fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 static int ext4_superblock_csum_verify(struct super_block *sb,
 141                                        struct ext4_super_block *es)
 142 {
 143         if (!ext4_has_metadata_csum(sb))
 144                 return 1;
 145
 146         return es->s_checksum == ext4_superblock_csum(sb, es);
 147 }
 148
 149 void ext4_superblock_csum_set(struct super_block *sb)
 150 {
 151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 152
 153         if (!ext4_has_metadata_csum(sb))
 154                 return;
 155
 156         es->s_checksum = ext4_superblock_csum(sb, es);
 157 }
 158
 159 void *ext4_kvmalloc(size_t size, gfp_t flags)
 160 {
 161         void *ret;
 162
 163         ret = kmalloc(size, flags | __GFP_NOWARN);
 164         if (!ret)
 165                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 166         return ret;
 167 }
 168
 169 void *ext4_kvzalloc(size_t size, gfp_t flags)
 170 {
 171         void *ret;
 172
 173         ret = kzalloc(size, flags | __GFP_NOWARN);
 174         if (!ret)
 175                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 176         return ret;
 177 }
 178
 179 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 180                                struct ext4_group_desc *bg)
 181 {
 182         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 183                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 184                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 185 }
 186
 187 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 188                                struct ext4_group_desc *bg)
 189 {
 190         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 191                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 192                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 193 }
 194
 195 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 196                               struct ext4_group_desc *bg)
 197 {
 198         return le32_to_cpu(bg->bg_inode_table_lo) |
 199                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 200                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 201 }
 202
 203 __u32 ext4_free_group_clusters(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 209 }
 210
 211 __u32 ext4_free_inodes_count(struct super_block *sb,
 212                               struct ext4_group_desc *bg)
 213 {
 214         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 217 }
 218
 219 __u32 ext4_used_dirs_count(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 225 }
 226
 227 __u32 ext4_itable_unused_count(struct super_block *sb,
 228                               struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_itable_unused_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 233 }
 234
 235 void ext4_block_bitmap_set(struct super_block *sb,
 236                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 237 {
 238         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 239         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 240                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 241 }
 242
 243 void ext4_inode_bitmap_set(struct super_block *sb,
 244                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 245 {
 246         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 247         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 248                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 249 }
 250
 251 void ext4_inode_table_set(struct super_block *sb,
 252                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 253 {
 254         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 255         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 256                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 257 }
 258
 259 void ext4_free_group_clusters_set(struct super_block *sb,
 260                                   struct ext4_group_desc *bg, __u32 count)
 261 {
 262         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 265 }
 266
 267 void ext4_free_inodes_set(struct super_block *sb,
 268                           struct ext4_group_desc *bg, __u32 count)
 269 {
 270         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 273 }
 274
 275 void ext4_used_dirs_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, __u32 count)
 277 {
 278         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 281 }
 282
 283 void ext4_itable_unused_set(struct super_block *sb,
 284                           struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291
 292 static void __save_error_info(struct super_block *sb, const char *func,
 293                             unsigned int line)
 294 {
 295         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 296
 297         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 298         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 299         es->s_last_error_time = cpu_to_le32(get_seconds());
 300         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 301         es->s_last_error_line = cpu_to_le32(line);
 302         if (!es->s_first_error_time) {
 303                 es->s_first_error_time = es->s_last_error_time;
 304                 strncpy(es->s_first_error_func, func,
 305                         sizeof(es->s_first_error_func));
 306                 es->s_first_error_line = cpu_to_le32(line);
 307                 es->s_first_error_ino = es->s_last_error_ino;
 308                 es->s_first_error_block = es->s_last_error_block;
 309         }
 310         /*
 311          * Start the daily error reporting function if it hasn't been
 312          * started already
 313          */
 314         if (!es->s_error_count)
 315                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 316         le32_add_cpu(&es->s_error_count, 1);
 317 }
 318
 319 static void save_error_info(struct super_block *sb, const char *func,
 320                             unsigned int line)
 321 {
 322         __save_error_info(sb, func, line);
 323         ext4_commit_super(sb, 1);
 324 }
 325
 326 /*
 327  * The del_gendisk() function uninitializes the disk-specific data
 328  * structures, including the bdi structure, without telling anyone
 329  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 330  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 331  * This is a kludge to prevent these oops until we can put in a proper
 332  * hook in del_gendisk() to inform the VFS and file system layers.
 333  */
 334 static int block_device_ejected(struct super_block *sb)
 335 {
 336         struct inode *bd_inode = sb->s_bdev->bd_inode;
 337         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 338
 339         return bdi->dev == NULL;
 340 }
 341
 342 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 343 {
 344         struct super_block              *sb = journal->j_private;
 345         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 346         int                             error = is_journal_aborted(journal);
 347         struct ext4_journal_cb_entry    *jce;
 348
 349         BUG_ON(txn->t_state == T_FINISHED);
 350         spin_lock(&sbi->s_md_lock);
 351         while (!list_empty(&txn->t_private_list)) {
 352                 jce = list_entry(txn->t_private_list.next,
 353                                  struct ext4_journal_cb_entry, jce_list);
 354                 list_del_init(&jce->jce_list);
 355                 spin_unlock(&sbi->s_md_lock);
 356                 jce->jce_func(sb, jce, error);
 357                 spin_lock(&sbi->s_md_lock);
 358         }
 359         spin_unlock(&sbi->s_md_lock);
 360 }
 361
 362 /* Deal with the reporting of failure conditions on a filesystem such as
 363  * inconsistencies detected or read IO failures.
 364  *
 365  * On ext2, we can store the error state of the filesystem in the
 366  * superblock.  That is not possible on ext4, because we may have other
 367  * write ordering constraints on the superblock which prevent us from
 368  * writing it out straight away; and given that the journal is about to
 369  * be aborted, we can't rely on the current, or future, transactions to
 370  * write out the superblock safely.
 371  *
 372  * We'll just use the jbd2_journal_abort() error code to record an error in
 373  * the journal instead.  On recovery, the journal will complain about
 374  * that error until we've noted it down and cleared it.
 375  */
 376
 377 static void ext4_handle_error(struct super_block *sb)
 378 {
 379         if (sb->s_flags & MS_RDONLY)
 380                 return;
 381
 382         if (!test_opt(sb, ERRORS_CONT)) {
 383                 journal_t *journal = EXT4_SB(sb)->s_journal;
 384
 385                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 386                 if (journal)
 387                         jbd2_journal_abort(journal, -EIO);
 388         }
 389         if (test_opt(sb, ERRORS_RO)) {
 390                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 391                 /*
 392                  * Make sure updated value of ->s_mount_flags will be visible
 393                  * before ->s_flags update
 394                  */
 395                 smp_wmb();
 396                 sb->s_flags |= MS_RDONLY;
 397         }
 398         if (test_opt(sb, ERRORS_PANIC))
 399                 panic("EXT4-fs (device %s): panic forced after error\n",
 400                         sb->s_id);
 401 }
 402
 403 #define ext4_error_ratelimit(sb)                                        \
 404                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 405                              "EXT4-fs error")
 406
 407 void __ext4_error(struct super_block *sb, const char *function,
 408                   unsigned int line, const char *fmt, ...)
 409 {
 410         struct va_format vaf;
 411         va_list args;
 412
 413         if (ext4_error_ratelimit(sb)) {
 414                 va_start(args, fmt);
 415                 vaf.fmt = fmt;
 416                 vaf.va = &args;
 417                 printk(KERN_CRIT
 418                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 419                        sb->s_id, function, line, current->comm, &vaf);
 420                 va_end(args);
 421         }
 422         save_error_info(sb, function, line);
 423         ext4_handle_error(sb);
 424 }
 425
 426 void __ext4_error_inode(struct inode *inode, const char *function,
 427                         unsigned int line, ext4_fsblk_t block,
 428                         const char *fmt, ...)
 429 {
 430         va_list args;
 431         struct va_format vaf;
 432         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 433
 434         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 435         es->s_last_error_block = cpu_to_le64(block);
 436         if (ext4_error_ratelimit(inode->i_sb)) {
 437                 va_start(args, fmt);
 438                 vaf.fmt = fmt;
 439                 vaf.va = &args;
 440                 if (block)
 441                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 442                                "inode #%lu: block %llu: comm %s: %pV\n",
 443                                inode->i_sb->s_id, function, line, inode->i_ino,
 444                                block, current->comm, &vaf);
 445                 else
 446                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 447                                "inode #%lu: comm %s: %pV\n",
 448                                inode->i_sb->s_id, function, line, inode->i_ino,
 449                                current->comm, &vaf);
 450                 va_end(args);
 451         }
 452         save_error_info(inode->i_sb, function, line);
 453         ext4_handle_error(inode->i_sb);
 454 }
 455
 456 void __ext4_error_file(struct file *file, const char *function,
 457                        unsigned int line, ext4_fsblk_t block,
 458                        const char *fmt, ...)
 459 {
 460         va_list args;
 461         struct va_format vaf;
 462         struct ext4_super_block *es;
 463         struct inode *inode = file_inode(file);
 464         char pathname[80], *path;
 465
 466         es = EXT4_SB(inode->i_sb)->s_es;
 467         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 468         if (ext4_error_ratelimit(inode->i_sb)) {
 469                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 470                 if (IS_ERR(path))
 471                         path = "(unknown)";
 472                 va_start(args, fmt);
 473                 vaf.fmt = fmt;
 474                 vaf.va = &args;
 475                 if (block)
 476                         printk(KERN_CRIT
 477                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 478                                "block %llu: comm %s: path %s: %pV\n",
 479                                inode->i_sb->s_id, function, line, inode->i_ino,
 480                                block, current->comm, path, &vaf);
 481                 else
 482                         printk(KERN_CRIT
 483                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 484                                "comm %s: path %s: %pV\n",
 485                                inode->i_sb->s_id, function, line, inode->i_ino,
 486                                current->comm, path, &vaf);
 487                 va_end(args);
 488         }
 489         save_error_info(inode->i_sb, function, line);
 490         ext4_handle_error(inode->i_sb);
 491 }
 492
 493 const char *ext4_decode_error(struct super_block *sb, int errno,
 494                               char nbuf[16])
 495 {
 496         char *errstr = NULL;
 497
 498         switch (errno) {
 499         case -EIO:
 500                 errstr = "IO failure";
 501                 break;
 502         case -ENOMEM:
 503                 errstr = "Out of memory";
 504                 break;
 505         case -EROFS:
 506                 if (!sb || (EXT4_SB(sb)->s_journal &&
 507                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 508                         errstr = "Journal has aborted";
 509                 else
 510                         errstr = "Readonly filesystem";
 511                 break;
 512         default:
 513                 /* If the caller passed in an extra buffer for unknown
 514                  * errors, textualise them now.  Else we just return
 515                  * NULL. */
 516                 if (nbuf) {
 517                         /* Check for truncated error codes... */
 518                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 519                                 errstr = nbuf;
 520                 }
 521                 break;
 522         }
 523
 524         return errstr;
 525 }
 526
 527 /* __ext4_std_error decodes expected errors from journaling functions
 528  * automatically and invokes the appropriate error response.  */
 529
 530 void __ext4_std_error(struct super_block *sb, const char *function,
 531                       unsigned int line, int errno)
 532 {
 533         char nbuf[16];
 534         const char *errstr;
 535
 536         /* Special case: if the error is EROFS, and we're not already
 537          * inside a transaction, then there's really no point in logging
 538          * an error. */
 539         if (errno == -EROFS && journal_current_handle() == NULL &&
 540             (sb->s_flags & MS_RDONLY))
 541                 return;
 542
 543         if (ext4_error_ratelimit(sb)) {
 544                 errstr = ext4_decode_error(sb, errno, nbuf);
 545                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 546                        sb->s_id, function, line, errstr);
 547         }
 548
 549         save_error_info(sb, function, line);
 550         ext4_handle_error(sb);
 551 }
 552
 553 /*
 554  * ext4_abort is a much stronger failure handler than ext4_error.  The
 555  * abort function may be used to deal with unrecoverable failures such
 556  * as journal IO errors or ENOMEM at a critical moment in log management.
 557  *
 558  * We unconditionally force the filesystem into an ABORT|READONLY state,
 559  * unless the error response on the fs has been set to panic in which
 560  * case we take the easy way out and panic immediately.
 561  */
 562
 563 void __ext4_abort(struct super_block *sb, const char *function,
 564                 unsigned int line, const char *fmt, ...)
 565 {
 566         va_list args;
 567
 568         save_error_info(sb, function, line);
 569         va_start(args, fmt);
 570         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 571                function, line);
 572         vprintk(fmt, args);
 573         printk("\n");
 574         va_end(args);
 575
 576         if ((sb->s_flags & MS_RDONLY) == 0) {
 577                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 578                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 579                 /*
 580                  * Make sure updated value of ->s_mount_flags will be visible
 581                  * before ->s_flags update
 582                  */
 583                 smp_wmb();
 584                 sb->s_flags |= MS_RDONLY;
 585                 if (EXT4_SB(sb)->s_journal)
 586                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 587                 save_error_info(sb, function, line);
 588         }
 589         if (test_opt(sb, ERRORS_PANIC))
 590                 panic("EXT4-fs panic from previous error\n");
 591 }
 592
 593 void __ext4_msg(struct super_block *sb,
 594                 const char *prefix, const char *fmt, ...)
 595 {
 596         struct va_format vaf;
 597         va_list args;
 598
 599         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 600                 return;
 601
 602         va_start(args, fmt);
 603         vaf.fmt = fmt;
 604         vaf.va = &args;
 605         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 606         va_end(args);
 607 }
 608
 609 void __ext4_warning(struct super_block *sb, const char *function,
 610                     unsigned int line, const char *fmt, ...)
 611 {
 612         struct va_format vaf;
 613         va_list args;
 614
 615         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 616                           "EXT4-fs warning"))
 617                 return;
 618
 619         va_start(args, fmt);
 620         vaf.fmt = fmt;
 621         vaf.va = &args;
 622         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 623                sb->s_id, function, line, &vaf);
 624         va_end(args);
 625 }
 626
 627 void __ext4_grp_locked_error(const char *function, unsigned int line,
 628                              struct super_block *sb, ext4_group_t grp,
 629                              unsigned long ino, ext4_fsblk_t block,
 630                              const char *fmt, ...)
 631 __releases(bitlock)
 632 __acquires(bitlock)
 633 {
 634         struct va_format vaf;
 635         va_list args;
 636         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 637
 638         es->s_last_error_ino = cpu_to_le32(ino);
 639         es->s_last_error_block = cpu_to_le64(block);
 640         __save_error_info(sb, function, line);
 641
 642         if (ext4_error_ratelimit(sb)) {
 643                 va_start(args, fmt);
 644                 vaf.fmt = fmt;
 645                 vaf.va = &args;
 646                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 647                        sb->s_id, function, line, grp);
 648                 if (ino)
 649                         printk(KERN_CONT "inode %lu: ", ino);
 650                 if (block)
 651                         printk(KERN_CONT "block %llu:",
 652                                (unsigned long long) block);
 653                 printk(KERN_CONT "%pV\n", &vaf);
 654                 va_end(args);
 655         }
 656
 657         if (test_opt(sb, ERRORS_CONT)) {
 658                 ext4_commit_super(sb, 0);
 659                 return;
 660         }
 661
 662         ext4_unlock_group(sb, grp);
 663         ext4_handle_error(sb);
 664         /*
 665          * We only get here in the ERRORS_RO case; relocking the group
 666          * may be dangerous, but nothing bad will happen since the
 667          * filesystem will have already been marked read/only and the
 668          * journal has been aborted.  We return 1 as a hint to callers
 669          * who might what to use the return value from
 670          * ext4_grp_locked_error() to distinguish between the
 671          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 672          * aggressively from the ext4 function in question, with a
 673          * more appropriate error code.
 674          */
 675         ext4_lock_group(sb, grp);
 676         return;
 677 }
 678
 679 void ext4_update_dynamic_rev(struct super_block *sb)
 680 {
 681         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 682
 683         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 684                 return;
 685
 686         ext4_warning(sb,
 687                      "updating to rev %d because of new feature flag, "
 688                      "running e2fsck is recommended",
 689                      EXT4_DYNAMIC_REV);
 690
 691         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 692         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 693         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 694         /* leave es->s_feature_*compat flags alone */
 695         /* es->s_uuid will be set by e2fsck if empty */
 696
 697         /*
 698          * The rest of the superblock fields should be zero, and if not it
 699          * means they are likely already in use, so leave them alone.  We
 700          * can leave it up to e2fsck to clean up any inconsistencies there.
 701          */
 702 }
 703
 704 /*
 705  * Open the external journal device
 706  */
 707 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 708 {
 709         struct block_device *bdev;
 710         char b[BDEVNAME_SIZE];
 711
 712         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 713         if (IS_ERR(bdev))
 714                 goto fail;
 715         return bdev;
 716
 717 fail:
 718         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 719                         __bdevname(dev, b), PTR_ERR(bdev));
 720         return NULL;
 721 }
 722
 723 /*
 724  * Release the journal device
 725  */
 726 static void ext4_blkdev_put(struct block_device *bdev)
 727 {
 728         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 729 }
 730
 731 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 732 {
 733         struct block_device *bdev;
 734         bdev = sbi->journal_bdev;
 735         if (bdev) {
 736                 ext4_blkdev_put(bdev);
 737                 sbi->journal_bdev = NULL;
 738         }
 739 }
 740
 741 static inline struct inode *orphan_list_entry(struct list_head *l)
 742 {
 743         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 744 }
 745
 746 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 747 {
 748         struct list_head *l;
 749
 750         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 751                  le32_to_cpu(sbi->s_es->s_last_orphan));
 752
 753         printk(KERN_ERR "sb_info orphan list:\n");
 754         list_for_each(l, &sbi->s_orphan) {
 755                 struct inode *inode = orphan_list_entry(l);
 756                 printk(KERN_ERR "  "
 757                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 758                        inode->i_sb->s_id, inode->i_ino, inode,
 759                        inode->i_mode, inode->i_nlink,
 760                        NEXT_ORPHAN(inode));
 761         }
 762 }
 763
 764 static void ext4_put_super(struct super_block *sb)
 765 {
 766         struct ext4_sb_info *sbi = EXT4_SB(sb);
 767         struct ext4_super_block *es = sbi->s_es;
 768         int i, err;
 769
 770         ext4_unregister_li_request(sb);
 771         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 772
 773         flush_workqueue(sbi->rsv_conversion_wq);
 774         destroy_workqueue(sbi->rsv_conversion_wq);
 775
 776         if (sbi->s_journal) {
 777                 err = jbd2_journal_destroy(sbi->s_journal);
 778                 sbi->s_journal = NULL;
 779                 if (err < 0)
 780                         ext4_abort(sb, "Couldn't clean up the journal");
 781         }
 782
 783         ext4_es_unregister_shrinker(sbi);
 784         del_timer_sync(&sbi->s_err_report);
 785         ext4_release_system_zone(sb);
 786         ext4_mb_release(sb);
 787         ext4_ext_release(sb);
 788         ext4_xattr_put_super(sb);
 789
 790         if (!(sb->s_flags & MS_RDONLY)) {
 791                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 792                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 793         }
 794         if (!(sb->s_flags & MS_RDONLY))
 795                 ext4_commit_super(sb, 1);
 796
 797         if (sbi->s_proc) {
 798                 remove_proc_entry("options", sbi->s_proc);
 799                 remove_proc_entry(sb->s_id, ext4_proc_root);
 800         }
 801         kobject_del(&sbi->s_kobj);
 802
 803         for (i = 0; i < sbi->s_gdb_count; i++)
 804                 brelse(sbi->s_group_desc[i]);
 805         kvfree(sbi->s_group_desc);
 806         kvfree(sbi->s_flex_groups);
 807         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 808         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 809         percpu_counter_destroy(&sbi->s_dirs_counter);
 810         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 811         brelse(sbi->s_sbh);
 812 #ifdef CONFIG_QUOTA
 813         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 814                 kfree(sbi->s_qf_names[i]);
 815 #endif
 816
 817         /* Debugging code just in case the in-memory inode orphan list
 818          * isn't empty.  The on-disk one can be non-empty if we've
 819          * detected an error and taken the fs readonly, but the
 820          * in-memory list had better be clean by this point. */
 821         if (!list_empty(&sbi->s_orphan))
 822                 dump_orphan_list(sb, sbi);
 823         J_ASSERT(list_empty(&sbi->s_orphan));
 824
 825         invalidate_bdev(sb->s_bdev);
 826         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 827                 /*
 828                  * Invalidate the journal device's buffers.  We don't want them
 829                  * floating about in memory - the physical journal device may
 830                  * hotswapped, and it breaks the `ro-after' testing code.
 831                  */
 832                 sync_blockdev(sbi->journal_bdev);
 833                 invalidate_bdev(sbi->journal_bdev);
 834                 ext4_blkdev_remove(sbi);
 835         }
 836         if (sbi->s_mb_cache) {
 837                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 838                 sbi->s_mb_cache = NULL;
 839         }
 840         if (sbi->s_mmp_tsk)
 841                 kthread_stop(sbi->s_mmp_tsk);
 842         sb->s_fs_info = NULL;
 843         /*
 844          * Now that we are completely done shutting down the
 845          * superblock, we need to actually destroy the kobject.
