         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1819             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1820                 SEQ_OPTS_PRINT("resgid=%u",
1821                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1822         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1823         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1824                 SEQ_OPTS_PUTS("errors=remount-ro");
1825         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1826                 SEQ_OPTS_PUTS("errors=continue");
1827         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1828                 SEQ_OPTS_PUTS("errors=panic");
1829         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1830                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1831         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1832                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1833         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1834                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1835         if (sb->s_flags & MS_I_VERSION)
1836                 SEQ_OPTS_PUTS("i_version");
1837         if (nodefs || sbi->s_stripe)
1838                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1839         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1840                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1841                         SEQ_OPTS_PUTS("data=journal");
1842                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1843                         SEQ_OPTS_PUTS("data=ordered");
1844                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1845                         SEQ_OPTS_PUTS("data=writeback");
1846         }
1847         if (nodefs ||
1848             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1849                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1850                                sbi->s_inode_readahead_blks);
1851
1852         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1853                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1854                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1855         if (nodefs || sbi->s_max_dir_size_kb)
1856                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1857
1858         ext4_show_quota_options(seq, sb);
1859         return 0;
1860 }
1861
1862 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1863 {
1864         return _ext4_show_options(seq, root->d_sb, 0);
1865 }
1866
1867 static int options_seq_show(struct seq_file *seq, void *offset)
1868 {
1869         struct super_block *sb = seq->private;
1870         int rc;
1871
1872         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1873         rc = _ext4_show_options(seq, sb, 1);
1874         seq_puts(seq, "\n");
1875         return rc;
1876 }
1877
1878 static int options_open_fs(struct inode *inode, struct file *file)
1879 {
1880         return single_open(file, options_seq_show, PDE_DATA(inode));
1881 }
1882
1883 static const struct file_operations ext4_seq_options_fops = {
1884         .owner = THIS_MODULE,
1885         .open = options_open_fs,
1886         .read = seq_read,
1887         .llseek = seq_lseek,
1888         .release = single_release,
1889 };
1890
1891 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1892                             int read_only)
1893 {
1894         struct ext4_sb_info *sbi = EXT4_SB(sb);
1895         int res = 0;
1896
1897         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1898                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1899                          "forcing read-only mode");
1900                 res = MS_RDONLY;
1901         }
1902         if (read_only)
1903                 goto done;
1904         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1905                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1906                          "running e2fsck is recommended");
1907         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1908                 ext4_msg(sb, KERN_WARNING,
1909                          "warning: mounting fs with errors, "
1910                          "running e2fsck is recommended");
1911         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1912                  le16_to_cpu(es->s_mnt_count) >=
1913                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1914                 ext4_msg(sb, KERN_WARNING,
1915                          "warning: maximal mount count reached, "
1916                          "running e2fsck is recommended");
1917         else if (le32_to_cpu(es->s_checkinterval) &&
1918                 (le32_to_cpu(es->s_lastcheck) +
1919                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1920                 ext4_msg(sb, KERN_WARNING,
1921                          "warning: checktime reached, "
1922                          "running e2fsck is recommended");
1923         if (!sbi->s_journal)
1924                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1925         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1926                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1927         le16_add_cpu(&es->s_mnt_count, 1);
1928         es->s_mtime = cpu_to_le32(get_seconds());
1929         ext4_update_dynamic_rev(sb);
1930         if (sbi->s_journal)
1931                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1932
1933         ext4_commit_super(sb, 1);
1934 done:
1935         if (test_opt(sb, DEBUG))
1936                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1937                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1938                         sb->s_blocksize,
1939                         sbi->s_groups_count,
1940                         EXT4_BLOCKS_PER_GROUP(sb),
1941                         EXT4_INODES_PER_GROUP(sb),
1942                         sbi->s_mount_opt, sbi->s_mount_opt2);
1943
1944         cleancache_init_fs(sb);
1945         return res;
1946 }
1947
1948 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1949 {
1950         struct ext4_sb_info *sbi = EXT4_SB(sb);
1951         struct flex_groups *new_groups;
1952         int size;
1953
1954         if (!sbi->s_log_groups_per_flex)
1955                 return 0;
1956
1957         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1958         if (size <= sbi->s_flex_groups_allocated)
1959                 return 0;
1960
1961         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1962         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1963         if (!new_groups) {
1964                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1965                          size / (int) sizeof(struct flex_groups));
1966                 return -ENOMEM;
1967         }
1968
1969         if (sbi->s_flex_groups) {
1970                 memcpy(new_groups, sbi->s_flex_groups,
1971                        (sbi->s_flex_groups_allocated *
1972                         sizeof(struct flex_groups)));
1973                 ext4_kvfree(sbi->s_flex_groups);
1974         }
1975         sbi->s_flex_groups = new_groups;
1976         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1977         return 0;
1978 }
1979
1980 static int ext4_fill_flex_info(struct super_block *sb)
1981 {
1982         struct ext4_sb_info *sbi = EXT4_SB(sb);
1983         struct ext4_group_desc *gdp = NULL;
1984         ext4_group_t flex_group;
1985         int i, err;
1986
1987         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1988         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1989                 sbi->s_log_groups_per_flex = 0;
1990                 return 1;
1991         }
1992
1993         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1994         if (err)
1995                 goto failed;
1996
1997         for (i = 0; i < sbi->s_groups_count; i++) {
1998                 gdp = ext4_get_group_desc(sb, i, NULL);
1999
2000                 flex_group = ext4_flex_group(sbi, i);
2001                 atomic_add(ext4_free_inodes_count(sb, gdp),
2002                            &sbi->s_flex_groups[flex_group].free_inodes);
2003                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2004                              &sbi->s_flex_groups[flex_group].free_clusters);
2005                 atomic_add(ext4_used_dirs_count(sb, gdp),
2006                            &sbi->s_flex_groups[flex_group].used_dirs);
2007         }
2008
2009         return 1;
2010 failed:
2011         return 0;
2012 }
2013
2014 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2015                                    struct ext4_group_desc *gdp)
2016 {
2017         int offset;
2018         __u16 crc = 0;
2019         __le32 le_group = cpu_to_le32(block_group);
2020
2021         if ((sbi->s_es->s_feature_ro_compat &
2022              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2023                 /* Use new metadata_csum algorithm */
2024                 __le16 save_csum;
2025                 __u32 csum32;
2026
2027                 save_csum = gdp->bg_checksum;
2028                 gdp->bg_checksum = 0;
2029                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2030                                      sizeof(le_group));
2031                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2032                                      sbi->s_desc_size);
2033                 gdp->bg_checksum = save_csum;
2034
2035                 crc = csum32 & 0xFFFF;
2036                 goto out;
2037         }
2038
2039         /* old crc16 code */
2040         offset = offsetof(struct ext4_group_desc, bg_checksum);
2041
2042         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2043         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2044         crc = crc16(crc, (__u8 *)gdp, offset);
2045         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2046         /* for checksum of struct ext4_group_desc do the rest...*/
2047         if ((sbi->s_es->s_feature_incompat &
2048              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2049             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2050                 crc = crc16(crc, (__u8 *)gdp + offset,
2051                             le16_to_cpu(sbi->s_es->s_desc_size) -
2052                                 offset);
2053
2054 out:
2055         return cpu_to_le16(crc);
2056 }
2057
2058 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2059                                 struct ext4_group_desc *gdp)
2060 {
2061         if (ext4_has_group_desc_csum(sb) &&
2062             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2063                                                       block_group, gdp)))
2064                 return 0;
2065
2066         return 1;
2067 }
2068
2069 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2070                               struct ext4_group_desc *gdp)
2071 {
2072         if (!ext4_has_group_desc_csum(sb))
2073                 return;
2074         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2075 }
2076
2077 /* Called at mount-time, super-block is locked */
2078 static int ext4_check_descriptors(struct super_block *sb,
2079                                   ext4_group_t *first_not_zeroed)
2080 {
2081         struct ext4_sb_info *sbi = EXT4_SB(sb);
2082         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2083         ext4_fsblk_t last_block;
2084         ext4_fsblk_t block_bitmap;
2085         ext4_fsblk_t inode_bitmap;
2086         ext4_fsblk_t inode_table;
2087         int flexbg_flag = 0;
2088         ext4_group_t i, grp = sbi->s_groups_count;
2089
2090         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2091                 flexbg_flag = 1;
2092
2093         ext4_debug("Checking group descriptors");
2094
2095         for (i = 0; i < sbi->s_groups_count; i++) {
2096                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2097
2098                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2099                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2100                 else
2101                         last_block = first_block +
2102                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2103
2104                 if ((grp == sbi->s_groups_count) &&
2105                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2106                         grp = i;
2107
2108                 block_bitmap = ext4_block_bitmap(sb, gdp);
2109                 if (block_bitmap < first_block || block_bitmap > last_block) {
2110                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2111                                "Block bitmap for group %u not in group "
2112                                "(block %llu)!", i, block_bitmap);
2113                         return 0;
2114                 }
2115                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2116                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2117                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2118                                "Inode bitmap for group %u not in group "
2119                                "(block %llu)!", i, inode_bitmap);
2120                         return 0;
2121                 }
2122                 inode_table = ext4_inode_table(sb, gdp);
2123                 if (inode_table < first_block ||
2124                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2125                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2126                                "Inode table for group %u not in group "
2127                                "(block %llu)!", i, inode_table);
2128                         return 0;
2129                 }
2130                 ext4_lock_group(sb, i);
2131                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2132                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2133                                  "Checksum for group %u failed (%u!=%u)",
2134                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2135                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2136                         if (!(sb->s_flags & MS_RDONLY)) {
2137                                 ext4_unlock_group(sb, i);
2138                                 return 0;
2139                         }
2140                 }
2141                 ext4_unlock_group(sb, i);
2142                 if (!flexbg_flag)
2143                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2144         }
2145         if (NULL != first_not_zeroed)
2146                 *first_not_zeroed = grp;
2147
2148         ext4_free_blocks_count_set(sbi->s_es,
2149                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2150         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2151         return 1;
2152 }
2153
2154 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2155  * the superblock) which were deleted from all directories, but held open by
2156  * a process at the time of a crash.  We walk the list and try to delete these
2157  * inodes at recovery time (only with a read-write filesystem).
2158  *
2159  * In order to keep the orphan inode chain consistent during traversal (in
2160  * case of crash during recovery), we link each inode into the superblock
2161  * orphan list_head and handle it the same way as an inode deletion during
2162  * normal operation (which journals the operations for us).
2163  *
2164  * We only do an iget() and an iput() on each inode, which is very safe if we
2165  * accidentally point at an in-use or already deleted inode.  The worst that
2166  * can happen in this case is that we get a "bit already cleared" message from
2167  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2168  * e2fsck was run on this filesystem, and it must have already done the orphan
2169  * inode cleanup for us, so we can safely abort without any further action.
2170  */
2171 static void ext4_orphan_cleanup(struct super_block *sb,
2172                                 struct ext4_super_block *es)
2173 {
2174         unsigned int s_flags = sb->s_flags;
2175         int nr_orphans = 0, nr_truncates = 0;
2176 #ifdef CONFIG_QUOTA
2177         int i;
2178 #endif
2179         if (!es->s_last_orphan) {
2180                 jbd_debug(4, "no orphan inodes to clean up\n");
2181                 return;
2182         }
2183
2184         if (bdev_read_only(sb->s_bdev)) {
2185                 ext4_msg(sb, KERN_ERR, "write access "
2186                         "unavailable, skipping orphan cleanup");
2187                 return;
2188         }
2189
2190         /* Check if feature set would not allow a r/w mount */
2191         if (!ext4_feature_set_ok(sb, 0)) {
2192                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2193                          "unknown ROCOMPAT features");
2194                 return;
2195         }
2196
2197         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2198                 /* don't clear list on RO mount w/ errors */
2199                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2200                         jbd_debug(1, "Errors on filesystem, "
2201                                   "clearing orphan list.\n");
2202                         es->s_last_orphan = 0;
2203                 }
2204                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2205                 return;
2206         }
2207
2208         if (s_flags & MS_RDONLY) {
2209                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2210                 sb->s_flags &= ~MS_RDONLY;
2211         }
2212 #ifdef CONFIG_QUOTA
2213         /* Needed for iput() to work correctly and not trash data */
2214         sb->s_flags |= MS_ACTIVE;
2215         /* Turn on quotas so that they are updated correctly */
2216         for (i = 0; i < MAXQUOTAS; i++) {
2217                 if (EXT4_SB(sb)->s_qf_names[i]) {
2218                         int ret = ext4_quota_on_mount(sb, i);
2219                         if (ret < 0)
2220                                 ext4_msg(sb, KERN_ERR,
2221                                         "Cannot turn on journaled "
2222                                         "quota: error %d", ret);
2223                 }
2224         }
2225 #endif
2226
2227         while (es->s_last_orphan) {
2228                 struct inode *inode;
2229
2230                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2231                 if (IS_ERR(inode)) {
2232                         es->s_last_orphan = 0;
2233                         break;
2234                 }
2235
2236                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2237                 dquot_initialize(inode);
2238                 if (inode->i_nlink) {
2239                         if (test_opt(sb, DEBUG))
2240                                 ext4_msg(sb, KERN_DEBUG,
2241                                         "%s: truncating inode %lu to %lld bytes",
2242                                         __func__, inode->i_ino, inode->i_size);
2243                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2244                                   inode->i_ino, inode->i_size);
2245                         mutex_lock(&inode->i_mutex);
2246                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2247                         ext4_truncate(inode);
2248                         mutex_unlock(&inode->i_mutex);
2249                         nr_truncates++;
2250                 } else {
2251                         if (test_opt(sb, DEBUG))
2252                                 ext4_msg(sb, KERN_DEBUG,
2253                                         "%s: deleting unreferenced inode %lu",
2254                                         __func__, inode->i_ino);
2255                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2256                                   inode->i_ino);
2257                         nr_orphans++;
2258                 }
2259                 iput(inode);  /* The delete magic happens here! */
2260         }
2261
2262 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2263
2264         if (nr_orphans)
2265                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2266                        PLURAL(nr_orphans));
2267         if (nr_truncates)
2268                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2269                        PLURAL(nr_truncates));
2270 #ifdef CONFIG_QUOTA
2271         /* Turn quotas off */
2272         for (i = 0; i < MAXQUOTAS; i++) {
2273                 if (sb_dqopt(sb)->files[i])
2274                         dquot_quota_off(sb, i);
2275         }
2276 #endif
2277         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2278 }
2279
2280 /*
2281  * Maximal extent format file size.
