fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_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_lazytime, Opt_nolazytime,
1130         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1131         Opt_inode_readahead_blks, Opt_journal_ioprio,
1132         Opt_dioread_nolock, Opt_dioread_lock,
1133         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1134         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1135 };
1136
1137 static const match_table_t tokens = {
1138         {Opt_bsd_df, "bsddf"},
1139         {Opt_minix_df, "minixdf"},
1140         {Opt_grpid, "grpid"},
1141         {Opt_grpid, "bsdgroups"},
1142         {Opt_nogrpid, "nogrpid"},
1143         {Opt_nogrpid, "sysvgroups"},
1144         {Opt_resgid, "resgid=%u"},
1145         {Opt_resuid, "resuid=%u"},
1146         {Opt_sb, "sb=%u"},
1147         {Opt_err_cont, "errors=continue"},
1148         {Opt_err_panic, "errors=panic"},
1149         {Opt_err_ro, "errors=remount-ro"},
1150         {Opt_nouid32, "nouid32"},
1151         {Opt_debug, "debug"},
1152         {Opt_removed, "oldalloc"},
1153         {Opt_removed, "orlov"},
1154         {Opt_user_xattr, "user_xattr"},
1155         {Opt_nouser_xattr, "nouser_xattr"},
1156         {Opt_acl, "acl"},
1157         {Opt_noacl, "noacl"},
1158         {Opt_noload, "norecovery"},
1159         {Opt_noload, "noload"},
1160         {Opt_removed, "nobh"},
1161         {Opt_removed, "bh"},
1162         {Opt_commit, "commit=%u"},
1163         {Opt_min_batch_time, "min_batch_time=%u"},
1164         {Opt_max_batch_time, "max_batch_time=%u"},
1165         {Opt_journal_dev, "journal_dev=%u"},
1166         {Opt_journal_path, "journal_path=%s"},
1167         {Opt_journal_checksum, "journal_checksum"},
1168         {Opt_nojournal_checksum, "nojournal_checksum"},
1169         {Opt_journal_async_commit, "journal_async_commit"},
1170         {Opt_abort, "abort"},
1171         {Opt_data_journal, "data=journal"},
1172         {Opt_data_ordered, "data=ordered"},
1173         {Opt_data_writeback, "data=writeback"},
1174         {Opt_data_err_abort, "data_err=abort"},
1175         {Opt_data_err_ignore, "data_err=ignore"},
1176         {Opt_offusrjquota, "usrjquota="},
1177         {Opt_usrjquota, "usrjquota=%s"},
1178         {Opt_offgrpjquota, "grpjquota="},
1179         {Opt_grpjquota, "grpjquota=%s"},
1180         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1181         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1182         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1183         {Opt_grpquota, "grpquota"},
1184         {Opt_noquota, "noquota"},
1185         {Opt_quota, "quota"},
1186         {Opt_usrquota, "usrquota"},
1187         {Opt_barrier, "barrier=%u"},
1188         {Opt_barrier, "barrier"},
1189         {Opt_nobarrier, "nobarrier"},
1190         {Opt_i_version, "i_version"},
1191         {Opt_dax, "dax"},
1192         {Opt_stripe, "stripe=%u"},
1193         {Opt_delalloc, "delalloc"},
1194         {Opt_lazytime, "lazytime"},
1195         {Opt_nolazytime, "nolazytime"},
1196         {Opt_nodelalloc, "nodelalloc"},
1197         {Opt_removed, "mblk_io_submit"},
1198         {Opt_removed, "nomblk_io_submit"},
1199         {Opt_block_validity, "block_validity"},
1200         {Opt_noblock_validity, "noblock_validity"},
1201         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1202         {Opt_journal_ioprio, "journal_ioprio=%u"},
1203         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1204         {Opt_auto_da_alloc, "auto_da_alloc"},
1205         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1206         {Opt_dioread_nolock, "dioread_nolock"},
1207         {Opt_dioread_lock, "dioread_lock"},
1208         {Opt_discard, "discard"},
1209         {Opt_nodiscard, "nodiscard"},
1210         {Opt_init_itable, "init_itable=%u"},
1211         {Opt_init_itable, "init_itable"},
1212         {Opt_noinit_itable, "noinit_itable"},
1213         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1214         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1215         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1216         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1217         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1218         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1219         {Opt_err, NULL},
1220 };
1221
1222 static ext4_fsblk_t get_sb_block(void **data)
1223 {
1224         ext4_fsblk_t    sb_block;
1225         char            *options = (char *) *data;
1226
1227         if (!options || strncmp(options, "sb=", 3) != 0)
1228                 return 1;       /* Default location */
1229
1230         options += 3;
1231         /* TODO: use simple_strtoll with >32bit ext4 */
1232         sb_block = simple_strtoul(options, &options, 0);
1233         if (*options && *options != ',') {
1234                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1235                        (char *) *data);
1236                 return 1;
1237         }
1238         if (*options == ',')
1239                 options++;
1240         *data = (void *) options;
1241
1242         return sb_block;
1243 }
1244
1245 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1246 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1247         "Contact [email protected] if you think we should keep it.\n";
1248
1249 #ifdef CONFIG_QUOTA
1250 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1251 {
1252         struct ext4_sb_info *sbi = EXT4_SB(sb);
1253         char *qname;
1254         int ret = -1;
1255
1256         if (sb_any_quota_loaded(sb) &&
1257                 !sbi->s_qf_names[qtype]) {
1258                 ext4_msg(sb, KERN_ERR,
1259                         "Cannot change journaled "
1260                         "quota options when quota turned on");
1261                 return -1;
1262         }
1263         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1264                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1265                          "when QUOTA feature is enabled");
1266                 return -1;
1267         }
1268         qname = match_strdup(args);
1269         if (!qname) {
1270                 ext4_msg(sb, KERN_ERR,
1271                         "Not enough memory for storing quotafile name");
1272                 return -1;
1273         }
1274         if (sbi->s_qf_names[qtype]) {
1275                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1276                         ret = 1;
1277                 else
1278                         ext4_msg(sb, KERN_ERR,
1279                                  "%s quota file already specified",
1280                                  QTYPE2NAME(qtype));
1281                 goto errout;
1282         }
1283         if (strchr(qname, '/')) {
1284                 ext4_msg(sb, KERN_ERR,
1285                         "quotafile must be on filesystem root");
1286                 goto errout;
1287         }
1288         sbi->s_qf_names[qtype] = qname;
1289         set_opt(sb, QUOTA);
1290         return 1;
1291 errout:
1292         kfree(qname);
1293         return ret;
1294 }
1295
1296 static int clear_qf_name(struct super_block *sb, int qtype)
1297 {
1298
1299         struct ext4_sb_info *sbi = EXT4_SB(sb);
1300
1301         if (sb_any_quota_loaded(sb) &&
1302                 sbi->s_qf_names[qtype]) {
1303                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1304                         " when quota turned on");
1305                 return -1;
1306         }
1307         kfree(sbi->s_qf_names[qtype]);
1308         sbi->s_qf_names[qtype] = NULL;
1309         return 1;
1310 }
1311 #endif
1312
1313 #define MOPT_SET        0x0001
1314 #define MOPT_CLEAR      0x0002
1315 #define MOPT_NOSUPPORT  0x0004
1316 #define MOPT_EXPLICIT   0x0008
1317 #define MOPT_CLEAR_ERR  0x0010
1318 #define MOPT_GTE0       0x0020
1319 #ifdef CONFIG_QUOTA
1320 #define MOPT_Q          0
1321 #define MOPT_QFMT       0x0040
1322 #else
1323 #define MOPT_Q          MOPT_NOSUPPORT
1324 #define MOPT_QFMT       MOPT_NOSUPPORT
1325 #endif
1326 #define MOPT_DATAJ      0x0080
1327 #define MOPT_NO_EXT2    0x0100
1328 #define MOPT_NO_EXT3    0x0200
1329 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1330 #define MOPT_STRING     0x0400
1331
1332 static const struct mount_opts {
1333         int     token;
1334         int     mount_opt;
1335         int     flags;
1336 } ext4_mount_opts[] = {
1337         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1338         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1339         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1340         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1341         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1342         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1343         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1344          MOPT_EXT4_ONLY | MOPT_SET},
1345         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1346          MOPT_EXT4_ONLY | MOPT_CLEAR},
1347         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1348         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1349         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1350          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1351         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1352          MOPT_EXT4_ONLY | MOPT_CLEAR},
1353         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1354          MOPT_EXT4_ONLY | MOPT_CLEAR},
1355         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1356          MOPT_EXT4_ONLY | MOPT_SET},
1357         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1358                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1359          MOPT_EXT4_ONLY | MOPT_SET},
1360         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1361         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1362         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1363         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1364         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1365          MOPT_NO_EXT2 | MOPT_SET},
1366         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1367          MOPT_NO_EXT2 | MOPT_CLEAR},
1368         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1369         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1370         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1371         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1372         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1373         {Opt_commit, 0, MOPT_GTE0},
1374         {Opt_max_batch_time, 0, MOPT_GTE0},
1375         {Opt_min_batch_time, 0, MOPT_GTE0},
1376         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1377         {Opt_init_itable, 0, MOPT_GTE0},
1378         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1379         {Opt_stripe, 0, MOPT_GTE0},
1380         {Opt_resuid, 0, MOPT_GTE0},
1381         {Opt_resgid, 0, MOPT_GTE0},
1382         {Opt_journal_dev, 0, MOPT_GTE0},
1383         {Opt_journal_path, 0, MOPT_STRING},
1384         {Opt_journal_ioprio, 0, MOPT_GTE0},
1385         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1386         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1387         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1388          MOPT_NO_EXT2 | MOPT_DATAJ},
1389         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1390         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1391 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1392         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1393         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1394 #else
1395         {Opt_acl, 0, MOPT_NOSUPPORT},
1396         {Opt_noacl, 0, MOPT_NOSUPPORT},
1397 #endif
1398         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1399         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1400         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1401         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1402                                                         MOPT_SET | MOPT_Q},
1403         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1404                                                         MOPT_SET | MOPT_Q},
1405         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1406                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1407         {Opt_usrjquota, 0, MOPT_Q},
1408         {Opt_grpjquota, 0, MOPT_Q},
1409         {Opt_offusrjquota, 0, MOPT_Q},
1410         {Opt_offgrpjquota, 0, MOPT_Q},
1411         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1412         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1413         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1414         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1415         {Opt_err, 0, 0}
1416 };
1417
1418 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1419                             substring_t *args, unsigned long *journal_devnum,
1420                             unsigned int *journal_ioprio, int is_remount)
1421 {
1422         struct ext4_sb_info *sbi = EXT4_SB(sb);
1423         const struct mount_opts *m;
1424         kuid_t uid;
1425         kgid_t gid;
1426         int arg = 0;
1427
1428 #ifdef CONFIG_QUOTA
1429         if (token == Opt_usrjquota)
1430                 return set_qf_name(sb, USRQUOTA, &args[0]);
1431         else if (token == Opt_grpjquota)
1432                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1433         else if (token == Opt_offusrjquota)
1434                 return clear_qf_name(sb, USRQUOTA);
1435         else if (token == Opt_offgrpjquota)
1436                 return clear_qf_name(sb, GRPQUOTA);
1437 #endif
1438         switch (token) {
1439         case Opt_noacl:
1440         case Opt_nouser_xattr:
1441                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1442                 break;
1443         case Opt_sb:
1444                 return 1;       /* handled by get_sb_block() */
1445         case Opt_removed:
1446                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1447                 return 1;
1448         case Opt_abort:
1449                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1450                 return 1;
1451         case Opt_i_version:
1452                 sb->s_flags |= MS_I_VERSION;
1453                 return 1;
1454         case Opt_lazytime:
1455                 sb->s_flags |= MS_LAZYTIME;
1456                 return 1;
1457         case Opt_nolazytime:
1458                 sb->s_flags &= ~MS_LAZYTIME;
1459                 return 1;
1460         }
1461
1462         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1463                 if (token == m->token)
1464                         break;
1465
1466         if (m->token == Opt_err) {
1467                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1468                          "or missing value", opt);
1469                 return -1;
1470         }
1471