 846          */
 847         kobject_put(&sbi->s_kobj);
 848         wait_for_completion(&sbi->s_kobj_unregister);
 849         if (sbi->s_chksum_driver)
 850                 crypto_free_shash(sbi->s_chksum_driver);
 851         kfree(sbi->s_blockgroup_lock);
 852         kfree(sbi);
 853 }
 854
 855 static struct kmem_cache *ext4_inode_cachep;
 856
 857 /*
 858  * Called inside transaction, so use GFP_NOFS
 859  */
 860 static struct inode *ext4_alloc_inode(struct super_block *sb)
 861 {
 862         struct ext4_inode_info *ei;
 863
 864         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 865         if (!ei)
 866                 return NULL;
 867
 868         ei->vfs_inode.i_version = 1;
 869         spin_lock_init(&ei->i_raw_lock);
 870         INIT_LIST_HEAD(&ei->i_prealloc_list);
 871         spin_lock_init(&ei->i_prealloc_lock);
 872         ext4_es_init_tree(&ei->i_es_tree);
 873         rwlock_init(&ei->i_es_lock);
 874         INIT_LIST_HEAD(&ei->i_es_list);
 875         ei->i_es_all_nr = 0;
 876         ei->i_es_shk_nr = 0;
 877         ei->i_es_shrink_lblk = 0;
 878         ei->i_reserved_data_blocks = 0;
 879         ei->i_reserved_meta_blocks = 0;
 880         ei->i_allocated_meta_blocks = 0;
 881         ei->i_da_metadata_calc_len = 0;
 882         ei->i_da_metadata_calc_last_lblock = 0;
 883         spin_lock_init(&(ei->i_block_reservation_lock));
 884 #ifdef CONFIG_QUOTA
 885         ei->i_reserved_quota = 0;
 886         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 887 #endif
 888         ei->jinode = NULL;
 889         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 890         spin_lock_init(&ei->i_completed_io_lock);
 891         ei->i_sync_tid = 0;
 892         ei->i_datasync_tid = 0;
 893         atomic_set(&ei->i_ioend_count, 0);
 894         atomic_set(&ei->i_unwritten, 0);
 895         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 896
 897         return &ei->vfs_inode;
 898 }
 899
 900 static int ext4_drop_inode(struct inode *inode)
 901 {
 902         int drop = generic_drop_inode(inode);
 903
 904         trace_ext4_drop_inode(inode, drop);
 905         return drop;
 906 }
 907
 908 static void ext4_i_callback(struct rcu_head *head)
 909 {
 910         struct inode *inode = container_of(head, struct inode, i_rcu);
 911         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 912 }
 913
 914 static void ext4_destroy_inode(struct inode *inode)
 915 {
 916         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 917                 ext4_msg(inode->i_sb, KERN_ERR,
 918                          "Inode %lu (%p): orphan list check failed!",
 919                          inode->i_ino, EXT4_I(inode));
 920                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 921                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 922                                 true);
 923                 dump_stack();
 924         }
 925         call_rcu(&inode->i_rcu, ext4_i_callback);
 926 }
 927
 928 static void init_once(void *foo)
 929 {
 930         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 931
 932         INIT_LIST_HEAD(&ei->i_orphan);
 933         init_rwsem(&ei->xattr_sem);
 934         init_rwsem(&ei->i_data_sem);
 935         inode_init_once(&ei->vfs_inode);
 936 }
 937
 938 static int __init init_inodecache(void)
 939 {
 940         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 941                                              sizeof(struct ext4_inode_info),
 942                                              0, (SLAB_RECLAIM_ACCOUNT|
 943                                                 SLAB_MEM_SPREAD),
 944                                              init_once);
 945         if (ext4_inode_cachep == NULL)
 946                 return -ENOMEM;
 947         return 0;
 948 }
 949
 950 static void destroy_inodecache(void)
 951 {
 952         /*
 953          * Make sure all delayed rcu free inodes are flushed before we
 954          * destroy cache.
 955          */
 956         rcu_barrier();
 957         kmem_cache_destroy(ext4_inode_cachep);
 958 }
 959
 960 void ext4_clear_inode(struct inode *inode)
 961 {
 962         invalidate_inode_buffers(inode);
 963         clear_inode(inode);
 964         dquot_drop(inode);
 965         ext4_discard_preallocations(inode);
 966         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 967         if (EXT4_I(inode)->jinode) {
 968                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 969                                                EXT4_I(inode)->jinode);
 970                 jbd2_free_inode(EXT4_I(inode)->jinode);
 971                 EXT4_I(inode)->jinode = NULL;
 972         }
 973 }
 974
 975 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 976                                         u64 ino, u32 generation)
 977 {
 978         struct inode *inode;
 979
 980         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 981                 return ERR_PTR(-ESTALE);
 982         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 983                 return ERR_PTR(-ESTALE);
 984
 985         /* iget isn't really right if the inode is currently unallocated!!
 986          *
 987          * ext4_read_inode will return a bad_inode if the inode had been
 988          * deleted, so we should be safe.
 989          *
 990          * Currently we don't know the generation for parent directory, so
 991          * a generation of 0 means "accept any"
 992          */
 993         inode = ext4_iget_normal(sb, ino);
 994         if (IS_ERR(inode))
 995                 return ERR_CAST(inode);
 996         if (generation && inode->i_generation != generation) {
 997                 iput(inode);
 998                 return ERR_PTR(-ESTALE);
 999         }
1000
1001         return inode;
1002 }
1003
1004 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1005                                         int fh_len, int fh_type)
1006 {
1007         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1008                                     ext4_nfs_get_inode);
1009 }
1010
1011 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1012                                         int fh_len, int fh_type)
1013 {
1014         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1015                                     ext4_nfs_get_inode);
1016 }
1017
1018 /*
1019  * Try to release metadata pages (indirect blocks, directories) which are
1020  * mapped via the block device.  Since these pages could have journal heads
1021  * which would prevent try_to_free_buffers() from freeing them, we must use
1022  * jbd2 layer's try_to_free_buffers() function to release them.
1023  */
1024 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1025                                  gfp_t wait)
1026 {
1027         journal_t *journal = EXT4_SB(sb)->s_journal;
1028
1029         WARN_ON(PageChecked(page));
1030         if (!page_has_buffers(page))
1031                 return 0;
1032         if (journal)
1033                 return jbd2_journal_try_to_free_buffers(journal, page,
1034                                                         wait & ~__GFP_WAIT);
1035         return try_to_free_buffers(page);
1036 }
1037
1038 #ifdef CONFIG_QUOTA
1039 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1040 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1041
1042 static int ext4_write_dquot(struct dquot *dquot);
1043 static int ext4_acquire_dquot(struct dquot *dquot);
1044 static int ext4_release_dquot(struct dquot *dquot);
1045 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1046 static int ext4_write_info(struct super_block *sb, int type);
1047 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1048                          struct path *path);
1049 static int ext4_quota_off(struct super_block *sb, int type);
1050 static int ext4_quota_on_mount(struct super_block *sb, int type);
1051 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1052                                size_t len, loff_t off);
1053 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1054                                 const char *data, size_t len, loff_t off);
1055 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1056                              unsigned int flags);
1057 static int ext4_enable_quotas(struct super_block *sb);
1058
1059 static struct dquot **ext4_get_dquots(struct inode *inode)
1060 {
1061         return EXT4_I(inode)->i_dquot;
1062 }
1063
1064 static const struct dquot_operations ext4_quota_operations = {
1065         .get_reserved_space = ext4_get_reserved_space,
1066         .write_dquot    = ext4_write_dquot,
1067         .acquire_dquot  = ext4_acquire_dquot,
1068         .release_dquot  = ext4_release_dquot,
1069         .mark_dirty     = ext4_mark_dquot_dirty,
1070         .write_info     = ext4_write_info,
1071         .alloc_dquot    = dquot_alloc,
1072         .destroy_dquot  = dquot_destroy,
1073 };
1074
1075 static const struct quotactl_ops ext4_qctl_operations = {
1076         .quota_on       = ext4_quota_on,
1077         .quota_off      = ext4_quota_off,
1078         .quota_sync     = dquot_quota_sync,
1079         .get_info       = dquot_get_dqinfo,
1080         .set_info       = dquot_set_dqinfo,
1081         .get_dqblk      = dquot_get_dqblk,
1082         .set_dqblk      = dquot_set_dqblk
1083 };
1084 #endif
1085
1086 static const struct super_operations ext4_sops = {
1087         .alloc_inode    = ext4_alloc_inode,
1088         .destroy_inode  = ext4_destroy_inode,
1089         .write_inode    = ext4_write_inode,
1090         .dirty_inode    = ext4_dirty_inode,
1091         .drop_inode     = ext4_drop_inode,
1092         .evict_inode    = ext4_evict_inode,
1093         .put_super      = ext4_put_super,
1094         .sync_fs        = ext4_sync_fs,
1095         .freeze_fs      = ext4_freeze,
1096         .unfreeze_fs    = ext4_unfreeze,
1097         .statfs         = ext4_statfs,
1098         .remount_fs     = ext4_remount,
1099         .show_options   = ext4_show_options,
1100 #ifdef CONFIG_QUOTA
1101         .quota_read     = ext4_quota_read,
1102         .quota_write    = ext4_quota_write,
1103         .get_dquots     = ext4_get_dquots,
1104 #endif
1105         .bdev_try_to_free_page = bdev_try_to_free_page,
1106 };
1107
1108 static const struct export_operations ext4_export_ops = {
1109         .fh_to_dentry = ext4_fh_to_dentry,
1110         .fh_to_parent = ext4_fh_to_parent,
1111         .get_parent = ext4_get_parent,
1112 };
1113
1114 enum {
1115         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1116         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1117         Opt_nouid32, Opt_debug, Opt_removed,
1118         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1119         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1120         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1121         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1122         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1123         Opt_data_err_abort, Opt_data_err_ignore,
1124         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1125         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1126         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1127         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1128         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1129         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1130         Opt_inode_readahead_blks, Opt_journal_ioprio,
1131         Opt_dioread_nolock, Opt_dioread_lock,
1132         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1133         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1134 };
1135
1136 static const match_table_t tokens = {
1137         {Opt_bsd_df, "bsddf"},
1138         {Opt_minix_df, "minixdf"},
1139         {Opt_grpid, "grpid"},
1140         {Opt_grpid, "bsdgroups"},
1141         {Opt_nogrpid, "nogrpid"},
1142         {Opt_nogrpid, "sysvgroups"},
1143         {Opt_resgid, "resgid=%u"},
1144         {Opt_resuid, "resuid=%u"},
1145         {Opt_sb, "sb=%u"},
1146         {Opt_err_cont, "errors=continue"},
1147         {Opt_err_panic, "errors=panic"},
1148         {Opt_err_ro, "errors=remount-ro"},
1149         {Opt_nouid32, "nouid32"},
1150         {Opt_debug, "debug"},
1151         {Opt_removed, "oldalloc"},
1152         {Opt_removed, "orlov"},
1153         {Opt_user_xattr, "user_xattr"},
1154         {Opt_nouser_xattr, "nouser_xattr"},
1155         {Opt_acl, "acl"},
1156         {Opt_noacl, "noacl"},
1157         {Opt_noload, "norecovery"},
1158         {Opt_noload, "noload"},
1159         {Opt_removed, "nobh"},
1160         {Opt_removed, "bh"},
1161         {Opt_commit, "commit=%u"},
1162         {Opt_min_batch_time, "min_batch_time=%u"},
1163         {Opt_max_batch_time, "max_batch_time=%u"},
1164         {Opt_journal_dev, "journal_dev=%u"},
1165         {Opt_journal_path, "journal_path=%s"},
1166         {Opt_journal_checksum, "journal_checksum"},
1167         {Opt_nojournal_checksum, "nojournal_checksum"},
1168         {Opt_journal_async_commit, "journal_async_commit"},
1169         {Opt_abort, "abort"},
1170         {Opt_data_journal, "data=journal"},
1171         {Opt_data_ordered, "data=ordered"},
1172         {Opt_data_writeback, "data=writeback"},
1173         {Opt_data_err_abort, "data_err=abort"},
1174         {Opt_data_err_ignore, "data_err=ignore"},
1175         {Opt_offusrjquota, "usrjquota="},
1176         {Opt_usrjquota, "usrjquota=%s"},
1177         {Opt_offgrpjquota, "grpjquota="},
1178         {Opt_grpjquota, "grpjquota=%s"},
1179         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1180         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1181         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1182         {Opt_grpquota, "grpquota"},
1183         {Opt_noquota, "noquota"},
1184         {Opt_quota, "quota"},
1185         {Opt_usrquota, "usrquota"},
1186         {Opt_barrier, "barrier=%u"},
1187         {Opt_barrier, "barrier"},
1188         {Opt_nobarrier, "nobarrier"},
1189         {Opt_i_version, "i_version"},
1190         {Opt_dax, "dax"},
1191         {Opt_stripe, "stripe=%u"},
1192         {Opt_delalloc, "delalloc"},
1193         {Opt_nodelalloc, "nodelalloc"},
1194         {Opt_removed, "mblk_io_submit"},
1195         {Opt_removed, "nomblk_io_submit"},
1196         {Opt_block_validity, "block_validity"},
1197         {Opt_noblock_validity, "noblock_validity"},
1198         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1199         {Opt_journal_ioprio, "journal_ioprio=%u"},
1200         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1201         {Opt_auto_da_alloc, "auto_da_alloc"},
1202         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1203         {Opt_dioread_nolock, "dioread_nolock"},
1204         {Opt_dioread_lock, "dioread_lock"},
1205         {Opt_discard, "discard"},
1206         {Opt_nodiscard, "nodiscard"},
1207         {Opt_init_itable, "init_itable=%u"},
1208         {Opt_init_itable, "init_itable"},
1209         {Opt_noinit_itable, "noinit_itable"},
1210         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1211         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1212         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1213         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1214         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1215         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1216         {Opt_err, NULL},
1217 };
1218
1219 static ext4_fsblk_t get_sb_block(void **data)
1220 {
1221         ext4_fsblk_t    sb_block;
1222         char            *options = (char *) *data;
1223
1224         if (!options || strncmp(options, "sb=", 3) != 0)
1225                 return 1;       /* Default location */
1226
1227         options += 3;
1228         /* TODO: use simple_strtoll with >32bit ext4 */
1229         sb_block = simple_strtoul(options, &options, 0);
1230         if (*options && *options != ',') {
1231                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1232                        (char *) *data);
1233                 return 1;
1234         }
1235         if (*options == ',')
1236                 options++;
1237         *data = (void *) options;
1238
1239         return sb_block;
1240 }
1241
1242 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1243 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1244         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1245
1246 #ifdef CONFIG_QUOTA
1247 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1248 {
1249         struct ext4_sb_info *sbi = EXT4_SB(sb);
1250         char *qname;
1251         int ret = -1;
1252
1253         if (sb_any_quota_loaded(sb) &&
1254                 !sbi->s_qf_names[qtype]) {
1255                 ext4_msg(sb, KERN_ERR,
1256                         "Cannot change journaled "
1257                         "quota options when quota turned on");
1258                 return -1;
1259         }
1260         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1261                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1262                          "when QUOTA feature is enabled");
1263                 return -1;
1264         }
1265         qname = match_strdup(args);
1266         if (!qname) {
1267                 ext4_msg(sb, KERN_ERR,
1268                         "Not enough memory for storing quotafile name");
1269                 return -1;
1270         }
1271         if (sbi->s_qf_names[qtype]) {
1272                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1273                         ret = 1;
1274                 else
1275                         ext4_msg(sb, KERN_ERR,
1276                                  "%s quota file already specified",
1277                                  QTYPE2NAME(qtype));
1278                 goto errout;
1279         }
1280         if (strchr(qname, '/')) {
1281                 ext4_msg(sb, KERN_ERR,
1282                         "quotafile must be on filesystem root");
1283                 goto errout;
1284         }
1285         sbi->s_qf_names[qtype] = qname;
1286         set_opt(sb, QUOTA);
1287         return 1;
1288 errout:
1289         kfree(qname);
1290         return ret;
1291 }
1292
1293 static int clear_qf_name(struct super_block *sb, int qtype)
1294 {
1295
1296         struct ext4_sb_info *sbi = EXT4_SB(sb);
1297
1298         if (sb_any_quota_loaded(sb) &&
1299                 sbi->s_qf_names[qtype]) {
1300                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1301                         " when quota turned on");
1302                 return -1;
1303         }
1304         kfree(sbi->s_qf_names[qtype]);
1305         sbi->s_qf_names[qtype] = NULL;
1306         return 1;
1307 }
1308 #endif
1309
1310 #define MOPT_SET        0x0001
1311 #define MOPT_CLEAR      0x0002
1312 #define MOPT_NOSUPPORT  0x0004
1313 #define MOPT_EXPLICIT   0x0008
1314 #define MOPT_CLEAR_ERR  0x0010
1315 #define MOPT_GTE0       0x0020
1316 #ifdef CONFIG_QUOTA
1317 #define MOPT_Q          0
1318 #define MOPT_QFMT       0x0040
1319 #else
1320 #define MOPT_Q          MOPT_NOSUPPORT
1321 #define MOPT_QFMT       MOPT_NOSUPPORT
1322 #endif
1323 #define MOPT_DATAJ      0x0080
1324 #define MOPT_NO_EXT2    0x0100
1325 #define MOPT_NO_EXT3    0x0200
1326 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1327 #define MOPT_STRING     0x0400
1328
1329 static const struct mount_opts {
1330         int     token;
1331         int     mount_opt;
1332         int     flags;
1333 } ext4_mount_opts[] = {
1334         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1335         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1336         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1337         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1338         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1339         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1340         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1341          MOPT_EXT4_ONLY | MOPT_SET},
1342         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1343          MOPT_EXT4_ONLY | MOPT_CLEAR},
1344         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1345         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1346         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1347          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1348         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1349          MOPT_EXT4_ONLY | MOPT_CLEAR},
1350         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1351          MOPT_EXT4_ONLY | MOPT_CLEAR},
1352         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1353          MOPT_EXT4_ONLY | MOPT_SET},
1354         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1355                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1356          MOPT_EXT4_ONLY | MOPT_SET},
1357         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1358         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1359         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1360         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1361         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1362          MOPT_NO_EXT2 | MOPT_SET},
1363         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1364          MOPT_NO_EXT2 | MOPT_CLEAR},
1365         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1366         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1367         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1368         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1369         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1370         {Opt_commit, 0, MOPT_GTE0},
1371         {Opt_max_batch_time, 0, MOPT_GTE0},
1372         {Opt_min_batch_time, 0, MOPT_GTE0},
1373         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1374         {Opt_init_itable, 0, MOPT_GTE0},
1375         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1376         {Opt_stripe, 0, MOPT_GTE0},
1377         {Opt_resuid, 0, MOPT_GTE0},
1378         {Opt_resgid, 0, MOPT_GTE0},
1379         {Opt_journal_dev, 0, MOPT_GTE0},
1380         {Opt_journal_path, 0, MOPT_STRING},
1381         {Opt_journal_ioprio, 0, MOPT_GTE0},
1382         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1383         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1384         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1385          MOPT_NO_EXT2 | MOPT_DATAJ},
1386         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1387         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1388 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1389         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1390         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1391 #else
1392         {Opt_acl, 0, MOPT_NOSUPPORT},
1393         {Opt_noacl, 0, MOPT_NOSUPPORT},
1394 #endif
1395         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1396         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1397         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1398         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1399                                                         MOPT_SET | MOPT_Q},
1400         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1401                                                         MOPT_SET | MOPT_Q},
1402         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1403                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1404         {Opt_usrjquota, 0, MOPT_Q},
1405         {Opt_grpjquota, 0, MOPT_Q},
1406         {Opt_offusrjquota, 0, MOPT_Q},
1407         {Opt_offgrpjquota, 0, MOPT_Q},
1408         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1409         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1410         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1411         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1412         {Opt_err, 0, 0}
1413 };
1414
1415 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1416                             substring_t *args, unsigned long *journal_devnum,
1417                             unsigned int *journal_ioprio, int is_remount)
1418 {
1419         struct ext4_sb_info *sbi = EXT4_SB(sb);
1420         const struct mount_opts *m;
1421         kuid_t uid;
1422         kgid_t gid;
1423         int arg = 0;
1424
1425 #ifdef CONFIG_QUOTA
1426         if (token == Opt_usrjquota)
1427                 return set_qf_name(sb, USRQUOTA, &args[0]);
1428         else if (token == Opt_grpjquota)
1429                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1430         else if (token == Opt_offusrjquota)
1431                 return clear_qf_name(sb, USRQUOTA);
1432         else if (token == Opt_offgrpjquota)
1433                 return clear_qf_name(sb, GRPQUOTA);
1434 #endif
1435         switch (token) {
1436         case Opt_noacl:
1437         case Opt_nouser_xattr:
1438                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1439                 break;
1440         case Opt_sb:
1441                 return 1;       /* handled by get_sb_block() */
1442         case Opt_removed:
1443                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1444                 return 1;
1445         case Opt_abort:
1446                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1447                 return 1;
1448         case Opt_i_version:
1449                 sb->s_flags |= MS_I_VERSION;
1450                 return 1;
1451         }
1452
1453         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1454                 if (token == m->token)
1455                         break;
1456
1457         if (m->token == Opt_err) {
1458                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1459                          "or missing value", opt);
1460                 return -1;
1461         }
1462
1463         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1464                 ext4_msg(sb, KERN_ERR,
1465                          "Mount option \"%s\" incompatible with ext2", opt);
1466                 return -1;
1467         }
1468         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1469                 ext4_msg(sb, KERN_ERR,
1470                          "Mount option \"%s\" incompatible with ext3", opt);
1471                 return -1;
1472         }
1473
1474         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1475                 return -1;
1476         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1477                 return -1;
1478         if (m->flags & MOPT_EXPLICIT)
1479                 set_opt2(sb, EXPLICIT_DELALLOC);
1480         if (m->flags & MOPT_CLEAR_ERR)
1481                 clear_opt(sb, ERRORS_MASK);
1482         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1483                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1484                          "options when quota turned on");
1485                 return -1;
1486         }
1487
1488         if (m->flags & MOPT_NOSUPPORT) {
1489                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1490         } else if (token == Opt_commit) {
1491                 if (arg == 0)
1492                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1493                 sbi->s_commit_interval = HZ * arg;
1494         } else if (token == Opt_max_batch_time) {
1495                 sbi->s_max_batch_time = arg;
1496         } else if (token == Opt_min_batch_time) {
1497                 sbi->s_min_batch_time = arg;
1498         } else if (token == Opt_inode_readahead_blks) {
1499                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1500                         ext4_msg(sb, KERN_ERR,
1501                                  "EXT4-fs: inode_readahead_blks must be "
1502                                  "0 or a power of 2 smaller than 2^31");
1503                         return -1;
