fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 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_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 static int ext4_superblock_csum_verify(struct super_block *sb,
 141                                        struct ext4_super_block *es)
 142 {
 143         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 144                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 145                 return 1;
 146
 147         return es->s_checksum == ext4_superblock_csum(sb, es);
 148 }
 149
 150 void ext4_superblock_csum_set(struct super_block *sb)
 151 {
 152         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 153
 154         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 155                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 156                 return;
 157
 158         es->s_checksum = ext4_superblock_csum(sb, es);
 159 }
 160
 161 void *ext4_kvmalloc(size_t size, gfp_t flags)
 162 {
 163         void *ret;
 164
 165         ret = kmalloc(size, flags | __GFP_NOWARN);
 166         if (!ret)
 167                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 168         return ret;
 169 }
 170
 171 void *ext4_kvzalloc(size_t size, gfp_t flags)
 172 {
 173         void *ret;
 174
 175         ret = kzalloc(size, flags | __GFP_NOWARN);
 176         if (!ret)
 177                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 178         return ret;
 179 }
 180
 181 void ext4_kvfree(void *ptr)
 182 {
 183         if (is_vmalloc_addr(ptr))
 184                 vfree(ptr);
 185         else
 186                 kfree(ptr);
 187
 188 }
 189
 190 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 191                                struct ext4_group_desc *bg)
 192 {
 193         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 196 }
 197
 198 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 199                                struct ext4_group_desc *bg)
 200 {
 201         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 202                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 203                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 204 }
 205
 206 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 207                               struct ext4_group_desc *bg)
 208 {
 209         return le32_to_cpu(bg->bg_inode_table_lo) |
 210                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 211                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 212 }
 213
 214 __u32 ext4_free_group_clusters(struct super_block *sb,
 215                                struct ext4_group_desc *bg)
 216 {
 217         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 218                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 219                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 220 }
 221
 222 __u32 ext4_free_inodes_count(struct super_block *sb,
 223                               struct ext4_group_desc *bg)
 224 {
 225         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 226                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 227                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 228 }
 229
 230 __u32 ext4_used_dirs_count(struct super_block *sb,
 231                               struct ext4_group_desc *bg)
 232 {
 233         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 234                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 235                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 236 }
 237
 238 __u32 ext4_itable_unused_count(struct super_block *sb,
 239                               struct ext4_group_desc *bg)
 240 {
 241         return le16_to_cpu(bg->bg_itable_unused_lo) |
 242                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 243                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 244 }
 245
 246 void ext4_block_bitmap_set(struct super_block *sb,
 247                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 248 {
 249         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 252 }
 253
 254 void ext4_inode_bitmap_set(struct super_block *sb,
 255                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 256 {
 257         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 258         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 259                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 260 }
 261
 262 void ext4_inode_table_set(struct super_block *sb,
 263                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 264 {
 265         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 266         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 267                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 268 }
 269
 270 void ext4_free_group_clusters_set(struct super_block *sb,
 271                                   struct ext4_group_desc *bg, __u32 count)
 272 {
 273         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 274         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 275                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 276 }
 277
 278 void ext4_free_inodes_set(struct super_block *sb,
 279                           struct ext4_group_desc *bg, __u32 count)
 280 {
 281         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 282         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 283                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 284 }
 285
 286 void ext4_used_dirs_set(struct super_block *sb,
 287                           struct ext4_group_desc *bg, __u32 count)
 288 {
 289         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 290         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 291                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 292 }
 293
 294 void ext4_itable_unused_set(struct super_block *sb,
 295                           struct ext4_group_desc *bg, __u32 count)
 296 {
 297         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 298         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 299                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 300 }
 301
 302
 303 static void __save_error_info(struct super_block *sb, const char *func,
 304                             unsigned int line)
 305 {
 306         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 307
 308         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 309         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 310         es->s_last_error_time = cpu_to_le32(get_seconds());
 311         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 312         es->s_last_error_line = cpu_to_le32(line);
 313         if (!es->s_first_error_time) {
 314                 es->s_first_error_time = es->s_last_error_time;
 315                 strncpy(es->s_first_error_func, func,
 316                         sizeof(es->s_first_error_func));
 317                 es->s_first_error_line = cpu_to_le32(line);
 318                 es->s_first_error_ino = es->s_last_error_ino;
 319                 es->s_first_error_block = es->s_last_error_block;
 320         }
 321         /*
 322          * Start the daily error reporting function if it hasn't been
 323          * started already
 324          */
 325         if (!es->s_error_count)
 326                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 327         le32_add_cpu(&es->s_error_count, 1);
 328 }
 329
 330 static void save_error_info(struct super_block *sb, const char *func,
 331                             unsigned int line)
 332 {
 333         __save_error_info(sb, func, line);
 334         ext4_commit_super(sb, 1);
 335 }
 336
 337 /*
 338  * The del_gendisk() function uninitializes the disk-specific data
 339  * structures, including the bdi structure, without telling anyone
 340  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 341  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 342  * This is a kludge to prevent these oops until we can put in a proper
 343  * hook in del_gendisk() to inform the VFS and file system layers.
 344  */
 345 static int block_device_ejected(struct super_block *sb)
 346 {
 347         struct inode *bd_inode = sb->s_bdev->bd_inode;
 348         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 349
 350         return bdi->dev == NULL;
 351 }
 352
 353 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 354 {
 355         struct super_block              *sb = journal->j_private;
 356         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 357         int                             error = is_journal_aborted(journal);
 358         struct ext4_journal_cb_entry    *jce;
 359
 360         BUG_ON(txn->t_state == T_FINISHED);
 361         spin_lock(&sbi->s_md_lock);
 362         while (!list_empty(&txn->t_private_list)) {
 363                 jce = list_entry(txn->t_private_list.next,
 364                                  struct ext4_journal_cb_entry, jce_list);
 365                 list_del_init(&jce->jce_list);
 366                 spin_unlock(&sbi->s_md_lock);
 367                 jce->jce_func(sb, jce, error);
 368                 spin_lock(&sbi->s_md_lock);
 369         }
 370         spin_unlock(&sbi->s_md_lock);
 371 }
 372
 373 /* Deal with the reporting of failure conditions on a filesystem such as
 374  * inconsistencies detected or read IO failures.
 375  *
 376  * On ext2, we can store the error state of the filesystem in the
 377  * superblock.  That is not possible on ext4, because we may have other
 378  * write ordering constraints on the superblock which prevent us from
 379  * writing it out straight away; and given that the journal is about to
 380  * be aborted, we can't rely on the current, or future, transactions to
 381  * write out the superblock safely.
 382  *
 383  * We'll just use the jbd2_journal_abort() error code to record an error in
 384  * the journal instead.  On recovery, the journal will complain about
 385  * that error until we've noted it down and cleared it.
 386  */
 387
 388 static void ext4_handle_error(struct super_block *sb)
 389 {
 390         if (sb->s_flags & MS_RDONLY)
 391                 return;
 392
 393         if (!test_opt(sb, ERRORS_CONT)) {
 394                 journal_t *journal = EXT4_SB(sb)->s_journal;
 395
 396                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 397                 if (journal)
 398                         jbd2_journal_abort(journal, -EIO);
 399         }
 400         if (test_opt(sb, ERRORS_RO)) {
 401                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 402                 /*
 403                  * Make sure updated value of ->s_mount_flags will be visible
 404                  * before ->s_flags update
 405                  */
 406                 smp_wmb();
 407                 sb->s_flags |= MS_RDONLY;
 408         }
 409         if (test_opt(sb, ERRORS_PANIC))
 410                 panic("EXT4-fs (device %s): panic forced after error\n",
 411                         sb->s_id);
 412 }
 413
 414 #define ext4_error_ratelimit(sb)                                        \
 415                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 416                              "EXT4-fs error")
 417
 418 void __ext4_error(struct super_block *sb, const char *function,
 419                   unsigned int line, const char *fmt, ...)
 420 {
 421         struct va_format vaf;
 422         va_list args;
 423
 424         if (ext4_error_ratelimit(sb)) {
 425                 va_start(args, fmt);
 426                 vaf.fmt = fmt;
 427                 vaf.va = &args;
 428                 printk(KERN_CRIT
 429                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 430                        sb->s_id, function, line, current->comm, &vaf);
 431                 va_end(args);
 432         }
 433         save_error_info(sb, function, line);
 434         ext4_handle_error(sb);
 435 }
 436
 437 void __ext4_error_inode(struct inode *inode, const char *function,
 438                         unsigned int line, ext4_fsblk_t block,
 439                         const char *fmt, ...)
 440 {
 441         va_list args;
 442         struct va_format vaf;
 443         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 444
 445         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 446         es->s_last_error_block = cpu_to_le64(block);
 447         if (ext4_error_ratelimit(inode->i_sb)) {
 448                 va_start(args, fmt);
 449                 vaf.fmt = fmt;
 450                 vaf.va = &args;
 451                 if (block)
 452                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 453                                "inode #%lu: block %llu: comm %s: %pV\n",
 454                                inode->i_sb->s_id, function, line, inode->i_ino,
 455                                block, current->comm, &vaf);
 456                 else
 457                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 458                                "inode #%lu: comm %s: %pV\n",
 459                                inode->i_sb->s_id, function, line, inode->i_ino,
 460                                current->comm, &vaf);
 461                 va_end(args);
 462         }
 463         save_error_info(inode->i_sb, function, line);
 464         ext4_handle_error(inode->i_sb);
 465 }
 466
 467 void __ext4_error_file(struct file *file, const char *function,
 468                        unsigned int line, ext4_fsblk_t block,
 469                        const char *fmt, ...)
 470 {
 471         va_list args;
 472         struct va_format vaf;
 473         struct ext4_super_block *es;
 474         struct inode *inode = file_inode(file);
 475         char pathname[80], *path;
 476
 477         es = EXT4_SB(inode->i_sb)->s_es;
 478         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 479         if (ext4_error_ratelimit(inode->i_sb)) {
 480                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 481                 if (IS_ERR(path))
 482                         path = "(unknown)";
 483                 va_start(args, fmt);
 484                 vaf.fmt = fmt;
 485                 vaf.va = &args;
 486                 if (block)
 487                         printk(KERN_CRIT
 488                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 489                                "block %llu: comm %s: path %s: %pV\n",
 490                                inode->i_sb->s_id, function, line, inode->i_ino,
 491                                block, current->comm, path, &vaf);
 492                 else
 493                         printk(KERN_CRIT
 494                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 495                                "comm %s: path %s: %pV\n",
 496                                inode->i_sb->s_id, function, line, inode->i_ino,
 497                                current->comm, path, &vaf);
 498                 va_end(args);
 499         }
 500         save_error_info(inode->i_sb, function, line);
 501         ext4_handle_error(inode->i_sb);
 502 }
 503
 504 const char *ext4_decode_error(struct super_block *sb, int errno,
 505                               char nbuf[16])
 506 {
 507         char *errstr = NULL;
 508
 509         switch (errno) {
 510         case -EIO:
 511                 errstr = "IO failure";
 512                 break;
 513         case -ENOMEM:
 514                 errstr = "Out of memory";
 515                 break;
 516         case -EROFS:
 517                 if (!sb || (EXT4_SB(sb)->s_journal &&
 518                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 519                         errstr = "Journal has aborted";
 520                 else
 521                         errstr = "Readonly filesystem";
 522                 break;
 523         default:
 524                 /* If the caller passed in an extra buffer for unknown
 525                  * errors, textualise them now.  Else we just return
 526                  * NULL. */
 527                 if (nbuf) {
 528                         /* Check for truncated error codes... */
 529                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 530                                 errstr = nbuf;
 531                 }
 532                 break;
 533         }
 534
 535         return errstr;
 536 }
 537
 538 /* __ext4_std_error decodes expected errors from journaling functions
 539  * automatically and invokes the appropriate error response.  */
 540
 541 void __ext4_std_error(struct super_block *sb, const char *function,
 542                       unsigned int line, int errno)
 543 {
 544         char nbuf[16];
 545         const char *errstr;
 546
 547         /* Special case: if the error is EROFS, and we're not already
 548          * inside a transaction, then there's really no point in logging
 549          * an error. */
 550         if (errno == -EROFS && journal_current_handle() == NULL &&
 551             (sb->s_flags & MS_RDONLY))
 552                 return;
 553
 554         if (ext4_error_ratelimit(sb)) {
 555                 errstr = ext4_decode_error(sb, errno, nbuf);
 556                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 557                        sb->s_id, function, line, errstr);
 558         }
 559
 560         save_error_info(sb, function, line);
 561         ext4_handle_error(sb);
 562 }
 563
 564 /*
 565  * ext4_abort is a much stronger failure handler than ext4_error.  The
 566  * abort function may be used to deal with unrecoverable failures such
 567  * as journal IO errors or ENOMEM at a critical moment in log management.
 568  *
 569  * We unconditionally force the filesystem into an ABORT|READONLY state,
 570  * unless the error response on the fs has been set to panic in which
 571  * case we take the easy way out and panic immediately.
 572  */
 573
 574 void __ext4_abort(struct super_block *sb, const char *function,
 575                 unsigned int line, const char *fmt, ...)
 576 {
 577         va_list args;
 578
 579         save_error_info(sb, function, line);
 580         va_start(args, fmt);
 581         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 582                function, line);
 583         vprintk(fmt, args);
 584         printk("\n");
 585         va_end(args);
 586
 587         if ((sb->s_flags & MS_RDONLY) == 0) {
 588                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 589                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 590                 /*
 591                  * Make sure updated value of ->s_mount_flags will be visible
 592                  * before ->s_flags update
 593                  */
 594                 smp_wmb();
 595                 sb->s_flags |= MS_RDONLY;
 596                 if (EXT4_SB(sb)->s_journal)
 597                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 598                 save_error_info(sb, function, line);
 599         }
 600         if (test_opt(sb, ERRORS_PANIC))
 601                 panic("EXT4-fs panic from previous error\n");
 602 }
 603
 604 void __ext4_msg(struct super_block *sb,
 605                 const char *prefix, const char *fmt, ...)
 606 {
 607         struct va_format vaf;
 608         va_list args;
 609
 610         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 611                 return;
 612
 613         va_start(args, fmt);
 614         vaf.fmt = fmt;
 615         vaf.va = &args;
 616         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 617         va_end(args);
 618 }
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 627                           "EXT4-fs warning"))
 628                 return;
 629
 630         va_start(args, fmt);
 631         vaf.fmt = fmt;
 632         vaf.va = &args;
 633         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 634                sb->s_id, function, line, &vaf);
 635         va_end(args);
 636 }
 637
 638 void __ext4_grp_locked_error(const char *function, unsigned int line,
 639                              struct super_block *sb, ext4_group_t grp,
 640                              unsigned long ino, ext4_fsblk_t block,
 641                              const char *fmt, ...)
 642 __releases(bitlock)
 643 __acquires(bitlock)
 644 {
 645         struct va_format vaf;
 646         va_list args;
 647         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 648
 649         es->s_last_error_ino = cpu_to_le32(ino);
 650         es->s_last_error_block = cpu_to_le64(block);
 651         __save_error_info(sb, function, line);
 652
 653         if (ext4_error_ratelimit(sb)) {
 654                 va_start(args, fmt);
 655                 vaf.fmt = fmt;
 656                 vaf.va = &args;
 657                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 658                        sb->s_id, function, line, grp);
 659                 if (ino)
 660                         printk(KERN_CONT "inode %lu: ", ino);
 661                 if (block)
 662                         printk(KERN_CONT "block %llu:",
 663                                (unsigned long long) block);
 664                 printk(KERN_CONT "%pV\n", &vaf);
 665                 va_end(args);
 666         }
 667
 668         if (test_opt(sb, ERRORS_CONT)) {
 669                 ext4_commit_super(sb, 0);
 670                 return;
 671         }
 672
 673         ext4_unlock_group(sb, grp);
 674         ext4_handle_error(sb);
 675         /*
 676          * We only get here in the ERRORS_RO case; relocking the group
 677          * may be dangerous, but nothing bad will happen since the
 678          * filesystem will have already been marked read/only and the
 679          * journal has been aborted.  We return 1 as a hint to callers
 680          * who might what to use the return value from
 681          * ext4_grp_locked_error() to distinguish between the
 682          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 683          * aggressively from the ext4 function in question, with a
 684          * more appropriate error code.
 685          */
 686         ext4_lock_group(sb, grp);
 687         return;
 688 }
 689
 690 void ext4_update_dynamic_rev(struct super_block *sb)
 691 {
 692         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 693
 694         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 695                 return;
 696
 697         ext4_warning(sb,
 698                      "updating to rev %d because of new feature flag, "
 699                      "running e2fsck is recommended",
 700                      EXT4_DYNAMIC_REV);
 701
 702         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 703         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 704         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 705         /* leave es->s_feature_*compat flags alone */
 706         /* es->s_uuid will be set by e2fsck if empty */
 707
 708         /*
 709          * The rest of the superblock fields should be zero, and if not it
 710          * means they are likely already in use, so leave them alone.  We
 711          * can leave it up to e2fsck to clean up any inconsistencies there.
 712          */
 713 }
 714
 715 /*
 716  * Open the external journal device
 717  */
 718 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 719 {
 720         struct block_device *bdev;
 721         char b[BDEVNAME_SIZE];
 722
 723         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 724         if (IS_ERR(bdev))
 725                 goto fail;
 726         return bdev;
 727
 728 fail:
 729         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 730                         __bdevname(dev, b), PTR_ERR(bdev));
 731         return NULL;
 732 }
 733
 734 /*
 735  * Release the journal device
 736  */
 737 static void ext4_blkdev_put(struct block_device *bdev)
 738 {
 739         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 740 }
 741
 742 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 743 {
 744         struct block_device *bdev;
 745         bdev = sbi->journal_bdev;
 746         if (bdev) {
 747                 ext4_blkdev_put(bdev);
 748                 sbi->journal_bdev = NULL;
 749         }
 750 }
 751
 752 static inline struct inode *orphan_list_entry(struct list_head *l)
 753 {
 754         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 755 }
 756
 757 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 758 {
 759         struct list_head *l;
 760
 761         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 762                  le32_to_cpu(sbi->s_es->s_last_orphan));
 763
 764         printk(KERN_ERR "sb_info orphan list:\n");
 765         list_for_each(l, &sbi->s_orphan) {
 766                 struct inode *inode = orphan_list_entry(l);
 767                 printk(KERN_ERR "  "
 768                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 769                        inode->i_sb->s_id, inode->i_ino, inode,
 770                        inode->i_mode, inode->i_nlink,
 771                        NEXT_ORPHAN(inode));
 772         }
 773 }
 774
 775 static void ext4_put_super(struct super_block *sb)
 776 {
 777         struct ext4_sb_info *sbi = EXT4_SB(sb);
 778         struct ext4_super_block *es = sbi->s_es;
 779         int i, err;
 780
 781         ext4_unregister_li_request(sb);
 782         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 783
 784         flush_workqueue(sbi->rsv_conversion_wq);
 785         destroy_workqueue(sbi->rsv_conversion_wq);
 786
 787         if (sbi->s_journal) {
 788                 err = jbd2_journal_destroy(sbi->s_journal);
 789                 sbi->s_journal = NULL;
 790                 if (err < 0)
 791                         ext4_abort(sb, "Couldn't clean up the journal");
 792         }
 793
 794         ext4_es_unregister_shrinker(sbi);
 795         del_timer_sync(&sbi->s_err_report);
 796         ext4_release_system_zone(sb);
 797         ext4_mb_release(sb);
 798         ext4_ext_release(sb);
 799         ext4_xattr_put_super(sb);
 800
 801         if (!(sb->s_flags & MS_RDONLY)) {
 802                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 803                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 804         }
 805         if (!(sb->s_flags & MS_RDONLY))
 806                 ext4_commit_super(sb, 1);
 807
 808         if (sbi->s_proc) {
 809                 remove_proc_entry("options", sbi->s_proc);
 810                 remove_proc_entry(sb->s_id, ext4_proc_root);
 811         }
 812         kobject_del(&sbi->s_kobj);
 813
 814         for (i = 0; i < sbi->s_gdb_count; i++)
 815                 brelse(sbi->s_group_desc[i]);
 816         ext4_kvfree(sbi->s_group_desc);
 817         ext4_kvfree(sbi->s_flex_groups);
 818         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 819         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 820         percpu_counter_destroy(&sbi->s_dirs_counter);
 821         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 822         brelse(sbi->s_sbh);
 823 #ifdef CONFIG_QUOTA
 824         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 825                 kfree(sbi->s_qf_names[i]);
 826 #endif
 827
 828         /* Debugging code just in case the in-memory inode orphan list
 829          * isn't empty.  The on-disk one can be non-empty if we've
 830          * detected an error and taken the fs readonly, but the
 831          * in-memory list had better be clean by this point. */
 832         if (!list_empty(&sbi->s_orphan))
 833                 dump_orphan_list(sb, sbi);
 834         J_ASSERT(list_empty(&sbi->s_orphan));
 835
 836         invalidate_bdev(sb->s_bdev);
 837         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 838                 /*
 839                  * Invalidate the journal device's buffers.  We don't want them
 840                  * floating about in memory - the physical journal device may
 841                  * hotswapped, and it breaks the `ro-after' testing code.