2282  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2283  * extent format containers, within a sector_t, and within i_blocks
2284  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2285  * so that won't be a limiting factor.
2286  *
2287  * However there is other limiting factor. We do store extents in the form
2288  * of starting block and length, hence the resulting length of the extent
2289  * covering maximum file size must fit into on-disk format containers as
2290  * well. Given that length is always by 1 unit bigger than max unit (because
2291  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2292  *
2293  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2294  */
2295 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2296 {
2297         loff_t res;
2298         loff_t upper_limit = MAX_LFS_FILESIZE;
2299
2300         /* small i_blocks in vfs inode? */
2301         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2302                 /*
2303                  * CONFIG_LBDAF is not enabled implies the inode
2304                  * i_block represent total blocks in 512 bytes
2305                  * 32 == size of vfs inode i_blocks * 8
2306                  */
2307                 upper_limit = (1LL << 32) - 1;
2308
2309                 /* total blocks in file system block size */
2310                 upper_limit >>= (blkbits - 9);
2311                 upper_limit <<= blkbits;
2312         }
2313
2314         /*
2315          * 32-bit extent-start container, ee_block. We lower the maxbytes
2316          * by one fs block, so ee_len can cover the extent of maximum file
2317          * size
2318          */
2319         res = (1LL << 32) - 1;
2320         res <<= blkbits;
2321
2322         /* Sanity check against vm- & vfs- imposed limits */
2323         if (res > upper_limit)
2324                 res = upper_limit;
2325
2326         return res;
2327 }
2328
2329 /*
2330  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2331  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2332  * We need to be 1 filesystem block less than the 2^48 sector limit.
2333  */
2334 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2335 {
2336         loff_t res = EXT4_NDIR_BLOCKS;
2337         int meta_blocks;
2338         loff_t upper_limit;
2339         /* This is calculated to be the largest file size for a dense, block
2340          * mapped file such that the file's total number of 512-byte sectors,
2341          * including data and all indirect blocks, does not exceed (2^48 - 1).
2342          *
2343          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2344          * number of 512-byte sectors of the file.
2345          */
2346
2347         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2348                 /*
2349                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2350                  * the inode i_block field represents total file blocks in
2351                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2352                  */
2353                 upper_limit = (1LL << 32) - 1;
2354
2355                 /* total blocks in file system block size */
2356                 upper_limit >>= (bits - 9);
2357
2358         } else {
2359                 /*
2360                  * We use 48 bit ext4_inode i_blocks
2361                  * With EXT4_HUGE_FILE_FL set the i_blocks
2362                  * represent total number of blocks in
2363                  * file system block size
2364                  */
2365                 upper_limit = (1LL << 48) - 1;
2366
2367         }
2368
2369         /* indirect blocks */
2370         meta_blocks = 1;
2371         /* double indirect blocks */
2372         meta_blocks += 1 + (1LL << (bits-2));
2373         /* tripple indirect blocks */
2374         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2375
2376         upper_limit -= meta_blocks;
2377         upper_limit <<= bits;
2378
2379         res += 1LL << (bits-2);
2380         res += 1LL << (2*(bits-2));
2381         res += 1LL << (3*(bits-2));
2382         res <<= bits;
2383         if (res > upper_limit)
2384                 res = upper_limit;
2385
2386         if (res > MAX_LFS_FILESIZE)
2387                 res = MAX_LFS_FILESIZE;
2388
2389         return res;
2390 }
2391
2392 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2393                                    ext4_fsblk_t logical_sb_block, int nr)
2394 {
2395         struct ext4_sb_info *sbi = EXT4_SB(sb);
2396         ext4_group_t bg, first_meta_bg;
2397         int has_super = 0;
2398
2399         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2400
2401         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2402             nr < first_meta_bg)
2403                 return logical_sb_block + nr + 1;
2404         bg = sbi->s_desc_per_block * nr;
2405         if (ext4_bg_has_super(sb, bg))
2406                 has_super = 1;
2407
2408         /*
2409          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2410          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2411          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2412          * compensate.
2413          */
2414         if (sb->s_blocksize == 1024 && nr == 0 &&
2415             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2416                 has_super++;
2417
2418         return (has_super + ext4_group_first_block_no(sb, bg));
2419 }
2420
2421 /**
2422  * ext4_get_stripe_size: Get the stripe size.
2423  * @sbi: In memory super block info
2424  *
2425  * If we have specified it via mount option, then
2426  * use the mount option value. If the value specified at mount time is
2427  * greater than the blocks per group use the super block value.
2428  * If the super block value is greater than blocks per group return 0.
2429  * Allocator needs it be less than blocks per group.
2430  *
2431  */
2432 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2433 {
2434         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2435         unsigned long stripe_width =
2436                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2437         int ret;
2438
2439         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2440                 ret = sbi->s_stripe;
2441         else if (stripe_width <= sbi->s_blocks_per_group)
2442                 ret = stripe_width;
2443         else if (stride <= sbi->s_blocks_per_group)
2444                 ret = stride;
2445         else
2446                 ret = 0;
2447
2448         /*
2449          * If the stripe width is 1, this makes no sense and
2450          * we set it to 0 to turn off stripe handling code.
2451          */
2452         if (ret <= 1)
2453                 ret = 0;
2454
2455         return ret;
2456 }
2457
2458 /* sysfs supprt */
2459
2460 struct ext4_attr {
2461         struct attribute attr;
2462         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2463         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2464                          const char *, size_t);
2465         union {
2466                 int offset;
2467                 int deprecated_val;
2468         } u;
2469 };
2470
2471 static int parse_strtoull(const char *buf,
2472                 unsigned long long max, unsigned long long *value)
2473 {
2474         int ret;
2475
2476         ret = kstrtoull(skip_spaces(buf), 0, value);
2477         if (!ret && *value > max)
2478                 ret = -EINVAL;
2479         return ret;
2480 }
2481
2482 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2483                                               struct ext4_sb_info *sbi,
2484                                               char *buf)
2485 {
2486         return snprintf(buf, PAGE_SIZE, "%llu\n",
2487                 (s64) EXT4_C2B(sbi,
2488                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2489 }
2490
2491 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2492                                          struct ext4_sb_info *sbi, char *buf)
2493 {
2494         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2495
2496         if (!sb->s_bdev->bd_part)
2497                 return snprintf(buf, PAGE_SIZE, "0\n");
2498         return snprintf(buf, PAGE_SIZE, "%lu\n",
2499                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2500                          sbi->s_sectors_written_start) >> 1);
2501 }
2502
2503 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2504                                           struct ext4_sb_info *sbi, char *buf)
2505 {
2506         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2507
2508         if (!sb->s_bdev->bd_part)
2509                 return snprintf(buf, PAGE_SIZE, "0\n");
2510         return snprintf(buf, PAGE_SIZE, "%llu\n",
2511                         (unsigned long long)(sbi->s_kbytes_written +
2512                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2513                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2514 }
2515
2516 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2517                                           struct ext4_sb_info *sbi,
2518                                           const char *buf, size_t count)
2519 {
2520         unsigned long t;
2521         int ret;
2522
2523         ret = kstrtoul(skip_spaces(buf), 0, &t);
2524         if (ret)
2525                 return ret;
2526
2527         if (t && (!is_power_of_2(t) || t > 0x40000000))
2528                 return -EINVAL;
2529
2530         sbi->s_inode_readahead_blks = t;
2531         return count;
2532 }
2533
2534 static ssize_t sbi_ui_show(struct ext4_attr *a,
2535                            struct ext4_sb_info *sbi, char *buf)
2536 {
2537         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2538
2539         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2540 }
2541
2542 static ssize_t sbi_ui_store(struct ext4_attr *a,
2543                             struct ext4_sb_info *sbi,
2544                             const char *buf, size_t count)
2545 {
2546         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2547         unsigned long t;
2548         int ret;
2549
2550         ret = kstrtoul(skip_spaces(buf), 0, &t);
2551         if (ret)
2552                 return ret;
2553         *ui = t;
2554         return count;
2555 }
2556
2557 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2558                                   struct ext4_sb_info *sbi, char *buf)
2559 {
2560         return snprintf(buf, PAGE_SIZE, "%llu\n",
2561                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2562 }
2563
2564 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2565                                    struct ext4_sb_info *sbi,
2566                                    const char *buf, size_t count)
2567 {
2568         unsigned long long val;
2569         int ret;
2570
2571         if (parse_strtoull(buf, -1ULL, &val))
2572                 return -EINVAL;
2573         ret = ext4_reserve_clusters(sbi, val);
2574
2575         return ret ? ret : count;
2576 }
2577
2578 static ssize_t trigger_test_error(struct ext4_attr *a,
2579                                   struct ext4_sb_info *sbi,
2580                                   const char *buf, size_t count)
2581 {
2582         int len = count;
2583
2584         if (!capable(CAP_SYS_ADMIN))
2585                 return -EPERM;
2586
2587         if (len && buf[len-1] == '\n')
2588                 len--;
2589
2590         if (len)
2591                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2592         return count;
2593 }
2594
2595 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2596                                    struct ext4_sb_info *sbi, char *buf)
2597 {
2598         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2599 }
2600
2601 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2602 static struct ext4_attr ext4_attr_##_name = {                   \
2603         .attr = {.name = __stringify(_name), .mode = _mode },   \
2604         .show   = _show,                                        \
2605         .store  = _store,                                       \
2606         .u = {                                                  \
2607                 .offset = offsetof(struct ext4_sb_info, _elname),\
2608         },                                                      \
2609 }
2610 #define EXT4_ATTR(name, mode, show, store) \
2611 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2612
2613 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2614 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2615 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2616 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2617         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2618 #define ATTR_LIST(name) &ext4_attr_##name.attr
2619 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2620 static struct ext4_attr ext4_attr_##_name = {                   \
2621         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2622         .show   = sbi_deprecated_show,                          \
2623         .u = {                                                  \
2624                 .deprecated_val = _val,                         \
2625         },                                                      \
2626 }
2627
2628 EXT4_RO_ATTR(delayed_allocation_blocks);
2629 EXT4_RO_ATTR(session_write_kbytes);
2630 EXT4_RO_ATTR(lifetime_write_kbytes);
2631 EXT4_RW_ATTR(reserved_clusters);
2632 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2633                  inode_readahead_blks_store, s_inode_readahead_blks);
2634 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2635 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2636 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2637 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2638 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2639 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2640 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2641 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2642 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2643 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2644 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2645 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2646 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2647 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2648 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2649 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2650
2651 static struct attribute *ext4_attrs[] = {
2652         ATTR_LIST(delayed_allocation_blocks),
2653         ATTR_LIST(session_write_kbytes),
2654         ATTR_LIST(lifetime_write_kbytes),
2655         ATTR_LIST(reserved_clusters),
2656         ATTR_LIST(inode_readahead_blks),
2657         ATTR_LIST(inode_goal),
2658         ATTR_LIST(mb_stats),
2659         ATTR_LIST(mb_max_to_scan),
2660         ATTR_LIST(mb_min_to_scan),
2661         ATTR_LIST(mb_order2_req),
2662         ATTR_LIST(mb_stream_req),
2663         ATTR_LIST(mb_group_prealloc),
2664         ATTR_LIST(max_writeback_mb_bump),
2665         ATTR_LIST(extent_max_zeroout_kb),
2666         ATTR_LIST(trigger_fs_error),
2667         ATTR_LIST(err_ratelimit_interval_ms),
2668         ATTR_LIST(err_ratelimit_burst),
2669         ATTR_LIST(warning_ratelimit_interval_ms),
2670         ATTR_LIST(warning_ratelimit_burst),
2671         ATTR_LIST(msg_ratelimit_interval_ms),
2672         ATTR_LIST(msg_ratelimit_burst),
2673         NULL,
2674 };
2675
2676 /* Features this copy of ext4 supports */
2677 EXT4_INFO_ATTR(lazy_itable_init);
2678 EXT4_INFO_ATTR(batched_discard);
2679 EXT4_INFO_ATTR(meta_bg_resize);
2680
2681 static struct attribute *ext4_feat_attrs[] = {
2682         ATTR_LIST(lazy_itable_init),
2683         ATTR_LIST(batched_discard),
2684         ATTR_LIST(meta_bg_resize),
2685         NULL,
2686 };
2687
2688 static ssize_t ext4_attr_show(struct kobject *kobj,
2689                               struct attribute *attr, char *buf)
2690 {
2691         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2692                                                 s_kobj);
2693         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2694
2695         return a->show ? a->show(a, sbi, buf) : 0;
2696 }
2697
2698 static ssize_t ext4_attr_store(struct kobject *kobj,
2699                                struct attribute *attr,
2700                                const char *buf, size_t len)
2701 {
2702         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2703                                                 s_kobj);
2704         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2705
2706         return a->store ? a->store(a, sbi, buf, len) : 0;
2707 }
2708
2709 static void ext4_sb_release(struct kobject *kobj)
2710 {
2711         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2712                                                 s_kobj);
2713         complete(&sbi->s_kobj_unregister);
2714 }
2715
2716 static const struct sysfs_ops ext4_attr_ops = {
2717         .show   = ext4_attr_show,
2718         .store  = ext4_attr_store,
2719 };
2720
2721 static struct kobj_type ext4_ktype = {
2722         .default_attrs  = ext4_attrs,
2723         .sysfs_ops      = &ext4_attr_ops,
2724         .release        = ext4_sb_release,
2725 };
2726
2727 static void ext4_feat_release(struct kobject *kobj)
2728 {
2729         complete(&ext4_feat->f_kobj_unregister);
2730 }
2731
2732 static struct kobj_type ext4_feat_ktype = {
2733         .default_attrs  = ext4_feat_attrs,
2734         .sysfs_ops      = &ext4_attr_ops,
2735         .release        = ext4_feat_release,
2736 };
2737
2738 /*
2739  * Check whether this filesystem can be mounted based on
2740  * the features present and the RDONLY/RDWR mount requested.
2741  * Returns 1 if this filesystem can be mounted as requested,
2742  * 0 if it cannot be.
