fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 MODULE_ALIAS_FS("ext2");
  97 MODULE_ALIAS("ext2");
  98 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  99 #else
 100 #define IS_EXT2_SB(sb) (0)
 101 #endif
 102
 103
 104 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 105 static struct file_system_type ext3_fs_type = {
 106         .owner          = THIS_MODULE,
 107         .name           = "ext3",
 108         .mount          = ext4_mount,
 109         .kill_sb        = kill_block_super,
 110         .fs_flags       = FS_REQUIRES_DEV,
 111 };
 112 MODULE_ALIAS_FS("ext3");
 113 MODULE_ALIAS("ext3");
 114 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 115 #else
 116 #define IS_EXT3_SB(sb) (0)
 117 #endif
 118
 119 static int ext4_verify_csum_type(struct super_block *sb,
 120                                  struct ext4_super_block *es)
 121 {
 122         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 123                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 124                 return 1;
 125
 126         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 127 }
 128
 129 static __le32 ext4_superblock_csum(struct super_block *sb,
 130                                    struct ext4_super_block *es)
 131 {
 132         struct ext4_sb_info *sbi = EXT4_SB(sb);
 133         int offset = offsetof(struct ext4_super_block, s_checksum);
 134         __u32 csum;
 135
 136         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 137
 138         return cpu_to_le32(csum);
 139 }
 140
 141 static int ext4_superblock_csum_verify(struct super_block *sb,
 142                                        struct ext4_super_block *es)
 143 {
 144         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 145                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 146                 return 1;
 147
 148         return es->s_checksum == ext4_superblock_csum(sb, es);
 149 }
 150
 151 void ext4_superblock_csum_set(struct super_block *sb)
 152 {
 153         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 154
 155         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 156                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 157                 return;
 158
 159         es->s_checksum = ext4_superblock_csum(sb, es);
 160 }
 161
 162 void *ext4_kvmalloc(size_t size, gfp_t flags)
 163 {
 164         void *ret;
 165
 166         ret = kmalloc(size, flags | __GFP_NOWARN);
 167         if (!ret)
 168                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 169         return ret;
 170 }
 171
 172 void *ext4_kvzalloc(size_t size, gfp_t flags)
 173 {
 174         void *ret;
 175
 176         ret = kzalloc(size, flags | __GFP_NOWARN);
 177         if (!ret)
 178                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 179         return ret;
 180 }
 181
 182 void ext4_kvfree(void *ptr)
 183 {
 184         if (is_vmalloc_addr(ptr))
 185                 vfree(ptr);
 186         else
 187                 kfree(ptr);
 188
 189 }
 190
 191 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 192                                struct ext4_group_desc *bg)
 193 {
 194         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 197 }
 198
 199 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 200                                struct ext4_group_desc *bg)
 201 {
 202         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 203                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 204                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 205 }
 206
 207 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 208                               struct ext4_group_desc *bg)
 209 {
 210         return le32_to_cpu(bg->bg_inode_table_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 213 }
 214
 215 __u32 ext4_free_group_clusters(struct super_block *sb,
 216                                struct ext4_group_desc *bg)
 217 {
 218         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 221 }
 222
 223 __u32 ext4_free_inodes_count(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 229 }
 230
 231 __u32 ext4_used_dirs_count(struct super_block *sb,
 232                               struct ext4_group_desc *bg)
 233 {
 234         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 235                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 236                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 237 }
 238
 239 __u32 ext4_itable_unused_count(struct super_block *sb,
 240                               struct ext4_group_desc *bg)
 241 {
 242         return le16_to_cpu(bg->bg_itable_unused_lo) |
 243                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 244                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 245 }
 246
 247 void ext4_block_bitmap_set(struct super_block *sb,
 248                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 249 {
 250         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 253 }
 254
 255 void ext4_inode_bitmap_set(struct super_block *sb,
 256                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 257 {
 258         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 259         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 260                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 261 }
 262
 263 void ext4_inode_table_set(struct super_block *sb,
 264                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 265 {
 266         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 269 }
 270
 271 void ext4_free_group_clusters_set(struct super_block *sb,
 272                                   struct ext4_group_desc *bg, __u32 count)
 273 {
 274         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 277 }
 278
 279 void ext4_free_inodes_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, __u32 count)
 281 {
 282         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 285 }
 286
 287 void ext4_used_dirs_set(struct super_block *sb,
 288                           struct ext4_group_desc *bg, __u32 count)
 289 {
 290         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 291         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 292                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 293 }
 294
 295 void ext4_itable_unused_set(struct super_block *sb,
 296                           struct ext4_group_desc *bg, __u32 count)
 297 {
 298         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 299         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 300                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 301 }
 302
 303
 304 static void __save_error_info(struct super_block *sb, const char *func,
 305                             unsigned int line)
 306 {
 307         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 308
 309         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 310         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 311         es->s_last_error_time = cpu_to_le32(get_seconds());
 312         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 313         es->s_last_error_line = cpu_to_le32(line);
 314         if (!es->s_first_error_time) {
 315                 es->s_first_error_time = es->s_last_error_time;
 316                 strncpy(es->s_first_error_func, func,
 317                         sizeof(es->s_first_error_func));
 318                 es->s_first_error_line = cpu_to_le32(line);
 319                 es->s_first_error_ino = es->s_last_error_ino;
 320                 es->s_first_error_block = es->s_last_error_block;
 321         }
 322         /*
 323          * Start the daily error reporting function if it hasn't been
 324          * started already
 325          */
 326         if (!es->s_error_count)
 327                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 328         le32_add_cpu(&es->s_error_count, 1);
 329 }
 330
 331 static void save_error_info(struct super_block *sb, const char *func,
 332                             unsigned int line)
 333 {
 334         __save_error_info(sb, func, line);
 335         ext4_commit_super(sb, 1);
 336 }
 337
 338 /*
 339  * The del_gendisk() function uninitializes the disk-specific data
 340  * structures, including the bdi structure, without telling anyone
 341  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 342  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 343  * This is a kludge to prevent these oops until we can put in a proper
 344  * hook in del_gendisk() to inform the VFS and file system layers.
 345  */
 346 static int block_device_ejected(struct super_block *sb)
 347 {
 348         struct inode *bd_inode = sb->s_bdev->bd_inode;
 349         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 350
 351         return bdi->dev == NULL;
 352 }
 353
 354 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 355 {
 356         struct super_block              *sb = journal->j_private;
 357         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 358         int                             error = is_journal_aborted(journal);
 359         struct ext4_journal_cb_entry    *jce;
 360
 361         BUG_ON(txn->t_state == T_FINISHED);
 362         spin_lock(&sbi->s_md_lock);
 363         while (!list_empty(&txn->t_private_list)) {
 364                 jce = list_entry(txn->t_private_list.next,
 365                                  struct ext4_journal_cb_entry, jce_list);
 366                 list_del_init(&jce->jce_list);
 367                 spin_unlock(&sbi->s_md_lock);
 368                 jce->jce_func(sb, jce, error);
 369                 spin_lock(&sbi->s_md_lock);
 370         }
 371         spin_unlock(&sbi->s_md_lock);
 372 }
 373
 374 /* Deal with the reporting of failure conditions on a filesystem such as
 375  * inconsistencies detected or read IO failures.
 376  *
 377  * On ext2, we can store the error state of the filesystem in the
 378  * superblock.  That is not possible on ext4, because we may have other
 379  * write ordering constraints on the superblock which prevent us from
 380  * writing it out straight away; and given that the journal is about to
 381  * be aborted, we can't rely on the current, or future, transactions to
 382  * write out the superblock safely.
 383  *
 384  * We'll just use the jbd2_journal_abort() error code to record an error in
 385  * the journal instead.  On recovery, the journal will complain about
 386  * that error until we've noted it down and cleared it.
 387  */
 388
 389 static void ext4_handle_error(struct super_block *sb)
 390 {
 391         if (sb->s_flags & MS_RDONLY)
 392                 return;
 393
 394         if (!test_opt(sb, ERRORS_CONT)) {
 395                 journal_t *journal = EXT4_SB(sb)->s_journal;
 396
 397                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 398                 if (journal)
 399                         jbd2_journal_abort(journal, -EIO);
 400         }
 401         if (test_opt(sb, ERRORS_RO)) {
 402                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 403                 /*
 404                  * Make sure updated value of ->s_mount_flags will be visible
 405                  * before ->s_flags update
 406                  */
 407                 smp_wmb();
 408                 sb->s_flags |= MS_RDONLY;
 409         }
 410         if (test_opt(sb, ERRORS_PANIC))
 411                 panic("EXT4-fs (device %s): panic forced after error\n",
 412                         sb->s_id);
 413 }
 414
 415 #define ext4_error_ratelimit(sb)                                        \
 416                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 417                              "EXT4-fs error")
 418
 419 void __ext4_error(struct super_block *sb, const char *function,
 420                   unsigned int line, const char *fmt, ...)
 421 {
 422         struct va_format vaf;
 423         va_list args;
 424
 425         if (ext4_error_ratelimit(sb)) {
 426                 va_start(args, fmt);
 427                 vaf.fmt = fmt;
 428                 vaf.va = &args;
 429                 printk(KERN_CRIT
 430                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 431                        sb->s_id, function, line, current->comm, &vaf);
 432                 va_end(args);
 433         }
 434         save_error_info(sb, function, line);
 435         ext4_handle_error(sb);
 436 }
 437
 438 void __ext4_error_inode(struct inode *inode, const char *function,
 439                         unsigned int line, ext4_fsblk_t block,
 440                         const char *fmt, ...)
 441 {
 442         va_list args;
 443         struct va_format vaf;
 444         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 445
 446         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 447         es->s_last_error_block = cpu_to_le64(block);
 448         if (ext4_error_ratelimit(inode->i_sb)) {
 449                 va_start(args, fmt);
 450                 vaf.fmt = fmt;
 451                 vaf.va = &args;
 452                 if (block)
 453                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 454                                "inode #%lu: block %llu: comm %s: %pV\n",
 455                                inode->i_sb->s_id, function, line, inode->i_ino,
 456                                block, current->comm, &vaf);
 457                 else
 458                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 459                                "inode #%lu: comm %s: %pV\n",
 460                                inode->i_sb->s_id, function, line, inode->i_ino,
 461                                current->comm, &vaf);
 462                 va_end(args);
 463         }
 464         save_error_info(inode->i_sb, function, line);
 465         ext4_handle_error(inode->i_sb);
 466 }
 467
 468 void __ext4_error_file(struct file *file, const char *function,
 469                        unsigned int line, ext4_fsblk_t block,
 470                        const char *fmt, ...)
 471 {
 472         va_list args;
 473         struct va_format vaf;
 474         struct ext4_super_block *es;
 475         struct inode *inode = file_inode(file);
 476         char pathname[80], *path;
 477
 478         es = EXT4_SB(inode->i_sb)->s_es;
 479         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 480         if (ext4_error_ratelimit(inode->i_sb)) {
 481                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 482                 if (IS_ERR(path))
 483                         path = "(unknown)";
 484                 va_start(args, fmt);
 485                 vaf.fmt = fmt;
 486                 vaf.va = &args;
 487                 if (block)
 488                         printk(KERN_CRIT
 489                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 490                                "block %llu: comm %s: path %s: %pV\n",
 491                                inode->i_sb->s_id, function, line, inode->i_ino,
 492                                block, current->comm, path, &vaf);
 493                 else
 494                         printk(KERN_CRIT
 495                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 496                                "comm %s: path %s: %pV\n",
 497                                inode->i_sb->s_id, function, line, inode->i_ino,
 498                                current->comm, path, &vaf);
 499                 va_end(args);
 500         }
 501         save_error_info(inode->i_sb, function, line);
 502         ext4_handle_error(inode->i_sb);
 503 }
 504
 505 const char *ext4_decode_error(struct super_block *sb, int errno,
 506                               char nbuf[16])
 507 {
 508         char *errstr = NULL;
 509
 510         switch (errno) {
 511         case -EIO:
 512                 errstr = "IO failure";
 513                 break;
 514         case -ENOMEM:
 515                 errstr = "Out of memory";
 516                 break;
 517         case -EROFS:
 518                 if (!sb || (EXT4_SB(sb)->s_journal &&
 519                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 520                         errstr = "Journal has aborted";
 521                 else
 522                         errstr = "Readonly filesystem";
 523                 break;
 524         default:
 525                 /* If the caller passed in an extra buffer for unknown
 526                  * errors, textualise them now.  Else we just return
 527                  * NULL. */
 528                 if (nbuf) {
 529                         /* Check for truncated error codes... */
 530                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 531                                 errstr = nbuf;
 532                 }
 533                 break;
 534         }
 535
 536         return errstr;
 537 }
 538
 539 /* __ext4_std_error decodes expected errors from journaling functions
 540  * automatically and invokes the appropriate error response.  */
 541
 542 void __ext4_std_error(struct super_block *sb, const char *function,
 543                       unsigned int line, int errno)
 544 {
 545         char nbuf[16];
 546         const char *errstr;
 547
 548         /* Special case: if the error is EROFS, and we're not already
 549          * inside a transaction, then there's really no point in logging
 550          * an error. */
 551         if (errno == -EROFS && journal_current_handle() == NULL &&
 552             (sb->s_flags & MS_RDONLY))
 553                 return;
 554
 555         if (ext4_error_ratelimit(sb)) {
 556                 errstr = ext4_decode_error(sb, errno, nbuf);
 557                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 558                        sb->s_id, function, line, errstr);
 559         }
 560
 561         save_error_info(sb, function, line);
 562         ext4_handle_error(sb);
 563 }
 564
 565 /*
 566  * ext4_abort is a much stronger failure handler than ext4_error.  The
 567  * abort function may be used to deal with unrecoverable failures such
 568  * as journal IO errors or ENOMEM at a critical moment in log management.
 569  *
 570  * We unconditionally force the filesystem into an ABORT|READONLY state,
 571  * unless the error response on the fs has been set to panic in which
 572  * case we take the easy way out and panic immediately.
 573  */
 574
 575 void __ext4_abort(struct super_block *sb, const char *function,
 576                 unsigned int line, const char *fmt, ...)
 577 {
 578         va_list args;
 579
 580         save_error_info(sb, function, line);
 581         va_start(args, fmt);
 582         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 583                function, line);
 584         vprintk(fmt, args);
 585         printk("\n");
 586         va_end(args);
 587
 588         if ((sb->s_flags & MS_RDONLY) == 0) {
 589                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 590                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 591                 /*
 592                  * Make sure updated value of ->s_mount_flags will be visible
 593                  * before ->s_flags update
 594                  */
 595                 smp_wmb();
 596                 sb->s_flags |= MS_RDONLY;
 597                 if (EXT4_SB(sb)->s_journal)
 598                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 599                 save_error_info(sb, function, line);
 600         }
 601         if (test_opt(sb, ERRORS_PANIC))
 602                 panic("EXT4-fs panic from previous error\n");
 603 }
 604
 605 void __ext4_msg(struct super_block *sb,
 606                 const char *prefix, const char *fmt, ...)
 607 {
 608         struct va_format vaf;
 609         va_list args;
 610
 611         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 612                 return;
 613
 614         va_start(args, fmt);
 615         vaf.fmt = fmt;
 616         vaf.va = &args;
 617         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 618         va_end(args);
 619 }
 620
 621 void __ext4_warning(struct super_block *sb, const char *function,
 622                     unsigned int line, const char *fmt, ...)
 623 {
 624         struct va_format vaf;
 625         va_list args;
 626
 627         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 628                           "EXT4-fs warning"))
 629                 return;
 630
 631         va_start(args, fmt);
 632         vaf.fmt = fmt;
 633         vaf.va = &args;
 634         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 635                sb->s_id, function, line, &vaf);
 636         va_end(args);
 637 }
 638
 639 void __ext4_grp_locked_error(const char *function, unsigned int line,
 640                              struct super_block *sb, ext4_group_t grp,
 641                              unsigned long ino, ext4_fsblk_t block,
 642                              const char *fmt, ...)
 643 __releases(bitlock)
 644 __acquires(bitlock)
 645 {
 646         struct va_format vaf;
 647         va_list args;
 648         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 649
 650         es->s_last_error_ino = cpu_to_le32(ino);
 651         es->s_last_error_block = cpu_to_le64(block);
 652         __save_error_info(sb, function, line);
 653
 654         if (ext4_error_ratelimit(sb)) {
 655                 va_start(args, fmt);
 656                 vaf.fmt = fmt;
 657                 vaf.va = &args;
 658                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 659                        sb->s_id, function, line, grp);
 660                 if (ino)
 661                         printk(KERN_CONT "inode %lu: ", ino);
 662                 if (block)
 663                         printk(KERN_CONT "block %llu:",
 664                                (unsigned long long) block);
 665                 printk(KERN_CONT "%pV\n", &vaf);
 666                 va_end(args);
 667         }
 668
 669         if (test_opt(sb, ERRORS_CONT)) {
 670                 ext4_commit_super(sb, 0);
 671                 return;
 672         }
 673
 674         ext4_unlock_group(sb, grp);
 675         ext4_handle_error(sb);
 676         /*
 677          * We only get here in the ERRORS_RO case; relocking the group
 678          * may be dangerous, but nothing bad will happen since the
 679          * filesystem will have already been marked read/only and the
 680          * journal has been aborted.  We return 1 as a hint to callers
 681          * who might what to use the return value from
 682          * ext4_grp_locked_error() to distinguish between the
 683          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 684          * aggressively from the ext4 function in question, with a
 685          * more appropriate error code.
 686          */
 687         ext4_lock_group(sb, grp);
 688         return;
 689 }
 690
 691 void ext4_update_dynamic_rev(struct super_block *sb)
 692 {
 693         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 694
 695         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 696                 return;
 697
 698         ext4_warning(sb,
 699                      "updating to rev %d because of new feature flag, "
 700                      "running e2fsck is recommended",
 701                      EXT4_DYNAMIC_REV);
 702
 703         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 704         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 705         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 706         /* leave es->s_feature_*compat flags alone */
 707         /* es->s_uuid will be set by e2fsck if empty */
 708
 709         /*
 710          * The rest of the superblock fields should be zero, and if not it
 711          * means they are likely already in use, so leave them alone.  We
 712          * can leave it up to e2fsck to clean up any inconsistencies there.
 713          */
 714 }
 715
 716 /*
 717  * Open the external journal device
 718  */
 719 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 720 {
 721         struct block_device *bdev;
 722         char b[BDEVNAME_SIZE];
 723
 724         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 725         if (IS_ERR(bdev))
 726                 goto fail;
 727         return bdev;
 728
 729 fail:
 730         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 731                         __bdevname(dev, b), PTR_ERR(bdev));
 732         return NULL;
 733 }
 734
 735 /*
 736  * Release the journal device
 737  */
 738 static void ext4_blkdev_put(struct block_device *bdev)
 739 {
 740         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 741 }
 742
 743 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 744 {
 745         struct block_device *bdev;
 746         bdev = sbi->journal_bdev;
 747         if (bdev) {
 748                 ext4_blkdev_put(bdev);
 749                 sbi->journal_bdev = NULL;
 750         }
 751 }
 752
 753 static inline struct inode *orphan_list_entry(struct list_head *l)
 754 {
 755         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 756 }
 757
 758 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 759 {
 760         struct list_head *l;
 761
 762         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 763                  le32_to_cpu(sbi->s_es->s_last_orphan));
 764
 765         printk(KERN_ERR "sb_info orphan list:\n");
 766         list_for_each(l, &sbi->s_orphan) {
 767                 struct inode *inode = orphan_list_entry(l);
 768                 printk(KERN_ERR "  "
 769                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 770                        inode->i_sb->s_id, inode->i_ino, inode,
 771                        inode->i_mode, inode->i_nlink,
 772                        NEXT_ORPHAN(inode));
 773         }
 774 }
 775
 776 static void ext4_put_super(struct super_block *sb)
 777 {
 778         struct ext4_sb_info *sbi = EXT4_SB(sb);
 779         struct ext4_super_block *es = sbi->s_es;
 780         int i, err;
 781
 782         ext4_unregister_li_request(sb);
 783         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 784
 785         flush_workqueue(sbi->rsv_conversion_wq);
 786         destroy_workqueue(sbi->rsv_conversion_wq);
 787
 788         if (sbi->s_journal) {
 789                 err = jbd2_journal_destroy(sbi->s_journal);
 790                 sbi->s_journal = NULL;
 791                 if (err < 0)
 792                         ext4_abort(sb, "Couldn't clean up the journal");
 793         }
 794
 795         ext4_es_unregister_shrinker(sbi);
 796         del_timer_sync(&sbi->s_err_report);
 797         ext4_release_system_zone(sb);
 798         ext4_mb_release(sb);
 799         ext4_ext_release(sb);
 800         ext4_xattr_put_super(sb);
 801
 802         if (!(sb->s_flags & MS_RDONLY)) {
 803                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 804                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 805         }
 806         if (!(sb->s_flags & MS_RDONLY))
 807                 ext4_commit_super(sb, 1);
 808
 809         if (sbi->s_proc) {
 810                 remove_proc_entry("options", sbi->s_proc);
 811                 remove_proc_entry(sb->s_id, ext4_proc_root);
 812         }
 813         kobject_del(&sbi->s_kobj);
 814
 815         for (i = 0; i < sbi->s_gdb_count; i++)
 816                 brelse(sbi->s_group_desc[i]);
 817         ext4_kvfree(sbi->s_group_desc);
 818         ext4_kvfree(sbi->s_flex_groups);
 819         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 820         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 821         percpu_counter_destroy(&sbi->s_dirs_counter);
 822         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 823         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 824         brelse(sbi->s_sbh);
 825 #ifdef CONFIG_QUOTA
 826         for (i = 0; i < MAXQUOTAS; i++)
 827                 kfree(sbi->s_qf_names[i]);
 828 #endif
 829
 830         /* Debugging code just in case the in-memory inode orphan list
 831          * isn't empty.  The on-disk one can be non-empty if we've
 832          * detected an error and taken the fs readonly, but the
 833          * in-memory list had better be clean by this point. */
 834         if (!list_empty(&sbi->s_orphan))
 835                 dump_orphan_list(sb, sbi);
 836         J_ASSERT(list_empty(&sbi->s_orphan));
 837
 838         invalidate_bdev(sb->s_bdev);
 839         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 840                 /*
 841                  * Invalidate the journal device's buffers.  We don't want them
 842                  * floating about in memory - the physical journal device may
 843                  * hotswapped, and it breaks the `ro-after' testing code.