 842                  */
 843                 sync_blockdev(sbi->journal_bdev);
 844                 invalidate_bdev(sbi->journal_bdev);
 845                 ext4_blkdev_remove(sbi);
 846         }
 847         if (sbi->s_mb_cache) {
 848                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 849                 sbi->s_mb_cache = NULL;
 850         }
 851         if (sbi->s_mmp_tsk)
 852                 kthread_stop(sbi->s_mmp_tsk);
 853         sb->s_fs_info = NULL;
 854         /*
 855          * Now that we are completely done shutting down the
 856          * superblock, we need to actually destroy the kobject.
 857          */
 858         kobject_put(&sbi->s_kobj);
 859         wait_for_completion(&sbi->s_kobj_unregister);
 860         if (sbi->s_chksum_driver)
 861                 crypto_free_shash(sbi->s_chksum_driver);
 862         kfree(sbi->s_blockgroup_lock);
 863         kfree(sbi);
 864 }
 865
 866 static struct kmem_cache *ext4_inode_cachep;
 867
 868 /*
 869  * Called inside transaction, so use GFP_NOFS
 870  */
 871 static struct inode *ext4_alloc_inode(struct super_block *sb)
 872 {
 873         struct ext4_inode_info *ei;
 874
 875         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 876         if (!ei)
 877                 return NULL;
 878
 879         ei->vfs_inode.i_version = 1;
 880         spin_lock_init(&ei->i_raw_lock);
 881         INIT_LIST_HEAD(&ei->i_prealloc_list);
 882         spin_lock_init(&ei->i_prealloc_lock);
 883         ext4_es_init_tree(&ei->i_es_tree);
 884         rwlock_init(&ei->i_es_lock);
 885         INIT_LIST_HEAD(&ei->i_es_lru);
 886         ei->i_es_all_nr = 0;
 887         ei->i_es_lru_nr = 0;
 888         ei->i_touch_when = 0;
 889         ei->i_reserved_data_blocks = 0;
 890         ei->i_reserved_meta_blocks = 0;
 891         ei->i_allocated_meta_blocks = 0;
 892         ei->i_da_metadata_calc_len = 0;
 893         ei->i_da_metadata_calc_last_lblock = 0;
 894         spin_lock_init(&(ei->i_block_reservation_lock));
 895 #ifdef CONFIG_QUOTA
 896         ei->i_reserved_quota = 0;
 897 #endif
 898         ei->jinode = NULL;
 899         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 900         spin_lock_init(&ei->i_completed_io_lock);
 901         ei->i_sync_tid = 0;
 902         ei->i_datasync_tid = 0;
 903         atomic_set(&ei->i_ioend_count, 0);
 904         atomic_set(&ei->i_unwritten, 0);
 905         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 906
 907         return &ei->vfs_inode;
 908 }
 909
 910 static int ext4_drop_inode(struct inode *inode)
 911 {
 912         int drop = generic_drop_inode(inode);
 913
 914         trace_ext4_drop_inode(inode, drop);
 915         return drop;
 916 }
 917
 918 static void ext4_i_callback(struct rcu_head *head)
 919 {
 920         struct inode *inode = container_of(head, struct inode, i_rcu);
 921         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 922 }
 923
 924 static void ext4_destroy_inode(struct inode *inode)
 925 {
 926         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 927                 ext4_msg(inode->i_sb, KERN_ERR,
 928                          "Inode %lu (%p): orphan list check failed!",
 929                          inode->i_ino, EXT4_I(inode));
 930                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 931                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 932                                 true);
 933                 dump_stack();
 934         }
 935         call_rcu(&inode->i_rcu, ext4_i_callback);
 936 }
 937
 938 static void init_once(void *foo)
 939 {
 940         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 941
 942         INIT_LIST_HEAD(&ei->i_orphan);
 943         init_rwsem(&ei->xattr_sem);
 944         init_rwsem(&ei->i_data_sem);
 945         inode_init_once(&ei->vfs_inode);
 946 }
 947
 948 static int __init init_inodecache(void)
 949 {
 950         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 951                                              sizeof(struct ext4_inode_info),
 952                                              0, (SLAB_RECLAIM_ACCOUNT|
 953                                                 SLAB_MEM_SPREAD),
 954                                              init_once);
 955         if (ext4_inode_cachep == NULL)
 956                 return -ENOMEM;
 957         return 0;
 958 }
 959
 960 static void destroy_inodecache(void)
 961 {
 962         /*
 963          * Make sure all delayed rcu free inodes are flushed before we
 964          * destroy cache.
 965          */
 966         rcu_barrier();
 967         kmem_cache_destroy(ext4_inode_cachep);
 968 }
 969
 970 void ext4_clear_inode(struct inode *inode)
 971 {
 972         invalidate_inode_buffers(inode);
 973         clear_inode(inode);
 974         dquot_drop(inode);
 975         ext4_discard_preallocations(inode);
 976         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 977         ext4_es_lru_del(inode);
 978         if (EXT4_I(inode)->jinode) {
 979                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 980                                                EXT4_I(inode)->jinode);
 981                 jbd2_free_inode(EXT4_I(inode)->jinode);
 982                 EXT4_I(inode)->jinode = NULL;
 983         }
 984 }
 985
 986 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 987                                         u64 ino, u32 generation)
 988 {
 989         struct inode *inode;
 990
 991         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 992                 return ERR_PTR(-ESTALE);
 993         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 994                 return ERR_PTR(-ESTALE);
 995
 996         /* iget isn't really right if the inode is currently unallocated!!
 997          *
 998          * ext4_read_inode will return a bad_inode if the inode had been
 999          * deleted, so we should be safe.
1000          *
1001          * Currently we don't know the generation for parent directory, so
1002          * a generation of 0 means "accept any"
1003          */
1004         inode = ext4_iget_normal(sb, ino);
1005         if (IS_ERR(inode))
1006                 return ERR_CAST(inode);
1007         if (generation && inode->i_generation != generation) {
1008                 iput(inode);
1009                 return ERR_PTR(-ESTALE);
1010         }
1011
1012         return inode;
1013 }
1014
1015 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1016                                         int fh_len, int fh_type)
1017 {
1018         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1019                                     ext4_nfs_get_inode);
1020 }
1021
1022 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1023                                         int fh_len, int fh_type)
1024 {
1025         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1026                                     ext4_nfs_get_inode);
1027 }
1028
1029 /*
1030  * Try to release metadata pages (indirect blocks, directories) which are
1031  * mapped via the block device.  Since these pages could have journal heads
1032  * which would prevent try_to_free_buffers() from freeing them, we must use
1033  * jbd2 layer's try_to_free_buffers() function to release them.
1034  */
1035 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1036                                  gfp_t wait)
1037 {
1038         journal_t *journal = EXT4_SB(sb)->s_journal;
1039
1040         WARN_ON(PageChecked(page));
1041         if (!page_has_buffers(page))
1042                 return 0;
1043         if (journal)
1044                 return jbd2_journal_try_to_free_buffers(journal, page,
1045                                                         wait & ~__GFP_WAIT);
1046         return try_to_free_buffers(page);
1047 }
1048
1049 #ifdef CONFIG_QUOTA
1050 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1051 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1052
1053 static int ext4_write_dquot(struct dquot *dquot);
1054 static int ext4_acquire_dquot(struct dquot *dquot);
1055 static int ext4_release_dquot(struct dquot *dquot);
1056 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1057 static int ext4_write_info(struct super_block *sb, int type);
1058 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1059                          struct path *path);
1060 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1061                                  int format_id);
1062 static int ext4_quota_off(struct super_block *sb, int type);
1063 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1064 static int ext4_quota_on_mount(struct super_block *sb, int type);
1065 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1066                                size_t len, loff_t off);
1067 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1068                                 const char *data, size_t len, loff_t off);
1069 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1070                              unsigned int flags);
1071 static int ext4_enable_quotas(struct super_block *sb);
1072
1073 static const struct dquot_operations ext4_quota_operations = {
1074         .get_reserved_space = ext4_get_reserved_space,
1075         .write_dquot    = ext4_write_dquot,
1076         .acquire_dquot  = ext4_acquire_dquot,
1077         .release_dquot  = ext4_release_dquot,
1078         .mark_dirty     = ext4_mark_dquot_dirty,
1079         .write_info     = ext4_write_info,
1080         .alloc_dquot    = dquot_alloc,
1081         .destroy_dquot  = dquot_destroy,
1082 };
1083
1084 static const struct quotactl_ops ext4_qctl_operations = {
1085         .quota_on       = ext4_quota_on,
1086         .quota_off      = ext4_quota_off,
1087         .quota_sync     = dquot_quota_sync,
1088         .get_info       = dquot_get_dqinfo,
1089         .set_info       = dquot_set_dqinfo,
1090         .get_dqblk      = dquot_get_dqblk,
1091         .set_dqblk      = dquot_set_dqblk
1092 };
1093
1094 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1095         .quota_on_meta  = ext4_quota_on_sysfile,
1096         .quota_off      = ext4_quota_off_sysfile,
1097         .quota_sync     = dquot_quota_sync,
1098         .get_info       = dquot_get_dqinfo,
1099         .set_info       = dquot_set_dqinfo,
1100         .get_dqblk      = dquot_get_dqblk,
1101         .set_dqblk      = dquot_set_dqblk
1102 };
1103 #endif
1104
1105 static const struct super_operations ext4_sops = {
1106         .alloc_inode    = ext4_alloc_inode,
1107         .destroy_inode  = ext4_destroy_inode,
1108         .write_inode    = ext4_write_inode,
1109         .dirty_inode    = ext4_dirty_inode,
1110         .drop_inode     = ext4_drop_inode,
1111         .evict_inode    = ext4_evict_inode,
1112         .put_super      = ext4_put_super,
1113         .sync_fs        = ext4_sync_fs,
1114         .freeze_fs      = ext4_freeze,
1115         .unfreeze_fs    = ext4_unfreeze,
1116         .statfs         = ext4_statfs,
1117         .remount_fs     = ext4_remount,
1118         .show_options   = ext4_show_options,
1119 #ifdef CONFIG_QUOTA
1120         .quota_read     = ext4_quota_read,
1121         .quota_write    = ext4_quota_write,
1122 #endif
1123         .bdev_try_to_free_page = bdev_try_to_free_page,
1124 };
1125
1126 static const struct export_operations ext4_export_ops = {
1127         .fh_to_dentry = ext4_fh_to_dentry,
1128         .fh_to_parent = ext4_fh_to_parent,
1129         .get_parent = ext4_get_parent,
1130 };
1131
1132 enum {
1133         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1134         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1135         Opt_nouid32, Opt_debug, Opt_removed,
1136         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1137         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1138         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1139         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1140         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1141         Opt_data_err_abort, Opt_data_err_ignore,
1142         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1143         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1144         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1145         Opt_usrquota, Opt_grpquota, Opt_i_version,
1146         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1147         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1148         Opt_inode_readahead_blks, Opt_journal_ioprio,
1149         Opt_dioread_nolock, Opt_dioread_lock,
1150         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1151         Opt_max_dir_size_kb,
1152 };
1153
1154 static const match_table_t tokens = {
1155         {Opt_bsd_df, "bsddf"},
1156         {Opt_minix_df, "minixdf"},
1157         {Opt_grpid, "grpid"},
1158         {Opt_grpid, "bsdgroups"},
1159         {Opt_nogrpid, "nogrpid"},
1160         {Opt_nogrpid, "sysvgroups"},
1161         {Opt_resgid, "resgid=%u"},
1162         {Opt_resuid, "resuid=%u"},
1163         {Opt_sb, "sb=%u"},
1164         {Opt_err_cont, "errors=continue"},
1165         {Opt_err_panic, "errors=panic"},
1166         {Opt_err_ro, "errors=remount-ro"},
1167         {Opt_nouid32, "nouid32"},
1168         {Opt_debug, "debug"},
1169         {Opt_removed, "oldalloc"},
1170         {Opt_removed, "orlov"},
1171         {Opt_user_xattr, "user_xattr"},
1172         {Opt_nouser_xattr, "nouser_xattr"},
1173         {Opt_acl, "acl"},
1174         {Opt_noacl, "noacl"},
1175         {Opt_noload, "norecovery"},
1176         {Opt_noload, "noload"},
1177         {Opt_removed, "nobh"},
1178         {Opt_removed, "bh"},
1179         {Opt_commit, "commit=%u"},
1180         {Opt_min_batch_time, "min_batch_time=%u"},
1181         {Opt_max_batch_time, "max_batch_time=%u"},
1182         {Opt_journal_dev, "journal_dev=%u"},
1183         {Opt_journal_path, "journal_path=%s"},
1184         {Opt_journal_checksum, "journal_checksum"},
1185         {Opt_journal_async_commit, "journal_async_commit"},
1186         {Opt_abort, "abort"},
1187         {Opt_data_journal, "data=journal"},
1188         {Opt_data_ordered, "data=ordered"},
1189         {Opt_data_writeback, "data=writeback"},
1190         {Opt_data_err_abort, "data_err=abort"},
1191         {Opt_data_err_ignore, "data_err=ignore"},
1192         {Opt_offusrjquota, "usrjquota="},
1193         {Opt_usrjquota, "usrjquota=%s"},
1194         {Opt_offgrpjquota, "grpjquota="},
1195         {Opt_grpjquota, "grpjquota=%s"},
1196         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1197         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1198         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1199         {Opt_grpquota, "grpquota"},
1200         {Opt_noquota, "noquota"},
1201         {Opt_quota, "quota"},
1202         {Opt_usrquota, "usrquota"},
1203         {Opt_barrier, "barrier=%u"},
1204         {Opt_barrier, "barrier"},
1205         {Opt_nobarrier, "nobarrier"},
1206         {Opt_i_version, "i_version"},
1207         {Opt_stripe, "stripe=%u"},
1208         {Opt_delalloc, "delalloc"},
1209         {Opt_nodelalloc, "nodelalloc"},
1210         {Opt_removed, "mblk_io_submit"},
1211         {Opt_removed, "nomblk_io_submit"},
1212         {Opt_block_validity, "block_validity"},
1213         {Opt_noblock_validity, "noblock_validity"},
1214         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1215         {Opt_journal_ioprio, "journal_ioprio=%u"},
1216         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1217         {Opt_auto_da_alloc, "auto_da_alloc"},
1218         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1219         {Opt_dioread_nolock, "dioread_nolock"},
1220         {Opt_dioread_lock, "dioread_lock"},
1221         {Opt_discard, "discard"},
1222         {Opt_nodiscard, "nodiscard"},
1223         {Opt_init_itable, "init_itable=%u"},
1224         {Opt_init_itable, "init_itable"},
1225         {Opt_noinit_itable, "noinit_itable"},
1226         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1227         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1228         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1229         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1230         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1231         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1232         {Opt_err, NULL},
1233 };
1234
1235 static ext4_fsblk_t get_sb_block(void **data)
1236 {
1237         ext4_fsblk_t    sb_block;
1238         char            *options = (char *) *data;
1239
1240         if (!options || strncmp(options, "sb=", 3) != 0)
1241                 return 1;       /* Default location */
1242
1243         options += 3;
1244         /* TODO: use simple_strtoll with >32bit ext4 */
1245         sb_block = simple_strtoul(options, &options, 0);
1246         if (*options && *options != ',') {
1247                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1248                        (char *) *data);
1249                 return 1;
1250         }
1251         if (*options == ',')
1252                 options++;
1253         *data = (void *) options;
1254
1255         return sb_block;
1256 }
1257
1258 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1259 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1260         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1261
1262 #ifdef CONFIG_QUOTA
1263 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1264 {
1265         struct ext4_sb_info *sbi = EXT4_SB(sb);
1266         char *qname;
1267         int ret = -1;
1268
1269         if (sb_any_quota_loaded(sb) &&
1270                 !sbi->s_qf_names[qtype]) {
1271                 ext4_msg(sb, KERN_ERR,
1272                         "Cannot change journaled "
1273                         "quota options when quota turned on");
1274                 return -1;
1275         }
1276         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1277                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1278                          "when QUOTA feature is enabled");
1279                 return -1;
1280         }
1281         qname = match_strdup(args);
1282         if (!qname) {
1283                 ext4_msg(sb, KERN_ERR,
1284                         "Not enough memory for storing quotafile name");
1285                 return -1;
1286         }
1287         if (sbi->s_qf_names[qtype]) {
1288                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1289                         ret = 1;
1290                 else
1291                         ext4_msg(sb, KERN_ERR,
1292                                  "%s quota file already specified",
1293                                  QTYPE2NAME(qtype));
1294                 goto errout;
1295         }
1296         if (strchr(qname, '/')) {
1297                 ext4_msg(sb, KERN_ERR,
1298                         "quotafile must be on filesystem root");
1299                 goto errout;
1300         }
1301         sbi->s_qf_names[qtype] = qname;
1302         set_opt(sb, QUOTA);
1303         return 1;
1304 errout:
1305         kfree(qname);
1306         return ret;
1307 }
1308
1309 static int clear_qf_name(struct super_block *sb, int qtype)
1310 {
1311
1312         struct ext4_sb_info *sbi = EXT4_SB(sb);
1313
1314         if (sb_any_quota_loaded(sb) &&
1315                 sbi->s_qf_names[qtype]) {
1316                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1317                         " when quota turned on");
1318                 return -1;
1319         }
1320         kfree(sbi->s_qf_names[qtype]);
1321         sbi->s_qf_names[qtype] = NULL;
1322         return 1;
1323 }
1324 #endif
1325
1326 #define MOPT_SET        0x0001
1327 #define MOPT_CLEAR      0x0002
1328 #define MOPT_NOSUPPORT  0x0004
1329 #define MOPT_EXPLICIT   0x0008
1330 #define MOPT_CLEAR_ERR  0x0010
1331 #define MOPT_GTE0       0x0020
1332 #ifdef CONFIG_QUOTA
1333 #define MOPT_Q          0
1334 #define MOPT_QFMT       0x0040
1335 #else
1336 #define MOPT_Q          MOPT_NOSUPPORT
1337 #define MOPT_QFMT       MOPT_NOSUPPORT
1338 #endif
1339 #define MOPT_DATAJ      0x0080
1340 #define MOPT_NO_EXT2    0x0100
1341 #define MOPT_NO_EXT3    0x0200
1342 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1343 #define MOPT_STRING     0x0400
1344
1345 static const struct mount_opts {
1346         int     token;
1347         int     mount_opt;
1348         int     flags;
1349 } ext4_mount_opts[] = {
1350         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1351         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1352         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1353         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1354         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1355         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1356         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1357          MOPT_EXT4_ONLY | MOPT_SET},
1358         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1359          MOPT_EXT4_ONLY | MOPT_CLEAR},
1360         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1361         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1362         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1363          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1364         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1365          MOPT_EXT4_ONLY | MOPT_CLEAR},
1366         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1367          MOPT_EXT4_ONLY | MOPT_SET},
1368         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1369                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1370          MOPT_EXT4_ONLY | MOPT_SET},
1371         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1372         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1373         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1374         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1375         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1376          MOPT_NO_EXT2 | MOPT_SET},
1377         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1378          MOPT_NO_EXT2 | MOPT_CLEAR},
1379         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1380         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1381         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1382         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1383         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1384         {Opt_commit, 0, MOPT_GTE0},
1385         {Opt_max_batch_time, 0, MOPT_GTE0},
1386         {Opt_min_batch_time, 0, MOPT_GTE0},
1387         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1388         {Opt_init_itable, 0, MOPT_GTE0},
1389         {Opt_stripe, 0, MOPT_GTE0},
1390         {Opt_resuid, 0, MOPT_GTE0},
1391         {Opt_resgid, 0, MOPT_GTE0},
1392         {Opt_journal_dev, 0, MOPT_GTE0},
1393         {Opt_journal_path, 0, MOPT_STRING},
1394         {Opt_journal_ioprio, 0, MOPT_GTE0},
1395         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1396         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1397         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1398          MOPT_NO_EXT2 | MOPT_DATAJ},
1399         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1400         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1401 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1402         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1403         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1404 #else
1405         {Opt_acl, 0, MOPT_NOSUPPORT},
1406         {Opt_noacl, 0, MOPT_NOSUPPORT},
1407 #endif
1408         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1409         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1410         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1411         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1412                                                         MOPT_SET | MOPT_Q},
1413         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1414                                                         MOPT_SET | MOPT_Q},
1415         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1416                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1417         {Opt_usrjquota, 0, MOPT_Q},
1418         {Opt_grpjquota, 0, MOPT_Q},
1419         {Opt_offusrjquota, 0, MOPT_Q},
1420         {Opt_offgrpjquota, 0, MOPT_Q},
1421         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1422         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1423         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1424         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1425         {Opt_err, 0, 0}
1426 };
1427
1428 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1429                             substring_t *args, unsigned long *journal_devnum,
1430                             unsigned int *journal_ioprio, int is_remount)
1431 {
1432         struct ext4_sb_info *sbi = EXT4_SB(sb);
1433         const struct mount_opts *m;
1434         kuid_t uid;
1435         kgid_t gid;
1436         int arg = 0;
1437
1438 #ifdef CONFIG_QUOTA
1439         if (token == Opt_usrjquota)
1440                 return set_qf_name(sb, USRQUOTA, &args[0]);
1441         else if (token == Opt_grpjquota)
1442                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1443         else if (token == Opt_offusrjquota)
1444                 return clear_qf_name(sb, USRQUOTA);
1445         else if (token == Opt_offgrpjquota)
1446                 return clear_qf_name(sb, GRPQUOTA);
1447 #endif
1448         switch (token) {
1449         case Opt_noacl:
1450         case Opt_nouser_xattr:
1451                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1452                 break;
1453         case Opt_sb:
1454                 return 1;       /* handled by get_sb_block() */
1455         case Opt_removed:
1456                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1457                 return 1;
1458         case Opt_abort:
1459                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1460                 return 1;
1461         case Opt_i_version:
1462                 sb->s_flags |= MS_I_VERSION;
1463                 return 1;
1464         }
1465
1466         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1467                 if (token == m->token)
1468                         break;
1469
1470         if (m->token == Opt_err) {
1471                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1472                          "or missing value", opt);
1473                 return -1;
1474         }
1475
1476         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1477                 ext4_msg(sb, KERN_ERR,
1478                          "Mount option \"%s\" incompatible with ext2", opt);
1479                 return -1;
1480         }
1481         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1482                 ext4_msg(sb, KERN_ERR,
1483                          "Mount option \"%s\" incompatible with ext3", opt);
1484                 return -1;
1485         }
1486
1487         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1488                 return -1;
1489         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1490                 return -1;
1491         if (m->flags & MOPT_EXPLICIT)
1492                 set_opt2(sb, EXPLICIT_DELALLOC);
1493         if (m->flags & MOPT_CLEAR_ERR)
1494                 clear_opt(sb, ERRORS_MASK);
1495         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1496                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1497                          "options when quota turned on");
