fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_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         brelse(sbi->s_sbh);
 824 #ifdef CONFIG_QUOTA
 825         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 826                 kfree(sbi->s_qf_names[i]);
 827 #endif
 828
 829         /* Debugging code just in case the in-memory inode orphan list
 830          * isn't empty.  The on-disk one can be non-empty if we've
 831          * detected an error and taken the fs readonly, but the
 832          * in-memory list had better be clean by this point. */
 833         if (!list_empty(&sbi->s_orphan))
 834                 dump_orphan_list(sb, sbi);
 835         J_ASSERT(list_empty(&sbi->s_orphan));
 836
 837         invalidate_bdev(sb->s_bdev);
 838         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 839                 /*
 840                  * Invalidate the journal device's buffers.  We don't want them
 841                  * floating about in memory - the physical journal device may
 842                  * hotswapped, and it breaks the `ro-after' testing code.
 843                  */
 844                 sync_blockdev(sbi->journal_bdev);
 845                 invalidate_bdev(sbi->journal_bdev);
 846                 ext4_blkdev_remove(sbi);
 847         }
 848         if (sbi->s_mb_cache) {
 849                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 850                 sbi->s_mb_cache = NULL;
 851         }
 852         if (sbi->s_mmp_tsk)
 853                 kthread_stop(sbi->s_mmp_tsk);
 854         sb->s_fs_info = NULL;
 855         /*
 856          * Now that we are completely done shutting down the
 857          * superblock, we need to actually destroy the kobject.
 858          */
 859         kobject_put(&sbi->s_kobj);
 860         wait_for_completion(&sbi->s_kobj_unregister);
 861         if (sbi->s_chksum_driver)
 862                 crypto_free_shash(sbi->s_chksum_driver);
 863         kfree(sbi->s_blockgroup_lock);
 864         kfree(sbi);
 865 }
 866
 867 static struct kmem_cache *ext4_inode_cachep;
 868
 869 /*
 870  * Called inside transaction, so use GFP_NOFS
 871  */
 872 static struct inode *ext4_alloc_inode(struct super_block *sb)
 873 {
 874         struct ext4_inode_info *ei;
 875
 876         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 877         if (!ei)
 878                 return NULL;
 879
 880         ei->vfs_inode.i_version = 1;
 881         spin_lock_init(&ei->i_raw_lock);
 882         INIT_LIST_HEAD(&ei->i_prealloc_list);
 883         spin_lock_init(&ei->i_prealloc_lock);
 884         ext4_es_init_tree(&ei->i_es_tree);
 885         rwlock_init(&ei->i_es_lock);
 886         INIT_LIST_HEAD(&ei->i_es_lru);
 887         ei->i_es_all_nr = 0;
 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 linux-ext4@vger.kernel.org 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                         ext4_msg(sb, KERN_INFO, "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 < EXT4_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 < EXT4_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 es_ui_show(struct ext4_attr *a,
2552                            struct ext4_sb_info *sbi, char *buf)
2553 {
2554
2555         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2556                            a->u.offset);
2557
2558         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2559 }
2560
2561 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2562                                   struct ext4_sb_info *sbi, char *buf)
2563 {
2564         return snprintf(buf, PAGE_SIZE, "%llu\n",
2565                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2566 }
2567
2568 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2569                                    struct ext4_sb_info *sbi,
2570                                    const char *buf, size_t count)
2571 {
2572         unsigned long long val;
2573         int ret;
2574
2575         if (parse_strtoull(buf, -1ULL, &val))
2576                 return -EINVAL;
2577         ret = ext4_reserve_clusters(sbi, val);
2578
2579         return ret ? ret : count;
2580 }
2581
2582 static ssize_t trigger_test_error(struct ext4_attr *a,
2583                                   struct ext4_sb_info *sbi,
2584                                   const char *buf, size_t count)
2585 {
2586         int len = count;
2587
2588         if (!capable(CAP_SYS_ADMIN))
2589                 return -EPERM;
2590
2591         if (len && buf[len-1] == '\n')
2592                 len--;
2593
2594         if (len)
2595                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2596         return count;
2597 }
2598
2599 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2600                                    struct ext4_sb_info *sbi, char *buf)
2601 {
2602         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2603 }
2604
2605 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2606 static struct ext4_attr ext4_attr_##_name = {                   \
2607         .attr = {.name = __stringify(_name), .mode = _mode },   \
2608         .show   = _show,                                        \
2609         .store  = _store,                                       \
2610         .u = {                                                  \
2611                 .offset = offsetof(struct ext4_sb_info, _elname),\
2612         },                                                      \
2613 }
2614
2615 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2616 static struct ext4_attr ext4_attr_##_name = {                           \
2617         .attr = {.name = __stringify(_name), .mode = _mode },           \
2618         .show   = _show,                                                \
2619         .store  = _store,                                               \
2620         .u = {                                                          \
2621                 .offset = offsetof(struct ext4_super_block, _elname),   \
2622         },                                                              \
2623 }
2624
2625 #define EXT4_ATTR(name, mode, show, store) \
2626 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2627
2628 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2629 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2630 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2631
2632 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2633         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2634 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2635         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2636
2637 #define ATTR_LIST(name) &ext4_attr_##name.attr
2638 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2639 static struct ext4_attr ext4_attr_##_name = {                   \
2640         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2641         .show   = sbi_deprecated_show,                          \
2642         .u = {                                                  \
2643                 .deprecated_val = _val,                         \
2644         },                                                      \
2645 }
2646
2647 EXT4_RO_ATTR(delayed_allocation_blocks);
2648 EXT4_RO_ATTR(session_write_kbytes);
2649 EXT4_RO_ATTR(lifetime_write_kbytes);
2650 EXT4_RW_ATTR(reserved_clusters);
2651 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2652                  inode_readahead_blks_store, s_inode_readahead_blks);
2653 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2654 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2655 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2656 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2657 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2658 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2659 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2660 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2661 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2662 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2663 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2664 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2665 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2666 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2667 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2668 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2669 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2670 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2671 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2672
2673 static struct attribute *ext4_attrs[] = {
2674         ATTR_LIST(delayed_allocation_blocks),
2675         ATTR_LIST(session_write_kbytes),
2676         ATTR_LIST(lifetime_write_kbytes),
2677         ATTR_LIST(reserved_clusters),
2678         ATTR_LIST(inode_readahead_blks),
2679         ATTR_LIST(inode_goal),
2680         ATTR_LIST(mb_stats),
2681         ATTR_LIST(mb_max_to_scan),
2682         ATTR_LIST(mb_min_to_scan),
2683         ATTR_LIST(mb_order2_req),
2684         ATTR_LIST(mb_stream_req),
2685         ATTR_LIST(mb_group_prealloc),
2686         ATTR_LIST(max_writeback_mb_bump),
2687         ATTR_LIST(extent_max_zeroout_kb),
2688         ATTR_LIST(trigger_fs_error),
2689         ATTR_LIST(err_ratelimit_interval_ms),
2690         ATTR_LIST(err_ratelimit_burst),
2691         ATTR_LIST(warning_ratelimit_interval_ms),
2692         ATTR_LIST(warning_ratelimit_burst),
2693         ATTR_LIST(msg_ratelimit_interval_ms),
2694         ATTR_LIST(msg_ratelimit_burst),
2695         ATTR_LIST(errors_count),
2696         ATTR_LIST(first_error_time),
2697         ATTR_LIST(last_error_time),
2698         NULL,
2699 };
2700
2701 /* Features this copy of ext4 supports */
2702 EXT4_INFO_ATTR(lazy_itable_init);
2703 EXT4_INFO_ATTR(batched_discard);
2704 EXT4_INFO_ATTR(meta_bg_resize);
2705
2706 static struct attribute *ext4_feat_attrs[] = {
2707         ATTR_LIST(lazy_itable_init),
2708         ATTR_LIST(batched_discard),
2709         ATTR_LIST(meta_bg_resize),
2710         NULL,
2711 };
2712
2713 static ssize_t ext4_attr_show(struct kobject *kobj,
2714                               struct attribute *attr, char *buf)
2715 {
2716         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2717                                                 s_kobj);
2718         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2719
2720         return a->show ? a->show(a, sbi, buf) : 0;
2721 }
2722
2723 static ssize_t ext4_attr_store(struct kobject *kobj,
2724                                struct attribute *attr,
2725                                const char *buf, size_t len)
2726 {
2727         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2728                                                 s_kobj);
2729         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2730
2731         return a->store ? a->store(a, sbi, buf, len) : 0;
2732 }
2733
2734 static void ext4_sb_release(struct kobject *kobj)
2735 {
2736         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2737                                                 s_kobj);
2738         complete(&sbi->s_kobj_unregister);
2739 }
2740
2741 static const struct sysfs_ops ext4_attr_ops = {
2742         .show   = ext4_attr_show,
2743         .store  = ext4_attr_store,
2744 };
2745
2746 static struct kobj_type ext4_ktype = {
2747         .default_attrs  = ext4_attrs,
2748         .sysfs_ops      = &ext4_attr_ops,
2749         .release        = ext4_sb_release,
2750 };
2751
2752 static void ext4_feat_release(struct kobject *kobj)
2753 {
2754         complete(&ext4_feat->f_kobj_unregister);
2755 }
2756
2757 static ssize_t ext4_feat_show(struct kobject *kobj,
2758                               struct attribute *attr, char *buf)
2759 {
2760         return snprintf(buf, PAGE_SIZE, "supported\n");
2761 }
2762
2763 /*
2764  * We can not use ext4_attr_show/store because it relies on the kobject
2765  * being embedded in the ext4_sb_info structure which is definitely not
2766  * true in this case.
2767  */
2768 static const struct sysfs_ops ext4_feat_ops = {
2769         .show   = ext4_feat_show,
2770         .store  = NULL,
2771 };
2772
2773 static struct kobj_type ext4_feat_ktype = {
2774         .default_attrs  = ext4_feat_attrs,
2775         .sysfs_ops      = &ext4_feat_ops,
2776         .release        = ext4_feat_release,
2777 };
2778
2779 /*
2780  * Check whether this filesystem can be mounted based on
2781  * the features present and the RDONLY/RDWR mount requested.
2782  * Returns 1 if this filesystem can be mounted as requested,
2783  * 0 if it cannot be.