1472         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1473                 ext4_msg(sb, KERN_ERR,
1474                          "Mount option \"%s\" incompatible with ext2", opt);
1475                 return -1;
1476         }
1477         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1478                 ext4_msg(sb, KERN_ERR,
1479                          "Mount option \"%s\" incompatible with ext3", opt);
1480                 return -1;
1481         }
1482
1483         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1484                 return -1;
1485         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1486                 return -1;
1487         if (m->flags & MOPT_EXPLICIT)
1488                 set_opt2(sb, EXPLICIT_DELALLOC);
1489         if (m->flags & MOPT_CLEAR_ERR)
1490                 clear_opt(sb, ERRORS_MASK);
1491         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1492                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1493                          "options when quota turned on");
1494                 return -1;
1495         }
1496
1497         if (m->flags & MOPT_NOSUPPORT) {
1498                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1499         } else if (token == Opt_commit) {
1500                 if (arg == 0)
1501                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1502                 sbi->s_commit_interval = HZ * arg;
1503         } else if (token == Opt_max_batch_time) {
1504                 sbi->s_max_batch_time = arg;
1505         } else if (token == Opt_min_batch_time) {
1506                 sbi->s_min_batch_time = arg;
1507         } else if (token == Opt_inode_readahead_blks) {
1508                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1509                         ext4_msg(sb, KERN_ERR,
1510                                  "EXT4-fs: inode_readahead_blks must be "
1511                                  "0 or a power of 2 smaller than 2^31");
1512                         return -1;
1513                 }
1514                 sbi->s_inode_readahead_blks = arg;
1515         } else if (token == Opt_init_itable) {
1516                 set_opt(sb, INIT_INODE_TABLE);
1517                 if (!args->from)
1518                         arg = EXT4_DEF_LI_WAIT_MULT;
1519                 sbi->s_li_wait_mult = arg;
1520         } else if (token == Opt_max_dir_size_kb) {
1521                 sbi->s_max_dir_size_kb = arg;
1522         } else if (token == Opt_stripe) {
1523                 sbi->s_stripe = arg;
1524         } else if (token == Opt_resuid) {
1525                 uid = make_kuid(current_user_ns(), arg);
1526                 if (!uid_valid(uid)) {
1527                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1528                         return -1;
1529                 }
1530                 sbi->s_resuid = uid;
1531         } else if (token == Opt_resgid) {
1532                 gid = make_kgid(current_user_ns(), arg);
1533                 if (!gid_valid(gid)) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1535                         return -1;
1536                 }
1537                 sbi->s_resgid = gid;
1538         } else if (token == Opt_journal_dev) {
1539                 if (is_remount) {
1540                         ext4_msg(sb, KERN_ERR,
1541                                  "Cannot specify journal on remount");
1542                         return -1;
1543                 }
1544                 *journal_devnum = arg;
1545         } else if (token == Opt_journal_path) {
1546                 char *journal_path;
1547                 struct inode *journal_inode;
1548                 struct path path;
1549                 int error;
1550
1551                 if (is_remount) {
1552                         ext4_msg(sb, KERN_ERR,
1553                                  "Cannot specify journal on remount");
1554                         return -1;
1555                 }
1556                 journal_path = match_strdup(&args[0]);
1557                 if (!journal_path) {
1558                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1559                                 "journal device string");
1560                         return -1;
1561                 }
1562
1563                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1564                 if (error) {
1565                         ext4_msg(sb, KERN_ERR, "error: could not find "
1566                                 "journal device path: error %d", error);
1567                         kfree(journal_path);
1568                         return -1;
1569                 }
1570
1571                 journal_inode = path.dentry->d_inode;
1572                 if (!S_ISBLK(journal_inode->i_mode)) {
1573                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1574                                 "is not a block device", journal_path);
1575                         path_put(&path);
1576                         kfree(journal_path);
1577                         return -1;
1578                 }
1579
1580                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1581                 path_put(&path);
1582                 kfree(journal_path);
1583         } else if (token == Opt_journal_ioprio) {
1584                 if (arg > 7) {
1585                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1586                                  " (must be 0-7)");
1587                         return -1;
1588                 }
1589                 *journal_ioprio =
1590                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1591         } else if (m->flags & MOPT_DATAJ) {
1592                 if (is_remount) {
1593                         if (!sbi->s_journal)
1594                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1595                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1596                                 ext4_msg(sb, KERN_ERR,
1597                                          "Cannot change data mode on remount");
1598                                 return -1;
1599                         }
1600                 } else {
1601                         clear_opt(sb, DATA_FLAGS);
1602                         sbi->s_mount_opt |= m->mount_opt;
1603                 }
1604 #ifdef CONFIG_QUOTA
1605         } else if (m->flags & MOPT_QFMT) {
1606                 if (sb_any_quota_loaded(sb) &&
1607                     sbi->s_jquota_fmt != m->mount_opt) {
1608                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1609                                  "quota options when quota turned on");
1610                         return -1;
1611                 }
1612                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1613                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1614                         ext4_msg(sb, KERN_ERR,
1615                                  "Cannot set journaled quota options "
1616                                  "when QUOTA feature is enabled");
1617                         return -1;
1618                 }
1619                 sbi->s_jquota_fmt = m->mount_opt;
1620 #endif
1621 #ifndef CONFIG_FS_DAX
1622         } else if (token == Opt_dax) {
1623                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1624                 return -1;
1625 #endif
1626         } else {
1627                 if (!args->from)
1628                         arg = 1;
1629                 if (m->flags & MOPT_CLEAR)
1630                         arg = !arg;
1631                 else if (unlikely(!(m->flags & MOPT_SET))) {
1632                         ext4_msg(sb, KERN_WARNING,
1633                                  "buggy handling of option %s", opt);
1634                         WARN_ON(1);
1635                         return -1;
1636                 }
1637                 if (arg != 0)
1638                         sbi->s_mount_opt |= m->mount_opt;
1639                 else
1640                         sbi->s_mount_opt &= ~m->mount_opt;
1641         }
1642         return 1;
1643 }
1644
1645 static int parse_options(char *options, struct super_block *sb,
1646                          unsigned long *journal_devnum,
1647                          unsigned int *journal_ioprio,
1648                          int is_remount)
1649 {
1650         struct ext4_sb_info *sbi = EXT4_SB(sb);
1651         char *p;
1652         substring_t args[MAX_OPT_ARGS];
1653         int token;
1654
1655         if (!options)
1656                 return 1;
1657
1658         while ((p = strsep(&options, ",")) != NULL) {
1659                 if (!*p)
1660                         continue;
1661                 /*
1662                  * Initialize args struct so we know whether arg was
1663                  * found; some options take optional arguments.
1664                  */
1665                 args[0].to = args[0].from = NULL;
1666                 token = match_token(p, tokens, args);
1667                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1668                                      journal_ioprio, is_remount) < 0)
1669                         return 0;
1670         }
1671 #ifdef CONFIG_QUOTA
1672         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1673             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1674                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1675                          "feature is enabled");
1676                 return 0;
1677         }
1678         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1679                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1680                         clear_opt(sb, USRQUOTA);
1681
1682                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1683                         clear_opt(sb, GRPQUOTA);
1684
1685                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1686                         ext4_msg(sb, KERN_ERR, "old and new quota "
1687                                         "format mixing");
1688                         return 0;
1689                 }
1690
1691                 if (!sbi->s_jquota_fmt) {
1692                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1693                                         "not specified");
1694                         return 0;
1695                 }
1696         }
1697 #endif
1698         if (test_opt(sb, DIOREAD_NOLOCK)) {
1699                 int blocksize =
1700                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1701
1702                 if (blocksize < PAGE_CACHE_SIZE) {
1703                         ext4_msg(sb, KERN_ERR, "can't mount with "
1704                                  "dioread_nolock if block size != PAGE_SIZE");
1705                         return 0;
1706                 }
1707         }
1708         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1709             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1710                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1711                          "in data=ordered mode");
1712                 return 0;
1713         }
1714         return 1;
1715 }
1716
1717 static inline void ext4_show_quota_options(struct seq_file *seq,
1718                                            struct super_block *sb)
1719 {
1720 #if defined(CONFIG_QUOTA)
1721         struct ext4_sb_info *sbi = EXT4_SB(sb);
1722
1723         if (sbi->s_jquota_fmt) {
1724                 char *fmtname = "";
1725
1726                 switch (sbi->s_jquota_fmt) {
1727                 case QFMT_VFS_OLD:
1728                         fmtname = "vfsold";
1729                         break;
1730                 case QFMT_VFS_V0:
1731                         fmtname = "vfsv0";
1732                         break;
1733                 case QFMT_VFS_V1:
1734                         fmtname = "vfsv1";
1735                         break;
1736                 }
1737                 seq_printf(seq, ",jqfmt=%s", fmtname);
1738         }
1739
1740         if (sbi->s_qf_names[USRQUOTA])
1741                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1742
1743         if (sbi->s_qf_names[GRPQUOTA])
1744                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1745 #endif
1746 }
1747
1748 static const char *token2str(int token)
1749 {
1750         const struct match_token *t;
1751
1752         for (t = tokens; t->token != Opt_err; t++)
1753                 if (t->token == token && !strchr(t->pattern, '='))
1754                         break;
1755         return t->pattern;
1756 }
1757
1758 /*
1759  * Show an option if
1760  *  - it's set to a non-default value OR
1761  *  - if the per-sb default is different from the global default
1762  */
1763 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1764                               int nodefs)
1765 {
1766         struct ext4_sb_info *sbi = EXT4_SB(sb);
1767         struct ext4_super_block *es = sbi->s_es;
1768         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1769         const struct mount_opts *m;
1770         char sep = nodefs ? '\n' : ',';
1771
1772 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1773 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1774
1775         if (sbi->s_sb_block != 1)
1776                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1777
1778         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1779                 int want_set = m->flags & MOPT_SET;
1780                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1781                     (m->flags & MOPT_CLEAR_ERR))
1782                         continue;
1783                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1784                         continue; /* skip if same as the default */
1785                 if ((want_set &&
1786                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1787                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1788                         continue; /* select Opt_noFoo vs Opt_Foo */
1789                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1790         }
1791
1792         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1793             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1794                 SEQ_OPTS_PRINT("resuid=%u",
1795                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1796         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1797             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1798                 SEQ_OPTS_PRINT("resgid=%u",
1799                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1800         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1801         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1802                 SEQ_OPTS_PUTS("errors=remount-ro");
1803         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1804                 SEQ_OPTS_PUTS("errors=continue");
1805         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1806                 SEQ_OPTS_PUTS("errors=panic");