1498                 return -1;
1499         }
1500
1501         if (m->flags & MOPT_NOSUPPORT) {
1502                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1503         } else if (token == Opt_commit) {
1504                 if (arg == 0)
1505                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1506                 sbi->s_commit_interval = HZ * arg;
1507         } else if (token == Opt_max_batch_time) {
1508                 sbi->s_max_batch_time = arg;
1509         } else if (token == Opt_min_batch_time) {
1510                 sbi->s_min_batch_time = arg;
1511         } else if (token == Opt_inode_readahead_blks) {
1512                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1513                         ext4_msg(sb, KERN_ERR,
1514                                  "EXT4-fs: inode_readahead_blks must be "
1515                                  "0 or a power of 2 smaller than 2^31");
1516                         return -1;
1517                 }
1518                 sbi->s_inode_readahead_blks = arg;
1519         } else if (token == Opt_init_itable) {
1520                 set_opt(sb, INIT_INODE_TABLE);
1521                 if (!args->from)
1522                         arg = EXT4_DEF_LI_WAIT_MULT;
1523                 sbi->s_li_wait_mult = arg;
1524         } else if (token == Opt_max_dir_size_kb) {
1525                 sbi->s_max_dir_size_kb = arg;
1526         } else if (token == Opt_stripe) {
1527                 sbi->s_stripe = arg;
1528         } else if (token == Opt_resuid) {
1529                 uid = make_kuid(current_user_ns(), arg);
1530                 if (!uid_valid(uid)) {
1531                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1532                         return -1;
1533                 }
1534                 sbi->s_resuid = uid;
1535         } else if (token == Opt_resgid) {
1536                 gid = make_kgid(current_user_ns(), arg);
1537                 if (!gid_valid(gid)) {
1538                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1539                         return -1;
1540                 }
1541                 sbi->s_resgid = gid;
1542         } else if (token == Opt_journal_dev) {
1543                 if (is_remount) {
1544                         ext4_msg(sb, KERN_ERR,
1545                                  "Cannot specify journal on remount");
1546                         return -1;
1547                 }
1548                 *journal_devnum = arg;
1549         } else if (token == Opt_journal_path) {
1550                 char *journal_path;
1551                 struct inode *journal_inode;
1552                 struct path path;
1553                 int error;
1554
1555                 if (is_remount) {
1556                         ext4_msg(sb, KERN_ERR,
1557                                  "Cannot specify journal on remount");
1558                         return -1;
1559                 }
1560                 journal_path = match_strdup(&args[0]);
1561                 if (!journal_path) {
1562                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1563                                 "journal device string");
1564                         return -1;
1565                 }
1566
1567                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1568                 if (error) {
1569                         ext4_msg(sb, KERN_ERR, "error: could not find "
1570                                 "journal device path: error %d", error);
1571                         kfree(journal_path);
1572                         return -1;
1573                 }
1574
1575                 journal_inode = path.dentry->d_inode;
1576                 if (!S_ISBLK(journal_inode->i_mode)) {
1577                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1578                                 "is not a block device", journal_path);
1579                         path_put(&path);
1580                         kfree(journal_path);
1581                         return -1;
1582                 }
1583
1584                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1585                 path_put(&path);
1586                 kfree(journal_path);
1587         } else if (token == Opt_journal_ioprio) {
1588                 if (arg > 7) {
1589                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1590                                  " (must be 0-7)");
1591                         return -1;
1592                 }
1593                 *journal_ioprio =
1594                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1595         } else if (m->flags & MOPT_DATAJ) {
1596                 if (is_remount) {
1597                         if (!sbi->s_journal)
1598                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1599                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1600                                 ext4_msg(sb, KERN_ERR,
1601                                          "Cannot change data mode on remount");
1602                                 return -1;
1603                         }
1604                 } else {
1605                         clear_opt(sb, DATA_FLAGS);
1606                         sbi->s_mount_opt |= m->mount_opt;
1607                 }
1608 #ifdef CONFIG_QUOTA
1609         } else if (m->flags & MOPT_QFMT) {
1610                 if (sb_any_quota_loaded(sb) &&
1611                     sbi->s_jquota_fmt != m->mount_opt) {
1612                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1613                                  "quota options when quota turned on");
1614                         return -1;
1615                 }
1616                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1617                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1618                         ext4_msg(sb, KERN_ERR,
1619                                  "Cannot set journaled quota options "
1620                                  "when QUOTA feature is enabled");
1621                         return -1;
1622                 }
1623                 sbi->s_jquota_fmt = m->mount_opt;
1624 #endif
1625         } else {
1626                 if (!args->from)
1627                         arg = 1;
1628                 if (m->flags & MOPT_CLEAR)
1629                         arg = !arg;
1630                 else if (unlikely(!(m->flags & MOPT_SET))) {
1631                         ext4_msg(sb, KERN_WARNING,
1632                                  "buggy handling of option %s", opt);
1633                         WARN_ON(1);
1634                         return -1;
1635                 }
1636                 if (arg != 0)
1637                         sbi->s_mount_opt |= m->mount_opt;
1638                 else
1639                         sbi->s_mount_opt &= ~m->mount_opt;
1640         }
1641         return 1;
1642 }
1643
1644 static int parse_options(char *options, struct super_block *sb,
1645                          unsigned long *journal_devnum,
1646                          unsigned int *journal_ioprio,
1647                          int is_remount)
1648 {
1649         struct ext4_sb_info *sbi = EXT4_SB(sb);
1650         char *p;
1651         substring_t args[MAX_OPT_ARGS];
1652         int token;
1653
1654         if (!options)
1655                 return 1;
1656
1657         while ((p = strsep(&options, ",")) != NULL) {
1658                 if (!*p)
1659                         continue;
1660                 /*
1661                  * Initialize args struct so we know whether arg was
1662                  * found; some options take optional arguments.
1663                  */
1664                 args[0].to = args[0].from = NULL;
1665                 token = match_token(p, tokens, args);
1666                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1667                                      journal_ioprio, is_remount) < 0)
1668                         return 0;
1669         }
1670 #ifdef CONFIG_QUOTA
1671         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1672             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1673                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1674                          "feature is enabled");
1675                 return 0;
1676         }
1677         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1678                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1679                         clear_opt(sb, USRQUOTA);
1680
1681                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1682                         clear_opt(sb, GRPQUOTA);
1683
1684                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1685                         ext4_msg(sb, KERN_ERR, "old and new quota "
1686                                         "format mixing");
1687                         return 0;
1688                 }
1689
1690                 if (!sbi->s_jquota_fmt) {
1691                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1692                                         "not specified");
1693                         return 0;
1694                 }
1695         }
1696 #endif
1697         if (test_opt(sb, DIOREAD_NOLOCK)) {
1698                 int blocksize =
1699                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1700
1701                 if (blocksize < PAGE_CACHE_SIZE) {
1702                         ext4_msg(sb, KERN_ERR, "can't mount with "
1703                                  "dioread_nolock if block size != PAGE_SIZE");
1704                         return 0;
1705                 }
1706         }
1707         return 1;
1708 }
1709
1710 static inline void ext4_show_quota_options(struct seq_file *seq,
1711                                            struct super_block *sb)
1712 {
1713 #if defined(CONFIG_QUOTA)
1714         struct ext4_sb_info *sbi = EXT4_SB(sb);
1715
1716         if (sbi->s_jquota_fmt) {
1717                 char *fmtname = "";
1718
1719                 switch (sbi->s_jquota_fmt) {
1720                 case QFMT_VFS_OLD:
1721                         fmtname = "vfsold";
1722                         break;
1723                 case QFMT_VFS_V0:
1724                         fmtname = "vfsv0";
1725                         break;
1726                 case QFMT_VFS_V1:
1727                         fmtname = "vfsv1";
1728                         break;
1729                 }
1730                 seq_printf(seq, ",jqfmt=%s", fmtname);
1731         }
1732
1733         if (sbi->s_qf_names[USRQUOTA])
1734                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1735
1736         if (sbi->s_qf_names[GRPQUOTA])
1737                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1738 #endif
1739 }
1740
1741 static const char *token2str(int token)
1742 {
1743         const struct match_token *t;
1744
1745         for (t = tokens; t->token != Opt_err; t++)
1746                 if (t->token == token && !strchr(t->pattern, '='))
1747                         break;
1748         return t->pattern;
1749 }
1750
1751 /*
1752  * Show an option if
1753  *  - it's set to a non-default value OR
1754  *  - if the per-sb default is different from the global default
1755  */
1756 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1757                               int nodefs)
1758 {
1759         struct ext4_sb_info *sbi = EXT4_SB(sb);
1760         struct ext4_super_block *es = sbi->s_es;
1761         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1762         const struct mount_opts *m;
1763         char sep = nodefs ? '\n' : ',';
1764
1765 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1766 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1767
1768         if (sbi->s_sb_block != 1)
1769                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1770
1771         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1772                 int want_set = m->flags & MOPT_SET;
1773                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1774                     (m->flags & MOPT_CLEAR_ERR))
1775                         continue;
1776                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1777                         continue; /* skip if same as the default */
1778                 if ((want_set &&
1779                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1780                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1781                         continue; /* select Opt_noFoo vs Opt_Foo */
1782                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1783         }
1784
1785         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1786             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1787                 SEQ_OPTS_PRINT("resuid=%u",
1788                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1789         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1790             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1791                 SEQ_OPTS_PRINT("resgid=%u",
1792                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1793         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1794         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1795                 SEQ_OPTS_PUTS("errors=remount-ro");
1796         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1797                 SEQ_OPTS_PUTS("errors=continue");
1798         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1799                 SEQ_OPTS_PUTS("errors=panic");
1800         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1801                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1802         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1803                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1804         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1805                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1806         if (sb->s_flags & MS_I_VERSION)
1807                 SEQ_OPTS_PUTS("i_version");
1808         if (nodefs || sbi->s_stripe)
1809                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1810         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1811                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1812                         SEQ_OPTS_PUTS("data=journal");
1813                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1814                         SEQ_OPTS_PUTS("data=ordered");
1815                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1816                         SEQ_OPTS_PUTS("data=writeback");
1817         }
1818         if (nodefs ||
1819             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1820                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1821                                sbi->s_inode_readahead_blks);
1822
1823         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1824                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1825                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1826         if (nodefs || sbi->s_max_dir_size_kb)
1827                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1828
1829         ext4_show_quota_options(seq, sb);
1830         return 0;
1831 }
1832
1833 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1834 {
1835         return _ext4_show_options(seq, root->d_sb, 0);
1836 }
1837
1838 static int options_seq_show(struct seq_file *seq, void *offset)
1839 {
1840         struct super_block *sb = seq->private;
1841         int rc;
1842
1843         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1844         rc = _ext4_show_options(seq, sb, 1);
1845         seq_puts(seq, "\n");
1846         return rc;
1847 }
1848
1849 static int options_open_fs(struct inode *inode, struct file *file)
1850 {
1851         return single_open(file, options_seq_show, PDE_DATA(inode));
1852 }
1853
1854 static const struct file_operations ext4_seq_options_fops = {
1855         .owner = THIS_MODULE,
1856         .open = options_open_fs,
1857         .read = seq_read,
1858         .llseek = seq_lseek,
1859         .release = single_release,
1860 };
1861
1862 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1863                             int read_only)
1864 {
1865         struct ext4_sb_info *sbi = EXT4_SB(sb);
1866         int res = 0;
1867
1868         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1869                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1870                          "forcing read-only mode");
1871                 res = MS_RDONLY;
1872         }
1873         if (read_only)
1874                 goto done;
1875         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1876                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1877                          "running e2fsck is recommended");
1878         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1879                 ext4_msg(sb, KERN_WARNING,
1880                          "warning: mounting fs with errors, "
1881                          "running e2fsck is recommended");
1882         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1883                  le16_to_cpu(es->s_mnt_count) >=
1884                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1885                 ext4_msg(sb, KERN_WARNING,
1886                          "warning: maximal mount count reached, "
1887                          "running e2fsck is recommended");
1888         else if (le32_to_cpu(es->s_checkinterval) &&
1889                 (le32_to_cpu(es->s_lastcheck) +
1890                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "warning: checktime reached, "
1893                          "running e2fsck is recommended");
1894         if (!sbi->s_journal)
1895                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1896         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1897                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1898         le16_add_cpu(&es->s_mnt_count, 1);
1899         es->s_mtime = cpu_to_le32(get_seconds());
1900         ext4_update_dynamic_rev(sb);
1901         if (sbi->s_journal)
1902                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1903
1904         ext4_commit_super(sb, 1);
1905 done:
1906         if (test_opt(sb, DEBUG))
1907                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1908                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1909                         sb->s_blocksize,
1910                         sbi->s_groups_count,
1911                         EXT4_BLOCKS_PER_GROUP(sb),
1912                         EXT4_INODES_PER_GROUP(sb),
1913                         sbi->s_mount_opt, sbi->s_mount_opt2);
1914
1915         cleancache_init_fs(sb);
1916         return res;
1917 }
1918
1919 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1920 {
1921         struct ext4_sb_info *sbi = EXT4_SB(sb);
1922         struct flex_groups *new_groups;
1923         int size;
1924
1925         if (!sbi->s_log_groups_per_flex)
1926                 return 0;
1927
1928         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1929         if (size <= sbi->s_flex_groups_allocated)
1930                 return 0;
1931
1932         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1933         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1934         if (!new_groups) {
1935                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1936                          size / (int) sizeof(struct flex_groups));
1937                 return -ENOMEM;
1938         }
1939
1940         if (sbi->s_flex_groups) {
1941                 memcpy(new_groups, sbi->s_flex_groups,
1942                        (sbi->s_flex_groups_allocated *
1943                         sizeof(struct flex_groups)));
1944                 ext4_kvfree(sbi->s_flex_groups);
1945         }
1946         sbi->s_flex_groups = new_groups;
1947         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1948         return 0;
1949 }
1950
1951 static int ext4_fill_flex_info(struct super_block *sb)
1952 {
1953         struct ext4_sb_info *sbi = EXT4_SB(sb);
1954         struct ext4_group_desc *gdp = NULL;
1955         ext4_group_t flex_group;
1956         int i, err;
1957
1958         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1959         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1960                 sbi->s_log_groups_per_flex = 0;
1961                 return 1;
1962         }
1963
1964         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1965         if (err)
1966                 goto failed;
1967
1968         for (i = 0; i < sbi->s_groups_count; i++) {
1969                 gdp = ext4_get_group_desc(sb, i, NULL);
1970
1971                 flex_group = ext4_flex_group(sbi, i);
1972                 atomic_add(ext4_free_inodes_count(sb, gdp),
1973                            &sbi->s_flex_groups[flex_group].free_inodes);
1974                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1975                              &sbi->s_flex_groups[flex_group].free_clusters);
1976                 atomic_add(ext4_used_dirs_count(sb, gdp),
1977                            &sbi->s_flex_groups[flex_group].used_dirs);
1978         }
1979
1980         return 1;
1981 failed:
1982         return 0;
1983 }
1984
1985 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1986                                    struct ext4_group_desc *gdp)
1987 {
1988         int offset;
1989         __u16 crc = 0;
1990         __le32 le_group = cpu_to_le32(block_group);
1991
1992         if ((sbi->s_es->s_feature_ro_compat &
1993              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1994                 /* Use new metadata_csum algorithm */
1995                 __le16 save_csum;
1996                 __u32 csum32;
1997
1998                 save_csum = gdp->bg_checksum;
1999                 gdp->bg_checksum = 0;
2000                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2001                                      sizeof(le_group));
2002                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2003                                      sbi->s_desc_size);
2004                 gdp->bg_checksum = save_csum;
2005
2006                 crc = csum32 & 0xFFFF;
2007                 goto out;
2008         }
2009
2010         /* old crc16 code */
2011         offset = offsetof(struct ext4_group_desc, bg_checksum);
2012
2013         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2014         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2015         crc = crc16(crc, (__u8 *)gdp, offset);
2016         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2017         /* for checksum of struct ext4_group_desc do the rest...*/
2018         if ((sbi->s_es->s_feature_incompat &
2019              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2020             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2021                 crc = crc16(crc, (__u8 *)gdp + offset,
2022                             le16_to_cpu(sbi->s_es->s_desc_size) -
2023                                 offset);
2024
2025 out:
2026         return cpu_to_le16(crc);
2027 }
2028
2029 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2030                                 struct ext4_group_desc *gdp)
2031 {
2032         if (ext4_has_group_desc_csum(sb) &&
2033             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2034                                                       block_group, gdp)))
2035                 return 0;
2036
2037         return 1;
2038 }
2039
2040 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2041                               struct ext4_group_desc *gdp)
2042 {
2043         if (!ext4_has_group_desc_csum(sb))
2044                 return;
2045         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2046 }
2047
2048 /* Called at mount-time, super-block is locked */
2049 static int ext4_check_descriptors(struct super_block *sb,
2050                                   ext4_group_t *first_not_zeroed)
2051 {
2052         struct ext4_sb_info *sbi = EXT4_SB(sb);
2053         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2054         ext4_fsblk_t last_block;
2055         ext4_fsblk_t block_bitmap;
2056         ext4_fsblk_t inode_bitmap;
2057         ext4_fsblk_t inode_table;
2058         int flexbg_flag = 0;
2059         ext4_group_t i, grp = sbi->s_groups_count;
2060
2061         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2062                 flexbg_flag = 1;
2063
2064         ext4_debug("Checking group descriptors");
2065
2066         for (i = 0; i < sbi->s_groups_count; i++) {
2067                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2068
2069                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2070                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2071                 else
2072                         last_block = first_block +
2073                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2074
2075                 if ((grp == sbi->s_groups_count) &&
2076                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2077                         grp = i;
2078
2079                 block_bitmap = ext4_block_bitmap(sb, gdp);
2080                 if (block_bitmap < first_block || block_bitmap > last_block) {
2081                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2082                                "Block bitmap for group %u not in group "
2083                                "(block %llu)!", i, block_bitmap);
2084                         return 0;
2085                 }
2086                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2087                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2088                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2089                                "Inode bitmap for group %u not in group "
2090                                "(block %llu)!", i, inode_bitmap);
2091                         return 0;
2092                 }
2093                 inode_table = ext4_inode_table(sb, gdp);
2094                 if (inode_table < first_block ||
2095                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2096                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2097                                "Inode table for group %u not in group "
2098                                "(block %llu)!", i, inode_table);
2099                         return 0;
2100                 }
2101                 ext4_lock_group(sb, i);
2102                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2103                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2104                                  "Checksum for group %u failed (%u!=%u)",
2105                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2106                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2107                         if (!(sb->s_flags & MS_RDONLY)) {
2108                                 ext4_unlock_group(sb, i);
2109                                 return 0;
2110                         }
2111                 }
2112                 ext4_unlock_group(sb, i);
2113                 if (!flexbg_flag)
2114                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2115         }
2116         if (NULL != first_not_zeroed)
2117                 *first_not_zeroed = grp;
2118         return 1;
2119 }
2120
2121 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2122  * the superblock) which were deleted from all directories, but held open by
2123  * a process at the time of a crash.  We walk the list and try to delete these
2124  * inodes at recovery time (only with a read-write filesystem).