2784  */
2785 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2786 {
2787         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2788                 ext4_msg(sb, KERN_ERR,
2789                         "Couldn't mount because of "
2790                         "unsupported optional features (%x)",
2791                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2792                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2793                 return 0;
2794         }
2795
2796         if (readonly)
2797                 return 1;
2798
2799         /* Check that feature set is OK for a read-write mount */
2800         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2801                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2802                          "unsupported optional features (%x)",
2803                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2804                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2805                 return 0;
2806         }
2807         /*
2808          * Large file size enabled file system can only be mounted
2809          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2810          */
2811         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2812                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2813                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2814                                  "cannot be mounted RDWR without "
2815                                  "CONFIG_LBDAF");
2816                         return 0;
2817                 }
2818         }
2819         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2820             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2821                 ext4_msg(sb, KERN_ERR,
2822                          "Can't support bigalloc feature without "
2823                          "extents feature\n");
2824                 return 0;
2825         }
2826
2827 #ifndef CONFIG_QUOTA
2828         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2829             !readonly) {
2830                 ext4_msg(sb, KERN_ERR,
2831                          "Filesystem with quota feature cannot be mounted RDWR "
2832                          "without CONFIG_QUOTA");
2833                 return 0;
2834         }
2835 #endif  /* CONFIG_QUOTA */
2836         return 1;
2837 }
2838
2839 /*
2840  * This function is called once a day if we have errors logged
2841  * on the file system
2842  */
2843 static void print_daily_error_info(unsigned long arg)
2844 {
2845         struct super_block *sb = (struct super_block *) arg;
2846         struct ext4_sb_info *sbi;
2847         struct ext4_super_block *es;
2848
2849         sbi = EXT4_SB(sb);
2850         es = sbi->s_es;
2851
2852         if (es->s_error_count)
2853                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2854                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2855                          le32_to_cpu(es->s_error_count));
2856         if (es->s_first_error_time) {
2857                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2858                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2859                        (int) sizeof(es->s_first_error_func),
2860                        es->s_first_error_func,
2861                        le32_to_cpu(es->s_first_error_line));
2862                 if (es->s_first_error_ino)
2863                         printk(": inode %u",
2864                                le32_to_cpu(es->s_first_error_ino));
2865                 if (es->s_first_error_block)
2866                         printk(": block %llu", (unsigned long long)
2867                                le64_to_cpu(es->s_first_error_block));
2868                 printk("\n");
2869         }
2870         if (es->s_last_error_time) {
2871                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2872                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2873                        (int) sizeof(es->s_last_error_func),
2874                        es->s_last_error_func,
2875                        le32_to_cpu(es->s_last_error_line));
2876                 if (es->s_last_error_ino)
2877                         printk(": inode %u",
2878                                le32_to_cpu(es->s_last_error_ino));
2879                 if (es->s_last_error_block)
2880                         printk(": block %llu", (unsigned long long)
2881                                le64_to_cpu(es->s_last_error_block));
2882                 printk("\n");
2883         }
2884         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2885 }
2886
2887 /* Find next suitable group and run ext4_init_inode_table */
2888 static int ext4_run_li_request(struct ext4_li_request *elr)
2889 {
2890         struct ext4_group_desc *gdp = NULL;
2891         ext4_group_t group, ngroups;
2892         struct super_block *sb;
2893         unsigned long timeout = 0;
2894         int ret = 0;
2895
2896         sb = elr->lr_super;
2897         ngroups = EXT4_SB(sb)->s_groups_count;
2898
2899         sb_start_write(sb);
2900         for (group = elr->lr_next_group; group < ngroups; group++) {
2901                 gdp = ext4_get_group_desc(sb, group, NULL);
2902                 if (!gdp) {
2903                         ret = 1;
2904                         break;
2905                 }
2906
2907                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2908                         break;
2909         }
2910
2911         if (group >= ngroups)
2912                 ret = 1;
2913
2914         if (!ret) {
2915                 timeout = jiffies;
2916                 ret = ext4_init_inode_table(sb, group,
2917                                             elr->lr_timeout ? 0 : 1);
2918                 if (elr->lr_timeout == 0) {
2919                         timeout = (jiffies - timeout) *
2920                                   elr->lr_sbi->s_li_wait_mult;
2921                         elr->lr_timeout = timeout;
2922                 }
2923                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2924                 elr->lr_next_group = group + 1;
2925         }
2926         sb_end_write(sb);
2927
2928         return ret;
2929 }
2930
2931 /*
2932  * Remove lr_request from the list_request and free the
2933  * request structure. Should be called with li_list_mtx held
2934  */
2935 static void ext4_remove_li_request(struct ext4_li_request *elr)
2936 {
2937         struct ext4_sb_info *sbi;
2938
2939         if (!elr)
2940                 return;
2941
2942         sbi = elr->lr_sbi;
2943
2944         list_del(&elr->lr_request);
2945         sbi->s_li_request = NULL;
2946         kfree(elr);
2947 }
2948
2949 static void ext4_unregister_li_request(struct super_block *sb)
2950 {
2951         mutex_lock(&ext4_li_mtx);
2952         if (!ext4_li_info) {
2953                 mutex_unlock(&ext4_li_mtx);
2954                 return;
2955         }
2956
2957         mutex_lock(&ext4_li_info->li_list_mtx);
2958         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2959         mutex_unlock(&ext4_li_info->li_list_mtx);
2960         mutex_unlock(&ext4_li_mtx);
2961 }
2962
2963 static struct task_struct *ext4_lazyinit_task;
2964
2965 /*
2966  * This is the function where ext4lazyinit thread lives. It walks
2967  * through the request list searching for next scheduled filesystem.
2968  * When such a fs is found, run the lazy initialization request
2969  * (ext4_rn_li_request) and keep track of the time spend in this
2970  * function. Based on that time we compute next schedule time of
2971  * the request. When walking through the list is complete, compute
2972  * next waking time and put itself into sleep.
2973  */
2974 static int ext4_lazyinit_thread(void *arg)
2975 {
2976         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2977         struct list_head *pos, *n;
2978         struct ext4_li_request *elr;
2979         unsigned long next_wakeup, cur;
2980
2981         BUG_ON(NULL == eli);
2982
2983 cont_thread:
2984         while (true) {
2985                 next_wakeup = MAX_JIFFY_OFFSET;
2986
2987                 mutex_lock(&eli->li_list_mtx);
2988                 if (list_empty(&eli->li_request_list)) {
2989                         mutex_unlock(&eli->li_list_mtx);
2990                         goto exit_thread;
2991                 }
2992
2993                 list_for_each_safe(pos, n, &eli->li_request_list) {
2994                         elr = list_entry(pos, struct ext4_li_request,
2995                                          lr_request);
2996
2997                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2998                                 if (ext4_run_li_request(elr) != 0) {
2999                                         /* error, remove the lazy_init job */
3000                                         ext4_remove_li_request(elr);
3001                                         continue;
3002                                 }
3003                         }
3004
3005                         if (time_before(elr->lr_next_sched, next_wakeup))
3006                                 next_wakeup = elr->lr_next_sched;
3007                 }
3008                 mutex_unlock(&eli->li_list_mtx);
3009
3010                 try_to_freeze();
3011
3012                 cur = jiffies;
3013                 if ((time_after_eq(cur, next_wakeup)) ||
3014                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3015                         cond_resched();
3016                         continue;
3017                 }
3018
3019                 schedule_timeout_interruptible(next_wakeup - cur);
3020
3021                 if (kthread_should_stop()) {
3022                         ext4_clear_request_list();
3023                         goto exit_thread;
3024                 }
3025         }
3026
3027 exit_thread:
3028         /*
3029          * It looks like the request list is empty, but we need
3030          * to check it under the li_list_mtx lock, to prevent any
3031          * additions into it, and of course we should lock ext4_li_mtx
3032          * to atomically free the list and ext4_li_info, because at
3033          * this point another ext4 filesystem could be registering
3034          * new one.
3035          */
3036         mutex_lock(&ext4_li_mtx);
3037         mutex_lock(&eli->li_list_mtx);
3038         if (!list_empty(&eli->li_request_list)) {
3039                 mutex_unlock(&eli->li_list_mtx);
3040                 mutex_unlock(&ext4_li_mtx);
3041                 goto cont_thread;
3042         }
3043         mutex_unlock(&eli->li_list_mtx);
3044         kfree(ext4_li_info);
3045         ext4_li_info = NULL;
3046         mutex_unlock(&ext4_li_mtx);
3047
3048         return 0;
3049 }
3050
3051 static void ext4_clear_request_list(void)
3052 {
3053         struct list_head *pos, *n;
3054         struct ext4_li_request *elr;
3055
3056         mutex_lock(&ext4_li_info->li_list_mtx);
3057         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3058                 elr = list_entry(pos, struct ext4_li_request,
3059                                  lr_request);
3060                 ext4_remove_li_request(elr);
3061         }
3062         mutex_unlock(&ext4_li_info->li_list_mtx);
3063 }
3064
3065 static int ext4_run_lazyinit_thread(void)
3066 {
3067         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3068                                          ext4_li_info, "ext4lazyinit");
3069         if (IS_ERR(ext4_lazyinit_task)) {
3070                 int err = PTR_ERR(ext4_lazyinit_task);
3071                 ext4_clear_request_list();
3072                 kfree(ext4_li_info);
3073                 ext4_li_info = NULL;
3074                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3075                                  "initialization thread\n",
3076                                  err);
3077                 return err;
3078         }
3079         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3080         return 0;
3081 }
3082
3083 /*
3084  * Check whether it make sense to run itable init. thread or not.
3085  * If there is at least one uninitialized inode table, return
3086  * corresponding group number, else the loop goes through all
3087  * groups and return total number of groups.
3088  */
3089 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3090 {
3091         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3092         struct ext4_group_desc *gdp = NULL;
3093
3094         for (group = 0; group < ngroups; group++) {
3095                 gdp = ext4_get_group_desc(sb, group, NULL);
3096                 if (!gdp)
3097                         continue;
3098
3099                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3100                         break;
3101         }
3102
3103         return group;
3104 }
3105
3106 static int ext4_li_info_new(void)
3107 {
3108         struct ext4_lazy_init *eli = NULL;
3109
3110         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3111         if (!eli)
3112                 return -ENOMEM;
3113
3114         INIT_LIST_HEAD(&eli->li_request_list);
3115         mutex_init(&eli->li_list_mtx);
3116
3117         eli->li_state |= EXT4_LAZYINIT_QUIT;
3118
3119         ext4_li_info = eli;
3120
3121         return 0;
3122 }
3123
3124 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3125                                             ext4_group_t start)
3126 {
3127         struct ext4_sb_info *sbi = EXT4_SB(sb);
3128         struct ext4_li_request *elr;
3129
3130         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3131         if (!elr)
3132                 return NULL;
3133
3134         elr->lr_super = sb;
3135         elr->lr_sbi = sbi;
3136         elr->lr_next_group = start;
3137
3138         /*
3139          * Randomize first schedule time of the request to
3140          * spread the inode table initialization requests
3141          * better.
3142          */
3143         elr->lr_next_sched = jiffies + (prandom_u32() %
3144                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3145         return elr;
3146 }
3147
3148 int ext4_register_li_request(struct super_block *sb,
3149                              ext4_group_t first_not_zeroed)
3150 {
3151         struct ext4_sb_info *sbi = EXT4_SB(sb);
3152         struct ext4_li_request *elr = NULL;
3153         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3154         int ret = 0;
3155
3156         mutex_lock(&ext4_li_mtx);
3157         if (sbi->s_li_request != NULL) {
3158                 /*
3159                  * Reset timeout so it can be computed again, because
3160                  * s_li_wait_mult might have changed.
3161                  */
3162                 sbi->s_li_request->lr_timeout = 0;
3163                 goto out;
3164         }
3165
3166         if (first_not_zeroed == ngroups ||
3167             (sb->s_flags & MS_RDONLY) ||
3168             !test_opt(sb, INIT_INODE_TABLE))
3169                 goto out;
3170
3171         elr = ext4_li_request_new(sb, first_not_zeroed);
3172         if (!elr) {
3173                 ret = -ENOMEM;
3174                 goto out;
3175         }
3176
3177         if (NULL == ext4_li_info) {
3178                 ret = ext4_li_info_new();
3179                 if (ret)
3180                         goto out;
3181         }
3182
3183         mutex_lock(&ext4_li_info->li_list_mtx);
3184         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3185         mutex_unlock(&ext4_li_info->li_list_mtx);
3186
3187         sbi->s_li_request = elr;
3188         /*
3189          * set elr to NULL here since it has been inserted to
3190          * the request_list and the removal and free of it is
3191          * handled by ext4_clear_request_list from now on.
3192          */
3193         elr = NULL;
3194
3195         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3196                 ret = ext4_run_lazyinit_thread();
3197                 if (ret)
3198                         goto out;
3199         }
3200 out:
3201         mutex_unlock(&ext4_li_mtx);
3202         if (ret)
3203                 kfree(elr);
3204         return ret;
3205 }
3206
3207 /*
3208  * We do not need to lock anything since this is called on
3209  * module unload.
3210  */
3211 static void ext4_destroy_lazyinit_thread(void)
3212 {
3213         /*
3214          * If thread exited earlier
3215          * there's nothing to be done.
3216          */
3217         if (!ext4_li_info || !ext4_lazyinit_task)
3218                 return;
3219
3220         kthread_stop(ext4_lazyinit_task);
3221 }
3222
3223 static int set_journal_csum_feature_set(struct super_block *sb)
3224 {
3225         int ret = 1;
3226         int compat, incompat;
3227         struct ext4_sb_info *sbi = EXT4_SB(sb);
3228
3229         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3230                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3231                 /* journal checksum v3 */
3232                 compat = 0;
3233                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3234         } else {
3235                 /* journal checksum v1 */
3236                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3237                 incompat = 0;
3238         }
3239
3240         jbd2_journal_clear_features(sbi->s_journal,
3241                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3242                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3243                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3244         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3245                 ret = jbd2_journal_set_features(sbi->s_journal,
3246                                 compat, 0,
3247                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3248                                 incompat);
3249         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3250                 ret = jbd2_journal_set_features(sbi->s_journal,
3251                                 compat, 0,
3252                                 incompat);
3253                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3254                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3255         } else {
3256                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3257                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3258         }
3259
3260         return ret;
3261 }
3262
3263 /*
3264  * Note: calculating the overhead so we can be compatible with
3265  * historical BSD practice is quite difficult in the face of
3266  * clusters/bigalloc.  This is because multiple metadata blocks from
3267  * different block group can end up in the same allocation cluster.