1807         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1808                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1809         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1810                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1811         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1812                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1813         if (sb->s_flags & MS_I_VERSION)
1814                 SEQ_OPTS_PUTS("i_version");
1815         if (nodefs || sbi->s_stripe)
1816                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1817         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1818                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1819                         SEQ_OPTS_PUTS("data=journal");
1820                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1821                         SEQ_OPTS_PUTS("data=ordered");
1822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1823                         SEQ_OPTS_PUTS("data=writeback");
1824         }
1825         if (nodefs ||
1826             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1827                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1828                                sbi->s_inode_readahead_blks);
1829
1830         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1831                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1832                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1833         if (nodefs || sbi->s_max_dir_size_kb)
1834                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1835
1836         ext4_show_quota_options(seq, sb);
1837         return 0;
1838 }
1839
1840 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1841 {
1842         return _ext4_show_options(seq, root->d_sb, 0);
1843 }
1844
1845 static int options_seq_show(struct seq_file *seq, void *offset)
1846 {
1847         struct super_block *sb = seq->private;
1848         int rc;
1849
1850         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1851         rc = _ext4_show_options(seq, sb, 1);
1852         seq_puts(seq, "\n");
1853         return rc;
1854 }
1855
1856 static int options_open_fs(struct inode *inode, struct file *file)
1857 {
1858         return single_open(file, options_seq_show, PDE_DATA(inode));
1859 }
1860
1861 static const struct file_operations ext4_seq_options_fops = {
1862         .owner = THIS_MODULE,
1863         .open = options_open_fs,
1864         .read = seq_read,
1865         .llseek = seq_lseek,
1866         .release = single_release,
1867 };
1868
1869 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1870                             int read_only)
1871 {
1872         struct ext4_sb_info *sbi = EXT4_SB(sb);
1873         int res = 0;
1874
1875         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1876                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1877                          "forcing read-only mode");
1878                 res = MS_RDONLY;
1879         }
1880         if (read_only)
1881                 goto done;
1882         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1883                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1884                          "running e2fsck is recommended");
1885         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1886                 ext4_msg(sb, KERN_WARNING,
1887                          "warning: mounting fs with errors, "
1888                          "running e2fsck is recommended");
1889         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1890                  le16_to_cpu(es->s_mnt_count) >=
1891                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1892                 ext4_msg(sb, KERN_WARNING,
1893                          "warning: maximal mount count reached, "
1894                          "running e2fsck is recommended");
1895         else if (le32_to_cpu(es->s_checkinterval) &&
1896                 (le32_to_cpu(es->s_lastcheck) +
1897                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1898                 ext4_msg(sb, KERN_WARNING,
1899                          "warning: checktime reached, "
1900                          "running e2fsck is recommended");
1901         if (!sbi->s_journal)
1902                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1903         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1904                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1905         le16_add_cpu(&es->s_mnt_count, 1);
1906         es->s_mtime = cpu_to_le32(get_seconds());
1907         ext4_update_dynamic_rev(sb);
1908         if (sbi->s_journal)
1909                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1910
1911         ext4_commit_super(sb, 1);
1912 done:
1913         if (test_opt(sb, DEBUG))
1914                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1915                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1916                         sb->s_blocksize,
1917                         sbi->s_groups_count,
1918                         EXT4_BLOCKS_PER_GROUP(sb),
1919                         EXT4_INODES_PER_GROUP(sb),
1920                         sbi->s_mount_opt, sbi->s_mount_opt2);
1921
1922         cleancache_init_fs(sb);
1923         return res;
1924 }
1925
1926 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1927 {
1928         struct ext4_sb_info *sbi = EXT4_SB(sb);
1929         struct flex_groups *new_groups;
1930         int size;
1931
1932         if (!sbi->s_log_groups_per_flex)
1933                 return 0;
1934
1935         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1936         if (size <= sbi->s_flex_groups_allocated)
1937                 return 0;
1938
1939         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1940         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1941         if (!new_groups) {
1942                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1943                          size / (int) sizeof(struct flex_groups));
1944                 return -ENOMEM;
1945         }
1946
1947         if (sbi->s_flex_groups) {
1948                 memcpy(new_groups, sbi->s_flex_groups,
1949                        (sbi->s_flex_groups_allocated *
1950                         sizeof(struct flex_groups)));
1951                 kvfree(sbi->s_flex_groups);
1952         }
1953         sbi->s_flex_groups = new_groups;
1954         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1955         return 0;
1956 }
1957
1958 static int ext4_fill_flex_info(struct super_block *sb)
1959 {
1960         struct ext4_sb_info *sbi = EXT4_SB(sb);
1961         struct ext4_group_desc *gdp = NULL;
1962         ext4_group_t flex_group;
1963         int i, err;
1964
1965         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1966         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1967                 sbi->s_log_groups_per_flex = 0;
1968                 return 1;
1969         }
1970
1971         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1972         if (err)
1973                 goto failed;
1974
1975         for (i = 0; i < sbi->s_groups_count; i++) {
1976                 gdp = ext4_get_group_desc(sb, i, NULL);
1977
1978                 flex_group = ext4_flex_group(sbi, i);
1979                 atomic_add(ext4_free_inodes_count(sb, gdp),
1980                            &sbi->s_flex_groups[flex_group].free_inodes);
1981                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1982                              &sbi->s_flex_groups[flex_group].free_clusters);
1983                 atomic_add(ext4_used_dirs_count(sb, gdp),
1984                            &sbi->s_flex_groups[flex_group].used_dirs);
1985         }
1986
1987         return 1;
1988 failed:
1989         return 0;
1990 }
1991
1992 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1993                                    struct ext4_group_desc *gdp)
1994 {
1995         int offset;
1996         __u16 crc = 0;
1997         __le32 le_group = cpu_to_le32(block_group);
1998
1999         if (ext4_has_metadata_csum(sbi->s_sb)) {
2000                 /* Use new metadata_csum algorithm */
2001                 __le16 save_csum;
2002                 __u32 csum32;
2003
2004                 save_csum = gdp->bg_checksum;
2005                 gdp->bg_checksum = 0;
2006                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2007                                      sizeof(le_group));
2008                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2009                                      sbi->s_desc_size);
2010                 gdp->bg_checksum = save_csum;
2011
2012                 crc = csum32 & 0xFFFF;
2013                 goto out;
2014         }
2015
2016         /* old crc16 code */
2017         if (!(sbi->s_es->s_feature_ro_compat &
2018               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2019                 return 0;
2020
2021         offset = offsetof(struct ext4_group_desc, bg_checksum);
2022
2023         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2024         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2025         crc = crc16(crc, (__u8 *)gdp, offset);
2026         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2027         /* for checksum of struct ext4_group_desc do the rest...*/
2028         if ((sbi->s_es->s_feature_incompat &
2029              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2030             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2031                 crc = crc16(crc, (__u8 *)gdp + offset,
2032                             le16_to_cpu(sbi->s_es->s_desc_size) -
2033                                 offset);
2034
2035 out:
2036         return cpu_to_le16(crc);
2037 }
2038
2039 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2040                                 struct ext4_group_desc *gdp)
2041 {
2042         if (ext4_has_group_desc_csum(sb) &&
2043             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2044                                                       block_group, gdp)))
2045                 return 0;
2046
2047         return 1;
2048 }
2049
2050 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2051                               struct ext4_group_desc *gdp)
2052 {
2053         if (!ext4_has_group_desc_csum(sb))
2054                 return;
2055         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2056 }
2057
2058 /* Called at mount-time, super-block is locked */
2059 static int ext4_check_descriptors(struct super_block *sb,
2060                                   ext4_group_t *first_not_zeroed)
2061 {
2062         struct ext4_sb_info *sbi = EXT4_SB(sb);
2063         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2064         ext4_fsblk_t last_block;
2065         ext4_fsblk_t block_bitmap;
2066         ext4_fsblk_t inode_bitmap;
2067         ext4_fsblk_t inode_table;
2068         int flexbg_flag = 0;
2069         ext4_group_t i, grp = sbi->s_groups_count;
2070
2071         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2072                 flexbg_flag = 1;
2073
2074         ext4_debug("Checking group descriptors");
2075
2076         for (i = 0; i < sbi->s_groups_count; i++) {
2077                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2078
2079                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2080                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2081                 else
2082                         last_block = first_block +
2083                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2084
2085                 if ((grp == sbi->s_groups_count) &&
2086                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2087                         grp = i;
2088
2089                 block_bitmap = ext4_block_bitmap(sb, gdp);
2090                 if (block_bitmap < first_block || block_bitmap > last_block) {
2091                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2092                                "Block bitmap for group %u not in group "
2093                                "(block %llu)!", i, block_bitmap);
2094                         return 0;
2095                 }
2096                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2097                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2098                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2099                                "Inode bitmap for group %u not in group "
2100                                "(block %llu)!", i, inode_bitmap);
2101                         return 0;
2102                 }
2103                 inode_table = ext4_inode_table(sb, gdp);
2104                 if (inode_table < first_block ||
2105                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2106                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2107                                "Inode table for group %u not in group "
2108                                "(block %llu)!", i, inode_table);
2109                         return 0;
2110                 }
2111                 ext4_lock_group(sb, i);
2112                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2113                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2114                                  "Checksum for group %u failed (%u!=%u)",
2115                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2116                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2117                         if (!(sb->s_flags & MS_RDONLY)) {
2118                                 ext4_unlock_group(sb, i);
2119                                 return 0;
2120                         }
2121                 }
2122                 ext4_unlock_group(sb, i);
2123                 if (!flexbg_flag)
2124                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2125         }
2126         if (NULL != first_not_zeroed)
2127                 *first_not_zeroed = grp;
2128         return 1;
2129 }
2130
2131 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2132  * the superblock) which were deleted from all directories, but held open by
2133  * a process at the time of a crash.  We walk the list and try to delete these
2134  * inodes at recovery time (only with a read-write filesystem).
2135  *
2136  * In order to keep the orphan inode chain consistent during traversal (in
2137  * case of crash during recovery), we link each inode into the superblock
2138  * orphan list_head and handle it the same way as an inode deletion during
2139  * normal operation (which journals the operations for us).
2140  *
2141  * We only do an iget() and an iput() on each inode, which is very safe if we
2142  * accidentally point at an in-use or already deleted inode.  The worst that
2143  * can happen in this case is that we get a "bit already cleared" message from
2144  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2145  * e2fsck was run on this filesystem, and it must have already done the orphan
2146  * inode cleanup for us, so we can safely abort without any further action.