2125  *
2126  * In order to keep the orphan inode chain consistent during traversal (in
2127  * case of crash during recovery), we link each inode into the superblock
2128  * orphan list_head and handle it the same way as an inode deletion during
2129  * normal operation (which journals the operations for us).
2130  *
2131  * We only do an iget() and an iput() on each inode, which is very safe if we
2132  * accidentally point at an in-use or already deleted inode.  The worst that
2133  * can happen in this case is that we get a "bit already cleared" message from
2134  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2135  * e2fsck was run on this filesystem, and it must have already done the orphan
2136  * inode cleanup for us, so we can safely abort without any further action.
2137  */
2138 static void ext4_orphan_cleanup(struct super_block *sb,
2139                                 struct ext4_super_block *es)
2140 {
2141         unsigned int s_flags = sb->s_flags;
2142         int nr_orphans = 0, nr_truncates = 0;
2143 #ifdef CONFIG_QUOTA
2144         int i;
2145 #endif
2146         if (!es->s_last_orphan) {
2147                 jbd_debug(4, "no orphan inodes to clean up\n");
2148                 return;
2149         }
2150
2151         if (bdev_read_only(sb->s_bdev)) {
2152                 ext4_msg(sb, KERN_ERR, "write access "
2153                         "unavailable, skipping orphan cleanup");
2154                 return;
2155         }
2156
2157         /* Check if feature set would not allow a r/w mount */
2158         if (!ext4_feature_set_ok(sb, 0)) {
2159                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2160                          "unknown ROCOMPAT features");
2161                 return;
2162         }
2163
2164         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2165                 /* don't clear list on RO mount w/ errors */
2166                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2167                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2168                                   "clearing orphan list.\n");
2169                         es->s_last_orphan = 0;
2170                 }
2171                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2172                 return;
2173         }
2174
2175         if (s_flags & MS_RDONLY) {
2176                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2177                 sb->s_flags &= ~MS_RDONLY;
2178         }
2179 #ifdef CONFIG_QUOTA
2180         /* Needed for iput() to work correctly and not trash data */
2181         sb->s_flags |= MS_ACTIVE;
2182         /* Turn on quotas so that they are updated correctly */
2183         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2184                 if (EXT4_SB(sb)->s_qf_names[i]) {
2185                         int ret = ext4_quota_on_mount(sb, i);
2186                         if (ret < 0)
2187                                 ext4_msg(sb, KERN_ERR,
2188                                         "Cannot turn on journaled "
2189                                         "quota: error %d", ret);
2190                 }
2191         }
2192 #endif
2193
2194         while (es->s_last_orphan) {
2195                 struct inode *inode;
2196
2197                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2198                 if (IS_ERR(inode)) {
2199                         es->s_last_orphan = 0;
2200                         break;
2201                 }
2202
2203                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2204                 dquot_initialize(inode);
2205                 if (inode->i_nlink) {
2206                         if (test_opt(sb, DEBUG))
2207                                 ext4_msg(sb, KERN_DEBUG,
2208                                         "%s: truncating inode %lu to %lld bytes",
2209                                         __func__, inode->i_ino, inode->i_size);
2210                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2211                                   inode->i_ino, inode->i_size);
2212                         mutex_lock(&inode->i_mutex);
2213                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2214                         ext4_truncate(inode);
2215                         mutex_unlock(&inode->i_mutex);
2216                         nr_truncates++;
2217                 } else {
2218                         if (test_opt(sb, DEBUG))
2219                                 ext4_msg(sb, KERN_DEBUG,
2220                                         "%s: deleting unreferenced inode %lu",
2221                                         __func__, inode->i_ino);
2222                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2223                                   inode->i_ino);
2224                         nr_orphans++;
2225                 }
2226                 iput(inode);  /* The delete magic happens here! */
2227         }
2228
2229 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2230
2231         if (nr_orphans)
2232                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2233                        PLURAL(nr_orphans));
2234         if (nr_truncates)
2235                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2236                        PLURAL(nr_truncates));
2237 #ifdef CONFIG_QUOTA
2238         /* Turn quotas off */
2239         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2240                 if (sb_dqopt(sb)->files[i])
2241                         dquot_quota_off(sb, i);
2242         }
2243 #endif
2244         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2245 }
2246
2247 /*
2248  * Maximal extent format file size.
2249  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2250  * extent format containers, within a sector_t, and within i_blocks
2251  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2252  * so that won't be a limiting factor.
2253  *
2254  * However there is other limiting factor. We do store extents in the form
2255  * of starting block and length, hence the resulting length of the extent
2256  * covering maximum file size must fit into on-disk format containers as
2257  * well. Given that length is always by 1 unit bigger than max unit (because
2258  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2259  *
2260  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2261  */
2262 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2263 {
2264         loff_t res;
2265         loff_t upper_limit = MAX_LFS_FILESIZE;
2266
2267         /* small i_blocks in vfs inode? */
2268         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2269                 /*
2270                  * CONFIG_LBDAF is not enabled implies the inode
2271                  * i_block represent total blocks in 512 bytes
2272                  * 32 == size of vfs inode i_blocks * 8
2273                  */
2274                 upper_limit = (1LL << 32) - 1;
2275
2276                 /* total blocks in file system block size */
2277                 upper_limit >>= (blkbits - 9);
2278                 upper_limit <<= blkbits;
2279         }
2280
2281         /*
2282          * 32-bit extent-start container, ee_block. We lower the maxbytes
2283          * by one fs block, so ee_len can cover the extent of maximum file
2284          * size
2285          */
2286         res = (1LL << 32) - 1;
2287         res <<= blkbits;
2288
2289         /* Sanity check against vm- & vfs- imposed limits */
2290         if (res > upper_limit)
2291                 res = upper_limit;
2292
2293         return res;
2294 }
2295
2296 /*
2297  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2298  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2299  * We need to be 1 filesystem block less than the 2^48 sector limit.
2300  */
2301 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2302 {
2303         loff_t res = EXT4_NDIR_BLOCKS;
2304         int meta_blocks;
2305         loff_t upper_limit;
2306         /* This is calculated to be the largest file size for a dense, block
2307          * mapped file such that the file's total number of 512-byte sectors,
2308          * including data and all indirect blocks, does not exceed (2^48 - 1).
2309          *
2310          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2311          * number of 512-byte sectors of the file.
2312          */
2313
2314         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2315                 /*
2316                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2317                  * the inode i_block field represents total file blocks in
2318                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2319                  */
2320                 upper_limit = (1LL << 32) - 1;
2321
2322                 /* total blocks in file system block size */
2323                 upper_limit >>= (bits - 9);
2324
2325         } else {
2326                 /*
2327                  * We use 48 bit ext4_inode i_blocks
2328                  * With EXT4_HUGE_FILE_FL set the i_blocks
2329                  * represent total number of blocks in
2330                  * file system block size
2331                  */
2332                 upper_limit = (1LL << 48) - 1;
2333
2334         }
2335
2336         /* indirect blocks */
2337         meta_blocks = 1;
2338         /* double indirect blocks */
2339         meta_blocks += 1 + (1LL << (bits-2));
2340         /* tripple indirect blocks */
2341         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2342
2343         upper_limit -= meta_blocks;
2344         upper_limit <<= bits;
2345
2346         res += 1LL << (bits-2);
2347         res += 1LL << (2*(bits-2));
2348         res += 1LL << (3*(bits-2));
2349         res <<= bits;
2350         if (res > upper_limit)
2351                 res = upper_limit;
2352
2353         if (res > MAX_LFS_FILESIZE)
2354                 res = MAX_LFS_FILESIZE;
2355
2356         return res;
2357 }
2358
2359 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2360                                    ext4_fsblk_t logical_sb_block, int nr)
2361 {
2362         struct ext4_sb_info *sbi = EXT4_SB(sb);
2363         ext4_group_t bg, first_meta_bg;
2364         int has_super = 0;
2365
2366         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2367
2368         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2369             nr < first_meta_bg)
2370                 return logical_sb_block + nr + 1;
2371         bg = sbi->s_desc_per_block * nr;
2372         if (ext4_bg_has_super(sb, bg))
2373                 has_super = 1;
2374
2375         /*
2376          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2377          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2378          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2379          * compensate.
2380          */
2381         if (sb->s_blocksize == 1024 && nr == 0 &&
2382             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2383                 has_super++;
2384
2385         return (has_super + ext4_group_first_block_no(sb, bg));
2386 }
2387
2388 /**
2389  * ext4_get_stripe_size: Get the stripe size.
2390  * @sbi: In memory super block info
2391  *
2392  * If we have specified it via mount option, then
2393  * use the mount option value. If the value specified at mount time is
2394  * greater than the blocks per group use the super block value.
2395  * If the super block value is greater than blocks per group return 0.
2396  * Allocator needs it be less than blocks per group.
2397  *
2398  */
2399 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2400 {
2401         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2402         unsigned long stripe_width =
2403                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2404         int ret;
2405
2406         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2407                 ret = sbi->s_stripe;
2408         else if (stripe_width <= sbi->s_blocks_per_group)
2409                 ret = stripe_width;
2410         else if (stride <= sbi->s_blocks_per_group)
2411                 ret = stride;
2412         else
2413                 ret = 0;
2414
2415         /*
2416          * If the stripe width is 1, this makes no sense and
2417          * we set it to 0 to turn off stripe handling code.
2418          */
2419         if (ret <= 1)
2420                 ret = 0;
2421
2422         return ret;
2423 }
2424
2425 /* sysfs supprt */
2426
2427 struct ext4_attr {
2428         struct attribute attr;
2429         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2430         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2431                          const char *, size_t);
2432         union {
2433                 int offset;
2434                 int deprecated_val;
2435         } u;
2436 };
2437
2438 static int parse_strtoull(const char *buf,
2439                 unsigned long long max, unsigned long long *value)
2440 {
2441         int ret;
2442
2443         ret = kstrtoull(skip_spaces(buf), 0, value);
2444         if (!ret && *value > max)
2445                 ret = -EINVAL;
2446         return ret;
2447 }
2448
2449 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2450                                               struct ext4_sb_info *sbi,
2451                                               char *buf)
2452 {
2453         return snprintf(buf, PAGE_SIZE, "%llu\n",
2454                 (s64) EXT4_C2B(sbi,
2455                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2456 }
2457
2458 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2459                                          struct ext4_sb_info *sbi, char *buf)
2460 {
2461         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2462
2463         if (!sb->s_bdev->bd_part)
2464                 return snprintf(buf, PAGE_SIZE, "0\n");
2465         return snprintf(buf, PAGE_SIZE, "%lu\n",
2466                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2467                          sbi->s_sectors_written_start) >> 1);
2468 }
2469
2470 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2471                                           struct ext4_sb_info *sbi, char *buf)
2472 {
2473         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2474
2475         if (!sb->s_bdev->bd_part)
2476                 return snprintf(buf, PAGE_SIZE, "0\n");
2477         return snprintf(buf, PAGE_SIZE, "%llu\n",
2478                         (unsigned long long)(sbi->s_kbytes_written +
2479                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2480                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2481 }
2482
2483 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2484                                           struct ext4_sb_info *sbi,
2485                                           const char *buf, size_t count)
2486 {
2487         unsigned long t;
2488         int ret;
2489
2490         ret = kstrtoul(skip_spaces(buf), 0, &t);
2491         if (ret)
2492                 return ret;
2493
2494         if (t && (!is_power_of_2(t) || t > 0x40000000))
2495                 return -EINVAL;
2496
2497         sbi->s_inode_readahead_blks = t;
2498         return count;
2499 }
2500
2501 static ssize_t sbi_ui_show(struct ext4_attr *a,
2502                            struct ext4_sb_info *sbi, char *buf)
2503 {
2504         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2505
2506         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2507 }
2508
2509 static ssize_t sbi_ui_store(struct ext4_attr *a,
2510                             struct ext4_sb_info *sbi,
2511                             const char *buf, size_t count)
2512 {
2513         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2514         unsigned long t;
2515         int ret;
2516
2517         ret = kstrtoul(skip_spaces(buf), 0, &t);
2518         if (ret)
2519                 return ret;
2520         *ui = t;
2521         return count;
2522 }
2523
2524 static ssize_t es_ui_show(struct ext4_attr *a,
2525                            struct ext4_sb_info *sbi, char *buf)
2526 {
2527
2528         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2529                            a->u.offset);
2530
2531         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2532 }
2533
2534 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2535                                   struct ext4_sb_info *sbi, char *buf)
2536 {
2537         return snprintf(buf, PAGE_SIZE, "%llu\n",
2538                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2539 }
2540
2541 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2542                                    struct ext4_sb_info *sbi,
2543                                    const char *buf, size_t count)
2544 {
2545         unsigned long long val;
2546         int ret;
2547
2548         if (parse_strtoull(buf, -1ULL, &val))
2549                 return -EINVAL;
2550         ret = ext4_reserve_clusters(sbi, val);
2551
2552         return ret ? ret : count;
2553 }
2554
2555 static ssize_t trigger_test_error(struct ext4_attr *a,
2556                                   struct ext4_sb_info *sbi,
2557                                   const char *buf, size_t count)
2558 {
2559         int len = count;
2560
2561         if (!capable(CAP_SYS_ADMIN))
2562                 return -EPERM;
2563
2564         if (len && buf[len-1] == '\n')
2565                 len--;
2566
2567         if (len)
2568                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2569         return count;
2570 }
2571
2572 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2573                                    struct ext4_sb_info *sbi, char *buf)
2574 {
2575         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2576 }
2577
2578 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2579 static struct ext4_attr ext4_attr_##_name = {                   \
2580         .attr = {.name = __stringify(_name), .mode = _mode },   \
2581         .show   = _show,                                        \
2582         .store  = _store,                                       \
2583         .u = {                                                  \
2584                 .offset = offsetof(struct ext4_sb_info, _elname),\
2585         },                                                      \
2586 }
2587
2588 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2589 static struct ext4_attr ext4_attr_##_name = {                           \
2590         .attr = {.name = __stringify(_name), .mode = _mode },           \
2591         .show   = _show,                                                \
2592         .store  = _store,                                               \
2593         .u = {                                                          \
2594                 .offset = offsetof(struct ext4_super_block, _elname),   \
2595         },                                                              \
2596 }
2597
2598 #define EXT4_ATTR(name, mode, show, store) \
2599 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2600
2601 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2602 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2603 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2604
2605 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2606         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2607 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2608         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2609
2610 #define ATTR_LIST(name) &ext4_attr_##name.attr
2611 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2612 static struct ext4_attr ext4_attr_##_name = {                   \
2613         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2614         .show   = sbi_deprecated_show,                          \
2615         .u = {                                                  \
2616                 .deprecated_val = _val,                         \
2617         },                                                      \
2618 }
2619
2620 EXT4_RO_ATTR(delayed_allocation_blocks);
2621 EXT4_RO_ATTR(session_write_kbytes);
2622 EXT4_RO_ATTR(lifetime_write_kbytes);
2623 EXT4_RW_ATTR(reserved_clusters);
2624 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2625                  inode_readahead_blks_store, s_inode_readahead_blks);
2626 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2627 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2628 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2629 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2630 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2631 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2632 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2633 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2634 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2635 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2636 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2637 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2638 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2639 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2640 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2641 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2642 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2643 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2644 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2645
2646 static struct attribute *ext4_attrs[] = {
2647         ATTR_LIST(delayed_allocation_blocks),
2648         ATTR_LIST(session_write_kbytes),
2649         ATTR_LIST(lifetime_write_kbytes),
2650         ATTR_LIST(reserved_clusters),
2651         ATTR_LIST(inode_readahead_blks),
2652         ATTR_LIST(inode_goal),
2653         ATTR_LIST(mb_stats),
2654         ATTR_LIST(mb_max_to_scan),
2655         ATTR_LIST(mb_min_to_scan),
2656         ATTR_LIST(mb_order2_req),
2657         ATTR_LIST(mb_stream_req),
2658         ATTR_LIST(mb_group_prealloc),
2659         ATTR_LIST(max_writeback_mb_bump),
2660         ATTR_LIST(extent_max_zeroout_kb),
2661         ATTR_LIST(trigger_fs_error),
2662         ATTR_LIST(err_ratelimit_interval_ms),
2663         ATTR_LIST(err_ratelimit_burst),
2664         ATTR_LIST(warning_ratelimit_interval_ms),
2665         ATTR_LIST(warning_ratelimit_burst),
2666         ATTR_LIST(msg_ratelimit_interval_ms),
2667         ATTR_LIST(msg_ratelimit_burst),
2668         ATTR_LIST(errors_count),
2669         ATTR_LIST(first_error_time),
2670         ATTR_LIST(last_error_time),
2671         NULL,
2672 };
2673
2674 /* Features this copy of ext4 supports */
2675 EXT4_INFO_ATTR(lazy_itable_init);
2676 EXT4_INFO_ATTR(batched_discard);
2677 EXT4_INFO_ATTR(meta_bg_resize);
2678
2679 static struct attribute *ext4_feat_attrs[] = {
2680         ATTR_LIST(lazy_itable_init),
2681         ATTR_LIST(batched_discard),
2682         ATTR_LIST(meta_bg_resize),
2683         NULL,
2684 };
2685
2686 static ssize_t ext4_attr_show(struct kobject *kobj,
2687                               struct attribute *attr, char *buf)
2688 {
2689         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2690                                                 s_kobj);
2691         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2692
2693         return a->show ? a->show(a, sbi, buf) : 0;
2694 }
2695
2696 static ssize_t ext4_attr_store(struct kobject *kobj,
2697                                struct attribute *attr,
2698                                const char *buf, size_t len)
2699 {
2700         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2701                                                 s_kobj);
2702         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2703
2704         return a->store ? a->store(a, sbi, buf, len) : 0;
2705 }
2706
2707 static void ext4_sb_release(struct kobject *kobj)
2708 {
2709         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2710                                                 s_kobj);
2711         complete(&sbi->s_kobj_unregister);
2712 }
2713
2714 static const struct sysfs_ops ext4_attr_ops = {
2715         .show   = ext4_attr_show,
2716         .store  = ext4_attr_store,
2717 };
2718
2719 static struct kobj_type ext4_ktype = {
2720         .default_attrs  = ext4_attrs,
2721         .sysfs_ops      = &ext4_attr_ops,
2722         .release        = ext4_sb_release,
2723 };
2724
2725 static void ext4_feat_release(struct kobject *kobj)
2726 {
2727         complete(&ext4_feat->f_kobj_unregister);
2728 }
2729
2730 static ssize_t ext4_feat_show(struct kobject *kobj,
2731                               struct attribute *attr, char *buf)
2732 {
2733         return snprintf(buf, PAGE_SIZE, "supported\n");
2734 }
2735
2736 /*
2737  * We can not use ext4_attr_show/store because it relies on the kobject
2738  * being embedded in the ext4_sb_info structure which is definitely not
2739  * true in this case.