3268  * Calculating the exact overhead in the face of clustered allocation
3269  * requires either O(all block bitmaps) in memory or O(number of block
3270  * groups**2) in time.  We will still calculate the superblock for
3271  * older file systems --- and if we come across with a bigalloc file
3272  * system with zero in s_overhead_clusters the estimate will be close to
3273  * correct especially for very large cluster sizes --- but for newer
3274  * file systems, it's better to calculate this figure once at mkfs
3275  * time, and store it in the superblock.  If the superblock value is
3276  * present (even for non-bigalloc file systems), we will use it.
3277  */
3278 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3279                           char *buf)
3280 {
3281         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3282         struct ext4_group_desc  *gdp;
3283         ext4_fsblk_t            first_block, last_block, b;
3284         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3285         int                     s, j, count = 0;
3286
3287         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3288                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3289                         sbi->s_itb_per_group + 2);
3290
3291         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3292                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3293         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3294         for (i = 0; i < ngroups; i++) {
3295                 gdp = ext4_get_group_desc(sb, i, NULL);
3296                 b = ext4_block_bitmap(sb, gdp);
3297                 if (b >= first_block && b <= last_block) {
3298                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3299                         count++;
3300                 }
3301                 b = ext4_inode_bitmap(sb, gdp);
3302                 if (b >= first_block && b <= last_block) {
3303                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3304                         count++;
3305                 }
3306                 b = ext4_inode_table(sb, gdp);
3307                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3308                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3309                                 int c = EXT4_B2C(sbi, b - first_block);
3310                                 ext4_set_bit(c, buf);
3311                                 count++;
3312                         }
3313                 if (i != grp)
3314                         continue;
3315                 s = 0;
3316                 if (ext4_bg_has_super(sb, grp)) {
3317                         ext4_set_bit(s++, buf);
3318                         count++;
3319                 }
3320                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3321                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3322                         count++;
3323                 }
3324         }
3325         if (!count)
3326                 return 0;
3327         return EXT4_CLUSTERS_PER_GROUP(sb) -
3328                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3329 }
3330
3331 /*
3332  * Compute the overhead and stash it in sbi->s_overhead
3333  */
3334 int ext4_calculate_overhead(struct super_block *sb)
3335 {
3336         struct ext4_sb_info *sbi = EXT4_SB(sb);
3337         struct ext4_super_block *es = sbi->s_es;
3338         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3339         ext4_fsblk_t overhead = 0;
3340         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3341
3342         if (!buf)
3343                 return -ENOMEM;
3344
3345         /*
3346          * Compute the overhead (FS structures).  This is constant
3347          * for a given filesystem unless the number of block groups
3348          * changes so we cache the previous value until it does.
3349          */
3350
3351         /*
3352          * All of the blocks before first_data_block are overhead
3353          */
3354         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3355
3356         /*
3357          * Add the overhead found in each block group
3358          */
3359         for (i = 0; i < ngroups; i++) {
3360                 int blks;
3361
3362                 blks = count_overhead(sb, i, buf);
3363                 overhead += blks;
3364                 if (blks)
3365                         memset(buf, 0, PAGE_SIZE);
3366                 cond_resched();
3367         }
3368         /* Add the journal blocks as well */
3369         if (sbi->s_journal)
3370                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3371
3372         sbi->s_overhead = overhead;
3373         smp_wmb();
3374         free_page((unsigned long) buf);
3375         return 0;
3376 }
3377
3378
3379 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3380 {
3381         ext4_fsblk_t resv_clusters;
3382
3383         /*
3384          * There's no need to reserve anything when we aren't using extents.
3385          * The space estimates are exact, there are no unwritten extents,
3386          * hole punching doesn't need new metadata... This is needed especially
3387          * to keep ext2/3 backward compatibility.
3388          */
3389         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3390                 return 0;
3391         /*
3392          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3393          * This should cover the situations where we can not afford to run
3394          * out of space like for example punch hole, or converting
3395          * unwritten extents in delalloc path. In most cases such
3396          * allocation would require 1, or 2 blocks, higher numbers are
3397          * very rare.
3398          */
3399         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3400                         EXT4_SB(sb)->s_cluster_bits;
3401
3402         do_div(resv_clusters, 50);
3403         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3404
3405         return resv_clusters;
3406 }
3407
3408
3409 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3410 {
3411         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3412                                 sbi->s_cluster_bits;
3413
3414         if (count >= clusters)
3415                 return -EINVAL;
3416
3417         atomic64_set(&sbi->s_resv_clusters, count);
3418         return 0;
3419 }
3420
3421 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3422 {
3423         char *orig_data = kstrdup(data, GFP_KERNEL);
3424         struct buffer_head *bh;
3425         struct ext4_super_block *es = NULL;
3426         struct ext4_sb_info *sbi;
3427         ext4_fsblk_t block;
3428         ext4_fsblk_t sb_block = get_sb_block(&data);
3429         ext4_fsblk_t logical_sb_block;
3430         unsigned long offset = 0;
3431         unsigned long journal_devnum = 0;
3432         unsigned long def_mount_opts;
3433         struct inode *root;
3434         char *cp;
3435         const char *descr;
3436         int ret = -ENOMEM;
3437         int blocksize, clustersize;
3438         unsigned int db_count;
3439         unsigned int i;
3440         int needs_recovery, has_huge_files, has_bigalloc;
3441         __u64 blocks_count;
3442         int err = 0;
3443         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3444         ext4_group_t first_not_zeroed;
3445
3446         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3447         if (!sbi)
3448                 goto out_free_orig;
3449
3450         sbi->s_blockgroup_lock =
3451                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3452         if (!sbi->s_blockgroup_lock) {
3453                 kfree(sbi);
3454                 goto out_free_orig;
3455         }
3456         sb->s_fs_info = sbi;
3457         sbi->s_sb = sb;
3458         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3459         sbi->s_sb_block = sb_block;
3460         if (sb->s_bdev->bd_part)
3461                 sbi->s_sectors_written_start =
3462                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3463
3464         /* Cleanup superblock name */
3465         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3466                 *cp = '!';
3467
3468         /* -EINVAL is default */
3469         ret = -EINVAL;
3470         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3471         if (!blocksize) {
3472                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3473                 goto out_fail;
3474         }
3475
3476         /*
3477          * The ext4 superblock will not be buffer aligned for other than 1kB
3478          * block sizes.  We need to calculate the offset from buffer start.
3479          */
3480         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3481                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3482                 offset = do_div(logical_sb_block, blocksize);
3483         } else {
3484                 logical_sb_block = sb_block;
3485         }
3486
3487         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3488                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3489                 goto out_fail;
3490         }
3491         /*
3492          * Note: s_es must be initialized as soon as possible because
3493          *       some ext4 macro-instructions depend on its value
3494          */
3495         es = (struct ext4_super_block *) (bh->b_data + offset);
3496         sbi->s_es = es;
3497         sb->s_magic = le16_to_cpu(es->s_magic);
3498         if (sb->s_magic != EXT4_SUPER_MAGIC)
3499                 goto cantfind_ext4;
3500         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3501
3502         /* Warn if metadata_csum and gdt_csum are both set. */
3503         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3504                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3505             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3506                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3507                              "redundant flags; please run fsck.");
3508
3509         /* Check for a known checksum algorithm */
3510         if (!ext4_verify_csum_type(sb, es)) {
3511                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3512                          "unknown checksum algorithm.");
3513                 silent = 1;
3514                 goto cantfind_ext4;
3515         }
3516
3517         /* Load the checksum driver */
3518         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3519                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3520                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3521                 if (IS_ERR(sbi->s_chksum_driver)) {
3522                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3523                         ret = PTR_ERR(sbi->s_chksum_driver);
3524                         sbi->s_chksum_driver = NULL;
3525                         goto failed_mount;
3526                 }
3527         }
3528
3529         /* Check superblock checksum */
3530         if (!ext4_superblock_csum_verify(sb, es)) {
3531                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3532                          "invalid superblock checksum.  Run e2fsck?");
3533                 silent = 1;
3534                 goto cantfind_ext4;
3535         }
3536
3537         /* Precompute checksum seed for all metadata */
3538         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3539                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3540                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3541                                                sizeof(es->s_uuid));
3542
3543         /* Set defaults before we parse the mount options */
3544         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3545         set_opt(sb, INIT_INODE_TABLE);
3546         if (def_mount_opts & EXT4_DEFM_DEBUG)
3547                 set_opt(sb, DEBUG);
3548         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3549                 set_opt(sb, GRPID);
3550         if (def_mount_opts & EXT4_DEFM_UID16)
3551                 set_opt(sb, NO_UID32);
3552         /* xattr user namespace & acls are now defaulted on */
3553         set_opt(sb, XATTR_USER);
3554 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3555         set_opt(sb, POSIX_ACL);
3556 #endif
3557         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3558                 set_opt(sb, JOURNAL_DATA);
3559         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3560                 set_opt(sb, ORDERED_DATA);
3561         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3562                 set_opt(sb, WRITEBACK_DATA);
3563
3564         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3565                 set_opt(sb, ERRORS_PANIC);
3566         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3567                 set_opt(sb, ERRORS_CONT);
3568         else
3569                 set_opt(sb, ERRORS_RO);
3570         /* block_validity enabled by default; disable with noblock_validity */
3571         set_opt(sb, BLOCK_VALIDITY);
3572         if (def_mount_opts & EXT4_DEFM_DISCARD)
3573                 set_opt(sb, DISCARD);
3574
3575         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3576         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3577         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3578         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3579         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3580
3581         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3582                 set_opt(sb, BARRIER);
3583
3584         /*
3585          * enable delayed allocation by default
3586          * Use -o nodelalloc to turn it off
3587          */
3588         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3589             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3590                 set_opt(sb, DELALLOC);
3591
3592         /*
3593          * set default s_li_wait_mult for lazyinit, for the case there is
3594          * no mount option specified.
3595          */
3596         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3597
3598         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3599                            &journal_devnum, &journal_ioprio, 0)) {
3600                 ext4_msg(sb, KERN_WARNING,
3601                          "failed to parse options in superblock: %s",
3602                          sbi->s_es->s_mount_opts);
3603         }
3604         sbi->s_def_mount_opt = sbi->s_mount_opt;
3605         if (!parse_options((char *) data, sb, &journal_devnum,
3606                            &journal_ioprio, 0))
3607                 goto failed_mount;
3608
3609         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3610                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3611                             "with data=journal disables delayed "
3612                             "allocation and O_DIRECT support!\n");
3613                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3614                         ext4_msg(sb, KERN_ERR, "can't mount with "
3615                                  "both data=journal and delalloc");
3616                         goto failed_mount;
3617                 }
3618                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3619                         ext4_msg(sb, KERN_ERR, "can't mount with "
3620                                  "both data=journal and dioread_nolock");
3621                         goto failed_mount;
3622                 }
3623                 if (test_opt(sb, DELALLOC))
3624                         clear_opt(sb, DELALLOC);
3625         }
3626
3627         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3628                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3629
3630         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3631             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3632              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3633              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3634                 ext4_msg(sb, KERN_WARNING,
3635                        "feature flags set on rev 0 fs, "
3636                        "running e2fsck is recommended");
3637
3638         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3639                 set_opt2(sb, HURD_COMPAT);
3640                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3641                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3642                         ext4_msg(sb, KERN_ERR,
3643                                  "The Hurd can't support 64-bit file systems");
3644                         goto failed_mount;
3645                 }
3646         }
3647
3648         if (IS_EXT2_SB(sb)) {
3649                 if (ext2_feature_set_ok(sb))
3650                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3651                                  "using the ext4 subsystem");
3652                 else {
3653                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3654                                  "to feature incompatibilities");
3655                         goto failed_mount;
3656                 }
3657         }
3658
3659         if (IS_EXT3_SB(sb)) {
3660                 if (ext3_feature_set_ok(sb))
3661                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3662                                  "using the ext4 subsystem");
3663                 else {
3664                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3665                                  "to feature incompatibilities");
3666                         goto failed_mount;
3667                 }
3668         }
3669
3670         /*
3671          * Check feature flags regardless of the revision level, since we
3672          * previously didn't change the revision level when setting the flags,
3673          * so there is a chance incompat flags are set on a rev 0 filesystem.