 844                  */
 845                 sync_blockdev(sbi->journal_bdev);
 846                 invalidate_bdev(sbi->journal_bdev);
 847                 ext4_blkdev_remove(sbi);
 848         }
 849         if (sbi->s_mb_cache) {
 850                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 851                 sbi->s_mb_cache = NULL;
 852         }
 853         if (sbi->s_mmp_tsk)
 854                 kthread_stop(sbi->s_mmp_tsk);
 855         sb->s_fs_info = NULL;
 856         /*
 857          * Now that we are completely done shutting down the
 858          * superblock, we need to actually destroy the kobject.
 859          */
 860         kobject_put(&sbi->s_kobj);
 861         wait_for_completion(&sbi->s_kobj_unregister);
 862         if (sbi->s_chksum_driver)
 863                 crypto_free_shash(sbi->s_chksum_driver);
 864         kfree(sbi->s_blockgroup_lock);
 865         kfree(sbi);
 866 }
 867
 868 static struct kmem_cache *ext4_inode_cachep;
 869
 870 /*
 871  * Called inside transaction, so use GFP_NOFS
 872  */
 873 static struct inode *ext4_alloc_inode(struct super_block *sb)
 874 {
 875         struct ext4_inode_info *ei;
 876
 877         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 878         if (!ei)
 879                 return NULL;
 880
 881         ei->vfs_inode.i_version = 1;
 882         spin_lock_init(&ei->i_raw_lock);
 883         INIT_LIST_HEAD(&ei->i_prealloc_list);
 884         spin_lock_init(&ei->i_prealloc_lock);
 885         ext4_es_init_tree(&ei->i_es_tree);
 886         rwlock_init(&ei->i_es_lock);
 887         INIT_LIST_HEAD(&ei->i_es_lru);
 888         ei->i_es_lru_nr = 0;
 889         ei->i_touch_when = 0;
 890         ei->i_reserved_data_blocks = 0;
 891         ei->i_reserved_meta_blocks = 0;
 892         ei->i_allocated_meta_blocks = 0;
 893         ei->i_da_metadata_calc_len = 0;
 894         ei->i_da_metadata_calc_last_lblock = 0;
 895         spin_lock_init(&(ei->i_block_reservation_lock));
 896 #ifdef CONFIG_QUOTA
 897         ei->i_reserved_quota = 0;
 898 #endif
 899         ei->jinode = NULL;
 900         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 901         spin_lock_init(&ei->i_completed_io_lock);
 902         ei->i_sync_tid = 0;
 903         ei->i_datasync_tid = 0;
 904         atomic_set(&ei->i_ioend_count, 0);
 905         atomic_set(&ei->i_unwritten, 0);
 906         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 907
 908         return &ei->vfs_inode;
 909 }
 910
 911 static int ext4_drop_inode(struct inode *inode)
 912 {
 913         int drop = generic_drop_inode(inode);
 914
 915         trace_ext4_drop_inode(inode, drop);
 916         return drop;
 917 }
 918
 919 static void ext4_i_callback(struct rcu_head *head)
 920 {
 921         struct inode *inode = container_of(head, struct inode, i_rcu);
 922         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 923 }
 924
 925 static void ext4_destroy_inode(struct inode *inode)
 926 {
 927         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 928                 ext4_msg(inode->i_sb, KERN_ERR,
 929                          "Inode %lu (%p): orphan list check failed!",
 930                          inode->i_ino, EXT4_I(inode));
 931                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 932                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 933                                 true);
 934                 dump_stack();
 935         }
 936         call_rcu(&inode->i_rcu, ext4_i_callback);
 937 }
 938
 939 static void init_once(void *foo)
 940 {
 941         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 942
 943         INIT_LIST_HEAD(&ei->i_orphan);
 944         init_rwsem(&ei->xattr_sem);
 945         init_rwsem(&ei->i_data_sem);
 946         inode_init_once(&ei->vfs_inode);
 947 }
 948
 949 static int __init init_inodecache(void)
 950 {
 951         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 952                                              sizeof(struct ext4_inode_info),
 953                                              0, (SLAB_RECLAIM_ACCOUNT|
 954                                                 SLAB_MEM_SPREAD),
 955                                              init_once);
 956         if (ext4_inode_cachep == NULL)
 957                 return -ENOMEM;
 958         return 0;
 959 }
 960
 961 static void destroy_inodecache(void)
 962 {
 963         /*
 964          * Make sure all delayed rcu free inodes are flushed before we
 965          * destroy cache.
 966          */
 967         rcu_barrier();
 968         kmem_cache_destroy(ext4_inode_cachep);
 969 }
 970
 971 void ext4_clear_inode(struct inode *inode)
 972 {
 973         invalidate_inode_buffers(inode);
 974         clear_inode(inode);
 975         dquot_drop(inode);
 976         ext4_discard_preallocations(inode);
 977         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 978         ext4_es_lru_del(inode);
 979         if (EXT4_I(inode)->jinode) {
 980                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 981                                                EXT4_I(inode)->jinode);
 982                 jbd2_free_inode(EXT4_I(inode)->jinode);
 983                 EXT4_I(inode)->jinode = NULL;
 984         }
 985 }
 986
 987 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 988                                         u64 ino, u32 generation)
 989 {
 990         struct inode *inode;
 991
 992         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 993                 return ERR_PTR(-ESTALE);
 994         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 995                 return ERR_PTR(-ESTALE);
 996
 997         /* iget isn't really right if the inode is currently unallocated!!
 998          *
 999          * ext4_read_inode will return a bad_inode if the inode had been
1000          * deleted, so we should be safe.
1001          *
1002          * Currently we don't know the generation for parent directory, so
1003          * a generation of 0 means "accept any"
1004          */
1005         inode = ext4_iget(sb, ino);
1006         if (IS_ERR(inode))
1007                 return ERR_CAST(inode);
1008         if (generation && inode->i_generation != generation) {
1009                 iput(inode);
1010                 return ERR_PTR(-ESTALE);
1011         }
1012
1013         return inode;
1014 }
1015
1016 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1017                                         int fh_len, int fh_type)
1018 {
1019         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1020                                     ext4_nfs_get_inode);
1021 }
1022
1023 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1024                                         int fh_len, int fh_type)
1025 {
1026         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1027                                     ext4_nfs_get_inode);
1028 }
1029
1030 /*
1031  * Try to release metadata pages (indirect blocks, directories) which are
1032  * mapped via the block device.  Since these pages could have journal heads
1033  * which would prevent try_to_free_buffers() from freeing them, we must use
1034  * jbd2 layer's try_to_free_buffers() function to release them.
1035  */
1036 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1037                                  gfp_t wait)
1038 {
1039         journal_t *journal = EXT4_SB(sb)->s_journal;
1040
1041         WARN_ON(PageChecked(page));
1042         if (!page_has_buffers(page))
1043                 return 0;
1044         if (journal)
1045                 return jbd2_journal_try_to_free_buffers(journal, page,
1046                                                         wait & ~__GFP_WAIT);
1047         return try_to_free_buffers(page);
1048 }
1049
1050 #ifdef CONFIG_QUOTA
1051 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1052 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1053
1054 static int ext4_write_dquot(struct dquot *dquot);
1055 static int ext4_acquire_dquot(struct dquot *dquot);
1056 static int ext4_release_dquot(struct dquot *dquot);
1057 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1058 static int ext4_write_info(struct super_block *sb, int type);
1059 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1060                          struct path *path);
1061 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1062                                  int format_id);
1063 static int ext4_quota_off(struct super_block *sb, int type);
1064 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1065 static int ext4_quota_on_mount(struct super_block *sb, int type);
1066 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1067                                size_t len, loff_t off);
1068 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1069                                 const char *data, size_t len, loff_t off);
1070 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1071                              unsigned int flags);
1072 static int ext4_enable_quotas(struct super_block *sb);
1073
1074 static const struct dquot_operations ext4_quota_operations = {
1075         .get_reserved_space = ext4_get_reserved_space,
1076         .write_dquot    = ext4_write_dquot,
1077         .acquire_dquot  = ext4_acquire_dquot,
1078         .release_dquot  = ext4_release_dquot,
1079         .mark_dirty     = ext4_mark_dquot_dirty,
1080         .write_info     = ext4_write_info,
1081         .alloc_dquot    = dquot_alloc,
1082         .destroy_dquot  = dquot_destroy,
1083 };
1084
1085 static const struct quotactl_ops ext4_qctl_operations = {
1086         .quota_on       = ext4_quota_on,
1087         .quota_off      = ext4_quota_off,
1088         .quota_sync     = dquot_quota_sync,
1089         .get_info       = dquot_get_dqinfo,
1090         .set_info       = dquot_set_dqinfo,
1091         .get_dqblk      = dquot_get_dqblk,
1092         .set_dqblk      = dquot_set_dqblk
1093 };
1094
1095 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1096         .quota_on_meta  = ext4_quota_on_sysfile,
1097         .quota_off      = ext4_quota_off_sysfile,
1098         .quota_sync     = dquot_quota_sync,
1099         .get_info       = dquot_get_dqinfo,
1100         .set_info       = dquot_set_dqinfo,
1101         .get_dqblk      = dquot_get_dqblk,
1102         .set_dqblk      = dquot_set_dqblk
1103 };
1104 #endif
1105
1106 static const struct super_operations ext4_sops = {
1107         .alloc_inode    = ext4_alloc_inode,
1108         .destroy_inode  = ext4_destroy_inode,
1109         .write_inode    = ext4_write_inode,
1110         .dirty_inode    = ext4_dirty_inode,
1111         .drop_inode     = ext4_drop_inode,
1112         .evict_inode    = ext4_evict_inode,
1113         .put_super      = ext4_put_super,
1114         .sync_fs        = ext4_sync_fs,
1115         .freeze_fs      = ext4_freeze,
1116         .unfreeze_fs    = ext4_unfreeze,
1117         .statfs         = ext4_statfs,
1118         .remount_fs     = ext4_remount,
1119         .show_options   = ext4_show_options,
1120 #ifdef CONFIG_QUOTA
1121         .quota_read     = ext4_quota_read,
1122         .quota_write    = ext4_quota_write,
1123 #endif
1124         .bdev_try_to_free_page = bdev_try_to_free_page,
1125 };
1126
1127 static const struct super_operations ext4_nojournal_sops = {
1128         .alloc_inode    = ext4_alloc_inode,
1129         .destroy_inode  = ext4_destroy_inode,
1130         .write_inode    = ext4_write_inode,
1131         .dirty_inode    = ext4_dirty_inode,
1132         .drop_inode     = ext4_drop_inode,
1133         .evict_inode    = ext4_evict_inode,
1134         .sync_fs        = ext4_sync_fs_nojournal,
1135         .put_super      = ext4_put_super,
1136         .statfs         = ext4_statfs,
1137         .remount_fs     = ext4_remount,
1138         .show_options   = ext4_show_options,
1139 #ifdef CONFIG_QUOTA
1140         .quota_read     = ext4_quota_read,
1141         .quota_write    = ext4_quota_write,
1142 #endif
1143         .bdev_try_to_free_page = bdev_try_to_free_page,
1144 };
1145
1146 static const struct export_operations ext4_export_ops = {
1147         .fh_to_dentry = ext4_fh_to_dentry,
1148         .fh_to_parent = ext4_fh_to_parent,
1149         .get_parent = ext4_get_parent,
1150 };
1151
1152 enum {
1153         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1154         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1155         Opt_nouid32, Opt_debug, Opt_removed,
1156         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1157         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1158         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1159         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1160         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1161         Opt_data_err_abort, Opt_data_err_ignore,
1162         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1163         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1164         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1165         Opt_usrquota, Opt_grpquota, Opt_i_version,
1166         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1167         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1168         Opt_inode_readahead_blks, Opt_journal_ioprio,
1169         Opt_dioread_nolock, Opt_dioread_lock,
1170         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1171         Opt_max_dir_size_kb,
1172 };
1173
1174 static const match_table_t tokens = {
1175         {Opt_bsd_df, "bsddf"},
1176         {Opt_minix_df, "minixdf"},
1177         {Opt_grpid, "grpid"},
1178         {Opt_grpid, "bsdgroups"},
1179         {Opt_nogrpid, "nogrpid"},
1180         {Opt_nogrpid, "sysvgroups"},
1181         {Opt_resgid, "resgid=%u"},
1182         {Opt_resuid, "resuid=%u"},
1183         {Opt_sb, "sb=%u"},
1184         {Opt_err_cont, "errors=continue"},
1185         {Opt_err_panic, "errors=panic"},
1186         {Opt_err_ro, "errors=remount-ro"},
1187         {Opt_nouid32, "nouid32"},
1188         {Opt_debug, "debug"},
1189         {Opt_removed, "oldalloc"},
1190         {Opt_removed, "orlov"},
1191         {Opt_user_xattr, "user_xattr"},
1192         {Opt_nouser_xattr, "nouser_xattr"},
1193         {Opt_acl, "acl"},
1194         {Opt_noacl, "noacl"},
1195         {Opt_noload, "norecovery"},
1196         {Opt_noload, "noload"},
1197         {Opt_removed, "nobh"},
1198         {Opt_removed, "bh"},
1199         {Opt_commit, "commit=%u"},
1200         {Opt_min_batch_time, "min_batch_time=%u"},
1201         {Opt_max_batch_time, "max_batch_time=%u"},
1202         {Opt_journal_dev, "journal_dev=%u"},
1203         {Opt_journal_path, "journal_path=%s"},
1204         {Opt_journal_checksum, "journal_checksum"},
1205         {Opt_journal_async_commit, "journal_async_commit"},
1206         {Opt_abort, "abort"},
1207         {Opt_data_journal, "data=journal"},
1208         {Opt_data_ordered, "data=ordered"},
1209         {Opt_data_writeback, "data=writeback"},
1210         {Opt_data_err_abort, "data_err=abort"},
1211         {Opt_data_err_ignore, "data_err=ignore"},
1212         {Opt_offusrjquota, "usrjquota="},
1213         {Opt_usrjquota, "usrjquota=%s"},
1214         {Opt_offgrpjquota, "grpjquota="},
1215         {Opt_grpjquota, "grpjquota=%s"},
1216         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1217         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1218         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1219         {Opt_grpquota, "grpquota"},
1220         {Opt_noquota, "noquota"},
1221         {Opt_quota, "quota"},
1222         {Opt_usrquota, "usrquota"},
1223         {Opt_barrier, "barrier=%u"},
1224         {Opt_barrier, "barrier"},
1225         {Opt_nobarrier, "nobarrier"},
1226         {Opt_i_version, "i_version"},
1227         {Opt_stripe, "stripe=%u"},
1228         {Opt_delalloc, "delalloc"},
1229         {Opt_nodelalloc, "nodelalloc"},
1230         {Opt_removed, "mblk_io_submit"},
1231         {Opt_removed, "nomblk_io_submit"},
1232         {Opt_block_validity, "block_validity"},
1233         {Opt_noblock_validity, "noblock_validity"},
1234         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1235         {Opt_journal_ioprio, "journal_ioprio=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1237         {Opt_auto_da_alloc, "auto_da_alloc"},
1238         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1239         {Opt_dioread_nolock, "dioread_nolock"},
1240         {Opt_dioread_lock, "dioread_lock"},
1241         {Opt_discard, "discard"},
1242         {Opt_nodiscard, "nodiscard"},
1243         {Opt_init_itable, "init_itable=%u"},
1244         {Opt_init_itable, "init_itable"},
1245         {Opt_noinit_itable, "noinit_itable"},
1246         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1247         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1248         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1249         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1250         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1251         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1252         {Opt_err, NULL},
1253 };
1254
1255 static ext4_fsblk_t get_sb_block(void **data)
1256 {
1257         ext4_fsblk_t    sb_block;
1258         char            *options = (char *) *data;
1259
1260         if (!options || strncmp(options, "sb=", 3) != 0)
1261                 return 1;       /* Default location */
1262
1263         options += 3;
1264         /* TODO: use simple_strtoll with >32bit ext4 */
1265         sb_block = simple_strtoul(options, &options, 0);
1266         if (*options && *options != ',') {
1267                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1268                        (char *) *data);
1269                 return 1;
1270         }
1271         if (*options == ',')
1272                 options++;
1273         *data = (void *) options;
1274
1275         return sb_block;
1276 }
1277
1278 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1279 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1280         "Contact [email protected] if you think we should keep it.\n";
1281
1282 #ifdef CONFIG_QUOTA
1283 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1284 {
1285         struct ext4_sb_info *sbi = EXT4_SB(sb);
1286         char *qname;
1287         int ret = -1;
1288
1289         if (sb_any_quota_loaded(sb) &&
1290                 !sbi->s_qf_names[qtype]) {
1291                 ext4_msg(sb, KERN_ERR,
1292                         "Cannot change journaled "
1293                         "quota options when quota turned on");
1294                 return -1;
1295         }
1296         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1297                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1298                          "when QUOTA feature is enabled");
1299                 return -1;
1300         }
1301         qname = match_strdup(args);
1302         if (!qname) {
1303                 ext4_msg(sb, KERN_ERR,
1304                         "Not enough memory for storing quotafile name");
1305                 return -1;
1306         }
1307         if (sbi->s_qf_names[qtype]) {
1308                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1309                         ret = 1;
1310                 else
1311                         ext4_msg(sb, KERN_ERR,
1312                                  "%s quota file already specified",
1313                                  QTYPE2NAME(qtype));
1314                 goto errout;
1315         }
1316         if (strchr(qname, '/')) {
1317                 ext4_msg(sb, KERN_ERR,
1318                         "quotafile must be on filesystem root");
1319                 goto errout;
1320         }
1321         sbi->s_qf_names[qtype] = qname;
1322         set_opt(sb, QUOTA);
1323         return 1;
1324 errout:
1325         kfree(qname);
1326         return ret;
1327 }
1328
1329 static int clear_qf_name(struct super_block *sb, int qtype)
1330 {
1331
1332         struct ext4_sb_info *sbi = EXT4_SB(sb);
1333
1334         if (sb_any_quota_loaded(sb) &&
1335                 sbi->s_qf_names[qtype]) {
1336                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1337                         " when quota turned on");
1338                 return -1;
1339         }
1340         kfree(sbi->s_qf_names[qtype]);
1341         sbi->s_qf_names[qtype] = NULL;
1342         return 1;
1343 }
1344 #endif
1345
1346 #define MOPT_SET        0x0001
1347 #define MOPT_CLEAR      0x0002
1348 #define MOPT_NOSUPPORT  0x0004
1349 #define MOPT_EXPLICIT   0x0008
1350 #define MOPT_CLEAR_ERR  0x0010
1351 #define MOPT_GTE0       0x0020
1352 #ifdef CONFIG_QUOTA
1353 #define MOPT_Q          0
1354 #define MOPT_QFMT       0x0040
1355 #else
1356 #define MOPT_Q          MOPT_NOSUPPORT
1357 #define MOPT_QFMT       MOPT_NOSUPPORT
1358 #endif
1359 #define MOPT_DATAJ      0x0080
1360 #define MOPT_NO_EXT2    0x0100
1361 #define MOPT_NO_EXT3    0x0200
1362 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1363 #define MOPT_STRING     0x0400
1364
1365 static const struct mount_opts {
1366         int     token;
1367         int     mount_opt;
1368         int     flags;
1369 } ext4_mount_opts[] = {
1370         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1371         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1372         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1373         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1374         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1375         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1376         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1377          MOPT_EXT4_ONLY | MOPT_SET},
1378         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1379          MOPT_EXT4_ONLY | MOPT_CLEAR},
1380         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1381         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1382         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1383          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1384         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1385          MOPT_EXT4_ONLY | MOPT_CLEAR},
1386         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1387          MOPT_EXT4_ONLY | MOPT_SET},
1388         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1389                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1390          MOPT_EXT4_ONLY | MOPT_SET},
1391         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1392         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1393         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1394         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1396          MOPT_NO_EXT2 | MOPT_SET},
1397         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1398          MOPT_NO_EXT2 | MOPT_CLEAR},
1399         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1400         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1401         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1402         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1403         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1404         {Opt_commit, 0, MOPT_GTE0},
1405         {Opt_max_batch_time, 0, MOPT_GTE0},
1406         {Opt_min_batch_time, 0, MOPT_GTE0},
1407         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1408         {Opt_init_itable, 0, MOPT_GTE0},
1409         {Opt_stripe, 0, MOPT_GTE0},
1410         {Opt_resuid, 0, MOPT_GTE0},
1411         {Opt_resgid, 0, MOPT_GTE0},
1412         {Opt_journal_dev, 0, MOPT_GTE0},
1413         {Opt_journal_path, 0, MOPT_STRING},
1414         {Opt_journal_ioprio, 0, MOPT_GTE0},
1415         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1416         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1417         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1418          MOPT_NO_EXT2 | MOPT_DATAJ},
1419         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1420         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1421 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1422         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1423         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1424 #else
1425         {Opt_acl, 0, MOPT_NOSUPPORT},
1426         {Opt_noacl, 0, MOPT_NOSUPPORT},
1427 #endif
1428         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1429         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1430         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1431         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1432                                                         MOPT_SET | MOPT_Q},
1433         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1434                                                         MOPT_SET | MOPT_Q},
1435         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1436                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1437         {Opt_usrjquota, 0, MOPT_Q},
1438         {Opt_grpjquota, 0, MOPT_Q},
1439         {Opt_offusrjquota, 0, MOPT_Q},
1440         {Opt_offgrpjquota, 0, MOPT_Q},
1441         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1442         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1443         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1444         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1445         {Opt_err, 0, 0}
1446 };
1447
1448 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1449                             substring_t *args, unsigned long *journal_devnum,
1450                             unsigned int *journal_ioprio, int is_remount)
1451 {
1452         struct ext4_sb_info *sbi = EXT4_SB(sb);
1453         const struct mount_opts *m;
1454         kuid_t uid;
1455         kgid_t gid;
1456         int arg = 0;
1457
1458 #ifdef CONFIG_QUOTA
1459         if (token == Opt_usrjquota)
1460                 return set_qf_name(sb, USRQUOTA, &args[0]);
1461         else if (token == Opt_grpjquota)
1462                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1463         else if (token == Opt_offusrjquota)
1464                 return clear_qf_name(sb, USRQUOTA);
1465         else if (token == Opt_offgrpjquota)
1466                 return clear_qf_name(sb, GRPQUOTA);
1467 #endif
1468         switch (token) {
1469         case Opt_noacl:
1470         case Opt_nouser_xattr:
1471                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1472                 break;
1473         case Opt_sb:
1474                 return 1;       /* handled by get_sb_block() */
1475         case Opt_removed:
1476                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1477                 return 1;
1478         case Opt_abort:
1479                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1480                 return 1;
1481         case Opt_i_version:
1482                 sb->s_flags |= MS_I_VERSION;
1483                 return 1;
1484         }
1485
1486         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1487                 if (token == m->token)
1488                         break;
1489
1490         if (m->token == Opt_err) {
1491                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1492                          "or missing value", opt);
1493                 return -1;
1494         }
1495
1496         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1497                 ext4_msg(sb, KERN_ERR,
1498                          "Mount option \"%s\" incompatible with ext2", opt);
1499                 return -1;
1500         }
1501         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1502                 ext4_msg(sb, KERN_ERR,
1503                          "Mount option \"%s\" incompatible with ext3", opt);
1504                 return -1;
1505         }
1506
1507         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1508                 return -1;
1509         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1510                 return -1;
1511         if (m->flags & MOPT_EXPLICIT)
1512                 set_opt2(sb, EXPLICIT_DELALLOC);
1513         if (m->flags & MOPT_CLEAR_ERR)
1514                 clear_opt(sb, ERRORS_MASK);