1504                 }
1505                 sbi->s_inode_readahead_blks = arg;
1506         } else if (token == Opt_init_itable) {
1507                 set_opt(sb, INIT_INODE_TABLE);
1508                 if (!args->from)
1509                         arg = EXT4_DEF_LI_WAIT_MULT;
1510                 sbi->s_li_wait_mult = arg;
1511         } else if (token == Opt_max_dir_size_kb) {
1512                 sbi->s_max_dir_size_kb = arg;
1513         } else if (token == Opt_stripe) {
1514                 sbi->s_stripe = arg;
1515         } else if (token == Opt_resuid) {
1516                 uid = make_kuid(current_user_ns(), arg);
1517                 if (!uid_valid(uid)) {
1518                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1519                         return -1;
1520                 }
1521                 sbi->s_resuid = uid;
1522         } else if (token == Opt_resgid) {
1523                 gid = make_kgid(current_user_ns(), arg);
1524                 if (!gid_valid(gid)) {
1525                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1526                         return -1;
1527                 }
1528                 sbi->s_resgid = gid;
1529         } else if (token == Opt_journal_dev) {
1530                 if (is_remount) {
1531                         ext4_msg(sb, KERN_ERR,
1532                                  "Cannot specify journal on remount");
1533                         return -1;
1534                 }
1535                 *journal_devnum = arg;
1536         } else if (token == Opt_journal_path) {
1537                 char *journal_path;
1538                 struct inode *journal_inode;
1539                 struct path path;
1540                 int error;
1541
1542                 if (is_remount) {
1543                         ext4_msg(sb, KERN_ERR,
1544                                  "Cannot specify journal on remount");
1545                         return -1;
1546                 }
1547                 journal_path = match_strdup(&args[0]);
1548                 if (!journal_path) {
1549                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1550                                 "journal device string");
1551                         return -1;
1552                 }
1553
1554                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1555                 if (error) {
1556                         ext4_msg(sb, KERN_ERR, "error: could not find "
1557                                 "journal device path: error %d", error);
1558                         kfree(journal_path);
1559                         return -1;
1560                 }
1561
1562                 journal_inode = path.dentry->d_inode;
1563                 if (!S_ISBLK(journal_inode->i_mode)) {
1564                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1565                                 "is not a block device", journal_path);
1566                         path_put(&path);
1567                         kfree(journal_path);
1568                         return -1;
1569                 }
1570
1571                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1572                 path_put(&path);
1573                 kfree(journal_path);
1574         } else if (token == Opt_journal_ioprio) {
1575                 if (arg > 7) {
1576                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1577                                  " (must be 0-7)");
1578                         return -1;
1579                 }
1580                 *journal_ioprio =
1581                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1582         } else if (m->flags & MOPT_DATAJ) {
1583                 if (is_remount) {
1584                         if (!sbi->s_journal)
1585                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1586                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1587                                 ext4_msg(sb, KERN_ERR,
1588                                          "Cannot change data mode on remount");
1589                                 return -1;
1590                         }
1591                 } else {
1592                         clear_opt(sb, DATA_FLAGS);
1593                         sbi->s_mount_opt |= m->mount_opt;
1594                 }
1595 #ifdef CONFIG_QUOTA
1596         } else if (m->flags & MOPT_QFMT) {
1597                 if (sb_any_quota_loaded(sb) &&
1598                     sbi->s_jquota_fmt != m->mount_opt) {
1599                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1600                                  "quota options when quota turned on");
1601                         return -1;
1602                 }
1603                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1604                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1605                         ext4_msg(sb, KERN_ERR,
1606                                  "Cannot set journaled quota options "
1607                                  "when QUOTA feature is enabled");
1608                         return -1;
1609                 }
1610                 sbi->s_jquota_fmt = m->mount_opt;
1611 #endif
1612 #ifndef CONFIG_FS_DAX
1613         } else if (token == Opt_dax) {
1614                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1615                 return -1;
1616 #endif
1617         } else {
1618                 if (!args->from)
1619                         arg = 1;
1620                 if (m->flags & MOPT_CLEAR)
1621                         arg = !arg;
1622                 else if (unlikely(!(m->flags & MOPT_SET))) {
1623                         ext4_msg(sb, KERN_WARNING,
1624                                  "buggy handling of option %s", opt);
1625                         WARN_ON(1);
1626                         return -1;
1627                 }
1628                 if (arg != 0)
1629                         sbi->s_mount_opt |= m->mount_opt;
1630                 else
1631                         sbi->s_mount_opt &= ~m->mount_opt;
1632         }
1633         return 1;
1634 }
1635
1636 static int parse_options(char *options, struct super_block *sb,
1637                          unsigned long *journal_devnum,
1638                          unsigned int *journal_ioprio,
1639                          int is_remount)
1640 {
1641         struct ext4_sb_info *sbi = EXT4_SB(sb);
1642         char *p;
1643         substring_t args[MAX_OPT_ARGS];
1644         int token;
1645
1646         if (!options)
1647                 return 1;
1648
1649         while ((p = strsep(&options, ",")) != NULL) {
1650                 if (!*p)
1651                         continue;
1652                 /*
1653                  * Initialize args struct so we know whether arg was
1654                  * found; some options take optional arguments.
1655                  */
1656                 args[0].to = args[0].from = NULL;
1657                 token = match_token(p, tokens, args);
1658                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1659                                      journal_ioprio, is_remount) < 0)
1660                         return 0;
1661         }
1662 #ifdef CONFIG_QUOTA
1663         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1664             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1665                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1666                          "feature is enabled");
1667                 return 0;
1668         }
1669         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1670                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1671                         clear_opt(sb, USRQUOTA);
1672
1673                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1674                         clear_opt(sb, GRPQUOTA);
1675
1676                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1677                         ext4_msg(sb, KERN_ERR, "old and new quota "
1678                                         "format mixing");
1679                         return 0;
1680                 }
1681
1682                 if (!sbi->s_jquota_fmt) {
1683                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1684                                         "not specified");
1685                         return 0;
1686                 }
1687         }
1688 #endif
1689         if (test_opt(sb, DIOREAD_NOLOCK)) {
1690                 int blocksize =
1691                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1692
1693                 if (blocksize < PAGE_CACHE_SIZE) {
1694                         ext4_msg(sb, KERN_ERR, "can't mount with "
1695                                  "dioread_nolock if block size != PAGE_SIZE");
1696                         return 0;
1697                 }
1698         }
1699         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1700             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1701                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1702                          "in data=ordered mode");
1703                 return 0;
1704         }
1705         return 1;
1706 }
1707
1708 static inline void ext4_show_quota_options(struct seq_file *seq,
1709                                            struct super_block *sb)
1710 {
1711 #if defined(CONFIG_QUOTA)
1712         struct ext4_sb_info *sbi = EXT4_SB(sb);
1713
1714         if (sbi->s_jquota_fmt) {
1715                 char *fmtname = "";
1716
1717                 switch (sbi->s_jquota_fmt) {
1718                 case QFMT_VFS_OLD:
1719                         fmtname = "vfsold";
1720                         break;
1721                 case QFMT_VFS_V0:
1722                         fmtname = "vfsv0";
1723                         break;
1724                 case QFMT_VFS_V1:
1725                         fmtname = "vfsv1";
1726                         break;
1727                 }
1728                 seq_printf(seq, ",jqfmt=%s", fmtname);
1729         }
1730
1731         if (sbi->s_qf_names[USRQUOTA])
1732                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1733
1734         if (sbi->s_qf_names[GRPQUOTA])
1735                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1736 #endif
1737 }
1738
1739 static const char *token2str(int token)
1740 {
1741         const struct match_token *t;
1742
1743         for (t = tokens; t->token != Opt_err; t++)
1744                 if (t->token == token && !strchr(t->pattern, '='))
1745                         break;
1746         return t->pattern;
1747 }
1748
1749 /*
1750  * Show an option if
1751  *  - it's set to a non-default value OR
1752  *  - if the per-sb default is different from the global default
1753  */
1754 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1755                               int nodefs)
1756 {
1757         struct ext4_sb_info *sbi = EXT4_SB(sb);
1758         struct ext4_super_block *es = sbi->s_es;
1759         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1760         const struct mount_opts *m;
1761         char sep = nodefs ? '\n' : ',';
1762
1763 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1764 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1765
1766         if (sbi->s_sb_block != 1)
1767                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1768
1769         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1770                 int want_set = m->flags & MOPT_SET;
1771                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1772                     (m->flags & MOPT_CLEAR_ERR))
1773                         continue;
1774                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1775                         continue; /* skip if same as the default */
1776                 if ((want_set &&
1777                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1778                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1779                         continue; /* select Opt_noFoo vs Opt_Foo */
1780                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1781         }
1782
1783         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1784             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1785                 SEQ_OPTS_PRINT("resuid=%u",
1786                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1787         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1788             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1789                 SEQ_OPTS_PRINT("resgid=%u",
1790                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1791         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1792         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1793                 SEQ_OPTS_PUTS("errors=remount-ro");
1794         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1795                 SEQ_OPTS_PUTS("errors=continue");
1796         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1797                 SEQ_OPTS_PUTS("errors=panic");
1798         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1799                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1800         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1801                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1802         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1803                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1804         if (sb->s_flags & MS_I_VERSION)
1805                 SEQ_OPTS_PUTS("i_version");
1806         if (nodefs || sbi->s_stripe)
1807                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1808         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1809                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1810                         SEQ_OPTS_PUTS("data=journal");
1811                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1812                         SEQ_OPTS_PUTS("data=ordered");
1813                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1814                         SEQ_OPTS_PUTS("data=writeback");
1815         }
1816         if (nodefs ||
1817             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1818                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1819                                sbi->s_inode_readahead_blks);
1820
1821         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1822                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1823                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1824         if (nodefs || sbi->s_max_dir_size_kb)
1825                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1826
1827         ext4_show_quota_options(seq, sb);
1828         return 0;
1829 }
1830
1831 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1832 {
1833         return _ext4_show_options(seq, root->d_sb, 0);
1834 }
1835
1836 static int options_seq_show(struct seq_file *seq, void *offset)
1837 {
1838         struct super_block *sb = seq->private;
1839         int rc;
1840
1841         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1842         rc = _ext4_show_options(seq, sb, 1);
1843         seq_puts(seq, "\n");
1844         return rc;
1845 }
1846
1847 static int options_open_fs(struct inode *inode, struct file *file)
1848 {
1849         return single_open(file, options_seq_show, PDE_DATA(inode));
1850 }
1851
1852 static const struct file_operations ext4_seq_options_fops = {
1853         .owner = THIS_MODULE,
1854         .open = options_open_fs,
1855         .read = seq_read,
1856         .llseek = seq_lseek,
1857         .release = single_release,
1858 };
1859
1860 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1861                             int read_only)
1862 {
1863         struct ext4_sb_info *sbi = EXT4_SB(sb);
1864         int res = 0;
1865
1866         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1867                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1868                          "forcing read-only mode");
1869                 res = MS_RDONLY;
1870         }
1871         if (read_only)
1872                 goto done;
1873         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1874                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1875                          "running e2fsck is recommended");
1876         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1877                 ext4_msg(sb, KERN_WARNING,
1878                          "warning: mounting fs with errors, "
1879                          "running e2fsck is recommended");
1880         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1881                  le16_to_cpu(es->s_mnt_count) >=
1882                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "warning: maximal mount count reached, "
1885                          "running e2fsck is recommended");
1886         else if (le32_to_cpu(es->s_checkinterval) &&
1887                 (le32_to_cpu(es->s_lastcheck) +
1888                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "warning: checktime reached, "
1891                          "running e2fsck is recommended");
1892         if (!sbi->s_journal)
1893                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1894         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1895                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1896         le16_add_cpu(&es->s_mnt_count, 1);
1897         es->s_mtime = cpu_to_le32(get_seconds());
1898         ext4_update_dynamic_rev(sb);
1899         if (sbi->s_journal)
1900                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1901
1902         ext4_commit_super(sb, 1);
1903 done:
1904         if (test_opt(sb, DEBUG))
1905                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1906                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1907                         sb->s_blocksize,
1908                         sbi->s_groups_count,
1909                         EXT4_BLOCKS_PER_GROUP(sb),
1910                         EXT4_INODES_PER_GROUP(sb),
1911                         sbi->s_mount_opt, sbi->s_mount_opt2);
1912
1913         cleancache_init_fs(sb);
1914         return res;
1915 }
1916
1917 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1918 {
1919         struct ext4_sb_info *sbi = EXT4_SB(sb);
1920         struct flex_groups *new_groups;
1921         int size;
1922
1923         if (!sbi->s_log_groups_per_flex)
1924                 return 0;
1925
1926         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1927         if (size <= sbi->s_flex_groups_allocated)
1928                 return 0;
1929
1930         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1931         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1932         if (!new_groups) {
1933                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1934                          size / (int) sizeof(struct flex_groups));
1935                 return -ENOMEM;
1936         }
1937
1938         if (sbi->s_flex_groups) {
1939                 memcpy(new_groups, sbi->s_flex_groups,
1940                        (sbi->s_flex_groups_allocated *
1941                         sizeof(struct flex_groups)));
1942                 kvfree(sbi->s_flex_groups);
1943         }
1944         sbi->s_flex_groups = new_groups;
1945         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1946         return 0;
1947 }
1948
1949 static int ext4_fill_flex_info(struct super_block *sb)
1950 {
1951         struct ext4_sb_info *sbi = EXT4_SB(sb);
1952         struct ext4_group_desc *gdp = NULL;
1953         ext4_group_t flex_group;
1954         int i, err;
1955
1956         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1957         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1958                 sbi->s_log_groups_per_flex = 0;
1959                 return 1;
1960         }
1961
1962         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1963         if (err)
1964                 goto failed;
1965
1966         for (i = 0; i < sbi->s_groups_count; i++) {
1967                 gdp = ext4_get_group_desc(sb, i, NULL);
1968
1969                 flex_group = ext4_flex_group(sbi, i);
1970                 atomic_add(ext4_free_inodes_count(sb, gdp),
1971                            &sbi->s_flex_groups[flex_group].free_inodes);
1972                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1973                              &sbi->s_flex_groups[flex_group].free_clusters);
1974                 atomic_add(ext4_used_dirs_count(sb, gdp),
1975                            &sbi->s_flex_groups[flex_group].used_dirs);
1976         }
1977
1978         return 1;
1979 failed:
1980         return 0;
1981 }
1982
1983 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1984                                    struct ext4_group_desc *gdp)
1985 {
1986         int offset;
1987         __u16 crc = 0;
1988         __le32 le_group = cpu_to_le32(block_group);
1989
1990         if (ext4_has_metadata_csum(sbi->s_sb)) {
1991                 /* Use new metadata_csum algorithm */
1992                 __le16 save_csum;
1993                 __u32 csum32;
1994
1995                 save_csum = gdp->bg_checksum;
1996                 gdp->bg_checksum = 0;
1997                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1998                                      sizeof(le_group));
1999                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2000                                      sbi->s_desc_size);
2001                 gdp->bg_checksum = save_csum;
2002
2003                 crc = csum32 & 0xFFFF;
2004                 goto out;
2005         }
2006
2007         /* old crc16 code */
2008         if (!(sbi->s_es->s_feature_ro_compat &
2009               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2010                 return 0;
2011
2012         offset = offsetof(struct ext4_group_desc, bg_checksum);
2013
2014         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2015         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2016         crc = crc16(crc, (__u8 *)gdp, offset);
2017         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2018         /* for checksum of struct ext4_group_desc do the rest...*/
2019         if ((sbi->s_es->s_feature_incompat &
2020              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2021             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2022                 crc = crc16(crc, (__u8 *)gdp + offset,
2023                             le16_to_cpu(sbi->s_es->s_desc_size) -
2024                                 offset);
2025
2026 out:
2027         return cpu_to_le16(crc);
2028 }
2029
2030 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2031                                 struct ext4_group_desc *gdp)
2032 {
2033         if (ext4_has_group_desc_csum(sb) &&
2034             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2035                                                       block_group, gdp)))
2036                 return 0;
2037
2038         return 1;
2039 }
2040
2041 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2042                               struct ext4_group_desc *gdp)
2043 {
2044         if (!ext4_has_group_desc_csum(sb))
2045                 return;
2046         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2047 }
2048
2049 /* Called at mount-time, super-block is locked */
2050 static int ext4_check_descriptors(struct super_block *sb,
2051                                   ext4_group_t *first_not_zeroed)
2052 {
2053         struct ext4_sb_info *sbi = EXT4_SB(sb);
2054         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2055         ext4_fsblk_t last_block;
2056         ext4_fsblk_t block_bitmap;
2057         ext4_fsblk_t inode_bitmap;
2058         ext4_fsblk_t inode_table;
2059         int flexbg_flag = 0;
2060         ext4_group_t i, grp = sbi->s_groups_count;
2061
2062         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2063                 flexbg_flag = 1;
2064
2065         ext4_debug("Checking group descriptors");
2066
2067         for (i = 0; i < sbi->s_groups_count; i++) {
2068                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2069
2070                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2071                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2072                 else
2073                         last_block = first_block +
2074                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2075
2076                 if ((grp == sbi->s_groups_count) &&
2077                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2078                         grp = i;
2079
2080                 block_bitmap = ext4_block_bitmap(sb, gdp);
2081                 if (block_bitmap < first_block || block_bitmap > last_block) {
2082                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2083                                "Block bitmap for group %u not in group "
2084                                "(block %llu)!", i, block_bitmap);
2085                         return 0;
2086                 }
2087                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2088                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2089                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2090                                "Inode bitmap for group %u not in group "
2091                                "(block %llu)!", i, inode_bitmap);
2092                         return 0;
2093                 }
2094                 inode_table = ext4_inode_table(sb, gdp);
2095                 if (inode_table < first_block ||
2096                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2097                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2098                                "Inode table for group %u not in group "
2099                                "(block %llu)!", i, inode_table);
2100                         return 0;
2101                 }
2102                 ext4_lock_group(sb, i);
2103                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2104                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2105                                  "Checksum for group %u failed (%u!=%u)",
2106                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2107                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2108                         if (!(sb->s_flags & MS_RDONLY)) {
2109                                 ext4_unlock_group(sb, i);
2110                                 return 0;
2111                         }
2112                 }
2113                 ext4_unlock_group(sb, i);
2114                 if (!flexbg_flag)
2115                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2116         }
2117         if (NULL != first_not_zeroed)
2118                 *first_not_zeroed = grp;
2119         return 1;
2120 }
2121
2122 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2123  * the superblock) which were deleted from all directories, but held open by
2124  * a process at the time of a crash.  We walk the list and try to delete these
2125  * inodes at recovery time (only with a read-write filesystem).
2126  *
2127  * In order to keep the orphan inode chain consistent during traversal (in
2128  * case of crash during recovery), we link each inode into the superblock
2129  * orphan list_head and handle it the same way as an inode deletion during
2130  * normal operation (which journals the operations for us).
2131  *
2132  * We only do an iget() and an iput() on each inode, which is very safe if we
2133  * accidentally point at an in-use or already deleted inode.  The worst that
2134  * can happen in this case is that we get a "bit already cleared" message from
2135  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2136  * e2fsck was run on this filesystem, and it must have already done the orphan
2137  * inode cleanup for us, so we can safely abort without any further action.