3674          */
3675         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3676                 goto failed_mount;
3677
3678         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3679         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3680             blocksize > EXT4_MAX_BLOCK_SIZE) {
3681                 ext4_msg(sb, KERN_ERR,
3682                        "Unsupported filesystem blocksize %d", blocksize);
3683                 goto failed_mount;
3684         }
3685
3686         if (sb->s_blocksize != blocksize) {
3687                 /* Validate the filesystem blocksize */
3688                 if (!sb_set_blocksize(sb, blocksize)) {
3689                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3690                                         blocksize);
3691                         goto failed_mount;
3692                 }
3693
3694                 brelse(bh);
3695                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3696                 offset = do_div(logical_sb_block, blocksize);
3697                 bh = sb_bread_unmovable(sb, logical_sb_block);
3698                 if (!bh) {
3699                         ext4_msg(sb, KERN_ERR,
3700                                "Can't read superblock on 2nd try");
3701                         goto failed_mount;
3702                 }
3703                 es = (struct ext4_super_block *)(bh->b_data + offset);
3704                 sbi->s_es = es;
3705                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3706                         ext4_msg(sb, KERN_ERR,
3707                                "Magic mismatch, very weird!");
3708                         goto failed_mount;
3709                 }
3710         }
3711
3712         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3713                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3714         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3715                                                       has_huge_files);
3716         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3717
3718         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3719                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3720                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3721         } else {
3722                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3723                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3724                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3725                     (!is_power_of_2(sbi->s_inode_size)) ||
3726                     (sbi->s_inode_size > blocksize)) {
3727                         ext4_msg(sb, KERN_ERR,
3728                                "unsupported inode size: %d",
3729                                sbi->s_inode_size);
3730                         goto failed_mount;
3731                 }
3732                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3733                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3734         }
3735
3736         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3737         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3738                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3739                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3740                     !is_power_of_2(sbi->s_desc_size)) {
3741                         ext4_msg(sb, KERN_ERR,
3742                                "unsupported descriptor size %lu",
3743                                sbi->s_desc_size);
3744                         goto failed_mount;
3745                 }
3746         } else
3747                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3748
3749         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3750         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3751         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3752                 goto cantfind_ext4;
3753
3754         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3755         if (sbi->s_inodes_per_block == 0)
3756                 goto cantfind_ext4;
3757         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3758                                         sbi->s_inodes_per_block;
3759         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3760         sbi->s_sbh = bh;
3761         sbi->s_mount_state = le16_to_cpu(es->s_state);
3762         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3763         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3764
3765         for (i = 0; i < 4; i++)
3766                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3767         sbi->s_def_hash_version = es->s_def_hash_version;
3768         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3769                 i = le32_to_cpu(es->s_flags);
3770                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3771                         sbi->s_hash_unsigned = 3;
3772                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3773 #ifdef __CHAR_UNSIGNED__
3774                         if (!(sb->s_flags & MS_RDONLY))
3775                                 es->s_flags |=
3776                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3777                         sbi->s_hash_unsigned = 3;
3778 #else
3779                         if (!(sb->s_flags & MS_RDONLY))
3780                                 es->s_flags |=
3781                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3782 #endif
3783                 }
3784         }
3785
3786         /* Handle clustersize */
3787         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3788         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3789                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3790         if (has_bigalloc) {
3791                 if (clustersize < blocksize) {
3792                         ext4_msg(sb, KERN_ERR,
3793                                  "cluster size (%d) smaller than "
3794                                  "block size (%d)", clustersize, blocksize);
3795                         goto failed_mount;
3796                 }
3797                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3798                         le32_to_cpu(es->s_log_block_size);
3799                 sbi->s_clusters_per_group =
3800                         le32_to_cpu(es->s_clusters_per_group);
3801                 if (sbi->s_clusters_per_group > blocksize * 8) {
3802                         ext4_msg(sb, KERN_ERR,
3803                                  "#clusters per group too big: %lu",
3804                                  sbi->s_clusters_per_group);
3805                         goto failed_mount;
3806                 }
3807                 if (sbi->s_blocks_per_group !=
3808                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3809                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3810                                  "clusters per group (%lu) inconsistent",
3811                                  sbi->s_blocks_per_group,
3812                                  sbi->s_clusters_per_group);
3813                         goto failed_mount;
3814                 }
3815         } else {
3816                 if (clustersize != blocksize) {
3817                         ext4_warning(sb, "fragment/cluster size (%d) != "
3818                                      "block size (%d)", clustersize,
3819                                      blocksize);
3820                         clustersize = blocksize;
3821                 }
3822                 if (sbi->s_blocks_per_group > blocksize * 8) {
3823                         ext4_msg(sb, KERN_ERR,
3824                                  "#blocks per group too big: %lu",
3825                                  sbi->s_blocks_per_group);
3826                         goto failed_mount;
3827                 }
3828                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3829                 sbi->s_cluster_bits = 0;
3830         }
3831         sbi->s_cluster_ratio = clustersize / blocksize;
3832
3833         if (sbi->s_inodes_per_group > blocksize * 8) {
3834                 ext4_msg(sb, KERN_ERR,
3835                        "#inodes per group too big: %lu",
3836                        sbi->s_inodes_per_group);
3837                 goto failed_mount;
3838         }
3839
3840         /* Do we have standard group size of clustersize * 8 blocks ? */
3841         if (sbi->s_blocks_per_group == clustersize << 3)
3842                 set_opt2(sb, STD_GROUP_SIZE);
3843
3844         /*
3845          * Test whether we have more sectors than will fit in sector_t,
3846          * and whether the max offset is addressable by the page cache.
3847          */
3848         err = generic_check_addressable(sb->s_blocksize_bits,
3849                                         ext4_blocks_count(es));
3850         if (err) {
3851                 ext4_msg(sb, KERN_ERR, "filesystem"
3852                          " too large to mount safely on this system");
3853                 if (sizeof(sector_t) < 8)
3854                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3855                 goto failed_mount;
3856         }
3857
3858         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3859                 goto cantfind_ext4;
3860
3861         /* check blocks count against device size */
3862         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3863         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3864                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3865                        "exceeds size of device (%llu blocks)",
3866                        ext4_blocks_count(es), blocks_count);
3867                 goto failed_mount;
3868         }
3869
3870         /*
3871          * It makes no sense for the first data block to be beyond the end
3872          * of the filesystem.
3873          */
3874         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3875                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3876                          "block %u is beyond end of filesystem (%llu)",
3877                          le32_to_cpu(es->s_first_data_block),
3878                          ext4_blocks_count(es));
3879                 goto failed_mount;
3880         }
3881         blocks_count = (ext4_blocks_count(es) -
3882                         le32_to_cpu(es->s_first_data_block) +
3883                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3884         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3885         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3886                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3887                        "(block count %llu, first data block %u, "
3888                        "blocks per group %lu)", sbi->s_groups_count,
3889                        ext4_blocks_count(es),
3890                        le32_to_cpu(es->s_first_data_block),
3891                        EXT4_BLOCKS_PER_GROUP(sb));
3892                 goto failed_mount;
3893         }
3894         sbi->s_groups_count = blocks_count;
3895         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3896                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3897         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3898                    EXT4_DESC_PER_BLOCK(sb);
3899         sbi->s_group_desc = ext4_kvmalloc(db_count *
3900                                           sizeof(struct buffer_head *),
3901                                           GFP_KERNEL);
3902         if (sbi->s_group_desc == NULL) {
3903                 ext4_msg(sb, KERN_ERR, "not enough memory");
3904                 ret = -ENOMEM;
3905                 goto failed_mount;
3906         }
3907
3908         if (ext4_proc_root)
3909                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3910
3911         if (sbi->s_proc)
3912                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3913                                  &ext4_seq_options_fops, sb);
3914
3915         bgl_lock_init(sbi->s_blockgroup_lock);
3916
3917         for (i = 0; i < db_count; i++) {
3918                 block = descriptor_loc(sb, logical_sb_block, i);
3919                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3920                 if (!sbi->s_group_desc[i]) {
3921                         ext4_msg(sb, KERN_ERR,
3922                                "can't read group descriptor %d", i);
3923                         db_count = i;
3924                         goto failed_mount2;
3925                 }
3926         }
3927         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3928                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3929                 goto failed_mount2;
3930         }
3931
3932         sbi->s_gdb_count = db_count;
3933         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3934         spin_lock_init(&sbi->s_next_gen_lock);
3935
3936         init_timer(&sbi->s_err_report);
3937         sbi->s_err_report.function = print_daily_error_info;
3938         sbi->s_err_report.data = (unsigned long) sb;
3939
3940         /* Register extent status tree shrinker */
3941         if (ext4_es_register_shrinker(sbi))
3942                 goto failed_mount3;
3943
3944         sbi->s_stripe = ext4_get_stripe_size(sbi);
3945         sbi->s_extent_max_zeroout_kb = 32;
3946
3947         /*
3948          * set up enough so that it can read an inode
3949          */
3950         if (!test_opt(sb, NOLOAD) &&
3951             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3952                 sb->s_op = &ext4_sops;
3953         else
3954                 sb->s_op = &ext4_nojournal_sops;
3955         sb->s_export_op = &ext4_export_ops;
3956         sb->s_xattr = ext4_xattr_handlers;
3957 #ifdef CONFIG_QUOTA
3958         sb->dq_op = &ext4_quota_operations;
3959         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3960                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3961         else
3962                 sb->s_qcop = &ext4_qctl_operations;
3963 #endif
3964         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3965
3966         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3967         mutex_init(&sbi->s_orphan_lock);
3968
3969         sb->s_root = NULL;
3970
3971         needs_recovery = (es->s_last_orphan != 0 ||
3972                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3973                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3974
3975         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3976             !(sb->s_flags & MS_RDONLY))
3977                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3978                         goto failed_mount3;
3979
3980         /*
3981          * The first inode we look at is the journal inode.  Don't try
3982          * root first: it may be modified in the journal!
3983          */
3984         if (!test_opt(sb, NOLOAD) &&
3985             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3986                 if (ext4_load_journal(sb, es, journal_devnum))
3987                         goto failed_mount3;
3988         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3989               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3990                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3991                        "suppressed and not mounted read-only");
3992                 goto failed_mount_wq;
3993         } else {
3994                 clear_opt(sb, DATA_FLAGS);
3995                 sbi->s_journal = NULL;
3996                 needs_recovery = 0;
3997                 goto no_journal;
3998         }
3999
4000         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
4001             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4002                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4003                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4004                 goto failed_mount_wq;
4005         }
4006
4007         if (!set_journal_csum_feature_set(sb)) {
4008                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4009                          "feature set");
4010                 goto failed_mount_wq;
4011         }
4012
4013         /* We have now updated the journal if required, so we can
4014          * validate the data journaling mode. */
4015         switch (test_opt(sb, DATA_FLAGS)) {
4016         case 0:
4017                 /* No mode set, assume a default based on the journal
4018                  * capabilities: ORDERED_DATA if the journal can
4019                  * cope, else JOURNAL_DATA
4020                  */
4021                 if (jbd2_journal_check_available_features
4022                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4023                         set_opt(sb, ORDERED_DATA);
4024                 else
4025                         set_opt(sb, JOURNAL_DATA);
4026                 break;
4027
4028         case EXT4_MOUNT_ORDERED_DATA:
4029         case EXT4_MOUNT_WRITEBACK_DATA:
4030                 if (!jbd2_journal_check_available_features
4031                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4032                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4033                                "requested data journaling mode");
4034                         goto failed_mount_wq;
4035                 }
4036         default:
4037                 break;
4038         }
4039         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4040
4041         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4042
4043 no_journal:
4044         if (ext4_mballoc_ready) {
4045                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4046                 if (!sbi->s_mb_cache) {
4047                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4048                         goto failed_mount_wq;
4049                 }
4050         }
4051
4052         /*
4053          * Get the # of file system overhead blocks from the
4054          * superblock if present.