1515         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1516                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1517                          "options when quota turned on");
1518                 return -1;
1519         }
1520
1521         if (m->flags & MOPT_NOSUPPORT) {
1522                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1523         } else if (token == Opt_commit) {
1524                 if (arg == 0)
1525                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1526                 sbi->s_commit_interval = HZ * arg;
1527         } else if (token == Opt_max_batch_time) {
1528                 sbi->s_max_batch_time = arg;
1529         } else if (token == Opt_min_batch_time) {
1530                 sbi->s_min_batch_time = arg;
1531         } else if (token == Opt_inode_readahead_blks) {
1532                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1533                         ext4_msg(sb, KERN_ERR,
1534                                  "EXT4-fs: inode_readahead_blks must be "
1535                                  "0 or a power of 2 smaller than 2^31");
1536                         return -1;
1537                 }
1538                 sbi->s_inode_readahead_blks = arg;
1539         } else if (token == Opt_init_itable) {
1540                 set_opt(sb, INIT_INODE_TABLE);
1541                 if (!args->from)
1542                         arg = EXT4_DEF_LI_WAIT_MULT;
1543                 sbi->s_li_wait_mult = arg;
1544         } else if (token == Opt_max_dir_size_kb) {
1545                 sbi->s_max_dir_size_kb = arg;
1546         } else if (token == Opt_stripe) {
1547                 sbi->s_stripe = arg;
1548         } else if (token == Opt_resuid) {
1549                 uid = make_kuid(current_user_ns(), arg);
1550                 if (!uid_valid(uid)) {
1551                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1552                         return -1;
1553                 }
1554                 sbi->s_resuid = uid;
1555         } else if (token == Opt_resgid) {
1556                 gid = make_kgid(current_user_ns(), arg);
1557                 if (!gid_valid(gid)) {
1558                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1559                         return -1;
1560                 }
1561                 sbi->s_resgid = gid;
1562         } else if (token == Opt_journal_dev) {
1563                 if (is_remount) {
1564                         ext4_msg(sb, KERN_ERR,
1565                                  "Cannot specify journal on remount");
1566                         return -1;
1567                 }
1568                 *journal_devnum = arg;
1569         } else if (token == Opt_journal_path) {
1570                 char *journal_path;
1571                 struct inode *journal_inode;
1572                 struct path path;
1573                 int error;
1574
1575                 if (is_remount) {
1576                         ext4_msg(sb, KERN_ERR,
1577                                  "Cannot specify journal on remount");
1578                         return -1;
1579                 }
1580                 journal_path = match_strdup(&args[0]);
1581                 if (!journal_path) {
1582                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1583                                 "journal device string");
1584                         return -1;
1585                 }
1586
1587                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1588                 if (error) {
1589                         ext4_msg(sb, KERN_ERR, "error: could not find "
1590                                 "journal device path: error %d", error);
1591                         kfree(journal_path);
1592                         return -1;
1593                 }
1594
1595                 journal_inode = path.dentry->d_inode;
1596                 if (!S_ISBLK(journal_inode->i_mode)) {
1597                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1598                                 "is not a block device", journal_path);
1599                         path_put(&path);
1600                         kfree(journal_path);
1601                         return -1;
1602                 }
1603
1604                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1605                 path_put(&path);
1606                 kfree(journal_path);
1607         } else if (token == Opt_journal_ioprio) {
1608                 if (arg > 7) {
1609                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1610                                  " (must be 0-7)");
1611                         return -1;
1612                 }
1613                 *journal_ioprio =
1614                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1615         } else if (m->flags & MOPT_DATAJ) {
1616                 if (is_remount) {
1617                         if (!sbi->s_journal)
1618                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1619                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1620                                 ext4_msg(sb, KERN_ERR,
1621                                          "Cannot change data mode on remount");
1622                                 return -1;
1623                         }
1624                 } else {
1625                         clear_opt(sb, DATA_FLAGS);
1626                         sbi->s_mount_opt |= m->mount_opt;
1627                 }
1628 #ifdef CONFIG_QUOTA
1629         } else if (m->flags & MOPT_QFMT) {
1630                 if (sb_any_quota_loaded(sb) &&
1631                     sbi->s_jquota_fmt != m->mount_opt) {
1632                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1633                                  "quota options when quota turned on");
1634                         return -1;
1635                 }
1636                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1637                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1638                         ext4_msg(sb, KERN_ERR,
1639                                  "Cannot set journaled quota options "
1640                                  "when QUOTA feature is enabled");
1641                         return -1;
1642                 }
1643                 sbi->s_jquota_fmt = m->mount_opt;
1644 #endif
1645         } else {
1646                 if (!args->from)
1647                         arg = 1;
1648                 if (m->flags & MOPT_CLEAR)
1649                         arg = !arg;
1650                 else if (unlikely(!(m->flags & MOPT_SET))) {
1651                         ext4_msg(sb, KERN_WARNING,
1652                                  "buggy handling of option %s", opt);
1653                         WARN_ON(1);
1654                         return -1;
1655                 }
1656                 if (arg != 0)
1657                         sbi->s_mount_opt |= m->mount_opt;
1658                 else
1659                         sbi->s_mount_opt &= ~m->mount_opt;
1660         }
1661         return 1;
1662 }
1663
1664 static int parse_options(char *options, struct super_block *sb,
1665                          unsigned long *journal_devnum,
1666                          unsigned int *journal_ioprio,
1667                          int is_remount)
1668 {
1669         struct ext4_sb_info *sbi = EXT4_SB(sb);
1670         char *p;
1671         substring_t args[MAX_OPT_ARGS];
1672         int token;
1673
1674         if (!options)
1675                 return 1;
1676
1677         while ((p = strsep(&options, ",")) != NULL) {
1678                 if (!*p)
1679                         continue;
1680                 /*
1681                  * Initialize args struct so we know whether arg was
1682                  * found; some options take optional arguments.
1683                  */
1684                 args[0].to = args[0].from = NULL;
1685                 token = match_token(p, tokens, args);
1686                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1687                                      journal_ioprio, is_remount) < 0)
1688                         return 0;
1689         }
1690 #ifdef CONFIG_QUOTA
1691         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1692             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1693                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1694                          "feature is enabled");
1695                 return 0;
1696         }
1697         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1698                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1699                         clear_opt(sb, USRQUOTA);
1700
1701                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1702                         clear_opt(sb, GRPQUOTA);
1703
1704                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1705                         ext4_msg(sb, KERN_ERR, "old and new quota "
1706                                         "format mixing");
1707                         return 0;
1708                 }
1709
1710                 if (!sbi->s_jquota_fmt) {
1711                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1712                                         "not specified");
1713                         return 0;
1714                 }
1715         } else {
1716                 if (sbi->s_jquota_fmt) {
1717                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1718                                         "specified with no journaling "
1719                                         "enabled");
1720                         return 0;
1721                 }
1722         }
1723 #endif
1724         if (test_opt(sb, DIOREAD_NOLOCK)) {
1725                 int blocksize =
1726                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1727
1728                 if (blocksize < PAGE_CACHE_SIZE) {
1729                         ext4_msg(sb, KERN_ERR, "can't mount with "
1730                                  "dioread_nolock if block size != PAGE_SIZE");
1731                         return 0;
1732                 }
1733         }
1734         return 1;
1735 }
1736
1737 static inline void ext4_show_quota_options(struct seq_file *seq,
1738                                            struct super_block *sb)
1739 {
1740 #if defined(CONFIG_QUOTA)
1741         struct ext4_sb_info *sbi = EXT4_SB(sb);
1742
1743         if (sbi->s_jquota_fmt) {
1744                 char *fmtname = "";
1745
1746                 switch (sbi->s_jquota_fmt) {
1747                 case QFMT_VFS_OLD:
1748                         fmtname = "vfsold";
1749                         break;
1750                 case QFMT_VFS_V0:
1751                         fmtname = "vfsv0";
1752                         break;
1753                 case QFMT_VFS_V1:
1754                         fmtname = "vfsv1";
1755                         break;
1756                 }
1757                 seq_printf(seq, ",jqfmt=%s", fmtname);
1758         }
1759
1760         if (sbi->s_qf_names[USRQUOTA])
1761                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1762
1763         if (sbi->s_qf_names[GRPQUOTA])
1764                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1765 #endif
1766 }
1767
1768 static const char *token2str(int token)
1769 {
1770         const struct match_token *t;
1771
1772         for (t = tokens; t->token != Opt_err; t++)
1773                 if (t->token == token && !strchr(t->pattern, '='))
1774                         break;
1775         return t->pattern;
1776 }
1777
1778 /*
1779  * Show an option if
1780  *  - it's set to a non-default value OR
1781  *  - if the per-sb default is different from the global default
1782  */
1783 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1784                               int nodefs)
1785 {
1786         struct ext4_sb_info *sbi = EXT4_SB(sb);
1787         struct ext4_super_block *es = sbi->s_es;
1788         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1789         const struct mount_opts *m;
1790         char sep = nodefs ? '\n' : ',';
1791
1792 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1793 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1794
1795         if (sbi->s_sb_block != 1)
1796                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1797
1798         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1799                 int want_set = m->flags & MOPT_SET;
1800                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1801                     (m->flags & MOPT_CLEAR_ERR))
1802                         continue;
1803                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1804                         continue; /* skip if same as the default */
1805                 if ((want_set &&
1806                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1807                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1808                         continue; /* select Opt_noFoo vs Opt_Foo */
1809                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1810         }
1811
1812         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1813             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1814                 SEQ_OPTS_PRINT("resuid=%u",
1815                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1816         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1817             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1818                 SEQ_OPTS_PRINT("resgid=%u",
1819                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1820         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1821         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1822                 SEQ_OPTS_PUTS("errors=remount-ro");
1823         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1824                 SEQ_OPTS_PUTS("errors=continue");
1825         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1826                 SEQ_OPTS_PUTS("errors=panic");
1827         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1828                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1829         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1830                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1831         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1832                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1833         if (sb->s_flags & MS_I_VERSION)
1834                 SEQ_OPTS_PUTS("i_version");
1835         if (nodefs || sbi->s_stripe)
1836                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1837         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1838                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1839                         SEQ_OPTS_PUTS("data=journal");
1840                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1841                         SEQ_OPTS_PUTS("data=ordered");
1842                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1843                         SEQ_OPTS_PUTS("data=writeback");
1844         }
1845         if (nodefs ||
1846             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1847                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1848                                sbi->s_inode_readahead_blks);
1849
1850         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1851                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1852                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1853         if (nodefs || sbi->s_max_dir_size_kb)
1854                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1855
1856         ext4_show_quota_options(seq, sb);
1857         return 0;
1858 }
1859
1860 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1861 {
1862         return _ext4_show_options(seq, root->d_sb, 0);
1863 }
1864
1865 static int options_seq_show(struct seq_file *seq, void *offset)
1866 {
1867         struct super_block *sb = seq->private;
1868         int rc;
1869
1870         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1871         rc = _ext4_show_options(seq, sb, 1);
1872         seq_puts(seq, "\n");
1873         return rc;
1874 }
1875
1876 static int options_open_fs(struct inode *inode, struct file *file)
1877 {
1878         return single_open(file, options_seq_show, PDE_DATA(inode));
1879 }
1880
1881 static const struct file_operations ext4_seq_options_fops = {
1882         .owner = THIS_MODULE,
1883         .open = options_open_fs,
1884         .read = seq_read,
1885         .llseek = seq_lseek,
1886         .release = single_release,
1887 };
1888
1889 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1890                             int read_only)
1891 {
1892         struct ext4_sb_info *sbi = EXT4_SB(sb);
1893         int res = 0;
1894
1895         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1896                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1897                          "forcing read-only mode");
1898                 res = MS_RDONLY;
1899         }
1900         if (read_only)
1901                 goto done;
1902         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1903                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1904                          "running e2fsck is recommended");
1905         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1906                 ext4_msg(sb, KERN_WARNING,
1907                          "warning: mounting fs with errors, "
1908                          "running e2fsck is recommended");
1909         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1910                  le16_to_cpu(es->s_mnt_count) >=
1911                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1912                 ext4_msg(sb, KERN_WARNING,
1913                          "warning: maximal mount count reached, "
1914                          "running e2fsck is recommended");
1915         else if (le32_to_cpu(es->s_checkinterval) &&
1916                 (le32_to_cpu(es->s_lastcheck) +
1917                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1918                 ext4_msg(sb, KERN_WARNING,
1919                          "warning: checktime reached, "
1920                          "running e2fsck is recommended");
1921         if (!sbi->s_journal)
1922                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1923         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1924                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1925         le16_add_cpu(&es->s_mnt_count, 1);
1926         es->s_mtime = cpu_to_le32(get_seconds());
1927         ext4_update_dynamic_rev(sb);
1928         if (sbi->s_journal)
1929                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1930
1931         ext4_commit_super(sb, 1);
1932 done:
1933         if (test_opt(sb, DEBUG))
1934                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1935                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1936                         sb->s_blocksize,
1937                         sbi->s_groups_count,
1938                         EXT4_BLOCKS_PER_GROUP(sb),
1939                         EXT4_INODES_PER_GROUP(sb),
1940                         sbi->s_mount_opt, sbi->s_mount_opt2);
1941
1942         cleancache_init_fs(sb);
1943         return res;
1944 }
1945
1946 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1947 {
1948         struct ext4_sb_info *sbi = EXT4_SB(sb);
1949         struct flex_groups *new_groups;
1950         int size;
1951
1952         if (!sbi->s_log_groups_per_flex)
1953                 return 0;
1954
1955         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1956         if (size <= sbi->s_flex_groups_allocated)
1957                 return 0;
1958
1959         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1960         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1961         if (!new_groups) {
1962                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1963                          size / (int) sizeof(struct flex_groups));
1964                 return -ENOMEM;
1965         }
1966
1967         if (sbi->s_flex_groups) {
1968                 memcpy(new_groups, sbi->s_flex_groups,
1969                        (sbi->s_flex_groups_allocated *
1970                         sizeof(struct flex_groups)));
1971                 ext4_kvfree(sbi->s_flex_groups);
1972         }
1973         sbi->s_flex_groups = new_groups;
1974         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1975         return 0;
1976 }
1977
1978 static int ext4_fill_flex_info(struct super_block *sb)
1979 {
1980         struct ext4_sb_info *sbi = EXT4_SB(sb);
1981         struct ext4_group_desc *gdp = NULL;
1982         ext4_group_t flex_group;
1983         int i, err;
1984
1985         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1986         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1987                 sbi->s_log_groups_per_flex = 0;
1988                 return 1;
1989         }
1990
1991         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1992         if (err)
1993                 goto failed;
1994
1995         for (i = 0; i < sbi->s_groups_count; i++) {
1996                 gdp = ext4_get_group_desc(sb, i, NULL);
1997
1998                 flex_group = ext4_flex_group(sbi, i);
1999                 atomic_add(ext4_free_inodes_count(sb, gdp),
2000                            &sbi->s_flex_groups[flex_group].free_inodes);
2001                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2002                              &sbi->s_flex_groups[flex_group].free_clusters);
2003                 atomic_add(ext4_used_dirs_count(sb, gdp),
2004                            &sbi->s_flex_groups[flex_group].used_dirs);
2005         }
2006
2007         return 1;
2008 failed:
2009         return 0;
2010 }
2011
2012 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2013                                    struct ext4_group_desc *gdp)
2014 {
2015         int offset;
2016         __u16 crc = 0;
2017         __le32 le_group = cpu_to_le32(block_group);
2018
2019         if ((sbi->s_es->s_feature_ro_compat &
2020              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2021                 /* Use new metadata_csum algorithm */
2022                 __le16 save_csum;
2023                 __u32 csum32;
2024
2025                 save_csum = gdp->bg_checksum;
2026                 gdp->bg_checksum = 0;
2027                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2028                                      sizeof(le_group));
2029                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2030                                      sbi->s_desc_size);
2031                 gdp->bg_checksum = save_csum;
2032
2033                 crc = csum32 & 0xFFFF;
2034                 goto out;
2035         }
2036
2037         /* old crc16 code */
2038         offset = offsetof(struct ext4_group_desc, bg_checksum);
2039
2040         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2041         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2042         crc = crc16(crc, (__u8 *)gdp, offset);
2043         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2044         /* for checksum of struct ext4_group_desc do the rest...*/
2045         if ((sbi->s_es->s_feature_incompat &
2046              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2047             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2048                 crc = crc16(crc, (__u8 *)gdp + offset,
2049                             le16_to_cpu(sbi->s_es->s_desc_size) -
2050                                 offset);
2051
2052 out:
2053         return cpu_to_le16(crc);
2054 }
2055
2056 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2057                                 struct ext4_group_desc *gdp)
2058 {
2059         if (ext4_has_group_desc_csum(sb) &&
2060             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2061                                                       block_group, gdp)))
2062                 return 0;
2063
2064         return 1;
2065 }
2066
2067 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2068                               struct ext4_group_desc *gdp)
2069 {
2070         if (!ext4_has_group_desc_csum(sb))
2071                 return;
2072         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2073 }
2074
2075 /* Called at mount-time, super-block is locked */
2076 static int ext4_check_descriptors(struct super_block *sb,
2077                                   ext4_group_t *first_not_zeroed)
2078 {
2079         struct ext4_sb_info *sbi = EXT4_SB(sb);
2080         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2081         ext4_fsblk_t last_block;
2082         ext4_fsblk_t block_bitmap;
2083         ext4_fsblk_t inode_bitmap;
2084         ext4_fsblk_t inode_table;
2085         int flexbg_flag = 0;
2086         ext4_group_t i, grp = sbi->s_groups_count;
2087
2088         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2089                 flexbg_flag = 1;
2090
2091         ext4_debug("Checking group descriptors");
2092
2093         for (i = 0; i < sbi->s_groups_count; i++) {
2094                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2095
2096                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2097                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2098                 else
2099                         last_block = first_block +
2100                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2101
2102                 if ((grp == sbi->s_groups_count) &&
2103                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2104                         grp = i;
2105
2106                 block_bitmap = ext4_block_bitmap(sb, gdp);
2107                 if (block_bitmap < first_block || block_bitmap > last_block) {
2108                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2109                                "Block bitmap for group %u not in group "
2110                                "(block %llu)!", i, block_bitmap);
2111                         return 0;
2112                 }
2113                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2114                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2115                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2116                                "Inode bitmap for group %u not in group "
2117                                "(block %llu)!", i, inode_bitmap);
2118                         return 0;
2119                 }
2120                 inode_table = ext4_inode_table(sb, gdp);
2121                 if (inode_table < first_block ||
2122                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2123                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2124                                "Inode table for group %u not in group "
2125                                "(block %llu)!", i, inode_table);
2126                         return 0;
2127                 }
2128                 ext4_lock_group(sb, i);
2129                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2130                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2131                                  "Checksum for group %u failed (%u!=%u)",
2132                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2133                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2134                         if (!(sb->s_flags & MS_RDONLY)) {
2135                                 ext4_unlock_group(sb, i);
2136                                 return 0;
2137                         }
2138                 }
2139                 ext4_unlock_group(sb, i);
2140                 if (!flexbg_flag)
2141                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2142         }
2143         if (NULL != first_not_zeroed)
2144                 *first_not_zeroed = grp;
2145         return 1;
2146 }
2147
2148 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2149  * the superblock) which were deleted from all directories, but held open by
2150  * a process at the time of a crash.  We walk the list and try to delete these
2151  * inodes at recovery time (only with a read-write filesystem).