2147  */
2148 static void ext4_orphan_cleanup(struct super_block *sb,
2149                                 struct ext4_super_block *es)
2150 {
2151         unsigned int s_flags = sb->s_flags;
2152         int nr_orphans = 0, nr_truncates = 0;
2153 #ifdef CONFIG_QUOTA
2154         int i;
2155 #endif
2156         if (!es->s_last_orphan) {
2157                 jbd_debug(4, "no orphan inodes to clean up\n");
2158                 return;
2159         }
2160
2161         if (bdev_read_only(sb->s_bdev)) {
2162                 ext4_msg(sb, KERN_ERR, "write access "
2163                         "unavailable, skipping orphan cleanup");
2164                 return;
2165         }
2166
2167         /* Check if feature set would not allow a r/w mount */
2168         if (!ext4_feature_set_ok(sb, 0)) {
2169                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2170                          "unknown ROCOMPAT features");
2171                 return;
2172         }
2173
2174         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2175                 /* don't clear list on RO mount w/ errors */
2176                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2177                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2178                                   "clearing orphan list.\n");
2179                         es->s_last_orphan = 0;
2180                 }
2181                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2182                 return;
2183         }
2184
2185         if (s_flags & MS_RDONLY) {
2186                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2187                 sb->s_flags &= ~MS_RDONLY;
2188         }
2189 #ifdef CONFIG_QUOTA
2190         /* Needed for iput() to work correctly and not trash data */
2191         sb->s_flags |= MS_ACTIVE;
2192         /* Turn on quotas so that they are updated correctly */
2193         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2194                 if (EXT4_SB(sb)->s_qf_names[i]) {
2195                         int ret = ext4_quota_on_mount(sb, i);
2196                         if (ret < 0)
2197                                 ext4_msg(sb, KERN_ERR,
2198                                         "Cannot turn on journaled "
2199                                         "quota: error %d", ret);
2200                 }
2201         }
2202 #endif
2203
2204         while (es->s_last_orphan) {
2205                 struct inode *inode;
2206
2207                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2208                 if (IS_ERR(inode)) {
2209                         es->s_last_orphan = 0;
2210                         break;
2211                 }
2212
2213                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2214                 dquot_initialize(inode);
2215                 if (inode->i_nlink) {
2216                         if (test_opt(sb, DEBUG))
2217                                 ext4_msg(sb, KERN_DEBUG,
2218                                         "%s: truncating inode %lu to %lld bytes",
2219                                         __func__, inode->i_ino, inode->i_size);
2220                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2221                                   inode->i_ino, inode->i_size);
2222                         mutex_lock(&inode->i_mutex);
2223                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2224                         ext4_truncate(inode);
2225                         mutex_unlock(&inode->i_mutex);
2226                         nr_truncates++;
2227                 } else {
2228                         if (test_opt(sb, DEBUG))
2229                                 ext4_msg(sb, KERN_DEBUG,
2230                                         "%s: deleting unreferenced inode %lu",
2231                                         __func__, inode->i_ino);
2232                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2233                                   inode->i_ino);
2234                         nr_orphans++;
2235                 }
2236                 iput(inode);  /* The delete magic happens here! */
2237         }
2238
2239 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2240
2241         if (nr_orphans)
2242                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2243                        PLURAL(nr_orphans));
2244         if (nr_truncates)
2245                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2246                        PLURAL(nr_truncates));
2247 #ifdef CONFIG_QUOTA
2248         /* Turn quotas off */
2249         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2250                 if (sb_dqopt(sb)->files[i])
2251                         dquot_quota_off(sb, i);
2252         }
2253 #endif
2254         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2255 }
2256
2257 /*
2258  * Maximal extent format file size.
2259  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2260  * extent format containers, within a sector_t, and within i_blocks
2261  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2262  * so that won't be a limiting factor.
2263  *
2264  * However there is other limiting factor. We do store extents in the form
2265  * of starting block and length, hence the resulting length of the extent
2266  * covering maximum file size must fit into on-disk format containers as
2267  * well. Given that length is always by 1 unit bigger than max unit (because
2268  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2269  *
2270  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2271  */
2272 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2273 {
2274         loff_t res;
2275         loff_t upper_limit = MAX_LFS_FILESIZE;
2276
2277         /* small i_blocks in vfs inode? */
2278         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2279                 /*
2280                  * CONFIG_LBDAF is not enabled implies the inode
2281                  * i_block represent total blocks in 512 bytes
2282                  * 32 == size of vfs inode i_blocks * 8
2283                  */
2284                 upper_limit = (1LL << 32) - 1;
2285
2286                 /* total blocks in file system block size */
2287                 upper_limit >>= (blkbits - 9);
2288                 upper_limit <<= blkbits;
2289         }
2290
2291         /*
2292          * 32-bit extent-start container, ee_block. We lower the maxbytes
2293          * by one fs block, so ee_len can cover the extent of maximum file
2294          * size
2295          */
2296         res = (1LL << 32) - 1;
2297         res <<= blkbits;
2298
2299         /* Sanity check against vm- & vfs- imposed limits */
2300         if (res > upper_limit)
2301                 res = upper_limit;
2302
2303         return res;
2304 }
2305
2306 /*
2307  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2308  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2309  * We need to be 1 filesystem block less than the 2^48 sector limit.
2310  */
2311 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2312 {
2313         loff_t res = EXT4_NDIR_BLOCKS;
2314         int meta_blocks;
2315         loff_t upper_limit;
2316         /* This is calculated to be the largest file size for a dense, block
2317          * mapped file such that the file's total number of 512-byte sectors,
2318          * including data and all indirect blocks, does not exceed (2^48 - 1).
2319          *
2320          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2321          * number of 512-byte sectors of the file.
2322          */
2323
2324         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2325                 /*
2326                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2327                  * the inode i_block field represents total file blocks in
2328                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2329                  */
2330                 upper_limit = (1LL << 32) - 1;
2331
2332                 /* total blocks in file system block size */
2333                 upper_limit >>= (bits - 9);
2334
2335         } else {
2336                 /*
2337                  * We use 48 bit ext4_inode i_blocks
2338                  * With EXT4_HUGE_FILE_FL set the i_blocks
2339                  * represent total number of blocks in
2340                  * file system block size
2341                  */
2342                 upper_limit = (1LL << 48) - 1;
2343
2344         }
2345
2346         /* indirect blocks */
2347         meta_blocks = 1;
2348         /* double indirect blocks */
2349         meta_blocks += 1 + (1LL << (bits-2));
2350         /* tripple indirect blocks */
2351         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2352
2353         upper_limit -= meta_blocks;
2354         upper_limit <<= bits;
2355
2356         res += 1LL << (bits-2);
2357         res += 1LL << (2*(bits-2));
2358         res += 1LL << (3*(bits-2));
2359         res <<= bits;
2360         if (res > upper_limit)
2361                 res = upper_limit;
2362
2363         if (res > MAX_LFS_FILESIZE)
2364                 res = MAX_LFS_FILESIZE;
2365
2366         return res;
2367 }
2368
2369 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2370                                    ext4_fsblk_t logical_sb_block, int nr)
2371 {
2372         struct ext4_sb_info *sbi = EXT4_SB(sb);
2373         ext4_group_t bg, first_meta_bg;
2374         int has_super = 0;
2375
2376         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2377
2378         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2379             nr < first_meta_bg)
2380                 return logical_sb_block + nr + 1;
2381         bg = sbi->s_desc_per_block * nr;
2382         if (ext4_bg_has_super(sb, bg))
2383                 has_super = 1;
2384
2385         /*
2386          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2387          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2388          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2389          * compensate.
2390          */
2391         if (sb->s_blocksize == 1024 && nr == 0 &&
2392             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2393                 has_super++;
2394
2395         return (has_super + ext4_group_first_block_no(sb, bg));
2396 }
2397
2398 /**
2399  * ext4_get_stripe_size: Get the stripe size.
2400  * @sbi: In memory super block info
2401  *
2402  * If we have specified it via mount option, then
2403  * use the mount option value. If the value specified at mount time is
2404  * greater than the blocks per group use the super block value.
2405  * If the super block value is greater than blocks per group return 0.
2406  * Allocator needs it be less than blocks per group.
2407  *
2408  */
2409 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2410 {
2411         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2412         unsigned long stripe_width =
2413                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2414         int ret;
2415
2416         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2417                 ret = sbi->s_stripe;
2418         else if (stripe_width <= sbi->s_blocks_per_group)
2419                 ret = stripe_width;
2420         else if (stride <= sbi->s_blocks_per_group)
2421                 ret = stride;
2422         else
2423                 ret = 0;
2424
2425         /*
2426          * If the stripe width is 1, this makes no sense and
2427          * we set it to 0 to turn off stripe handling code.
2428          */
2429         if (ret <= 1)
2430                 ret = 0;
2431
2432         return ret;
2433 }
2434
2435 /* sysfs supprt */
2436
2437 struct ext4_attr {
2438         struct attribute attr;
2439         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2440         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2441                          const char *, size_t);
2442         union {
2443                 int offset;
2444                 int deprecated_val;
2445         } u;
2446 };
2447
2448 static int parse_strtoull(const char *buf,
2449                 unsigned long long max, unsigned long long *value)
2450 {
2451         int ret;
2452
2453         ret = kstrtoull(skip_spaces(buf), 0, value);
2454         if (!ret && *value > max)
2455                 ret = -EINVAL;
2456         return ret;
2457 }
2458
2459 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2460                                               struct ext4_sb_info *sbi,
2461                                               char *buf)
2462 {
2463         return snprintf(buf, PAGE_SIZE, "%llu\n",
2464                 (s64) EXT4_C2B(sbi,
2465                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2466 }
2467
2468 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2469                                          struct ext4_sb_info *sbi, char *buf)
2470 {
2471         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2472
2473         if (!sb->s_bdev->bd_part)
2474                 return snprintf(buf, PAGE_SIZE, "0\n");
2475         return snprintf(buf, PAGE_SIZE, "%lu\n",
2476                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2477                          sbi->s_sectors_written_start) >> 1);
2478 }
2479
2480 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2481                                           struct ext4_sb_info *sbi, char *buf)
2482 {
2483         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2484
2485         if (!sb->s_bdev->bd_part)
2486                 return snprintf(buf, PAGE_SIZE, "0\n");
2487         return snprintf(buf, PAGE_SIZE, "%llu\n",
2488                         (unsigned long long)(sbi->s_kbytes_written +
2489                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2490                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2491 }
2492
2493 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2494                                           struct ext4_sb_info *sbi,
2495                                           const char *buf, size_t count)
2496 {
2497         unsigned long t;
2498         int ret;
2499
2500         ret = kstrtoul(skip_spaces(buf), 0, &t);
2501         if (ret)
2502                 return ret;
2503
2504         if (t && (!is_power_of_2(t) || t > 0x40000000))
2505                 return -EINVAL;
2506
2507         sbi->s_inode_readahead_blks = t;
2508         return count;
2509 }
2510
2511 static ssize_t sbi_ui_show(struct ext4_attr *a,
2512                            struct ext4_sb_info *sbi, char *buf)
2513 {
2514         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2515
2516         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2517 }
2518
2519 static ssize_t sbi_ui_store(struct ext4_attr *a,
2520                             struct ext4_sb_info *sbi,
2521                             const char *buf, size_t count)
2522 {
2523         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2524         unsigned long t;
2525         int ret;
2526
2527         ret = kstrtoul(skip_spaces(buf), 0, &t);
2528         if (ret)
2529                 return ret;
2530         *ui = t;
2531         return count;
2532 }
2533
2534 static ssize_t es_ui_show(struct ext4_attr *a,
2535                            struct ext4_sb_info *sbi, char *buf)
2536 {
2537
2538         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2539                            a->u.offset);
2540
2541         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2542 }
2543
2544 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2545                                   struct ext4_sb_info *sbi, char *buf)
2546 {
2547         return snprintf(buf, PAGE_SIZE, "%llu\n",
2548                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2549 }
2550
2551 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2552                                    struct ext4_sb_info *sbi,
2553                                    const char *buf, size_t count)
2554 {
2555         unsigned long long val;
2556         int ret;
2557
2558         if (parse_strtoull(buf, -1ULL, &val))
2559                 return -EINVAL;
2560         ret = ext4_reserve_clusters(sbi, val);
2561