2138  */
2139 static void ext4_orphan_cleanup(struct super_block *sb,
2140                                 struct ext4_super_block *es)
2141 {
2142         unsigned int s_flags = sb->s_flags;
2143         int nr_orphans = 0, nr_truncates = 0;
2144 #ifdef CONFIG_QUOTA
2145         int i;
2146 #endif
2147         if (!es->s_last_orphan) {
2148                 jbd_debug(4, "no orphan inodes to clean up\n");
2149                 return;
2150         }
2151
2152         if (bdev_read_only(sb->s_bdev)) {
2153                 ext4_msg(sb, KERN_ERR, "write access "
2154                         "unavailable, skipping orphan cleanup");
2155                 return;
2156         }
2157
2158         /* Check if feature set would not allow a r/w mount */
2159         if (!ext4_feature_set_ok(sb, 0)) {
2160                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2161                          "unknown ROCOMPAT features");
2162                 return;
2163         }
2164
2165         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2166                 /* don't clear list on RO mount w/ errors */
2167                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2168                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2169                                   "clearing orphan list.\n");
2170                         es->s_last_orphan = 0;
2171                 }
2172                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2173                 return;
2174         }
2175
2176         if (s_flags & MS_RDONLY) {
2177                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2178                 sb->s_flags &= ~MS_RDONLY;
2179         }
2180 #ifdef CONFIG_QUOTA
2181         /* Needed for iput() to work correctly and not trash data */
2182         sb->s_flags |= MS_ACTIVE;
2183         /* Turn on quotas so that they are updated correctly */
2184         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2185                 if (EXT4_SB(sb)->s_qf_names[i]) {
2186                         int ret = ext4_quota_on_mount(sb, i);
2187                         if (ret < 0)
2188                                 ext4_msg(sb, KERN_ERR,
2189                                         "Cannot turn on journaled "
2190                                         "quota: error %d", ret);
2191                 }
2192         }
2193 #endif
2194
2195         while (es->s_last_orphan) {
2196                 struct inode *inode;
2197
2198                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2199                 if (IS_ERR(inode)) {
2200                         es->s_last_orphan = 0;
2201                         break;
2202                 }
2203
2204                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2205                 dquot_initialize(inode);
2206                 if (inode->i_nlink) {
2207                         if (test_opt(sb, DEBUG))
2208                                 ext4_msg(sb, KERN_DEBUG,
2209                                         "%s: truncating inode %lu to %lld bytes",
2210                                         __func__, inode->i_ino, inode->i_size);
2211                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2212                                   inode->i_ino, inode->i_size);
2213                         mutex_lock(&inode->i_mutex);
2214                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2215                         ext4_truncate(inode);
2216                         mutex_unlock(&inode->i_mutex);
2217                         nr_truncates++;
2218                 } else {
2219                         if (test_opt(sb, DEBUG))
2220                                 ext4_msg(sb, KERN_DEBUG,
2221                                         "%s: deleting unreferenced inode %lu",
2222                                         __func__, inode->i_ino);
2223                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2224                                   inode->i_ino);
2225                         nr_orphans++;
2226                 }
2227                 iput(inode);  /* The delete magic happens here! */
2228         }
2229
2230 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2231
2232         if (nr_orphans)
2233                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2234                        PLURAL(nr_orphans));
2235         if (nr_truncates)
2236                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2237                        PLURAL(nr_truncates));
2238 #ifdef CONFIG_QUOTA
2239         /* Turn quotas off */
2240         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2241                 if (sb_dqopt(sb)->files[i])
2242                         dquot_quota_off(sb, i);
2243         }
2244 #endif
2245         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2246 }
2247
2248 /*
2249  * Maximal extent format file size.
2250  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2251  * extent format containers, within a sector_t, and within i_blocks
2252  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2253  * so that won't be a limiting factor.
2254  *
2255  * However there is other limiting factor. We do store extents in the form
2256  * of starting block and length, hence the resulting length of the extent
2257  * covering maximum file size must fit into on-disk format containers as
2258  * well. Given that length is always by 1 unit bigger than max unit (because
2259  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2260  *
2261  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2262  */
2263 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2264 {
2265         loff_t res;
2266         loff_t upper_limit = MAX_LFS_FILESIZE;
2267
2268         /* small i_blocks in vfs inode? */
2269         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2270                 /*
2271                  * CONFIG_LBDAF is not enabled implies the inode
2272                  * i_block represent total blocks in 512 bytes
2273                  * 32 == size of vfs inode i_blocks * 8
2274                  */
2275                 upper_limit = (1LL << 32) - 1;
2276
2277                 /* total blocks in file system block size */
2278                 upper_limit >>= (blkbits - 9);
2279                 upper_limit <<= blkbits;
2280         }
2281
2282         /*
2283          * 32-bit extent-start container, ee_block. We lower the maxbytes
2284          * by one fs block, so ee_len can cover the extent of maximum file
2285          * size
2286          */
2287         res = (1LL << 32) - 1;
2288         res <<= blkbits;
2289
2290         /* Sanity check against vm- & vfs- imposed limits */
2291         if (res > upper_limit)
2292                 res = upper_limit;
2293
2294         return res;
2295 }
2296
2297 /*
2298  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2299  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2300  * We need to be 1 filesystem block less than the 2^48 sector limit.
2301  */
2302 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2303 {
2304         loff_t res = EXT4_NDIR_BLOCKS;
2305         int meta_blocks;
2306         loff_t upper_limit;
2307         /* This is calculated to be the largest file size for a dense, block
2308          * mapped file such that the file's total number of 512-byte sectors,
2309          * including data and all indirect blocks, does not exceed (2^48 - 1).
2310          *
2311          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2312          * number of 512-byte sectors of the file.
2313          */
2314
2315         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2316                 /*
2317                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2318                  * the inode i_block field represents total file blocks in
2319                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2320                  */
2321                 upper_limit = (1LL << 32) - 1;
2322
2323                 /* total blocks in file system block size */
2324                 upper_limit >>= (bits - 9);
2325
2326         } else {
2327                 /*
2328                  * We use 48 bit ext4_inode i_blocks
2329                  * With EXT4_HUGE_FILE_FL set the i_blocks
2330                  * represent total number of blocks in
2331                  * file system block size
2332                  */
2333                 upper_limit = (1LL << 48) - 1;
2334
2335         }
2336
2337         /* indirect blocks */
2338         meta_blocks = 1;
2339         /* double indirect blocks */
2340         meta_blocks += 1 + (1LL << (bits-2));
2341         /* tripple indirect blocks */
2342         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2343
2344         upper_limit -= meta_blocks;
2345         upper_limit <<= bits;
2346
2347         res += 1LL << (bits-2);
2348         res += 1LL << (2*(bits-2));
2349         res += 1LL << (3*(bits-2));
2350         res <<= bits;
2351         if (res > upper_limit)
2352                 res = upper_limit;
2353
2354         if (res > MAX_LFS_FILESIZE)
2355                 res = MAX_LFS_FILESIZE;
2356
2357         return res;
2358 }
2359
2360 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2361                                    ext4_fsblk_t logical_sb_block, int nr)
2362 {
2363         struct ext4_sb_info *sbi = EXT4_SB(sb);
2364         ext4_group_t bg, first_meta_bg;
2365         int has_super = 0;
2366
2367         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2368
2369         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2370             nr < first_meta_bg)
2371                 return logical_sb_block + nr + 1;
2372         bg = sbi->s_desc_per_block * nr;
2373         if (ext4_bg_has_super(sb, bg))
2374                 has_super = 1;
2375
2376         /*
2377          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2378          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2379          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2380          * compensate.
2381          */
2382         if (sb->s_blocksize == 1024 && nr == 0 &&
2383             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2384                 has_super++;
2385
2386         return (has_super + ext4_group_first_block_no(sb, bg));
2387 }
2388
2389 /**
2390  * ext4_get_stripe_size: Get the stripe size.
2391  * @sbi: In memory super block info
2392  *
2393  * If we have specified it via mount option, then
2394  * use the mount option value. If the value specified at mount time is
2395  * greater than the blocks per group use the super block value.
2396  * If the super block value is greater than blocks per group return 0.
2397  * Allocator needs it be less than blocks per group.
2398  *
2399  */
2400 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2401 {
2402         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2403         unsigned long stripe_width =
2404                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2405         int ret;
2406
2407         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2408                 ret = sbi->s_stripe;
2409         else if (stripe_width <= sbi->s_blocks_per_group)
2410                 ret = stripe_width;
2411         else if (stride <= sbi->s_blocks_per_group)
2412                 ret = stride;
2413         else
2414                 ret = 0;
2415
2416         /*
2417          * If the stripe width is 1, this makes no sense and
2418          * we set it to 0 to turn off stripe handling code.
2419          */
2420         if (ret <= 1)
2421                 ret = 0;
2422
2423         return ret;
2424 }
2425
2426 /* sysfs supprt */
2427
2428 struct ext4_attr {
2429         struct attribute attr;
2430         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2431         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2432                          const char *, size_t);
2433         union {
2434                 int offset;
2435                 int deprecated_val;
2436         } u;
2437 };
2438
2439 static int parse_strtoull(const char *buf,
2440                 unsigned long long max, unsigned long long *value)
2441 {
2442         int ret;
2443
2444         ret = kstrtoull(skip_spaces(buf), 0, value);
2445         if (!ret && *value > max)
2446                 ret = -EINVAL;
2447         return ret;
2448 }
2449
2450 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2451                                               struct ext4_sb_info *sbi,
2452                                               char *buf)
2453 {
2454         return snprintf(buf, PAGE_SIZE, "%llu\n",
2455                 (s64) EXT4_C2B(sbi,
2456                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2457 }
2458
2459 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2460                                          struct ext4_sb_info *sbi, char *buf)
2461 {
2462         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2463
2464         if (!sb->s_bdev->bd_part)
2465                 return snprintf(buf, PAGE_SIZE, "0\n");
2466         return snprintf(buf, PAGE_SIZE, "%lu\n",
2467                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2468                          sbi->s_sectors_written_start) >> 1);
2469 }
2470
2471 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2472                                           struct ext4_sb_info *sbi, char *buf)
2473 {
2474         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2475
2476         if (!sb->s_bdev->bd_part)
2477                 return snprintf(buf, PAGE_SIZE, "0\n");
2478         return snprintf(buf, PAGE_SIZE, "%llu\n",
2479                         (unsigned long long)(sbi->s_kbytes_written +
2480                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2481                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2482 }
2483
2484 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2485                                           struct ext4_sb_info *sbi,
2486                                           const char *buf, size_t count)
2487 {
2488         unsigned long t;
2489         int ret;
2490
2491         ret = kstrtoul(skip_spaces(buf), 0, &t);
2492         if (ret)
2493                 return ret;
2494
2495         if (t && (!is_power_of_2(t) || t > 0x40000000))
2496                 return -EINVAL;
2497
2498         sbi->s_inode_readahead_blks = t;
2499         return count;
2500 }
2501
2502 static ssize_t sbi_ui_show(struct ext4_attr *a,
2503                            struct ext4_sb_info *sbi, char *buf)
2504 {
2505         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2506
2507         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2508 }
2509
2510 static ssize_t sbi_ui_store(struct ext4_attr *a,
2511                             struct ext4_sb_info *sbi,
2512                             const char *buf, size_t count)
2513 {
2514         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2515         unsigned long t;
2516         int ret;
2517
2518         ret = kstrtoul(skip_spaces(buf), 0, &t);
2519         if (ret)
2520                 return ret;
2521         *ui = t;
2522         return count;
2523 }
2524
2525 static ssize_t es_ui_show(struct ext4_attr *a,
2526                            struct ext4_sb_info *sbi, char *buf)
2527 {
2528
2529         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2530                            a->u.offset);
2531
2532         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2533 }
2534
2535 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2536                                   struct ext4_sb_info *sbi, char *buf)
2537 {
2538         return snprintf(buf, PAGE_SIZE, "%llu\n",
2539                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2540 }
2541
2542 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2543                                    struct ext4_sb_info *sbi,
2544                                    const char *buf, size_t count)
2545 {
2546         unsigned long long val;
2547         int ret;
2548
2549         if (parse_strtoull(buf, -1ULL, &val))
2550                 return -EINVAL;
2551         ret = ext4_reserve_clusters(sbi, val);
2552
2553         return ret ? ret : count;
2554 }
2555
2556 static ssize_t trigger_test_error(struct ext4_attr *a,
2557                                   struct ext4_sb_info *sbi,
2558                                   const char *buf, size_t count)
2559 {
2560         int len = count;
2561
2562         if (!capable(CAP_SYS_ADMIN))
2563                 return -EPERM;
2564
2565         if (len && buf[len-1] == '\n')
2566                 len--;
2567
2568         if (len)
2569                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2570         return count;
2571 }
2572
2573 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2574                                    struct ext4_sb_info *sbi, char *buf)
2575 {
2576         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2577 }
2578
2579 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2580 static struct ext4_attr ext4_attr_##_name = {                   \
2581         .attr = {.name = __stringify(_name), .mode = _mode },   \
2582         .show   = _show,                                        \
2583         .store  = _store,                                       \
2584         .u = {                                                  \
2585                 .offset = offsetof(struct ext4_sb_info, _elname),\
2586         },                                                      \
2587 }
2588
2589 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2590 static struct ext4_attr ext4_attr_##_name = {                           \
2591         .attr = {.name = __stringify(_name), .mode = _mode },           \
2592         .show   = _show,                                                \
2593         .store  = _store,                                               \
2594         .u = {                                                          \
2595                 .offset = offsetof(struct ext4_super_block, _elname),   \
2596         },                                                              \
2597 }
2598
2599 #define EXT4_ATTR(name, mode, show, store) \
2600 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2601
2602 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2603 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2604 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2605
2606 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2607         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2608 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2609         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2610
2611 #define ATTR_LIST(name) &ext4_attr_##name.attr
2612 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2613 static struct ext4_attr ext4_attr_##_name = {                   \
2614         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2615         .show   = sbi_deprecated_show,                          \
2616         .u = {                                                  \
2617                 .deprecated_val = _val,                         \
2618         },                                                      \
2619 }
2620
2621 EXT4_RO_ATTR(delayed_allocation_blocks);
2622 EXT4_RO_ATTR(session_write_kbytes);
2623 EXT4_RO_ATTR(lifetime_write_kbytes);
2624 EXT4_RW_ATTR(reserved_clusters);
2625 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2626                  inode_readahead_blks_store, s_inode_readahead_blks);
2627 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2628 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2629 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2630 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2631 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2632 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2633 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2634 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2635 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2636 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2637 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2638 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2639 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2640 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2641 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2642 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2643 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2644 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2645 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2646
2647 static struct attribute *ext4_attrs[] = {
2648         ATTR_LIST(delayed_allocation_blocks),
2649         ATTR_LIST(session_write_kbytes),
2650         ATTR_LIST(lifetime_write_kbytes),
2651         ATTR_LIST(reserved_clusters),
2652         ATTR_LIST(inode_readahead_blks),
2653         ATTR_LIST(inode_goal),
2654         ATTR_LIST(mb_stats),
2655         ATTR_LIST(mb_max_to_scan),
2656         ATTR_LIST(mb_min_to_scan),
2657         ATTR_LIST(mb_order2_req),
2658         ATTR_LIST(mb_stream_req),
2659         ATTR_LIST(mb_group_prealloc),
2660         ATTR_LIST(max_writeback_mb_bump),
2661         ATTR_LIST(extent_max_zeroout_kb),
2662         ATTR_LIST(trigger_fs_error),
2663         ATTR_LIST(err_ratelimit_interval_ms),
2664         ATTR_LIST(err_ratelimit_burst),
2665         ATTR_LIST(warning_ratelimit_interval_ms),
2666         ATTR_LIST(warning_ratelimit_burst),
2667         ATTR_LIST(msg_ratelimit_interval_ms),
2668         ATTR_LIST(msg_ratelimit_burst),
2669         ATTR_LIST(errors_count),
2670         ATTR_LIST(first_error_time),
2671         ATTR_LIST(last_error_time),
2672         NULL,
2673 };
2674
2675 /* Features this copy of ext4 supports */
2676 EXT4_INFO_ATTR(lazy_itable_init);
2677 EXT4_INFO_ATTR(batched_discard);
2678 EXT4_INFO_ATTR(meta_bg_resize);
2679
2680 static struct attribute *ext4_feat_attrs[] = {
2681         ATTR_LIST(lazy_itable_init),
2682         ATTR_LIST(batched_discard),
2683         ATTR_LIST(meta_bg_resize),
2684         NULL,
2685 };
2686
2687 static ssize_t ext4_attr_show(struct kobject *kobj,
2688                               struct attribute *attr, char *buf)
2689 {
2690         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2691                                                 s_kobj);
2692         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2693
2694         return a->show ? a->show(a, sbi, buf) : 0;
2695 }
2696
2697 static ssize_t ext4_attr_store(struct kobject *kobj,
2698                                struct attribute *attr,
2699                                const char *buf, size_t len)
2700 {
2701         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2702                                                 s_kobj);
2703         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2704
2705         return a->store ? a->store(a, sbi, buf, len) : 0;
2706 }
2707
2708 static void ext4_sb_release(struct kobject *kobj)
2709 {
2710         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2711                                                 s_kobj);
2712         complete(&sbi->s_kobj_unregister);
2713 }
2714
2715 static const struct sysfs_ops ext4_attr_ops = {
2716         .show   = ext4_attr_show,
2717         .store  = ext4_attr_store,
2718 };
2719
2720 static struct kobj_type ext4_ktype = {
2721         .default_attrs  = ext4_attrs,
2722         .sysfs_ops      = &ext4_attr_ops,
2723         .release        = ext4_sb_release,
2724 };
2725
2726 static void ext4_feat_release(struct kobject *kobj)
2727 {
2728         complete(&ext4_feat->f_kobj_unregister);
2729 }
2730
2731 static ssize_t ext4_feat_show(struct kobject *kobj,
2732                               struct attribute *attr, char *buf)
2733 {
2734         return snprintf(buf, PAGE_SIZE, "supported\n");
2735 }
2736
2737 /*
2738  * We can not use ext4_attr_show/store because it relies on the kobject
2739  * being embedded in the ext4_sb_info structure which is definitely not
2740  * true in this case.
2741  */
2742 static const struct sysfs_ops ext4_feat_ops = {
2743         .show   = ext4_feat_show,
2744         .store  = NULL,
2745 };
2746
2747 static struct kobj_type ext4_feat_ktype = {
2748         .default_attrs  = ext4_feat_attrs,
2749         .sysfs_ops      = &ext4_feat_ops,
2750         .release        = ext4_feat_release,
2751 };
2752
2753 /*
2754  * Check whether this filesystem can be mounted based on
2755  * the features present and the RDONLY/RDWR mount requested.
2756  * Returns 1 if this filesystem can be mounted as requested,
2757  * 0 if it cannot be.