2152  *
2153  * In order to keep the orphan inode chain consistent during traversal (in
2154  * case of crash during recovery), we link each inode into the superblock
2155  * orphan list_head and handle it the same way as an inode deletion during
2156  * normal operation (which journals the operations for us).
2157  *
2158  * We only do an iget() and an iput() on each inode, which is very safe if we
2159  * accidentally point at an in-use or already deleted inode.  The worst that
2160  * can happen in this case is that we get a "bit already cleared" message from
2161  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2162  * e2fsck was run on this filesystem, and it must have already done the orphan
2163  * inode cleanup for us, so we can safely abort without any further action.
2164  */
2165 static void ext4_orphan_cleanup(struct super_block *sb,
2166                                 struct ext4_super_block *es)
2167 {
2168         unsigned int s_flags = sb->s_flags;
2169         int nr_orphans = 0, nr_truncates = 0;
2170 #ifdef CONFIG_QUOTA
2171         int i;
2172 #endif
2173         if (!es->s_last_orphan) {
2174                 jbd_debug(4, "no orphan inodes to clean up\n");
2175                 return;
2176         }
2177
2178         if (bdev_read_only(sb->s_bdev)) {
2179                 ext4_msg(sb, KERN_ERR, "write access "
2180                         "unavailable, skipping orphan cleanup");
2181                 return;
2182         }
2183
2184         /* Check if feature set would not allow a r/w mount */
2185         if (!ext4_feature_set_ok(sb, 0)) {
2186                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2187                          "unknown ROCOMPAT features");
2188                 return;
2189         }
2190
2191         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2192                 /* don't clear list on RO mount w/ errors */
2193                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2194                         jbd_debug(1, "Errors on filesystem, "
2195                                   "clearing orphan list.\n");
2196                         es->s_last_orphan = 0;
2197                 }
2198                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2199                 return;
2200         }
2201
2202         if (s_flags & MS_RDONLY) {
2203                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2204                 sb->s_flags &= ~MS_RDONLY;
2205         }
2206 #ifdef CONFIG_QUOTA
2207         /* Needed for iput() to work correctly and not trash data */
2208         sb->s_flags |= MS_ACTIVE;
2209         /* Turn on quotas so that they are updated correctly */
2210         for (i = 0; i < MAXQUOTAS; i++) {
2211                 if (EXT4_SB(sb)->s_qf_names[i]) {
2212                         int ret = ext4_quota_on_mount(sb, i);
2213                         if (ret < 0)
2214                                 ext4_msg(sb, KERN_ERR,
2215                                         "Cannot turn on journaled "
2216                                         "quota: error %d", ret);
2217                 }
2218         }
2219 #endif
2220
2221         while (es->s_last_orphan) {
2222                 struct inode *inode;
2223
2224                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2225                 if (IS_ERR(inode)) {
2226                         es->s_last_orphan = 0;
2227                         break;
2228                 }
2229
2230                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2231                 dquot_initialize(inode);
2232                 if (inode->i_nlink) {
2233                         if (test_opt(sb, DEBUG))
2234                                 ext4_msg(sb, KERN_DEBUG,
2235                                         "%s: truncating inode %lu to %lld bytes",
2236                                         __func__, inode->i_ino, inode->i_size);
2237                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2238                                   inode->i_ino, inode->i_size);
2239                         mutex_lock(&inode->i_mutex);
2240                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2241                         ext4_truncate(inode);
2242                         mutex_unlock(&inode->i_mutex);
2243                         nr_truncates++;
2244                 } else {
2245                         if (test_opt(sb, DEBUG))
2246                                 ext4_msg(sb, KERN_DEBUG,
2247                                         "%s: deleting unreferenced inode %lu",
2248                                         __func__, inode->i_ino);
2249                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2250                                   inode->i_ino);
2251                         nr_orphans++;
2252                 }
2253                 iput(inode);  /* The delete magic happens here! */
2254         }
2255
2256 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2257
2258         if (nr_orphans)
2259                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2260                        PLURAL(nr_orphans));
2261         if (nr_truncates)
2262                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2263                        PLURAL(nr_truncates));
2264 #ifdef CONFIG_QUOTA
2265         /* Turn quotas off */
2266         for (i = 0; i < MAXQUOTAS; i++) {
2267                 if (sb_dqopt(sb)->files[i])
2268                         dquot_quota_off(sb, i);
2269         }
2270 #endif
2271         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2272 }
2273
2274 /*
2275  * Maximal extent format file size.
2276  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2277  * extent format containers, within a sector_t, and within i_blocks
2278  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2279  * so that won't be a limiting factor.
2280  *
2281  * However there is other limiting factor. We do store extents in the form
2282  * of starting block and length, hence the resulting length of the extent
2283  * covering maximum file size must fit into on-disk format containers as
2284  * well. Given that length is always by 1 unit bigger than max unit (because
2285  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2286  *
2287  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2288  */
2289 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2290 {
2291         loff_t res;
2292         loff_t upper_limit = MAX_LFS_FILESIZE;
2293
2294         /* small i_blocks in vfs inode? */
2295         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2296                 /*
2297                  * CONFIG_LBDAF is not enabled implies the inode
2298                  * i_block represent total blocks in 512 bytes
2299                  * 32 == size of vfs inode i_blocks * 8
2300                  */
2301                 upper_limit = (1LL << 32) - 1;
2302
2303                 /* total blocks in file system block size */
2304                 upper_limit >>= (blkbits - 9);
2305                 upper_limit <<= blkbits;
2306         }
2307
2308         /*
2309          * 32-bit extent-start container, ee_block. We lower the maxbytes
2310          * by one fs block, so ee_len can cover the extent of maximum file
2311          * size
2312          */
2313         res = (1LL << 32) - 1;
2314         res <<= blkbits;
2315
2316         /* Sanity check against vm- & vfs- imposed limits */
2317         if (res > upper_limit)
2318                 res = upper_limit;
2319
2320         return res;
2321 }
2322
2323 /*
2324  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2325  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2326  * We need to be 1 filesystem block less than the 2^48 sector limit.
2327  */
2328 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2329 {
2330         loff_t res = EXT4_NDIR_BLOCKS;
2331         int meta_blocks;
2332         loff_t upper_limit;
2333         /* This is calculated to be the largest file size for a dense, block
2334          * mapped file such that the file's total number of 512-byte sectors,
2335          * including data and all indirect blocks, does not exceed (2^48 - 1).
2336          *
2337          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2338          * number of 512-byte sectors of the file.
2339          */
2340
2341         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2342                 /*
2343                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2344                  * the inode i_block field represents total file blocks in
2345                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2346                  */
2347                 upper_limit = (1LL << 32) - 1;
2348
2349                 /* total blocks in file system block size */
2350                 upper_limit >>= (bits - 9);
2351
2352         } else {
2353                 /*
2354                  * We use 48 bit ext4_inode i_blocks
2355                  * With EXT4_HUGE_FILE_FL set the i_blocks
2356                  * represent total number of blocks in
2357                  * file system block size
2358                  */
2359                 upper_limit = (1LL << 48) - 1;
2360
2361         }
2362
2363         /* indirect blocks */
2364         meta_blocks = 1;
2365         /* double indirect blocks */
2366         meta_blocks += 1 + (1LL << (bits-2));
2367         /* tripple indirect blocks */
2368         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2369
2370         upper_limit -= meta_blocks;
2371         upper_limit <<= bits;
2372
2373         res += 1LL << (bits-2);
2374         res += 1LL << (2*(bits-2));
2375         res += 1LL << (3*(bits-2));
2376         res <<= bits;
2377         if (res > upper_limit)
2378                 res = upper_limit;
2379
2380         if (res > MAX_LFS_FILESIZE)
2381                 res = MAX_LFS_FILESIZE;
2382
2383         return res;
2384 }
2385
2386 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2387                                    ext4_fsblk_t logical_sb_block, int nr)
2388 {
2389         struct ext4_sb_info *sbi = EXT4_SB(sb);
2390         ext4_group_t bg, first_meta_bg;
2391         int has_super = 0;
2392
2393         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2394
2395         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2396             nr < first_meta_bg)
2397                 return logical_sb_block + nr + 1;
2398         bg = sbi->s_desc_per_block * nr;
2399         if (ext4_bg_has_super(sb, bg))
2400                 has_super = 1;
2401
2402         /*
2403          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2404          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2405          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2406          * compensate.
2407          */
2408         if (sb->s_blocksize == 1024 && nr == 0 &&
2409             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2410                 has_super++;
2411
2412         return (has_super + ext4_group_first_block_no(sb, bg));
2413 }
2414
2415 /**
2416  * ext4_get_stripe_size: Get the stripe size.
2417  * @sbi: In memory super block info
2418  *
2419  * If we have specified it via mount option, then
2420  * use the mount option value. If the value specified at mount time is
2421  * greater than the blocks per group use the super block value.
2422  * If the super block value is greater than blocks per group return 0.
2423  * Allocator needs it be less than blocks per group.
2424  *
2425  */
2426 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2427 {
2428         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2429         unsigned long stripe_width =
2430                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2431         int ret;
2432
2433         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2434                 ret = sbi->s_stripe;
2435         else if (stripe_width <= sbi->s_blocks_per_group)
2436                 ret = stripe_width;
2437         else if (stride <= sbi->s_blocks_per_group)
2438                 ret = stride;
2439         else
2440                 ret = 0;
2441
2442         /*
2443          * If the stripe width is 1, this makes no sense and
2444          * we set it to 0 to turn off stripe handling code.
2445          */
2446         if (ret <= 1)
2447                 ret = 0;
2448
2449         return ret;
2450 }
2451
2452 /* sysfs supprt */
2453
2454 struct ext4_attr {
2455         struct attribute attr;
2456         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2457         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2458                          const char *, size_t);
2459         union {
2460                 int offset;
2461                 int deprecated_val;
2462         } u;
2463 };
2464
2465 static int parse_strtoull(const char *buf,
2466                 unsigned long long max, unsigned long long *value)
2467 {
2468         int ret;
2469
2470         ret = kstrtoull(skip_spaces(buf), 0, value);
2471         if (!ret && *value > max)
2472                 ret = -EINVAL;
2473         return ret;
2474 }
2475
2476 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2477                                               struct ext4_sb_info *sbi,
2478                                               char *buf)
2479 {
2480         return snprintf(buf, PAGE_SIZE, "%llu\n",
2481                 (s64) EXT4_C2B(sbi,
2482                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2483 }
2484
2485 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2486                                          struct ext4_sb_info *sbi, char *buf)
2487 {
2488         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2489
2490         if (!sb->s_bdev->bd_part)
2491                 return snprintf(buf, PAGE_SIZE, "0\n");
2492         return snprintf(buf, PAGE_SIZE, "%lu\n",
2493                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2494                          sbi->s_sectors_written_start) >> 1);
2495 }
2496
2497 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2498                                           struct ext4_sb_info *sbi, char *buf)
2499 {
2500         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2501
2502         if (!sb->s_bdev->bd_part)
2503                 return snprintf(buf, PAGE_SIZE, "0\n");
2504         return snprintf(buf, PAGE_SIZE, "%llu\n",
2505                         (unsigned long long)(sbi->s_kbytes_written +
2506                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2507                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2508 }
2509
2510 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2511                                           struct ext4_sb_info *sbi,
2512                                           const char *buf, size_t count)
2513 {
2514         unsigned long t;
2515         int ret;
2516
2517         ret = kstrtoul(skip_spaces(buf), 0, &t);
2518         if (ret)
2519                 return ret;
2520
2521         if (t && (!is_power_of_2(t) || t > 0x40000000))
2522                 return -EINVAL;
2523
2524         sbi->s_inode_readahead_blks = t;
2525         return count;
2526 }
2527
2528 static ssize_t sbi_ui_show(struct ext4_attr *a,
2529                            struct ext4_sb_info *sbi, char *buf)
2530 {
2531         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2532
2533         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2534 }
2535
2536 static ssize_t sbi_ui_store(struct ext4_attr *a,
2537                             struct ext4_sb_info *sbi,
2538                             const char *buf, size_t count)
2539 {
2540         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2541         unsigned long t;
2542         int ret;
2543
2544         ret = kstrtoul(skip_spaces(buf), 0, &t);
2545         if (ret)
2546                 return ret;
2547         *ui = t;
2548         return count;
2549 }
2550
2551 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2552                                   struct ext4_sb_info *sbi, char *buf)
2553 {
2554         return snprintf(buf, PAGE_SIZE, "%llu\n",
2555                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2556 }
2557
2558 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2559                                    struct ext4_sb_info *sbi,
2560                                    const char *buf, size_t count)
2561 {
2562         unsigned long long val;
2563         int ret;
2564
2565         if (parse_strtoull(buf, -1ULL, &val))
2566                 return -EINVAL;
2567         ret = ext4_reserve_clusters(sbi, val);
2568
2569         return ret ? ret : count;
2570 }
2571
2572 static ssize_t trigger_test_error(struct ext4_attr *a,
2573                                   struct ext4_sb_info *sbi,
2574                                   const char *buf, size_t count)
2575 {
2576         int len = count;
2577
2578         if (!capable(CAP_SYS_ADMIN))
2579                 return -EPERM;
2580
2581         if (len && buf[len-1] == '\n')
2582                 len--;
2583
2584         if (len)
2585                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2586         return count;
2587 }
2588
2589 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2590                                    struct ext4_sb_info *sbi, char *buf)
2591 {
2592         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2593 }
2594
2595 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2596 static struct ext4_attr ext4_attr_##_name = {                   \
2597         .attr = {.name = __stringify(_name), .mode = _mode },   \
2598         .show   = _show,                                        \
2599         .store  = _store,                                       \
2600         .u = {                                                  \
2601                 .offset = offsetof(struct ext4_sb_info, _elname),\
2602         },                                                      \
2603 }
2604 #define EXT4_ATTR(name, mode, show, store) \
2605 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2606
2607 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2608 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2609 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2610 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2611         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2612 #define ATTR_LIST(name) &ext4_attr_##name.attr
2613 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2614 static struct ext4_attr ext4_attr_##_name = {                   \
2615         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2616         .show   = sbi_deprecated_show,                          \
2617         .u = {                                                  \
2618                 .deprecated_val = _val,                         \
2619         },                                                      \
2620 }
2621
2622 EXT4_RO_ATTR(delayed_allocation_blocks);
2623 EXT4_RO_ATTR(session_write_kbytes);
2624 EXT4_RO_ATTR(lifetime_write_kbytes);
2625 EXT4_RW_ATTR(reserved_clusters);
2626 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2627                  inode_readahead_blks_store, s_inode_readahead_blks);
2628 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2629 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2630 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2631 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2632 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2633 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2634 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2635 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2636 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2637 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2638 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2639 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2640 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2641 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2642 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2643 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2644
2645 static struct attribute *ext4_attrs[] = {
2646         ATTR_LIST(delayed_allocation_blocks),
2647         ATTR_LIST(session_write_kbytes),
2648         ATTR_LIST(lifetime_write_kbytes),
2649         ATTR_LIST(reserved_clusters),
2650         ATTR_LIST(inode_readahead_blks),
2651         ATTR_LIST(inode_goal),
2652         ATTR_LIST(mb_stats),
2653         ATTR_LIST(mb_max_to_scan),
2654         ATTR_LIST(mb_min_to_scan),
2655         ATTR_LIST(mb_order2_req),
2656         ATTR_LIST(mb_stream_req),
2657         ATTR_LIST(mb_group_prealloc),
2658         ATTR_LIST(max_writeback_mb_bump),
2659         ATTR_LIST(extent_max_zeroout_kb),
2660         ATTR_LIST(trigger_fs_error),
2661         ATTR_LIST(err_ratelimit_interval_ms),
2662         ATTR_LIST(err_ratelimit_burst),
2663         ATTR_LIST(warning_ratelimit_interval_ms),
2664         ATTR_LIST(warning_ratelimit_burst),
2665         ATTR_LIST(msg_ratelimit_interval_ms),
2666         ATTR_LIST(msg_ratelimit_burst),
2667         NULL,
2668 };
2669
2670 /* Features this copy of ext4 supports */
2671 EXT4_INFO_ATTR(lazy_itable_init);
2672 EXT4_INFO_ATTR(batched_discard);
2673 EXT4_INFO_ATTR(meta_bg_resize);
2674
2675 static struct attribute *ext4_feat_attrs[] = {
2676         ATTR_LIST(lazy_itable_init),
2677         ATTR_LIST(batched_discard),
2678         ATTR_LIST(meta_bg_resize),
2679         NULL,
2680 };
2681
2682 static ssize_t ext4_attr_show(struct kobject *kobj,
2683                               struct attribute *attr, char *buf)
2684 {
2685         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2686                                                 s_kobj);
2687         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2688
2689         return a->show ? a->show(a, sbi, buf) : 0;
2690 }
2691
2692 static ssize_t ext4_attr_store(struct kobject *kobj,
2693                                struct attribute *attr,
2694                                const char *buf, size_t len)
2695 {
2696         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2697                                                 s_kobj);
2698         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2699
2700         return a->store ? a->store(a, sbi, buf, len) : 0;
2701 }
2702
2703 static void ext4_sb_release(struct kobject *kobj)
2704 {
2705         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2706                                                 s_kobj);
2707         complete(&sbi->s_kobj_unregister);
2708 }
2709
2710 static const struct sysfs_ops ext4_attr_ops = {
2711         .show   = ext4_attr_show,
2712         .store  = ext4_attr_store,
2713 };
2714
2715 static struct kobj_type ext4_ktype = {
2716         .default_attrs  = ext4_attrs,
2717         .sysfs_ops      = &ext4_attr_ops,
2718         .release        = ext4_sb_release,
2719 };
2720
2721 static void ext4_feat_release(struct kobject *kobj)
2722 {
2723         complete(&ext4_feat->f_kobj_unregister);
2724 }
2725
2726 static struct kobj_type ext4_feat_ktype = {
2727         .default_attrs  = ext4_feat_attrs,
2728         .sysfs_ops      = &ext4_attr_ops,
2729         .release        = ext4_feat_release,
2730 };
2731
2732 /*
2733  * Check whether this filesystem can be mounted based on
2734  * the features present and the RDONLY/RDWR mount requested.
2735  * Returns 1 if this filesystem can be mounted as requested,
2736  * 0 if it cannot be.