2743  */
2744 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2745 {
2746         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2747                 ext4_msg(sb, KERN_ERR,
2748                         "Couldn't mount because of "
2749                         "unsupported optional features (%x)",
2750                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2751                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2752                 return 0;
2753         }
2754
2755         if (readonly)
2756                 return 1;
2757
2758         /* Check that feature set is OK for a read-write mount */
2759         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2760                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2761                          "unsupported optional features (%x)",
2762                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2763                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2764                 return 0;
2765         }
2766         /*
2767          * Large file size enabled file system can only be mounted
2768          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2769          */
2770         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2771                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2772                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2773                                  "cannot be mounted RDWR without "
2774                                  "CONFIG_LBDAF");
2775                         return 0;
2776                 }
2777         }
2778         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2779             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2780                 ext4_msg(sb, KERN_ERR,
2781                          "Can't support bigalloc feature without "
2782                          "extents feature\n");
2783                 return 0;
2784         }
2785
2786 #ifndef CONFIG_QUOTA
2787         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2788             !readonly) {
2789                 ext4_msg(sb, KERN_ERR,
2790                          "Filesystem with quota feature cannot be mounted RDWR "
2791                          "without CONFIG_QUOTA");
2792                 return 0;
2793         }
2794 #endif  /* CONFIG_QUOTA */
2795         return 1;
2796 }
2797
2798 /*
2799  * This function is called once a day if we have errors logged
2800  * on the file system
2801  */
2802 static void print_daily_error_info(unsigned long arg)
2803 {
2804         struct super_block *sb = (struct super_block *) arg;
2805         struct ext4_sb_info *sbi;
2806         struct ext4_super_block *es;
2807
2808         sbi = EXT4_SB(sb);
2809         es = sbi->s_es;
2810
2811         if (es->s_error_count)
2812                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2813                          le32_to_cpu(es->s_error_count));
2814         if (es->s_first_error_time) {
2815                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2816                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2817                        (int) sizeof(es->s_first_error_func),
2818                        es->s_first_error_func,
2819                        le32_to_cpu(es->s_first_error_line));
2820                 if (es->s_first_error_ino)
2821                         printk(": inode %u",
2822                                le32_to_cpu(es->s_first_error_ino));
2823                 if (es->s_first_error_block)
2824                         printk(": block %llu", (unsigned long long)
2825                                le64_to_cpu(es->s_first_error_block));
2826                 printk("\n");
2827         }
2828         if (es->s_last_error_time) {
2829                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2830                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2831                        (int) sizeof(es->s_last_error_func),
2832                        es->s_last_error_func,
2833                        le32_to_cpu(es->s_last_error_line));
2834                 if (es->s_last_error_ino)
2835                         printk(": inode %u",
2836                                le32_to_cpu(es->s_last_error_ino));
2837                 if (es->s_last_error_block)
2838                         printk(": block %llu", (unsigned long long)
2839                                le64_to_cpu(es->s_last_error_block));
2840                 printk("\n");
2841         }
2842         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2843 }
2844
2845 /* Find next suitable group and run ext4_init_inode_table */
2846 static int ext4_run_li_request(struct ext4_li_request *elr)
2847 {
2848         struct ext4_group_desc *gdp = NULL;
2849         ext4_group_t group, ngroups;
2850         struct super_block *sb;
2851         unsigned long timeout = 0;
2852         int ret = 0;
2853
2854         sb = elr->lr_super;
2855         ngroups = EXT4_SB(sb)->s_groups_count;
2856
2857         sb_start_write(sb);
2858         for (group = elr->lr_next_group; group < ngroups; group++) {
2859                 gdp = ext4_get_group_desc(sb, group, NULL);
2860                 if (!gdp) {
2861                         ret = 1;
2862                         break;
2863                 }
2864
2865                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2866                         break;
2867         }
2868
2869         if (group >= ngroups)
2870                 ret = 1;
2871
2872         if (!ret) {
2873                 timeout = jiffies;
2874                 ret = ext4_init_inode_table(sb, group,
2875                                             elr->lr_timeout ? 0 : 1);
2876                 if (elr->lr_timeout == 0) {
2877                         timeout = (jiffies - timeout) *
2878                                   elr->lr_sbi->s_li_wait_mult;
2879                         elr->lr_timeout = timeout;
2880                 }
2881                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2882                 elr->lr_next_group = group + 1;
2883         }
2884         sb_end_write(sb);
2885
2886         return ret;
2887 }
2888
2889 /*
2890  * Remove lr_request from the list_request and free the
2891  * request structure. Should be called with li_list_mtx held
2892  */
2893 static void ext4_remove_li_request(struct ext4_li_request *elr)
2894 {
2895         struct ext4_sb_info *sbi;
2896
2897         if (!elr)
2898                 return;
2899
2900         sbi = elr->lr_sbi;
2901
2902         list_del(&elr->lr_request);
2903         sbi->s_li_request = NULL;
2904         kfree(elr);
2905 }
2906
2907 static void ext4_unregister_li_request(struct super_block *sb)
2908 {
2909         mutex_lock(&ext4_li_mtx);
2910         if (!ext4_li_info) {
2911                 mutex_unlock(&ext4_li_mtx);
2912                 return;
2913         }
2914
2915         mutex_lock(&ext4_li_info->li_list_mtx);
2916         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2917         mutex_unlock(&ext4_li_info->li_list_mtx);
2918         mutex_unlock(&ext4_li_mtx);
2919 }
2920
2921 static struct task_struct *ext4_lazyinit_task;
2922
2923 /*
2924  * This is the function where ext4lazyinit thread lives. It walks
2925  * through the request list searching for next scheduled filesystem.
2926  * When such a fs is found, run the lazy initialization request
2927  * (ext4_rn_li_request) and keep track of the time spend in this
2928  * function. Based on that time we compute next schedule time of
2929  * the request. When walking through the list is complete, compute
2930  * next waking time and put itself into sleep.
2931  */
2932 static int ext4_lazyinit_thread(void *arg)
2933 {
2934         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2935         struct list_head *pos, *n;
2936         struct ext4_li_request *elr;
2937         unsigned long next_wakeup, cur;
2938
2939         BUG_ON(NULL == eli);
2940
2941 cont_thread:
2942         while (true) {
2943                 next_wakeup = MAX_JIFFY_OFFSET;
2944
2945                 mutex_lock(&eli->li_list_mtx);
2946                 if (list_empty(&eli->li_request_list)) {
2947                         mutex_unlock(&eli->li_list_mtx);
2948                         goto exit_thread;
2949                 }
2950
2951                 list_for_each_safe(pos, n, &eli->li_request_list) {
2952                         elr = list_entry(pos, struct ext4_li_request,
2953                                          lr_request);
2954
2955                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2956                                 if (ext4_run_li_request(elr) != 0) {
2957                                         /* error, remove the lazy_init job */
2958                                         ext4_remove_li_request(elr);
2959                                         continue;
2960                                 }
2961                         }
2962
2963                         if (time_before(elr->lr_next_sched, next_wakeup))
2964                                 next_wakeup = elr->lr_next_sched;
2965                 }
2966                 mutex_unlock(&eli->li_list_mtx);
2967
2968                 try_to_freeze();
2969
2970                 cur = jiffies;
2971                 if ((time_after_eq(cur, next_wakeup)) ||
2972                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2973                         cond_resched();
2974                         continue;
2975                 }
2976
2977                 schedule_timeout_interruptible(next_wakeup - cur);
2978
2979                 if (kthread_should_stop()) {
2980                         ext4_clear_request_list();
2981                         goto exit_thread;
2982                 }
2983         }
2984
2985 exit_thread:
2986         /*
2987          * It looks like the request list is empty, but we need
2988          * to check it under the li_list_mtx lock, to prevent any
2989          * additions into it, and of course we should lock ext4_li_mtx
2990          * to atomically free the list and ext4_li_info, because at
2991          * this point another ext4 filesystem could be registering
2992          * new one.
2993          */
2994         mutex_lock(&ext4_li_mtx);
2995         mutex_lock(&eli->li_list_mtx);
2996         if (!list_empty(&eli->li_request_list)) {
2997                 mutex_unlock(&eli->li_list_mtx);
2998                 mutex_unlock(&ext4_li_mtx);
2999                 goto cont_thread;
3000         }
3001         mutex_unlock(&eli->li_list_mtx);
3002         kfree(ext4_li_info);
3003         ext4_li_info = NULL;
3004         mutex_unlock(&ext4_li_mtx);
3005
3006         return 0;
3007 }
3008
3009 static void ext4_clear_request_list(void)
3010 {
3011         struct list_head *pos, *n;
3012         struct ext4_li_request *elr;
3013
3014         mutex_lock(&ext4_li_info->li_list_mtx);
3015         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3016                 elr = list_entry(pos, struct ext4_li_request,
3017                                  lr_request);
3018                 ext4_remove_li_request(elr);
3019         }
3020         mutex_unlock(&ext4_li_info->li_list_mtx);
3021 }
3022
3023 static int ext4_run_lazyinit_thread(void)
3024 {
3025         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3026                                          ext4_li_info, "ext4lazyinit");
3027         if (IS_ERR(ext4_lazyinit_task)) {
3028                 int err = PTR_ERR(ext4_lazyinit_task);
3029                 ext4_clear_request_list();
3030                 kfree(ext4_li_info);
3031                 ext4_li_info = NULL;
3032                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3033                                  "initialization thread\n",
3034                                  err);
3035                 return err;
3036         }
3037         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3038         return 0;
3039 }
3040
3041 /*
3042  * Check whether it make sense to run itable init. thread or not.
3043  * If there is at least one uninitialized inode table, return
3044  * corresponding group number, else the loop goes through all
3045  * groups and return total number of groups.
3046  */
3047 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3048 {
3049         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3050         struct ext4_group_desc *gdp = NULL;
3051
3052         for (group = 0; group < ngroups; group++) {
3053                 gdp = ext4_get_group_desc(sb, group, NULL);
3054                 if (!gdp)
3055                         continue;
3056
3057                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3058                         break;
3059         }
3060
3061         return group;
3062 }
3063
3064 static int ext4_li_info_new(void)
3065 {
3066         struct ext4_lazy_init *eli = NULL;
3067
3068         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3069         if (!eli)
3070                 return -ENOMEM;
3071
3072         INIT_LIST_HEAD(&eli->li_request_list);
3073         mutex_init(&eli->li_list_mtx);
3074
3075         eli->li_state |= EXT4_LAZYINIT_QUIT;
3076
3077         ext4_li_info = eli;
3078
3079         return 0;
3080 }
3081
3082 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3083                                             ext4_group_t start)
3084 {
3085         struct ext4_sb_info *sbi = EXT4_SB(sb);
3086         struct ext4_li_request *elr;
3087
3088         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3089         if (!elr)
3090                 return NULL;
3091
3092         elr->lr_super = sb;
3093         elr->lr_sbi = sbi;
3094         elr->lr_next_group = start;
3095
3096         /*
3097          * Randomize first schedule time of the request to
3098          * spread the inode table initialization requests
3099          * better.
3100          */
3101         elr->lr_next_sched = jiffies + (prandom_u32() %
3102                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3103         return elr;
3104 }
3105
3106 int ext4_register_li_request(struct super_block *sb,
3107                              ext4_group_t first_not_zeroed)
3108 {
3109         struct ext4_sb_info *sbi = EXT4_SB(sb);
3110         struct ext4_li_request *elr = NULL;
3111         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3112         int ret = 0;
3113
3114         mutex_lock(&ext4_li_mtx);
3115         if (sbi->s_li_request != NULL) {
3116                 /*
3117                  * Reset timeout so it can be computed again, because
3118                  * s_li_wait_mult might have changed.
3119                  */
3120                 sbi->s_li_request->lr_timeout = 0;
3121                 goto out;
3122         }
3123
3124         if (first_not_zeroed == ngroups ||
3125             (sb->s_flags & MS_RDONLY) ||
3126             !test_opt(sb, INIT_INODE_TABLE))
3127                 goto out;
3128
3129         elr = ext4_li_request_new(sb, first_not_zeroed);
3130         if (!elr) {
3131                 ret = -ENOMEM;
3132                 goto out;
3133         }
3134
3135         if (NULL == ext4_li_info) {
3136                 ret = ext4_li_info_new();
3137                 if (ret)
3138                         goto out;
3139         }
3140
3141         mutex_lock(&ext4_li_info->li_list_mtx);
3142         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3143         mutex_unlock(&ext4_li_info->li_list_mtx);
3144
3145         sbi->s_li_request = elr;
3146         /*
3147          * set elr to NULL here since it has been inserted to
3148          * the request_list and the removal and free of it is
3149          * handled by ext4_clear_request_list from now on.
3150          */
3151         elr = NULL;
3152
3153         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3154                 ret = ext4_run_lazyinit_thread();
3155                 if (ret)
3156                         goto out;
3157         }
3158 out:
3159         mutex_unlock(&ext4_li_mtx);
3160         if (ret)
3161                 kfree(elr);
3162         return ret;
3163 }
3164
3165 /*
3166  * We do not need to lock anything since this is called on
3167  * module unload.
3168  */
3169 static void ext4_destroy_lazyinit_thread(void)
3170 {
3171         /*
3172          * If thread exited earlier
3173          * there's nothing to be done.
3174          */
3175         if (!ext4_li_info || !ext4_lazyinit_task)
3176                 return;
3177
3178         kthread_stop(ext4_lazyinit_task);
3179 }
3180
3181 static int set_journal_csum_feature_set(struct super_block *sb)
3182 {
3183         int ret = 1;
3184         int compat, incompat;
3185         struct ext4_sb_info *sbi = EXT4_SB(sb);
3186
3187         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3188                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3189                 /* journal checksum v2 */
3190                 compat = 0;
3191                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3192         } else {
3193                 /* journal checksum v1 */
3194                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3195                 incompat = 0;
3196         }
3197
3198         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3199                 ret = jbd2_journal_set_features(sbi->s_journal,
3200                                 compat, 0,
3201                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3202                                 incompat);
3203         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3204                 ret = jbd2_journal_set_features(sbi->s_journal,
3205                                 compat, 0,
3206                                 incompat);
3207                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3208                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3209         } else {
3210                 jbd2_journal_clear_features(sbi->s_journal,
3211                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3212                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3213                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3214         }
3215
3216         return ret;
3217 }
3218
3219 /*
3220  * Note: calculating the overhead so we can be compatible with
3221  * historical BSD practice is quite difficult in the face of
3222  * clusters/bigalloc.  This is because multiple metadata blocks from
3223  * different block group can end up in the same allocation cluster.