4055          */
4056         if (es->s_overhead_clusters)
4057                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4058         else {
4059                 err = ext4_calculate_overhead(sb);
4060                 if (err)
4061                         goto failed_mount_wq;
4062         }
4063
4064         /*
4065          * The maximum number of concurrent works can be high and
4066          * concurrency isn't really necessary.  Limit it to 1.
4067          */
4068         EXT4_SB(sb)->rsv_conversion_wq =
4069                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4070         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4071                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4072                 ret = -ENOMEM;
4073                 goto failed_mount4;
4074         }
4075
4076         /*
4077          * The jbd2_journal_load will have done any necessary log recovery,
4078          * so we can safely mount the rest of the filesystem now.
4079          */
4080
4081         root = ext4_iget(sb, EXT4_ROOT_INO);
4082         if (IS_ERR(root)) {
4083                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4084                 ret = PTR_ERR(root);
4085                 root = NULL;
4086                 goto failed_mount4;
4087         }
4088         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4089                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4090                 iput(root);
4091                 goto failed_mount4;
4092         }
4093         sb->s_root = d_make_root(root);
4094         if (!sb->s_root) {
4095                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4096                 ret = -ENOMEM;
4097                 goto failed_mount4;
4098         }
4099
4100         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4101                 sb->s_flags |= MS_RDONLY;
4102
4103         /* determine the minimum size of new large inodes, if present */
4104         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4105                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4106                                                      EXT4_GOOD_OLD_INODE_SIZE;
4107                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4108                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4109                         if (sbi->s_want_extra_isize <
4110                             le16_to_cpu(es->s_want_extra_isize))
4111                                 sbi->s_want_extra_isize =
4112                                         le16_to_cpu(es->s_want_extra_isize);
4113                         if (sbi->s_want_extra_isize <
4114                             le16_to_cpu(es->s_min_extra_isize))
4115                                 sbi->s_want_extra_isize =
4116                                         le16_to_cpu(es->s_min_extra_isize);
4117                 }
4118         }
4119         /* Check if enough inode space is available */
4120         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4121                                                         sbi->s_inode_size) {
4122                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4123                                                        EXT4_GOOD_OLD_INODE_SIZE;
4124                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4125                          "available");
4126         }
4127
4128         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4129         if (err) {
4130                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4131                          "reserved pool", ext4_calculate_resv_clusters(sb));
4132                 goto failed_mount4a;
4133         }
4134
4135         err = ext4_setup_system_zone(sb);
4136         if (err) {
4137                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4138                          "zone (%d)", err);
4139                 goto failed_mount4a;
4140         }
4141
4142         ext4_ext_init(sb);
4143         err = ext4_mb_init(sb);
4144         if (err) {
4145                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4146                          err);
4147                 goto failed_mount5;
4148         }
4149
4150         block = ext4_count_free_clusters(sb);
4151         ext4_free_blocks_count_set(sbi->s_es,
4152                                    EXT4_C2B(sbi, block));
4153         err = percpu_counter_init(&sbi->s_freeclusters_counter, block);
4154         if (!err) {
4155                 unsigned long freei = ext4_count_free_inodes(sb);
4156                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4157                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei);
4158         }
4159         if (!err)
4160                 err = percpu_counter_init(&sbi->s_dirs_counter,
4161                                           ext4_count_dirs(sb));
4162         if (!err)
4163                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
4164         if (err) {
4165                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4166                 goto failed_mount6;
4167         }
4168
4169         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4170                 if (!ext4_fill_flex_info(sb)) {
4171                         ext4_msg(sb, KERN_ERR,
4172                                "unable to initialize "
4173                                "flex_bg meta info!");
4174                         goto failed_mount6;
4175                 }
4176
4177         err = ext4_register_li_request(sb, first_not_zeroed);
4178         if (err)
4179                 goto failed_mount6;
4180
4181         sbi->s_kobj.kset = ext4_kset;
4182         init_completion(&sbi->s_kobj_unregister);
4183         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4184                                    "%s", sb->s_id);
4185         if (err)
4186                 goto failed_mount7;
4187
4188 #ifdef CONFIG_QUOTA
4189         /* Enable quota usage during mount. */
4190         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4191             !(sb->s_flags & MS_RDONLY)) {
4192                 err = ext4_enable_quotas(sb);
4193                 if (err)
4194                         goto failed_mount8;
4195         }
4196 #endif  /* CONFIG_QUOTA */
4197
4198         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4199         ext4_orphan_cleanup(sb, es);
4200         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4201         if (needs_recovery) {
4202                 ext4_msg(sb, KERN_INFO, "recovery complete");
4203                 ext4_mark_recovery_complete(sb, es);
4204         }
4205         if (EXT4_SB(sb)->s_journal) {
4206                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4207                         descr = " journalled data mode";
4208                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4209                         descr = " ordered data mode";
4210                 else
4211                         descr = " writeback data mode";
4212         } else
4213                 descr = "out journal";
4214
4215         if (test_opt(sb, DISCARD)) {
4216                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4217                 if (!blk_queue_discard(q))
4218                         ext4_msg(sb, KERN_WARNING,
4219                                  "mounting with \"discard\" option, but "
4220                                  "the device does not support discard");
4221         }
4222
4223         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4224                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4225                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4226
4227         if (es->s_error_count)
4228                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4229
4230         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4231         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4232         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4233         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4234
4235         kfree(orig_data);
4236         return 0;
4237
4238 cantfind_ext4:
4239         if (!silent)
4240                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4241         goto failed_mount;
4242
4243 #ifdef CONFIG_QUOTA
4244 failed_mount8:
4245         kobject_del(&sbi->s_kobj);
4246 #endif
4247 failed_mount7:
4248         ext4_unregister_li_request(sb);
4249 failed_mount6:
4250         ext4_mb_release(sb);
4251         if (sbi->s_flex_groups)
4252                 ext4_kvfree(sbi->s_flex_groups);
4253         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4254         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4255         percpu_counter_destroy(&sbi->s_dirs_counter);
4256         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4257 failed_mount5:
4258         ext4_ext_release(sb);
4259         ext4_release_system_zone(sb);
4260 failed_mount4a:
4261         dput(sb->s_root);
4262         sb->s_root = NULL;
4263 failed_mount4:
4264         ext4_msg(sb, KERN_ERR, "mount failed");
4265         if (EXT4_SB(sb)->rsv_conversion_wq)
4266                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4267 failed_mount_wq:
4268         if (sbi->s_journal) {
4269                 jbd2_journal_destroy(sbi->s_journal);
4270                 sbi->s_journal = NULL;
4271         }
4272         ext4_es_unregister_shrinker(sbi);
4273 failed_mount3:
4274         del_timer_sync(&sbi->s_err_report);
4275         if (sbi->s_mmp_tsk)
4276                 kthread_stop(sbi->s_mmp_tsk);
4277 failed_mount2:
4278         for (i = 0; i < db_count; i++)
4279                 brelse(sbi->s_group_desc[i]);
4280         ext4_kvfree(sbi->s_group_desc);
4281 failed_mount:
4282         if (sbi->s_chksum_driver)
4283                 crypto_free_shash(sbi->s_chksum_driver);
4284         if (sbi->s_proc) {
4285                 remove_proc_entry("options", sbi->s_proc);
4286                 remove_proc_entry(sb->s_id, ext4_proc_root);
4287         }
4288 #ifdef CONFIG_QUOTA
4289         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4290                 kfree(sbi->s_qf_names[i]);
4291 #endif
4292         ext4_blkdev_remove(sbi);
4293         brelse(bh);
4294 out_fail:
4295         sb->s_fs_info = NULL;
4296         kfree(sbi->s_blockgroup_lock);
4297         kfree(sbi);
4298 out_free_orig:
4299         kfree(orig_data);
4300         return err ? err : ret;
4301 }
4302
4303 /*
4304  * Setup any per-fs journal parameters now.  We'll do this both on
4305  * initial mount, once the journal has been initialised but before we've
4306  * done any recovery; and again on any subsequent remount.
4307  */
4308 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4309 {
4310         struct ext4_sb_info *sbi = EXT4_SB(sb);
4311
4312         journal->j_commit_interval = sbi->s_commit_interval;
4313         journal->j_min_batch_time = sbi->s_min_batch_time;
4314         journal->j_max_batch_time = sbi->s_max_batch_time;
4315
4316         write_lock(&journal->j_state_lock);
4317         if (test_opt(sb, BARRIER))
4318                 journal->j_flags |= JBD2_BARRIER;
4319         else
4320                 journal->j_flags &= ~JBD2_BARRIER;
4321         if (test_opt(sb, DATA_ERR_ABORT))
4322                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4323         else
4324                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4325         write_unlock(&journal->j_state_lock);
4326 }
4327
4328 static journal_t *ext4_get_journal(struct super_block *sb,
4329                                    unsigned int journal_inum)
4330 {
4331         struct inode *journal_inode;
4332         journal_t *journal;
4333
4334         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4335
4336         /* First, test for the existence of a valid inode on disk.  Bad
4337          * things happen if we iget() an unused inode, as the subsequent
4338          * iput() will try to delete it. */
4339
4340         journal_inode = ext4_iget(sb, journal_inum);
4341         if (IS_ERR(journal_inode)) {
4342                 ext4_msg(sb, KERN_ERR, "no journal found");
4343                 return NULL;
4344         }
4345         if (!journal_inode->i_nlink) {
4346                 make_bad_inode(journal_inode);
4347                 iput(journal_inode);
4348                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4349                 return NULL;
4350         }
4351
4352         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4353                   journal_inode, journal_inode->i_size);
4354         if (!S_ISREG(journal_inode->i_mode)) {
4355                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4356                 iput(journal_inode);
4357                 return NULL;
4358         }
4359
4360         journal = jbd2_journal_init_inode(journal_inode);
4361         if (!journal) {
4362                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4363                 iput(journal_inode);
4364                 return NULL;
4365         }
4366         journal->j_private = sb;
4367         ext4_init_journal_params(sb, journal);
4368         return journal;
4369 }
4370
4371 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4372                                        dev_t j_dev)
4373 {
4374         struct buffer_head *bh;
4375         journal_t *journal;
4376         ext4_fsblk_t start;
4377         ext4_fsblk_t len;
4378         int hblock, blocksize;
4379         ext4_fsblk_t sb_block;
4380         unsigned long offset;
4381         struct ext4_super_block *es;
4382         struct block_device *bdev;
4383
4384         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4385
4386         bdev = ext4_blkdev_get(j_dev, sb);
4387         if (bdev == NULL)
4388                 return NULL;
4389
4390         blocksize = sb->s_blocksize;
4391         hblock = bdev_logical_block_size(bdev);
4392         if (blocksize < hblock) {
4393                 ext4_msg(sb, KERN_ERR,
4394                         "blocksize too small for journal device");
4395                 goto out_bdev;
4396         }
4397
4398         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4399         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4400         set_blocksize(bdev, blocksize);
4401         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4402                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4403                        "external journal");
4404                 goto out_bdev;
4405         }
4406
4407         es = (struct ext4_super_block *) (bh->b_data + offset);
4408         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4409             !(le32_to_cpu(es->s_feature_incompat) &
4410               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4411                 ext4_msg(sb, KERN_ERR, "external journal has "
4412                                         "bad superblock");
4413                 brelse(bh);
4414                 goto out_bdev;
4415         }
4416
4417         if ((le32_to_cpu(es->s_feature_ro_compat) &
4418              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4419             es->s_checksum != ext4_superblock_csum(sb, es)) {
4420                 ext4_msg(sb, KERN_ERR, "external journal has "
4421                                        "corrupt superblock");
4422                 brelse(bh);
4423                 goto out_bdev;
4424         }
4425
4426         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4427                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4428                 brelse(bh);
4429                 goto out_bdev;
4430         }
4431
4432         len = ext4_blocks_count(es);
4433         start = sb_block + 1;
4434         brelse(bh);     /* we're done with the superblock */
4435
4436         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4437                                         start, len, blocksize);
4438         if (!journal) {
4439                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4440                 goto out_bdev;
4441         }
4442         journal->j_private = sb;
4443         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4444         wait_on_buffer(journal->j_sb_buffer);
4445         if (!buffer_uptodate(journal->j_sb_buffer)) {
4446                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4447                 goto out_journal;
4448         }
4449         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4450                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4451                                         "user (unsupported) - %d",
4452                         be32_to_cpu(journal->j_superblock->s_nr_users));
4453                 goto out_journal;
4454         }
4455         EXT4_SB(sb)->journal_bdev = bdev;
4456         ext4_init_journal_params(sb, journal);
4457         return journal;
4458
4459 out_journal:
4460         jbd2_journal_destroy(journal);
4461 out_bdev:
4462         ext4_blkdev_put(bdev);
4463         return NULL;
4464 }
4465
4466 static int ext4_load_journal(struct super_block *sb,
4467                              struct ext4_super_block *es,
4468                              unsigned long journal_devnum)
4469 {
4470         journal_t *journal;
4471         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4472         dev_t journal_dev;
4473         int err = 0;
4474         int really_read_only;
4475
4476         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4477
4478         if (journal_devnum &&
4479             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4480                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4481                         "numbers have changed");
4482                 journal_dev = new_decode_dev(journal_devnum);
4483         } else
4484                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4485
4486         really_read_only = bdev_read_only(sb->s_bdev);
4487
4488         /*
4489          * Are we loading a blank journal or performing recovery after a
4490          * crash?  For recovery, we need to check in advance whether we
4491          * can get read-write access to the device.