2737  */
2738 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2739 {
2740         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2741                 ext4_msg(sb, KERN_ERR,
2742                         "Couldn't mount because of "
2743                         "unsupported optional features (%x)",
2744                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2745                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2746                 return 0;
2747         }
2748
2749         if (readonly)
2750                 return 1;
2751
2752         /* Check that feature set is OK for a read-write mount */
2753         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2754                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2755                          "unsupported optional features (%x)",
2756                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2757                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2758                 return 0;
2759         }
2760         /*
2761          * Large file size enabled file system can only be mounted
2762          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2763          */
2764         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2765                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2766                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2767                                  "cannot be mounted RDWR without "
2768                                  "CONFIG_LBDAF");
2769                         return 0;
2770                 }
2771         }
2772         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2773             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2774                 ext4_msg(sb, KERN_ERR,
2775                          "Can't support bigalloc feature without "
2776                          "extents feature\n");
2777                 return 0;
2778         }
2779
2780 #ifndef CONFIG_QUOTA
2781         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2782             !readonly) {
2783                 ext4_msg(sb, KERN_ERR,
2784                          "Filesystem with quota feature cannot be mounted RDWR "
2785                          "without CONFIG_QUOTA");
2786                 return 0;
2787         }
2788 #endif  /* CONFIG_QUOTA */
2789         return 1;
2790 }
2791
2792 /*
2793  * This function is called once a day if we have errors logged
2794  * on the file system
2795  */
2796 static void print_daily_error_info(unsigned long arg)
2797 {
2798         struct super_block *sb = (struct super_block *) arg;
2799         struct ext4_sb_info *sbi;
2800         struct ext4_super_block *es;
2801
2802         sbi = EXT4_SB(sb);
2803         es = sbi->s_es;
2804
2805         if (es->s_error_count)
2806                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2807                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2808                          le32_to_cpu(es->s_error_count));
2809         if (es->s_first_error_time) {
2810                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2811                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2812                        (int) sizeof(es->s_first_error_func),
2813                        es->s_first_error_func,
2814                        le32_to_cpu(es->s_first_error_line));
2815                 if (es->s_first_error_ino)
2816                         printk(": inode %u",
2817                                le32_to_cpu(es->s_first_error_ino));
2818                 if (es->s_first_error_block)
2819                         printk(": block %llu", (unsigned long long)
2820                                le64_to_cpu(es->s_first_error_block));
2821                 printk("\n");
2822         }
2823         if (es->s_last_error_time) {
2824                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2825                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2826                        (int) sizeof(es->s_last_error_func),
2827                        es->s_last_error_func,
2828                        le32_to_cpu(es->s_last_error_line));
2829                 if (es->s_last_error_ino)
2830                         printk(": inode %u",
2831                                le32_to_cpu(es->s_last_error_ino));
2832                 if (es->s_last_error_block)
2833                         printk(": block %llu", (unsigned long long)
2834                                le64_to_cpu(es->s_last_error_block));
2835                 printk("\n");
2836         }
2837         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2838 }
2839
2840 /* Find next suitable group and run ext4_init_inode_table */
2841 static int ext4_run_li_request(struct ext4_li_request *elr)
2842 {
2843         struct ext4_group_desc *gdp = NULL;
2844         ext4_group_t group, ngroups;
2845         struct super_block *sb;
2846         unsigned long timeout = 0;
2847         int ret = 0;
2848
2849         sb = elr->lr_super;
2850         ngroups = EXT4_SB(sb)->s_groups_count;
2851
2852         sb_start_write(sb);
2853         for (group = elr->lr_next_group; group < ngroups; group++) {
2854                 gdp = ext4_get_group_desc(sb, group, NULL);
2855                 if (!gdp) {
2856                         ret = 1;
2857                         break;
2858                 }
2859
2860                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2861                         break;
2862         }
2863
2864         if (group >= ngroups)
2865                 ret = 1;
2866
2867         if (!ret) {
2868                 timeout = jiffies;
2869                 ret = ext4_init_inode_table(sb, group,
2870                                             elr->lr_timeout ? 0 : 1);
2871                 if (elr->lr_timeout == 0) {
2872                         timeout = (jiffies - timeout) *
2873                                   elr->lr_sbi->s_li_wait_mult;
2874                         elr->lr_timeout = timeout;
2875                 }
2876                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2877                 elr->lr_next_group = group + 1;
2878         }
2879         sb_end_write(sb);
2880
2881         return ret;
2882 }
2883
2884 /*
2885  * Remove lr_request from the list_request and free the
2886  * request structure. Should be called with li_list_mtx held
2887  */
2888 static void ext4_remove_li_request(struct ext4_li_request *elr)
2889 {
2890         struct ext4_sb_info *sbi;
2891
2892         if (!elr)
2893                 return;
2894
2895         sbi = elr->lr_sbi;
2896
2897         list_del(&elr->lr_request);
2898         sbi->s_li_request = NULL;
2899         kfree(elr);
2900 }
2901
2902 static void ext4_unregister_li_request(struct super_block *sb)
2903 {
2904         mutex_lock(&ext4_li_mtx);
2905         if (!ext4_li_info) {
2906                 mutex_unlock(&ext4_li_mtx);
2907                 return;
2908         }
2909
2910         mutex_lock(&ext4_li_info->li_list_mtx);
2911         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2912         mutex_unlock(&ext4_li_info->li_list_mtx);
2913         mutex_unlock(&ext4_li_mtx);
2914 }
2915
2916 static struct task_struct *ext4_lazyinit_task;
2917
2918 /*
2919  * This is the function where ext4lazyinit thread lives. It walks
2920  * through the request list searching for next scheduled filesystem.
2921  * When such a fs is found, run the lazy initialization request
2922  * (ext4_rn_li_request) and keep track of the time spend in this
2923  * function. Based on that time we compute next schedule time of
2924  * the request. When walking through the list is complete, compute
2925  * next waking time and put itself into sleep.
2926  */
2927 static int ext4_lazyinit_thread(void *arg)
2928 {
2929         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2930         struct list_head *pos, *n;
2931         struct ext4_li_request *elr;
2932         unsigned long next_wakeup, cur;
2933
2934         BUG_ON(NULL == eli);
2935
2936 cont_thread:
2937         while (true) {
2938                 next_wakeup = MAX_JIFFY_OFFSET;
2939
2940                 mutex_lock(&eli->li_list_mtx);
2941                 if (list_empty(&eli->li_request_list)) {
2942                         mutex_unlock(&eli->li_list_mtx);
2943                         goto exit_thread;
2944                 }
2945
2946                 list_for_each_safe(pos, n, &eli->li_request_list) {
2947                         elr = list_entry(pos, struct ext4_li_request,
2948                                          lr_request);
2949
2950                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2951                                 if (ext4_run_li_request(elr) != 0) {
2952                                         /* error, remove the lazy_init job */
2953                                         ext4_remove_li_request(elr);
2954                                         continue;
2955                                 }
2956                         }
2957
2958                         if (time_before(elr->lr_next_sched, next_wakeup))
2959                                 next_wakeup = elr->lr_next_sched;
2960                 }
2961                 mutex_unlock(&eli->li_list_mtx);
2962
2963                 try_to_freeze();
2964
2965                 cur = jiffies;
2966                 if ((time_after_eq(cur, next_wakeup)) ||
2967                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2968                         cond_resched();
2969                         continue;
2970                 }
2971
2972                 schedule_timeout_interruptible(next_wakeup - cur);
2973
2974                 if (kthread_should_stop()) {
2975                         ext4_clear_request_list();
2976                         goto exit_thread;
2977                 }
2978         }
2979
2980 exit_thread:
2981         /*
2982          * It looks like the request list is empty, but we need
2983          * to check it under the li_list_mtx lock, to prevent any
2984          * additions into it, and of course we should lock ext4_li_mtx
2985          * to atomically free the list and ext4_li_info, because at
2986          * this point another ext4 filesystem could be registering
2987          * new one.
2988          */
2989         mutex_lock(&ext4_li_mtx);
2990         mutex_lock(&eli->li_list_mtx);
2991         if (!list_empty(&eli->li_request_list)) {
2992                 mutex_unlock(&eli->li_list_mtx);
2993                 mutex_unlock(&ext4_li_mtx);
2994                 goto cont_thread;
2995         }
2996         mutex_unlock(&eli->li_list_mtx);
2997         kfree(ext4_li_info);
2998         ext4_li_info = NULL;
2999         mutex_unlock(&ext4_li_mtx);
3000
3001         return 0;
3002 }
3003
3004 static void ext4_clear_request_list(void)
3005 {
3006         struct list_head *pos, *n;
3007         struct ext4_li_request *elr;
3008
3009         mutex_lock(&ext4_li_info->li_list_mtx);
3010         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3011                 elr = list_entry(pos, struct ext4_li_request,
3012                                  lr_request);
3013                 ext4_remove_li_request(elr);
3014         }
3015         mutex_unlock(&ext4_li_info->li_list_mtx);
3016 }
3017
3018 static int ext4_run_lazyinit_thread(void)
3019 {
3020         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3021                                          ext4_li_info, "ext4lazyinit");
3022         if (IS_ERR(ext4_lazyinit_task)) {
3023                 int err = PTR_ERR(ext4_lazyinit_task);
3024                 ext4_clear_request_list();
3025                 kfree(ext4_li_info);
3026                 ext4_li_info = NULL;
3027                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3028                                  "initialization thread\n",
3029                                  err);
3030                 return err;
3031         }
3032         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3033         return 0;
3034 }
3035
3036 /*
3037  * Check whether it make sense to run itable init. thread or not.
3038  * If there is at least one uninitialized inode table, return
3039  * corresponding group number, else the loop goes through all
3040  * groups and return total number of groups.
3041  */
3042 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3043 {
3044         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3045         struct ext4_group_desc *gdp = NULL;
3046
3047         for (group = 0; group < ngroups; group++) {
3048                 gdp = ext4_get_group_desc(sb, group, NULL);
3049                 if (!gdp)
3050                         continue;
3051
3052                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3053                         break;
3054         }
3055
3056         return group;
3057 }
3058
3059 static int ext4_li_info_new(void)
3060 {
3061         struct ext4_lazy_init *eli = NULL;
3062
3063         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3064         if (!eli)
3065                 return -ENOMEM;
3066
3067         INIT_LIST_HEAD(&eli->li_request_list);
3068         mutex_init(&eli->li_list_mtx);
3069
3070         eli->li_state |= EXT4_LAZYINIT_QUIT;
3071
3072         ext4_li_info = eli;
3073
3074         return 0;
3075 }
3076
3077 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3078                                             ext4_group_t start)
3079 {
3080         struct ext4_sb_info *sbi = EXT4_SB(sb);
3081         struct ext4_li_request *elr;
3082
3083         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3084         if (!elr)
3085                 return NULL;
3086
3087         elr->lr_super = sb;
3088         elr->lr_sbi = sbi;
3089         elr->lr_next_group = start;
3090
3091         /*
3092          * Randomize first schedule time of the request to
3093          * spread the inode table initialization requests
3094          * better.
3095          */
3096         elr->lr_next_sched = jiffies + (prandom_u32() %
3097                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3098         return elr;
3099 }
3100
3101 int ext4_register_li_request(struct super_block *sb,
3102                              ext4_group_t first_not_zeroed)
3103 {
3104         struct ext4_sb_info *sbi = EXT4_SB(sb);
3105         struct ext4_li_request *elr = NULL;
3106         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3107         int ret = 0;
3108
3109         mutex_lock(&ext4_li_mtx);
3110         if (sbi->s_li_request != NULL) {
3111                 /*
3112                  * Reset timeout so it can be computed again, because
3113                  * s_li_wait_mult might have changed.
3114                  */
3115                 sbi->s_li_request->lr_timeout = 0;
3116                 goto out;
3117         }
3118
3119         if (first_not_zeroed == ngroups ||
3120             (sb->s_flags & MS_RDONLY) ||
3121             !test_opt(sb, INIT_INODE_TABLE))
3122                 goto out;
3123
3124         elr = ext4_li_request_new(sb, first_not_zeroed);
3125         if (!elr) {
3126                 ret = -ENOMEM;
3127                 goto out;
3128         }
3129
3130         if (NULL == ext4_li_info) {
3131                 ret = ext4_li_info_new();
3132                 if (ret)
3133                         goto out;
3134         }
3135
3136         mutex_lock(&ext4_li_info->li_list_mtx);
3137         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3138         mutex_unlock(&ext4_li_info->li_list_mtx);
3139
3140         sbi->s_li_request = elr;
3141         /*
3142          * set elr to NULL here since it has been inserted to
3143          * the request_list and the removal and free of it is
3144          * handled by ext4_clear_request_list from now on.
3145          */
3146         elr = NULL;
3147
3148         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3149                 ret = ext4_run_lazyinit_thread();
3150                 if (ret)
3151                         goto out;
3152         }
3153 out:
3154         mutex_unlock(&ext4_li_mtx);
3155         if (ret)
3156                 kfree(elr);
3157         return ret;
3158 }
3159
3160 /*
3161  * We do not need to lock anything since this is called on
3162  * module unload.
3163  */
3164 static void ext4_destroy_lazyinit_thread(void)
3165 {
3166         /*
3167          * If thread exited earlier
3168          * there's nothing to be done.
3169          */
3170         if (!ext4_li_info || !ext4_lazyinit_task)
3171                 return;
3172
3173         kthread_stop(ext4_lazyinit_task);
3174 }
3175
3176 static int set_journal_csum_feature_set(struct super_block *sb)
3177 {
3178         int ret = 1;
3179         int compat, incompat;
3180         struct ext4_sb_info *sbi = EXT4_SB(sb);
3181
3182         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3183                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3184                 /* journal checksum v3 */
3185                 compat = 0;
3186                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3187         } else {
3188                 /* journal checksum v1 */
3189                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3190                 incompat = 0;
3191         }
3192
3193         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3194                 ret = jbd2_journal_set_features(sbi->s_journal,
3195                                 compat, 0,
3196                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3197                                 incompat);
3198         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3199                 ret = jbd2_journal_set_features(sbi->s_journal,
3200                                 compat, 0,
3201                                 incompat);
3202                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3203                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3204         } else {
3205                 jbd2_journal_clear_features(sbi->s_journal,
3206                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3207                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3208                                 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3209                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3210         }
3211
3212         return ret;
3213 }
3214
3215 /*
3216  * Note: calculating the overhead so we can be compatible with
3217  * historical BSD practice is quite difficult in the face of
3218  * clusters/bigalloc.  This is because multiple metadata blocks from
3219  * different block group can end up in the same allocation cluster.
3220  * Calculating the exact overhead in the face of clustered allocation
3221  * requires either O(all block bitmaps) in memory or O(number of block
3222  * groups**2) in time.  We will still calculate the superblock for
3223  * older file systems --- and if we come across with a bigalloc file
3224  * system with zero in s_overhead_clusters the estimate will be close to
3225  * correct especially for very large cluster sizes --- but for newer
3226  * file systems, it's better to calculate this figure once at mkfs
3227  * time, and store it in the superblock.  If the superblock value is
3228  * present (even for non-bigalloc file systems), we will use it.
3229  */
3230 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3231                           char *buf)
3232 {
3233         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3234         struct ext4_group_desc  *gdp;
3235         ext4_fsblk_t            first_block, last_block, b;
3236         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3237         int                     s, j, count = 0;
3238
3239         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3240                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3241                         sbi->s_itb_per_group + 2);
3242
3243         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3244                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3245         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3246         for (i = 0; i < ngroups; i++) {
3247                 gdp = ext4_get_group_desc(sb, i, NULL);
3248                 b = ext4_block_bitmap(sb, gdp);
3249                 if (b >= first_block && b <= last_block) {
3250                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3251                         count++;
3252                 }
3253                 b = ext4_inode_bitmap(sb, gdp);
3254                 if (b >= first_block && b <= last_block) {
3255                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3256                         count++;
3257                 }
3258                 b = ext4_inode_table(sb, gdp);
3259                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3260                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3261                                 int c = EXT4_B2C(sbi, b - first_block);
3262                                 ext4_set_bit(c, buf);
3263                                 count++;
3264                         }
3265                 if (i != grp)
3266                         continue;
3267                 s = 0;
3268                 if (ext4_bg_has_super(sb, grp)) {
3269                         ext4_set_bit(s++, buf);
3270                         count++;
3271                 }
3272                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3273                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3274                         count++;
3275                 }
3276         }
3277         if (!count)
3278                 return 0;
3279         return EXT4_CLUSTERS_PER_GROUP(sb) -
3280                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3281 }
3282
3283 /*
3284  * Compute the overhead and stash it in sbi->s_overhead
3285  */
3286 int ext4_calculate_overhead(struct super_block *sb)
3287 {
3288         struct ext4_sb_info *sbi = EXT4_SB(sb);
3289         struct ext4_super_block *es = sbi->s_es;
3290         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3291         ext4_fsblk_t overhead = 0;
3292         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3293
3294         if (!buf)
3295                 return -ENOMEM;
3296
3297         /*
3298          * Compute the overhead (FS structures).  This is constant
3299          * for a given filesystem unless the number of block groups
3300          * changes so we cache the previous value until it does.
3301          */
3302
3303         /*
3304          * All of the blocks before first_data_block are overhead
3305          */
3306         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3307
3308         /*
3309          * Add the overhead found in each block group
3310          */
3311         for (i = 0; i < ngroups; i++) {
3312                 int blks;
3313
3314                 blks = count_overhead(sb, i, buf);
3315                 overhead += blks;
3316                 if (blks)
3317                         memset(buf, 0, PAGE_SIZE);
3318                 cond_resched();
3319         }
3320         /* Add the journal blocks as well */
3321         if (sbi->s_journal)
3322                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3323
3324         sbi->s_overhead = overhead;
3325         smp_wmb();
3326         free_page((unsigned long) buf);
3327         return 0;
3328 }
3329
3330
3331 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3332 {
3333         ext4_fsblk_t resv_clusters;
3334
3335         /*
3336          * There's no need to reserve anything when we aren't using extents.
3337          * The space estimates are exact, there are no unwritten extents,
3338          * hole punching doesn't need new metadata... This is needed especially
3339          * to keep ext2/3 backward compatibility.
3340          */
3341         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3342                 return 0;
3343         /*
3344          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3345          * This should cover the situations where we can not afford to run
3346          * out of space like for example punch hole, or converting
3347          * unwritten extents in delalloc path. In most cases such
3348          * allocation would require 1, or 2 blocks, higher numbers are
3349          * very rare.
3350          */
3351         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3352                         EXT4_SB(sb)->s_cluster_bits;
3353
3354         do_div(resv_clusters, 50);
3355         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3356
3357         return resv_clusters;
3358 }
3359
3360
3361 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3362 {
3363         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3364                                 sbi->s_cluster_bits;
3365
3366         if (count >= clusters)
3367                 return -EINVAL;
3368
3369         atomic64_set(&sbi->s_resv_clusters, count);
3370         return 0;
3371 }
3372
3373 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3374 {
3375         char *orig_data = kstrdup(data, GFP_KERNEL);
3376         struct buffer_head *bh;
3377         struct ext4_super_block *es = NULL;
3378         struct ext4_sb_info *sbi;
3379         ext4_fsblk_t block;
3380         ext4_fsblk_t sb_block = get_sb_block(&data);
3381         ext4_fsblk_t logical_sb_block;
3382         unsigned long offset = 0;
3383         unsigned long journal_devnum = 0;
3384         unsigned long def_mount_opts;
3385         struct inode *root;
3386         char *cp;
3387         const char *descr;
3388         int ret = -ENOMEM;
3389         int blocksize, clustersize;
3390         unsigned int db_count;
3391         unsigned int i;
3392         int needs_recovery, has_huge_files, has_bigalloc;
3393         __u64 blocks_count;
3394         int err = 0;
3395         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3396         ext4_group_t first_not_zeroed;
3397
3398         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3399         if (!sbi)
3400                 goto out_free_orig;
3401
3402         sbi->s_blockgroup_lock =
3403                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3404         if (!sbi->s_blockgroup_lock) {
3405                 kfree(sbi);
3406                 goto out_free_orig;
3407         }
3408         sb->s_fs_info = sbi;
3409         sbi->s_sb = sb;
3410         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3411         sbi->s_sb_block = sb_block;
3412         if (sb->s_bdev->bd_part)
3413                 sbi->s_sectors_written_start =
3414                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3415
3416         /* Cleanup superblock name */
3417         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3418                 *cp = '!';
3419
3420         /* -EINVAL is default */
3421         ret = -EINVAL;
3422         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3423         if (!blocksize) {
3424                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3425                 goto out_fail;
3426         }
3427
3428         /*
3429          * The ext4 superblock will not be buffer aligned for other than 1kB
3430          * block sizes.  We need to calculate the offset from buffer start.