2740  */
2741 static const struct sysfs_ops ext4_feat_ops = {
2742         .show   = ext4_feat_show,
2743         .store  = NULL,
2744 };
2745
2746 static struct kobj_type ext4_feat_ktype = {
2747         .default_attrs  = ext4_feat_attrs,
2748         .sysfs_ops      = &ext4_feat_ops,
2749         .release        = ext4_feat_release,
2750 };
2751
2752 /*
2753  * Check whether this filesystem can be mounted based on
2754  * the features present and the RDONLY/RDWR mount requested.
2755  * Returns 1 if this filesystem can be mounted as requested,
2756  * 0 if it cannot be.
2757  */
2758 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2759 {
2760         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2761                 ext4_msg(sb, KERN_ERR,
2762                         "Couldn't mount because of "
2763                         "unsupported optional features (%x)",
2764                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2765                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2766                 return 0;
2767         }
2768
2769         if (readonly)
2770                 return 1;
2771
2772         /* Check that feature set is OK for a read-write mount */
2773         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2774                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2775                          "unsupported optional features (%x)",
2776                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2777                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2778                 return 0;
2779         }
2780         /*
2781          * Large file size enabled file system can only be mounted
2782          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2783          */
2784         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2785                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2786                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2787                                  "cannot be mounted RDWR without "
2788                                  "CONFIG_LBDAF");
2789                         return 0;
2790                 }
2791         }
2792         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2793             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2794                 ext4_msg(sb, KERN_ERR,
2795                          "Can't support bigalloc feature without "
2796                          "extents feature\n");
2797                 return 0;
2798         }
2799
2800 #ifndef CONFIG_QUOTA
2801         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2802             !readonly) {
2803                 ext4_msg(sb, KERN_ERR,
2804                          "Filesystem with quota feature cannot be mounted RDWR "
2805                          "without CONFIG_QUOTA");
2806                 return 0;
2807         }
2808 #endif  /* CONFIG_QUOTA */
2809         return 1;
2810 }
2811
2812 /*
2813  * This function is called once a day if we have errors logged
2814  * on the file system
2815  */
2816 static void print_daily_error_info(unsigned long arg)
2817 {
2818         struct super_block *sb = (struct super_block *) arg;
2819         struct ext4_sb_info *sbi;
2820         struct ext4_super_block *es;
2821
2822         sbi = EXT4_SB(sb);
2823         es = sbi->s_es;
2824
2825         if (es->s_error_count)
2826                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2827                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2828                          le32_to_cpu(es->s_error_count));
2829         if (es->s_first_error_time) {
2830                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2831                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2832                        (int) sizeof(es->s_first_error_func),
2833                        es->s_first_error_func,
2834                        le32_to_cpu(es->s_first_error_line));
2835                 if (es->s_first_error_ino)
2836                         printk(": inode %u",
2837                                le32_to_cpu(es->s_first_error_ino));
2838                 if (es->s_first_error_block)
2839                         printk(": block %llu", (unsigned long long)
2840                                le64_to_cpu(es->s_first_error_block));
2841                 printk("\n");
2842         }
2843         if (es->s_last_error_time) {
2844                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2845                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2846                        (int) sizeof(es->s_last_error_func),
2847                        es->s_last_error_func,
2848                        le32_to_cpu(es->s_last_error_line));
2849                 if (es->s_last_error_ino)
2850                         printk(": inode %u",
2851                                le32_to_cpu(es->s_last_error_ino));
2852                 if (es->s_last_error_block)
2853                         printk(": block %llu", (unsigned long long)
2854                                le64_to_cpu(es->s_last_error_block));
2855                 printk("\n");
2856         }
2857         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2858 }
2859
2860 /* Find next suitable group and run ext4_init_inode_table */
2861 static int ext4_run_li_request(struct ext4_li_request *elr)
2862 {
2863         struct ext4_group_desc *gdp = NULL;
2864         ext4_group_t group, ngroups;
2865         struct super_block *sb;
2866         unsigned long timeout = 0;
2867         int ret = 0;
2868
2869         sb = elr->lr_super;
2870         ngroups = EXT4_SB(sb)->s_groups_count;
2871
2872         sb_start_write(sb);
2873         for (group = elr->lr_next_group; group < ngroups; group++) {
2874                 gdp = ext4_get_group_desc(sb, group, NULL);
2875                 if (!gdp) {
2876                         ret = 1;
2877                         break;
2878                 }
2879
2880                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2881                         break;
2882         }
2883
2884         if (group >= ngroups)
2885                 ret = 1;
2886
2887         if (!ret) {
2888                 timeout = jiffies;
2889                 ret = ext4_init_inode_table(sb, group,
2890                                             elr->lr_timeout ? 0 : 1);
2891                 if (elr->lr_timeout == 0) {
2892                         timeout = (jiffies - timeout) *
2893                                   elr->lr_sbi->s_li_wait_mult;
2894                         elr->lr_timeout = timeout;
2895                 }
2896                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2897                 elr->lr_next_group = group + 1;
2898         }
2899         sb_end_write(sb);
2900
2901         return ret;
2902 }
2903
2904 /*
2905  * Remove lr_request from the list_request and free the
2906  * request structure. Should be called with li_list_mtx held
2907  */
2908 static void ext4_remove_li_request(struct ext4_li_request *elr)
2909 {
2910         struct ext4_sb_info *sbi;
2911
2912         if (!elr)
2913                 return;
2914
2915         sbi = elr->lr_sbi;
2916
2917         list_del(&elr->lr_request);
2918         sbi->s_li_request = NULL;
2919         kfree(elr);
2920 }
2921
2922 static void ext4_unregister_li_request(struct super_block *sb)
2923 {
2924         mutex_lock(&ext4_li_mtx);
2925         if (!ext4_li_info) {
2926                 mutex_unlock(&ext4_li_mtx);
2927                 return;
2928         }
2929
2930         mutex_lock(&ext4_li_info->li_list_mtx);
2931         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2932         mutex_unlock(&ext4_li_info->li_list_mtx);
2933         mutex_unlock(&ext4_li_mtx);
2934 }
2935
2936 static struct task_struct *ext4_lazyinit_task;
2937
2938 /*
2939  * This is the function where ext4lazyinit thread lives. It walks
2940  * through the request list searching for next scheduled filesystem.
2941  * When such a fs is found, run the lazy initialization request
2942  * (ext4_rn_li_request) and keep track of the time spend in this
2943  * function. Based on that time we compute next schedule time of
2944  * the request. When walking through the list is complete, compute
2945  * next waking time and put itself into sleep.
2946  */
2947 static int ext4_lazyinit_thread(void *arg)
2948 {
2949         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2950         struct list_head *pos, *n;
2951         struct ext4_li_request *elr;
2952         unsigned long next_wakeup, cur;
2953
2954         BUG_ON(NULL == eli);
2955
2956 cont_thread:
2957         while (true) {
2958                 next_wakeup = MAX_JIFFY_OFFSET;
2959
2960                 mutex_lock(&eli->li_list_mtx);
2961                 if (list_empty(&eli->li_request_list)) {
2962                         mutex_unlock(&eli->li_list_mtx);
2963                         goto exit_thread;
2964                 }
2965
2966                 list_for_each_safe(pos, n, &eli->li_request_list) {
2967                         elr = list_entry(pos, struct ext4_li_request,
2968                                          lr_request);
2969
2970                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2971                                 if (ext4_run_li_request(elr) != 0) {
2972                                         /* error, remove the lazy_init job */
2973                                         ext4_remove_li_request(elr);
2974                                         continue;
2975                                 }
2976                         }
2977
2978                         if (time_before(elr->lr_next_sched, next_wakeup))
2979                                 next_wakeup = elr->lr_next_sched;
2980                 }
2981                 mutex_unlock(&eli->li_list_mtx);
2982
2983                 try_to_freeze();
2984
2985                 cur = jiffies;
2986                 if ((time_after_eq(cur, next_wakeup)) ||
2987                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2988                         cond_resched();
2989                         continue;
2990                 }
2991
2992                 schedule_timeout_interruptible(next_wakeup - cur);
2993
2994                 if (kthread_should_stop()) {
2995                         ext4_clear_request_list();
2996                         goto exit_thread;
2997                 }
2998         }
2999
3000 exit_thread:
3001         /*
3002          * It looks like the request list is empty, but we need
3003          * to check it under the li_list_mtx lock, to prevent any
3004          * additions into it, and of course we should lock ext4_li_mtx
3005          * to atomically free the list and ext4_li_info, because at
3006          * this point another ext4 filesystem could be registering
3007          * new one.
3008          */
3009         mutex_lock(&ext4_li_mtx);
3010         mutex_lock(&eli->li_list_mtx);
3011         if (!list_empty(&eli->li_request_list)) {
3012                 mutex_unlock(&eli->li_list_mtx);
3013                 mutex_unlock(&ext4_li_mtx);
3014                 goto cont_thread;
3015         }
3016         mutex_unlock(&eli->li_list_mtx);
3017         kfree(ext4_li_info);
3018         ext4_li_info = NULL;
3019         mutex_unlock(&ext4_li_mtx);
3020
3021         return 0;
3022 }
3023
3024 static void ext4_clear_request_list(void)
3025 {
3026         struct list_head *pos, *n;
3027         struct ext4_li_request *elr;
3028
3029         mutex_lock(&ext4_li_info->li_list_mtx);
3030         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3031                 elr = list_entry(pos, struct ext4_li_request,
3032                                  lr_request);
3033                 ext4_remove_li_request(elr);
3034         }
3035         mutex_unlock(&ext4_li_info->li_list_mtx);
3036 }
3037
3038 static int ext4_run_lazyinit_thread(void)
3039 {
3040         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3041                                          ext4_li_info, "ext4lazyinit");
3042         if (IS_ERR(ext4_lazyinit_task)) {
3043                 int err = PTR_ERR(ext4_lazyinit_task);
3044                 ext4_clear_request_list();
3045                 kfree(ext4_li_info);
3046                 ext4_li_info = NULL;
3047                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3048                                  "initialization thread\n",
3049                                  err);
3050                 return err;
3051         }
3052         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3053         return 0;
3054 }
3055
3056 /*
3057  * Check whether it make sense to run itable init. thread or not.
3058  * If there is at least one uninitialized inode table, return
3059  * corresponding group number, else the loop goes through all
3060  * groups and return total number of groups.
3061  */
3062 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3063 {
3064         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3065         struct ext4_group_desc *gdp = NULL;
3066
3067         for (group = 0; group < ngroups; group++) {
3068                 gdp = ext4_get_group_desc(sb, group, NULL);
3069                 if (!gdp)
3070                         continue;
3071
3072                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3073                         break;
3074         }
3075
3076         return group;
3077 }
3078
3079 static int ext4_li_info_new(void)
3080 {
3081         struct ext4_lazy_init *eli = NULL;
3082
3083         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3084         if (!eli)
3085                 return -ENOMEM;
3086
3087         INIT_LIST_HEAD(&eli->li_request_list);
3088         mutex_init(&eli->li_list_mtx);
3089
3090         eli->li_state |= EXT4_LAZYINIT_QUIT;
3091
3092         ext4_li_info = eli;
3093
3094         return 0;
3095 }
3096
3097 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3098                                             ext4_group_t start)
3099 {
3100         struct ext4_sb_info *sbi = EXT4_SB(sb);
3101         struct ext4_li_request *elr;
3102
3103         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3104         if (!elr)
3105                 return NULL;
3106
3107         elr->lr_super = sb;
3108         elr->lr_sbi = sbi;
3109         elr->lr_next_group = start;
3110
3111         /*
3112          * Randomize first schedule time of the request to
3113          * spread the inode table initialization requests
3114          * better.
3115          */
3116         elr->lr_next_sched = jiffies + (prandom_u32() %
3117                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3118         return elr;
3119 }
3120
3121 int ext4_register_li_request(struct super_block *sb,
3122                              ext4_group_t first_not_zeroed)
3123 {
3124         struct ext4_sb_info *sbi = EXT4_SB(sb);
3125         struct ext4_li_request *elr = NULL;
3126         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3127         int ret = 0;
3128
3129         mutex_lock(&ext4_li_mtx);
3130         if (sbi->s_li_request != NULL) {
3131                 /*
3132                  * Reset timeout so it can be computed again, because
3133                  * s_li_wait_mult might have changed.
3134                  */
3135                 sbi->s_li_request->lr_timeout = 0;
3136                 goto out;
3137         }
3138
3139         if (first_not_zeroed == ngroups ||
3140             (sb->s_flags & MS_RDONLY) ||
3141             !test_opt(sb, INIT_INODE_TABLE))
3142                 goto out;
3143
3144         elr = ext4_li_request_new(sb, first_not_zeroed);
3145         if (!elr) {
3146                 ret = -ENOMEM;
3147                 goto out;
3148         }
3149
3150         if (NULL == ext4_li_info) {
3151                 ret = ext4_li_info_new();
3152                 if (ret)
3153                         goto out;
3154         }
3155
3156         mutex_lock(&ext4_li_info->li_list_mtx);
3157         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3158         mutex_unlock(&ext4_li_info->li_list_mtx);
3159
3160         sbi->s_li_request = elr;
3161         /*
3162          * set elr to NULL here since it has been inserted to
3163          * the request_list and the removal and free of it is
3164          * handled by ext4_clear_request_list from now on.
3165          */
3166         elr = NULL;
3167
3168         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3169                 ret = ext4_run_lazyinit_thread();
3170                 if (ret)
3171                         goto out;
3172         }
3173 out:
3174         mutex_unlock(&ext4_li_mtx);
3175         if (ret)
3176                 kfree(elr);
3177         return ret;
3178 }
3179
3180 /*
3181  * We do not need to lock anything since this is called on
3182  * module unload.
3183  */
3184 static void ext4_destroy_lazyinit_thread(void)
3185 {
3186         /*
3187          * If thread exited earlier
3188          * there's nothing to be done.
3189          */
3190         if (!ext4_li_info || !ext4_lazyinit_task)
3191                 return;
3192
3193         kthread_stop(ext4_lazyinit_task);
3194 }
3195
3196 static int set_journal_csum_feature_set(struct super_block *sb)
3197 {
3198         int ret = 1;
3199         int compat, incompat;
3200         struct ext4_sb_info *sbi = EXT4_SB(sb);
3201
3202         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3203                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3204                 /* journal checksum v3 */
3205                 compat = 0;
3206                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3207         } else {
3208                 /* journal checksum v1 */
3209                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3210                 incompat = 0;
3211         }
3212
3213         jbd2_journal_clear_features(sbi->s_journal,
3214                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3215                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3216                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3217         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3218                 ret = jbd2_journal_set_features(sbi->s_journal,
3219                                 compat, 0,
3220                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3221                                 incompat);
3222         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3223                 ret = jbd2_journal_set_features(sbi->s_journal,
3224                                 compat, 0,
3225                                 incompat);
3226                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3227                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3228         } else {
3229                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3230                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3231         }
3232
3233         return ret;
3234 }
3235
3236 /*
3237  * Note: calculating the overhead so we can be compatible with
3238  * historical BSD practice is quite difficult in the face of
3239  * clusters/bigalloc.  This is because multiple metadata blocks from
3240  * different block group can end up in the same allocation cluster.
3241  * Calculating the exact overhead in the face of clustered allocation
3242  * requires either O(all block bitmaps) in memory or O(number of block
3243  * groups**2) in time.  We will still calculate the superblock for
3244  * older file systems --- and if we come across with a bigalloc file
3245  * system with zero in s_overhead_clusters the estimate will be close to
3246  * correct especially for very large cluster sizes --- but for newer
3247  * file systems, it's better to calculate this figure once at mkfs
3248  * time, and store it in the superblock.  If the superblock value is
3249  * present (even for non-bigalloc file systems), we will use it.
3250  */
3251 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3252                           char *buf)
3253 {
3254         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3255         struct ext4_group_desc  *gdp;
3256         ext4_fsblk_t            first_block, last_block, b;
3257         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3258         int                     s, j, count = 0;
3259
3260         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3261                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3262                         sbi->s_itb_per_group + 2);
3263
3264         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3265                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3266         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3267         for (i = 0; i < ngroups; i++) {
3268                 gdp = ext4_get_group_desc(sb, i, NULL);
3269                 b = ext4_block_bitmap(sb, gdp);
3270                 if (b >= first_block && b <= last_block) {
3271                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3272                         count++;
3273                 }
3274                 b = ext4_inode_bitmap(sb, gdp);
3275                 if (b >= first_block && b <= last_block) {
3276                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3277                         count++;
3278                 }
3279                 b = ext4_inode_table(sb, gdp);
3280                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3281                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3282                                 int c = EXT4_B2C(sbi, b - first_block);
3283                                 ext4_set_bit(c, buf);
3284                                 count++;
3285                         }
3286                 if (i != grp)
3287                         continue;
3288                 s = 0;
3289                 if (ext4_bg_has_super(sb, grp)) {
3290                         ext4_set_bit(s++, buf);
3291                         count++;
3292                 }
3293                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3294                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3295                         count++;
3296                 }
3297         }
3298         if (!count)
3299                 return 0;
3300         return EXT4_CLUSTERS_PER_GROUP(sb) -
3301                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3302 }
3303
3304 /*
3305  * Compute the overhead and stash it in sbi->s_overhead
3306  */
3307 int ext4_calculate_overhead(struct super_block *sb)
3308 {
3309         struct ext4_sb_info *sbi = EXT4_SB(sb);
3310         struct ext4_super_block *es = sbi->s_es;
3311         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3312         ext4_fsblk_t overhead = 0;
3313         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3314
3315         if (!buf)
3316                 return -ENOMEM;
3317
3318         /*
3319          * Compute the overhead (FS structures).  This is constant
3320          * for a given filesystem unless the number of block groups
3321          * changes so we cache the previous value until it does.
3322          */
3323
3324         /*
3325          * All of the blocks before first_data_block are overhead
3326          */
3327         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3328
3329         /*
3330          * Add the overhead found in each block group
3331          */
3332         for (i = 0; i < ngroups; i++) {
3333                 int blks;
3334
3335                 blks = count_overhead(sb, i, buf);
3336                 overhead += blks;
3337                 if (blks)
3338                         memset(buf, 0, PAGE_SIZE);
3339                 cond_resched();
3340         }
3341         /* Add the journal blocks as well */
3342         if (sbi->s_journal)
3343                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3344
3345         sbi->s_overhead = overhead;
3346         smp_wmb();
3347         free_page((unsigned long) buf);
3348         return 0;
3349 }
3350
3351
3352 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3353 {
3354         ext4_fsblk_t resv_clusters;
3355
3356         /*
3357          * There's no need to reserve anything when we aren't using extents.
3358          * The space estimates are exact, there are no unwritten extents,
3359          * hole punching doesn't need new metadata... This is needed especially
3360          * to keep ext2/3 backward compatibility.
3361          */
3362         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3363                 return 0;
3364         /*
3365          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3366          * This should cover the situations where we can not afford to run
3367          * out of space like for example punch hole, or converting
3368          * unwritten extents in delalloc path. In most cases such
3369          * allocation would require 1, or 2 blocks, higher numbers are
3370          * very rare.