3224  * Calculating the exact overhead in the face of clustered allocation
3225  * requires either O(all block bitmaps) in memory or O(number of block
3226  * groups**2) in time.  We will still calculate the superblock for
3227  * older file systems --- and if we come across with a bigalloc file
3228  * system with zero in s_overhead_clusters the estimate will be close to
3229  * correct especially for very large cluster sizes --- but for newer
3230  * file systems, it's better to calculate this figure once at mkfs
3231  * time, and store it in the superblock.  If the superblock value is
3232  * present (even for non-bigalloc file systems), we will use it.
3233  */
3234 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3235                           char *buf)
3236 {
3237         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3238         struct ext4_group_desc  *gdp;
3239         ext4_fsblk_t            first_block, last_block, b;
3240         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3241         int                     s, j, count = 0;
3242
3243         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3244                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3245                         sbi->s_itb_per_group + 2);
3246
3247         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3248                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3249         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3250         for (i = 0; i < ngroups; i++) {
3251                 gdp = ext4_get_group_desc(sb, i, NULL);
3252                 b = ext4_block_bitmap(sb, gdp);
3253                 if (b >= first_block && b <= last_block) {
3254                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3255                         count++;
3256                 }
3257                 b = ext4_inode_bitmap(sb, gdp);
3258                 if (b >= first_block && b <= last_block) {
3259                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3260                         count++;
3261                 }
3262                 b = ext4_inode_table(sb, gdp);
3263                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3264                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3265                                 int c = EXT4_B2C(sbi, b - first_block);
3266                                 ext4_set_bit(c, buf);
3267                                 count++;
3268                         }
3269                 if (i != grp)
3270                         continue;
3271                 s = 0;
3272                 if (ext4_bg_has_super(sb, grp)) {
3273                         ext4_set_bit(s++, buf);
3274                         count++;
3275                 }
3276                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3277                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3278                         count++;
3279                 }
3280         }
3281         if (!count)
3282                 return 0;
3283         return EXT4_CLUSTERS_PER_GROUP(sb) -
3284                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3285 }
3286
3287 /*
3288  * Compute the overhead and stash it in sbi->s_overhead
3289  */
3290 int ext4_calculate_overhead(struct super_block *sb)
3291 {
3292         struct ext4_sb_info *sbi = EXT4_SB(sb);
3293         struct ext4_super_block *es = sbi->s_es;
3294         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3295         ext4_fsblk_t overhead = 0;
3296         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3297
3298         if (!buf)
3299                 return -ENOMEM;
3300
3301         /*
3302          * Compute the overhead (FS structures).  This is constant
3303          * for a given filesystem unless the number of block groups
3304          * changes so we cache the previous value until it does.
3305          */
3306
3307         /*
3308          * All of the blocks before first_data_block are overhead
3309          */
3310         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3311
3312         /*
3313          * Add the overhead found in each block group
3314          */
3315         for (i = 0; i < ngroups; i++) {
3316                 int blks;
3317
3318                 blks = count_overhead(sb, i, buf);
3319                 overhead += blks;
3320                 if (blks)
3321                         memset(buf, 0, PAGE_SIZE);
3322                 cond_resched();
3323         }
3324         /* Add the journal blocks as well */
3325         if (sbi->s_journal)
3326                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3327
3328         sbi->s_overhead = overhead;
3329         smp_wmb();
3330         free_page((unsigned long) buf);
3331         return 0;
3332 }
3333
3334
3335 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3336 {
3337         ext4_fsblk_t resv_clusters;
3338
3339         /*
3340          * There's no need to reserve anything when we aren't using extents.
3341          * The space estimates are exact, there are no unwritten extents,
3342          * hole punching doesn't need new metadata... This is needed especially
3343          * to keep ext2/3 backward compatibility.
3344          */
3345         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3346                 return 0;
3347         /*
3348          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3349          * This should cover the situations where we can not afford to run
3350          * out of space like for example punch hole, or converting
3351          * unwritten extents in delalloc path. In most cases such
3352          * allocation would require 1, or 2 blocks, higher numbers are
3353          * very rare.
3354          */
3355         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3356                         EXT4_SB(sb)->s_cluster_bits;
3357
3358         do_div(resv_clusters, 50);
3359         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3360
3361         return resv_clusters;
3362 }
3363
3364
3365 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3366 {
3367         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3368                                 sbi->s_cluster_bits;
3369
3370         if (count >= clusters)
3371                 return -EINVAL;
3372
3373         atomic64_set(&sbi->s_resv_clusters, count);
3374         return 0;
3375 }
3376
3377 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3378 {
3379         char *orig_data = kstrdup(data, GFP_KERNEL);
3380         struct buffer_head *bh;
3381         struct ext4_super_block *es = NULL;
3382         struct ext4_sb_info *sbi;
3383         ext4_fsblk_t block;
3384         ext4_fsblk_t sb_block = get_sb_block(&data);
3385         ext4_fsblk_t logical_sb_block;
3386         unsigned long offset = 0;
3387         unsigned long journal_devnum = 0;
3388         unsigned long def_mount_opts;
3389         struct inode *root;
3390         char *cp;
3391         const char *descr;
3392         int ret = -ENOMEM;
3393         int blocksize, clustersize;
3394         unsigned int db_count;
3395         unsigned int i;
3396         int needs_recovery, has_huge_files, has_bigalloc;
3397         __u64 blocks_count;
3398         int err = 0;
3399         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3400         ext4_group_t first_not_zeroed;
3401
3402         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3403         if (!sbi)
3404                 goto out_free_orig;
3405
3406         sbi->s_blockgroup_lock =
3407                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3408         if (!sbi->s_blockgroup_lock) {
3409                 kfree(sbi);
3410                 goto out_free_orig;
3411         }
3412         sb->s_fs_info = sbi;
3413         sbi->s_sb = sb;
3414         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3415         sbi->s_sb_block = sb_block;
3416         if (sb->s_bdev->bd_part)
3417                 sbi->s_sectors_written_start =
3418                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3419
3420         /* Cleanup superblock name */
3421         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3422                 *cp = '!';
3423
3424         /* -EINVAL is default */
3425         ret = -EINVAL;
3426         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3427         if (!blocksize) {
3428                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3429                 goto out_fail;
3430         }
3431
3432         /*
3433          * The ext4 superblock will not be buffer aligned for other than 1kB
3434          * block sizes.  We need to calculate the offset from buffer start.
3435          */
3436         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3437                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3438                 offset = do_div(logical_sb_block, blocksize);
3439         } else {
3440                 logical_sb_block = sb_block;
3441         }
3442
3443         if (!(bh = sb_bread(sb, logical_sb_block))) {
3444                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3445                 goto out_fail;
3446         }
3447         /*
3448          * Note: s_es must be initialized as soon as possible because
3449          *       some ext4 macro-instructions depend on its value
3450          */
3451         es = (struct ext4_super_block *) (bh->b_data + offset);
3452         sbi->s_es = es;
3453         sb->s_magic = le16_to_cpu(es->s_magic);
3454         if (sb->s_magic != EXT4_SUPER_MAGIC)
3455                 goto cantfind_ext4;
3456         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3457
3458         /* Warn if metadata_csum and gdt_csum are both set. */
3459         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3460                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3461             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3462                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3463                              "redundant flags; please run fsck.");
3464
3465         /* Check for a known checksum algorithm */
3466         if (!ext4_verify_csum_type(sb, es)) {
3467                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3468                          "unknown checksum algorithm.");
3469                 silent = 1;
3470                 goto cantfind_ext4;
3471         }
3472
3473         /* Load the checksum driver */
3474         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3475                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3476                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3477                 if (IS_ERR(sbi->s_chksum_driver)) {
3478                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3479                         ret = PTR_ERR(sbi->s_chksum_driver);
3480                         sbi->s_chksum_driver = NULL;
3481                         goto failed_mount;
3482                 }
3483         }
3484
3485         /* Check superblock checksum */
3486         if (!ext4_superblock_csum_verify(sb, es)) {
3487                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3488                          "invalid superblock checksum.  Run e2fsck?");
3489                 silent = 1;
3490                 goto cantfind_ext4;
3491         }
3492
3493         /* Precompute checksum seed for all metadata */
3494         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3495                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3496                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3497                                                sizeof(es->s_uuid));
3498
3499         /* Set defaults before we parse the mount options */
3500         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3501         set_opt(sb, INIT_INODE_TABLE);
3502         if (def_mount_opts & EXT4_DEFM_DEBUG)
3503                 set_opt(sb, DEBUG);
3504         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3505                 set_opt(sb, GRPID);
3506         if (def_mount_opts & EXT4_DEFM_UID16)
3507                 set_opt(sb, NO_UID32);
3508         /* xattr user namespace & acls are now defaulted on */
3509         set_opt(sb, XATTR_USER);
3510 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3511         set_opt(sb, POSIX_ACL);
3512 #endif
3513         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3514                 set_opt(sb, JOURNAL_DATA);
3515         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3516                 set_opt(sb, ORDERED_DATA);
3517         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3518                 set_opt(sb, WRITEBACK_DATA);
3519
3520         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3521                 set_opt(sb, ERRORS_PANIC);
3522         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3523                 set_opt(sb, ERRORS_CONT);
3524         else
3525                 set_opt(sb, ERRORS_RO);
3526         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3527                 set_opt(sb, BLOCK_VALIDITY);
3528         if (def_mount_opts & EXT4_DEFM_DISCARD)
3529                 set_opt(sb, DISCARD);
3530
3531         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3532         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3533         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3534         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3535         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3536
3537         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3538                 set_opt(sb, BARRIER);
3539
3540         /*
3541          * enable delayed allocation by default
3542          * Use -o nodelalloc to turn it off
3543          */
3544         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3545             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3546                 set_opt(sb, DELALLOC);
3547
3548         /*
3549          * set default s_li_wait_mult for lazyinit, for the case there is
3550          * no mount option specified.
3551          */
3552         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3553
3554         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3555                            &journal_devnum, &journal_ioprio, 0)) {
3556                 ext4_msg(sb, KERN_WARNING,
3557                          "failed to parse options in superblock: %s",
3558                          sbi->s_es->s_mount_opts);
3559         }
3560         sbi->s_def_mount_opt = sbi->s_mount_opt;
3561         if (!parse_options((char *) data, sb, &journal_devnum,
3562                            &journal_ioprio, 0))
3563                 goto failed_mount;
3564
3565         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3566                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3567                             "with data=journal disables delayed "
3568                             "allocation and O_DIRECT support!\n");
3569                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3570                         ext4_msg(sb, KERN_ERR, "can't mount with "
3571                                  "both data=journal and delalloc");
3572                         goto failed_mount;
3573                 }
3574                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3575                         ext4_msg(sb, KERN_ERR, "can't mount with "
3576                                  "both data=journal and dioread_nolock");
3577                         goto failed_mount;
3578                 }
3579                 if (test_opt(sb, DELALLOC))
3580                         clear_opt(sb, DELALLOC);
3581         }
3582
3583         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3584                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3585
3586         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3587             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3588              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3589              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3590                 ext4_msg(sb, KERN_WARNING,
3591                        "feature flags set on rev 0 fs, "
3592                        "running e2fsck is recommended");
3593
3594         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3595                 set_opt2(sb, HURD_COMPAT);
3596                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3597                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3598                         ext4_msg(sb, KERN_ERR,
3599                                  "The Hurd can't support 64-bit file systems");
3600                         goto failed_mount;
3601                 }
3602         }
3603
3604         if (IS_EXT2_SB(sb)) {
3605                 if (ext2_feature_set_ok(sb))
3606                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3607                                  "using the ext4 subsystem");
3608                 else {
3609                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3610                                  "to feature incompatibilities");
3611                         goto failed_mount;
3612                 }
3613         }
3614
3615         if (IS_EXT3_SB(sb)) {
3616                 if (ext3_feature_set_ok(sb))
3617                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3618                                  "using the ext4 subsystem");
3619                 else {
3620                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3621                                  "to feature incompatibilities");
3622                         goto failed_mount;
3623                 }
3624         }
3625
3626         /*
3627          * Check feature flags regardless of the revision level, since we
3628          * previously didn't change the revision level when setting the flags,
3629          * so there is a chance incompat flags are set on a rev 0 filesystem.