2562         return ret ? ret : count;
2563 }
2564
2565 static ssize_t trigger_test_error(struct ext4_attr *a,
2566                                   struct ext4_sb_info *sbi,
2567                                   const char *buf, size_t count)
2568 {
2569         int len = count;
2570
2571         if (!capable(CAP_SYS_ADMIN))
2572                 return -EPERM;
2573
2574         if (len && buf[len-1] == '\n')
2575                 len--;
2576
2577         if (len)
2578                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2579         return count;
2580 }
2581
2582 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2583                                    struct ext4_sb_info *sbi, char *buf)
2584 {
2585         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2586 }
2587
2588 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2589 static struct ext4_attr ext4_attr_##_name = {                   \
2590         .attr = {.name = __stringify(_name), .mode = _mode },   \
2591         .show   = _show,                                        \
2592         .store  = _store,                                       \
2593         .u = {                                                  \
2594                 .offset = offsetof(struct ext4_sb_info, _elname),\
2595         },                                                      \
2596 }
2597
2598 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2599 static struct ext4_attr ext4_attr_##_name = {                           \
2600         .attr = {.name = __stringify(_name), .mode = _mode },           \
2601         .show   = _show,                                                \
2602         .store  = _store,                                               \
2603         .u = {                                                          \
2604                 .offset = offsetof(struct ext4_super_block, _elname),   \
2605         },                                                              \
2606 }
2607
2608 #define EXT4_ATTR(name, mode, show, store) \
2609 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2610
2611 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2612 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2613 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2614
2615 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2616         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2617 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2618         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2619
2620 #define ATTR_LIST(name) &ext4_attr_##name.attr
2621 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2622 static struct ext4_attr ext4_attr_##_name = {                   \
2623         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2624         .show   = sbi_deprecated_show,                          \
2625         .u = {                                                  \
2626                 .deprecated_val = _val,                         \
2627         },                                                      \
2628 }
2629
2630 EXT4_RO_ATTR(delayed_allocation_blocks);
2631 EXT4_RO_ATTR(session_write_kbytes);
2632 EXT4_RO_ATTR(lifetime_write_kbytes);
2633 EXT4_RW_ATTR(reserved_clusters);
2634 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2635                  inode_readahead_blks_store, s_inode_readahead_blks);
2636 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2637 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2638 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2639 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2640 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2641 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2642 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2643 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2644 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2645 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2646 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2647 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2648 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2649 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2650 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2651 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2652 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2653 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2654 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2655
2656 static struct attribute *ext4_attrs[] = {
2657         ATTR_LIST(delayed_allocation_blocks),
2658         ATTR_LIST(session_write_kbytes),
2659         ATTR_LIST(lifetime_write_kbytes),
2660         ATTR_LIST(reserved_clusters),
2661         ATTR_LIST(inode_readahead_blks),
2662         ATTR_LIST(inode_goal),
2663         ATTR_LIST(mb_stats),
2664         ATTR_LIST(mb_max_to_scan),
2665         ATTR_LIST(mb_min_to_scan),
2666         ATTR_LIST(mb_order2_req),
2667         ATTR_LIST(mb_stream_req),
2668         ATTR_LIST(mb_group_prealloc),
2669         ATTR_LIST(max_writeback_mb_bump),
2670         ATTR_LIST(extent_max_zeroout_kb),
2671         ATTR_LIST(trigger_fs_error),
2672         ATTR_LIST(err_ratelimit_interval_ms),
2673         ATTR_LIST(err_ratelimit_burst),
2674         ATTR_LIST(warning_ratelimit_interval_ms),
2675         ATTR_LIST(warning_ratelimit_burst),
2676         ATTR_LIST(msg_ratelimit_interval_ms),
2677         ATTR_LIST(msg_ratelimit_burst),
2678         ATTR_LIST(errors_count),
2679         ATTR_LIST(first_error_time),
2680         ATTR_LIST(last_error_time),
2681         NULL,
2682 };
2683
2684 /* Features this copy of ext4 supports */
2685 EXT4_INFO_ATTR(lazy_itable_init);
2686 EXT4_INFO_ATTR(batched_discard);
2687 EXT4_INFO_ATTR(meta_bg_resize);
2688
2689 static struct attribute *ext4_feat_attrs[] = {
2690         ATTR_LIST(lazy_itable_init),
2691         ATTR_LIST(batched_discard),
2692         ATTR_LIST(meta_bg_resize),
2693         NULL,
2694 };
2695
2696 static ssize_t ext4_attr_show(struct kobject *kobj,
2697                               struct attribute *attr, char *buf)
2698 {
2699         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2700                                                 s_kobj);
2701         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2702
2703         return a->show ? a->show(a, sbi, buf) : 0;
2704 }
2705
2706 static ssize_t ext4_attr_store(struct kobject *kobj,
2707                                struct attribute *attr,
2708                                const char *buf, size_t len)
2709 {
2710         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2711                                                 s_kobj);
2712         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2713
2714         return a->store ? a->store(a, sbi, buf, len) : 0;
2715 }
2716
2717 static void ext4_sb_release(struct kobject *kobj)
2718 {
2719         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2720                                                 s_kobj);
2721         complete(&sbi->s_kobj_unregister);
2722 }
2723
2724 static const struct sysfs_ops ext4_attr_ops = {
2725         .show   = ext4_attr_show,
2726         .store  = ext4_attr_store,
2727 };
2728
2729 static struct kobj_type ext4_ktype = {
2730         .default_attrs  = ext4_attrs,
2731         .sysfs_ops      = &ext4_attr_ops,
2732         .release        = ext4_sb_release,
2733 };
2734
2735 static void ext4_feat_release(struct kobject *kobj)
2736 {
2737         complete(&ext4_feat->f_kobj_unregister);
2738 }
2739
2740 static ssize_t ext4_feat_show(struct kobject *kobj,
2741                               struct attribute *attr, char *buf)
2742 {
2743         return snprintf(buf, PAGE_SIZE, "supported\n");
2744 }
2745
2746 /*
2747  * We can not use ext4_attr_show/store because it relies on the kobject
2748  * being embedded in the ext4_sb_info structure which is definitely not
2749  * true in this case.
2750  */
2751 static const struct sysfs_ops ext4_feat_ops = {
2752         .show   = ext4_feat_show,
2753         .store  = NULL,
2754 };
2755
2756 static struct kobj_type ext4_feat_ktype = {
2757         .default_attrs  = ext4_feat_attrs,
2758         .sysfs_ops      = &ext4_feat_ops,
2759         .release        = ext4_feat_release,
2760 };
2761
2762 /*
2763  * Check whether this filesystem can be mounted based on
2764  * the features present and the RDONLY/RDWR mount requested.
2765  * Returns 1 if this filesystem can be mounted as requested,
2766  * 0 if it cannot be.
2767  */
2768 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2769 {
2770         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2771                 ext4_msg(sb, KERN_ERR,
2772                         "Couldn't mount because of "
2773                         "unsupported optional features (%x)",
2774                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2775                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2776                 return 0;
2777         }
2778
2779         if (readonly)
2780                 return 1;
2781
2782         /* Check that feature set is OK for a read-write mount */
2783         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2784                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2785                          "unsupported optional features (%x)",
2786                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2787                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2788                 return 0;
2789         }
2790         /*
2791          * Large file size enabled file system can only be mounted
2792          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2793          */
2794         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2795                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2796                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2797                                  "cannot be mounted RDWR without "
2798                                  "CONFIG_LBDAF");
2799                         return 0;
2800                 }
2801         }
2802         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2803             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2804                 ext4_msg(sb, KERN_ERR,
2805                          "Can't support bigalloc feature without "
2806                          "extents feature\n");
2807                 return 0;
2808         }
2809
2810 #ifndef CONFIG_QUOTA
2811         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2812             !readonly) {
2813                 ext4_msg(sb, KERN_ERR,
2814                          "Filesystem with quota feature cannot be mounted RDWR "
2815                          "without CONFIG_QUOTA");
2816                 return 0;
2817         }
2818 #endif  /* CONFIG_QUOTA */
2819         return 1;
2820 }
2821
2822 /*
2823  * This function is called once a day if we have errors logged
2824  * on the file system
2825  */
2826 static void print_daily_error_info(unsigned long arg)
2827 {
2828         struct super_block *sb = (struct super_block *) arg;
2829         struct ext4_sb_info *sbi;
2830         struct ext4_super_block *es;
2831
2832         sbi = EXT4_SB(sb);
2833         es = sbi->s_es;
2834
2835         if (es->s_error_count)
2836                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2837                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2838                          le32_to_cpu(es->s_error_count));
2839         if (es->s_first_error_time) {
2840                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2841                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2842                        (int) sizeof(es->s_first_error_func),
2843                        es->s_first_error_func,
2844                        le32_to_cpu(es->s_first_error_line));
2845                 if (es->s_first_error_ino)
2846                         printk(": inode %u",
2847                                le32_to_cpu(es->s_first_error_ino));
2848                 if (es->s_first_error_block)
2849                         printk(": block %llu", (unsigned long long)
2850                                le64_to_cpu(es->s_first_error_block));
2851                 printk("\n");
2852         }
2853         if (es->s_last_error_time) {
2854                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2855                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2856                        (int) sizeof(es->s_last_error_func),
2857                        es->s_last_error_func,
2858                        le32_to_cpu(es->s_last_error_line));
2859                 if (es->s_last_error_ino)
2860                         printk(": inode %u",
2861                                le32_to_cpu(es->s_last_error_ino));
2862                 if (es->s_last_error_block)
2863                         printk(": block %llu", (unsigned long long)
2864                                le64_to_cpu(es->s_last_error_block));
2865                 printk("\n");
2866         }
2867         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2868 }
2869
2870 /* Find next suitable group and run ext4_init_inode_table */
2871 static int ext4_run_li_request(struct ext4_li_request *elr)
2872 {
2873         struct ext4_group_desc *gdp = NULL;
2874         ext4_group_t group, ngroups;
2875         struct super_block *sb;
2876         unsigned long timeout = 0;
2877         int ret = 0;
2878
2879         sb = elr->lr_super;
2880         ngroups = EXT4_SB(sb)->s_groups_count;
2881
2882         sb_start_write(sb);
2883         for (group = elr->lr_next_group; group < ngroups; group++) {
2884                 gdp = ext4_get_group_desc(sb, group, NULL);
2885                 if (!gdp) {
2886                         ret = 1;
2887                         break;
2888                 }
2889
2890                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2891                         break;
2892         }
2893
2894         if (group >= ngroups)
2895                 ret = 1;
2896
2897         if (!ret) {
2898                 timeout = jiffies;
2899                 ret = ext4_init_inode_table(sb, group,
2900                                             elr->lr_timeout ? 0 : 1);
2901                 if (elr->lr_timeout == 0) {
2902                         timeout = (jiffies - timeout) *
2903                                   elr->lr_sbi->s_li_wait_mult;
2904                         elr->lr_timeout = timeout;
2905                 }
2906                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2907                 elr->lr_next_group = group + 1;
2908         }
2909         sb_end_write(sb);
2910
2911         return ret;
2912 }
2913
2914 /*
2915  * Remove lr_request from the list_request and free the
2916  * request structure. Should be called with li_list_mtx held
2917  */
2918 static void ext4_remove_li_request(struct ext4_li_request *elr)
2919 {
2920         struct ext4_sb_info *sbi;
2921
2922         if (!elr)
2923                 return;
2924
2925         sbi = elr->lr_sbi;
2926
2927         list_del(&elr->lr_request);
2928         sbi->s_li_request = NULL;
2929         kfree(elr);
2930 }
2931
2932 static void ext4_unregister_li_request(struct super_block *sb)
2933 {
2934         mutex_lock(&ext4_li_mtx);
2935         if (!ext4_li_info) {
2936                 mutex_unlock(&ext4_li_mtx);
2937                 return;
2938         }
2939
2940         mutex_lock(&ext4_li_info->li_list_mtx);
2941         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2942         mutex_unlock(&ext4_li_info->li_list_mtx);
2943         mutex_unlock(&ext4_li_mtx);
2944 }
2945
2946 static struct task_struct *ext4_lazyinit_task;
2947
2948 /*
2949  * This is the function where ext4lazyinit thread lives. It walks
2950  * through the request list searching for next scheduled filesystem.
2951  * When such a fs is found, run the lazy initialization request
2952  * (ext4_rn_li_request) and keep track of the time spend in this
2953  * function. Based on that time we compute next schedule time of
2954  * the request. When walking through the list is complete, compute
2955  * next waking time and put itself into sleep.