2758  */
2759 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2760 {
2761         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2762                 ext4_msg(sb, KERN_ERR,
2763                         "Couldn't mount because of "
2764                         "unsupported optional features (%x)",
2765                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2766                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2767                 return 0;
2768         }
2769
2770         if (readonly)
2771                 return 1;
2772
2773         /* Check that feature set is OK for a read-write mount */
2774         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2775                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2776                          "unsupported optional features (%x)",
2777                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2778                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2779                 return 0;
2780         }
2781         /*
2782          * Large file size enabled file system can only be mounted
2783          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2784          */
2785         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2786                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2787                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2788                                  "cannot be mounted RDWR without "
2789                                  "CONFIG_LBDAF");
2790                         return 0;
2791                 }
2792         }
2793         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2794             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2795                 ext4_msg(sb, KERN_ERR,
2796                          "Can't support bigalloc feature without "
2797                          "extents feature\n");
2798                 return 0;
2799         }
2800
2801 #ifndef CONFIG_QUOTA
2802         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2803             !readonly) {
2804                 ext4_msg(sb, KERN_ERR,
2805                          "Filesystem with quota feature cannot be mounted RDWR "
2806                          "without CONFIG_QUOTA");
2807                 return 0;
2808         }
2809 #endif  /* CONFIG_QUOTA */
2810         return 1;
2811 }
2812
2813 /*
2814  * This function is called once a day if we have errors logged
2815  * on the file system
2816  */
2817 static void print_daily_error_info(unsigned long arg)
2818 {
2819         struct super_block *sb = (struct super_block *) arg;
2820         struct ext4_sb_info *sbi;
2821         struct ext4_super_block *es;
2822
2823         sbi = EXT4_SB(sb);
2824         es = sbi->s_es;
2825
2826         if (es->s_error_count)
2827                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2828                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2829                          le32_to_cpu(es->s_error_count));
2830         if (es->s_first_error_time) {
2831                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2832                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2833                        (int) sizeof(es->s_first_error_func),
2834                        es->s_first_error_func,
2835                        le32_to_cpu(es->s_first_error_line));
2836                 if (es->s_first_error_ino)
2837                         printk(": inode %u",
2838                                le32_to_cpu(es->s_first_error_ino));
2839                 if (es->s_first_error_block)
2840                         printk(": block %llu", (unsigned long long)
2841                                le64_to_cpu(es->s_first_error_block));
2842                 printk("\n");
2843         }
2844         if (es->s_last_error_time) {
2845                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2846                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2847                        (int) sizeof(es->s_last_error_func),
2848                        es->s_last_error_func,
2849                        le32_to_cpu(es->s_last_error_line));
2850                 if (es->s_last_error_ino)
2851                         printk(": inode %u",
2852                                le32_to_cpu(es->s_last_error_ino));
2853                 if (es->s_last_error_block)
2854                         printk(": block %llu", (unsigned long long)
2855                                le64_to_cpu(es->s_last_error_block));
2856                 printk("\n");
2857         }
2858         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2859 }
2860
2861 /* Find next suitable group and run ext4_init_inode_table */
2862 static int ext4_run_li_request(struct ext4_li_request *elr)
2863 {
2864         struct ext4_group_desc *gdp = NULL;
2865         ext4_group_t group, ngroups;
2866         struct super_block *sb;
2867         unsigned long timeout = 0;
2868         int ret = 0;
2869
2870         sb = elr->lr_super;
2871         ngroups = EXT4_SB(sb)->s_groups_count;
2872
2873         sb_start_write(sb);
2874         for (group = elr->lr_next_group; group < ngroups; group++) {
2875                 gdp = ext4_get_group_desc(sb, group, NULL);
2876                 if (!gdp) {
2877                         ret = 1;
2878                         break;
2879                 }
2880
2881                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2882                         break;
2883         }
2884
2885         if (group >= ngroups)
2886                 ret = 1;
2887
2888         if (!ret) {
2889                 timeout = jiffies;
2890                 ret = ext4_init_inode_table(sb, group,
2891                                             elr->lr_timeout ? 0 : 1);
2892                 if (elr->lr_timeout == 0) {
2893                         timeout = (jiffies - timeout) *
2894                                   elr->lr_sbi->s_li_wait_mult;
2895                         elr->lr_timeout = timeout;
2896                 }
2897                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2898                 elr->lr_next_group = group + 1;
2899         }
2900         sb_end_write(sb);
2901
2902         return ret;
2903 }
2904
2905 /*
2906  * Remove lr_request from the list_request and free the
2907  * request structure. Should be called with li_list_mtx held
2908  */
2909 static void ext4_remove_li_request(struct ext4_li_request *elr)
2910 {
2911         struct ext4_sb_info *sbi;
2912
2913         if (!elr)
2914                 return;
2915
2916         sbi = elr->lr_sbi;
2917
2918         list_del(&elr->lr_request);
2919         sbi->s_li_request = NULL;
2920         kfree(elr);
2921 }
2922
2923 static void ext4_unregister_li_request(struct super_block *sb)
2924 {
2925         mutex_lock(&ext4_li_mtx);
2926         if (!ext4_li_info) {
2927                 mutex_unlock(&ext4_li_mtx);
2928                 return;
2929         }
2930
2931         mutex_lock(&ext4_li_info->li_list_mtx);
2932         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2933         mutex_unlock(&ext4_li_info->li_list_mtx);
2934         mutex_unlock(&ext4_li_mtx);
2935 }
2936
2937 static struct task_struct *ext4_lazyinit_task;
2938
2939 /*
2940  * This is the function where ext4lazyinit thread lives. It walks
2941  * through the request list searching for next scheduled filesystem.
2942  * When such a fs is found, run the lazy initialization request
2943  * (ext4_rn_li_request) and keep track of the time spend in this
2944  * function. Based on that time we compute next schedule time of
2945  * the request. When walking through the list is complete, compute
2946  * next waking time and put itself into sleep.
2947  */
2948 static int ext4_lazyinit_thread(void *arg)
2949 {
2950         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2951         struct list_head *pos, *n;
2952         struct ext4_li_request *elr;
2953         unsigned long next_wakeup, cur;
2954
2955         BUG_ON(NULL == eli);
2956
2957 cont_thread:
2958         while (true) {
2959                 next_wakeup = MAX_JIFFY_OFFSET;
2960
2961                 mutex_lock(&eli->li_list_mtx);
2962                 if (list_empty(&eli->li_request_list)) {
2963                         mutex_unlock(&eli->li_list_mtx);
2964                         goto exit_thread;
2965                 }
2966
2967                 list_for_each_safe(pos, n, &eli->li_request_list) {
2968                         elr = list_entry(pos, struct ext4_li_request,
2969                                          lr_request);
2970
2971                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2972                                 if (ext4_run_li_request(elr) != 0) {
2973                                         /* error, remove the lazy_init job */
2974                                         ext4_remove_li_request(elr);
2975                                         continue;
2976                                 }
2977                         }
2978
2979                         if (time_before(elr->lr_next_sched, next_wakeup))
2980                                 next_wakeup = elr->lr_next_sched;
2981                 }
2982                 mutex_unlock(&eli->li_list_mtx);
2983
2984                 try_to_freeze();
2985
2986                 cur = jiffies;
2987                 if ((time_after_eq(cur, next_wakeup)) ||
2988                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2989                         cond_resched();
2990                         continue;
2991                 }
2992
2993                 schedule_timeout_interruptible(next_wakeup - cur);
2994
2995                 if (kthread_should_stop()) {
2996                         ext4_clear_request_list();
2997                         goto exit_thread;
2998                 }
2999         }
3000
3001 exit_thread:
3002         /*
3003          * It looks like the request list is empty, but we need
3004          * to check it under the li_list_mtx lock, to prevent any
3005          * additions into it, and of course we should lock ext4_li_mtx
3006          * to atomically free the list and ext4_li_info, because at
3007          * this point another ext4 filesystem could be registering
3008          * new one.
3009          */
3010         mutex_lock(&ext4_li_mtx);
3011         mutex_lock(&eli->li_list_mtx);
3012         if (!list_empty(&eli->li_request_list)) {
3013                 mutex_unlock(&eli->li_list_mtx);
3014                 mutex_unlock(&ext4_li_mtx);
3015                 goto cont_thread;
3016         }
3017         mutex_unlock(&eli->li_list_mtx);
3018         kfree(ext4_li_info);
3019         ext4_li_info = NULL;
3020         mutex_unlock(&ext4_li_mtx);
3021
3022         return 0;
3023 }
3024
3025 static void ext4_clear_request_list(void)
3026 {
3027         struct list_head *pos, *n;
3028         struct ext4_li_request *elr;
3029
3030         mutex_lock(&ext4_li_info->li_list_mtx);
3031         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3032                 elr = list_entry(pos, struct ext4_li_request,
3033                                  lr_request);
3034                 ext4_remove_li_request(elr);
3035         }
3036         mutex_unlock(&ext4_li_info->li_list_mtx);
3037 }
3038
3039 static int ext4_run_lazyinit_thread(void)
3040 {
3041         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3042                                          ext4_li_info, "ext4lazyinit");
3043         if (IS_ERR(ext4_lazyinit_task)) {
3044                 int err = PTR_ERR(ext4_lazyinit_task);
3045                 ext4_clear_request_list();
3046                 kfree(ext4_li_info);
3047                 ext4_li_info = NULL;
3048                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3049                                  "initialization thread\n",
3050                                  err);
3051                 return err;
3052         }
3053         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3054         return 0;
3055 }
3056
3057 /*
3058  * Check whether it make sense to run itable init. thread or not.
3059  * If there is at least one uninitialized inode table, return
3060  * corresponding group number, else the loop goes through all
3061  * groups and return total number of groups.
3062  */
3063 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3064 {
3065         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3066         struct ext4_group_desc *gdp = NULL;
3067
3068         for (group = 0; group < ngroups; group++) {
3069                 gdp = ext4_get_group_desc(sb, group, NULL);
3070                 if (!gdp)
3071                         continue;
3072
3073                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3074                         break;
3075         }
3076
3077         return group;
3078 }
3079
3080 static int ext4_li_info_new(void)
3081 {
3082         struct ext4_lazy_init *eli = NULL;
3083
3084         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3085         if (!eli)
3086                 return -ENOMEM;
3087
3088         INIT_LIST_HEAD(&eli->li_request_list);
3089         mutex_init(&eli->li_list_mtx);
3090
3091         eli->li_state |= EXT4_LAZYINIT_QUIT;
3092
3093         ext4_li_info = eli;
3094
3095         return 0;
3096 }
3097
3098 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3099                                             ext4_group_t start)
3100 {
3101         struct ext4_sb_info *sbi = EXT4_SB(sb);
3102         struct ext4_li_request *elr;
3103
3104         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3105         if (!elr)
3106                 return NULL;
3107
3108         elr->lr_super = sb;
3109         elr->lr_sbi = sbi;
3110         elr->lr_next_group = start;
3111
3112         /*
3113          * Randomize first schedule time of the request to
3114          * spread the inode table initialization requests
3115          * better.
3116          */
3117         elr->lr_next_sched = jiffies + (prandom_u32() %
3118                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3119         return elr;
3120 }
3121
3122 int ext4_register_li_request(struct super_block *sb,
3123                              ext4_group_t first_not_zeroed)
3124 {
3125         struct ext4_sb_info *sbi = EXT4_SB(sb);
3126         struct ext4_li_request *elr = NULL;
3127         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3128         int ret = 0;
3129
3130         mutex_lock(&ext4_li_mtx);
3131         if (sbi->s_li_request != NULL) {
3132                 /*
3133                  * Reset timeout so it can be computed again, because
3134                  * s_li_wait_mult might have changed.
3135                  */
3136                 sbi->s_li_request->lr_timeout = 0;
3137                 goto out;
3138         }
3139
3140         if (first_not_zeroed == ngroups ||
3141             (sb->s_flags & MS_RDONLY) ||
3142             !test_opt(sb, INIT_INODE_TABLE))
3143                 goto out;
3144
3145         elr = ext4_li_request_new(sb, first_not_zeroed);
3146         if (!elr) {
3147                 ret = -ENOMEM;
3148                 goto out;
3149         }
3150
3151         if (NULL == ext4_li_info) {
3152                 ret = ext4_li_info_new();
3153                 if (ret)
3154                         goto out;
3155         }
3156
3157         mutex_lock(&ext4_li_info->li_list_mtx);
3158         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3159         mutex_unlock(&ext4_li_info->li_list_mtx);
3160
3161         sbi->s_li_request = elr;
3162         /*
3163          * set elr to NULL here since it has been inserted to
3164          * the request_list and the removal and free of it is
3165          * handled by ext4_clear_request_list from now on.
3166          */
3167         elr = NULL;
3168
3169         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3170                 ret = ext4_run_lazyinit_thread();
3171                 if (ret)
3172                         goto out;
3173         }
3174 out:
3175         mutex_unlock(&ext4_li_mtx);
3176         if (ret)
3177                 kfree(elr);
3178         return ret;
3179 }
3180
3181 /*
3182  * We do not need to lock anything since this is called on
3183  * module unload.
3184  */
3185 static void ext4_destroy_lazyinit_thread(void)
3186 {
3187         /*
3188          * If thread exited earlier
3189          * there's nothing to be done.
3190          */
3191         if (!ext4_li_info || !ext4_lazyinit_task)
3192                 return;
3193
3194         kthread_stop(ext4_lazyinit_task);
3195 }
3196
3197 static int set_journal_csum_feature_set(struct super_block *sb)
3198 {
3199         int ret = 1;
3200         int compat, incompat;
3201         struct ext4_sb_info *sbi = EXT4_SB(sb);
3202
3203         if (ext4_has_metadata_csum(sb)) {
3204                 /* journal checksum v3 */
3205                 compat = 0;
3206                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3207         } else {
3208                 /* journal checksum v1 */
3209                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3210                 incompat = 0;
3211         }
3212
3213         jbd2_journal_clear_features(sbi->s_journal,
3214                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3215                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3216                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3217         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3218                 ret = jbd2_journal_set_features(sbi->s_journal,
3219                                 compat, 0,
3220                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3221                                 incompat);
3222         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3223                 ret = jbd2_journal_set_features(sbi->s_journal,
3224                                 compat, 0,
3225                                 incompat);
3226                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3227                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3228         } else {
3229                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3230                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3231         }
3232
3233         return ret;
3234 }
3235
3236 /*
3237  * Note: calculating the overhead so we can be compatible with
3238  * historical BSD practice is quite difficult in the face of
3239  * clusters/bigalloc.  This is because multiple metadata blocks from
3240  * different block group can end up in the same allocation cluster.
3241  * Calculating the exact overhead in the face of clustered allocation
3242  * requires either O(all block bitmaps) in memory or O(number of block
3243  * groups**2) in time.  We will still calculate the superblock for
3244  * older file systems --- and if we come across with a bigalloc file
3245  * system with zero in s_overhead_clusters the estimate will be close to
3246  * correct especially for very large cluster sizes --- but for newer
3247  * file systems, it's better to calculate this figure once at mkfs
3248  * time, and store it in the superblock.  If the superblock value is
3249  * present (even for non-bigalloc file systems), we will use it.
3250  */
3251 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3252                           char *buf)
3253 {
3254         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3255         struct ext4_group_desc  *gdp;
3256         ext4_fsblk_t            first_block, last_block, b;
3257         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3258         int                     s, j, count = 0;
3259
3260         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3261                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3262                         sbi->s_itb_per_group + 2);
3263
3264         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3265                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3266         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3267         for (i = 0; i < ngroups; i++) {
3268                 gdp = ext4_get_group_desc(sb, i, NULL);
3269                 b = ext4_block_bitmap(sb, gdp);
3270                 if (b >= first_block && b <= last_block) {
3271                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3272                         count++;
3273                 }
3274                 b = ext4_inode_bitmap(sb, gdp);
3275                 if (b >= first_block && b <= last_block) {
3276                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3277                         count++;
3278                 }
3279                 b = ext4_inode_table(sb, gdp);
3280                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3281                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3282                                 int c = EXT4_B2C(sbi, b - first_block);
3283                                 ext4_set_bit(c, buf);
3284                                 count++;
3285                         }
3286                 if (i != grp)
3287                         continue;
3288                 s = 0;
3289                 if (ext4_bg_has_super(sb, grp)) {
3290                         ext4_set_bit(s++, buf);
3291                         count++;
3292                 }
3293                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3294                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3295                         count++;
3296                 }
3297         }
3298         if (!count)
3299                 return 0;
3300         return EXT4_CLUSTERS_PER_GROUP(sb) -
3301                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3302 }
3303
3304 /*
3305  * Compute the overhead and stash it in sbi->s_overhead
3306  */
3307 int ext4_calculate_overhead(struct super_block *sb)
3308 {
3309         struct ext4_sb_info *sbi = EXT4_SB(sb);
3310         struct ext4_super_block *es = sbi->s_es;
3311         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3312         ext4_fsblk_t overhead = 0;
3313         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3314
3315         if (!buf)
3316                 return -ENOMEM;
3317
3318         /*
3319          * Compute the overhead (FS structures).  This is constant
3320          * for a given filesystem unless the number of block groups
3321          * changes so we cache the previous value until it does.
3322          */
3323
3324         /*
3325          * All of the blocks before first_data_block are overhead
3326          */
3327         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3328
3329         /*
3330          * Add the overhead found in each block group
3331          */
3332         for (i = 0; i < ngroups; i++) {
3333                 int blks;
3334
3335                 blks = count_overhead(sb, i, buf);
3336                 overhead += blks;
3337                 if (blks)
3338                         memset(buf, 0, PAGE_SIZE);
3339                 cond_resched();
3340         }
3341         /* Add the internal journal blocks as well */
3342         if (sbi->s_journal && !sbi->journal_bdev)
3343                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3344
3345         sbi->s_overhead = overhead;
3346         smp_wmb();
3347         free_page((unsigned long) buf);
3348         return 0;
3349 }
3350
3351
3352 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3353 {
3354         ext4_fsblk_t resv_clusters;
3355
3356         /*
3357          * There's no need to reserve anything when we aren't using extents.
3358          * The space estimates are exact, there are no unwritten extents,
3359          * hole punching doesn't need new metadata... This is needed especially
3360          * to keep ext2/3 backward compatibility.
3361          */
3362         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3363                 return 0;
3364         /*
3365          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3366          * This should cover the situations where we can not afford to run
3367          * out of space like for example punch hole, or converting
3368          * unwritten extents in delalloc path. In most cases such
3369          * allocation would require 1, or 2 blocks, higher numbers are
3370          * very rare.
3371          */
3372         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3373                         EXT4_SB(sb)->s_cluster_bits;
3374
3375         do_div(resv_clusters, 50);
3376         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3377
3378         return resv_clusters;
3379 }
3380
3381
3382 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3383 {
3384         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3385                                 sbi->s_cluster_bits;
3386
3387         if (count >= clusters)
3388                 return -EINVAL;
3389
3390         atomic64_set(&sbi->s_resv_clusters, count);
3391         return 0;
3392 }
3393
3394 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3395 {
3396         char *orig_data = kstrdup(data, GFP_KERNEL);
3397         struct buffer_head *bh;
3398         struct ext4_super_block *es = NULL;
3399         struct ext4_sb_info *sbi;
3400         ext4_fsblk_t block;
3401         ext4_fsblk_t sb_block = get_sb_block(&data);
3402         ext4_fsblk_t logical_sb_block;
3403         unsigned long offset = 0;
3404         unsigned long journal_devnum = 0;
3405         unsigned long def_mount_opts;
3406         struct inode *root;
3407         char *cp;
3408         const char *descr;
3409         int ret = -ENOMEM;
3410         int blocksize, clustersize;
3411         unsigned int db_count;
3412         unsigned int i;
3413         int needs_recovery, has_huge_files, has_bigalloc;
3414         __u64 blocks_count;
3415         int err = 0;
3416         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3417         ext4_group_t first_not_zeroed;
3418
3419         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3420         if (!sbi)
3421                 goto out_free_orig;
3422
3423         sbi->s_blockgroup_lock =
3424                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3425         if (!sbi->s_blockgroup_lock) {
3426                 kfree(sbi);
3427                 goto out_free_orig;
3428         }
3429         sb->s_fs_info = sbi;
3430         sbi->s_sb = sb;
3431         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3432         sbi->s_sb_block = sb_block;
3433         if (sb->s_bdev->bd_part)
3434                 sbi->s_sectors_written_start =
3435                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3436
3437         /* Cleanup superblock name */
3438         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3439                 *cp = '!';
3440
3441         /* -EINVAL is default */
3442         ret = -EINVAL;
3443         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3444         if (!blocksize) {
3445                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3446                 goto out_fail;
3447         }
3448
3449         /*
3450          * The ext4 superblock will not be buffer aligned for other than 1kB
3451          * block sizes.  We need to calculate the offset from buffer start.
3452          */
3453         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3454                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3455                 offset = do_div(logical_sb_block, blocksize);
3456         } else {
3457                 logical_sb_block = sb_block;
3458         }
3459
3460         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3461                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3462                 goto out_fail;
3463         }
3464         /*
3465          * Note: s_es must be initialized as soon as possible because
3466          *       some ext4 macro-instructions depend on its value
3467          */
3468         es = (struct ext4_super_block *) (bh->b_data + offset);
3469         sbi->s_es = es;
3470         sb->s_magic = le16_to_cpu(es->s_magic);
3471         if (sb->s_magic != EXT4_SUPER_MAGIC)
3472                 goto cantfind_ext4;
3473         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3474
3475         /* Warn if metadata_csum and gdt_csum are both set. */
3476         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3477                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3478             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3479                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3480                              "redundant flags; please run fsck.");
3481
3482         /* Check for a known checksum algorithm */
3483         if (!ext4_verify_csum_type(sb, es)) {
3484                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3485                          "unknown checksum algorithm.");
3486                 silent = 1;
3487                 goto cantfind_ext4;
3488         }
3489
3490         /* Load the checksum driver */
3491         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3492                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3493                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3494                 if (IS_ERR(sbi->s_chksum_driver)) {
3495                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3496                         ret = PTR_ERR(sbi->s_chksum_driver);
3497                         sbi->s_chksum_driver = NULL;
3498                         goto failed_mount;
3499                 }
3500         }
3501
3502         /* Check superblock checksum */
3503         if (!ext4_superblock_csum_verify(sb, es)) {
3504                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3505                          "invalid superblock checksum.  Run e2fsck?");
3506                 silent = 1;
3507                 goto cantfind_ext4;
3508         }
3509
3510         /* Precompute checksum seed for all metadata */
3511         if (ext4_has_metadata_csum(sb))
3512                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3513                                                sizeof(es->s_uuid));
3514
3515         /* Set defaults before we parse the mount options */
3516         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3517         set_opt(sb, INIT_INODE_TABLE);
3518         if (def_mount_opts & EXT4_DEFM_DEBUG)
3519                 set_opt(sb, DEBUG);
3520         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3521                 set_opt(sb, GRPID);
3522         if (def_mount_opts & EXT4_DEFM_UID16)
3523                 set_opt(sb, NO_UID32);
3524         /* xattr user namespace & acls are now defaulted on */
3525         set_opt(sb, XATTR_USER);
3526 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3527         set_opt(sb, POSIX_ACL);
3528 #endif
3529         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3530         if (ext4_has_metadata_csum(sb))
3531                 set_opt(sb, JOURNAL_CHECKSUM);
3532
3533         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3534                 set_opt(sb, JOURNAL_DATA);
3535         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3536                 set_opt(sb, ORDERED_DATA);
3537         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3538                 set_opt(sb, WRITEBACK_DATA);
3539
3540         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3541                 set_opt(sb, ERRORS_PANIC);
3542         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3543                 set_opt(sb, ERRORS_CONT);
3544         else
3545                 set_opt(sb, ERRORS_RO);
3546         /* block_validity enabled by default; disable with noblock_validity */
3547         set_opt(sb, BLOCK_VALIDITY);
3548         if (def_mount_opts & EXT4_DEFM_DISCARD)
3549                 set_opt(sb, DISCARD);
3550
3551         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3552         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3553         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3554         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3555         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3556
3557         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3558                 set_opt(sb, BARRIER);
3559
3560         /*
3561          * enable delayed allocation by default
3562          * Use -o nodelalloc to turn it off
3563          */
3564         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3565             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3566                 set_opt(sb, DELALLOC);
3567
3568         /*
3569          * set default s_li_wait_mult for lazyinit, for the case there is
3570          * no mount option specified.