3630          */
3631         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3632                 goto failed_mount;
3633
3634         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3635         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3636             blocksize > EXT4_MAX_BLOCK_SIZE) {
3637                 ext4_msg(sb, KERN_ERR,
3638                        "Unsupported filesystem blocksize %d", blocksize);
3639                 goto failed_mount;
3640         }
3641
3642         if (sb->s_blocksize != blocksize) {
3643                 /* Validate the filesystem blocksize */
3644                 if (!sb_set_blocksize(sb, blocksize)) {
3645                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3646                                         blocksize);
3647                         goto failed_mount;
3648                 }
3649
3650                 brelse(bh);
3651                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3652                 offset = do_div(logical_sb_block, blocksize);
3653                 bh = sb_bread(sb, logical_sb_block);
3654                 if (!bh) {
3655                         ext4_msg(sb, KERN_ERR,
3656                                "Can't read superblock on 2nd try");
3657                         goto failed_mount;
3658                 }
3659                 es = (struct ext4_super_block *)(bh->b_data + offset);
3660                 sbi->s_es = es;
3661                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3662                         ext4_msg(sb, KERN_ERR,
3663                                "Magic mismatch, very weird!");
3664                         goto failed_mount;
3665                 }
3666         }
3667
3668         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3669                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3670         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3671                                                       has_huge_files);
3672         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3673
3674         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3675                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3676                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3677         } else {
3678                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3679                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3680                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3681                     (!is_power_of_2(sbi->s_inode_size)) ||
3682                     (sbi->s_inode_size > blocksize)) {
3683                         ext4_msg(sb, KERN_ERR,
3684                                "unsupported inode size: %d",
3685                                sbi->s_inode_size);
3686                         goto failed_mount;
3687                 }
3688                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3689                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3690         }
3691
3692         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3693         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3694                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3695                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3696                     !is_power_of_2(sbi->s_desc_size)) {
3697                         ext4_msg(sb, KERN_ERR,
3698                                "unsupported descriptor size %lu",
3699                                sbi->s_desc_size);
3700                         goto failed_mount;
3701                 }
3702         } else
3703                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3704
3705         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3706         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3707         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3708                 goto cantfind_ext4;
3709
3710         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3711         if (sbi->s_inodes_per_block == 0)
3712                 goto cantfind_ext4;
3713         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3714                                         sbi->s_inodes_per_block;
3715         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3716         sbi->s_sbh = bh;
3717         sbi->s_mount_state = le16_to_cpu(es->s_state);
3718         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3719         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3720
3721         for (i = 0; i < 4; i++)
3722                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3723         sbi->s_def_hash_version = es->s_def_hash_version;
3724         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3725                 i = le32_to_cpu(es->s_flags);
3726                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3727                         sbi->s_hash_unsigned = 3;
3728                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3729 #ifdef __CHAR_UNSIGNED__
3730                         if (!(sb->s_flags & MS_RDONLY))
3731                                 es->s_flags |=
3732                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3733                         sbi->s_hash_unsigned = 3;
3734 #else
3735                         if (!(sb->s_flags & MS_RDONLY))
3736                                 es->s_flags |=
3737                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3738 #endif
3739                 }
3740         }
3741
3742         /* Handle clustersize */
3743         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3744         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3745                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3746         if (has_bigalloc) {
3747                 if (clustersize < blocksize) {
3748                         ext4_msg(sb, KERN_ERR,
3749                                  "cluster size (%d) smaller than "
3750                                  "block size (%d)", clustersize, blocksize);
3751                         goto failed_mount;
3752                 }
3753                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3754                         le32_to_cpu(es->s_log_block_size);
3755                 sbi->s_clusters_per_group =
3756                         le32_to_cpu(es->s_clusters_per_group);
3757                 if (sbi->s_clusters_per_group > blocksize * 8) {
3758                         ext4_msg(sb, KERN_ERR,
3759                                  "#clusters per group too big: %lu",
3760                                  sbi->s_clusters_per_group);
3761                         goto failed_mount;
3762                 }
3763                 if (sbi->s_blocks_per_group !=
3764                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3765                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3766                                  "clusters per group (%lu) inconsistent",
3767                                  sbi->s_blocks_per_group,
3768                                  sbi->s_clusters_per_group);
3769                         goto failed_mount;
3770                 }
3771         } else {
3772                 if (clustersize != blocksize) {
3773                         ext4_warning(sb, "fragment/cluster size (%d) != "
3774                                      "block size (%d)", clustersize,
3775                                      blocksize);
3776                         clustersize = blocksize;
3777                 }
3778                 if (sbi->s_blocks_per_group > blocksize * 8) {
3779                         ext4_msg(sb, KERN_ERR,
3780                                  "#blocks per group too big: %lu",
3781                                  sbi->s_blocks_per_group);
3782                         goto failed_mount;
3783                 }
3784                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3785                 sbi->s_cluster_bits = 0;
3786         }
3787         sbi->s_cluster_ratio = clustersize / blocksize;
3788
3789         if (sbi->s_inodes_per_group > blocksize * 8) {
3790                 ext4_msg(sb, KERN_ERR,
3791                        "#inodes per group too big: %lu",
3792                        sbi->s_inodes_per_group);
3793                 goto failed_mount;
3794         }
3795
3796         /* Do we have standard group size of clustersize * 8 blocks ? */
3797         if (sbi->s_blocks_per_group == clustersize << 3)
3798                 set_opt2(sb, STD_GROUP_SIZE);
3799
3800         /*
3801          * Test whether we have more sectors than will fit in sector_t,
3802          * and whether the max offset is addressable by the page cache.
3803          */
3804         err = generic_check_addressable(sb->s_blocksize_bits,
3805                                         ext4_blocks_count(es));
3806         if (err) {
3807                 ext4_msg(sb, KERN_ERR, "filesystem"
3808                          " too large to mount safely on this system");
3809                 if (sizeof(sector_t) < 8)
3810                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3811                 goto failed_mount;
3812         }
3813
3814         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3815                 goto cantfind_ext4;
3816
3817         /* check blocks count against device size */
3818         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3819         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3820                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3821                        "exceeds size of device (%llu blocks)",
3822                        ext4_blocks_count(es), blocks_count);
3823                 goto failed_mount;
3824         }
3825
3826         /*
3827          * It makes no sense for the first data block to be beyond the end
3828          * of the filesystem.
3829          */
3830         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3831                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3832                          "block %u is beyond end of filesystem (%llu)",
3833                          le32_to_cpu(es->s_first_data_block),
3834                          ext4_blocks_count(es));
3835                 goto failed_mount;
3836         }
3837         blocks_count = (ext4_blocks_count(es) -
3838                         le32_to_cpu(es->s_first_data_block) +
3839                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3840         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3841         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3842                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3843                        "(block count %llu, first data block %u, "
3844                        "blocks per group %lu)", sbi->s_groups_count,
3845                        ext4_blocks_count(es),
3846                        le32_to_cpu(es->s_first_data_block),
3847                        EXT4_BLOCKS_PER_GROUP(sb));
3848                 goto failed_mount;
3849         }
3850         sbi->s_groups_count = blocks_count;
3851         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3852                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3853         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3854                    EXT4_DESC_PER_BLOCK(sb);
3855         sbi->s_group_desc = ext4_kvmalloc(db_count *
3856                                           sizeof(struct buffer_head *),
3857                                           GFP_KERNEL);
3858         if (sbi->s_group_desc == NULL) {
3859                 ext4_msg(sb, KERN_ERR, "not enough memory");
3860                 ret = -ENOMEM;
3861                 goto failed_mount;
3862         }
3863
3864         if (ext4_proc_root)
3865                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3866
3867         if (sbi->s_proc)
3868                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3869                                  &ext4_seq_options_fops, sb);
3870
3871         bgl_lock_init(sbi->s_blockgroup_lock);
3872
3873         for (i = 0; i < db_count; i++) {
3874                 block = descriptor_loc(sb, logical_sb_block, i);
3875                 sbi->s_group_desc[i] = sb_bread(sb, block);
3876                 if (!sbi->s_group_desc[i]) {
3877                         ext4_msg(sb, KERN_ERR,
3878                                "can't read group descriptor %d", i);
3879                         db_count = i;
3880                         goto failed_mount2;
3881                 }
3882         }
3883
3884         /*
3885          * set up enough so that it can read an inode,
3886          * and create new inode for buddy allocator
3887          */
3888         sbi->s_gdb_count = db_count;
3889         if (!test_opt(sb, NOLOAD) &&
3890             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3891                 sb->s_op = &ext4_sops;
3892         else
3893                 sb->s_op = &ext4_nojournal_sops;
3894
3895         ext4_ext_init(sb);
3896         err = ext4_mb_init(sb);
3897         if (err) {
3898                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3899                          err);
3900                 goto failed_mount2;
3901         }
3902
3903         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3904                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3905                 goto failed_mount2a;
3906         }
3907         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3908                 if (!ext4_fill_flex_info(sb)) {
3909                         ext4_msg(sb, KERN_ERR,
3910                                "unable to initialize "
3911                                "flex_bg meta info!");
3912                         goto failed_mount2a;
3913                 }
3914
3915         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3916         spin_lock_init(&sbi->s_next_gen_lock);
3917
3918         init_timer(&sbi->s_err_report);
3919         sbi->s_err_report.function = print_daily_error_info;
3920         sbi->s_err_report.data = (unsigned long) sb;
3921
3922         /* Register extent status tree shrinker */
3923         ext4_es_register_shrinker(sbi);
3924
3925         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3926                         ext4_count_free_clusters(sb));
3927         if (!err) {
3928                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3929                                 ext4_count_free_inodes(sb));
3930         }
3931         if (!err) {
3932                 err = percpu_counter_init(&sbi->s_dirs_counter,
3933                                 ext4_count_dirs(sb));
3934         }
3935         if (!err) {
3936                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3937         }
3938         if (!err) {
3939                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3940         }
3941         if (err) {
3942                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3943                 goto failed_mount3;
3944         }
3945
3946         sbi->s_stripe = ext4_get_stripe_size(sbi);
3947         sbi->s_extent_max_zeroout_kb = 32;
3948
3949         sb->s_export_op = &ext4_export_ops;
3950         sb->s_xattr = ext4_xattr_handlers;
3951 #ifdef CONFIG_QUOTA
3952         sb->dq_op = &ext4_quota_operations;
3953         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3954                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3955         else
3956                 sb->s_qcop = &ext4_qctl_operations;
3957 #endif
3958         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3959
3960         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3961         mutex_init(&sbi->s_orphan_lock);
3962
3963         sb->s_root = NULL;
3964
3965         needs_recovery = (es->s_last_orphan != 0 ||
3966                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3967                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3968
3969         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3970             !(sb->s_flags & MS_RDONLY))
3971                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3972                         goto failed_mount3;
3973
3974         /*
3975          * The first inode we look at is the journal inode.  Don't try
3976          * root first: it may be modified in the journal!
3977          */
3978         if (!test_opt(sb, NOLOAD) &&
3979             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3980                 if (ext4_load_journal(sb, es, journal_devnum))
3981                         goto failed_mount3;
3982         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3983               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3984                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3985                        "suppressed and not mounted read-only");
3986                 goto failed_mount_wq;
3987         } else {
3988                 clear_opt(sb, DATA_FLAGS);
3989                 sbi->s_journal = NULL;
3990                 needs_recovery = 0;
3991                 goto no_journal;
3992         }
3993
3994         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3995             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3996                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3997                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3998                 goto failed_mount_wq;
3999         }
4000
4001         if (!set_journal_csum_feature_set(sb)) {
4002                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4003                          "feature set");
4004                 goto failed_mount_wq;
4005         }
4006
4007         /* We have now updated the journal if required, so we can
4008          * validate the data journaling mode. */
4009         switch (test_opt(sb, DATA_FLAGS)) {
4010         case 0:
4011                 /* No mode set, assume a default based on the journal
4012                  * capabilities: ORDERED_DATA if the journal can
4013                  * cope, else JOURNAL_DATA
4014                  */
4015                 if (jbd2_journal_check_available_features
4016                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4017                         set_opt(sb, ORDERED_DATA);
4018                 else
4019                         set_opt(sb, JOURNAL_DATA);
4020                 break;
4021
4022         case EXT4_MOUNT_ORDERED_DATA:
4023         case EXT4_MOUNT_WRITEBACK_DATA:
4024                 if (!jbd2_journal_check_available_features
4025                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4026                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4027                                "requested data journaling mode");
4028                         goto failed_mount_wq;
4029                 }
4030         default:
4031                 break;
4032         }
4033         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4034
4035         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4036
4037         /*
4038          * The journal may have updated the bg summary counts, so we
4039          * need to update the global counters.
4040          */
4041         percpu_counter_set(&sbi->s_freeclusters_counter,
4042                            ext4_count_free_clusters(sb));
4043         percpu_counter_set(&sbi->s_freeinodes_counter,
4044                            ext4_count_free_inodes(sb));
4045         percpu_counter_set(&sbi->s_dirs_counter,
4046                            ext4_count_dirs(sb));
4047         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
4048
4049 no_journal:
4050         if (ext4_mballoc_ready) {
4051                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4052                 if (!sbi->s_mb_cache) {
4053                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4054                         goto failed_mount_wq;
4055                 }
4056         }
4057
4058         /*
4059          * Get the # of file system overhead blocks from the
4060          * superblock if present.
4061          */
4062         if (es->s_overhead_clusters)
4063                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4064         else {
4065                 err = ext4_calculate_overhead(sb);
4066                 if (err)
4067                         goto failed_mount_wq;
4068         }
4069
4070         /*
4071          * The maximum number of concurrent works can be high and
4072          * concurrency isn't really necessary.  Limit it to 1.
4073          */
4074         EXT4_SB(sb)->rsv_conversion_wq =
4075                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4076         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4077                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4078                 ret = -ENOMEM;
4079                 goto failed_mount4;
4080         }
4081
4082         /*
4083          * The jbd2_journal_load will have done any necessary log recovery,
4084          * so we can safely mount the rest of the filesystem now.