4492          */
4493         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4494                 if (sb->s_flags & MS_RDONLY) {
4495                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4496                                         "required on readonly filesystem");
4497                         if (really_read_only) {
4498                                 ext4_msg(sb, KERN_ERR, "write access "
4499                                         "unavailable, cannot proceed");
4500                                 return -EROFS;
4501                         }
4502                         ext4_msg(sb, KERN_INFO, "write access will "
4503                                "be enabled during recovery");
4504                 }
4505         }
4506
4507         if (journal_inum && journal_dev) {
4508                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4509                        "and inode journals!");
4510                 return -EINVAL;
4511         }
4512
4513         if (journal_inum) {
4514                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4515                         return -EINVAL;
4516         } else {
4517                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4518                         return -EINVAL;
4519         }
4520
4521         if (!(journal->j_flags & JBD2_BARRIER))
4522                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4523
4524         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4525                 err = jbd2_journal_wipe(journal, !really_read_only);
4526         if (!err) {
4527                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4528                 if (save)
4529                         memcpy(save, ((char *) es) +
4530                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4531                 err = jbd2_journal_load(journal);
4532                 if (save)
4533                         memcpy(((char *) es) + EXT4_S_ERR_START,
4534                                save, EXT4_S_ERR_LEN);
4535                 kfree(save);
4536         }
4537
4538         if (err) {
4539                 ext4_msg(sb, KERN_ERR, "error loading journal");
4540                 jbd2_journal_destroy(journal);
4541                 return err;
4542         }
4543
4544         EXT4_SB(sb)->s_journal = journal;
4545         ext4_clear_journal_err(sb, es);
4546
4547         if (!really_read_only && journal_devnum &&
4548             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4549                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4550
4551                 /* Make sure we flush the recovery flag to disk. */
4552                 ext4_commit_super(sb, 1);
4553         }
4554
4555         return 0;
4556 }
4557
4558 static int ext4_commit_super(struct super_block *sb, int sync)
4559 {
4560         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4561         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4562         int error = 0;
4563
4564         if (!sbh || block_device_ejected(sb))
4565                 return error;
4566         if (buffer_write_io_error(sbh)) {
4567                 /*
4568                  * Oh, dear.  A previous attempt to write the
4569                  * superblock failed.  This could happen because the
4570                  * USB device was yanked out.  Or it could happen to
4571                  * be a transient write error and maybe the block will
4572                  * be remapped.  Nothing we can do but to retry the
4573                  * write and hope for the best.
4574                  */
4575                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4576                        "superblock detected");
4577                 clear_buffer_write_io_error(sbh);
4578                 set_buffer_uptodate(sbh);
4579         }
4580         /*
4581          * If the file system is mounted read-only, don't update the
4582          * superblock write time.  This avoids updating the superblock
4583          * write time when we are mounting the root file system
4584          * read/only but we need to replay the journal; at that point,
4585          * for people who are east of GMT and who make their clock
4586          * tick in localtime for Windows bug-for-bug compatibility,
4587          * the clock is set in the future, and this will cause e2fsck
4588          * to complain and force a full file system check.
4589          */
4590         if (!(sb->s_flags & MS_RDONLY))
4591                 es->s_wtime = cpu_to_le32(get_seconds());
4592         if (sb->s_bdev->bd_part)
4593                 es->s_kbytes_written =
4594                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4595                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4596                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4597         else
4598                 es->s_kbytes_written =
4599                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4600         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4601                 ext4_free_blocks_count_set(es,
4602                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4603                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4604         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4605                 es->s_free_inodes_count =
4606                         cpu_to_le32(percpu_counter_sum_positive(
4607                                 &EXT4_SB(sb)->s_freeinodes_counter));
4608         BUFFER_TRACE(sbh, "marking dirty");
4609         ext4_superblock_csum_set(sb);
4610         mark_buffer_dirty(sbh);
4611         if (sync) {
4612                 error = sync_dirty_buffer(sbh);
4613                 if (error)
4614                         return error;
4615
4616                 error = buffer_write_io_error(sbh);
4617                 if (error) {
4618                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4619                                "superblock");
4620                         clear_buffer_write_io_error(sbh);
4621                         set_buffer_uptodate(sbh);
4622                 }
4623         }
4624         return error;
4625 }
4626
4627 /*
4628  * Have we just finished recovery?  If so, and if we are mounting (or
4629  * remounting) the filesystem readonly, then we will end up with a
4630  * consistent fs on disk.  Record that fact.
4631  */
4632 static void ext4_mark_recovery_complete(struct super_block *sb,
4633                                         struct ext4_super_block *es)
4634 {
4635         journal_t *journal = EXT4_SB(sb)->s_journal;
4636
4637         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4638                 BUG_ON(journal != NULL);
4639                 return;
4640         }
4641         jbd2_journal_lock_updates(journal);
4642         if (jbd2_journal_flush(journal) < 0)
4643                 goto out;
4644
4645         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4646             sb->s_flags & MS_RDONLY) {
4647                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4648                 ext4_commit_super(sb, 1);
4649         }
4650
4651 out:
4652         jbd2_journal_unlock_updates(journal);
4653 }
4654
4655 /*
4656  * If we are mounting (or read-write remounting) a filesystem whose journal
4657  * has recorded an error from a previous lifetime, move that error to the
4658  * main filesystem now.
4659  */
4660 static void ext4_clear_journal_err(struct super_block *sb,
4661                                    struct ext4_super_block *es)
4662 {
4663         journal_t *journal;
4664         int j_errno;
4665         const char *errstr;
4666
4667         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4668
4669         journal = EXT4_SB(sb)->s_journal;
4670
4671         /*
4672          * Now check for any error status which may have been recorded in the
4673          * journal by a prior ext4_error() or ext4_abort()
4674          */
4675
4676         j_errno = jbd2_journal_errno(journal);
4677         if (j_errno) {
4678                 char nbuf[16];
4679
4680                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4681                 ext4_warning(sb, "Filesystem error recorded "
4682                              "from previous mount: %s", errstr);
4683                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4684
4685                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4686                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4687                 ext4_commit_super(sb, 1);
4688
4689                 jbd2_journal_clear_err(journal);
4690                 jbd2_journal_update_sb_errno(journal);
4691         }
4692 }
4693
4694 /*
4695  * Force the running and committing transactions to commit,
4696  * and wait on the commit.
4697  */
4698 int ext4_force_commit(struct super_block *sb)
4699 {
4700         journal_t *journal;
4701
4702         if (sb->s_flags & MS_RDONLY)
4703                 return 0;
4704
4705         journal = EXT4_SB(sb)->s_journal;
4706         return ext4_journal_force_commit(journal);
4707 }
4708
4709 static int ext4_sync_fs(struct super_block *sb, int wait)
4710 {
4711         int ret = 0;
4712         tid_t target;
4713         bool needs_barrier = false;
4714         struct ext4_sb_info *sbi = EXT4_SB(sb);
4715
4716         trace_ext4_sync_fs(sb, wait);
4717         flush_workqueue(sbi->rsv_conversion_wq);
4718         /*
4719          * Writeback quota in non-journalled quota case - journalled quota has
4720          * no dirty dquots
4721          */
4722         dquot_writeback_dquots(sb, -1);
4723         /*
4724          * Data writeback is possible w/o journal transaction, so barrier must
4725          * being sent at the end of the function. But we can skip it if
4726          * transaction_commit will do it for us.
4727          */
4728         target = jbd2_get_latest_transaction(sbi->s_journal);
4729         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4730             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4731                 needs_barrier = true;
4732
4733         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4734                 if (wait)
4735                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4736         }
4737         if (needs_barrier) {
4738                 int err;
4739                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4740                 if (!ret)
4741                         ret = err;
4742         }
4743
4744         return ret;
4745 }
4746
4747 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4748 {
4749         int ret = 0;
4750
4751         trace_ext4_sync_fs(sb, wait);
4752         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4753         dquot_writeback_dquots(sb, -1);
4754         if (wait && test_opt(sb, BARRIER))
4755                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4756
4757         return ret;
4758 }
4759
4760 /*
4761  * LVM calls this function before a (read-only) snapshot is created.  This
4762  * gives us a chance to flush the journal completely and mark the fs clean.
4763  *
4764  * Note that only this function cannot bring a filesystem to be in a clean
4765  * state independently. It relies on upper layer to stop all data & metadata
4766  * modifications.
4767  */
4768 static int ext4_freeze(struct super_block *sb)
4769 {
4770         int error = 0;
4771         journal_t *journal;
4772
4773         if (sb->s_flags & MS_RDONLY)
4774                 return 0;
4775
4776         journal = EXT4_SB(sb)->s_journal;
4777
4778         /* Now we set up the journal barrier. */
4779         jbd2_journal_lock_updates(journal);
4780
4781         /*
4782          * Don't clear the needs_recovery flag if we failed to flush
4783          * the journal.
4784          */
4785         error = jbd2_journal_flush(journal);
4786         if (error < 0)
4787                 goto out;
4788
4789         /* Journal blocked and flushed, clear needs_recovery flag. */
4790         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4791         error = ext4_commit_super(sb, 1);
4792 out:
4793         /* we rely on upper layer to stop further updates */
4794         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4795         return error;
4796 }
4797
4798 /*
4799  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4800  * flag here, even though the filesystem is not technically dirty yet.