3371          */
3372         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3373                         EXT4_SB(sb)->s_cluster_bits;
3374
3375         do_div(resv_clusters, 50);
3376         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3377
3378         return resv_clusters;
3379 }
3380
3381
3382 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3383 {
3384         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3385                                 sbi->s_cluster_bits;
3386
3387         if (count >= clusters)
3388                 return -EINVAL;
3389
3390         atomic64_set(&sbi->s_resv_clusters, count);
3391         return 0;
3392 }
3393
3394 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3395 {
3396         char *orig_data = kstrdup(data, GFP_KERNEL);
3397         struct buffer_head *bh;
3398         struct ext4_super_block *es = NULL;
3399         struct ext4_sb_info *sbi;
3400         ext4_fsblk_t block;
3401         ext4_fsblk_t sb_block = get_sb_block(&data);
3402         ext4_fsblk_t logical_sb_block;
3403         unsigned long offset = 0;
3404         unsigned long journal_devnum = 0;
3405         unsigned long def_mount_opts;
3406         struct inode *root;
3407         char *cp;
3408         const char *descr;
3409         int ret = -ENOMEM;
3410         int blocksize, clustersize;
3411         unsigned int db_count;
3412         unsigned int i;
3413         int needs_recovery, has_huge_files, has_bigalloc;
3414         __u64 blocks_count;
3415         int err = 0;
3416         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3417         ext4_group_t first_not_zeroed;
3418
3419         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3420         if (!sbi)
3421                 goto out_free_orig;
3422
3423         sbi->s_blockgroup_lock =
3424                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3425         if (!sbi->s_blockgroup_lock) {
3426                 kfree(sbi);
3427                 goto out_free_orig;
3428         }
3429         sb->s_fs_info = sbi;
3430         sbi->s_sb = sb;
3431         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3432         sbi->s_sb_block = sb_block;
3433         if (sb->s_bdev->bd_part)
3434                 sbi->s_sectors_written_start =
3435                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3436
3437         /* Cleanup superblock name */
3438         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3439                 *cp = '!';
3440
3441         /* -EINVAL is default */
3442         ret = -EINVAL;
3443         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3444         if (!blocksize) {
3445                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3446                 goto out_fail;
3447         }
3448
3449         /*
3450          * The ext4 superblock will not be buffer aligned for other than 1kB
3451          * block sizes.  We need to calculate the offset from buffer start.
3452          */
3453         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3454                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3455                 offset = do_div(logical_sb_block, blocksize);
3456         } else {
3457                 logical_sb_block = sb_block;
3458         }
3459
3460         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3461                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3462                 goto out_fail;
3463         }
3464         /*
3465          * Note: s_es must be initialized as soon as possible because
3466          *       some ext4 macro-instructions depend on its value
3467          */
3468         es = (struct ext4_super_block *) (bh->b_data + offset);
3469         sbi->s_es = es;
3470         sb->s_magic = le16_to_cpu(es->s_magic);
3471         if (sb->s_magic != EXT4_SUPER_MAGIC)
3472                 goto cantfind_ext4;
3473         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3474
3475         /* Warn if metadata_csum and gdt_csum are both set. */
3476         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3477                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3478             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3479                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3480                              "redundant flags; please run fsck.");
3481
3482         /* Check for a known checksum algorithm */
3483         if (!ext4_verify_csum_type(sb, es)) {
3484                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3485                          "unknown checksum algorithm.");
3486                 silent = 1;
3487                 goto cantfind_ext4;
3488         }
3489
3490         /* Load the checksum driver */
3491         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3492                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3493                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3494                 if (IS_ERR(sbi->s_chksum_driver)) {
3495                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3496                         ret = PTR_ERR(sbi->s_chksum_driver);
3497                         sbi->s_chksum_driver = NULL;
3498                         goto failed_mount;
3499                 }
3500         }
3501
3502         /* Check superblock checksum */
3503         if (!ext4_superblock_csum_verify(sb, es)) {
3504                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3505                          "invalid superblock checksum.  Run e2fsck?");
3506                 silent = 1;
3507                 goto cantfind_ext4;
3508         }
3509
3510         /* Precompute checksum seed for all metadata */
3511         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3512                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3513                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3514                                                sizeof(es->s_uuid));
3515
3516         /* Set defaults before we parse the mount options */
3517         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3518         set_opt(sb, INIT_INODE_TABLE);
3519         if (def_mount_opts & EXT4_DEFM_DEBUG)
3520                 set_opt(sb, DEBUG);
3521         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3522                 set_opt(sb, GRPID);
3523         if (def_mount_opts & EXT4_DEFM_UID16)
3524                 set_opt(sb, NO_UID32);
3525         /* xattr user namespace & acls are now defaulted on */
3526         set_opt(sb, XATTR_USER);
3527 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3528         set_opt(sb, POSIX_ACL);
3529 #endif
3530         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3531                 set_opt(sb, JOURNAL_DATA);
3532         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3533                 set_opt(sb, ORDERED_DATA);
3534         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3535                 set_opt(sb, WRITEBACK_DATA);
3536
3537         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3538                 set_opt(sb, ERRORS_PANIC);
3539         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3540                 set_opt(sb, ERRORS_CONT);
3541         else
3542                 set_opt(sb, ERRORS_RO);
3543         /* block_validity enabled by default; disable with noblock_validity */
3544         set_opt(sb, BLOCK_VALIDITY);
3545         if (def_mount_opts & EXT4_DEFM_DISCARD)
3546                 set_opt(sb, DISCARD);
3547
3548         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3549         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3550         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3551         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3552         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3553
3554         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3555                 set_opt(sb, BARRIER);
3556
3557         /*
3558          * enable delayed allocation by default
3559          * Use -o nodelalloc to turn it off
3560          */
3561         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3562             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3563                 set_opt(sb, DELALLOC);
3564
3565         /*
3566          * set default s_li_wait_mult for lazyinit, for the case there is
3567          * no mount option specified.
3568          */
3569         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3570
3571         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3572                            &journal_devnum, &journal_ioprio, 0)) {
3573                 ext4_msg(sb, KERN_WARNING,
3574                          "failed to parse options in superblock: %s",
3575                          sbi->s_es->s_mount_opts);
3576         }
3577         sbi->s_def_mount_opt = sbi->s_mount_opt;
3578         if (!parse_options((char *) data, sb, &journal_devnum,
3579                            &journal_ioprio, 0))
3580                 goto failed_mount;
3581
3582         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3583                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3584                             "with data=journal disables delayed "
3585                             "allocation and O_DIRECT support!\n");
3586                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3587                         ext4_msg(sb, KERN_ERR, "can't mount with "
3588                                  "both data=journal and delalloc");
3589                         goto failed_mount;
3590                 }
3591                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3592                         ext4_msg(sb, KERN_ERR, "can't mount with "
3593                                  "both data=journal and dioread_nolock");
3594                         goto failed_mount;
3595                 }
3596                 if (test_opt(sb, DELALLOC))
3597                         clear_opt(sb, DELALLOC);
3598         }
3599
3600         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3601                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3602
3603         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3604             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3605              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3606              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3607                 ext4_msg(sb, KERN_WARNING,
3608                        "feature flags set on rev 0 fs, "
3609                        "running e2fsck is recommended");
3610
3611         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3612                 set_opt2(sb, HURD_COMPAT);
3613                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3614                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3615                         ext4_msg(sb, KERN_ERR,
3616                                  "The Hurd can't support 64-bit file systems");
3617                         goto failed_mount;
3618                 }
3619         }
3620
3621         if (IS_EXT2_SB(sb)) {
3622                 if (ext2_feature_set_ok(sb))
3623                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3624                                  "using the ext4 subsystem");
3625                 else {
3626                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3627                                  "to feature incompatibilities");
3628                         goto failed_mount;
3629                 }
3630         }
3631
3632         if (IS_EXT3_SB(sb)) {
3633                 if (ext3_feature_set_ok(sb))
3634                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3635                                  "using the ext4 subsystem");
3636                 else {
3637                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3638                                  "to feature incompatibilities");
3639                         goto failed_mount;
3640                 }
3641         }
3642
3643         /*
3644          * Check feature flags regardless of the revision level, since we
3645          * previously didn't change the revision level when setting the flags,
3646          * so there is a chance incompat flags are set on a rev 0 filesystem.
3647          */
3648         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3649                 goto failed_mount;
3650
3651         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3652         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3653             blocksize > EXT4_MAX_BLOCK_SIZE) {
3654                 ext4_msg(sb, KERN_ERR,
3655                        "Unsupported filesystem blocksize %d", blocksize);
3656                 goto failed_mount;
3657         }
3658
3659         if (sb->s_blocksize != blocksize) {
3660                 /* Validate the filesystem blocksize */
3661                 if (!sb_set_blocksize(sb, blocksize)) {
3662                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3663                                         blocksize);
3664                         goto failed_mount;
3665                 }
3666
3667                 brelse(bh);
3668                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3669                 offset = do_div(logical_sb_block, blocksize);
3670                 bh = sb_bread_unmovable(sb, logical_sb_block);
3671                 if (!bh) {
3672                         ext4_msg(sb, KERN_ERR,
3673                                "Can't read superblock on 2nd try");
3674                         goto failed_mount;
3675                 }
3676                 es = (struct ext4_super_block *)(bh->b_data + offset);
3677                 sbi->s_es = es;
3678                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3679                         ext4_msg(sb, KERN_ERR,
3680                                "Magic mismatch, very weird!");
3681                         goto failed_mount;
3682                 }
3683         }
3684
3685         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3686                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3687         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3688                                                       has_huge_files);
3689         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3690
3691         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3692                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3693                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3694         } else {
3695                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3696                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3697                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3698                     (!is_power_of_2(sbi->s_inode_size)) ||
3699                     (sbi->s_inode_size > blocksize)) {
3700                         ext4_msg(sb, KERN_ERR,
3701                                "unsupported inode size: %d",
3702                                sbi->s_inode_size);
3703                         goto failed_mount;
3704                 }
3705                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3706                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3707         }
3708
3709         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3710         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3711                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3712                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3713                     !is_power_of_2(sbi->s_desc_size)) {
3714                         ext4_msg(sb, KERN_ERR,
3715                                "unsupported descriptor size %lu",
3716                                sbi->s_desc_size);
3717                         goto failed_mount;
3718                 }
3719         } else
3720                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3721
3722         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3723         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3724         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3725                 goto cantfind_ext4;
3726
3727         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3728         if (sbi->s_inodes_per_block == 0)
3729                 goto cantfind_ext4;
3730         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3731                                         sbi->s_inodes_per_block;
3732         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3733         sbi->s_sbh = bh;
3734         sbi->s_mount_state = le16_to_cpu(es->s_state);
3735         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3736         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3737
3738         for (i = 0; i < 4; i++)
3739                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3740         sbi->s_def_hash_version = es->s_def_hash_version;
3741         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3742                 i = le32_to_cpu(es->s_flags);
3743                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3744                         sbi->s_hash_unsigned = 3;
3745                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3746 #ifdef __CHAR_UNSIGNED__
3747                         if (!(sb->s_flags & MS_RDONLY))
3748                                 es->s_flags |=
3749                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3750                         sbi->s_hash_unsigned = 3;
3751 #else
3752                         if (!(sb->s_flags & MS_RDONLY))
3753                                 es->s_flags |=
3754                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3755 #endif
3756                 }
3757         }
3758
3759         /* Handle clustersize */
3760         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3761         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3762                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3763         if (has_bigalloc) {
3764                 if (clustersize < blocksize) {
3765                         ext4_msg(sb, KERN_ERR,
3766                                  "cluster size (%d) smaller than "
3767                                  "block size (%d)", clustersize, blocksize);
3768                         goto failed_mount;
3769                 }
3770                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3771                         le32_to_cpu(es->s_log_block_size);
3772                 sbi->s_clusters_per_group =
3773                         le32_to_cpu(es->s_clusters_per_group);
3774                 if (sbi->s_clusters_per_group > blocksize * 8) {
3775                         ext4_msg(sb, KERN_ERR,
3776                                  "#clusters per group too big: %lu",
3777                                  sbi->s_clusters_per_group);
3778                         goto failed_mount;
3779                 }
3780                 if (sbi->s_blocks_per_group !=
3781                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3782                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3783                                  "clusters per group (%lu) inconsistent",
3784                                  sbi->s_blocks_per_group,
3785                                  sbi->s_clusters_per_group);
3786                         goto failed_mount;
3787                 }
3788         } else {
3789                 if (clustersize != blocksize) {
3790                         ext4_warning(sb, "fragment/cluster size (%d) != "
3791                                      "block size (%d)", clustersize,
3792                                      blocksize);
3793                         clustersize = blocksize;
3794                 }
3795                 if (sbi->s_blocks_per_group > blocksize * 8) {
3796                         ext4_msg(sb, KERN_ERR,
3797                                  "#blocks per group too big: %lu",
3798                                  sbi->s_blocks_per_group);
3799                         goto failed_mount;
3800                 }
3801                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3802                 sbi->s_cluster_bits = 0;
3803         }
3804         sbi->s_cluster_ratio = clustersize / blocksize;
3805
3806         if (sbi->s_inodes_per_group > blocksize * 8) {
3807                 ext4_msg(sb, KERN_ERR,
3808                        "#inodes per group too big: %lu",
3809                        sbi->s_inodes_per_group);
3810                 goto failed_mount;
3811         }
3812
3813         /* Do we have standard group size of clustersize * 8 blocks ? */
3814         if (sbi->s_blocks_per_group == clustersize << 3)
3815                 set_opt2(sb, STD_GROUP_SIZE);
3816
3817         /*
3818          * Test whether we have more sectors than will fit in sector_t,
3819          * and whether the max offset is addressable by the page cache.
3820          */
3821         err = generic_check_addressable(sb->s_blocksize_bits,
3822                                         ext4_blocks_count(es));
3823         if (err) {
3824                 ext4_msg(sb, KERN_ERR, "filesystem"
3825                          " too large to mount safely on this system");
3826                 if (sizeof(sector_t) < 8)
3827                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3828                 goto failed_mount;
3829         }
3830
3831         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3832                 goto cantfind_ext4;
3833
3834         /* check blocks count against device size */
3835         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3836         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3837                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3838                        "exceeds size of device (%llu blocks)",
3839                        ext4_blocks_count(es), blocks_count);
3840                 goto failed_mount;
3841         }
3842
3843         /*
3844          * It makes no sense for the first data block to be beyond the end
3845          * of the filesystem.
3846          */
3847         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3848                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3849                          "block %u is beyond end of filesystem (%llu)",
3850                          le32_to_cpu(es->s_first_data_block),
3851                          ext4_blocks_count(es));
3852                 goto failed_mount;
3853         }
3854         blocks_count = (ext4_blocks_count(es) -
3855                         le32_to_cpu(es->s_first_data_block) +
3856                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3857         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3858         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3859                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3860                        "(block count %llu, first data block %u, "
3861                        "blocks per group %lu)", sbi->s_groups_count,
3862                        ext4_blocks_count(es),
3863                        le32_to_cpu(es->s_first_data_block),
3864                        EXT4_BLOCKS_PER_GROUP(sb));
3865                 goto failed_mount;
3866         }
3867         sbi->s_groups_count = blocks_count;
3868         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3869                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3870         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3871                    EXT4_DESC_PER_BLOCK(sb);
3872         sbi->s_group_desc = ext4_kvmalloc(db_count *
3873                                           sizeof(struct buffer_head *),
3874                                           GFP_KERNEL);
3875         if (sbi->s_group_desc == NULL) {
3876                 ext4_msg(sb, KERN_ERR, "not enough memory");
3877                 ret = -ENOMEM;
3878                 goto failed_mount;
3879         }
3880
3881         if (ext4_proc_root)
3882                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3883
3884         if (sbi->s_proc)
3885                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3886                                  &ext4_seq_options_fops, sb);
3887
3888         bgl_lock_init(sbi->s_blockgroup_lock);
3889
3890         for (i = 0; i < db_count; i++) {
3891                 block = descriptor_loc(sb, logical_sb_block, i);
3892                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3893                 if (!sbi->s_group_desc[i]) {
3894                         ext4_msg(sb, KERN_ERR,
3895                                "can't read group descriptor %d", i);
3896                         db_count = i;
3897                         goto failed_mount2;
3898                 }
3899         }
3900         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3901                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3902                 goto failed_mount2;
3903         }
3904
3905         sbi->s_gdb_count = db_count;
3906         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3907         spin_lock_init(&sbi->s_next_gen_lock);
3908
3909         init_timer(&sbi->s_err_report);
3910         sbi->s_err_report.function = print_daily_error_info;
3911         sbi->s_err_report.data = (unsigned long) sb;
3912
3913         /* Register extent status tree shrinker */
3914         if (ext4_es_register_shrinker(sbi))
3915                 goto failed_mount3;
3916
3917         sbi->s_stripe = ext4_get_stripe_size(sbi);
3918         sbi->s_extent_max_zeroout_kb = 32;
3919
3920         /*
3921          * set up enough so that it can read an inode
3922          */
3923         sb->s_op = &ext4_sops;
3924         sb->s_export_op = &ext4_export_ops;
3925         sb->s_xattr = ext4_xattr_handlers;
3926 #ifdef CONFIG_QUOTA
3927         sb->dq_op = &ext4_quota_operations;
3928         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3929                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3930         else
3931                 sb->s_qcop = &ext4_qctl_operations;
3932 #endif
3933         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3934
3935         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3936         mutex_init(&sbi->s_orphan_lock);
3937
3938         sb->s_root = NULL;
3939
3940         needs_recovery = (es->s_last_orphan != 0 ||
3941                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3942                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3943
3944         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3945             !(sb->s_flags & MS_RDONLY))
3946                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3947                         goto failed_mount3;
3948
3949         /*
3950          * The first inode we look at is the journal inode.  Don't try
3951          * root first: it may be modified in the journal!
3952          */
3953         if (!test_opt(sb, NOLOAD) &&
3954             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3955                 if (ext4_load_journal(sb, es, journal_devnum))
3956                         goto failed_mount3;
3957         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3958               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3959                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3960                        "suppressed and not mounted read-only");
3961                 goto failed_mount_wq;
3962         } else {
3963                 clear_opt(sb, DATA_FLAGS);
3964                 sbi->s_journal = NULL;
3965                 needs_recovery = 0;
3966                 goto no_journal;
3967         }
3968
3969         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3970             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3971                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3972                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3973                 goto failed_mount_wq;
3974         }
3975
3976         if (!set_journal_csum_feature_set(sb)) {
3977                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3978                          "feature set");
3979                 goto failed_mount_wq;
3980         }
3981
3982         /* We have now updated the journal if required, so we can
3983          * validate the data journaling mode. */
3984         switch (test_opt(sb, DATA_FLAGS)) {
3985         case 0:
3986                 /* No mode set, assume a default based on the journal
3987                  * capabilities: ORDERED_DATA if the journal can
3988                  * cope, else JOURNAL_DATA
3989                  */
3990                 if (jbd2_journal_check_available_features
3991                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3992                         set_opt(sb, ORDERED_DATA);
3993                 else
3994                         set_opt(sb, JOURNAL_DATA);
3995                 break;
3996
3997         case EXT4_MOUNT_ORDERED_DATA:
3998         case EXT4_MOUNT_WRITEBACK_DATA:
3999                 if (!jbd2_journal_check_available_features
4000                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4001                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4002                                "requested data journaling mode");
4003                         goto failed_mount_wq;
4004                 }
4005         default:
4006                 break;
4007         }
4008         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4009
4010         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4011
4012 no_journal:
4013         if (ext4_mballoc_ready) {
4014                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4015                 if (!sbi->s_mb_cache) {
4016                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4017                         goto failed_mount_wq;
4018                 }
4019         }
4020
4021         /*
4022          * Get the # of file system overhead blocks from the
4023          * superblock if present.
4024          */
4025         if (es->s_overhead_clusters)
4026                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4027         else {
4028                 err = ext4_calculate_overhead(sb);
4029                 if (err)
4030                         goto failed_mount_wq;
4031         }
4032
4033         /*
4034          * The maximum number of concurrent works can be high and
4035          * concurrency isn't really necessary.  Limit it to 1.