4085          */
4086
4087         root = ext4_iget(sb, EXT4_ROOT_INO);
4088         if (IS_ERR(root)) {
4089                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4090                 ret = PTR_ERR(root);
4091                 root = NULL;
4092                 goto failed_mount4;
4093         }
4094         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4095                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4096                 iput(root);
4097                 goto failed_mount4;
4098         }
4099         sb->s_root = d_make_root(root);
4100         if (!sb->s_root) {
4101                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4102                 ret = -ENOMEM;
4103                 goto failed_mount4;
4104         }
4105
4106         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4107                 sb->s_flags |= MS_RDONLY;
4108
4109         /* determine the minimum size of new large inodes, if present */
4110         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4111                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4112                                                      EXT4_GOOD_OLD_INODE_SIZE;
4113                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4114                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4115                         if (sbi->s_want_extra_isize <
4116                             le16_to_cpu(es->s_want_extra_isize))
4117                                 sbi->s_want_extra_isize =
4118                                         le16_to_cpu(es->s_want_extra_isize);
4119                         if (sbi->s_want_extra_isize <
4120                             le16_to_cpu(es->s_min_extra_isize))
4121                                 sbi->s_want_extra_isize =
4122                                         le16_to_cpu(es->s_min_extra_isize);
4123                 }
4124         }
4125         /* Check if enough inode space is available */
4126         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4127                                                         sbi->s_inode_size) {
4128                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4129                                                        EXT4_GOOD_OLD_INODE_SIZE;
4130                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4131                          "available");
4132         }
4133
4134         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4135         if (err) {
4136                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4137                          "reserved pool", ext4_calculate_resv_clusters(sb));
4138                 goto failed_mount5;
4139         }
4140
4141         err = ext4_setup_system_zone(sb);
4142         if (err) {
4143                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4144                          "zone (%d)", err);
4145                 goto failed_mount5;
4146         }
4147
4148         err = ext4_register_li_request(sb, first_not_zeroed);
4149         if (err)
4150                 goto failed_mount6;
4151
4152         sbi->s_kobj.kset = ext4_kset;
4153         init_completion(&sbi->s_kobj_unregister);
4154         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4155                                    "%s", sb->s_id);
4156         if (err)
4157                 goto failed_mount7;
4158
4159 #ifdef CONFIG_QUOTA
4160         /* Enable quota usage during mount. */
4161         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4162             !(sb->s_flags & MS_RDONLY)) {
4163                 err = ext4_enable_quotas(sb);
4164                 if (err)
4165                         goto failed_mount8;
4166         }
4167 #endif  /* CONFIG_QUOTA */
4168
4169         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4170         ext4_orphan_cleanup(sb, es);
4171         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4172         if (needs_recovery) {
4173                 ext4_msg(sb, KERN_INFO, "recovery complete");
4174                 ext4_mark_recovery_complete(sb, es);
4175         }
4176         if (EXT4_SB(sb)->s_journal) {
4177                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4178                         descr = " journalled data mode";
4179                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4180                         descr = " ordered data mode";
4181                 else
4182                         descr = " writeback data mode";
4183         } else
4184                 descr = "out journal";
4185
4186         if (test_opt(sb, DISCARD)) {
4187                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4188                 if (!blk_queue_discard(q))
4189                         ext4_msg(sb, KERN_WARNING,
4190                                  "mounting with \"discard\" option, but "
4191                                  "the device does not support discard");
4192         }
4193
4194         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4195                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4196                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4197
4198         if (es->s_error_count)
4199                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4200
4201         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4202         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4203         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4204         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4205
4206         kfree(orig_data);
4207         return 0;
4208
4209 cantfind_ext4:
4210         if (!silent)
4211                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4212         goto failed_mount;
4213
4214 #ifdef CONFIG_QUOTA
4215 failed_mount8:
4216         kobject_del(&sbi->s_kobj);
4217 #endif
4218 failed_mount7:
4219         ext4_unregister_li_request(sb);
4220 failed_mount6:
4221         ext4_release_system_zone(sb);
4222 failed_mount5:
4223         dput(sb->s_root);
4224         sb->s_root = NULL;
4225 failed_mount4:
4226         ext4_msg(sb, KERN_ERR, "mount failed");
4227         if (EXT4_SB(sb)->rsv_conversion_wq)
4228                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4229 failed_mount_wq:
4230         if (sbi->s_journal) {
4231                 jbd2_journal_destroy(sbi->s_journal);
4232                 sbi->s_journal = NULL;
4233         }
4234 failed_mount3:
4235         ext4_es_unregister_shrinker(sbi);
4236         del_timer_sync(&sbi->s_err_report);
4237         if (sbi->s_flex_groups)
4238                 ext4_kvfree(sbi->s_flex_groups);
4239         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4240         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4241         percpu_counter_destroy(&sbi->s_dirs_counter);
4242         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4243         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4244         if (sbi->s_mmp_tsk)
4245                 kthread_stop(sbi->s_mmp_tsk);
4246 failed_mount2a:
4247         ext4_mb_release(sb);
4248 failed_mount2:
4249         for (i = 0; i < db_count; i++)
4250                 brelse(sbi->s_group_desc[i]);
4251         ext4_kvfree(sbi->s_group_desc);
4252 failed_mount:
4253         ext4_ext_release(sb);
4254         if (sbi->s_chksum_driver)
4255                 crypto_free_shash(sbi->s_chksum_driver);
4256         if (sbi->s_proc) {
4257                 remove_proc_entry("options", sbi->s_proc);
4258                 remove_proc_entry(sb->s_id, ext4_proc_root);
4259         }
4260 #ifdef CONFIG_QUOTA
4261         for (i = 0; i < MAXQUOTAS; i++)
4262                 kfree(sbi->s_qf_names[i]);
4263 #endif
4264         ext4_blkdev_remove(sbi);
4265         brelse(bh);
4266 out_fail:
4267         sb->s_fs_info = NULL;
4268         kfree(sbi->s_blockgroup_lock);
4269         kfree(sbi);
4270 out_free_orig:
4271         kfree(orig_data);
4272         return err ? err : ret;
4273 }
4274
4275 /*
4276  * Setup any per-fs journal parameters now.  We'll do this both on
4277  * initial mount, once the journal has been initialised but before we've
4278  * done any recovery; and again on any subsequent remount.
4279  */
4280 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4281 {
4282         struct ext4_sb_info *sbi = EXT4_SB(sb);
4283
4284         journal->j_commit_interval = sbi->s_commit_interval;
4285         journal->j_min_batch_time = sbi->s_min_batch_time;
4286         journal->j_max_batch_time = sbi->s_max_batch_time;
4287
4288         write_lock(&journal->j_state_lock);
4289         if (test_opt(sb, BARRIER))
4290                 journal->j_flags |= JBD2_BARRIER;
4291         else
4292                 journal->j_flags &= ~JBD2_BARRIER;
4293         if (test_opt(sb, DATA_ERR_ABORT))
4294                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4295         else
4296                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4297         write_unlock(&journal->j_state_lock);
4298 }
4299
4300 static journal_t *ext4_get_journal(struct super_block *sb,
4301                                    unsigned int journal_inum)
4302 {
4303         struct inode *journal_inode;
4304         journal_t *journal;
4305
4306         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4307
4308         /* First, test for the existence of a valid inode on disk.  Bad
4309          * things happen if we iget() an unused inode, as the subsequent
4310          * iput() will try to delete it. */
4311
4312         journal_inode = ext4_iget(sb, journal_inum);
4313         if (IS_ERR(journal_inode)) {
4314                 ext4_msg(sb, KERN_ERR, "no journal found");
4315                 return NULL;
4316         }
4317         if (!journal_inode->i_nlink) {
4318                 make_bad_inode(journal_inode);
4319                 iput(journal_inode);
4320                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4321                 return NULL;
4322         }
4323
4324         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4325                   journal_inode, journal_inode->i_size);
4326         if (!S_ISREG(journal_inode->i_mode)) {
4327                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4328                 iput(journal_inode);
4329                 return NULL;
4330         }
4331
4332         journal = jbd2_journal_init_inode(journal_inode);
4333         if (!journal) {
4334                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4335                 iput(journal_inode);
4336                 return NULL;
4337         }
4338         journal->j_private = sb;
4339         ext4_init_journal_params(sb, journal);
4340         return journal;
4341 }
4342
4343 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4344                                        dev_t j_dev)
4345 {
4346         struct buffer_head *bh;
4347         journal_t *journal;
4348         ext4_fsblk_t start;
4349         ext4_fsblk_t len;
4350         int hblock, blocksize;
4351         ext4_fsblk_t sb_block;
4352         unsigned long offset;
4353         struct ext4_super_block *es;
4354         struct block_device *bdev;
4355
4356         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4357
4358         bdev = ext4_blkdev_get(j_dev, sb);
4359         if (bdev == NULL)
4360                 return NULL;
4361
4362         blocksize = sb->s_blocksize;
4363         hblock = bdev_logical_block_size(bdev);
4364         if (blocksize < hblock) {
4365                 ext4_msg(sb, KERN_ERR,
4366                         "blocksize too small for journal device");
4367                 goto out_bdev;
4368         }
4369
4370         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4371         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4372         set_blocksize(bdev, blocksize);
4373         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4374                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4375                        "external journal");
4376                 goto out_bdev;
4377         }
4378
4379         es = (struct ext4_super_block *) (bh->b_data + offset);
4380         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4381             !(le32_to_cpu(es->s_feature_incompat) &
4382               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4383                 ext4_msg(sb, KERN_ERR, "external journal has "
4384                                         "bad superblock");
4385                 brelse(bh);
4386                 goto out_bdev;
4387         }
4388
4389         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4390                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4391                 brelse(bh);
4392                 goto out_bdev;
4393         }
4394
4395         len = ext4_blocks_count(es);
4396         start = sb_block + 1;
4397         brelse(bh);     /* we're done with the superblock */
4398
4399         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4400                                         start, len, blocksize);
4401         if (!journal) {
4402                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4403                 goto out_bdev;
4404         }
4405         journal->j_private = sb;
4406         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4407         wait_on_buffer(journal->j_sb_buffer);
4408         if (!buffer_uptodate(journal->j_sb_buffer)) {
4409                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4410                 goto out_journal;
4411         }
4412         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4413                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4414                                         "user (unsupported) - %d",
4415                         be32_to_cpu(journal->j_superblock->s_nr_users));
4416                 goto out_journal;
4417         }
4418         EXT4_SB(sb)->journal_bdev = bdev;
4419         ext4_init_journal_params(sb, journal);
4420         return journal;
4421
4422 out_journal:
4423         jbd2_journal_destroy(journal);
4424 out_bdev:
4425         ext4_blkdev_put(bdev);
4426         return NULL;
4427 }
4428
4429 static int ext4_load_journal(struct super_block *sb,
4430                              struct ext4_super_block *es,
4431                              unsigned long journal_devnum)
4432 {
4433         journal_t *journal;
4434         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4435         dev_t journal_dev;
4436         int err = 0;
4437         int really_read_only;
4438
4439         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4440
4441         if (journal_devnum &&
4442             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4443                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4444                         "numbers have changed");
4445                 journal_dev = new_decode_dev(journal_devnum);
4446         } else
4447                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4448
4449         really_read_only = bdev_read_only(sb->s_bdev);
4450
4451         /*
4452          * Are we loading a blank journal or performing recovery after a
4453          * crash?  For recovery, we need to check in advance whether we
4454          * can get read-write access to the device.
4455          */
4456         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4457                 if (sb->s_flags & MS_RDONLY) {
4458                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4459                                         "required on readonly filesystem");
4460                         if (really_read_only) {
4461                                 ext4_msg(sb, KERN_ERR, "write access "
4462                                         "unavailable, cannot proceed");
4463                                 return -EROFS;
4464                         }
4465                         ext4_msg(sb, KERN_INFO, "write access will "
4466                                "be enabled during recovery");
4467                 }
4468         }
4469
4470         if (journal_inum && journal_dev) {
4471                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4472                        "and inode journals!");
4473                 return -EINVAL;
4474         }
4475
4476         if (journal_inum) {
4477                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4478                         return -EINVAL;
4479         } else {
4480                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4481                         return -EINVAL;
4482         }
4483
4484         if (!(journal->j_flags & JBD2_BARRIER))
4485                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4486
4487         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4488                 err = jbd2_journal_wipe(journal, !really_read_only);
4489         if (!err) {
4490                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4491                 if (save)
4492                         memcpy(save, ((char *) es) +
4493                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4494                 err = jbd2_journal_load(journal);
4495                 if (save)
4496                         memcpy(((char *) es) + EXT4_S_ERR_START,
4497                                save, EXT4_S_ERR_LEN);
4498                 kfree(save);
4499         }
4500
4501         if (err) {
4502                 ext4_msg(sb, KERN_ERR, "error loading journal");
4503                 jbd2_journal_destroy(journal);
4504                 return err;
4505         }
4506
4507         EXT4_SB(sb)->s_journal = journal;
4508         ext4_clear_journal_err(sb, es);
4509
4510         if (!really_read_only && journal_devnum &&
4511             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4512                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4513
4514                 /* Make sure we flush the recovery flag to disk. */
4515                 ext4_commit_super(sb, 1);
4516         }
4517
4518         return 0;
4519 }
4520
4521 static int ext4_commit_super(struct super_block *sb, int sync)
4522 {
4523         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4524         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4525         int error = 0;
4526
4527         if (!sbh || block_device_ejected(sb))
4528                 return error;
4529         if (buffer_write_io_error(sbh)) {
4530                 /*
4531                  * Oh, dear.  A previous attempt to write the
4532                  * superblock failed.  This could happen because the
4533                  * USB device was yanked out.  Or it could happen to
4534                  * be a transient write error and maybe the block will
4535                  * be remapped.  Nothing we can do but to retry the
4536                  * write and hope for the best.
4537                  */
4538                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4539                        "superblock detected");
4540                 clear_buffer_write_io_error(sbh);
4541                 set_buffer_uptodate(sbh);
4542         }
4543         /*
4544          * If the file system is mounted read-only, don't update the
4545          * superblock write time.  This avoids updating the superblock
4546          * write time when we are mounting the root file system
4547          * read/only but we need to replay the journal; at that point,
4548          * for people who are east of GMT and who make their clock
4549          * tick in localtime for Windows bug-for-bug compatibility,
4550          * the clock is set in the future, and this will cause e2fsck
4551          * to complain and force a full file system check.
4552          */
4553         if (!(sb->s_flags & MS_RDONLY))
4554                 es->s_wtime = cpu_to_le32(get_seconds());
4555         if (sb->s_bdev->bd_part)
4556                 es->s_kbytes_written =
4557                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4558                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4559                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4560         else
4561                 es->s_kbytes_written =
4562                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4563         ext4_free_blocks_count_set(es,
4564                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4565                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4566         es->s_free_inodes_count =
4567                 cpu_to_le32(percpu_counter_sum_positive(
4568                                 &EXT4_SB(sb)->s_freeinodes_counter));
4569         BUFFER_TRACE(sbh, "marking dirty");
4570         ext4_superblock_csum_set(sb);
4571         mark_buffer_dirty(sbh);
4572         if (sync) {
4573                 error = sync_dirty_buffer(sbh);
4574                 if (error)
4575                         return error;
4576
4577                 error = buffer_write_io_error(sbh);
4578                 if (error) {
4579                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4580                                "superblock");
4581                         clear_buffer_write_io_error(sbh);
4582                         set_buffer_uptodate(sbh);
4583                 }
4584         }
4585         return error;
4586 }
4587
4588 /*
4589  * Have we just finished recovery?  If so, and if we are mounting (or
4590  * remounting) the filesystem readonly, then we will end up with a
4591  * consistent fs on disk.  Record that fact.
4592  */
4593 static void ext4_mark_recovery_complete(struct super_block *sb,
4594                                         struct ext4_super_block *es)
4595 {
4596         journal_t *journal = EXT4_SB(sb)->s_journal;
4597
4598         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4599                 BUG_ON(journal != NULL);
4600                 return;
4601         }
4602         jbd2_journal_lock_updates(journal);
4603         if (jbd2_journal_flush(journal) < 0)
4604                 goto out;
4605
4606         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4607             sb->s_flags & MS_RDONLY) {
4608                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4609                 ext4_commit_super(sb, 1);
4610         }
4611
4612 out:
4613         jbd2_journal_unlock_updates(journal);
4614 }
4615
4616 /*
4617  * If we are mounting (or read-write remounting) a filesystem whose journal
4618  * has recorded an error from a previous lifetime, move that error to the
4619  * main filesystem now.