4801  */
4802 static int ext4_unfreeze(struct super_block *sb)
4803 {
4804         if (sb->s_flags & MS_RDONLY)
4805                 return 0;
4806
4807         /* Reset the needs_recovery flag before the fs is unlocked. */
4808         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4809         ext4_commit_super(sb, 1);
4810         return 0;
4811 }
4812
4813 /*
4814  * Structure to save mount options for ext4_remount's benefit
4815  */
4816 struct ext4_mount_options {
4817         unsigned long s_mount_opt;
4818         unsigned long s_mount_opt2;
4819         kuid_t s_resuid;
4820         kgid_t s_resgid;
4821         unsigned long s_commit_interval;
4822         u32 s_min_batch_time, s_max_batch_time;
4823 #ifdef CONFIG_QUOTA
4824         int s_jquota_fmt;
4825         char *s_qf_names[EXT4_MAXQUOTAS];
4826 #endif
4827 };
4828
4829 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4830 {
4831         struct ext4_super_block *es;
4832         struct ext4_sb_info *sbi = EXT4_SB(sb);
4833         unsigned long old_sb_flags;
4834         struct ext4_mount_options old_opts;
4835         int enable_quota = 0;
4836         ext4_group_t g;
4837         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4838         int err = 0;
4839 #ifdef CONFIG_QUOTA
4840         int i, j;
4841 #endif
4842         char *orig_data = kstrdup(data, GFP_KERNEL);
4843
4844         /* Store the original options */
4845         old_sb_flags = sb->s_flags;
4846         old_opts.s_mount_opt = sbi->s_mount_opt;
4847         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4848         old_opts.s_resuid = sbi->s_resuid;
4849         old_opts.s_resgid = sbi->s_resgid;
4850         old_opts.s_commit_interval = sbi->s_commit_interval;
4851         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4852         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4853 #ifdef CONFIG_QUOTA
4854         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4855         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4856                 if (sbi->s_qf_names[i]) {
4857                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4858                                                          GFP_KERNEL);
4859                         if (!old_opts.s_qf_names[i]) {
4860                                 for (j = 0; j < i; j++)
4861                                         kfree(old_opts.s_qf_names[j]);
4862                                 kfree(orig_data);
4863                                 return -ENOMEM;
4864                         }
4865                 } else
4866                         old_opts.s_qf_names[i] = NULL;
4867 #endif
4868         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4869                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4870
4871         /*
4872          * Allow the "check" option to be passed as a remount option.
4873          */
4874         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4875                 err = -EINVAL;
4876                 goto restore_opts;
4877         }
4878
4879         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4880                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4881                         ext4_msg(sb, KERN_ERR, "can't mount with "
4882                                  "both data=journal and delalloc");
4883                         err = -EINVAL;
4884                         goto restore_opts;
4885                 }
4886                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4887                         ext4_msg(sb, KERN_ERR, "can't mount with "
4888                                  "both data=journal and dioread_nolock");
4889                         err = -EINVAL;
4890                         goto restore_opts;
4891                 }
4892         }
4893
4894         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4895                 ext4_abort(sb, "Abort forced by user");
4896
4897         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4898                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4899
4900         es = sbi->s_es;
4901
4902         if (sbi->s_journal) {
4903                 ext4_init_journal_params(sb, sbi->s_journal);
4904                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4905         }
4906
4907         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4908                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4909                         err = -EROFS;
4910                         goto restore_opts;
4911                 }
4912
4913                 if (*flags & MS_RDONLY) {
4914                         err = sync_filesystem(sb);
4915                         if (err < 0)
4916                                 goto restore_opts;
4917                         err = dquot_suspend(sb, -1);
4918                         if (err < 0)
4919                                 goto restore_opts;
4920
4921                         /*
4922                          * First of all, the unconditional stuff we have to do
4923                          * to disable replay of the journal when we next remount
4924                          */
4925                         sb->s_flags |= MS_RDONLY;
4926
4927                         /*
4928                          * OK, test if we are remounting a valid rw partition
4929                          * readonly, and if so set the rdonly flag and then
4930                          * mark the partition as valid again.
4931                          */
4932                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4933                             (sbi->s_mount_state & EXT4_VALID_FS))
4934                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4935
4936                         if (sbi->s_journal)
4937                                 ext4_mark_recovery_complete(sb, es);
4938                 } else {
4939                         /* Make sure we can mount this feature set readwrite */
4940                         if (!ext4_feature_set_ok(sb, 0)) {
4941                                 err = -EROFS;
4942                                 goto restore_opts;
4943                         }
4944                         /*
4945                          * Make sure the group descriptor checksums
4946                          * are sane.  If they aren't, refuse to remount r/w.
4947                          */
4948                         for (g = 0; g < sbi->s_groups_count; g++) {
4949                                 struct ext4_group_desc *gdp =
4950                                         ext4_get_group_desc(sb, g, NULL);
4951
4952                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4953                                         ext4_msg(sb, KERN_ERR,
4954                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4955                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4956                                                le16_to_cpu(gdp->bg_checksum));
4957                                         err = -EINVAL;
4958                                         goto restore_opts;
4959                                 }
4960                         }
4961
4962                         /*
4963                          * If we have an unprocessed orphan list hanging
4964                          * around from a previously readonly bdev mount,
4965                          * require a full umount/remount for now.
4966                          */
4967                         if (es->s_last_orphan) {
4968                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4969                                        "remount RDWR because of unprocessed "
4970                                        "orphan inode list.  Please "
4971                                        "umount/remount instead");
4972                                 err = -EINVAL;
4973                                 goto restore_opts;
4974                         }
4975
4976                         /*
4977                          * Mounting a RDONLY partition read-write, so reread
4978                          * and store the current valid flag.  (It may have
4979                          * been changed by e2fsck since we originally mounted
4980                          * the partition.)
4981                          */
4982                         if (sbi->s_journal)
4983                                 ext4_clear_journal_err(sb, es);
4984                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4985                         if (!ext4_setup_super(sb, es, 0))
4986                                 sb->s_flags &= ~MS_RDONLY;
4987                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4988                                                      EXT4_FEATURE_INCOMPAT_MMP))
4989                                 if (ext4_multi_mount_protect(sb,
4990                                                 le64_to_cpu(es->s_mmp_block))) {
4991                                         err = -EROFS;
4992                                         goto restore_opts;
4993                                 }
4994                         enable_quota = 1;
4995                 }
4996         }
4997
4998         /*
4999          * Reinitialize lazy itable initialization thread based on
5000          * current settings
5001          */
5002         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5003                 ext4_unregister_li_request(sb);
5004         else {
5005                 ext4_group_t first_not_zeroed;
5006                 first_not_zeroed = ext4_has_uninit_itable(sb);
5007                 ext4_register_li_request(sb, first_not_zeroed);
5008         }
5009
5010         ext4_setup_system_zone(sb);
5011         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5012                 ext4_commit_super(sb, 1);
5013
5014 #ifdef CONFIG_QUOTA
5015         /* Release old quota file names */
5016         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5017                 kfree(old_opts.s_qf_names[i]);
5018         if (enable_quota) {
5019                 if (sb_any_quota_suspended(sb))
5020                         dquot_resume(sb, -1);
5021                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5022                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5023                         err = ext4_enable_quotas(sb);
5024                         if (err)
5025                                 goto restore_opts;
5026                 }
5027         }
5028 #endif
5029
5030         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5031         kfree(orig_data);
5032         return 0;
5033
5034 restore_opts:
5035         sb->s_flags = old_sb_flags;
5036         sbi->s_mount_opt = old_opts.s_mount_opt;
5037         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5038         sbi->s_resuid = old_opts.s_resuid;
5039         sbi->s_resgid = old_opts.s_resgid;
5040         sbi->s_commit_interval = old_opts.s_commit_interval;
5041         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5042         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5043 #ifdef CONFIG_QUOTA
5044         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5045         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5046                 kfree(sbi->s_qf_names[i]);
5047                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5048         }
5049 #endif
5050         kfree(orig_data);
5051         return err;
5052 }
5053
5054 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5055 {
5056         struct super_block *sb = dentry->d_sb;
5057         struct ext4_sb_info *sbi = EXT4_SB(sb);
5058         struct ext4_super_block *es = sbi->s_es;
5059         ext4_fsblk_t overhead = 0, resv_blocks;
5060         u64 fsid;
5061         s64 bfree;
5062         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5063
5064         if (!test_opt(sb, MINIX_DF))
5065                 overhead = sbi->s_overhead;
5066
5067         buf->f_type = EXT4_SUPER_MAGIC;
5068         buf->f_bsize = sb->s_blocksize;
5069         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5070         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5071                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5072         /* prevent underflow in case that few free space is available */
5073         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5074         buf->f_bavail = buf->f_bfree -
5075                         (ext4_r_blocks_count(es) + resv_blocks);
5076         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5077                 buf->f_bavail = 0;
5078         buf->f_files = le32_to_cpu(es->s_inodes_count);
5079         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5080         buf->f_namelen = EXT4_NAME_LEN;
5081         fsid = le64_to_cpup((void *)es->s_uuid) ^
5082                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5083         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5084         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5085
5086         return 0;
5087 }
5088
5089 /* Helper function for writing quotas on sync - we need to start transaction
5090  * before quota file is locked for write. Otherwise the are possible deadlocks:
5091  * Process 1                         Process 2
5092  * ext4_create()                     quota_sync()
5093  *   jbd2_journal_start()                  write_dquot()
5094  *   dquot_initialize()                         down(dqio_mutex)
5095  *     down(dqio_mutex)                    jbd2_journal_start()
5096  *
5097  */
5098
5099 #ifdef CONFIG_QUOTA
5100
5101 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5102 {
5103         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5104 }
5105
5106 static int ext4_write_dquot(struct dquot *dquot)
5107 {
5108         int ret, err;
5109         handle_t *handle;
5110         struct inode *inode;
5111
5112         inode = dquot_to_inode(dquot);
5113         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5114                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5115         if (IS_ERR(handle))
5116                 return PTR_ERR(handle);
5117         ret = dquot_commit(dquot);
5118         err = ext4_journal_stop(handle);
5119         if (!ret)
5120                 ret = err;
5121         return ret;
5122 }
5123
5124 static int ext4_acquire_dquot(struct dquot *dquot)
5125 {
5126         int ret, err;
5127         handle_t *handle;
5128
5129         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5130                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5131         if (IS_ERR(handle))
5132                 return PTR_ERR(handle);
5133         ret = dquot_acquire(dquot);
5134         err = ext4_journal_stop(handle);
5135         if (!ret)
5136                 ret = err;
5137         return ret;
5138 }
5139
5140 static int ext4_release_dquot(struct dquot *dquot)
5141 {
5142         int ret, err;
5143         handle_t *handle;
5144
5145         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5146                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5147         if (IS_ERR(handle)) {
5148                 /* Release dquot anyway to avoid endless cycle in dqput() */
5149                 dquot_release(dquot);
5150                 return PTR_ERR(handle);
5151         }
5152         ret = dquot_release(dquot);
5153         err = ext4_journal_stop(handle);
5154         if (!ret)
5155                 ret = err;
5156         return ret;
5157 }
5158
5159 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5160 {
5161         struct super_block *sb = dquot->dq_sb;
5162         struct ext4_sb_info *sbi = EXT4_SB(sb);
5163
5164         /* Are we journaling quotas? */
5165         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5166             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5167                 dquot_mark_dquot_dirty(dquot);
5168                 return ext4_write_dquot(dquot);
5169         } else {
5170                 return dquot_mark_dquot_dirty(dquot);
5171         }
5172 }
5173
5174 static int ext4_write_info(struct super_block *sb, int type)
5175 {
5176         int ret, err;
5177         handle_t *handle;
5178
5179         /* Data block + inode block */
5180         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5181         if (IS_ERR(handle))
5182                 return PTR_ERR(handle);
5183         ret = dquot_commit_info(sb, type);
5184         err = ext4_journal_stop(handle);
5185         if (!ret)
5186                 ret = err;
5187         return ret;
5188 }
5189
5190 /*
5191  * Turn on quotas during mount time - we need to find
5192  * the quota file and such...
5193  */
5194 static int ext4_quota_on_mount(struct super_block *sb, int type)
5195 {
5196         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5197                                         EXT4_SB(sb)->s_jquota_fmt, type);
5198 }
5199
5200 /*
5201  * Standard function to be called on quota_on
5202  */
5203 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5204                          struct path *path)
5205 {
5206         int err;
5207
5208         if (!test_opt(sb, QUOTA))
5209                 return -EINVAL;
5210
5211         /* Quotafile not on the same filesystem? */
5212         if (path->dentry->d_sb != sb)
5213                 return -EXDEV;
5214         /* Journaling quota? */
5215         if (EXT4_SB(sb)->s_qf_names[type]) {
5216                 /* Quotafile not in fs root? */
5217                 if (path->dentry->d_parent != sb->s_root)
5218                         ext4_msg(sb, KERN_WARNING,
5219                                 "Quota file not on filesystem root. "
5220                                 "Journaled quota will not work");
5221         }
5222
5223         /*
5224          * When we journal data on quota file, we have to flush journal to see
5225          * all updates to the file when we bypass pagecache...