4036          */
4037         EXT4_SB(sb)->rsv_conversion_wq =
4038                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4039         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4040                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4041                 ret = -ENOMEM;
4042                 goto failed_mount4;
4043         }
4044
4045         /*
4046          * The jbd2_journal_load will have done any necessary log recovery,
4047          * so we can safely mount the rest of the filesystem now.
4048          */
4049
4050         root = ext4_iget(sb, EXT4_ROOT_INO);
4051         if (IS_ERR(root)) {
4052                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4053                 ret = PTR_ERR(root);
4054                 root = NULL;
4055                 goto failed_mount4;
4056         }
4057         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4058                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4059                 iput(root);
4060                 goto failed_mount4;
4061         }
4062         sb->s_root = d_make_root(root);
4063         if (!sb->s_root) {
4064                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4065                 ret = -ENOMEM;
4066                 goto failed_mount4;
4067         }
4068
4069         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4070                 sb->s_flags |= MS_RDONLY;
4071
4072         /* determine the minimum size of new large inodes, if present */
4073         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4074                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4075                                                      EXT4_GOOD_OLD_INODE_SIZE;
4076                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4077                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4078                         if (sbi->s_want_extra_isize <
4079                             le16_to_cpu(es->s_want_extra_isize))
4080                                 sbi->s_want_extra_isize =
4081                                         le16_to_cpu(es->s_want_extra_isize);
4082                         if (sbi->s_want_extra_isize <
4083                             le16_to_cpu(es->s_min_extra_isize))
4084                                 sbi->s_want_extra_isize =
4085                                         le16_to_cpu(es->s_min_extra_isize);
4086                 }
4087         }
4088         /* Check if enough inode space is available */
4089         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4090                                                         sbi->s_inode_size) {
4091                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4092                                                        EXT4_GOOD_OLD_INODE_SIZE;
4093                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4094                          "available");
4095         }
4096
4097         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4098         if (err) {
4099                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4100                          "reserved pool", ext4_calculate_resv_clusters(sb));
4101                 goto failed_mount4a;
4102         }
4103
4104         err = ext4_setup_system_zone(sb);
4105         if (err) {
4106                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4107                          "zone (%d)", err);
4108                 goto failed_mount4a;
4109         }
4110
4111         ext4_ext_init(sb);
4112         err = ext4_mb_init(sb);
4113         if (err) {
4114                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4115                          err);
4116                 goto failed_mount5;
4117         }
4118
4119         block = ext4_count_free_clusters(sb);
4120         ext4_free_blocks_count_set(sbi->s_es,
4121                                    EXT4_C2B(sbi, block));
4122         err = percpu_counter_init(&sbi->s_freeclusters_counter, block);
4123         if (!err) {
4124                 unsigned long freei = ext4_count_free_inodes(sb);
4125                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4126                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei);
4127         }
4128         if (!err)
4129                 err = percpu_counter_init(&sbi->s_dirs_counter,
4130                                           ext4_count_dirs(sb));
4131         if (!err)
4132                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
4133         if (err) {
4134                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4135                 goto failed_mount6;
4136         }
4137
4138         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4139                 if (!ext4_fill_flex_info(sb)) {
4140                         ext4_msg(sb, KERN_ERR,
4141                                "unable to initialize "
4142                                "flex_bg meta info!");
4143                         goto failed_mount6;
4144                 }
4145
4146         err = ext4_register_li_request(sb, first_not_zeroed);
4147         if (err)
4148                 goto failed_mount6;
4149
4150         sbi->s_kobj.kset = ext4_kset;
4151         init_completion(&sbi->s_kobj_unregister);
4152         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4153                                    "%s", sb->s_id);
4154         if (err)
4155                 goto failed_mount7;
4156
4157 #ifdef CONFIG_QUOTA
4158         /* Enable quota usage during mount. */
4159         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4160             !(sb->s_flags & MS_RDONLY)) {
4161                 err = ext4_enable_quotas(sb);
4162                 if (err)
4163                         goto failed_mount8;
4164         }
4165 #endif  /* CONFIG_QUOTA */
4166
4167         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4168         ext4_orphan_cleanup(sb, es);
4169         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4170         if (needs_recovery) {
4171                 ext4_msg(sb, KERN_INFO, "recovery complete");
4172                 ext4_mark_recovery_complete(sb, es);
4173         }
4174         if (EXT4_SB(sb)->s_journal) {
4175                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4176                         descr = " journalled data mode";
4177                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4178                         descr = " ordered data mode";
4179                 else
4180                         descr = " writeback data mode";
4181         } else
4182                 descr = "out journal";
4183
4184         if (test_opt(sb, DISCARD)) {
4185                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4186                 if (!blk_queue_discard(q))
4187                         ext4_msg(sb, KERN_WARNING,
4188                                  "mounting with \"discard\" option, but "
4189                                  "the device does not support discard");
4190         }
4191
4192         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4193                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4194                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4195
4196         if (es->s_error_count)
4197                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4198
4199         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4200         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4201         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4202         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4203
4204         kfree(orig_data);
4205         return 0;
4206
4207 cantfind_ext4:
4208         if (!silent)
4209                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4210         goto failed_mount;
4211
4212 #ifdef CONFIG_QUOTA
4213 failed_mount8:
4214         kobject_del(&sbi->s_kobj);
4215 #endif
4216 failed_mount7:
4217         ext4_unregister_li_request(sb);
4218 failed_mount6:
4219         ext4_mb_release(sb);
4220         if (sbi->s_flex_groups)
4221                 ext4_kvfree(sbi->s_flex_groups);
4222         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4223         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4224         percpu_counter_destroy(&sbi->s_dirs_counter);
4225         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4226 failed_mount5:
4227         ext4_ext_release(sb);
4228         ext4_release_system_zone(sb);
4229 failed_mount4a:
4230         dput(sb->s_root);
4231         sb->s_root = NULL;
4232 failed_mount4:
4233         ext4_msg(sb, KERN_ERR, "mount failed");
4234         if (EXT4_SB(sb)->rsv_conversion_wq)
4235                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4236 failed_mount_wq:
4237         if (sbi->s_journal) {
4238                 jbd2_journal_destroy(sbi->s_journal);
4239                 sbi->s_journal = NULL;
4240         }
4241         ext4_es_unregister_shrinker(sbi);
4242 failed_mount3:
4243         del_timer_sync(&sbi->s_err_report);
4244         if (sbi->s_mmp_tsk)
4245                 kthread_stop(sbi->s_mmp_tsk);
4246 failed_mount2:
4247         for (i = 0; i < db_count; i++)
4248                 brelse(sbi->s_group_desc[i]);
4249         ext4_kvfree(sbi->s_group_desc);
4250 failed_mount:
4251         if (sbi->s_chksum_driver)
4252                 crypto_free_shash(sbi->s_chksum_driver);
4253         if (sbi->s_proc) {
4254                 remove_proc_entry("options", sbi->s_proc);
4255                 remove_proc_entry(sb->s_id, ext4_proc_root);
4256         }
4257 #ifdef CONFIG_QUOTA
4258         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4259                 kfree(sbi->s_qf_names[i]);
4260 #endif
4261         ext4_blkdev_remove(sbi);
4262         brelse(bh);
4263 out_fail:
4264         sb->s_fs_info = NULL;
4265         kfree(sbi->s_blockgroup_lock);
4266         kfree(sbi);
4267 out_free_orig:
4268         kfree(orig_data);
4269         return err ? err : ret;
4270 }
4271
4272 /*
4273  * Setup any per-fs journal parameters now.  We'll do this both on
4274  * initial mount, once the journal has been initialised but before we've
4275  * done any recovery; and again on any subsequent remount.
4276  */
4277 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4278 {
4279         struct ext4_sb_info *sbi = EXT4_SB(sb);
4280
4281         journal->j_commit_interval = sbi->s_commit_interval;
4282         journal->j_min_batch_time = sbi->s_min_batch_time;
4283         journal->j_max_batch_time = sbi->s_max_batch_time;
4284
4285         write_lock(&journal->j_state_lock);
4286         if (test_opt(sb, BARRIER))
4287                 journal->j_flags |= JBD2_BARRIER;
4288         else
4289                 journal->j_flags &= ~JBD2_BARRIER;
4290         if (test_opt(sb, DATA_ERR_ABORT))
4291                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4292         else
4293                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4294         write_unlock(&journal->j_state_lock);
4295 }
4296
4297 static journal_t *ext4_get_journal(struct super_block *sb,
4298                                    unsigned int journal_inum)
4299 {
4300         struct inode *journal_inode;
4301         journal_t *journal;
4302
4303         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4304
4305         /* First, test for the existence of a valid inode on disk.  Bad
4306          * things happen if we iget() an unused inode, as the subsequent
4307          * iput() will try to delete it. */
4308
4309         journal_inode = ext4_iget(sb, journal_inum);
4310         if (IS_ERR(journal_inode)) {
4311                 ext4_msg(sb, KERN_ERR, "no journal found");
4312                 return NULL;
4313         }
4314         if (!journal_inode->i_nlink) {
4315                 make_bad_inode(journal_inode);
4316                 iput(journal_inode);
4317                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4318                 return NULL;
4319         }
4320
4321         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4322                   journal_inode, journal_inode->i_size);
4323         if (!S_ISREG(journal_inode->i_mode)) {
4324                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4325                 iput(journal_inode);
4326                 return NULL;
4327         }
4328
4329         journal = jbd2_journal_init_inode(journal_inode);
4330         if (!journal) {
4331                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4332                 iput(journal_inode);
4333                 return NULL;
4334         }
4335         journal->j_private = sb;
4336         ext4_init_journal_params(sb, journal);
4337         return journal;
4338 }
4339
4340 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4341                                        dev_t j_dev)
4342 {
4343         struct buffer_head *bh;
4344         journal_t *journal;
4345         ext4_fsblk_t start;
4346         ext4_fsblk_t len;
4347         int hblock, blocksize;
4348         ext4_fsblk_t sb_block;
4349         unsigned long offset;
4350         struct ext4_super_block *es;
4351         struct block_device *bdev;
4352
4353         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4354
4355         bdev = ext4_blkdev_get(j_dev, sb);
4356         if (bdev == NULL)
4357                 return NULL;
4358
4359         blocksize = sb->s_blocksize;
4360         hblock = bdev_logical_block_size(bdev);
4361         if (blocksize < hblock) {
4362                 ext4_msg(sb, KERN_ERR,
4363                         "blocksize too small for journal device");
4364                 goto out_bdev;
4365         }
4366
4367         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4368         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4369         set_blocksize(bdev, blocksize);
4370         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4371                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4372                        "external journal");
4373                 goto out_bdev;
4374         }
4375
4376         es = (struct ext4_super_block *) (bh->b_data + offset);
4377         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4378             !(le32_to_cpu(es->s_feature_incompat) &
4379               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4380                 ext4_msg(sb, KERN_ERR, "external journal has "
4381                                         "bad superblock");
4382                 brelse(bh);
4383                 goto out_bdev;
4384         }
4385
4386         if ((le32_to_cpu(es->s_feature_ro_compat) &
4387              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4388             es->s_checksum != ext4_superblock_csum(sb, es)) {
4389                 ext4_msg(sb, KERN_ERR, "external journal has "
4390                                        "corrupt superblock");
4391                 brelse(bh);
4392                 goto out_bdev;
4393         }
4394
4395         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4396                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4397                 brelse(bh);
4398                 goto out_bdev;
4399         }
4400
4401         len = ext4_blocks_count(es);
4402         start = sb_block + 1;
4403         brelse(bh);     /* we're done with the superblock */
4404
4405         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4406                                         start, len, blocksize);
4407         if (!journal) {
4408                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4409                 goto out_bdev;
4410         }
4411         journal->j_private = sb;
4412         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4413         wait_on_buffer(journal->j_sb_buffer);
4414         if (!buffer_uptodate(journal->j_sb_buffer)) {
4415                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4416                 goto out_journal;
4417         }
4418         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4419                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4420                                         "user (unsupported) - %d",
4421                         be32_to_cpu(journal->j_superblock->s_nr_users));
4422                 goto out_journal;
4423         }
4424         EXT4_SB(sb)->journal_bdev = bdev;
4425         ext4_init_journal_params(sb, journal);
4426         return journal;
4427
4428 out_journal:
4429         jbd2_journal_destroy(journal);
4430 out_bdev:
4431         ext4_blkdev_put(bdev);
4432         return NULL;
4433 }
4434
4435 static int ext4_load_journal(struct super_block *sb,
4436                              struct ext4_super_block *es,
4437                              unsigned long journal_devnum)
4438 {
4439         journal_t *journal;
4440         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4441         dev_t journal_dev;
4442         int err = 0;
4443         int really_read_only;
4444
4445         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4446
4447         if (journal_devnum &&
4448             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4449                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4450                         "numbers have changed");
4451                 journal_dev = new_decode_dev(journal_devnum);
4452         } else
4453                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4454
4455         really_read_only = bdev_read_only(sb->s_bdev);
4456
4457         /*
4458          * Are we loading a blank journal or performing recovery after a
4459          * crash?  For recovery, we need to check in advance whether we
4460          * can get read-write access to the device.
4461          */
4462         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4463                 if (sb->s_flags & MS_RDONLY) {
4464                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4465                                         "required on readonly filesystem");
4466                         if (really_read_only) {
4467                                 ext4_msg(sb, KERN_ERR, "write access "
4468                                         "unavailable, cannot proceed");
4469                                 return -EROFS;
4470                         }
4471                         ext4_msg(sb, KERN_INFO, "write access will "
4472                                "be enabled during recovery");
4473                 }
4474         }
4475
4476         if (journal_inum && journal_dev) {
4477                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4478                        "and inode journals!");
4479                 return -EINVAL;
4480         }
4481
4482         if (journal_inum) {
4483                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4484                         return -EINVAL;
4485         } else {
4486                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4487                         return -EINVAL;
4488         }
4489
4490         if (!(journal->j_flags & JBD2_BARRIER))
4491                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4492
4493         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4494                 err = jbd2_journal_wipe(journal, !really_read_only);
4495         if (!err) {
4496                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4497                 if (save)
4498                         memcpy(save, ((char *) es) +
4499                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4500                 err = jbd2_journal_load(journal);
4501                 if (save)
4502                         memcpy(((char *) es) + EXT4_S_ERR_START,
4503                                save, EXT4_S_ERR_LEN);
4504                 kfree(save);
4505         }
4506
4507         if (err) {
4508                 ext4_msg(sb, KERN_ERR, "error loading journal");
4509                 jbd2_journal_destroy(journal);
4510                 return err;
4511         }
4512
4513         EXT4_SB(sb)->s_journal = journal;
4514         ext4_clear_journal_err(sb, es);
4515
4516         if (!really_read_only && journal_devnum &&
4517             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4518                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4519
4520                 /* Make sure we flush the recovery flag to disk. */
4521                 ext4_commit_super(sb, 1);
4522         }
4523
4524         return 0;
4525 }
4526
4527 static int ext4_commit_super(struct super_block *sb, int sync)
4528 {
4529         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4530         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4531         int error = 0;
4532
4533         if (!sbh || block_device_ejected(sb))
4534                 return error;
4535         if (buffer_write_io_error(sbh)) {
4536                 /*
4537                  * Oh, dear.  A previous attempt to write the
4538                  * superblock failed.  This could happen because the
4539                  * USB device was yanked out.  Or it could happen to
4540                  * be a transient write error and maybe the block will
4541                  * be remapped.  Nothing we can do but to retry the
4542                  * write and hope for the best.
4543                  */
4544                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4545                        "superblock detected");
4546                 clear_buffer_write_io_error(sbh);
4547                 set_buffer_uptodate(sbh);
4548         }
4549         /*
4550          * If the file system is mounted read-only, don't update the
4551          * superblock write time.  This avoids updating the superblock
4552          * write time when we are mounting the root file system
4553          * read/only but we need to replay the journal; at that point,
4554          * for people who are east of GMT and who make their clock
4555          * tick in localtime for Windows bug-for-bug compatibility,
4556          * the clock is set in the future, and this will cause e2fsck
4557          * to complain and force a full file system check.
4558          */
4559         if (!(sb->s_flags & MS_RDONLY))
4560                 es->s_wtime = cpu_to_le32(get_seconds());
4561         if (sb->s_bdev->bd_part)
4562                 es->s_kbytes_written =
4563                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4564                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4565                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4566         else
4567                 es->s_kbytes_written =
4568                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4569         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4570                 ext4_free_blocks_count_set(es,
4571                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4572                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4573         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4574                 es->s_free_inodes_count =
4575                         cpu_to_le32(percpu_counter_sum_positive(
4576                                 &EXT4_SB(sb)->s_freeinodes_counter));
4577         BUFFER_TRACE(sbh, "marking dirty");
4578         ext4_superblock_csum_set(sb);
4579         mark_buffer_dirty(sbh);
4580         if (sync) {
4581                 error = sync_dirty_buffer(sbh);
4582                 if (error)
4583                         return error;
4584
4585                 error = buffer_write_io_error(sbh);
4586                 if (error) {
4587                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4588                                "superblock");
4589                         clear_buffer_write_io_error(sbh);
4590                         set_buffer_uptodate(sbh);
4591                 }
4592         }
4593         return error;
4594 }
4595
4596 /*
4597  * Have we just finished recovery?  If so, and if we are mounting (or
4598  * remounting) the filesystem readonly, then we will end up with a
4599  * consistent fs on disk.  Record that fact.
4600  */
4601 static void ext4_mark_recovery_complete(struct super_block *sb,
4602                                         struct ext4_super_block *es)
4603 {
4604         journal_t *journal = EXT4_SB(sb)->s_journal;
4605
4606         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4607                 BUG_ON(journal != NULL);
4608                 return;
4609         }
4610         jbd2_journal_lock_updates(journal);
4611         if (jbd2_journal_flush(journal) < 0)
4612                 goto out;
4613
4614         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4615             sb->s_flags & MS_RDONLY) {
4616                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4617                 ext4_commit_super(sb, 1);
4618         }
4619
4620 out:
4621         jbd2_journal_unlock_updates(journal);
4622 }
4623
4624 /*
4625  * If we are mounting (or read-write remounting) a filesystem whose journal
4626  * has recorded an error from a previous lifetime, move that error to the
4627  * main filesystem now.
4628  */
4629 static void ext4_clear_journal_err(struct super_block *sb,
4630                                    struct ext4_super_block *es)
4631 {
4632         journal_t *journal;
4633         int j_errno;
4634         const char *errstr;
4635
4636         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4637
4638         journal = EXT4_SB(sb)->s_journal;
4639
4640         /*
4641          * Now check for any error status which may have been recorded in the
4642          * journal by a prior ext4_error() or ext4_abort()
4643          */
4644
4645         j_errno = jbd2_journal_errno(journal);
4646         if (j_errno) {
4647                 char nbuf[16];
4648
4649                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4650                 ext4_warning(sb, "Filesystem error recorded "
4651                              "from previous mount: %s", errstr);
4652                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4653
4654                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4655                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4656                 ext4_commit_super(sb, 1);
4657
4658                 jbd2_journal_clear_err(journal);
4659                 jbd2_journal_update_sb_errno(journal);
4660         }
4661 }
4662
4663 /*
4664  * Force the running and committing transactions to commit,
4665  * and wait on the commit.
4666  */
4667 int ext4_force_commit(struct super_block *sb)
4668 {
4669         journal_t *journal;
4670
4671         if (sb->s_flags & MS_RDONLY)
4672                 return 0;
4673
4674         journal = EXT4_SB(sb)->s_journal;
4675         return ext4_journal_force_commit(journal);
4676 }
4677
4678 static int ext4_sync_fs(struct super_block *sb, int wait)
4679 {
4680         int ret = 0;
4681         tid_t target;
4682         bool needs_barrier = false;
4683         struct ext4_sb_info *sbi = EXT4_SB(sb);
4684
4685         trace_ext4_sync_fs(sb, wait);
4686         flush_workqueue(sbi->rsv_conversion_wq);
4687         /*
4688          * Writeback quota in non-journalled quota case - journalled quota has
4689          * no dirty dquots
4690          */
4691         dquot_writeback_dquots(sb, -1);
4692         /*
4693          * Data writeback is possible w/o journal transaction, so barrier must
4694          * being sent at the end of the function. But we can skip it if
4695          * transaction_commit will do it for us.