2956  */
2957 static int ext4_lazyinit_thread(void *arg)
2958 {
2959         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2960         struct list_head *pos, *n;
2961         struct ext4_li_request *elr;
2962         unsigned long next_wakeup, cur;
2963
2964         BUG_ON(NULL == eli);
2965
2966 cont_thread:
2967         while (true) {
2968                 next_wakeup = MAX_JIFFY_OFFSET;
2969
2970                 mutex_lock(&eli->li_list_mtx);
2971                 if (list_empty(&eli->li_request_list)) {
2972                         mutex_unlock(&eli->li_list_mtx);
2973                         goto exit_thread;
2974                 }
2975
2976                 list_for_each_safe(pos, n, &eli->li_request_list) {
2977                         elr = list_entry(pos, struct ext4_li_request,
2978                                          lr_request);
2979
2980                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2981                                 if (ext4_run_li_request(elr) != 0) {
2982                                         /* error, remove the lazy_init job */
2983                                         ext4_remove_li_request(elr);
2984                                         continue;
2985                                 }
2986                         }
2987
2988                         if (time_before(elr->lr_next_sched, next_wakeup))
2989                                 next_wakeup = elr->lr_next_sched;
2990                 }
2991                 mutex_unlock(&eli->li_list_mtx);
2992
2993                 try_to_freeze();
2994
2995                 cur = jiffies;
2996                 if ((time_after_eq(cur, next_wakeup)) ||
2997                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2998                         cond_resched();
2999                         continue;
3000                 }
3001
3002                 schedule_timeout_interruptible(next_wakeup - cur);
3003
3004                 if (kthread_should_stop()) {
3005                         ext4_clear_request_list();
3006                         goto exit_thread;
3007                 }
3008         }
3009
3010 exit_thread:
3011         /*
3012          * It looks like the request list is empty, but we need
3013          * to check it under the li_list_mtx lock, to prevent any
3014          * additions into it, and of course we should lock ext4_li_mtx
3015          * to atomically free the list and ext4_li_info, because at
3016          * this point another ext4 filesystem could be registering
3017          * new one.
3018          */
3019         mutex_lock(&ext4_li_mtx);
3020         mutex_lock(&eli->li_list_mtx);
3021         if (!list_empty(&eli->li_request_list)) {
3022                 mutex_unlock(&eli->li_list_mtx);
3023                 mutex_unlock(&ext4_li_mtx);
3024                 goto cont_thread;
3025         }
3026         mutex_unlock(&eli->li_list_mtx);
3027         kfree(ext4_li_info);
3028         ext4_li_info = NULL;
3029         mutex_unlock(&ext4_li_mtx);
3030
3031         return 0;
3032 }
3033
3034 static void ext4_clear_request_list(void)
3035 {
3036         struct list_head *pos, *n;
3037         struct ext4_li_request *elr;
3038
3039         mutex_lock(&ext4_li_info->li_list_mtx);
3040         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3041                 elr = list_entry(pos, struct ext4_li_request,
3042                                  lr_request);
3043                 ext4_remove_li_request(elr);
3044         }
3045         mutex_unlock(&ext4_li_info->li_list_mtx);
3046 }
3047
3048 static int ext4_run_lazyinit_thread(void)
3049 {
3050         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3051                                          ext4_li_info, "ext4lazyinit");
3052         if (IS_ERR(ext4_lazyinit_task)) {
3053                 int err = PTR_ERR(ext4_lazyinit_task);
3054                 ext4_clear_request_list();
3055                 kfree(ext4_li_info);
3056                 ext4_li_info = NULL;
3057                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3058                                  "initialization thread\n",
3059                                  err);
3060                 return err;
3061         }
3062         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3063         return 0;
3064 }
3065
3066 /*
3067  * Check whether it make sense to run itable init. thread or not.
3068  * If there is at least one uninitialized inode table, return
3069  * corresponding group number, else the loop goes through all
3070  * groups and return total number of groups.
3071  */
3072 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3073 {
3074         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3075         struct ext4_group_desc *gdp = NULL;
3076
3077         for (group = 0; group < ngroups; group++) {
3078                 gdp = ext4_get_group_desc(sb, group, NULL);
3079                 if (!gdp)
3080                         continue;
3081
3082                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3083                         break;
3084         }
3085
3086         return group;
3087 }
3088
3089 static int ext4_li_info_new(void)
3090 {
3091         struct ext4_lazy_init *eli = NULL;
3092
3093         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3094         if (!eli)
3095                 return -ENOMEM;
3096
3097         INIT_LIST_HEAD(&eli->li_request_list);
3098         mutex_init(&eli->li_list_mtx);
3099
3100         eli->li_state |= EXT4_LAZYINIT_QUIT;
3101
3102         ext4_li_info = eli;
3103
3104         return 0;
3105 }
3106
3107 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3108                                             ext4_group_t start)
3109 {
3110         struct ext4_sb_info *sbi = EXT4_SB(sb);
3111         struct ext4_li_request *elr;
3112
3113         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3114         if (!elr)
3115                 return NULL;
3116
3117         elr->lr_super = sb;
3118         elr->lr_sbi = sbi;
3119         elr->lr_next_group = start;
3120
3121         /*
3122          * Randomize first schedule time of the request to
3123          * spread the inode table initialization requests
3124          * better.
3125          */
3126         elr->lr_next_sched = jiffies + (prandom_u32() %
3127                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3128         return elr;
3129 }
3130
3131 int ext4_register_li_request(struct super_block *sb,
3132                              ext4_group_t first_not_zeroed)
3133 {
3134         struct ext4_sb_info *sbi = EXT4_SB(sb);
3135         struct ext4_li_request *elr = NULL;
3136         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3137         int ret = 0;
3138
3139         mutex_lock(&ext4_li_mtx);
3140         if (sbi->s_li_request != NULL) {
3141                 /*
3142                  * Reset timeout so it can be computed again, because
3143                  * s_li_wait_mult might have changed.
3144                  */
3145                 sbi->s_li_request->lr_timeout = 0;
3146                 goto out;
3147         }
3148
3149         if (first_not_zeroed == ngroups ||
3150             (sb->s_flags & MS_RDONLY) ||
3151             !test_opt(sb, INIT_INODE_TABLE))
3152                 goto out;
3153
3154         elr = ext4_li_request_new(sb, first_not_zeroed);
3155         if (!elr) {
3156                 ret = -ENOMEM;
3157                 goto out;
3158         }
3159
3160         if (NULL == ext4_li_info) {
3161                 ret = ext4_li_info_new();
3162                 if (ret)
3163                         goto out;
3164         }
3165
3166         mutex_lock(&ext4_li_info->li_list_mtx);
3167         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3168         mutex_unlock(&ext4_li_info->li_list_mtx);
3169
3170         sbi->s_li_request = elr;
3171         /*
3172          * set elr to NULL here since it has been inserted to
3173          * the request_list and the removal and free of it is
3174          * handled by ext4_clear_request_list from now on.
3175          */
3176         elr = NULL;
3177
3178         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3179                 ret = ext4_run_lazyinit_thread();
3180                 if (ret)
3181                         goto out;
3182         }
3183 out:
3184         mutex_unlock(&ext4_li_mtx);
3185         if (ret)
3186                 kfree(elr);
3187         return ret;
3188 }
3189
3190 /*
3191  * We do not need to lock anything since this is called on
3192  * module unload.
3193  */
3194 static void ext4_destroy_lazyinit_thread(void)
3195 {
3196         /*
3197          * If thread exited earlier
3198          * there's nothing to be done.
3199          */
3200         if (!ext4_li_info || !ext4_lazyinit_task)
3201                 return;
3202
3203         kthread_stop(ext4_lazyinit_task);
3204 }
3205
3206 static int set_journal_csum_feature_set(struct super_block *sb)
3207 {
3208         int ret = 1;
3209         int compat, incompat;
3210         struct ext4_sb_info *sbi = EXT4_SB(sb);
3211
3212         if (ext4_has_metadata_csum(sb)) {
3213                 /* journal checksum v3 */
3214                 compat = 0;
3215                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3216         } else {
3217                 /* journal checksum v1 */
3218                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3219                 incompat = 0;
3220         }
3221
3222         jbd2_journal_clear_features(sbi->s_journal,
3223                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3224                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3225                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3226         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3227                 ret = jbd2_journal_set_features(sbi->s_journal,
3228                                 compat, 0,
3229                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3230                                 incompat);
3231         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3232                 ret = jbd2_journal_set_features(sbi->s_journal,
3233                                 compat, 0,
3234                                 incompat);
3235                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3236                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3237         } else {
3238                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3239                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3240         }
3241
3242         return ret;
3243 }
3244
3245 /*
3246  * Note: calculating the overhead so we can be compatible with
3247  * historical BSD practice is quite difficult in the face of
3248  * clusters/bigalloc.  This is because multiple metadata blocks from
3249  * different block group can end up in the same allocation cluster.
3250  * Calculating the exact overhead in the face of clustered allocation
3251  * requires either O(all block bitmaps) in memory or O(number of block
3252  * groups**2) in time.  We will still calculate the superblock for
3253  * older file systems --- and if we come across with a bigalloc file
3254  * system with zero in s_overhead_clusters the estimate will be close to
3255  * correct especially for very large cluster sizes --- but for newer
3256  * file systems, it's better to calculate this figure once at mkfs
3257  * time, and store it in the superblock.  If the superblock value is
3258  * present (even for non-bigalloc file systems), we will use it.
3259  */
3260 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3261                           char *buf)
3262 {
3263         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3264         struct ext4_group_desc  *gdp;
3265         ext4_fsblk_t            first_block, last_block, b;
3266         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3267         int                     s, j, count = 0;
3268
3269         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3270                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3271                         sbi->s_itb_per_group + 2);
3272
3273         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3274                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3275         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3276         for (i = 0; i < ngroups; i++) {
3277                 gdp = ext4_get_group_desc(sb, i, NULL);
3278                 b = ext4_block_bitmap(sb, gdp);
3279                 if (b >= first_block && b <= last_block) {
3280                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3281                         count++;
3282                 }
3283                 b = ext4_inode_bitmap(sb, gdp);
3284                 if (b >= first_block && b <= last_block) {
3285                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3286                         count++;
3287                 }
3288                 b = ext4_inode_table(sb, gdp);
3289                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3290                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3291                                 int c = EXT4_B2C(sbi, b - first_block);
3292                                 ext4_set_bit(c, buf);
3293                                 count++;
3294                         }
3295                 if (i != grp)
3296                         continue;
3297                 s = 0;
3298                 if (ext4_bg_has_super(sb, grp)) {
3299                         ext4_set_bit(s++, buf);
3300                         count++;
3301                 }
3302                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3303                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3304                         count++;
3305                 }
3306         }
3307         if (!count)
3308                 return 0;
3309         return EXT4_CLUSTERS_PER_GROUP(sb) -
3310                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3311 }
3312
3313 /*
3314  * Compute the overhead and stash it in sbi->s_overhead
3315  */
3316 int ext4_calculate_overhead(struct super_block *sb)
3317 {
3318         struct ext4_sb_info *sbi = EXT4_SB(sb);
3319         struct ext4_super_block *es = sbi->s_es;
3320         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3321         ext4_fsblk_t overhead = 0;
3322         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3323
3324         if (!buf)
3325                 return -ENOMEM;
3326
3327         /*
3328          * Compute the overhead (FS structures).  This is constant
3329          * for a given filesystem unless the number of block groups
3330          * changes so we cache the previous value until it does.