3571          */
3572         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3573
3574         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3575                            &journal_devnum, &journal_ioprio, 0)) {
3576                 ext4_msg(sb, KERN_WARNING,
3577                          "failed to parse options in superblock: %s",
3578                          sbi->s_es->s_mount_opts);
3579         }
3580         sbi->s_def_mount_opt = sbi->s_mount_opt;
3581         if (!parse_options((char *) data, sb, &journal_devnum,
3582                            &journal_ioprio, 0))
3583                 goto failed_mount;
3584
3585         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3586                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3587                             "with data=journal disables delayed "
3588                             "allocation and O_DIRECT support!\n");
3589                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3590                         ext4_msg(sb, KERN_ERR, "can't mount with "
3591                                  "both data=journal and delalloc");
3592                         goto failed_mount;
3593                 }
3594                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3595                         ext4_msg(sb, KERN_ERR, "can't mount with "
3596                                  "both data=journal and dioread_nolock");
3597                         goto failed_mount;
3598                 }
3599                 if (test_opt(sb, DAX)) {
3600                         ext4_msg(sb, KERN_ERR, "can't mount with "
3601                                  "both data=journal and dax");
3602                         goto failed_mount;
3603                 }
3604                 if (test_opt(sb, DELALLOC))
3605                         clear_opt(sb, DELALLOC);
3606         }
3607
3608         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3609                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3610
3611         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3612             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3613              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3614              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3615                 ext4_msg(sb, KERN_WARNING,
3616                        "feature flags set on rev 0 fs, "
3617                        "running e2fsck is recommended");
3618
3619         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3620                 set_opt2(sb, HURD_COMPAT);
3621                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3622                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3623                         ext4_msg(sb, KERN_ERR,
3624                                  "The Hurd can't support 64-bit file systems");
3625                         goto failed_mount;
3626                 }
3627         }
3628
3629         if (IS_EXT2_SB(sb)) {
3630                 if (ext2_feature_set_ok(sb))
3631                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3632                                  "using the ext4 subsystem");
3633                 else {
3634                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3635                                  "to feature incompatibilities");
3636                         goto failed_mount;
3637                 }
3638         }
3639
3640         if (IS_EXT3_SB(sb)) {
3641                 if (ext3_feature_set_ok(sb))
3642                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3643                                  "using the ext4 subsystem");
3644                 else {
3645                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3646                                  "to feature incompatibilities");
3647                         goto failed_mount;
3648                 }
3649         }
3650
3651         /*
3652          * Check feature flags regardless of the revision level, since we
3653          * previously didn't change the revision level when setting the flags,
3654          * so there is a chance incompat flags are set on a rev 0 filesystem.
3655          */
3656         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3657                 goto failed_mount;
3658
3659         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3660         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3661             blocksize > EXT4_MAX_BLOCK_SIZE) {
3662                 ext4_msg(sb, KERN_ERR,
3663                        "Unsupported filesystem blocksize %d", blocksize);
3664                 goto failed_mount;
3665         }
3666
3667         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3668                 if (blocksize != PAGE_SIZE) {
3669                         ext4_msg(sb, KERN_ERR,
3670                                         "error: unsupported blocksize for dax");
3671                         goto failed_mount;
3672                 }
3673                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3674                         ext4_msg(sb, KERN_ERR,
3675                                         "error: device does not support dax");
3676                         goto failed_mount;
3677                 }
3678         }
3679
3680         if (sb->s_blocksize != blocksize) {
3681                 /* Validate the filesystem blocksize */
3682                 if (!sb_set_blocksize(sb, blocksize)) {
3683                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3684                                         blocksize);
3685                         goto failed_mount;
3686                 }
3687
3688                 brelse(bh);
3689                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3690                 offset = do_div(logical_sb_block, blocksize);
3691                 bh = sb_bread_unmovable(sb, logical_sb_block);
3692                 if (!bh) {
3693                         ext4_msg(sb, KERN_ERR,
3694                                "Can't read superblock on 2nd try");
3695                         goto failed_mount;
3696                 }
3697                 es = (struct ext4_super_block *)(bh->b_data + offset);
3698                 sbi->s_es = es;
3699                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3700                         ext4_msg(sb, KERN_ERR,
3701                                "Magic mismatch, very weird!");
3702                         goto failed_mount;
3703                 }
3704         }
3705
3706         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3707                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3708         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3709                                                       has_huge_files);
3710         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3711
3712         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3713                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3714                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3715         } else {
3716                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3717                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3718                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3719                     (!is_power_of_2(sbi->s_inode_size)) ||
3720                     (sbi->s_inode_size > blocksize)) {
3721                         ext4_msg(sb, KERN_ERR,
3722                                "unsupported inode size: %d",
3723                                sbi->s_inode_size);
3724                         goto failed_mount;
3725                 }
3726                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3727                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3728         }
3729
3730         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3731         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3732                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3733                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3734                     !is_power_of_2(sbi->s_desc_size)) {
3735                         ext4_msg(sb, KERN_ERR,
3736                                "unsupported descriptor size %lu",
3737                                sbi->s_desc_size);
3738                         goto failed_mount;
3739                 }
3740         } else
3741                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3742
3743         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3744         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3745         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3746                 goto cantfind_ext4;
3747
3748         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3749         if (sbi->s_inodes_per_block == 0)
3750                 goto cantfind_ext4;
3751         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3752                                         sbi->s_inodes_per_block;
3753         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3754         sbi->s_sbh = bh;
3755         sbi->s_mount_state = le16_to_cpu(es->s_state);
3756         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3757         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3758
3759         for (i = 0; i < 4; i++)
3760                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3761         sbi->s_def_hash_version = es->s_def_hash_version;
3762         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3763                 i = le32_to_cpu(es->s_flags);
3764                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3765                         sbi->s_hash_unsigned = 3;
3766                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3767 #ifdef __CHAR_UNSIGNED__
3768                         if (!(sb->s_flags & MS_RDONLY))
3769                                 es->s_flags |=
3770                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3771                         sbi->s_hash_unsigned = 3;
3772 #else
3773                         if (!(sb->s_flags & MS_RDONLY))
3774                                 es->s_flags |=
3775                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3776 #endif
3777                 }
3778         }
3779
3780         /* Handle clustersize */
3781         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3782         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3783                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3784         if (has_bigalloc) {
3785                 if (clustersize < blocksize) {
3786                         ext4_msg(sb, KERN_ERR,
3787                                  "cluster size (%d) smaller than "
3788                                  "block size (%d)", clustersize, blocksize);
3789                         goto failed_mount;
3790                 }
3791                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3792                         le32_to_cpu(es->s_log_block_size);
3793                 sbi->s_clusters_per_group =
3794                         le32_to_cpu(es->s_clusters_per_group);
3795                 if (sbi->s_clusters_per_group > blocksize * 8) {
3796                         ext4_msg(sb, KERN_ERR,
3797                                  "#clusters per group too big: %lu",
3798                                  sbi->s_clusters_per_group);
3799                         goto failed_mount;
3800                 }
3801                 if (sbi->s_blocks_per_group !=
3802                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3803                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3804                                  "clusters per group (%lu) inconsistent",
3805                                  sbi->s_blocks_per_group,
3806                                  sbi->s_clusters_per_group);
3807                         goto failed_mount;
3808                 }
3809         } else {
3810                 if (clustersize != blocksize) {
3811                         ext4_warning(sb, "fragment/cluster size (%d) != "
3812                                      "block size (%d)", clustersize,
3813                                      blocksize);
3814                         clustersize = blocksize;
3815                 }
3816                 if (sbi->s_blocks_per_group > blocksize * 8) {
3817                         ext4_msg(sb, KERN_ERR,
3818                                  "#blocks per group too big: %lu",
3819                                  sbi->s_blocks_per_group);
3820                         goto failed_mount;
3821                 }
3822                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3823                 sbi->s_cluster_bits = 0;
3824         }
3825         sbi->s_cluster_ratio = clustersize / blocksize;
3826
3827         if (sbi->s_inodes_per_group > blocksize * 8) {
3828                 ext4_msg(sb, KERN_ERR,
3829                        "#inodes per group too big: %lu",
3830                        sbi->s_inodes_per_group);
3831                 goto failed_mount;
3832         }
3833
3834         /* Do we have standard group size of clustersize * 8 blocks ? */
3835         if (sbi->s_blocks_per_group == clustersize << 3)
3836                 set_opt2(sb, STD_GROUP_SIZE);
3837
3838         /*
3839          * Test whether we have more sectors than will fit in sector_t,
3840          * and whether the max offset is addressable by the page cache.
3841          */
3842         err = generic_check_addressable(sb->s_blocksize_bits,
3843                                         ext4_blocks_count(es));
3844         if (err) {
3845                 ext4_msg(sb, KERN_ERR, "filesystem"
3846                          " too large to mount safely on this system");
3847                 if (sizeof(sector_t) < 8)
3848                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3849                 goto failed_mount;
3850         }
3851
3852         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3853                 goto cantfind_ext4;
3854
3855         /* check blocks count against device size */
3856         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3857         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3858                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3859                        "exceeds size of device (%llu blocks)",
3860                        ext4_blocks_count(es), blocks_count);
3861                 goto failed_mount;
3862         }
3863
3864         /*
3865          * It makes no sense for the first data block to be beyond the end
3866          * of the filesystem.
3867          */
3868         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3869                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3870                          "block %u is beyond end of filesystem (%llu)",
3871                          le32_to_cpu(es->s_first_data_block),
3872                          ext4_blocks_count(es));
3873                 goto failed_mount;
3874         }
3875         blocks_count = (ext4_blocks_count(es) -
3876                         le32_to_cpu(es->s_first_data_block) +
3877                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3878         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3879         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3880                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3881                        "(block count %llu, first data block %u, "
3882                        "blocks per group %lu)", sbi->s_groups_count,
3883                        ext4_blocks_count(es),
3884                        le32_to_cpu(es->s_first_data_block),
3885                        EXT4_BLOCKS_PER_GROUP(sb));
3886                 goto failed_mount;
3887         }
3888         sbi->s_groups_count = blocks_count;
3889         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3890                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3891         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3892                    EXT4_DESC_PER_BLOCK(sb);
3893         sbi->s_group_desc = ext4_kvmalloc(db_count *
3894                                           sizeof(struct buffer_head *),
3895                                           GFP_KERNEL);
3896         if (sbi->s_group_desc == NULL) {
3897                 ext4_msg(sb, KERN_ERR, "not enough memory");
3898                 ret = -ENOMEM;
3899                 goto failed_mount;
3900         }
3901
3902         if (ext4_proc_root)
3903                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3904
3905         if (sbi->s_proc)
3906                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3907                                  &ext4_seq_options_fops, sb);
3908
3909         bgl_lock_init(sbi->s_blockgroup_lock);
3910
3911         for (i = 0; i < db_count; i++) {
3912                 block = descriptor_loc(sb, logical_sb_block, i);
3913                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3914                 if (!sbi->s_group_desc[i]) {
3915                         ext4_msg(sb, KERN_ERR,
3916                                "can't read group descriptor %d", i);
3917                         db_count = i;
3918                         goto failed_mount2;
3919                 }
3920         }
3921         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3922                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3923                 goto failed_mount2;
3924         }
3925
3926         sbi->s_gdb_count = db_count;
3927         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3928         spin_lock_init(&sbi->s_next_gen_lock);
3929
3930         init_timer(&sbi->s_err_report);
3931         sbi->s_err_report.function = print_daily_error_info;
3932         sbi->s_err_report.data = (unsigned long) sb;
3933
3934         /* Register extent status tree shrinker */
3935         if (ext4_es_register_shrinker(sbi))
3936                 goto failed_mount3;
3937
3938         sbi->s_stripe = ext4_get_stripe_size(sbi);
3939         sbi->s_extent_max_zeroout_kb = 32;
3940
3941         /*
3942          * set up enough so that it can read an inode
3943          */
3944         sb->s_op = &ext4_sops;
3945         sb->s_export_op = &ext4_export_ops;
3946         sb->s_xattr = ext4_xattr_handlers;
3947 #ifdef CONFIG_QUOTA
3948         sb->dq_op = &ext4_quota_operations;
3949         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3950                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3951         else
3952                 sb->s_qcop = &ext4_qctl_operations;
3953         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3954 #endif
3955         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3956
3957         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3958         mutex_init(&sbi->s_orphan_lock);
3959
3960         sb->s_root = NULL;
3961
3962         needs_recovery = (es->s_last_orphan != 0 ||
3963                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3964                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3965
3966         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3967             !(sb->s_flags & MS_RDONLY))
3968                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3969                         goto failed_mount3a;
3970
3971         /*
3972          * The first inode we look at is the journal inode.  Don't try
3973          * root first: it may be modified in the journal!
3974          */
3975         if (!test_opt(sb, NOLOAD) &&
3976             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3977                 if (ext4_load_journal(sb, es, journal_devnum))
3978                         goto failed_mount3a;
3979         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3980               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3981                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3982                        "suppressed and not mounted read-only");
3983                 goto failed_mount_wq;
3984         } else {
3985                 clear_opt(sb, DATA_FLAGS);
3986                 sbi->s_journal = NULL;
3987                 needs_recovery = 0;
3988                 goto no_journal;
3989         }
3990
3991         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3992             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3993                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3994                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3995                 goto failed_mount_wq;
3996         }
3997
3998         if (!set_journal_csum_feature_set(sb)) {
3999                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4000                          "feature set");
4001                 goto failed_mount_wq;
4002         }
4003
4004         /* We have now updated the journal if required, so we can
4005          * validate the data journaling mode. */
4006         switch (test_opt(sb, DATA_FLAGS)) {
4007         case 0:
4008                 /* No mode set, assume a default based on the journal
4009                  * capabilities: ORDERED_DATA if the journal can
4010                  * cope, else JOURNAL_DATA
4011                  */
4012                 if (jbd2_journal_check_available_features
4013                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4014                         set_opt(sb, ORDERED_DATA);
4015                 else
4016                         set_opt(sb, JOURNAL_DATA);
4017                 break;
4018
4019         case EXT4_MOUNT_ORDERED_DATA:
4020         case EXT4_MOUNT_WRITEBACK_DATA:
4021                 if (!jbd2_journal_check_available_features
4022                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4023                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4024                                "requested data journaling mode");
4025                         goto failed_mount_wq;
4026                 }
4027         default:
4028                 break;
4029         }
4030         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4031
4032         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4033
4034 no_journal:
4035         if (ext4_mballoc_ready) {
4036                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4037                 if (!sbi->s_mb_cache) {
4038                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4039                         goto failed_mount_wq;
4040                 }
4041         }
4042
4043         /*
4044          * Get the # of file system overhead blocks from the
4045          * superblock if present.
4046          */
4047         if (es->s_overhead_clusters)
4048                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4049         else {
4050                 err = ext4_calculate_overhead(sb);
4051                 if (err)
4052                         goto failed_mount_wq;
4053         }
4054
4055         /*
4056          * The maximum number of concurrent works can be high and
4057          * concurrency isn't really necessary.  Limit it to 1.
4058          */
4059         EXT4_SB(sb)->rsv_conversion_wq =
4060                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4061         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4062                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4063                 ret = -ENOMEM;
4064                 goto failed_mount4;
4065         }
4066
4067         /*
4068          * The jbd2_journal_load will have done any necessary log recovery,
4069          * so we can safely mount the rest of the filesystem now.
4070          */
4071
4072         root = ext4_iget(sb, EXT4_ROOT_INO);
4073         if (IS_ERR(root)) {
4074                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4075                 ret = PTR_ERR(root);
4076                 root = NULL;
4077                 goto failed_mount4;
4078         }
4079         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4080                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4081                 iput(root);
4082                 goto failed_mount4;
4083         }
4084         sb->s_root = d_make_root(root);
4085         if (!sb->s_root) {
4086                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4087                 ret = -ENOMEM;
4088                 goto failed_mount4;
4089         }
4090
4091         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4092                 sb->s_flags |= MS_RDONLY;
4093
4094         /* determine the minimum size of new large inodes, if present */
4095         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4096                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4097                                                      EXT4_GOOD_OLD_INODE_SIZE;
4098                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4099                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4100                         if (sbi->s_want_extra_isize <
4101                             le16_to_cpu(es->s_want_extra_isize))
4102                                 sbi->s_want_extra_isize =
4103                                         le16_to_cpu(es->s_want_extra_isize);
4104                         if (sbi->s_want_extra_isize <
4105                             le16_to_cpu(es->s_min_extra_isize))
4106                                 sbi->s_want_extra_isize =
4107                                         le16_to_cpu(es->s_min_extra_isize);
4108                 }
4109         }
4110         /* Check if enough inode space is available */
4111         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4112                                                         sbi->s_inode_size) {
4113                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4114                                                        EXT4_GOOD_OLD_INODE_SIZE;
4115                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4116                          "available");
4117         }
4118
4119         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4120         if (err) {
4121                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4122                          "reserved pool", ext4_calculate_resv_clusters(sb));
4123                 goto failed_mount4a;
4124         }
4125
4126         err = ext4_setup_system_zone(sb);
4127         if (err) {
4128                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4129                          "zone (%d)", err);
4130                 goto failed_mount4a;
4131         }
4132
4133         ext4_ext_init(sb);
4134         err = ext4_mb_init(sb);
4135         if (err) {
4136                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4137                          err);
4138                 goto failed_mount5;
4139         }
4140
4141         block = ext4_count_free_clusters(sb);
4142         ext4_free_blocks_count_set(sbi->s_es,
4143                                    EXT4_C2B(sbi, block));
4144         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4145                                   GFP_KERNEL);
4146         if (!err) {
4147                 unsigned long freei = ext4_count_free_inodes(sb);
4148                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4149                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4150                                           GFP_KERNEL);
4151         }
4152         if (!err)
4153                 err = percpu_counter_init(&sbi->s_dirs_counter,
4154                                           ext4_count_dirs(sb), GFP_KERNEL);
4155         if (!err)
4156                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4157                                           GFP_KERNEL);
4158         if (err) {
4159                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4160                 goto failed_mount6;
4161         }
4162
4163         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4164                 if (!ext4_fill_flex_info(sb)) {
4165                         ext4_msg(sb, KERN_ERR,
4166                                "unable to initialize "
4167                                "flex_bg meta info!");
4168                         goto failed_mount6;
4169                 }
4170
4171         err = ext4_register_li_request(sb, first_not_zeroed);
4172         if (err)
4173                 goto failed_mount6;
4174
4175         sbi->s_kobj.kset = ext4_kset;
4176         init_completion(&sbi->s_kobj_unregister);
4177         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4178                                    "%s", sb->s_id);
4179         if (err)
4180                 goto failed_mount7;
4181
4182 #ifdef CONFIG_QUOTA
4183         /* Enable quota usage during mount. */
4184         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4185             !(sb->s_flags & MS_RDONLY)) {
4186                 err = ext4_enable_quotas(sb);
4187                 if (err)
4188                         goto failed_mount8;
4189         }
4190 #endif  /* CONFIG_QUOTA */
4191
4192         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4193         ext4_orphan_cleanup(sb, es);
4194         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4195         if (needs_recovery) {
4196                 ext4_msg(sb, KERN_INFO, "recovery complete");
4197                 ext4_mark_recovery_complete(sb, es);
4198         }
4199         if (EXT4_SB(sb)->s_journal) {
4200                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4201                         descr = " journalled data mode";
4202                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4203                         descr = " ordered data mode";
4204                 else
4205                         descr = " writeback data mode";
4206         } else
4207                 descr = "out journal";
4208
4209         if (test_opt(sb, DISCARD)) {
4210                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4211                 if (!blk_queue_discard(q))
4212                         ext4_msg(sb, KERN_WARNING,
4213                                  "mounting with \"discard\" option, but "
4214                                  "the device does not support discard");
4215         }
4216
4217         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4218                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4219                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4220
4221         if (es->s_error_count)
4222                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4223
4224         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4225         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4226         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4227         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4228
4229         kfree(orig_data);
4230         return 0;
4231
4232 cantfind_ext4:
4233         if (!silent)
4234                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4235         goto failed_mount;
4236
4237 #ifdef CONFIG_QUOTA
4238 failed_mount8:
4239         kobject_del(&sbi->s_kobj);
4240 #endif
4241 failed_mount7:
4242         ext4_unregister_li_request(sb);
4243 failed_mount6:
4244         ext4_mb_release(sb);
4245         if (sbi->s_flex_groups)
4246                 kvfree(sbi->s_flex_groups);
4247         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4248         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4249         percpu_counter_destroy(&sbi->s_dirs_counter);
4250         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4251 failed_mount5:
4252         ext4_ext_release(sb);
4253         ext4_release_system_zone(sb);
4254 failed_mount4a:
4255         dput(sb->s_root);
4256         sb->s_root = NULL;
4257 failed_mount4:
4258         ext4_msg(sb, KERN_ERR, "mount failed");
4259         if (EXT4_SB(sb)->rsv_conversion_wq)
4260                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4261 failed_mount_wq:
4262         if (sbi->s_journal) {
4263                 jbd2_journal_destroy(sbi->s_journal);
4264                 sbi->s_journal = NULL;
4265         }
4266 failed_mount3a:
4267         ext4_es_unregister_shrinker(sbi);
4268 failed_mount3:
4269         del_timer_sync(&sbi->s_err_report);
4270         if (sbi->s_mmp_tsk)
4271                 kthread_stop(sbi->s_mmp_tsk);
4272 failed_mount2:
4273         for (i = 0; i < db_count; i++)
4274                 brelse(sbi->s_group_desc[i]);
4275         kvfree(sbi->s_group_desc);
4276 failed_mount:
4277         if (sbi->s_chksum_driver)
4278                 crypto_free_shash(sbi->s_chksum_driver);
4279         if (sbi->s_proc) {
4280                 remove_proc_entry("options", sbi->s_proc);
4281                 remove_proc_entry(sb->s_id, ext4_proc_root);
4282         }
4283 #ifdef CONFIG_QUOTA
4284         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4285                 kfree(sbi->s_qf_names[i]);
4286 #endif
4287         ext4_blkdev_remove(sbi);
4288         brelse(bh);
4289 out_fail:
4290         sb->s_fs_info = NULL;
4291         kfree(sbi->s_blockgroup_lock);
4292         kfree(sbi);
4293 out_free_orig:
4294         kfree(orig_data);
4295         return err ? err : ret;
4296 }
4297
4298 /*
4299  * Setup any per-fs journal parameters now.  We'll do this both on
4300  * initial mount, once the journal has been initialised but before we've
4301  * done any recovery; and again on any subsequent remount.