4696          */
4697         if (sbi->s_journal) {
4698                 target = jbd2_get_latest_transaction(sbi->s_journal);
4699                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4700                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4701                         needs_barrier = true;
4702
4703                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4704                         if (wait)
4705                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4706                                                            target);
4707                 }
4708         } else if (wait && test_opt(sb, BARRIER))
4709                 needs_barrier = true;
4710         if (needs_barrier) {
4711                 int err;
4712                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4713                 if (!ret)
4714                         ret = err;
4715         }
4716
4717         return ret;
4718 }
4719
4720 /*
4721  * LVM calls this function before a (read-only) snapshot is created.  This
4722  * gives us a chance to flush the journal completely and mark the fs clean.
4723  *
4724  * Note that only this function cannot bring a filesystem to be in a clean
4725  * state independently. It relies on upper layer to stop all data & metadata
4726  * modifications.
4727  */
4728 static int ext4_freeze(struct super_block *sb)
4729 {
4730         int error = 0;
4731         journal_t *journal;
4732
4733         if (sb->s_flags & MS_RDONLY)
4734                 return 0;
4735
4736         journal = EXT4_SB(sb)->s_journal;
4737
4738         if (journal) {
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
4744                  * flush the journal.
4745                  */
4746                 error = jbd2_journal_flush(journal);
4747                 if (error < 0)
4748                         goto out;
4749         }
4750
4751         /* Journal blocked and flushed, clear needs_recovery flag. */
4752         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4753         error = ext4_commit_super(sb, 1);
4754 out:
4755         if (journal)
4756                 /* we rely on upper layer to stop further updates */
4757                 jbd2_journal_unlock_updates(journal);
4758         return error;
4759 }
4760
4761 /*
4762  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4763  * flag here, even though the filesystem is not technically dirty yet.
4764  */
4765 static int ext4_unfreeze(struct super_block *sb)
4766 {
4767         if (sb->s_flags & MS_RDONLY)
4768                 return 0;
4769
4770         /* Reset the needs_recovery flag before the fs is unlocked. */
4771         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4772         ext4_commit_super(sb, 1);
4773         return 0;
4774 }
4775
4776 /*
4777  * Structure to save mount options for ext4_remount's benefit
4778  */
4779 struct ext4_mount_options {
4780         unsigned long s_mount_opt;
4781         unsigned long s_mount_opt2;
4782         kuid_t s_resuid;
4783         kgid_t s_resgid;
4784         unsigned long s_commit_interval;
4785         u32 s_min_batch_time, s_max_batch_time;
4786 #ifdef CONFIG_QUOTA
4787         int s_jquota_fmt;
4788         char *s_qf_names[EXT4_MAXQUOTAS];
4789 #endif
4790 };
4791
4792 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4793 {
4794         struct ext4_super_block *es;
4795         struct ext4_sb_info *sbi = EXT4_SB(sb);
4796         unsigned long old_sb_flags;
4797         struct ext4_mount_options old_opts;
4798         int enable_quota = 0;
4799         ext4_group_t g;
4800         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4801         int err = 0;
4802 #ifdef CONFIG_QUOTA
4803         int i, j;
4804 #endif
4805         char *orig_data = kstrdup(data, GFP_KERNEL);
4806
4807         /* Store the original options */
4808         old_sb_flags = sb->s_flags;
4809         old_opts.s_mount_opt = sbi->s_mount_opt;
4810         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4811         old_opts.s_resuid = sbi->s_resuid;
4812         old_opts.s_resgid = sbi->s_resgid;
4813         old_opts.s_commit_interval = sbi->s_commit_interval;
4814         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4815         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4816 #ifdef CONFIG_QUOTA
4817         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4818         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4819                 if (sbi->s_qf_names[i]) {
4820                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4821                                                          GFP_KERNEL);
4822                         if (!old_opts.s_qf_names[i]) {
4823                                 for (j = 0; j < i; j++)
4824                                         kfree(old_opts.s_qf_names[j]);
4825                                 kfree(orig_data);
4826                                 return -ENOMEM;
4827                         }
4828                 } else
4829                         old_opts.s_qf_names[i] = NULL;
4830 #endif
4831         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4832                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4833
4834         /*
4835          * Allow the "check" option to be passed as a remount option.
4836          */
4837         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4838                 err = -EINVAL;
4839                 goto restore_opts;
4840         }
4841
4842         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4843                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4844                         ext4_msg(sb, KERN_ERR, "can't mount with "
4845                                  "both data=journal and delalloc");
4846                         err = -EINVAL;
4847                         goto restore_opts;
4848                 }
4849                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4850                         ext4_msg(sb, KERN_ERR, "can't mount with "
4851                                  "both data=journal and dioread_nolock");
4852                         err = -EINVAL;
4853                         goto restore_opts;
4854                 }
4855         }
4856
4857         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4858                 ext4_abort(sb, "Abort forced by user");
4859
4860         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4861                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4862
4863         es = sbi->s_es;
4864
4865         if (sbi->s_journal) {
4866                 ext4_init_journal_params(sb, sbi->s_journal);
4867                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4868         }
4869
4870         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4871                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4872                         err = -EROFS;
4873                         goto restore_opts;
4874                 }
4875
4876                 if (*flags & MS_RDONLY) {
4877                         err = sync_filesystem(sb);
4878                         if (err < 0)
4879                                 goto restore_opts;
4880                         err = dquot_suspend(sb, -1);
4881                         if (err < 0)
4882                                 goto restore_opts;
4883
4884                         /*
4885                          * First of all, the unconditional stuff we have to do
4886                          * to disable replay of the journal when we next remount
4887                          */
4888                         sb->s_flags |= MS_RDONLY;
4889
4890                         /*
4891                          * OK, test if we are remounting a valid rw partition
4892                          * readonly, and if so set the rdonly flag and then
4893                          * mark the partition as valid again.
4894                          */
4895                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4896                             (sbi->s_mount_state & EXT4_VALID_FS))
4897                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4898
4899                         if (sbi->s_journal)
4900                                 ext4_mark_recovery_complete(sb, es);
4901                 } else {
4902                         /* Make sure we can mount this feature set readwrite */
4903                         if (!ext4_feature_set_ok(sb, 0)) {
4904                                 err = -EROFS;
4905                                 goto restore_opts;
4906                         }
4907                         /*
4908                          * Make sure the group descriptor checksums
4909                          * are sane.  If they aren't, refuse to remount r/w.
4910                          */
4911                         for (g = 0; g < sbi->s_groups_count; g++) {
4912                                 struct ext4_group_desc *gdp =
4913                                         ext4_get_group_desc(sb, g, NULL);
4914
4915                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4916                                         ext4_msg(sb, KERN_ERR,
4917                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4918                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4919                                                le16_to_cpu(gdp->bg_checksum));
4920                                         err = -EINVAL;
4921                                         goto restore_opts;
4922                                 }
4923                         }
4924
4925                         /*
4926                          * If we have an unprocessed orphan list hanging
4927                          * around from a previously readonly bdev mount,
4928                          * require a full umount/remount for now.
4929                          */
4930                         if (es->s_last_orphan) {
4931                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4932                                        "remount RDWR because of unprocessed "
4933                                        "orphan inode list.  Please "
4934                                        "umount/remount instead");
4935                                 err = -EINVAL;
4936                                 goto restore_opts;
4937                         }
4938
4939                         /*
4940                          * Mounting a RDONLY partition read-write, so reread
4941                          * and store the current valid flag.  (It may have
4942                          * been changed by e2fsck since we originally mounted
4943                          * the partition.)
4944                          */
4945                         if (sbi->s_journal)
4946                                 ext4_clear_journal_err(sb, es);
4947                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4948                         if (!ext4_setup_super(sb, es, 0))
4949                                 sb->s_flags &= ~MS_RDONLY;
4950                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4951                                                      EXT4_FEATURE_INCOMPAT_MMP))
4952                                 if (ext4_multi_mount_protect(sb,
4953                                                 le64_to_cpu(es->s_mmp_block))) {
4954                                         err = -EROFS;
4955                                         goto restore_opts;
4956                                 }
4957                         enable_quota = 1;
4958                 }
4959         }
4960
4961         /*
4962          * Reinitialize lazy itable initialization thread based on
4963          * current settings
4964          */
4965         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4966                 ext4_unregister_li_request(sb);
4967         else {
4968                 ext4_group_t first_not_zeroed;
4969                 first_not_zeroed = ext4_has_uninit_itable(sb);
4970                 ext4_register_li_request(sb, first_not_zeroed);
4971         }
4972
4973         ext4_setup_system_zone(sb);
4974         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4975                 ext4_commit_super(sb, 1);
4976
4977 #ifdef CONFIG_QUOTA
4978         /* Release old quota file names */
4979         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4980                 kfree(old_opts.s_qf_names[i]);
4981         if (enable_quota) {
4982                 if (sb_any_quota_suspended(sb))
4983                         dquot_resume(sb, -1);
4984                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4985                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4986                         err = ext4_enable_quotas(sb);
4987                         if (err)
4988                                 goto restore_opts;
4989                 }
4990         }
4991 #endif
4992
4993         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4994         kfree(orig_data);
4995         return 0;
4996
4997 restore_opts:
4998         sb->s_flags = old_sb_flags;
4999         sbi->s_mount_opt = old_opts.s_mount_opt;
5000         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5001         sbi->s_resuid = old_opts.s_resuid;
5002         sbi->s_resgid = old_opts.s_resgid;
5003         sbi->s_commit_interval = old_opts.s_commit_interval;
5004         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5005         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5006 #ifdef CONFIG_QUOTA
5007         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5008         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5009                 kfree(sbi->s_qf_names[i]);
5010                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5011         }
5012 #endif
5013         kfree(orig_data);
5014         return err;
5015 }
5016
5017 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5018 {
5019         struct super_block *sb = dentry->d_sb;
5020         struct ext4_sb_info *sbi = EXT4_SB(sb);
5021         struct ext4_super_block *es = sbi->s_es;
5022         ext4_fsblk_t overhead = 0, resv_blocks;
5023         u64 fsid;
5024         s64 bfree;
5025         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5026
5027         if (!test_opt(sb, MINIX_DF))
5028                 overhead = sbi->s_overhead;
5029
5030         buf->f_type = EXT4_SUPER_MAGIC;
5031         buf->f_bsize = sb->s_blocksize;
5032         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5033         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5034                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5035         /* prevent underflow in case that few free space is available */
5036         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5037         buf->f_bavail = buf->f_bfree -
5038                         (ext4_r_blocks_count(es) + resv_blocks);
5039         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5040                 buf->f_bavail = 0;
5041         buf->f_files = le32_to_cpu(es->s_inodes_count);
5042         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5043         buf->f_namelen = EXT4_NAME_LEN;
5044         fsid = le64_to_cpup((void *)es->s_uuid) ^
5045                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5046         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5047         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5048
5049         return 0;
5050 }
5051
5052 /* Helper function for writing quotas on sync - we need to start transaction
5053  * before quota file is locked for write. Otherwise the are possible deadlocks:
5054  * Process 1                         Process 2
5055  * ext4_create()                     quota_sync()
5056  *   jbd2_journal_start()                  write_dquot()
5057  *   dquot_initialize()                         down(dqio_mutex)
5058  *     down(dqio_mutex)                    jbd2_journal_start()
5059  *
5060  */
5061
5062 #ifdef CONFIG_QUOTA
5063
5064 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5065 {
5066         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5067 }
5068
5069 static int ext4_write_dquot(struct dquot *dquot)
5070 {
5071         int ret, err;
5072         handle_t *handle;
5073         struct inode *inode;
5074
5075         inode = dquot_to_inode(dquot);
5076         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5077                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5078         if (IS_ERR(handle))
5079                 return PTR_ERR(handle);
5080         ret = dquot_commit(dquot);
5081         err = ext4_journal_stop(handle);
5082         if (!ret)
5083                 ret = err;
5084         return ret;
5085 }
5086
5087 static int ext4_acquire_dquot(struct dquot *dquot)
5088 {
5089         int ret, err;
5090         handle_t *handle;
5091
5092         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5093                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5094         if (IS_ERR(handle))
5095                 return PTR_ERR(handle);
5096         ret = dquot_acquire(dquot);
5097         err = ext4_journal_stop(handle);
5098         if (!ret)
5099                 ret = err;
5100         return ret;
5101 }
5102
5103 static int ext4_release_dquot(struct dquot *dquot)
5104 {
5105         int ret, err;
5106         handle_t *handle;
5107
5108         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5109                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5110         if (IS_ERR(handle)) {
5111                 /* Release dquot anyway to avoid endless cycle in dqput() */
5112                 dquot_release(dquot);
5113                 return PTR_ERR(handle);
5114         }
5115         ret = dquot_release(dquot);
5116         err = ext4_journal_stop(handle);
5117         if (!ret)
5118                 ret = err;
5119         return ret;
5120 }
5121
5122 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5123 {
5124         struct super_block *sb = dquot->dq_sb;
5125         struct ext4_sb_info *sbi = EXT4_SB(sb);
5126
5127         /* Are we journaling quotas? */
5128         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5129             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5130                 dquot_mark_dquot_dirty(dquot);
5131                 return ext4_write_dquot(dquot);
5132         } else {
5133                 return dquot_mark_dquot_dirty(dquot);
5134         }
5135 }
5136
5137 static int ext4_write_info(struct super_block *sb, int type)
5138 {
5139         int ret, err;
5140         handle_t *handle;
5141
5142         /* Data block + inode block */
5143         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5144         if (IS_ERR(handle))
5145                 return PTR_ERR(handle);
5146         ret = dquot_commit_info(sb, type);
5147         err = ext4_journal_stop(handle);
5148         if (!ret)
5149                 ret = err;
5150         return ret;
5151 }
5152
5153 /*
5154  * Turn on quotas during mount time - we need to find
5155  * the quota file and such...
5156  */
5157 static int ext4_quota_on_mount(struct super_block *sb, int type)
5158 {
5159         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5160                                         EXT4_SB(sb)->s_jquota_fmt, type);
5161 }
5162
5163 /*
5164  * Standard function to be called on quota_on
5165  */
5166 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5167                          struct path *path)
5168 {
5169         int err;
5170
5171         if (!test_opt(sb, QUOTA))
5172                 return -EINVAL;
5173
5174         /* Quotafile not on the same filesystem? */
5175         if (path->dentry->d_sb != sb)
5176                 return -EXDEV;
5177         /* Journaling quota? */
5178         if (EXT4_SB(sb)->s_qf_names[type]) {
5179                 /* Quotafile not in fs root? */
5180                 if (path->dentry->d_parent != sb->s_root)
5181                         ext4_msg(sb, KERN_WARNING,
5182                                 "Quota file not on filesystem root. "
5183                                 "Journaled quota will not work");
5184         }
5185
5186         /*
5187          * When we journal data on quota file, we have to flush journal to see
5188          * all updates to the file when we bypass pagecache...
5189          */
5190         if (EXT4_SB(sb)->s_journal &&
5191             ext4_should_journal_data(path->dentry->d_inode)) {
5192                 /*
5193                  * We don't need to lock updates but journal_flush() could
5194                  * otherwise be livelocked...
5195                  */
5196                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5197                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5198                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5199                 if (err)
5200                         return err;
5201         }
5202
5203         return dquot_quota_on(sb, type, format_id, path);
5204 }
5205
5206 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5207                              unsigned int flags)
5208 {
5209         int err;
5210         struct inode *qf_inode;
5211         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5212                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5213                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5214         };
5215
5216         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5217
5218         if (!qf_inums[type])
5219                 return -EPERM;
5220
5221         qf_inode = ext4_iget(sb, qf_inums[type]);
5222         if (IS_ERR(qf_inode)) {
5223                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5224                 return PTR_ERR(qf_inode);
5225         }
5226
5227         /* Don't account quota for quota files to avoid recursion */
5228         qf_inode->i_flags |= S_NOQUOTA;
5229         err = dquot_enable(qf_inode, type, format_id, flags);
5230         iput(qf_inode);
5231
5232         return err;
5233 }
5234
5235 /* Enable usage tracking for all quota types. */
5236 static int ext4_enable_quotas(struct super_block *sb)
5237 {
5238         int type, err = 0;
5239         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5240                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5241                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5242         };
5243
5244         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5245         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5246                 if (qf_inums[type]) {
5247                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5248                                                 DQUOT_USAGE_ENABLED);
5249                         if (err) {
5250                                 ext4_warning(sb,
5251                                         "Failed to enable quota tracking "
5252                                         "(type=%d, err=%d). Please run "
5253                                         "e2fsck to fix.", type, err);
5254                                 return err;
5255                         }
5256                 }
5257         }
5258         return 0;
5259 }
5260
5261 /*
5262  * quota_on function that is used when QUOTA feature is set.
5263  */
5264 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5265                                  int format_id)
5266 {
5267         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5268                 return -EINVAL;
5269
5270         /*
5271          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5272          */
5273         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5274 }
5275
5276 static int ext4_quota_off(struct super_block *sb, int type)
5277 {
5278         struct inode *inode = sb_dqopt(sb)->files[type];
5279         handle_t *handle;
5280
5281         /* Force all delayed allocation blocks to be allocated.
5282          * Caller already holds s_umount sem */
5283         if (test_opt(sb, DELALLOC))
5284                 sync_filesystem(sb);
5285
5286         if (!inode)
5287                 goto out;
5288
5289         /* Update modification times of quota files when userspace can
5290          * start looking at them */
5291         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5292         if (IS_ERR(handle))
5293                 goto out;
5294         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5295         ext4_mark_inode_dirty(handle, inode);
5296         ext4_journal_stop(handle);
5297
5298 out:
5299         return dquot_quota_off(sb, type);
5300 }
5301
5302 /*
5303  * quota_off function that is used when QUOTA feature is set.
5304  */
5305 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5306 {
5307         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5308                 return -EINVAL;
5309
5310         /* Disable only the limits. */
5311         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5312 }
5313
5314 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5315  * acquiring the locks... As quota files are never truncated and quota code
5316  * itself serializes the operations (and no one else should touch the files)
5317  * we don't have to be afraid of races */
5318 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5319                                size_t len, loff_t off)
5320 {
5321         struct inode *inode = sb_dqopt(sb)->files[type];
5322         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5323         int offset = off & (sb->s_blocksize - 1);
5324         int tocopy;
5325         size_t toread;
5326         struct buffer_head *bh;
5327         loff_t i_size = i_size_read(inode);
5328
5329         if (off > i_size)
5330                 return 0;
5331         if (off+len > i_size)
5332                 len = i_size-off;
5333         toread = len;
5334         while (toread > 0) {
5335                 tocopy = sb->s_blocksize - offset < toread ?
5336                                 sb->s_blocksize - offset : toread;
5337                 bh = ext4_bread(NULL, inode, blk, 0);
5338                 if (IS_ERR(bh))
5339                         return PTR_ERR(bh);
5340                 if (!bh)        /* A hole? */
5341                         memset(data, 0, tocopy);
5342                 else
5343                         memcpy(data, bh->b_data+offset, tocopy);
5344                 brelse(bh);
5345                 offset = 0;
5346                 toread -= tocopy;
5347                 data += tocopy;
5348                 blk++;
5349         }
5350         return len;
5351 }
5352
5353 /* Write to quotafile (we know the transaction is already started and has
5354  * enough credits) */
5355 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5356                                 const char *data, size_t len, loff_t off)
5357 {
5358         struct inode *inode = sb_dqopt(sb)->files[type];
5359         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5360         int err, 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);
5382         if (IS_ERR(bh))
5383                 return PTR_ERR(bh);
5384         if (!bh)
5385                 goto out;
5386         BUFFER_TRACE(bh, "get write access");
5387         err = ext4_journal_get_write_access(handle, bh);
5388         if (err) {
5389                 brelse(bh);
5390                 return err;
5391         }
5392         lock_buffer(bh);
5393         memcpy(bh->b_data+offset, data, len);
5394         flush_dcache_page(bh->b_page);
5395         unlock_buffer(bh);
5396         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5397         brelse(bh);
5398 out:
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)