3431          */
3432         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3433                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3434                 offset = do_div(logical_sb_block, blocksize);
3435         } else {
3436                 logical_sb_block = sb_block;
3437         }
3438
3439         if (!(bh = sb_bread(sb, logical_sb_block))) {
3440                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3441                 goto out_fail;
3442         }
3443         /*
3444          * Note: s_es must be initialized as soon as possible because
3445          *       some ext4 macro-instructions depend on its value
3446          */
3447         es = (struct ext4_super_block *) (bh->b_data + offset);
3448         sbi->s_es = es;
3449         sb->s_magic = le16_to_cpu(es->s_magic);
3450         if (sb->s_magic != EXT4_SUPER_MAGIC)
3451                 goto cantfind_ext4;
3452         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3453
3454         /* Warn if metadata_csum and gdt_csum are both set. */
3455         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3456                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3457             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3458                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3459                              "redundant flags; please run fsck.");
3460
3461         /* Check for a known checksum algorithm */
3462         if (!ext4_verify_csum_type(sb, es)) {
3463                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3464                          "unknown checksum algorithm.");
3465                 silent = 1;
3466                 goto cantfind_ext4;
3467         }
3468
3469         /* Load the checksum driver */
3470         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3471                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3472                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3473                 if (IS_ERR(sbi->s_chksum_driver)) {
3474                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3475                         ret = PTR_ERR(sbi->s_chksum_driver);
3476                         sbi->s_chksum_driver = NULL;
3477                         goto failed_mount;
3478                 }
3479         }
3480
3481         /* Check superblock checksum */
3482         if (!ext4_superblock_csum_verify(sb, es)) {
3483                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3484                          "invalid superblock checksum.  Run e2fsck?");
3485                 silent = 1;
3486                 goto cantfind_ext4;
3487         }
3488
3489         /* Precompute checksum seed for all metadata */
3490         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3491                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3492                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3493                                                sizeof(es->s_uuid));
3494
3495         /* Set defaults before we parse the mount options */
3496         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3497         set_opt(sb, INIT_INODE_TABLE);
3498         if (def_mount_opts & EXT4_DEFM_DEBUG)
3499                 set_opt(sb, DEBUG);
3500         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3501                 set_opt(sb, GRPID);
3502         if (def_mount_opts & EXT4_DEFM_UID16)
3503                 set_opt(sb, NO_UID32);
3504         /* xattr user namespace & acls are now defaulted on */
3505         set_opt(sb, XATTR_USER);
3506 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3507         set_opt(sb, POSIX_ACL);
3508 #endif
3509         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3510                 set_opt(sb, JOURNAL_DATA);
3511         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3512                 set_opt(sb, ORDERED_DATA);
3513         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3514                 set_opt(sb, WRITEBACK_DATA);
3515
3516         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3517                 set_opt(sb, ERRORS_PANIC);
3518         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3519                 set_opt(sb, ERRORS_CONT);
3520         else
3521                 set_opt(sb, ERRORS_RO);
3522         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3523                 set_opt(sb, BLOCK_VALIDITY);
3524         if (def_mount_opts & EXT4_DEFM_DISCARD)
3525                 set_opt(sb, DISCARD);
3526
3527         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3528         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3529         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3530         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3531         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3532
3533         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3534                 set_opt(sb, BARRIER);
3535
3536         /*
3537          * enable delayed allocation by default
3538          * Use -o nodelalloc to turn it off
3539          */
3540         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3541             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3542                 set_opt(sb, DELALLOC);
3543
3544         /*
3545          * set default s_li_wait_mult for lazyinit, for the case there is
3546          * no mount option specified.
3547          */
3548         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3549
3550         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3551                            &journal_devnum, &journal_ioprio, 0)) {
3552                 ext4_msg(sb, KERN_WARNING,
3553                          "failed to parse options in superblock: %s",
3554                          sbi->s_es->s_mount_opts);
3555         }
3556         sbi->s_def_mount_opt = sbi->s_mount_opt;
3557         if (!parse_options((char *) data, sb, &journal_devnum,
3558                            &journal_ioprio, 0))
3559                 goto failed_mount;
3560
3561         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3562                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3563                             "with data=journal disables delayed "
3564                             "allocation and O_DIRECT support!\n");
3565                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3566                         ext4_msg(sb, KERN_ERR, "can't mount with "
3567                                  "both data=journal and delalloc");
3568                         goto failed_mount;
3569                 }
3570                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3571                         ext4_msg(sb, KERN_ERR, "can't mount with "
3572                                  "both data=journal and dioread_nolock");
3573                         goto failed_mount;
3574                 }
3575                 if (test_opt(sb, DELALLOC))
3576                         clear_opt(sb, DELALLOC);
3577         }
3578
3579         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3580                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3581
3582         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3583             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3584              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3585              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3586                 ext4_msg(sb, KERN_WARNING,
3587                        "feature flags set on rev 0 fs, "
3588                        "running e2fsck is recommended");
3589
3590         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3591                 set_opt2(sb, HURD_COMPAT);
3592                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3593                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3594                         ext4_msg(sb, KERN_ERR,
3595                                  "The Hurd can't support 64-bit file systems");
3596                         goto failed_mount;
3597                 }
3598         }
3599
3600         if (IS_EXT2_SB(sb)) {
3601                 if (ext2_feature_set_ok(sb))
3602                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3603                                  "using the ext4 subsystem");
3604                 else {
3605                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3606                                  "to feature incompatibilities");
3607                         goto failed_mount;
3608                 }
3609         }
3610
3611         if (IS_EXT3_SB(sb)) {
3612                 if (ext3_feature_set_ok(sb))
3613                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3614                                  "using the ext4 subsystem");
3615                 else {
3616                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3617                                  "to feature incompatibilities");
3618                         goto failed_mount;
3619                 }
3620         }
3621
3622         /*
3623          * Check feature flags regardless of the revision level, since we
3624          * previously didn't change the revision level when setting the flags,
3625          * so there is a chance incompat flags are set on a rev 0 filesystem.
3626          */
3627         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3628                 goto failed_mount;
3629
3630         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3631         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3632             blocksize > EXT4_MAX_BLOCK_SIZE) {
3633                 ext4_msg(sb, KERN_ERR,
3634                        "Unsupported filesystem blocksize %d", blocksize);
3635                 goto failed_mount;
3636         }
3637
3638         if (sb->s_blocksize != blocksize) {
3639                 /* Validate the filesystem blocksize */
3640                 if (!sb_set_blocksize(sb, blocksize)) {
3641                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3642                                         blocksize);
3643                         goto failed_mount;
3644                 }
3645
3646                 brelse(bh);
3647                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3648                 offset = do_div(logical_sb_block, blocksize);
3649                 bh = sb_bread(sb, logical_sb_block);
3650                 if (!bh) {
3651                         ext4_msg(sb, KERN_ERR,
3652                                "Can't read superblock on 2nd try");
3653                         goto failed_mount;
3654                 }
3655                 es = (struct ext4_super_block *)(bh->b_data + offset);
3656                 sbi->s_es = es;
3657                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3658                         ext4_msg(sb, KERN_ERR,
3659                                "Magic mismatch, very weird!");
3660                         goto failed_mount;
3661                 }
3662         }
3663
3664         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3665                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3666         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3667                                                       has_huge_files);
3668         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3669
3670         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3671                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3672                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3673         } else {
3674                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3675                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3676                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3677                     (!is_power_of_2(sbi->s_inode_size)) ||
3678                     (sbi->s_inode_size > blocksize)) {
3679                         ext4_msg(sb, KERN_ERR,
3680                                "unsupported inode size: %d",
3681                                sbi->s_inode_size);
3682                         goto failed_mount;
3683                 }
3684                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3685                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3686         }
3687
3688         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3689         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3690                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3691                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3692                     !is_power_of_2(sbi->s_desc_size)) {
3693                         ext4_msg(sb, KERN_ERR,
3694                                "unsupported descriptor size %lu",
3695                                sbi->s_desc_size);
3696                         goto failed_mount;
3697                 }
3698         } else
3699                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3700
3701         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3702         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3703         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3704                 goto cantfind_ext4;
3705
3706         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3707         if (sbi->s_inodes_per_block == 0)
3708                 goto cantfind_ext4;
3709         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3710                                         sbi->s_inodes_per_block;
3711         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3712         sbi->s_sbh = bh;
3713         sbi->s_mount_state = le16_to_cpu(es->s_state);
3714         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3715         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3716
3717         for (i = 0; i < 4; i++)
3718                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3719         sbi->s_def_hash_version = es->s_def_hash_version;
3720         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3721                 i = le32_to_cpu(es->s_flags);
3722                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3723                         sbi->s_hash_unsigned = 3;
3724                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3725 #ifdef __CHAR_UNSIGNED__
3726                         if (!(sb->s_flags & MS_RDONLY))
3727                                 es->s_flags |=
3728                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3729                         sbi->s_hash_unsigned = 3;
3730 #else
3731                         if (!(sb->s_flags & MS_RDONLY))
3732                                 es->s_flags |=
3733                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3734 #endif
3735                 }
3736         }
3737
3738         /* Handle clustersize */
3739         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3740         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3741                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3742         if (has_bigalloc) {
3743                 if (clustersize < blocksize) {
3744                         ext4_msg(sb, KERN_ERR,
3745                                  "cluster size (%d) smaller than "
3746                                  "block size (%d)", clustersize, blocksize);
3747                         goto failed_mount;
3748                 }
3749                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3750                         le32_to_cpu(es->s_log_block_size);
3751                 sbi->s_clusters_per_group =
3752                         le32_to_cpu(es->s_clusters_per_group);
3753                 if (sbi->s_clusters_per_group > blocksize * 8) {
3754                         ext4_msg(sb, KERN_ERR,
3755                                  "#clusters per group too big: %lu",
3756                                  sbi->s_clusters_per_group);
3757                         goto failed_mount;
3758                 }
3759                 if (sbi->s_blocks_per_group !=
3760                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3761                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3762                                  "clusters per group (%lu) inconsistent",
3763                                  sbi->s_blocks_per_group,
3764                                  sbi->s_clusters_per_group);
3765                         goto failed_mount;
3766                 }
3767         } else {
3768                 if (clustersize != blocksize) {
3769                         ext4_warning(sb, "fragment/cluster size (%d) != "
3770                                      "block size (%d)", clustersize,
3771                                      blocksize);
3772                         clustersize = blocksize;
3773                 }
3774                 if (sbi->s_blocks_per_group > blocksize * 8) {
3775                         ext4_msg(sb, KERN_ERR,
3776                                  "#blocks per group too big: %lu",
3777                                  sbi->s_blocks_per_group);
3778                         goto failed_mount;
3779                 }
3780                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3781                 sbi->s_cluster_bits = 0;
3782         }
3783         sbi->s_cluster_ratio = clustersize / blocksize;
3784
3785         if (sbi->s_inodes_per_group > blocksize * 8) {
3786                 ext4_msg(sb, KERN_ERR,
3787                        "#inodes per group too big: %lu",
3788                        sbi->s_inodes_per_group);
3789                 goto failed_mount;
3790         }
3791
3792         /* Do we have standard group size of clustersize * 8 blocks ? */
3793         if (sbi->s_blocks_per_group == clustersize << 3)
3794                 set_opt2(sb, STD_GROUP_SIZE);
3795
3796         /*
3797          * Test whether we have more sectors than will fit in sector_t,
3798          * and whether the max offset is addressable by the page cache.
3799          */
3800         err = generic_check_addressable(sb->s_blocksize_bits,
3801                                         ext4_blocks_count(es));
3802         if (err) {
3803                 ext4_msg(sb, KERN_ERR, "filesystem"
3804                          " too large to mount safely on this system");
3805                 if (sizeof(sector_t) < 8)
3806                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3807                 goto failed_mount;
3808         }
3809
3810         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3811                 goto cantfind_ext4;
3812
3813         /* check blocks count against device size */
3814         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3815         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3816                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3817                        "exceeds size of device (%llu blocks)",
3818                        ext4_blocks_count(es), blocks_count);
3819                 goto failed_mount;
3820         }
3821
3822         /*
3823          * It makes no sense for the first data block to be beyond the end
3824          * of the filesystem.
3825          */
3826         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3827                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3828                          "block %u is beyond end of filesystem (%llu)",
3829                          le32_to_cpu(es->s_first_data_block),
3830                          ext4_blocks_count(es));
3831                 goto failed_mount;
3832         }
3833         blocks_count = (ext4_blocks_count(es) -
3834                         le32_to_cpu(es->s_first_data_block) +
3835                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3836         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3837         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3838                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3839                        "(block count %llu, first data block %u, "
3840                        "blocks per group %lu)", sbi->s_groups_count,
3841                        ext4_blocks_count(es),
3842                        le32_to_cpu(es->s_first_data_block),
3843                        EXT4_BLOCKS_PER_GROUP(sb));
3844                 goto failed_mount;
3845         }
3846         sbi->s_groups_count = blocks_count;
3847         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3848                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3849         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3850                    EXT4_DESC_PER_BLOCK(sb);
3851         sbi->s_group_desc = ext4_kvmalloc(db_count *
3852                                           sizeof(struct buffer_head *),
3853                                           GFP_KERNEL);
3854         if (sbi->s_group_desc == NULL) {
3855                 ext4_msg(sb, KERN_ERR, "not enough memory");
3856                 ret = -ENOMEM;
3857                 goto failed_mount;
3858         }
3859
3860         if (ext4_proc_root)
3861                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3862
3863         if (sbi->s_proc)
3864                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3865                                  &ext4_seq_options_fops, sb);
3866
3867         bgl_lock_init(sbi->s_blockgroup_lock);
3868
3869         for (i = 0; i < db_count; i++) {
3870                 block = descriptor_loc(sb, logical_sb_block, i);
3871                 sbi->s_group_desc[i] = sb_bread(sb, block);
3872                 if (!sbi->s_group_desc[i]) {
3873                         ext4_msg(sb, KERN_ERR,
3874                                "can't read group descriptor %d", i);
3875                         db_count = i;
3876                         goto failed_mount2;
3877                 }
3878         }
3879         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3880                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3881                 goto failed_mount2;
3882         }
3883
3884         sbi->s_gdb_count = db_count;
3885         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3886         spin_lock_init(&sbi->s_next_gen_lock);
3887
3888         init_timer(&sbi->s_err_report);
3889         sbi->s_err_report.function = print_daily_error_info;
3890         sbi->s_err_report.data = (unsigned long) sb;
3891
3892         /* Register extent status tree shrinker */
3893         ext4_es_register_shrinker(sbi);
3894
3895         err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0, GFP_KERNEL);
3896         if (err) {
3897                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3898                 goto failed_mount3;
3899         }
3900
3901         sbi->s_stripe = ext4_get_stripe_size(sbi);
3902         sbi->s_extent_max_zeroout_kb = 32;
3903
3904         /*
3905          * set up enough so that it can read an inode
3906          */
3907         if (!test_opt(sb, NOLOAD) &&
3908             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3909                 sb->s_op = &ext4_sops;
3910         else
3911                 sb->s_op = &ext4_nojournal_sops;
3912         sb->s_export_op = &ext4_export_ops;
3913         sb->s_xattr = ext4_xattr_handlers;
3914 #ifdef CONFIG_QUOTA
3915         sb->dq_op = &ext4_quota_operations;
3916         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3917                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3918         else
3919                 sb->s_qcop = &ext4_qctl_operations;
3920 #endif
3921         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3922
3923         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3924         mutex_init(&sbi->s_orphan_lock);
3925
3926         sb->s_root = NULL;
3927
3928         needs_recovery = (es->s_last_orphan != 0 ||
3929                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3930                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3931
3932         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3933             !(sb->s_flags & MS_RDONLY))
3934                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3935                         goto failed_mount3;
3936
3937         /*
3938          * The first inode we look at is the journal inode.  Don't try
3939          * root first: it may be modified in the journal!
3940          */
3941         if (!test_opt(sb, NOLOAD) &&
3942             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3943                 if (ext4_load_journal(sb, es, journal_devnum))
3944                         goto failed_mount3;
3945         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3946               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3947                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3948                        "suppressed and not mounted read-only");
3949                 goto failed_mount_wq;
3950         } else {
3951                 clear_opt(sb, DATA_FLAGS);
3952                 sbi->s_journal = NULL;
3953                 needs_recovery = 0;
3954                 goto no_journal;
3955         }
3956
3957         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3958             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3959                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3960                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3961                 goto failed_mount_wq;
3962         }
3963
3964         if (!set_journal_csum_feature_set(sb)) {
3965                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3966                          "feature set");
3967                 goto failed_mount_wq;
3968         }
3969
3970         /* We have now updated the journal if required, so we can
3971          * validate the data journaling mode. */
3972         switch (test_opt(sb, DATA_FLAGS)) {
3973         case 0:
3974                 /* No mode set, assume a default based on the journal
3975                  * capabilities: ORDERED_DATA if the journal can
3976                  * cope, else JOURNAL_DATA
3977                  */
3978                 if (jbd2_journal_check_available_features
3979                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3980                         set_opt(sb, ORDERED_DATA);
3981                 else
3982                         set_opt(sb, JOURNAL_DATA);
3983                 break;
3984
3985         case EXT4_MOUNT_ORDERED_DATA:
3986         case EXT4_MOUNT_WRITEBACK_DATA:
3987                 if (!jbd2_journal_check_available_features
3988                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3989                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3990                                "requested data journaling mode");
3991                         goto failed_mount_wq;
3992                 }
3993         default:
3994                 break;
3995         }
3996         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3997
3998         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3999
4000 no_journal:
4001         if (ext4_mballoc_ready) {
4002                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4003                 if (!sbi->s_mb_cache) {
4004                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4005                         goto failed_mount_wq;
4006                 }
4007         }
4008
4009         /*
4010          * Get the # of file system overhead blocks from the
4011          * superblock if present.
4012          */
4013         if (es->s_overhead_clusters)
4014                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4015         else {
4016                 err = ext4_calculate_overhead(sb);
4017                 if (err)
4018                         goto failed_mount_wq;
4019         }
4020
4021         /*
4022          * The maximum number of concurrent works can be high and
4023          * concurrency isn't really necessary.  Limit it to 1.
4024          */
4025         EXT4_SB(sb)->rsv_conversion_wq =
4026                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4027         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4028                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4029                 ret = -ENOMEM;
4030                 goto failed_mount4;
4031         }
4032
4033         /*
4034          * The jbd2_journal_load will have done any necessary log recovery,
4035          * so we can safely mount the rest of the filesystem now.
4036          */
4037
4038         root = ext4_iget(sb, EXT4_ROOT_INO);
4039         if (IS_ERR(root)) {
4040                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4041                 ret = PTR_ERR(root);
4042                 root = NULL;
4043                 goto failed_mount4;
4044         }
4045         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4046                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4047                 iput(root);
4048                 goto failed_mount4;
4049         }
4050         sb->s_root = d_make_root(root);
4051         if (!sb->s_root) {
4052                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4053                 ret = -ENOMEM;
4054                 goto failed_mount4;
4055         }
4056
4057         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4058                 sb->s_flags |= MS_RDONLY;
4059
4060         /* determine the minimum size of new large inodes, if present */
4061         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4062                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4063                                                      EXT4_GOOD_OLD_INODE_SIZE;
4064                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4065                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4066                         if (sbi->s_want_extra_isize <
4067                             le16_to_cpu(es->s_want_extra_isize))
4068                                 sbi->s_want_extra_isize =
4069                                         le16_to_cpu(es->s_want_extra_isize);
4070                         if (sbi->s_want_extra_isize <
4071                             le16_to_cpu(es->s_min_extra_isize))
4072                                 sbi->s_want_extra_isize =
4073                                         le16_to_cpu(es->s_min_extra_isize);
4074                 }
4075         }
4076         /* Check if enough inode space is available */
4077         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4078                                                         sbi->s_inode_size) {
4079                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4080                                                        EXT4_GOOD_OLD_INODE_SIZE;
4081                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4082                          "available");
4083         }
4084
4085         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4086         if (err) {
4087                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4088                          "reserved pool", ext4_calculate_resv_clusters(sb));
4089                 goto failed_mount4a;
4090         }
4091
4092         err = ext4_setup_system_zone(sb);
4093         if (err) {
4094                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4095                          "zone (%d)", err);
4096                 goto failed_mount4a;
4097         }
4098
4099         ext4_ext_init(sb);
4100         err = ext4_mb_init(sb);
4101         if (err) {
4102                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4103                          err);
4104                 goto failed_mount5;
4105         }
4106
4107         block = ext4_count_free_clusters(sb);
4108         ext4_free_blocks_count_set(sbi->s_es,
4109                                    EXT4_C2B(sbi, block));
4110         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4111                                   GFP_KERNEL);
4112         if (!err) {
4113                 unsigned long freei = ext4_count_free_inodes(sb);
4114                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4115                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4116                                           GFP_KERNEL);
4117         }
4118         if (!err)
4119                 err = percpu_counter_init(&sbi->s_dirs_counter,
4120                                           ext4_count_dirs(sb), GFP_KERNEL);
4121         if (!err)
4122                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4123                                           GFP_KERNEL);
4124         if (err) {
4125                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4126                 goto failed_mount6;
4127         }
4128
4129         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4130                 if (!ext4_fill_flex_info(sb)) {
4131                         ext4_msg(sb, KERN_ERR,
4132                                "unable to initialize "
4133                                "flex_bg meta info!");
4134                         goto failed_mount6;
4135                 }
4136
4137         err = ext4_register_li_request(sb, first_not_zeroed);
4138         if (err)
4139                 goto failed_mount6;
4140
4141         sbi->s_kobj.kset = ext4_kset;
4142         init_completion(&sbi->s_kobj_unregister);
4143         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4144                                    "%s", sb->s_id);
4145         if (err)
4146                 goto failed_mount7;
4147
4148 #ifdef CONFIG_QUOTA
4149         /* Enable quota usage during mount. */
4150         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4151             !(sb->s_flags & MS_RDONLY)) {
4152                 err = ext4_enable_quotas(sb);
4153                 if (err)
4154                         goto failed_mount8;
4155         }
4156 #endif  /* CONFIG_QUOTA */
4157
4158         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4159         ext4_orphan_cleanup(sb, es);
4160         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4161         if (needs_recovery) {
4162                 ext4_msg(sb, KERN_INFO, "recovery complete");
4163                 ext4_mark_recovery_complete(sb, es);
4164         }
4165         if (EXT4_SB(sb)->s_journal) {
4166                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4167                         descr = " journalled data mode";
4168                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4169                         descr = " ordered data mode";
4170                 else
4171                         descr = " writeback data mode";
4172         } else
4173                 descr = "out journal";
4174
4175         if (test_opt(sb, DISCARD)) {
4176                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4177                 if (!blk_queue_discard(q))
4178                         ext4_msg(sb, KERN_WARNING,
4179                                  "mounting with \"discard\" option, but "
4180                                  "the device does not support discard");
4181         }
4182
4183         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4184                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4185                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4186
4187         if (es->s_error_count)
4188                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4189
4190         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4191         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4192         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4193         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4194
4195         kfree(orig_data);
4196         return 0;
4197
4198 cantfind_ext4:
4199         if (!silent)
4200                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4201         goto failed_mount;
4202
4203 #ifdef CONFIG_QUOTA
4204 failed_mount8:
4205         kobject_del(&sbi->s_kobj);
4206 #endif
4207 failed_mount7:
4208         ext4_unregister_li_request(sb);
4209 failed_mount6:
4210         ext4_mb_release(sb);
4211         if (sbi->s_flex_groups)
4212                 ext4_kvfree(sbi->s_flex_groups);
4213         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4214         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4215         percpu_counter_destroy(&sbi->s_dirs_counter);
4216         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4217 failed_mount5:
4218         ext4_ext_release(sb);
4219         ext4_release_system_zone(sb);
4220 failed_mount4a:
4221         dput(sb->s_root);
4222         sb->s_root = NULL;
4223 failed_mount4:
4224         ext4_msg(sb, KERN_ERR, "mount failed");
4225         if (EXT4_SB(sb)->rsv_conversion_wq)
4226                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4227 failed_mount_wq:
4228         if (sbi->s_journal) {
4229                 jbd2_journal_destroy(sbi->s_journal);
4230                 sbi->s_journal = NULL;
4231         }
4232 failed_mount3:
4233         ext4_es_unregister_shrinker(sbi);
4234         del_timer_sync(&sbi->s_err_report);
4235         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4236         if (sbi->s_mmp_tsk)
4237                 kthread_stop(sbi->s_mmp_tsk);
4238 failed_mount2:
4239         for (i = 0; i < db_count; i++)
4240                 brelse(sbi->s_group_desc[i]);
4241         ext4_kvfree(sbi->s_group_desc);
4242 failed_mount:
4243         if (sbi->s_chksum_driver)
4244                 crypto_free_shash(sbi->s_chksum_driver);
4245         if (sbi->s_proc) {
4246                 remove_proc_entry("options", sbi->s_proc);
4247                 remove_proc_entry(sb->s_id, ext4_proc_root);
4248         }
4249 #ifdef CONFIG_QUOTA
4250         for (i = 0; i < MAXQUOTAS; i++)
4251                 kfree(sbi->s_qf_names[i]);
4252 #endif
4253         ext4_blkdev_remove(sbi);
4254         brelse(bh);
4255 out_fail:
4256         sb->s_fs_info = NULL;
4257         kfree(sbi->s_blockgroup_lock);
4258         kfree(sbi);
4259 out_free_orig:
4260         kfree(orig_data);
4261         return err ? err : ret;
4262 }
4263
4264 /*
4265  * Setup any per-fs journal parameters now.  We'll do this both on
4266  * initial mount, once the journal has been initialised but before we've
4267  * done any recovery; and again on any subsequent remount.