4302  */
4303 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4304 {
4305         struct ext4_sb_info *sbi = EXT4_SB(sb);
4306
4307         journal->j_commit_interval = sbi->s_commit_interval;
4308         journal->j_min_batch_time = sbi->s_min_batch_time;
4309         journal->j_max_batch_time = sbi->s_max_batch_time;
4310
4311         write_lock(&journal->j_state_lock);
4312         if (test_opt(sb, BARRIER))
4313                 journal->j_flags |= JBD2_BARRIER;
4314         else
4315                 journal->j_flags &= ~JBD2_BARRIER;
4316         if (test_opt(sb, DATA_ERR_ABORT))
4317                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4318         else
4319                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4320         write_unlock(&journal->j_state_lock);
4321 }
4322
4323 static journal_t *ext4_get_journal(struct super_block *sb,
4324                                    unsigned int journal_inum)
4325 {
4326         struct inode *journal_inode;
4327         journal_t *journal;
4328
4329         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4330
4331         /* First, test for the existence of a valid inode on disk.  Bad
4332          * things happen if we iget() an unused inode, as the subsequent
4333          * iput() will try to delete it. */
4334
4335         journal_inode = ext4_iget(sb, journal_inum);
4336         if (IS_ERR(journal_inode)) {
4337                 ext4_msg(sb, KERN_ERR, "no journal found");
4338                 return NULL;
4339         }
4340         if (!journal_inode->i_nlink) {
4341                 make_bad_inode(journal_inode);
4342                 iput(journal_inode);
4343                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4344                 return NULL;
4345         }
4346
4347         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4348                   journal_inode, journal_inode->i_size);
4349         if (!S_ISREG(journal_inode->i_mode)) {
4350                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4351                 iput(journal_inode);
4352                 return NULL;
4353         }
4354
4355         journal = jbd2_journal_init_inode(journal_inode);
4356         if (!journal) {
4357                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4358                 iput(journal_inode);
4359                 return NULL;
4360         }
4361         journal->j_private = sb;
4362         ext4_init_journal_params(sb, journal);
4363         return journal;
4364 }
4365
4366 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4367                                        dev_t j_dev)
4368 {
4369         struct buffer_head *bh;
4370         journal_t *journal;
4371         ext4_fsblk_t start;
4372         ext4_fsblk_t len;
4373         int hblock, blocksize;
4374         ext4_fsblk_t sb_block;
4375         unsigned long offset;
4376         struct ext4_super_block *es;
4377         struct block_device *bdev;
4378
4379         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4380
4381         bdev = ext4_blkdev_get(j_dev, sb);
4382         if (bdev == NULL)
4383                 return NULL;
4384
4385         blocksize = sb->s_blocksize;
4386         hblock = bdev_logical_block_size(bdev);
4387         if (blocksize < hblock) {
4388                 ext4_msg(sb, KERN_ERR,
4389                         "blocksize too small for journal device");
4390                 goto out_bdev;
4391         }
4392
4393         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4394         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4395         set_blocksize(bdev, blocksize);
4396         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4397                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4398                        "external journal");
4399                 goto out_bdev;
4400         }
4401
4402         es = (struct ext4_super_block *) (bh->b_data + offset);
4403         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4404             !(le32_to_cpu(es->s_feature_incompat) &
4405               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4406                 ext4_msg(sb, KERN_ERR, "external journal has "
4407                                         "bad superblock");
4408                 brelse(bh);
4409                 goto out_bdev;
4410         }
4411
4412         if ((le32_to_cpu(es->s_feature_ro_compat) &
4413              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4414             es->s_checksum != ext4_superblock_csum(sb, es)) {
4415                 ext4_msg(sb, KERN_ERR, "external journal has "
4416                                        "corrupt superblock");
4417                 brelse(bh);
4418                 goto out_bdev;
4419         }
4420
4421         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4422                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4423                 brelse(bh);
4424                 goto out_bdev;
4425         }
4426
4427         len = ext4_blocks_count(es);
4428         start = sb_block + 1;
4429         brelse(bh);     /* we're done with the superblock */
4430
4431         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4432                                         start, len, blocksize);
4433         if (!journal) {
4434                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4435                 goto out_bdev;
4436         }
4437         journal->j_private = sb;
4438         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4439         wait_on_buffer(journal->j_sb_buffer);
4440         if (!buffer_uptodate(journal->j_sb_buffer)) {
4441                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4442                 goto out_journal;
4443         }
4444         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4445                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4446                                         "user (unsupported) - %d",
4447                         be32_to_cpu(journal->j_superblock->s_nr_users));
4448                 goto out_journal;
4449         }
4450         EXT4_SB(sb)->journal_bdev = bdev;
4451         ext4_init_journal_params(sb, journal);
4452         return journal;
4453
4454 out_journal:
4455         jbd2_journal_destroy(journal);
4456 out_bdev:
4457         ext4_blkdev_put(bdev);
4458         return NULL;
4459 }
4460
4461 static int ext4_load_journal(struct super_block *sb,
4462                              struct ext4_super_block *es,
4463                              unsigned long journal_devnum)
4464 {
4465         journal_t *journal;
4466         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4467         dev_t journal_dev;
4468         int err = 0;
4469         int really_read_only;
4470
4471         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4472
4473         if (journal_devnum &&
4474             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4475                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4476                         "numbers have changed");
4477                 journal_dev = new_decode_dev(journal_devnum);
4478         } else
4479                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4480
4481         really_read_only = bdev_read_only(sb->s_bdev);
4482
4483         /*
4484          * Are we loading a blank journal or performing recovery after a
4485          * crash?  For recovery, we need to check in advance whether we
4486          * can get read-write access to the device.
4487          */
4488         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4489                 if (sb->s_flags & MS_RDONLY) {
4490                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4491                                         "required on readonly filesystem");
4492                         if (really_read_only) {
4493                                 ext4_msg(sb, KERN_ERR, "write access "
4494                                         "unavailable, cannot proceed");
4495                                 return -EROFS;
4496                         }
4497                         ext4_msg(sb, KERN_INFO, "write access will "
4498                                "be enabled during recovery");
4499                 }
4500         }
4501
4502         if (journal_inum && journal_dev) {
4503                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4504                        "and inode journals!");
4505                 return -EINVAL;
4506         }
4507
4508         if (journal_inum) {
4509                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4510                         return -EINVAL;
4511         } else {
4512                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4513                         return -EINVAL;
4514         }
4515
4516         if (!(journal->j_flags & JBD2_BARRIER))
4517                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4518
4519         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4520                 err = jbd2_journal_wipe(journal, !really_read_only);
4521         if (!err) {
4522                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4523                 if (save)
4524                         memcpy(save, ((char *) es) +
4525                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4526                 err = jbd2_journal_load(journal);
4527                 if (save)
4528                         memcpy(((char *) es) + EXT4_S_ERR_START,
4529                                save, EXT4_S_ERR_LEN);
4530                 kfree(save);
4531         }
4532
4533         if (err) {
4534                 ext4_msg(sb, KERN_ERR, "error loading journal");
4535                 jbd2_journal_destroy(journal);
4536                 return err;
4537         }
4538
4539         EXT4_SB(sb)->s_journal = journal;
4540         ext4_clear_journal_err(sb, es);
4541
4542         if (!really_read_only && journal_devnum &&
4543             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4544                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4545
4546                 /* Make sure we flush the recovery flag to disk. */
4547                 ext4_commit_super(sb, 1);
4548         }
4549
4550         return 0;
4551 }
4552
4553 static int ext4_commit_super(struct super_block *sb, int sync)
4554 {
4555         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4556         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4557         int error = 0;
4558
4559         if (!sbh || block_device_ejected(sb))
4560                 return error;
4561         if (buffer_write_io_error(sbh)) {
4562                 /*
4563                  * Oh, dear.  A previous attempt to write the
4564                  * superblock failed.  This could happen because the
4565                  * USB device was yanked out.  Or it could happen to
4566                  * be a transient write error and maybe the block will
4567                  * be remapped.  Nothing we can do but to retry the
4568                  * write and hope for the best.
4569                  */
4570                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4571                        "superblock detected");
4572                 clear_buffer_write_io_error(sbh);
4573                 set_buffer_uptodate(sbh);
4574         }
4575         /*
4576          * If the file system is mounted read-only, don't update the
4577          * superblock write time.  This avoids updating the superblock
4578          * write time when we are mounting the root file system
4579          * read/only but we need to replay the journal; at that point,
4580          * for people who are east of GMT and who make their clock
4581          * tick in localtime for Windows bug-for-bug compatibility,
4582          * the clock is set in the future, and this will cause e2fsck
4583          * to complain and force a full file system check.
4584          */
4585         if (!(sb->s_flags & MS_RDONLY))
4586                 es->s_wtime = cpu_to_le32(get_seconds());
4587         if (sb->s_bdev->bd_part)
4588                 es->s_kbytes_written =
4589                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4590                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4591                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4592         else
4593                 es->s_kbytes_written =
4594                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4595         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4596                 ext4_free_blocks_count_set(es,
4597                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4598                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4599         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4600                 es->s_free_inodes_count =
4601                         cpu_to_le32(percpu_counter_sum_positive(
4602                                 &EXT4_SB(sb)->s_freeinodes_counter));
4603         BUFFER_TRACE(sbh, "marking dirty");
4604         ext4_superblock_csum_set(sb);
4605         mark_buffer_dirty(sbh);
4606         if (sync) {
4607                 error = sync_dirty_buffer(sbh);
4608                 if (error)
4609                         return error;
4610
4611                 error = buffer_write_io_error(sbh);
4612                 if (error) {
4613                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4614                                "superblock");
4615                         clear_buffer_write_io_error(sbh);
4616                         set_buffer_uptodate(sbh);
4617                 }
4618         }
4619         return error;
4620 }
4621
4622 /*
4623  * Have we just finished recovery?  If so, and if we are mounting (or
4624  * remounting) the filesystem readonly, then we will end up with a
4625  * consistent fs on disk.  Record that fact.
4626  */
4627 static void ext4_mark_recovery_complete(struct super_block *sb,
4628                                         struct ext4_super_block *es)
4629 {
4630         journal_t *journal = EXT4_SB(sb)->s_journal;
4631
4632         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4633                 BUG_ON(journal != NULL);
4634                 return;
4635         }
4636         jbd2_journal_lock_updates(journal);
4637         if (jbd2_journal_flush(journal) < 0)
4638                 goto out;
4639
4640         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4641             sb->s_flags & MS_RDONLY) {
4642                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4643                 ext4_commit_super(sb, 1);
4644         }
4645
4646 out:
4647         jbd2_journal_unlock_updates(journal);
4648 }
4649
4650 /*
4651  * If we are mounting (or read-write remounting) a filesystem whose journal
4652  * has recorded an error from a previous lifetime, move that error to the
4653  * main filesystem now.
4654  */
4655 static void ext4_clear_journal_err(struct super_block *sb,
4656                                    struct ext4_super_block *es)
4657 {
4658         journal_t *journal;
4659         int j_errno;
4660         const char *errstr;
4661
4662         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4663
4664         journal = EXT4_SB(sb)->s_journal;
4665
4666         /*
4667          * Now check for any error status which may have been recorded in the
4668          * journal by a prior ext4_error() or ext4_abort()
4669          */
4670
4671         j_errno = jbd2_journal_errno(journal);
4672         if (j_errno) {
4673                 char nbuf[16];
4674
4675                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4676                 ext4_warning(sb, "Filesystem error recorded "
4677                              "from previous mount: %s", errstr);
4678                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4679
4680                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4681                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4682                 ext4_commit_super(sb, 1);
4683
4684                 jbd2_journal_clear_err(journal);
4685                 jbd2_journal_update_sb_errno(journal);
4686         }
4687 }
4688
4689 /*
4690  * Force the running and committing transactions to commit,
4691  * and wait on the commit.
4692  */
4693 int ext4_force_commit(struct super_block *sb)
4694 {
4695         journal_t *journal;
4696
4697         if (sb->s_flags & MS_RDONLY)
4698                 return 0;
4699
4700         journal = EXT4_SB(sb)->s_journal;
4701         return ext4_journal_force_commit(journal);
4702 }
4703
4704 static int ext4_sync_fs(struct super_block *sb, int wait)
4705 {
4706         int ret = 0;
4707         tid_t target;
4708         bool needs_barrier = false;
4709         struct ext4_sb_info *sbi = EXT4_SB(sb);
4710
4711         trace_ext4_sync_fs(sb, wait);
4712         flush_workqueue(sbi->rsv_conversion_wq);
4713         /*
4714          * Writeback quota in non-journalled quota case - journalled quota has
4715          * no dirty dquots
4716          */
4717         dquot_writeback_dquots(sb, -1);
4718         /*
4719          * Data writeback is possible w/o journal transaction, so barrier must
4720          * being sent at the end of the function. But we can skip it if
4721          * transaction_commit will do it for us.
4722          */
4723         if (sbi->s_journal) {
4724                 target = jbd2_get_latest_transaction(sbi->s_journal);
4725                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4726                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4727                         needs_barrier = true;
4728
4729                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4730                         if (wait)
4731                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4732                                                            target);
4733                 }
4734         } else if (wait && test_opt(sb, BARRIER))
4735                 needs_barrier = true;
4736         if (needs_barrier) {
4737                 int err;
4738                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4739                 if (!ret)
4740                         ret = err;
4741         }
4742
4743         return ret;
4744 }
4745
4746 /*
4747  * LVM calls this function before a (read-only) snapshot is created.  This
4748  * gives us a chance to flush the journal completely and mark the fs clean.
4749  *
4750  * Note that only this function cannot bring a filesystem to be in a clean
4751  * state independently. It relies on upper layer to stop all data & metadata
4752  * modifications.
4753  */
4754 static int ext4_freeze(struct super_block *sb)
4755 {
4756         int error = 0;
4757         journal_t *journal;
4758
4759         if (sb->s_flags & MS_RDONLY)
4760                 return 0;
4761
4762         journal = EXT4_SB(sb)->s_journal;
4763
4764         if (journal) {
4765                 /* Now we set up the journal barrier. */
4766                 jbd2_journal_lock_updates(journal);
4767
4768                 /*
4769                  * Don't clear the needs_recovery flag if we failed to
4770                  * flush the journal.
4771                  */
4772                 error = jbd2_journal_flush(journal);
4773                 if (error < 0)
4774                         goto out;
4775         }
4776
4777         /* Journal blocked and flushed, clear needs_recovery flag. */
4778         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4779         error = ext4_commit_super(sb, 1);
4780 out:
4781         if (journal)
4782                 /* we rely on upper layer to stop further updates */
4783                 jbd2_journal_unlock_updates(journal);
4784         return error;
4785 }
4786
4787 /*
4788  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4789  * flag here, even though the filesystem is not technically dirty yet.
4790  */
4791 static int ext4_unfreeze(struct super_block *sb)
4792 {
4793         if (sb->s_flags & MS_RDONLY)
4794                 return 0;
4795
4796         /* Reset the needs_recovery flag before the fs is unlocked. */
4797         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4798         ext4_commit_super(sb, 1);
4799         return 0;
4800 }
4801
4802 /*
4803  * Structure to save mount options for ext4_remount's benefit
4804  */
4805 struct ext4_mount_options {
4806         unsigned long s_mount_opt;
4807         unsigned long s_mount_opt2;
4808         kuid_t s_resuid;
4809         kgid_t s_resgid;
4810         unsigned long s_commit_interval;
4811         u32 s_min_batch_time, s_max_batch_time;
4812 #ifdef CONFIG_QUOTA
4813         int s_jquota_fmt;
4814         char *s_qf_names[EXT4_MAXQUOTAS];
4815 #endif
4816 };
4817
4818 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4819 {
4820         struct ext4_super_block *es;
4821         struct ext4_sb_info *sbi = EXT4_SB(sb);
4822         unsigned long old_sb_flags;
4823         struct ext4_mount_options old_opts;
4824         int enable_quota = 0;
4825         ext4_group_t g;
4826         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4827         int err = 0;
4828 #ifdef CONFIG_QUOTA
4829         int i, j;
4830 #endif
4831         char *orig_data = kstrdup(data, GFP_KERNEL);
4832
4833         /* Store the original options */
4834         old_sb_flags = sb->s_flags;
4835         old_opts.s_mount_opt = sbi->s_mount_opt;
4836         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4837         old_opts.s_resuid = sbi->s_resuid;
4838         old_opts.s_resgid = sbi->s_resgid;
4839         old_opts.s_commit_interval = sbi->s_commit_interval;
4840         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4841         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4842 #ifdef CONFIG_QUOTA
4843         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4844         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4845                 if (sbi->s_qf_names[i]) {
4846                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4847                                                          GFP_KERNEL);
4848                         if (!old_opts.s_qf_names[i]) {
4849                                 for (j = 0; j < i; j++)
4850                                         kfree(old_opts.s_qf_names[j]);
4851                                 kfree(orig_data);
4852                                 return -ENOMEM;
4853                         }
4854                 } else
4855                         old_opts.s_qf_names[i] = NULL;
4856 #endif
4857         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4858                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4859
4860         /*
4861          * Allow the "check" option to be passed as a remount option.
4862          */
4863         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4864                 err = -EINVAL;
4865                 goto restore_opts;
4866         }
4867
4868         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4869             test_opt(sb, JOURNAL_CHECKSUM)) {
4870                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4871                          "during remount not supported");
4872                 err = -EINVAL;
4873                 goto restore_opts;
4874         }
4875
4876         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4877             test_opt(sb, JOURNAL_CHECKSUM)) {
4878                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4879                          "during remount not supported");
4880                 err = -EINVAL;
4881                 goto restore_opts;
4882         }
4883
4884         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4885                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4886                         ext4_msg(sb, KERN_ERR, "can't mount with "
4887                                  "both data=journal and delalloc");
4888                         err = -EINVAL;
4889                         goto restore_opts;
4890                 }
4891                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4892                         ext4_msg(sb, KERN_ERR, "can't mount with "
4893                                  "both data=journal and dioread_nolock");
4894                         err = -EINVAL;
4895                         goto restore_opts;
4896                 }
4897                 if (test_opt(sb, DAX)) {
4898                         ext4_msg(sb, KERN_ERR, "can't mount with "
4899                                  "both data=journal and dax");
4900                         err = -EINVAL;
4901                         goto restore_opts;
4902                 }
4903         }
4904
4905         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4906                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4907                         "dax flag with busy inodes while remounting");
4908                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4909         }
4910
4911         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4912                 ext4_abort(sb, "Abort forced by user");
4913
4914         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4915                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4916
4917         es = sbi->s_es;
4918
4919         if (sbi->s_journal) {
4920                 ext4_init_journal_params(sb, sbi->s_journal);
4921                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4922         }
4923
4924         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4925                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4926                         err = -EROFS;
4927                         goto restore_opts;
4928                 }
4929
4930                 if (*flags & MS_RDONLY) {
4931                         err = sync_filesystem(sb);
4932                         if (err < 0)
4933                                 goto restore_opts;
4934                         err = dquot_suspend(sb, -1);
4935                         if (err < 0)
4936                                 goto restore_opts;
4937
4938                         /*
4939                          * First of all, the unconditional stuff we have to do
4940                          * to disable replay of the journal when we next remount
4941                          */
4942                         sb->s_flags |= MS_RDONLY;
4943
4944                         /*
4945                          * OK, test if we are remounting a valid rw partition
4946                          * readonly, and if so set the rdonly flag and then
4947                          * mark the partition as valid again.