3331          */
3332
3333         /*
3334          * All of the blocks before first_data_block are overhead
3335          */
3336         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3337
3338         /*
3339          * Add the overhead found in each block group
3340          */
3341         for (i = 0; i < ngroups; i++) {
3342                 int blks;
3343
3344                 blks = count_overhead(sb, i, buf);
3345                 overhead += blks;
3346                 if (blks)
3347                         memset(buf, 0, PAGE_SIZE);
3348                 cond_resched();
3349         }
3350         /* Add the internal journal blocks as well */
3351         if (sbi->s_journal && !sbi->journal_bdev)
3352                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3353
3354         sbi->s_overhead = overhead;
3355         smp_wmb();
3356         free_page((unsigned long) buf);
3357         return 0;
3358 }
3359
3360
3361 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3362 {
3363         ext4_fsblk_t resv_clusters;
3364
3365         /*
3366          * There's no need to reserve anything when we aren't using extents.
3367          * The space estimates are exact, there are no unwritten extents,
3368          * hole punching doesn't need new metadata... This is needed especially
3369          * to keep ext2/3 backward compatibility.
3370          */
3371         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3372                 return 0;
3373         /*
3374          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3375          * This should cover the situations where we can not afford to run
3376          * out of space like for example punch hole, or converting
3377          * unwritten extents in delalloc path. In most cases such
3378          * allocation would require 1, or 2 blocks, higher numbers are
3379          * very rare.
3380          */
3381         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3382                         EXT4_SB(sb)->s_cluster_bits;
3383
3384         do_div(resv_clusters, 50);
3385         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3386
3387         return resv_clusters;
3388 }
3389
3390
3391 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3392 {
3393         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3394                                 sbi->s_cluster_bits;
3395
3396         if (count >= clusters)
3397                 return -EINVAL;
3398
3399         atomic64_set(&sbi->s_resv_clusters, count);
3400         return 0;
3401 }
3402
3403 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3404 {
3405         char *orig_data = kstrdup(data, GFP_KERNEL);
3406         struct buffer_head *bh;
3407         struct ext4_super_block *es = NULL;
3408         struct ext4_sb_info *sbi;
3409         ext4_fsblk_t block;
3410         ext4_fsblk_t sb_block = get_sb_block(&data);
3411         ext4_fsblk_t logical_sb_block;
3412         unsigned long offset = 0;
3413         unsigned long journal_devnum = 0;
3414         unsigned long def_mount_opts;
3415         struct inode *root;
3416         char *cp;
3417         const char *descr;
3418         int ret = -ENOMEM;
3419         int blocksize, clustersize;
3420         unsigned int db_count;
3421         unsigned int i;
3422         int needs_recovery, has_huge_files, has_bigalloc;
3423         __u64 blocks_count;
3424         int err = 0;
3425         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3426         ext4_group_t first_not_zeroed;
3427
3428         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3429         if (!sbi)
3430                 goto out_free_orig;
3431
3432         sbi->s_blockgroup_lock =
3433                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3434         if (!sbi->s_blockgroup_lock) {
3435                 kfree(sbi);
3436                 goto out_free_orig;
3437         }
3438         sb->s_fs_info = sbi;
3439         sbi->s_sb = sb;
3440         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3441         sbi->s_sb_block = sb_block;
3442         if (sb->s_bdev->bd_part)
3443                 sbi->s_sectors_written_start =
3444                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3445
3446         /* Cleanup superblock name */
3447         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3448                 *cp = '!';
3449
3450         /* -EINVAL is default */
3451         ret = -EINVAL;
3452         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3453         if (!blocksize) {
3454                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3455                 goto out_fail;
3456         }
3457
3458         /*
3459          * The ext4 superblock will not be buffer aligned for other than 1kB
3460          * block sizes.  We need to calculate the offset from buffer start.
3461          */
3462         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3463                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3464                 offset = do_div(logical_sb_block, blocksize);
3465         } else {
3466                 logical_sb_block = sb_block;
3467         }
3468
3469         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3470                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3471                 goto out_fail;
3472         }
3473         /*
3474          * Note: s_es must be initialized as soon as possible because
3475          *       some ext4 macro-instructions depend on its value
3476          */
3477         es = (struct ext4_super_block *) (bh->b_data + offset);
3478         sbi->s_es = es;
3479         sb->s_magic = le16_to_cpu(es->s_magic);
3480         if (sb->s_magic != EXT4_SUPER_MAGIC)
3481                 goto cantfind_ext4;
3482         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3483
3484         /* Warn if metadata_csum and gdt_csum are both set. */
3485         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3486                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3487             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3488                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3489                              "redundant flags; please run fsck.");
3490
3491         /* Check for a known checksum algorithm */
3492         if (!ext4_verify_csum_type(sb, es)) {
3493                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3494                          "unknown checksum algorithm.");
3495                 silent = 1;
3496                 goto cantfind_ext4;
3497         }
3498
3499         /* Load the checksum driver */
3500         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3501                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3502                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3503                 if (IS_ERR(sbi->s_chksum_driver)) {
3504                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3505                         ret = PTR_ERR(sbi->s_chksum_driver);
3506                         sbi->s_chksum_driver = NULL;
3507                         goto failed_mount;
3508                 }
3509         }
3510
3511         /* Check superblock checksum */
3512         if (!ext4_superblock_csum_verify(sb, es)) {
3513                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3514                          "invalid superblock checksum.  Run e2fsck?");
3515                 silent = 1;
3516                 goto cantfind_ext4;
3517         }
3518
3519         /* Precompute checksum seed for all metadata */
3520         if (ext4_has_metadata_csum(sb))
3521                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3522                                                sizeof(es->s_uuid));
3523
3524         /* Set defaults before we parse the mount options */
3525         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3526         set_opt(sb, INIT_INODE_TABLE);
3527         if (def_mount_opts & EXT4_DEFM_DEBUG)
3528                 set_opt(sb, DEBUG);
3529         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3530                 set_opt(sb, GRPID);
3531         if (def_mount_opts & EXT4_DEFM_UID16)
3532                 set_opt(sb, NO_UID32);
3533         /* xattr user namespace & acls are now defaulted on */
3534         set_opt(sb, XATTR_USER);
3535 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3536         set_opt(sb, POSIX_ACL);
3537 #endif
3538         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3539         if (ext4_has_metadata_csum(sb))
3540                 set_opt(sb, JOURNAL_CHECKSUM);
3541
3542         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3543                 set_opt(sb, JOURNAL_DATA);
3544         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3545                 set_opt(sb, ORDERED_DATA);
3546         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3547                 set_opt(sb, WRITEBACK_DATA);
3548
3549         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3550                 set_opt(sb, ERRORS_PANIC);
3551         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3552                 set_opt(sb, ERRORS_CONT);
3553         else
3554                 set_opt(sb, ERRORS_RO);
3555         /* block_validity enabled by default; disable with noblock_validity */
3556         set_opt(sb, BLOCK_VALIDITY);
3557         if (def_mount_opts & EXT4_DEFM_DISCARD)
3558                 set_opt(sb, DISCARD);
3559
3560         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3561         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3562         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3563         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3564         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3565
3566         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3567                 set_opt(sb, BARRIER);
3568
3569         /*
3570          * enable delayed allocation by default
3571          * Use -o nodelalloc to turn it off
3572          */
3573         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3574             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3575                 set_opt(sb, DELALLOC);
3576
3577         /*
3578          * set default s_li_wait_mult for lazyinit, for the case there is
3579          * no mount option specified.
3580          */
3581         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3582
3583         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3584                            &journal_devnum, &journal_ioprio, 0)) {
3585                 ext4_msg(sb, KERN_WARNING,
3586                          "failed to parse options in superblock: %s",
3587                          sbi->s_es->s_mount_opts);
3588         }
3589         sbi->s_def_mount_opt = sbi->s_mount_opt;
3590         if (!parse_options((char *) data, sb, &journal_devnum,
3591                            &journal_ioprio, 0))
3592                 goto failed_mount;
3593
3594         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3595                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3596                             "with data=journal disables delayed "
3597                             "allocation and O_DIRECT support!\n");
3598                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3599                         ext4_msg(sb, KERN_ERR, "can't mount with "
3600                                  "both data=journal and delalloc");
3601                         goto failed_mount;
3602                 }
3603                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3604                         ext4_msg(sb, KERN_ERR, "can't mount with "
3605                                  "both data=journal and dioread_nolock");
3606                         goto failed_mount;
3607                 }
3608                 if (test_opt(sb, DAX)) {
3609                         ext4_msg(sb, KERN_ERR, "can't mount with "
3610                                  "both data=journal and dax");
3611                         goto failed_mount;
3612                 }
3613                 if (test_opt(sb, DELALLOC))
3614                         clear_opt(sb, DELALLOC);
3615         }
3616
3617         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3618                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3619
3620         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3621             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3622              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3623              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3624                 ext4_msg(sb, KERN_WARNING,
3625                        "feature flags set on rev 0 fs, "
3626                        "running e2fsck is recommended");
3627
3628         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3629                 set_opt2(sb, HURD_COMPAT);
3630                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3631                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3632                         ext4_msg(sb, KERN_ERR,
3633                                  "The Hurd can't support 64-bit file systems");
3634                         goto failed_mount;
3635                 }
3636         }
3637
3638         if (IS_EXT2_SB(sb)) {
3639                 if (ext2_feature_set_ok(sb))
3640                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3641                                  "using the ext4 subsystem");
3642                 else {
3643                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3644                                  "to feature incompatibilities");
3645                         goto failed_mount;
3646                 }
3647         }
3648
3649         if (IS_EXT3_SB(sb)) {
3650                 if (ext3_feature_set_ok(sb))
3651                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3652                                  "using the ext4 subsystem");
3653                 else {
3654                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3655                                  "to feature incompatibilities");
3656                         goto failed_mount;
3657                 }
3658         }
3659
3660         /*
3661          * Check feature flags regardless of the revision level, since we
3662          * previously didn't change the revision level when setting the flags,
3663          * so there is a chance incompat flags are set on a rev 0 filesystem.