4620  */
4621 static void ext4_clear_journal_err(struct super_block *sb,
4622                                    struct ext4_super_block *es)
4623 {
4624         journal_t *journal;
4625         int j_errno;
4626         const char *errstr;
4627
4628         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4629
4630         journal = EXT4_SB(sb)->s_journal;
4631
4632         /*
4633          * Now check for any error status which may have been recorded in the
4634          * journal by a prior ext4_error() or ext4_abort()
4635          */
4636
4637         j_errno = jbd2_journal_errno(journal);
4638         if (j_errno) {
4639                 char nbuf[16];
4640
4641                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4642                 ext4_warning(sb, "Filesystem error recorded "
4643                              "from previous mount: %s", errstr);
4644                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4645
4646                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4647                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4648                 ext4_commit_super(sb, 1);
4649
4650                 jbd2_journal_clear_err(journal);
4651                 jbd2_journal_update_sb_errno(journal);
4652         }
4653 }
4654
4655 /*
4656  * Force the running and committing transactions to commit,
4657  * and wait on the commit.
4658  */
4659 int ext4_force_commit(struct super_block *sb)
4660 {
4661         journal_t *journal;
4662
4663         if (sb->s_flags & MS_RDONLY)
4664                 return 0;
4665
4666         journal = EXT4_SB(sb)->s_journal;
4667         return ext4_journal_force_commit(journal);
4668 }
4669
4670 static int ext4_sync_fs(struct super_block *sb, int wait)
4671 {
4672         int ret = 0;
4673         tid_t target;
4674         bool needs_barrier = false;
4675         struct ext4_sb_info *sbi = EXT4_SB(sb);
4676
4677         trace_ext4_sync_fs(sb, wait);
4678         flush_workqueue(sbi->rsv_conversion_wq);
4679         /*
4680          * Writeback quota in non-journalled quota case - journalled quota has
4681          * no dirty dquots
4682          */
4683         dquot_writeback_dquots(sb, -1);
4684         /*
4685          * Data writeback is possible w/o journal transaction, so barrier must
4686          * being sent at the end of the function. But we can skip it if
4687          * transaction_commit will do it for us.
4688          */
4689         target = jbd2_get_latest_transaction(sbi->s_journal);
4690         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4691             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4692                 needs_barrier = true;
4693
4694         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4695                 if (wait)
4696                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4697         }
4698         if (needs_barrier) {
4699                 int err;
4700                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4701                 if (!ret)
4702                         ret = err;
4703         }
4704
4705         return ret;
4706 }
4707
4708 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4709 {
4710         int ret = 0;
4711
4712         trace_ext4_sync_fs(sb, wait);
4713         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4714         dquot_writeback_dquots(sb, -1);
4715         if (wait && test_opt(sb, BARRIER))
4716                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4717
4718         return ret;
4719 }
4720
4721 /*
4722  * LVM calls this function before a (read-only) snapshot is created.  This
4723  * gives us a chance to flush the journal completely and mark the fs clean.
4724  *
4725  * Note that only this function cannot bring a filesystem to be in a clean
4726  * state independently. It relies on upper layer to stop all data & metadata
4727  * modifications.
4728  */
4729 static int ext4_freeze(struct super_block *sb)
4730 {
4731         int error = 0;
4732         journal_t *journal;
4733
4734         if (sb->s_flags & MS_RDONLY)
4735                 return 0;
4736
4737         journal = EXT4_SB(sb)->s_journal;
4738
4739         /* Now we set up the journal barrier. */
4740         jbd2_journal_lock_updates(journal);
4741
4742         /*
4743          * Don't clear the needs_recovery flag if we failed to flush
4744          * the journal.
4745          */
4746         error = jbd2_journal_flush(journal);
4747         if (error < 0)
4748                 goto out;
4749
4750         /* Journal blocked and flushed, clear needs_recovery flag. */
4751         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4752         error = ext4_commit_super(sb, 1);
4753 out:
4754         /* we rely on upper layer to stop further updates */
4755         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4756         return error;
4757 }
4758
4759 /*
4760  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4761  * flag here, even though the filesystem is not technically dirty yet.
4762  */
4763 static int ext4_unfreeze(struct super_block *sb)
4764 {
4765         if (sb->s_flags & MS_RDONLY)
4766                 return 0;
4767
4768         /* Reset the needs_recovery flag before the fs is unlocked. */
4769         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4770         ext4_commit_super(sb, 1);
4771         return 0;
4772 }
4773
4774 /*
4775  * Structure to save mount options for ext4_remount's benefit
4776  */
4777 struct ext4_mount_options {
4778         unsigned long s_mount_opt;
4779         unsigned long s_mount_opt2;
4780         kuid_t s_resuid;
4781         kgid_t s_resgid;
4782         unsigned long s_commit_interval;
4783         u32 s_min_batch_time, s_max_batch_time;
4784 #ifdef CONFIG_QUOTA
4785         int s_jquota_fmt;
4786         char *s_qf_names[MAXQUOTAS];
4787 #endif
4788 };
4789
4790 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4791 {
4792         struct ext4_super_block *es;
4793         struct ext4_sb_info *sbi = EXT4_SB(sb);
4794         unsigned long old_sb_flags;
4795         struct ext4_mount_options old_opts;
4796         int enable_quota = 0;
4797         ext4_group_t g;
4798         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4799         int err = 0;
4800 #ifdef CONFIG_QUOTA
4801         int i, j;
4802 #endif
4803         char *orig_data = kstrdup(data, GFP_KERNEL);
4804
4805         /* Store the original options */
4806         old_sb_flags = sb->s_flags;
4807         old_opts.s_mount_opt = sbi->s_mount_opt;
4808         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4809         old_opts.s_resuid = sbi->s_resuid;
4810         old_opts.s_resgid = sbi->s_resgid;
4811         old_opts.s_commit_interval = sbi->s_commit_interval;
4812         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4813         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4814 #ifdef CONFIG_QUOTA
4815         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4816         for (i = 0; i < MAXQUOTAS; i++)
4817                 if (sbi->s_qf_names[i]) {
4818                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4819                                                          GFP_KERNEL);
4820                         if (!old_opts.s_qf_names[i]) {
4821                                 for (j = 0; j < i; j++)
4822                                         kfree(old_opts.s_qf_names[j]);
4823                                 kfree(orig_data);
4824                                 return -ENOMEM;
4825                         }
4826                 } else
4827                         old_opts.s_qf_names[i] = NULL;
4828 #endif
4829         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4830                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4831
4832         /*
4833          * Allow the "check" option to be passed as a remount option.
4834          */
4835         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4836                 err = -EINVAL;
4837                 goto restore_opts;
4838         }
4839
4840         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4841                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4842                         ext4_msg(sb, KERN_ERR, "can't mount with "
4843                                  "both data=journal and delalloc");
4844                         err = -EINVAL;
4845                         goto restore_opts;
4846                 }
4847                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4848                         ext4_msg(sb, KERN_ERR, "can't mount with "
4849                                  "both data=journal and dioread_nolock");
4850                         err = -EINVAL;
4851                         goto restore_opts;
4852                 }
4853         }
4854
4855         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4856                 ext4_abort(sb, "Abort forced by user");
4857
4858         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4859                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4860
4861         es = sbi->s_es;
4862
4863         if (sbi->s_journal) {
4864                 ext4_init_journal_params(sb, sbi->s_journal);
4865                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4866         }
4867
4868         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4869                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4870                         err = -EROFS;
4871                         goto restore_opts;
4872                 }
4873
4874                 if (*flags & MS_RDONLY) {
4875                         err = sync_filesystem(sb);
4876                         if (err < 0)
4877                                 goto restore_opts;
4878                         err = dquot_suspend(sb, -1);
4879                         if (err < 0)
4880                                 goto restore_opts;
4881
4882                         /*
4883                          * First of all, the unconditional stuff we have to do
4884                          * to disable replay of the journal when we next remount
4885                          */
4886                         sb->s_flags |= MS_RDONLY;
4887
4888                         /*
4889                          * OK, test if we are remounting a valid rw partition
4890                          * readonly, and if so set the rdonly flag and then
4891                          * mark the partition as valid again.
4892                          */
4893                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4894                             (sbi->s_mount_state & EXT4_VALID_FS))
4895                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4896
4897                         if (sbi->s_journal)
4898                                 ext4_mark_recovery_complete(sb, es);
4899                 } else {
4900                         /* Make sure we can mount this feature set readwrite */
4901                         if (!ext4_feature_set_ok(sb, 0)) {
4902                                 err = -EROFS;
4903                                 goto restore_opts;
4904                         }
4905                         /*
4906                          * Make sure the group descriptor checksums
4907                          * are sane.  If they aren't, refuse to remount r/w.
4908                          */
4909                         for (g = 0; g < sbi->s_groups_count; g++) {
4910                                 struct ext4_group_desc *gdp =
4911                                         ext4_get_group_desc(sb, g, NULL);
4912
4913                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4914                                         ext4_msg(sb, KERN_ERR,
4915                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4916                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4917                                                le16_to_cpu(gdp->bg_checksum));
4918                                         err = -EINVAL;
4919                                         goto restore_opts;
4920                                 }
4921                         }
4922
4923                         /*
4924                          * If we have an unprocessed orphan list hanging
4925                          * around from a previously readonly bdev mount,
4926                          * require a full umount/remount for now.
4927                          */
4928                         if (es->s_last_orphan) {
4929                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4930                                        "remount RDWR because of unprocessed "
4931                                        "orphan inode list.  Please "
4932                                        "umount/remount instead");
4933                                 err = -EINVAL;
4934                                 goto restore_opts;
4935                         }
4936
4937                         /*
4938                          * Mounting a RDONLY partition read-write, so reread
4939                          * and store the current valid flag.  (It may have
4940                          * been changed by e2fsck since we originally mounted
4941                          * the partition.)
4942                          */
4943                         if (sbi->s_journal)
4944                                 ext4_clear_journal_err(sb, es);
4945                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4946                         if (!ext4_setup_super(sb, es, 0))
4947                                 sb->s_flags &= ~MS_RDONLY;
4948                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4949                                                      EXT4_FEATURE_INCOMPAT_MMP))
4950                                 if (ext4_multi_mount_protect(sb,
4951                                                 le64_to_cpu(es->s_mmp_block))) {
4952                                         err = -EROFS;
4953                                         goto restore_opts;
4954                                 }
4955                         enable_quota = 1;
4956                 }
4957         }
4958
4959         /*
4960          * Reinitialize lazy itable initialization thread based on
4961          * current settings
4962          */
4963         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4964                 ext4_unregister_li_request(sb);
4965         else {
4966                 ext4_group_t first_not_zeroed;
4967                 first_not_zeroed = ext4_has_uninit_itable(sb);
4968                 ext4_register_li_request(sb, first_not_zeroed);
4969         }
4970
4971         ext4_setup_system_zone(sb);
4972         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4973                 ext4_commit_super(sb, 1);
4974
4975 #ifdef CONFIG_QUOTA
4976         /* Release old quota file names */
4977         for (i = 0; i < MAXQUOTAS; i++)
4978                 kfree(old_opts.s_qf_names[i]);
4979         if (enable_quota) {
4980                 if (sb_any_quota_suspended(sb))
4981                         dquot_resume(sb, -1);
4982                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4983                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4984                         err = ext4_enable_quotas(sb);
4985                         if (err)
4986                                 goto restore_opts;
4987                 }
4988         }
4989 #endif
4990
4991         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4992         kfree(orig_data);
4993         return 0;
4994
4995 restore_opts:
4996         sb->s_flags = old_sb_flags;
4997         sbi->s_mount_opt = old_opts.s_mount_opt;
4998         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4999         sbi->s_resuid = old_opts.s_resuid;
5000         sbi->s_resgid = old_opts.s_resgid;
5001         sbi->s_commit_interval = old_opts.s_commit_interval;
5002         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5003         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5004 #ifdef CONFIG_QUOTA
5005         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5006         for (i = 0; i < MAXQUOTAS; i++) {
5007                 kfree(sbi->s_qf_names[i]);
5008                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5009         }
5010 #endif
5011         kfree(orig_data);
5012         return err;
5013 }
5014
5015 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5016 {
5017         struct super_block *sb = dentry->d_sb;
5018         struct ext4_sb_info *sbi = EXT4_SB(sb);
5019         struct ext4_super_block *es = sbi->s_es;
5020         ext4_fsblk_t overhead = 0, resv_blocks;
5021         u64 fsid;
5022         s64 bfree;
5023         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5024
5025         if (!test_opt(sb, MINIX_DF))
5026                 overhead = sbi->s_overhead;
5027
5028         buf->f_type = EXT4_SUPER_MAGIC;
5029         buf->f_bsize = sb->s_blocksize;
5030         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5031         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5032                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5033         /* prevent underflow in case that few free space is available */
5034         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5035         buf->f_bavail = buf->f_bfree -
5036                         (ext4_r_blocks_count(es) + resv_blocks);
5037         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5038                 buf->f_bavail = 0;
5039         buf->f_files = le32_to_cpu(es->s_inodes_count);
5040         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5041         buf->f_namelen = EXT4_NAME_LEN;
5042         fsid = le64_to_cpup((void *)es->s_uuid) ^
5043                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5044         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5045         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5046
5047         return 0;
5048 }
5049
5050 /* Helper function for writing quotas on sync - we need to start transaction
5051  * before quota file is locked for write. Otherwise the are possible deadlocks:
5052  * Process 1                         Process 2
5053  * ext4_create()                     quota_sync()
5054  *   jbd2_journal_start()                  write_dquot()
5055  *   dquot_initialize()                         down(dqio_mutex)
5056  *     down(dqio_mutex)                    jbd2_journal_start()
5057  *
5058  */
5059
5060 #ifdef CONFIG_QUOTA
5061
5062 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5063 {
5064         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5065 }
5066
5067 static int ext4_write_dquot(struct dquot *dquot)
5068 {
5069         int ret, err;
5070         handle_t *handle;
5071         struct inode *inode;
5072
5073         inode = dquot_to_inode(dquot);
5074         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5075                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5076         if (IS_ERR(handle))
5077                 return PTR_ERR(handle);
5078         ret = dquot_commit(dquot);
5079         err = ext4_journal_stop(handle);
5080         if (!ret)
5081                 ret = err;
5082         return ret;
5083 }
5084
5085 static int ext4_acquire_dquot(struct dquot *dquot)
5086 {
5087         int ret, err;
5088         handle_t *handle;
5089
5090         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5091                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5092         if (IS_ERR(handle))
5093                 return PTR_ERR(handle);
5094         ret = dquot_acquire(dquot);
5095         err = ext4_journal_stop(handle);
5096         if (!ret)
5097                 ret = err;
5098         return ret;
5099 }
5100
5101 static int ext4_release_dquot(struct dquot *dquot)
5102 {
5103         int ret, err;
5104         handle_t *handle;
5105
5106         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5107                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5108         if (IS_ERR(handle)) {
5109                 /* Release dquot anyway to avoid endless cycle in dqput() */
5110                 dquot_release(dquot);
5111                 return PTR_ERR(handle);
5112         }
5113         ret = dquot_release(dquot);
5114         err = ext4_journal_stop(handle);
5115         if (!ret)
5116                 ret = err;
5117         return ret;
5118 }
5119
5120 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5121 {
5122         struct super_block *sb = dquot->dq_sb;
5123         struct ext4_sb_info *sbi = EXT4_SB(sb);
5124
5125         /* Are we journaling quotas? */
5126         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5127             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5128                 dquot_mark_dquot_dirty(dquot);
5129                 return ext4_write_dquot(dquot);
5130         } else {
5131                 return dquot_mark_dquot_dirty(dquot);
5132         }
5133 }
5134
5135 static int ext4_write_info(struct super_block *sb, int type)
5136 {
5137         int ret, err;
5138         handle_t *handle;
5139
5140         /* Data block + inode block */
5141         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5142         if (IS_ERR(handle))
5143                 return PTR_ERR(handle);
5144         ret = dquot_commit_info(sb, type);
5145         err = ext4_journal_stop(handle);
5146         if (!ret)
5147                 ret = err;
5148         return ret;
5149 }
5150
5151 /*
5152  * Turn on quotas during mount time - we need to find
5153  * the quota file and such...