4948                          */
4949                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4950                             (sbi->s_mount_state & EXT4_VALID_FS))
4951                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4952
4953                         if (sbi->s_journal)
4954                                 ext4_mark_recovery_complete(sb, es);
4955                 } else {
4956                         /* Make sure we can mount this feature set readwrite */
4957                         if (!ext4_feature_set_ok(sb, 0)) {
4958                                 err = -EROFS;
4959                                 goto restore_opts;
4960                         }
4961                         /*
4962                          * Make sure the group descriptor checksums
4963                          * are sane.  If they aren't, refuse to remount r/w.
4964                          */
4965                         for (g = 0; g < sbi->s_groups_count; g++) {
4966                                 struct ext4_group_desc *gdp =
4967                                         ext4_get_group_desc(sb, g, NULL);
4968
4969                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4970                                         ext4_msg(sb, KERN_ERR,
4971                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4972                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4973                                                le16_to_cpu(gdp->bg_checksum));
4974                                         err = -EINVAL;
4975                                         goto restore_opts;
4976                                 }
4977                         }
4978
4979                         /*
4980                          * If we have an unprocessed orphan list hanging
4981                          * around from a previously readonly bdev mount,
4982                          * require a full umount/remount for now.
4983                          */
4984                         if (es->s_last_orphan) {
4985                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4986                                        "remount RDWR because of unprocessed "
4987                                        "orphan inode list.  Please "
4988                                        "umount/remount instead");
4989                                 err = -EINVAL;
4990                                 goto restore_opts;
4991                         }
4992
4993                         /*
4994                          * Mounting a RDONLY partition read-write, so reread
4995                          * and store the current valid flag.  (It may have
4996                          * been changed by e2fsck since we originally mounted
4997                          * the partition.)
4998                          */
4999                         if (sbi->s_journal)
5000                                 ext4_clear_journal_err(sb, es);
5001                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5002                         if (!ext4_setup_super(sb, es, 0))
5003                                 sb->s_flags &= ~MS_RDONLY;
5004                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
5005                                                      EXT4_FEATURE_INCOMPAT_MMP))
5006                                 if (ext4_multi_mount_protect(sb,
5007                                                 le64_to_cpu(es->s_mmp_block))) {
5008                                         err = -EROFS;
5009                                         goto restore_opts;
5010                                 }
5011                         enable_quota = 1;
5012                 }
5013         }
5014
5015         /*
5016          * Reinitialize lazy itable initialization thread based on
5017          * current settings
5018          */
5019         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5020                 ext4_unregister_li_request(sb);
5021         else {
5022                 ext4_group_t first_not_zeroed;
5023                 first_not_zeroed = ext4_has_uninit_itable(sb);
5024                 ext4_register_li_request(sb, first_not_zeroed);
5025         }
5026
5027         ext4_setup_system_zone(sb);
5028         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5029                 ext4_commit_super(sb, 1);
5030
5031 #ifdef CONFIG_QUOTA
5032         /* Release old quota file names */
5033         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5034                 kfree(old_opts.s_qf_names[i]);
5035         if (enable_quota) {
5036                 if (sb_any_quota_suspended(sb))
5037                         dquot_resume(sb, -1);
5038                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5039                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5040                         err = ext4_enable_quotas(sb);
5041                         if (err)
5042                                 goto restore_opts;
5043                 }
5044         }
5045 #endif
5046
5047         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5048         kfree(orig_data);
5049         return 0;
5050
5051 restore_opts:
5052         sb->s_flags = old_sb_flags;
5053         sbi->s_mount_opt = old_opts.s_mount_opt;
5054         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5055         sbi->s_resuid = old_opts.s_resuid;
5056         sbi->s_resgid = old_opts.s_resgid;
5057         sbi->s_commit_interval = old_opts.s_commit_interval;
5058         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5059         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5060 #ifdef CONFIG_QUOTA
5061         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5062         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5063                 kfree(sbi->s_qf_names[i]);
5064                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5065         }
5066 #endif
5067         kfree(orig_data);
5068         return err;
5069 }
5070
5071 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5072 {
5073         struct super_block *sb = dentry->d_sb;
5074         struct ext4_sb_info *sbi = EXT4_SB(sb);
5075         struct ext4_super_block *es = sbi->s_es;
5076         ext4_fsblk_t overhead = 0, resv_blocks;
5077         u64 fsid;
5078         s64 bfree;
5079         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5080
5081         if (!test_opt(sb, MINIX_DF))
5082                 overhead = sbi->s_overhead;
5083
5084         buf->f_type = EXT4_SUPER_MAGIC;
5085         buf->f_bsize = sb->s_blocksize;
5086         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5087         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5088                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5089         /* prevent underflow in case that few free space is available */
5090         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5091         buf->f_bavail = buf->f_bfree -
5092                         (ext4_r_blocks_count(es) + resv_blocks);
5093         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5094                 buf->f_bavail = 0;
5095         buf->f_files = le32_to_cpu(es->s_inodes_count);
5096         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5097         buf->f_namelen = EXT4_NAME_LEN;
5098         fsid = le64_to_cpup((void *)es->s_uuid) ^
5099                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5100         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5101         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5102
5103         return 0;
5104 }
5105
5106 /* Helper function for writing quotas on sync - we need to start transaction
5107  * before quota file is locked for write. Otherwise the are possible deadlocks:
5108  * Process 1                         Process 2
5109  * ext4_create()                     quota_sync()
5110  *   jbd2_journal_start()                  write_dquot()
5111  *   dquot_initialize()                         down(dqio_mutex)
5112  *     down(dqio_mutex)                    jbd2_journal_start()
5113  *
5114  */
5115
5116 #ifdef CONFIG_QUOTA
5117
5118 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5119 {
5120         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5121 }
5122
5123 static int ext4_write_dquot(struct dquot *dquot)
5124 {
5125         int ret, err;
5126         handle_t *handle;
5127         struct inode *inode;
5128
5129         inode = dquot_to_inode(dquot);
5130         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5131                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5132         if (IS_ERR(handle))
5133                 return PTR_ERR(handle);
5134         ret = dquot_commit(dquot);
5135         err = ext4_journal_stop(handle);
5136         if (!ret)
5137                 ret = err;
5138         return ret;
5139 }
5140
5141 static int ext4_acquire_dquot(struct dquot *dquot)
5142 {
5143         int ret, err;
5144         handle_t *handle;
5145
5146         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5147                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5148         if (IS_ERR(handle))
5149                 return PTR_ERR(handle);
5150         ret = dquot_acquire(dquot);
5151         err = ext4_journal_stop(handle);
5152         if (!ret)
5153                 ret = err;
5154         return ret;
5155 }
5156
5157 static int ext4_release_dquot(struct dquot *dquot)
5158 {
5159         int ret, err;
5160         handle_t *handle;
5161
5162         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5163                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5164         if (IS_ERR(handle)) {
5165                 /* Release dquot anyway to avoid endless cycle in dqput() */
5166                 dquot_release(dquot);
5167                 return PTR_ERR(handle);
5168         }
5169         ret = dquot_release(dquot);
5170         err = ext4_journal_stop(handle);
5171         if (!ret)
5172                 ret = err;
5173         return ret;
5174 }
5175
5176 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5177 {
5178         struct super_block *sb = dquot->dq_sb;
5179         struct ext4_sb_info *sbi = EXT4_SB(sb);
5180
5181         /* Are we journaling quotas? */
5182         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5183             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5184                 dquot_mark_dquot_dirty(dquot);
5185                 return ext4_write_dquot(dquot);
5186         } else {
5187                 return dquot_mark_dquot_dirty(dquot);
5188         }
5189 }
5190
5191 static int ext4_write_info(struct super_block *sb, int type)
5192 {
5193         int ret, err;
5194         handle_t *handle;
5195
5196         /* Data block + inode block */
5197         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5198         if (IS_ERR(handle))
5199                 return PTR_ERR(handle);
5200         ret = dquot_commit_info(sb, type);
5201         err = ext4_journal_stop(handle);
5202         if (!ret)
5203                 ret = err;
5204         return ret;
5205 }
5206
5207 /*
5208  * Turn on quotas during mount time - we need to find
5209  * the quota file and such...
5210  */
5211 static int ext4_quota_on_mount(struct super_block *sb, int type)
5212 {
5213         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5214                                         EXT4_SB(sb)->s_jquota_fmt, type);
5215 }
5216
5217 /*
5218  * Standard function to be called on quota_on
5219  */
5220 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5221                          struct path *path)
5222 {
5223         int err;
5224
5225         if (!test_opt(sb, QUOTA))
5226                 return -EINVAL;
5227
5228         /* Quotafile not on the same filesystem? */
5229         if (path->dentry->d_sb != sb)
5230                 return -EXDEV;
5231         /* Journaling quota? */
5232         if (EXT4_SB(sb)->s_qf_names[type]) {
5233                 /* Quotafile not in fs root? */
5234                 if (path->dentry->d_parent != sb->s_root)
5235                         ext4_msg(sb, KERN_WARNING,
5236                                 "Quota file not on filesystem root. "
5237                                 "Journaled quota will not work");
5238         }
5239
5240         /*
5241          * When we journal data on quota file, we have to flush journal to see
5242          * all updates to the file when we bypass pagecache...
5243          */
5244         if (EXT4_SB(sb)->s_journal &&
5245             ext4_should_journal_data(path->dentry->d_inode)) {
5246                 /*
5247                  * We don't need to lock updates but journal_flush() could
5248                  * otherwise be livelocked...
5249                  */
5250                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5251                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5252                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5253                 if (err)
5254                         return err;
5255         }
5256
5257         return dquot_quota_on(sb, type, format_id, path);
5258 }
5259
5260 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5261                              unsigned int flags)
5262 {
5263         int err;
5264         struct inode *qf_inode;
5265         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5266                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5267                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5268         };
5269
5270         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5271
5272         if (!qf_inums[type])
5273                 return -EPERM;
5274
5275         qf_inode = ext4_iget(sb, qf_inums[type]);
5276         if (IS_ERR(qf_inode)) {
5277                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5278                 return PTR_ERR(qf_inode);
5279         }
5280
5281         /* Don't account quota for quota files to avoid recursion */
5282         qf_inode->i_flags |= S_NOQUOTA;
5283         err = dquot_enable(qf_inode, type, format_id, flags);
5284         iput(qf_inode);
5285
5286         return err;
5287 }
5288
5289 /* Enable usage tracking for all quota types. */
5290 static int ext4_enable_quotas(struct super_block *sb)
5291 {
5292         int type, err = 0;
5293         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5294                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5295                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5296         };
5297
5298         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5299         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5300                 if (qf_inums[type]) {
5301                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5302                                                 DQUOT_USAGE_ENABLED);
5303                         if (err) {
5304                                 ext4_warning(sb,
5305                                         "Failed to enable quota tracking "
5306                                         "(type=%d, err=%d). Please run "
5307                                         "e2fsck to fix.", type, err);
5308                                 return err;
5309                         }
5310                 }
5311         }
5312         return 0;
5313 }
5314
5315 static int ext4_quota_off(struct super_block *sb, int type)
5316 {
5317         struct inode *inode = sb_dqopt(sb)->files[type];
5318         handle_t *handle;
5319
5320         /* Force all delayed allocation blocks to be allocated.
5321          * Caller already holds s_umount sem */
5322         if (test_opt(sb, DELALLOC))
5323                 sync_filesystem(sb);
5324
5325         if (!inode)
5326                 goto out;
5327
5328         /* Update modification times of quota files when userspace can
5329          * start looking at them */
5330         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5331         if (IS_ERR(handle))
5332                 goto out;
5333         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5334         ext4_mark_inode_dirty(handle, inode);
5335         ext4_journal_stop(handle);
5336
5337 out:
5338         return dquot_quota_off(sb, type);
5339 }
5340
5341 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5342  * acquiring the locks... As quota files are never truncated and quota code
5343  * itself serializes the operations (and no one else should touch the files)
5344  * we don't have to be afraid of races */
5345 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5346                                size_t len, loff_t off)
5347 {
5348         struct inode *inode = sb_dqopt(sb)->files[type];
5349         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5350         int offset = off & (sb->s_blocksize - 1);
5351         int tocopy;
5352         size_t toread;
5353         struct buffer_head *bh;
5354         loff_t i_size = i_size_read(inode);
5355
5356         if (off > i_size)
5357                 return 0;
5358         if (off+len > i_size)
5359                 len = i_size-off;
5360         toread = len;
5361         while (toread > 0) {
5362                 tocopy = sb->s_blocksize - offset < toread ?
5363                                 sb->s_blocksize - offset : toread;
5364                 bh = ext4_bread(NULL, inode, blk, 0);
5365                 if (IS_ERR(bh))
5366                         return PTR_ERR(bh);
5367                 if (!bh)        /* A hole? */
5368                         memset(data, 0, tocopy);
5369                 else
5370                         memcpy(data, bh->b_data+offset, tocopy);
5371                 brelse(bh);
5372                 offset = 0;
5373                 toread -= tocopy;
5374                 data += tocopy;
5375                 blk++;
5376         }
5377         return len;
5378 }
5379
5380 /* Write to quotafile (we know the transaction is already started and has
5381  * enough credits) */
5382 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5383                                 const char *data, size_t len, loff_t off)
5384 {
5385         struct inode *inode = sb_dqopt(sb)->files[type];
5386         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5387         int err, offset = off & (sb->s_blocksize - 1);
5388         struct buffer_head *bh;
5389         handle_t *handle = journal_current_handle();
5390
5391         if (EXT4_SB(sb)->s_journal && !handle) {
5392                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5393                         " cancelled because transaction is not started",
5394                         (unsigned long long)off, (unsigned long long)len);
5395                 return -EIO;
5396         }
5397         /*
5398          * Since we account only one data block in transaction credits,
5399          * then it is impossible to cross a block boundary.
5400          */
5401         if (sb->s_blocksize - offset < len) {
5402                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5403                         " cancelled because not block aligned",
5404                         (unsigned long long)off, (unsigned long long)len);
5405                 return -EIO;
5406         }
5407
5408         bh = ext4_bread(handle, inode, blk, 1);
5409         if (IS_ERR(bh))
5410                 return PTR_ERR(bh);
5411         if (!bh)
5412                 goto out;
5413         BUFFER_TRACE(bh, "get write access");
5414         err = ext4_journal_get_write_access(handle, bh);
5415         if (err) {
5416                 brelse(bh);
5417                 return err;
5418         }
5419         lock_buffer(bh);
5420         memcpy(bh->b_data+offset, data, len);
5421         flush_dcache_page(bh->b_page);
5422         unlock_buffer(bh);
5423         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5424         brelse(bh);
5425 out:
5426         if (inode->i_size < off + len) {
5427                 i_size_write(inode, off + len);
5428                 EXT4_I(inode)->i_disksize = inode->i_size;
5429                 ext4_mark_inode_dirty(handle, inode);
5430         }
5431         return len;
5432 }
5433
5434 #endif
5435
5436 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5437                        const char *dev_name, void *data)
5438 {
5439         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5440 }
5441
5442 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5443 static inline void register_as_ext2(void)
5444 {
5445         int err = register_filesystem(&ext2_fs_type);
5446         if (err)
5447                 printk(KERN_WARNING
5448                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5449 }
5450
5451 static inline void unregister_as_ext2(void)
5452 {
5453         unregister_filesystem(&ext2_fs_type);
5454 }
5455
5456 static inline int ext2_feature_set_ok(struct super_block *sb)
5457 {
5458         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5459                 return 0;
5460         if (sb->s_flags & MS_RDONLY)
5461                 return 1;
5462         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5463                 return 0;
5464         return 1;
5465 }
5466 #else
5467 static inline void register_as_ext2(void) { }
5468 static inline void unregister_as_ext2(void) { }
5469 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5470 #endif
5471
5472 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5473 static inline void register_as_ext3(void)
5474 {
5475         int err = register_filesystem(&ext3_fs_type);
5476         if (err)
5477                 printk(KERN_WARNING
5478                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5479 }
5480
5481 static inline void unregister_as_ext3(void)
5482 {
5483         unregister_filesystem(&ext3_fs_type);
5484 }
5485
5486 static inline int ext3_feature_set_ok(struct super_block *sb)
5487 {
5488         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5489                 return 0;
5490         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5491                 return 0;
5492         if (sb->s_flags & MS_RDONLY)
5493                 return 1;
5494         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5495                 return 0;
5496         return 1;
5497 }
5498 #else
5499 static inline void register_as_ext3(void) { }
5500 static inline void unregister_as_ext3(void) { }
5501 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5502 #endif
5503
5504 static struct file_system_type ext4_fs_type = {
5505         .owner          = THIS_MODULE,
5506         .name           = "ext4",
5507         .mount          = ext4_mount,
5508         .kill_sb        = kill_block_super,
5509         .fs_flags       = FS_REQUIRES_DEV,
5510 };
5511 MODULE_ALIAS_FS("ext4");
5512
5513 static int __init ext4_init_feat_adverts(void)
5514 {
5515         struct ext4_features *ef;
5516         int ret = -ENOMEM;
5517
5518         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5519         if (!ef)
5520                 goto out;
5521
5522         ef->f_kobj.kset = ext4_kset;
5523         init_completion(&ef->f_kobj_unregister);
5524         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5525                                    "features");
5526         if (ret) {
5527                 kfree(ef);
5528                 goto out;
5529         }
5530
5531         ext4_feat = ef;
5532         ret = 0;
5533 out:
5534         return ret;
5535 }
5536
5537 static void ext4_exit_feat_adverts(void)
5538 {
5539         kobject_put(&ext4_feat->f_kobj);
5540         wait_for_completion(&ext4_feat->f_kobj_unregister);
5541         kfree(ext4_feat);
5542 }
5543
5544 /* Shared across all ext4 file systems */
5545 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5546 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5547
5548 static int __init ext4_init_fs(void)
5549 {
5550         int i, err;
5551
5552         ext4_li_info = NULL;
5553         mutex_init(&ext4_li_mtx);
5554
5555         /* Build-time check for flags consistency */
5556         ext4_check_flag_values();
5557
5558         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5559                 mutex_init(&ext4__aio_mutex[i]);
5560                 init_waitqueue_head(&ext4__ioend_wq[i]);
5561         }
5562
5563         err = ext4_init_es();
5564         if (err)
5565                 return err;
5566
5567         err = ext4_init_pageio();
5568         if (err)
5569                 goto out7;
5570
5571         err = ext4_init_system_zone();
5572         if (err)
5573                 goto out6;
5574         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5575         if (!ext4_kset) {
5576                 err = -ENOMEM;
5577                 goto out5;
5578         }
5579         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5580
5581         err = ext4_init_feat_adverts();
5582         if (err)
5583                 goto out4;
5584
5585         err = ext4_init_mballoc();
5586         if (err)
5587                 goto out2;
5588         else
5589                 ext4_mballoc_ready = 1;
5590         err = init_inodecache();
5591         if (err)
5592                 goto out1;
5593         register_as_ext3();
5594         register_as_ext2();
5595         err = register_filesystem(&ext4_fs_type);
5596         if (err)
5597                 goto out;
5598
5599         return 0;
5600 out:
5601         unregister_as_ext2();
5602         unregister_as_ext3();
5603         destroy_inodecache();
5604 out1:
5605         ext4_mballoc_ready = 0;
5606         ext4_exit_mballoc();
5607 out2:
5608         ext4_exit_feat_adverts();
5609 out4:
5610         if (ext4_proc_root)
5611                 remove_proc_entry("fs/ext4", NULL);
5612         kset_unregister(ext4_kset);
5613 out5:
5614         ext4_exit_system_zone();
5615 out6:
5616         ext4_exit_pageio();
5617 out7:
5618         ext4_exit_es();
5619
5620         return err;
5621 }
5622
5623 static void __exit ext4_exit_fs(void)
5624 {
5625         ext4_destroy_lazyinit_thread();
5626         unregister_as_ext2();
5627         unregister_as_ext3();
5628         unregister_filesystem(&ext4_fs_type);
5629         destroy_inodecache();
5630         ext4_exit_mballoc();
5631         ext4_exit_feat_adverts();
5632         remove_proc_entry("fs/ext4", NULL);
5633         kset_unregister(ext4_kset);
5634         ext4_exit_system_zone();
5635         ext4_exit_pageio();
5636         ext4_exit_es();
5637 }
5638
5639 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5640 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5641 MODULE_LICENSE("GPL");
5642 module_init(ext4_init_fs)
5643 module_exit(ext4_exit_fs)