4268  */
4269 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4270 {
4271         struct ext4_sb_info *sbi = EXT4_SB(sb);
4272
4273         journal->j_commit_interval = sbi->s_commit_interval;
4274         journal->j_min_batch_time = sbi->s_min_batch_time;
4275         journal->j_max_batch_time = sbi->s_max_batch_time;
4276
4277         write_lock(&journal->j_state_lock);
4278         if (test_opt(sb, BARRIER))
4279                 journal->j_flags |= JBD2_BARRIER;
4280         else
4281                 journal->j_flags &= ~JBD2_BARRIER;
4282         if (test_opt(sb, DATA_ERR_ABORT))
4283                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4284         else
4285                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4286         write_unlock(&journal->j_state_lock);
4287 }
4288
4289 static journal_t *ext4_get_journal(struct super_block *sb,
4290                                    unsigned int journal_inum)
4291 {
4292         struct inode *journal_inode;
4293         journal_t *journal;
4294
4295         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4296
4297         /* First, test for the existence of a valid inode on disk.  Bad
4298          * things happen if we iget() an unused inode, as the subsequent
4299          * iput() will try to delete it. */
4300
4301         journal_inode = ext4_iget(sb, journal_inum);
4302         if (IS_ERR(journal_inode)) {
4303                 ext4_msg(sb, KERN_ERR, "no journal found");
4304                 return NULL;
4305         }
4306         if (!journal_inode->i_nlink) {
4307                 make_bad_inode(journal_inode);
4308                 iput(journal_inode);
4309                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4310                 return NULL;
4311         }
4312
4313         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4314                   journal_inode, journal_inode->i_size);
4315         if (!S_ISREG(journal_inode->i_mode)) {
4316                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4317                 iput(journal_inode);
4318                 return NULL;
4319         }
4320
4321         journal = jbd2_journal_init_inode(journal_inode);
4322         if (!journal) {
4323                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4324                 iput(journal_inode);
4325                 return NULL;
4326         }
4327         journal->j_private = sb;
4328         ext4_init_journal_params(sb, journal);
4329         return journal;
4330 }
4331
4332 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4333                                        dev_t j_dev)
4334 {
4335         struct buffer_head *bh;
4336         journal_t *journal;
4337         ext4_fsblk_t start;
4338         ext4_fsblk_t len;
4339         int hblock, blocksize;
4340         ext4_fsblk_t sb_block;
4341         unsigned long offset;
4342         struct ext4_super_block *es;
4343         struct block_device *bdev;
4344
4345         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4346
4347         bdev = ext4_blkdev_get(j_dev, sb);
4348         if (bdev == NULL)
4349                 return NULL;
4350
4351         blocksize = sb->s_blocksize;
4352         hblock = bdev_logical_block_size(bdev);
4353         if (blocksize < hblock) {
4354                 ext4_msg(sb, KERN_ERR,
4355                         "blocksize too small for journal device");
4356                 goto out_bdev;
4357         }
4358
4359         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4360         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4361         set_blocksize(bdev, blocksize);
4362         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4363                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4364                        "external journal");
4365                 goto out_bdev;
4366         }
4367
4368         es = (struct ext4_super_block *) (bh->b_data + offset);
4369         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4370             !(le32_to_cpu(es->s_feature_incompat) &
4371               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4372                 ext4_msg(sb, KERN_ERR, "external journal has "
4373                                         "bad superblock");
4374                 brelse(bh);
4375                 goto out_bdev;
4376         }
4377
4378         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4379                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4380                 brelse(bh);
4381                 goto out_bdev;
4382         }
4383
4384         len = ext4_blocks_count(es);
4385         start = sb_block + 1;
4386         brelse(bh);     /* we're done with the superblock */
4387
4388         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4389                                         start, len, blocksize);
4390         if (!journal) {
4391                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4392                 goto out_bdev;
4393         }
4394         journal->j_private = sb;
4395         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4396         wait_on_buffer(journal->j_sb_buffer);
4397         if (!buffer_uptodate(journal->j_sb_buffer)) {
4398                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4399                 goto out_journal;
4400         }
4401         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4402                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4403                                         "user (unsupported) - %d",
4404                         be32_to_cpu(journal->j_superblock->s_nr_users));
4405                 goto out_journal;
4406         }
4407         EXT4_SB(sb)->journal_bdev = bdev;
4408         ext4_init_journal_params(sb, journal);
4409         return journal;
4410
4411 out_journal:
4412         jbd2_journal_destroy(journal);
4413 out_bdev:
4414         ext4_blkdev_put(bdev);
4415         return NULL;
4416 }
4417
4418 static int ext4_load_journal(struct super_block *sb,
4419                              struct ext4_super_block *es,
4420                              unsigned long journal_devnum)
4421 {
4422         journal_t *journal;
4423         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4424         dev_t journal_dev;
4425         int err = 0;
4426         int really_read_only;
4427
4428         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4429
4430         if (journal_devnum &&
4431             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4432                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4433                         "numbers have changed");
4434                 journal_dev = new_decode_dev(journal_devnum);
4435         } else
4436                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4437
4438         really_read_only = bdev_read_only(sb->s_bdev);
4439
4440         /*
4441          * Are we loading a blank journal or performing recovery after a
4442          * crash?  For recovery, we need to check in advance whether we
4443          * can get read-write access to the device.
4444          */
4445         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4446                 if (sb->s_flags & MS_RDONLY) {
4447                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4448                                         "required on readonly filesystem");
4449                         if (really_read_only) {
4450                                 ext4_msg(sb, KERN_ERR, "write access "
4451                                         "unavailable, cannot proceed");
4452                                 return -EROFS;
4453                         }
4454                         ext4_msg(sb, KERN_INFO, "write access will "
4455                                "be enabled during recovery");
4456                 }
4457         }
4458
4459         if (journal_inum && journal_dev) {
4460                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4461                        "and inode journals!");
4462                 return -EINVAL;
4463         }
4464
4465         if (journal_inum) {
4466                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4467                         return -EINVAL;
4468         } else {
4469                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4470                         return -EINVAL;
4471         }
4472
4473         if (!(journal->j_flags & JBD2_BARRIER))
4474                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4475
4476         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4477                 err = jbd2_journal_wipe(journal, !really_read_only);
4478         if (!err) {
4479                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4480                 if (save)
4481                         memcpy(save, ((char *) es) +
4482                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4483                 err = jbd2_journal_load(journal);
4484                 if (save)
4485                         memcpy(((char *) es) + EXT4_S_ERR_START,
4486                                save, EXT4_S_ERR_LEN);
4487                 kfree(save);
4488         }
4489
4490         if (err) {
4491                 ext4_msg(sb, KERN_ERR, "error loading journal");
4492                 jbd2_journal_destroy(journal);
4493                 return err;
4494         }
4495
4496         EXT4_SB(sb)->s_journal = journal;
4497         ext4_clear_journal_err(sb, es);
4498
4499         if (!really_read_only && journal_devnum &&
4500             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4501                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4502
4503                 /* Make sure we flush the recovery flag to disk. */
4504                 ext4_commit_super(sb, 1);
4505         }
4506
4507         return 0;
4508 }
4509
4510 static int ext4_commit_super(struct super_block *sb, int sync)
4511 {
4512         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4513         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4514         int error = 0;
4515
4516         if (!sbh || block_device_ejected(sb))
4517                 return error;
4518         if (buffer_write_io_error(sbh)) {
4519                 /*
4520                  * Oh, dear.  A previous attempt to write the
4521                  * superblock failed.  This could happen because the
4522                  * USB device was yanked out.  Or it could happen to
4523                  * be a transient write error and maybe the block will
4524                  * be remapped.  Nothing we can do but to retry the
4525                  * write and hope for the best.
4526                  */
4527                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4528                        "superblock detected");
4529                 clear_buffer_write_io_error(sbh);
4530                 set_buffer_uptodate(sbh);
4531         }
4532         /*
4533          * If the file system is mounted read-only, don't update the
4534          * superblock write time.  This avoids updating the superblock
4535          * write time when we are mounting the root file system
4536          * read/only but we need to replay the journal; at that point,
4537          * for people who are east of GMT and who make their clock
4538          * tick in localtime for Windows bug-for-bug compatibility,
4539          * the clock is set in the future, and this will cause e2fsck
4540          * to complain and force a full file system check.
4541          */
4542         if (!(sb->s_flags & MS_RDONLY))
4543                 es->s_wtime = cpu_to_le32(get_seconds());
4544         if (sb->s_bdev->bd_part)
4545                 es->s_kbytes_written =
4546                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4547                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4548                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4549         else
4550                 es->s_kbytes_written =
4551                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4552         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4553                 ext4_free_blocks_count_set(es,
4554                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4555                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4556         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4557                 es->s_free_inodes_count =
4558                         cpu_to_le32(percpu_counter_sum_positive(
4559                                 &EXT4_SB(sb)->s_freeinodes_counter));
4560         BUFFER_TRACE(sbh, "marking dirty");
4561         ext4_superblock_csum_set(sb);
4562         mark_buffer_dirty(sbh);
4563         if (sync) {
4564                 error = sync_dirty_buffer(sbh);
4565                 if (error)
4566                         return error;
4567
4568                 error = buffer_write_io_error(sbh);
4569                 if (error) {
4570                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4571                                "superblock");
4572                         clear_buffer_write_io_error(sbh);
4573                         set_buffer_uptodate(sbh);
4574                 }
4575         }
4576         return error;
4577 }
4578
4579 /*
4580  * Have we just finished recovery?  If so, and if we are mounting (or
4581  * remounting) the filesystem readonly, then we will end up with a
4582  * consistent fs on disk.  Record that fact.
4583  */
4584 static void ext4_mark_recovery_complete(struct super_block *sb,
4585                                         struct ext4_super_block *es)
4586 {
4587         journal_t *journal = EXT4_SB(sb)->s_journal;
4588
4589         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4590                 BUG_ON(journal != NULL);
4591                 return;
4592         }
4593         jbd2_journal_lock_updates(journal);
4594         if (jbd2_journal_flush(journal) < 0)
4595                 goto out;
4596
4597         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4598             sb->s_flags & MS_RDONLY) {
4599                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4600                 ext4_commit_super(sb, 1);
4601         }
4602
4603 out:
4604         jbd2_journal_unlock_updates(journal);
4605 }
4606
4607 /*
4608  * If we are mounting (or read-write remounting) a filesystem whose journal
4609  * has recorded an error from a previous lifetime, move that error to the
4610  * main filesystem now.
4611  */
4612 static void ext4_clear_journal_err(struct super_block *sb,
4613                                    struct ext4_super_block *es)
4614 {
4615         journal_t *journal;
4616         int j_errno;
4617         const char *errstr;
4618
4619         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4620
4621         journal = EXT4_SB(sb)->s_journal;
4622
4623         /*
4624          * Now check for any error status which may have been recorded in the
4625          * journal by a prior ext4_error() or ext4_abort()
4626          */
4627
4628         j_errno = jbd2_journal_errno(journal);
4629         if (j_errno) {
4630                 char nbuf[16];
4631
4632                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4633                 ext4_warning(sb, "Filesystem error recorded "
4634                              "from previous mount: %s", errstr);
4635                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4636
4637                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4638                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4639                 ext4_commit_super(sb, 1);
4640
4641                 jbd2_journal_clear_err(journal);
4642                 jbd2_journal_update_sb_errno(journal);
4643         }
4644 }
4645
4646 /*
4647  * Force the running and committing transactions to commit,
4648  * and wait on the commit.
4649  */
4650 int ext4_force_commit(struct super_block *sb)
4651 {
4652         journal_t *journal;
4653
4654         if (sb->s_flags & MS_RDONLY)
4655                 return 0;
4656
4657         journal = EXT4_SB(sb)->s_journal;
4658         return ext4_journal_force_commit(journal);
4659 }
4660
4661 static int ext4_sync_fs(struct super_block *sb, int wait)
4662 {
4663         int ret = 0;
4664         tid_t target;
4665         bool needs_barrier = false;
4666         struct ext4_sb_info *sbi = EXT4_SB(sb);
4667
4668         trace_ext4_sync_fs(sb, wait);
4669         flush_workqueue(sbi->rsv_conversion_wq);
4670         /*
4671          * Writeback quota in non-journalled quota case - journalled quota has
4672          * no dirty dquots
4673          */
4674         dquot_writeback_dquots(sb, -1);
4675         /*
4676          * Data writeback is possible w/o journal transaction, so barrier must
4677          * being sent at the end of the function. But we can skip it if
4678          * transaction_commit will do it for us.
4679          */
4680         target = jbd2_get_latest_transaction(sbi->s_journal);
4681         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4682             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4683                 needs_barrier = true;
4684
4685         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4686                 if (wait)
4687                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4688         }
4689         if (needs_barrier) {
4690                 int err;
4691                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4692                 if (!ret)
4693                         ret = err;
4694         }
4695
4696         return ret;
4697 }
4698
4699 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4700 {
4701         int ret = 0;
4702
4703         trace_ext4_sync_fs(sb, wait);
4704         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4705         dquot_writeback_dquots(sb, -1);
4706         if (wait && test_opt(sb, BARRIER))
4707                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4708
4709         return ret;
4710 }
4711
4712 /*
4713  * LVM calls this function before a (read-only) snapshot is created.  This
4714  * gives us a chance to flush the journal completely and mark the fs clean.
4715  *
4716  * Note that only this function cannot bring a filesystem to be in a clean
4717  * state independently. It relies on upper layer to stop all data & metadata
4718  * modifications.
4719  */
4720 static int ext4_freeze(struct super_block *sb)
4721 {
4722         int error = 0;
4723         journal_t *journal;
4724
4725         if (sb->s_flags & MS_RDONLY)
4726                 return 0;
4727
4728         journal = EXT4_SB(sb)->s_journal;
4729
4730         /* Now we set up the journal barrier. */
4731         jbd2_journal_lock_updates(journal);
4732
4733         /*
4734          * Don't clear the needs_recovery flag if we failed to flush
4735          * the journal.
4736          */
4737         error = jbd2_journal_flush(journal);
4738         if (error < 0)
4739                 goto out;
4740
4741         /* Journal blocked and flushed, clear needs_recovery flag. */
4742         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4743         error = ext4_commit_super(sb, 1);
4744 out:
4745         /* we rely on upper layer to stop further updates */
4746         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4747         return error;
4748 }
4749
4750 /*
4751  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4752  * flag here, even though the filesystem is not technically dirty yet.
4753  */
4754 static int ext4_unfreeze(struct super_block *sb)
4755 {
4756         if (sb->s_flags & MS_RDONLY)
4757                 return 0;
4758
4759         /* Reset the needs_recovery flag before the fs is unlocked. */
4760         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4761         ext4_commit_super(sb, 1);
4762         return 0;
4763 }
4764
4765 /*
4766  * Structure to save mount options for ext4_remount's benefit
4767  */
4768 struct ext4_mount_options {
4769         unsigned long s_mount_opt;
4770         unsigned long s_mount_opt2;
4771         kuid_t s_resuid;
4772         kgid_t s_resgid;
4773         unsigned long s_commit_interval;
4774         u32 s_min_batch_time, s_max_batch_time;
4775 #ifdef CONFIG_QUOTA
4776         int s_jquota_fmt;
4777         char *s_qf_names[MAXQUOTAS];
4778 #endif
4779 };
4780
4781 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4782 {
4783         struct ext4_super_block *es;
4784         struct ext4_sb_info *sbi = EXT4_SB(sb);
4785         unsigned long old_sb_flags;
4786         struct ext4_mount_options old_opts;
4787         int enable_quota = 0;
4788         ext4_group_t g;
4789         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4790         int err = 0;
4791 #ifdef CONFIG_QUOTA
4792         int i, j;
4793 #endif
4794         char *orig_data = kstrdup(data, GFP_KERNEL);
4795
4796         /* Store the original options */
4797         old_sb_flags = sb->s_flags;
4798         old_opts.s_mount_opt = sbi->s_mount_opt;
4799         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4800         old_opts.s_resuid = sbi->s_resuid;
4801         old_opts.s_resgid = sbi->s_resgid;
4802         old_opts.s_commit_interval = sbi->s_commit_interval;
4803         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4804         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4805 #ifdef CONFIG_QUOTA
4806         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4807         for (i = 0; i < MAXQUOTAS; i++)
4808                 if (sbi->s_qf_names[i]) {
4809                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4810                                                          GFP_KERNEL);
4811                         if (!old_opts.s_qf_names[i]) {
4812                                 for (j = 0; j < i; j++)
4813                                         kfree(old_opts.s_qf_names[j]);
4814                                 kfree(orig_data);
4815                                 return -ENOMEM;
4816                         }
4817                 } else
4818                         old_opts.s_qf_names[i] = NULL;
4819 #endif
4820         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4821                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4822
4823         /*
4824          * Allow the "check" option to be passed as a remount option.
4825          */
4826         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4827                 err = -EINVAL;
4828                 goto restore_opts;
4829         }
4830
4831         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4832                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4833                         ext4_msg(sb, KERN_ERR, "can't mount with "
4834                                  "both data=journal and delalloc");
4835                         err = -EINVAL;
4836                         goto restore_opts;
4837                 }
4838                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4839                         ext4_msg(sb, KERN_ERR, "can't mount with "
4840                                  "both data=journal and dioread_nolock");
4841                         err = -EINVAL;
4842                         goto restore_opts;
4843                 }
4844         }
4845
4846         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4847                 ext4_abort(sb, "Abort forced by user");
4848
4849         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4850                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4851
4852         es = sbi->s_es;
4853
4854         if (sbi->s_journal) {
4855                 ext4_init_journal_params(sb, sbi->s_journal);
4856                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4857         }
4858
4859         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4860                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4861                         err = -EROFS;
4862                         goto restore_opts;
4863                 }
4864
4865                 if (*flags & MS_RDONLY) {
4866                         err = sync_filesystem(sb);
4867                         if (err < 0)
4868                                 goto restore_opts;
4869                         err = dquot_suspend(sb, -1);
4870                         if (err < 0)
4871                                 goto restore_opts;
4872
4873                         /*
4874                          * First of all, the unconditional stuff we have to do
4875                          * to disable replay of the journal when we next remount
4876                          */
4877                         sb->s_flags |= MS_RDONLY;
4878
4879                         /*
4880                          * OK, test if we are remounting a valid rw partition
4881                          * readonly, and if so set the rdonly flag and then
4882                          * mark the partition as valid again.