5154  */
5155 static int ext4_quota_on_mount(struct super_block *sb, int type)
5156 {
5157         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5158                                         EXT4_SB(sb)->s_jquota_fmt, type);
5159 }
5160
5161 /*
5162  * Standard function to be called on quota_on
5163  */
5164 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5165                          struct path *path)
5166 {
5167         int err;
5168
5169         if (!test_opt(sb, QUOTA))
5170                 return -EINVAL;
5171
5172         /* Quotafile not on the same filesystem? */
5173         if (path->dentry->d_sb != sb)
5174                 return -EXDEV;
5175         /* Journaling quota? */
5176         if (EXT4_SB(sb)->s_qf_names[type]) {
5177                 /* Quotafile not in fs root? */
5178                 if (path->dentry->d_parent != sb->s_root)
5179                         ext4_msg(sb, KERN_WARNING,
5180                                 "Quota file not on filesystem root. "
5181                                 "Journaled quota will not work");
5182         }
5183
5184         /*
5185          * When we journal data on quota file, we have to flush journal to see
5186          * all updates to the file when we bypass pagecache...
5187          */
5188         if (EXT4_SB(sb)->s_journal &&
5189             ext4_should_journal_data(path->dentry->d_inode)) {
5190                 /*
5191                  * We don't need to lock updates but journal_flush() could
5192                  * otherwise be livelocked...
5193                  */
5194                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5195                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5196                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5197                 if (err)
5198                         return err;
5199         }
5200
5201         return dquot_quota_on(sb, type, format_id, path);
5202 }
5203
5204 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5205                              unsigned int flags)
5206 {
5207         int err;
5208         struct inode *qf_inode;
5209         unsigned long qf_inums[MAXQUOTAS] = {
5210                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5211                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5212         };
5213
5214         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5215
5216         if (!qf_inums[type])
5217                 return -EPERM;
5218
5219         qf_inode = ext4_iget(sb, qf_inums[type]);
5220         if (IS_ERR(qf_inode)) {
5221                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5222                 return PTR_ERR(qf_inode);
5223         }
5224
5225         /* Don't account quota for quota files to avoid recursion */
5226         qf_inode->i_flags |= S_NOQUOTA;
5227         err = dquot_enable(qf_inode, type, format_id, flags);
5228         iput(qf_inode);
5229
5230         return err;
5231 }
5232
5233 /* Enable usage tracking for all quota types. */
5234 static int ext4_enable_quotas(struct super_block *sb)
5235 {
5236         int type, err = 0;
5237         unsigned long qf_inums[MAXQUOTAS] = {
5238                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5239                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5240         };
5241
5242         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5243         for (type = 0; type < MAXQUOTAS; type++) {
5244                 if (qf_inums[type]) {
5245                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5246                                                 DQUOT_USAGE_ENABLED);
5247                         if (err) {
5248                                 ext4_warning(sb,
5249                                         "Failed to enable quota tracking "
5250                                         "(type=%d, err=%d). Please run "
5251                                         "e2fsck to fix.", type, err);
5252                                 return err;
5253                         }
5254                 }
5255         }
5256         return 0;
5257 }
5258
5259 /*
5260  * quota_on function that is used when QUOTA feature is set.
5261  */
5262 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5263                                  int format_id)
5264 {
5265         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5266                 return -EINVAL;
5267
5268         /*
5269          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5270          */
5271         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5272 }
5273
5274 static int ext4_quota_off(struct super_block *sb, int type)
5275 {
5276         struct inode *inode = sb_dqopt(sb)->files[type];
5277         handle_t *handle;
5278
5279         /* Force all delayed allocation blocks to be allocated.
5280          * Caller already holds s_umount sem */
5281         if (test_opt(sb, DELALLOC))
5282                 sync_filesystem(sb);
5283
5284         if (!inode)
5285                 goto out;
5286
5287         /* Update modification times of quota files when userspace can
5288          * start looking at them */
5289         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5290         if (IS_ERR(handle))
5291                 goto out;
5292         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5293         ext4_mark_inode_dirty(handle, inode);
5294         ext4_journal_stop(handle);
5295
5296 out:
5297         return dquot_quota_off(sb, type);
5298 }
5299
5300 /*
5301  * quota_off function that is used when QUOTA feature is set.
5302  */
5303 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5304 {
5305         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5306                 return -EINVAL;
5307
5308         /* Disable only the limits. */
5309         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5310 }
5311
5312 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5313  * acquiring the locks... As quota files are never truncated and quota code
5314  * itself serializes the operations (and no one else should touch the files)
5315  * we don't have to be afraid of races */
5316 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5317                                size_t len, loff_t off)
5318 {
5319         struct inode *inode = sb_dqopt(sb)->files[type];
5320         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5321         int err = 0;
5322         int offset = off & (sb->s_blocksize - 1);
5323         int tocopy;
5324         size_t toread;
5325         struct buffer_head *bh;
5326         loff_t i_size = i_size_read(inode);
5327
5328         if (off > i_size)
5329                 return 0;
5330         if (off+len > i_size)
5331                 len = i_size-off;
5332         toread = len;
5333         while (toread > 0) {
5334                 tocopy = sb->s_blocksize - offset < toread ?
5335                                 sb->s_blocksize - offset : toread;
5336                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5337                 if (err)
5338                         return err;
5339                 if (!bh)        /* A hole? */
5340                         memset(data, 0, tocopy);
5341                 else
5342                         memcpy(data, bh->b_data+offset, tocopy);
5343                 brelse(bh);
5344                 offset = 0;
5345                 toread -= tocopy;
5346                 data += tocopy;
5347                 blk++;
5348         }
5349         return len;
5350 }
5351
5352 /* Write to quotafile (we know the transaction is already started and has
5353  * enough credits) */
5354 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5355                                 const char *data, size_t len, loff_t off)
5356 {
5357         struct inode *inode = sb_dqopt(sb)->files[type];
5358         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5359         int err = 0;
5360         int offset = off & (sb->s_blocksize - 1);
5361         struct buffer_head *bh;
5362         handle_t *handle = journal_current_handle();
5363
5364         if (EXT4_SB(sb)->s_journal && !handle) {
5365                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5366                         " cancelled because transaction is not started",
5367                         (unsigned long long)off, (unsigned long long)len);
5368                 return -EIO;
5369         }
5370         /*
5371          * Since we account only one data block in transaction credits,
5372          * then it is impossible to cross a block boundary.
5373          */
5374         if (sb->s_blocksize - offset < len) {
5375                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5376                         " cancelled because not block aligned",
5377                         (unsigned long long)off, (unsigned long long)len);
5378                 return -EIO;
5379         }
5380
5381         bh = ext4_bread(handle, inode, blk, 1, &err);
5382         if (!bh)
5383                 goto out;
5384         BUFFER_TRACE(bh, "get write access");
5385         err = ext4_journal_get_write_access(handle, bh);
5386         if (err) {
5387                 brelse(bh);
5388                 goto out;
5389         }
5390         lock_buffer(bh);
5391         memcpy(bh->b_data+offset, data, len);
5392         flush_dcache_page(bh->b_page);
5393         unlock_buffer(bh);
5394         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5395         brelse(bh);
5396 out:
5397         if (err)
5398                 return err;
5399         if (inode->i_size < off + len) {
5400                 i_size_write(inode, off + len);
5401                 EXT4_I(inode)->i_disksize = inode->i_size;
5402                 ext4_mark_inode_dirty(handle, inode);
5403         }
5404         return len;
5405 }
5406
5407 #endif
5408
5409 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5410                        const char *dev_name, void *data)
5411 {
5412         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5413 }
5414
5415 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5416 static inline void register_as_ext2(void)
5417 {
5418         int err = register_filesystem(&ext2_fs_type);
5419         if (err)
5420                 printk(KERN_WARNING
5421                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5422 }
5423
5424 static inline void unregister_as_ext2(void)
5425 {
5426         unregister_filesystem(&ext2_fs_type);
5427 }
5428
5429 static inline int ext2_feature_set_ok(struct super_block *sb)
5430 {
5431         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5432                 return 0;
5433         if (sb->s_flags & MS_RDONLY)
5434                 return 1;
5435         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5436                 return 0;
5437         return 1;
5438 }
5439 #else
5440 static inline void register_as_ext2(void) { }
5441 static inline void unregister_as_ext2(void) { }
5442 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5443 #endif
5444
5445 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5446 static inline void register_as_ext3(void)
5447 {
5448         int err = register_filesystem(&ext3_fs_type);
5449         if (err)
5450                 printk(KERN_WARNING
5451                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5452 }
5453
5454 static inline void unregister_as_ext3(void)
5455 {
5456         unregister_filesystem(&ext3_fs_type);
5457 }
5458
5459 static inline int ext3_feature_set_ok(struct super_block *sb)
5460 {
5461         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5462                 return 0;
5463         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5464                 return 0;
5465         if (sb->s_flags & MS_RDONLY)
5466                 return 1;
5467         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5468                 return 0;
5469         return 1;
5470 }
5471 #else
5472 static inline void register_as_ext3(void) { }
5473 static inline void unregister_as_ext3(void) { }
5474 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5475 #endif
5476
5477 static struct file_system_type ext4_fs_type = {
5478         .owner          = THIS_MODULE,
5479         .name           = "ext4",
5480         .mount          = ext4_mount,
5481         .kill_sb        = kill_block_super,
5482         .fs_flags       = FS_REQUIRES_DEV,
5483 };
5484 MODULE_ALIAS_FS("ext4");
5485
5486 static int __init ext4_init_feat_adverts(void)
5487 {
5488         struct ext4_features *ef;
5489         int ret = -ENOMEM;
5490
5491         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5492         if (!ef)
5493                 goto out;
5494
5495         ef->f_kobj.kset = ext4_kset;
5496         init_completion(&ef->f_kobj_unregister);
5497         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5498                                    "features");
5499         if (ret) {
5500                 kfree(ef);
5501                 goto out;
5502         }
5503
5504         ext4_feat = ef;
5505         ret = 0;
5506 out:
5507         return ret;
5508 }
5509
5510 static void ext4_exit_feat_adverts(void)
5511 {
5512         kobject_put(&ext4_feat->f_kobj);
5513         wait_for_completion(&ext4_feat->f_kobj_unregister);
5514         kfree(ext4_feat);
5515 }
5516
5517 /* Shared across all ext4 file systems */
5518 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5519 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5520
5521 static int __init ext4_init_fs(void)
5522 {
5523         int i, err;
5524
5525         ext4_li_info = NULL;
5526         mutex_init(&ext4_li_mtx);
5527
5528         /* Build-time check for flags consistency */
5529         ext4_check_flag_values();
5530
5531         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5532                 mutex_init(&ext4__aio_mutex[i]);
5533                 init_waitqueue_head(&ext4__ioend_wq[i]);
5534         }
5535
5536         err = ext4_init_es();
5537         if (err)
5538                 return err;
5539
5540         err = ext4_init_pageio();
5541         if (err)
5542                 goto out7;
5543
5544         err = ext4_init_system_zone();
5545         if (err)
5546                 goto out6;
5547         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5548         if (!ext4_kset) {
5549                 err = -ENOMEM;
5550                 goto out5;
5551         }
5552         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5553
5554         err = ext4_init_feat_adverts();
5555         if (err)
5556                 goto out4;
5557
5558         err = ext4_init_mballoc();
5559         if (err)
5560                 goto out2;
5561         else
5562                 ext4_mballoc_ready = 1;
5563         err = init_inodecache();
5564         if (err)
5565                 goto out1;
5566         register_as_ext3();
5567         register_as_ext2();
5568         err = register_filesystem(&ext4_fs_type);
5569         if (err)
5570                 goto out;
5571
5572         return 0;
5573 out:
5574         unregister_as_ext2();
5575         unregister_as_ext3();
5576         destroy_inodecache();
5577 out1:
5578         ext4_mballoc_ready = 0;
5579         ext4_exit_mballoc();
5580 out2:
5581         ext4_exit_feat_adverts();
5582 out4:
5583         if (ext4_proc_root)
5584                 remove_proc_entry("fs/ext4", NULL);
5585         kset_unregister(ext4_kset);
5586 out5:
5587         ext4_exit_system_zone();
5588 out6:
5589         ext4_exit_pageio();
5590 out7:
5591         ext4_exit_es();
5592
5593         return err;
5594 }
5595
5596 static void __exit ext4_exit_fs(void)
5597 {
5598         ext4_destroy_lazyinit_thread();
5599         unregister_as_ext2();
5600         unregister_as_ext3();
5601         unregister_filesystem(&ext4_fs_type);
5602         destroy_inodecache();
5603         ext4_exit_mballoc();
5604         ext4_exit_feat_adverts();
5605         remove_proc_entry("fs/ext4", NULL);
5606         kset_unregister(ext4_kset);
5607         ext4_exit_system_zone();
5608         ext4_exit_pageio();
5609         ext4_exit_es();
5610 }
5611
5612 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5613 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5614 MODULE_LICENSE("GPL");
5615 module_init(ext4_init_fs)
5616 module_exit(ext4_exit_fs)