5226          */
5227         if (EXT4_SB(sb)->s_journal &&
5228             ext4_should_journal_data(path->dentry->d_inode)) {
5229                 /*
5230                  * We don't need to lock updates but journal_flush() could
5231                  * otherwise be livelocked...
5232                  */
5233                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5234                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5235                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5236                 if (err)
5237                         return err;
5238         }
5239
5240         return dquot_quota_on(sb, type, format_id, path);
5241 }
5242
5243 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5244                              unsigned int flags)
5245 {
5246         int err;
5247         struct inode *qf_inode;
5248         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5249                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5250                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5251         };
5252
5253         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5254
5255         if (!qf_inums[type])
5256                 return -EPERM;
5257
5258         qf_inode = ext4_iget(sb, qf_inums[type]);
5259         if (IS_ERR(qf_inode)) {
5260                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5261                 return PTR_ERR(qf_inode);
5262         }
5263
5264         /* Don't account quota for quota files to avoid recursion */
5265         qf_inode->i_flags |= S_NOQUOTA;
5266         err = dquot_enable(qf_inode, type, format_id, flags);
5267         iput(qf_inode);
5268
5269         return err;
5270 }
5271
5272 /* Enable usage tracking for all quota types. */
5273 static int ext4_enable_quotas(struct super_block *sb)
5274 {
5275         int type, err = 0;
5276         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5277                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5278                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5279         };
5280
5281         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5282         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5283                 if (qf_inums[type]) {
5284                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5285                                                 DQUOT_USAGE_ENABLED);
5286                         if (err) {
5287                                 ext4_warning(sb,
5288                                         "Failed to enable quota tracking "
5289                                         "(type=%d, err=%d). Please run "
5290                                         "e2fsck to fix.", type, err);
5291                                 return err;
5292                         }
5293                 }
5294         }
5295         return 0;
5296 }
5297
5298 /*
5299  * quota_on function that is used when QUOTA feature is set.
5300  */
5301 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5302                                  int format_id)
5303 {
5304         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5305                 return -EINVAL;
5306
5307         /*
5308          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5309          */
5310         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5311 }
5312
5313 static int ext4_quota_off(struct super_block *sb, int type)
5314 {
5315         struct inode *inode = sb_dqopt(sb)->files[type];
5316         handle_t *handle;
5317
5318         /* Force all delayed allocation blocks to be allocated.
5319          * Caller already holds s_umount sem */
5320         if (test_opt(sb, DELALLOC))
5321                 sync_filesystem(sb);
5322
5323         if (!inode)
5324                 goto out;
5325
5326         /* Update modification times of quota files when userspace can
5327          * start looking at them */
5328         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5329         if (IS_ERR(handle))
5330                 goto out;
5331         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5332         ext4_mark_inode_dirty(handle, inode);
5333         ext4_journal_stop(handle);
5334
5335 out:
5336         return dquot_quota_off(sb, type);
5337 }
5338
5339 /*
5340  * quota_off function that is used when QUOTA feature is set.
5341  */
5342 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5343 {
5344         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5345                 return -EINVAL;
5346
5347         /* Disable only the limits. */
5348         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5349 }
5350
5351 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5352  * acquiring the locks... As quota files are never truncated and quota code
5353  * itself serializes the operations (and no one else should touch the files)
5354  * we don't have to be afraid of races */
5355 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5356                                size_t len, loff_t off)
5357 {
5358         struct inode *inode = sb_dqopt(sb)->files[type];
5359         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5360         int offset = off & (sb->s_blocksize - 1);
5361         int tocopy;
5362         size_t toread;
5363         struct buffer_head *bh;
5364         loff_t i_size = i_size_read(inode);
5365
5366         if (off > i_size)
5367                 return 0;
5368         if (off+len > i_size)
5369                 len = i_size-off;
5370         toread = len;
5371         while (toread > 0) {
5372                 tocopy = sb->s_blocksize - offset < toread ?
5373                                 sb->s_blocksize - offset : toread;
5374                 bh = ext4_bread(NULL, inode, blk, 0);
5375                 if (IS_ERR(bh))
5376                         return PTR_ERR(bh);
5377                 if (!bh)        /* A hole? */
5378                         memset(data, 0, tocopy);
5379                 else
5380                         memcpy(data, bh->b_data+offset, tocopy);
5381                 brelse(bh);
5382                 offset = 0;
5383                 toread -= tocopy;
5384                 data += tocopy;
5385                 blk++;
5386         }
5387         return len;
5388 }
5389
5390 /* Write to quotafile (we know the transaction is already started and has
5391  * enough credits) */
5392 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5393                                 const char *data, size_t len, loff_t off)
5394 {
5395         struct inode *inode = sb_dqopt(sb)->files[type];
5396         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5397         int err, offset = off & (sb->s_blocksize - 1);
5398         struct buffer_head *bh;
5399         handle_t *handle = journal_current_handle();
5400
5401         if (EXT4_SB(sb)->s_journal && !handle) {
5402                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5403                         " cancelled because transaction is not started",
5404                         (unsigned long long)off, (unsigned long long)len);
5405                 return -EIO;
5406         }
5407         /*
5408          * Since we account only one data block in transaction credits,
5409          * then it is impossible to cross a block boundary.
5410          */
5411         if (sb->s_blocksize - offset < len) {
5412                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5413                         " cancelled because not block aligned",
5414                         (unsigned long long)off, (unsigned long long)len);
5415                 return -EIO;
5416         }
5417
5418         bh = ext4_bread(handle, inode, blk, 1);
5419         if (IS_ERR(bh))
5420                 return PTR_ERR(bh);
5421         if (!bh)
5422                 goto out;
5423         BUFFER_TRACE(bh, "get write access");
5424         err = ext4_journal_get_write_access(handle, bh);
5425         if (err) {
5426                 brelse(bh);
5427                 return err;
5428         }
5429         lock_buffer(bh);
5430         memcpy(bh->b_data+offset, data, len);
5431         flush_dcache_page(bh->b_page);
5432         unlock_buffer(bh);
5433         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5434         brelse(bh);
5435 out:
5436         if (inode->i_size < off + len) {
5437                 i_size_write(inode, off + len);
5438                 EXT4_I(inode)->i_disksize = inode->i_size;
5439                 ext4_mark_inode_dirty(handle, inode);
5440         }
5441         return len;
5442 }
5443
5444 #endif
5445
5446 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5447                        const char *dev_name, void *data)
5448 {
5449         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5450 }
5451
5452 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5453 static inline void register_as_ext2(void)
5454 {
5455         int err = register_filesystem(&ext2_fs_type);
5456         if (err)
5457                 printk(KERN_WARNING
5458                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5459 }
5460
5461 static inline void unregister_as_ext2(void)
5462 {
5463         unregister_filesystem(&ext2_fs_type);
5464 }
5465
5466 static inline int ext2_feature_set_ok(struct super_block *sb)
5467 {
5468         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5469                 return 0;
5470         if (sb->s_flags & MS_RDONLY)
5471                 return 1;
5472         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5473                 return 0;
5474         return 1;
5475 }
5476 #else
5477 static inline void register_as_ext2(void) { }
5478 static inline void unregister_as_ext2(void) { }
5479 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5480 #endif
5481
5482 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5483 static inline void register_as_ext3(void)
5484 {
5485         int err = register_filesystem(&ext3_fs_type);
5486         if (err)
5487                 printk(KERN_WARNING
5488                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5489 }
5490
5491 static inline void unregister_as_ext3(void)
5492 {
5493         unregister_filesystem(&ext3_fs_type);
5494 }
5495
5496 static inline int ext3_feature_set_ok(struct super_block *sb)
5497 {
5498         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5499                 return 0;
5500         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5501                 return 0;
5502         if (sb->s_flags & MS_RDONLY)
5503                 return 1;
5504         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5505                 return 0;
5506         return 1;
5507 }
5508 #else
5509 static inline void register_as_ext3(void) { }
5510 static inline void unregister_as_ext3(void) { }
5511 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5512 #endif
5513
5514 static struct file_system_type ext4_fs_type = {
5515         .owner          = THIS_MODULE,
5516         .name           = "ext4",
5517         .mount          = ext4_mount,
5518         .kill_sb        = kill_block_super,
5519         .fs_flags       = FS_REQUIRES_DEV,
5520 };
5521 MODULE_ALIAS_FS("ext4");
5522
5523 static int __init ext4_init_feat_adverts(void)
5524 {
5525         struct ext4_features *ef;
5526         int ret = -ENOMEM;
5527
5528         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5529         if (!ef)
5530                 goto out;
5531
5532         ef->f_kobj.kset = ext4_kset;
5533         init_completion(&ef->f_kobj_unregister);
5534         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5535                                    "features");
5536         if (ret) {
5537                 kfree(ef);
5538                 goto out;
5539         }
5540
5541         ext4_feat = ef;
5542         ret = 0;
5543 out:
5544         return ret;
5545 }
5546
5547 static void ext4_exit_feat_adverts(void)
5548 {
5549         kobject_put(&ext4_feat->f_kobj);
5550         wait_for_completion(&ext4_feat->f_kobj_unregister);
5551         kfree(ext4_feat);
5552 }
5553
5554 /* Shared across all ext4 file systems */
5555 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5556 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5557
5558 static int __init ext4_init_fs(void)
5559 {
5560         int i, err;
5561
5562         ext4_li_info = NULL;
5563         mutex_init(&ext4_li_mtx);
5564
5565         /* Build-time check for flags consistency */
5566         ext4_check_flag_values();
5567
5568         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5569                 mutex_init(&ext4__aio_mutex[i]);
5570                 init_waitqueue_head(&ext4__ioend_wq[i]);
5571         }
5572
5573         err = ext4_init_es();
5574         if (err)
5575                 return err;
5576
5577         err = ext4_init_pageio();
5578         if (err)
5579                 goto out7;
5580
5581         err = ext4_init_system_zone();
5582         if (err)
5583                 goto out6;
5584         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5585         if (!ext4_kset) {
5586                 err = -ENOMEM;
5587                 goto out5;
5588         }
5589         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5590
5591         err = ext4_init_feat_adverts();
5592         if (err)
5593                 goto out4;
5594
5595         err = ext4_init_mballoc();
5596         if (err)
5597                 goto out2;
5598         else
5599                 ext4_mballoc_ready = 1;
5600         err = init_inodecache();
5601         if (err)
5602                 goto out1;
5603         register_as_ext3();
5604         register_as_ext2();
5605         err = register_filesystem(&ext4_fs_type);
5606         if (err)
5607                 goto out;
5608
5609         return 0;
5610 out:
5611         unregister_as_ext2();
5612         unregister_as_ext3();
5613         destroy_inodecache();
5614 out1:
5615         ext4_mballoc_ready = 0;
5616         ext4_exit_mballoc();
5617 out2:
5618         ext4_exit_feat_adverts();
5619 out4:
5620         if (ext4_proc_root)
5621                 remove_proc_entry("fs/ext4", NULL);
5622         kset_unregister(ext4_kset);
5623 out5:
5624         ext4_exit_system_zone();
5625 out6:
5626         ext4_exit_pageio();
5627 out7:
5628         ext4_exit_es();
5629
5630         return err;
5631 }
5632
5633 static void __exit ext4_exit_fs(void)
5634 {
5635         ext4_destroy_lazyinit_thread();
5636         unregister_as_ext2();
5637         unregister_as_ext3();
5638         unregister_filesystem(&ext4_fs_type);
5639         destroy_inodecache();
5640         ext4_exit_mballoc();
5641         ext4_exit_feat_adverts();
5642         remove_proc_entry("fs/ext4", NULL);
5643         kset_unregister(ext4_kset);
5644         ext4_exit_system_zone();
5645         ext4_exit_pageio();
5646         ext4_exit_es();
5647 }
5648
5649 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5650 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5651 MODULE_LICENSE("GPL");
5652 module_init(ext4_init_fs)
5653 module_exit(ext4_exit_fs)