4883                          */
4884                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4885                             (sbi->s_mount_state & EXT4_VALID_FS))
4886                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4887
4888                         if (sbi->s_journal)
4889                                 ext4_mark_recovery_complete(sb, es);
4890                 } else {
4891                         /* Make sure we can mount this feature set readwrite */
4892                         if (!ext4_feature_set_ok(sb, 0)) {
4893                                 err = -EROFS;
4894                                 goto restore_opts;
4895                         }
4896                         /*
4897                          * Make sure the group descriptor checksums
4898                          * are sane.  If they aren't, refuse to remount r/w.
4899                          */
4900                         for (g = 0; g < sbi->s_groups_count; g++) {
4901                                 struct ext4_group_desc *gdp =
4902                                         ext4_get_group_desc(sb, g, NULL);
4903
4904                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4905                                         ext4_msg(sb, KERN_ERR,
4906                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4907                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4908                                                le16_to_cpu(gdp->bg_checksum));
4909                                         err = -EINVAL;
4910                                         goto restore_opts;
4911                                 }
4912                         }
4913
4914                         /*
4915                          * If we have an unprocessed orphan list hanging
4916                          * around from a previously readonly bdev mount,
4917                          * require a full umount/remount for now.
4918                          */
4919                         if (es->s_last_orphan) {
4920                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4921                                        "remount RDWR because of unprocessed "
4922                                        "orphan inode list.  Please "
4923                                        "umount/remount instead");
4924                                 err = -EINVAL;
4925                                 goto restore_opts;
4926                         }
4927
4928                         /*
4929                          * Mounting a RDONLY partition read-write, so reread
4930                          * and store the current valid flag.  (It may have
4931                          * been changed by e2fsck since we originally mounted
4932                          * the partition.)
4933                          */
4934                         if (sbi->s_journal)
4935                                 ext4_clear_journal_err(sb, es);
4936                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4937                         if (!ext4_setup_super(sb, es, 0))
4938                                 sb->s_flags &= ~MS_RDONLY;
4939                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4940                                                      EXT4_FEATURE_INCOMPAT_MMP))
4941                                 if (ext4_multi_mount_protect(sb,
4942                                                 le64_to_cpu(es->s_mmp_block))) {
4943                                         err = -EROFS;
4944                                         goto restore_opts;
4945                                 }
4946                         enable_quota = 1;
4947                 }
4948         }
4949
4950         /*
4951          * Reinitialize lazy itable initialization thread based on
4952          * current settings
4953          */
4954         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4955                 ext4_unregister_li_request(sb);
4956         else {
4957                 ext4_group_t first_not_zeroed;
4958                 first_not_zeroed = ext4_has_uninit_itable(sb);
4959                 ext4_register_li_request(sb, first_not_zeroed);
4960         }
4961
4962         ext4_setup_system_zone(sb);
4963         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4964                 ext4_commit_super(sb, 1);
4965
4966 #ifdef CONFIG_QUOTA
4967         /* Release old quota file names */
4968         for (i = 0; i < MAXQUOTAS; i++)
4969                 kfree(old_opts.s_qf_names[i]);
4970         if (enable_quota) {
4971                 if (sb_any_quota_suspended(sb))
4972                         dquot_resume(sb, -1);
4973                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4974                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4975                         err = ext4_enable_quotas(sb);
4976                         if (err)
4977                                 goto restore_opts;
4978                 }
4979         }
4980 #endif
4981
4982         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4983         kfree(orig_data);
4984         return 0;
4985
4986 restore_opts:
4987         sb->s_flags = old_sb_flags;
4988         sbi->s_mount_opt = old_opts.s_mount_opt;
4989         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4990         sbi->s_resuid = old_opts.s_resuid;
4991         sbi->s_resgid = old_opts.s_resgid;
4992         sbi->s_commit_interval = old_opts.s_commit_interval;
4993         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4994         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4995 #ifdef CONFIG_QUOTA
4996         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4997         for (i = 0; i < MAXQUOTAS; i++) {
4998                 kfree(sbi->s_qf_names[i]);
4999                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5000         }
5001 #endif
5002         kfree(orig_data);
5003         return err;
5004 }
5005
5006 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5007 {
5008         struct super_block *sb = dentry->d_sb;
5009         struct ext4_sb_info *sbi = EXT4_SB(sb);
5010         struct ext4_super_block *es = sbi->s_es;
5011         ext4_fsblk_t overhead = 0, resv_blocks;
5012         u64 fsid;
5013         s64 bfree;
5014         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5015
5016         if (!test_opt(sb, MINIX_DF))
5017                 overhead = sbi->s_overhead;
5018
5019         buf->f_type = EXT4_SUPER_MAGIC;
5020         buf->f_bsize = sb->s_blocksize;
5021         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5022         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5023                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5024         /* prevent underflow in case that few free space is available */
5025         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5026         buf->f_bavail = buf->f_bfree -
5027                         (ext4_r_blocks_count(es) + resv_blocks);
5028         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5029                 buf->f_bavail = 0;
5030         buf->f_files = le32_to_cpu(es->s_inodes_count);
5031         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5032         buf->f_namelen = EXT4_NAME_LEN;
5033         fsid = le64_to_cpup((void *)es->s_uuid) ^
5034                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5035         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5036         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5037
5038         return 0;
5039 }
5040
5041 /* Helper function for writing quotas on sync - we need to start transaction
5042  * before quota file is locked for write. Otherwise the are possible deadlocks:
5043  * Process 1                         Process 2
5044  * ext4_create()                     quota_sync()
5045  *   jbd2_journal_start()                  write_dquot()
5046  *   dquot_initialize()                         down(dqio_mutex)
5047  *     down(dqio_mutex)                    jbd2_journal_start()
5048  *
5049  */
5050
5051 #ifdef CONFIG_QUOTA
5052
5053 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5054 {
5055         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5056 }
5057
5058 static int ext4_write_dquot(struct dquot *dquot)
5059 {
5060         int ret, err;
5061         handle_t *handle;
5062         struct inode *inode;
5063
5064         inode = dquot_to_inode(dquot);
5065         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5066                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5067         if (IS_ERR(handle))
5068                 return PTR_ERR(handle);
5069         ret = dquot_commit(dquot);
5070         err = ext4_journal_stop(handle);
5071         if (!ret)
5072                 ret = err;
5073         return ret;
5074 }
5075
5076 static int ext4_acquire_dquot(struct dquot *dquot)
5077 {
5078         int ret, err;
5079         handle_t *handle;
5080
5081         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5082                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5083         if (IS_ERR(handle))
5084                 return PTR_ERR(handle);
5085         ret = dquot_acquire(dquot);
5086         err = ext4_journal_stop(handle);
5087         if (!ret)
5088                 ret = err;
5089         return ret;
5090 }
5091
5092 static int ext4_release_dquot(struct dquot *dquot)
5093 {
5094         int ret, err;
5095         handle_t *handle;
5096
5097         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5098                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5099         if (IS_ERR(handle)) {
5100                 /* Release dquot anyway to avoid endless cycle in dqput() */
5101                 dquot_release(dquot);
5102                 return PTR_ERR(handle);
5103         }
5104         ret = dquot_release(dquot);
5105         err = ext4_journal_stop(handle);
5106         if (!ret)
5107                 ret = err;
5108         return ret;
5109 }
5110
5111 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5112 {
5113         struct super_block *sb = dquot->dq_sb;
5114         struct ext4_sb_info *sbi = EXT4_SB(sb);
5115
5116         /* Are we journaling quotas? */
5117         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5118             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5119                 dquot_mark_dquot_dirty(dquot);
5120                 return ext4_write_dquot(dquot);
5121         } else {
5122                 return dquot_mark_dquot_dirty(dquot);
5123         }
5124 }
5125
5126 static int ext4_write_info(struct super_block *sb, int type)
5127 {
5128         int ret, err;
5129         handle_t *handle;
5130
5131         /* Data block + inode block */
5132         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5133         if (IS_ERR(handle))
5134                 return PTR_ERR(handle);
5135         ret = dquot_commit_info(sb, type);
5136         err = ext4_journal_stop(handle);
5137         if (!ret)
5138                 ret = err;
5139         return ret;
5140 }
5141
5142 /*
5143  * Turn on quotas during mount time - we need to find
5144  * the quota file and such...
5145  */
5146 static int ext4_quota_on_mount(struct super_block *sb, int type)
5147 {
5148         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5149                                         EXT4_SB(sb)->s_jquota_fmt, type);
5150 }
5151
5152 /*
5153  * Standard function to be called on quota_on
5154  */
5155 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5156                          struct path *path)
5157 {
5158         int err;
5159
5160         if (!test_opt(sb, QUOTA))
5161                 return -EINVAL;
5162
5163         /* Quotafile not on the same filesystem? */
5164         if (path->dentry->d_sb != sb)
5165                 return -EXDEV;
5166         /* Journaling quota? */
5167         if (EXT4_SB(sb)->s_qf_names[type]) {
5168                 /* Quotafile not in fs root? */
5169                 if (path->dentry->d_parent != sb->s_root)
5170                         ext4_msg(sb, KERN_WARNING,
5171                                 "Quota file not on filesystem root. "
5172                                 "Journaled quota will not work");
5173         }
5174
5175         /*
5176          * When we journal data on quota file, we have to flush journal to see
5177          * all updates to the file when we bypass pagecache...
5178          */
5179         if (EXT4_SB(sb)->s_journal &&
5180             ext4_should_journal_data(path->dentry->d_inode)) {
5181                 /*
5182                  * We don't need to lock updates but journal_flush() could
5183                  * otherwise be livelocked...
5184                  */
5185                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5186                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5187                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5188                 if (err)
5189                         return err;
5190         }
5191
5192         return dquot_quota_on(sb, type, format_id, path);
5193 }
5194
5195 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5196                              unsigned int flags)
5197 {
5198         int err;
5199         struct inode *qf_inode;
5200         unsigned long qf_inums[MAXQUOTAS] = {
5201                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5202                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5203         };
5204
5205         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5206
5207         if (!qf_inums[type])
5208                 return -EPERM;
5209
5210         qf_inode = ext4_iget(sb, qf_inums[type]);
5211         if (IS_ERR(qf_inode)) {
5212                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5213                 return PTR_ERR(qf_inode);
5214         }
5215
5216         /* Don't account quota for quota files to avoid recursion */
5217         qf_inode->i_flags |= S_NOQUOTA;
5218         err = dquot_enable(qf_inode, type, format_id, flags);
5219         iput(qf_inode);
5220
5221         return err;
5222 }
5223
5224 /* Enable usage tracking for all quota types. */
5225 static int ext4_enable_quotas(struct super_block *sb)
5226 {
5227         int type, err = 0;
5228         unsigned long qf_inums[MAXQUOTAS] = {
5229                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5230                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5231         };
5232
5233         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5234         for (type = 0; type < MAXQUOTAS; type++) {
5235                 if (qf_inums[type]) {
5236                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5237                                                 DQUOT_USAGE_ENABLED);
5238                         if (err) {
5239                                 ext4_warning(sb,
5240                                         "Failed to enable quota tracking "
5241                                         "(type=%d, err=%d). Please run "
5242                                         "e2fsck to fix.", type, err);
5243                                 return err;
5244                         }
5245                 }
5246         }
5247         return 0;
5248 }
5249
5250 /*
5251  * quota_on function that is used when QUOTA feature is set.
5252  */
5253 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5254                                  int format_id)
5255 {
5256         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5257                 return -EINVAL;
5258
5259         /*
5260          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5261          */
5262         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5263 }
5264
5265 static int ext4_quota_off(struct super_block *sb, int type)
5266 {
5267         struct inode *inode = sb_dqopt(sb)->files[type];
5268         handle_t *handle;
5269
5270         /* Force all delayed allocation blocks to be allocated.
5271          * Caller already holds s_umount sem */
5272         if (test_opt(sb, DELALLOC))
5273                 sync_filesystem(sb);
5274
5275         if (!inode)
5276                 goto out;
5277
5278         /* Update modification times of quota files when userspace can
5279          * start looking at them */
5280         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5281         if (IS_ERR(handle))
5282                 goto out;
5283         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5284         ext4_mark_inode_dirty(handle, inode);
5285         ext4_journal_stop(handle);
5286
5287 out:
5288         return dquot_quota_off(sb, type);
5289 }
5290
5291 /*
5292  * quota_off function that is used when QUOTA feature is set.
5293  */
5294 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5295 {
5296         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5297                 return -EINVAL;
5298
5299         /* Disable only the limits. */
5300         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5301 }
5302
5303 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5304  * acquiring the locks... As quota files are never truncated and quota code
5305  * itself serializes the operations (and no one else should touch the files)
5306  * we don't have to be afraid of races */
5307 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5308                                size_t len, loff_t off)
5309 {
5310         struct inode *inode = sb_dqopt(sb)->files[type];
5311         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5312         int err = 0;
5313         int offset = off & (sb->s_blocksize - 1);
5314         int tocopy;
5315         size_t toread;
5316         struct buffer_head *bh;
5317         loff_t i_size = i_size_read(inode);
5318
5319         if (off > i_size)
5320                 return 0;
5321         if (off+len > i_size)
5322                 len = i_size-off;
5323         toread = len;
5324         while (toread > 0) {
5325                 tocopy = sb->s_blocksize - offset < toread ?
5326                                 sb->s_blocksize - offset : toread;
5327                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5328                 if (err)
5329                         return err;
5330                 if (!bh)        /* A hole? */
5331                         memset(data, 0, tocopy);
5332                 else
5333                         memcpy(data, bh->b_data+offset, tocopy);
5334                 brelse(bh);
5335                 offset = 0;
5336                 toread -= tocopy;
5337                 data += tocopy;
5338                 blk++;
5339         }
5340         return len;
5341 }
5342
5343 /* Write to quotafile (we know the transaction is already started and has
5344  * enough credits) */
5345 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5346                                 const char *data, size_t len, loff_t off)
5347 {
5348         struct inode *inode = sb_dqopt(sb)->files[type];
5349         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5350         int err = 0;
5351         int offset = off & (sb->s_blocksize - 1);
5352         struct buffer_head *bh;
5353         handle_t *handle = journal_current_handle();
5354
5355         if (EXT4_SB(sb)->s_journal && !handle) {
5356                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5357                         " cancelled because transaction is not started",
5358                         (unsigned long long)off, (unsigned long long)len);
5359                 return -EIO;
5360         }
5361         /*
5362          * Since we account only one data block in transaction credits,
5363          * then it is impossible to cross a block boundary.
5364          */
5365         if (sb->s_blocksize - offset < len) {
5366                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5367                         " cancelled because not block aligned",
5368                         (unsigned long long)off, (unsigned long long)len);
5369                 return -EIO;
5370         }
5371
5372         bh = ext4_bread(handle, inode, blk, 1, &err);
5373         if (!bh)
5374                 goto out;
5375         BUFFER_TRACE(bh, "get write access");
5376         err = ext4_journal_get_write_access(handle, bh);
5377         if (err) {
5378                 brelse(bh);
5379                 goto out;
5380         }
5381         lock_buffer(bh);
5382         memcpy(bh->b_data+offset, data, len);
5383         flush_dcache_page(bh->b_page);
5384         unlock_buffer(bh);
5385         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5386         brelse(bh);
5387 out:
5388         if (err)
5389                 return err;
5390         if (inode->i_size < off + len) {
5391                 i_size_write(inode, off + len);
5392                 EXT4_I(inode)->i_disksize = inode->i_size;
5393                 ext4_mark_inode_dirty(handle, inode);
5394         }
5395         return len;
5396 }
5397
5398 #endif
5399
5400 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5401                        const char *dev_name, void *data)
5402 {
5403         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5404 }
5405
5406 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5407 static inline void register_as_ext2(void)
5408 {
5409         int err = register_filesystem(&ext2_fs_type);
5410         if (err)
5411                 printk(KERN_WARNING
5412                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5413 }
5414
5415 static inline void unregister_as_ext2(void)
5416 {
5417         unregister_filesystem(&ext2_fs_type);
5418 }
5419
5420 static inline int ext2_feature_set_ok(struct super_block *sb)
5421 {
5422         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5423                 return 0;
5424         if (sb->s_flags & MS_RDONLY)
5425                 return 1;
5426         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5427                 return 0;
5428         return 1;
5429 }
5430 #else
5431 static inline void register_as_ext2(void) { }
5432 static inline void unregister_as_ext2(void) { }
5433 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5434 #endif
5435
5436 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5437 static inline void register_as_ext3(void)
5438 {
5439         int err = register_filesystem(&ext3_fs_type);
5440         if (err)
5441                 printk(KERN_WARNING
5442                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5443 }
5444
5445 static inline void unregister_as_ext3(void)
5446 {
5447         unregister_filesystem(&ext3_fs_type);
5448 }
5449
5450 static inline int ext3_feature_set_ok(struct super_block *sb)
5451 {
5452         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5453                 return 0;
5454         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5455                 return 0;
5456         if (sb->s_flags & MS_RDONLY)
5457                 return 1;
5458         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5459                 return 0;
5460         return 1;
5461 }
5462 #else
5463 static inline void register_as_ext3(void) { }
5464 static inline void unregister_as_ext3(void) { }
5465 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5466 #endif
5467
5468 static struct file_system_type ext4_fs_type = {
5469         .owner          = THIS_MODULE,
5470         .name           = "ext4",
5471         .mount          = ext4_mount,
5472         .kill_sb        = kill_block_super,
5473         .fs_flags       = FS_REQUIRES_DEV,
5474 };
5475 MODULE_ALIAS_FS("ext4");
5476
5477 static int __init ext4_init_feat_adverts(void)
5478 {
5479         struct ext4_features *ef;
5480         int ret = -ENOMEM;
5481
5482         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5483         if (!ef)
5484                 goto out;
5485
5486         ef->f_kobj.kset = ext4_kset;
5487         init_completion(&ef->f_kobj_unregister);
5488         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5489                                    "features");
5490         if (ret) {
5491                 kfree(ef);
5492                 goto out;
5493         }
5494
5495         ext4_feat = ef;
5496         ret = 0;
5497 out:
5498         return ret;
5499 }
5500
5501 static void ext4_exit_feat_adverts(void)
5502 {
5503         kobject_put(&ext4_feat->f_kobj);
5504         wait_for_completion(&ext4_feat->f_kobj_unregister);
5505         kfree(ext4_feat);
5506 }
5507
5508 /* Shared across all ext4 file systems */
5509 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5510 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5511
5512 static int __init ext4_init_fs(void)
5513 {
5514         int i, err;
5515
5516         ext4_li_info = NULL;
5517         mutex_init(&ext4_li_mtx);
5518
5519         /* Build-time check for flags consistency */
5520         ext4_check_flag_values();
5521
5522         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5523                 mutex_init(&ext4__aio_mutex[i]);
5524                 init_waitqueue_head(&ext4__ioend_wq[i]);
5525         }
5526
5527         err = ext4_init_es();
5528         if (err)
5529                 return err;
5530
5531         err = ext4_init_pageio();
5532         if (err)
5533                 goto out7;
5534
5535         err = ext4_init_system_zone();
5536         if (err)
5537                 goto out6;
5538         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5539         if (!ext4_kset) {
5540                 err = -ENOMEM;
5541                 goto out5;
5542         }
5543         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5544
5545         err = ext4_init_feat_adverts();
5546         if (err)
5547                 goto out4;
5548
5549         err = ext4_init_mballoc();
5550         if (err)
5551                 goto out2;
5552         else
5553                 ext4_mballoc_ready = 1;
5554         err = init_inodecache();
5555         if (err)
5556                 goto out1;
5557         register_as_ext3();
5558         register_as_ext2();
5559         err = register_filesystem(&ext4_fs_type);
5560         if (err)
5561                 goto out;
5562
5563         return 0;
5564 out:
5565         unregister_as_ext2();
5566         unregister_as_ext3();
5567         destroy_inodecache();
5568 out1:
5569         ext4_mballoc_ready = 0;
5570         ext4_exit_mballoc();
5571 out2:
5572         ext4_exit_feat_adverts();
5573 out4:
5574         if (ext4_proc_root)
5575                 remove_proc_entry("fs/ext4", NULL);
5576         kset_unregister(ext4_kset);
5577 out5:
5578         ext4_exit_system_zone();
5579 out6:
5580         ext4_exit_pageio();
5581 out7:
5582         ext4_exit_es();
5583
5584         return err;
5585 }
5586
5587 static void __exit ext4_exit_fs(void)
5588 {
5589         ext4_destroy_lazyinit_thread();
5590         unregister_as_ext2();
5591         unregister_as_ext3();
5592         unregister_filesystem(&ext4_fs_type);
5593         destroy_inodecache();
5594         ext4_exit_mballoc();
5595         ext4_exit_feat_adverts();
5596         remove_proc_entry("fs/ext4", NULL);
5597         kset_unregister(ext4_kset);
5598         ext4_exit_system_zone();
5599         ext4_exit_pageio();
5600         ext4_exit_es();
5601 }
5602
5603 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5604 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5605 MODULE_LICENSE("GPL");
5606 module_init(ext4_init_fs)
5607 module_exit(ext4_exit_fs)