3664          */
3665         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3666                 goto failed_mount;
3667
3668         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3669         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3670             blocksize > EXT4_MAX_BLOCK_SIZE) {
3671                 ext4_msg(sb, KERN_ERR,
3672                        "Unsupported filesystem blocksize %d", blocksize);
3673                 goto failed_mount;
3674         }
3675
3676         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3677                 if (blocksize != PAGE_SIZE) {
3678                         ext4_msg(sb, KERN_ERR,
3679                                         "error: unsupported blocksize for dax");
3680                         goto failed_mount;
3681                 }
3682                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3683                         ext4_msg(sb, KERN_ERR,
3684                                         "error: device does not support dax");
3685                         goto failed_mount;
3686                 }
3687         }
3688
3689         if (sb->s_blocksize != blocksize) {
3690                 /* Validate the filesystem blocksize */
3691                 if (!sb_set_blocksize(sb, blocksize)) {
3692                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3693                                         blocksize);
3694                         goto failed_mount;
3695                 }
3696
3697                 brelse(bh);
3698                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3699                 offset = do_div(logical_sb_block, blocksize);
3700                 bh = sb_bread_unmovable(sb, logical_sb_block);
3701                 if (!bh) {
3702                         ext4_msg(sb, KERN_ERR,
3703                                "Can't read superblock on 2nd try");
3704                         goto failed_mount;
3705                 }
3706                 es = (struct ext4_super_block *)(bh->b_data + offset);
3707                 sbi->s_es = es;
3708                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3709                         ext4_msg(sb, KERN_ERR,
3710                                "Magic mismatch, very weird!");
3711                         goto failed_mount;
3712                 }
3713         }
3714
3715         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3716                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3717         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3718                                                       has_huge_files);
3719         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3720
3721         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3722                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3723                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3724         } else {
3725                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3726                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3727                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3728                     (!is_power_of_2(sbi->s_inode_size)) ||
3729                     (sbi->s_inode_size > blocksize)) {
3730                         ext4_msg(sb, KERN_ERR,
3731                                "unsupported inode size: %d",
3732                                sbi->s_inode_size);
3733                         goto failed_mount;
3734                 }
3735                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3736                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3737         }
3738
3739         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3740         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3741                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3742                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3743                     !is_power_of_2(sbi->s_desc_size)) {
3744                         ext4_msg(sb, KERN_ERR,
3745                                "unsupported descriptor size %lu",
3746                                sbi->s_desc_size);
3747                         goto failed_mount;
3748                 }
3749         } else
3750                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3751
3752         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3753         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3754         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3755                 goto cantfind_ext4;
3756
3757         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3758         if (sbi->s_inodes_per_block == 0)
3759                 goto cantfind_ext4;
3760         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3761                                         sbi->s_inodes_per_block;
3762         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3763         sbi->s_sbh = bh;
3764         sbi->s_mount_state = le16_to_cpu(es->s_state);
3765         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3766         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3767
3768         for (i = 0; i < 4; i++)
3769                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3770         sbi->s_def_hash_version = es->s_def_hash_version;
3771         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3772                 i = le32_to_cpu(es->s_flags);
3773                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3774                         sbi->s_hash_unsigned = 3;
3775                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3776 #ifdef __CHAR_UNSIGNED__
3777                         if (!(sb->s_flags & MS_RDONLY))
3778                                 es->s_flags |=
3779                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3780                         sbi->s_hash_unsigned = 3;
3781 #else
3782                         if (!(sb->s_flags & MS_RDONLY))
3783                                 es->s_flags |=
3784                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3785 #endif
3786                 }
3787         }
3788
3789         /* Handle clustersize */
3790         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3791         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3792                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3793         if (has_bigalloc) {
3794                 if (clustersize < blocksize) {
3795                         ext4_msg(sb, KERN_ERR,
3796                                  "cluster size (%d) smaller than "
3797                                  "block size (%d)", clustersize, blocksize);
3798                         goto failed_mount;
3799                 }
3800                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3801                         le32_to_cpu(es->s_log_block_size);
3802                 sbi->s_clusters_per_group =
3803                         le32_to_cpu(es->s_clusters_per_group);
3804                 if (sbi->s_clusters_per_group > blocksize * 8) {
3805                         ext4_msg(sb, KERN_ERR,
3806                                  "#clusters per group too big: %lu",
3807                                  sbi->s_clusters_per_group);
3808                         goto failed_mount;
3809                 }
3810                 if (sbi->s_blocks_per_group !=
3811                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3812                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3813                                  "clusters per group (%lu) inconsistent",
3814                                  sbi->s_blocks_per_group,
3815                                  sbi->s_clusters_per_group);
3816                         goto failed_mount;
3817                 }
3818         } else {
3819                 if (clustersize != blocksize) {
3820                         ext4_warning(sb, "fragment/cluster size (%d) != "
3821                                      "block size (%d)", clustersize,
3822                                      blocksize);
3823                         clustersize = blocksize;
3824                 }
3825                 if (sbi->s_blocks_per_group > blocksize * 8) {
3826                         ext4_msg(sb, KERN_ERR,
3827                                  "#blocks per group too big: %lu",
3828                                  sbi->s_blocks_per_group);
3829                         goto failed_mount;
3830                 }
3831                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3832                 sbi->s_cluster_bits = 0;
3833         }
3834         sbi->s_cluster_ratio = clustersize / blocksize;
3835
3836         if (sbi->s_inodes_per_group > blocksize * 8) {
3837                 ext4_msg(sb, KERN_ERR,
3838                        "#inodes per group too big: %lu",
3839                        sbi->s_inodes_per_group);
3840                 goto failed_mount;
3841         }
3842
3843         /* Do we have standard group size of clustersize * 8 blocks ? */
3844         if (sbi->s_blocks_per_group == clustersize << 3)
3845                 set_opt2(sb, STD_GROUP_SIZE);
3846
3847         /*
3848          * Test whether we have more sectors than will fit in sector_t,
3849          * and whether the max offset is addressable by the page cache.
3850          */
3851         err = generic_check_addressable(sb->s_blocksize_bits,
3852                                         ext4_blocks_count(es));
3853         if (err) {
3854                 ext4_msg(sb, KERN_ERR, "filesystem"
3855                          " too large to mount safely on this system");
3856                 if (sizeof(sector_t) < 8)
3857                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3858                 goto failed_mount;
3859         }
3860
3861         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3862                 goto cantfind_ext4;
3863
3864         /* check blocks count against device size */
3865         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3866         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3867                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3868                        "exceeds size of device (%llu blocks)",
3869                        ext4_blocks_count(es), blocks_count);
3870                 goto failed_mount;
3871         }
3872
3873         /*
3874          * It makes no sense for the first data block to be beyond the end
3875          * of the filesystem.
3876          */
3877         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3878                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3879                          "block %u is beyond end of filesystem (%llu)",
3880                          le32_to_cpu(es->s_first_data_block),
3881                          ext4_blocks_count(es));
3882                 goto failed_mount;
3883         }
3884         blocks_count = (ext4_blocks_count(es) -
3885                         le32_to_cpu(es->s_first_data_block) +
3886                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3887         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3888         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3889                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3890                        "(block count %llu, first data block %u, "
3891                        "blocks per group %lu)", sbi->s_groups_count,
3892                        ext4_blocks_count(es),
3893                        le32_to_cpu(es->s_first_data_block),
3894                        EXT4_BLOCKS_PER_GROUP(sb));
3895                 goto failed_mount;
3896         }
3897         sbi->s_groups_count = blocks_count;
3898         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3899                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3900         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3901                    EXT4_DESC_PER_BLOCK(sb);
3902         sbi->s_group_desc = ext4_kvmalloc(db_count *
3903                                           sizeof(struct buffer_head *),
3904                                           GFP_KERNEL);
3905         if (sbi->s_group_desc == NULL) {
3906                 ext4_msg(sb, KERN_ERR, "not enough memory");
3907                 ret = -ENOMEM;
3908                 goto failed_mount;
3909         }
3910
3911         if (ext4_proc_root)
3912                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3913
3914         if (sbi->s_proc)
3915                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3916                                  &ext4_seq_options_fops, sb);
3917
3918         bgl_lock_init(sbi->s_blockgroup_lock);
3919
3920         for (i = 0; i < db_count; i++) {
3921                 block = descriptor_loc(sb, logical_sb_block, i);
3922                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3923                 if (!sbi->s_group_desc[i]) {
3924                         ext4_msg(sb, KERN_ERR,
3925                                "can't read group descriptor %d", i);
3926                         db_count = i;
3927                         goto failed_mount2;
3928                 }
3929         }
3930         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3931                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3932                 goto failed_mount2;
3933         }
3934
3935         sbi->s_gdb_count = db_count;
3936         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3937         spin_lock_init(&sbi->s_next_gen_lock);
3938
3939         init_timer(&sbi->s_err_report);
3940         sbi->s_err_report.function = print_daily_error_info;
3941         sbi->s_err_report.data = (unsigned long) sb;
3942
3943         /* Register extent status tree shrinker */
3944         if (ext4_es_register_shrinker(sbi))
3945                 goto failed_mount3;
3946
3947         sbi->s_stripe = ext4_get_stripe_size(sbi);
3948         sbi->s_extent_max_zeroout_kb = 32;
3949
3950         /*
3951          * set up enough so that it can read an inode
3952          */
3953         sb->s_op = &ext4_sops;
3954         sb->s_export_op = &ext4_export_ops;
3955         sb->s_xattr = ext4_xattr_handlers;
3956 #ifdef CONFIG_QUOTA
3957         sb->dq_op = &ext4_quota_operations;
3958         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3959                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3960         else
3961                 sb->s_qcop = &ext4_qctl_operations;
3962         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3963 #endif
3964         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3965
3966         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3967         mutex_init(&sbi->s_orphan_lock);
3968
3969         sb->s_root = NULL;
3970
3971         needs_recovery = (es->s_last_orphan != 0 ||
3972                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3973                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3974
3975         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3976             !(sb->s_flags & MS_RDONLY))
3977                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3978                         goto failed_mount3a;
3979
3980         /*
3981          * The first inode we look at is the journal inode.  Don't try
3982          * root first: it may be modified in the journal!
3983          */
3984         if (!test_opt(sb, NOLOAD) &&
3985             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3986                 if (ext4_load_journal(sb, es, journal_devnum))
3987                         goto failed_mount3a;
3988         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3989               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3990                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3991                        "suppressed and not mounted read-only");
3992                 goto failed_mount_wq;
3993         } else {
3994                 clear_opt(sb, DATA_FLAGS);
3995                 sbi->s_journal = NULL;
3996                 needs_recovery = 0;
3997                 goto no_journal;
3998         }
3999
4000         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
4001             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4002                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4003                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4004                 goto failed_mount_wq;
4005         }
4006
4007         if (!set_journal_csum_feature_set(sb)) {
4008                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4009                          "feature set");
4010                 goto failed_mount_wq;
4011         }
4012
4013         /* We have now updated the journal if required, so we can
4014          * validate the data journaling mode. */
4015         switch (test_opt(sb, DATA_FLAGS)) {
4016         case 0:
4017                 /* No mode set, assume a default based on the journal
4018                  * capabilities: ORDERED_DATA if the journal can
4019                  * cope, else JOURNAL_DATA
4020                  */
4021                 if (jbd2_journal_check_available_features
4022                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4023                         set_opt(sb, ORDERED_DATA);
4024                 else
4025                         set_opt(sb, JOURNAL_DATA);
4026                 break;
4027
4028         case EXT4_MOUNT_ORDERED_DATA:
4029         case EXT4_MOUNT_WRITEBACK_DATA:
4030                 if (!jbd2_journal_check_available_features
4031                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4032                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4033                                "requested data journaling mode");
4034                         goto failed_mount_wq;
4035                 }
4036         default:
4037                 break;
4038         }
4039         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4040
4041         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4042
4043 no_journal:
4044         if (ext4_mballoc_ready) {
4045                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4046                 if (!sbi->s_mb_cache) {
4047                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4048                         goto failed_mount_wq;
4049                 }
4050         }
4051
4052         /*
4053          * Get the # of file system overhead blocks from the
4054          * superblock if present.
4055          */
4056         if (es->s_overhead_clusters)
4057                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4058         else {
4059                 err = ext4_calculate_overhead(sb);
4060                 if (err)
4061                         goto failed_mount_wq;
4062         }
4063
4064         /*
4065          * The maximum number of concurrent works can be high and
4066          * concurrency isn't really necessary.  Limit it to 1.
4067          */
4068         EXT4_SB(sb)->rsv_conversion_wq =
4069                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4070         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4071                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4072                 ret = -ENOMEM;
4073                 goto failed_mount4;
4074         }
4075
4076         /*
4077          * The jbd2_journal_load will have done any necessary log recovery,
4078          * so we can safely mount the rest of the filesystem now.
4079          */
4080
4081         root = ext4_iget(sb, EXT4_ROOT_INO);
4082         if (IS_ERR(root)) {
4083                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4084                 ret = PTR_ERR(root);
4085                 root = NULL;
4086                 goto failed_mount4;
4087         }
4088         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4089                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4090                 iput(root);
4091                 goto failed_mount4;
4092         }
4093         sb->s_root = d_make_root(root);
4094         if (!sb->s_root) {
4095                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4096                 ret = -ENOMEM;
4097                 goto failed_mount4;
4098         }
4099
4100         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4101                 sb->s_flags |= MS_RDONLY;
4102
4103         /* determine the minimum size of new large inodes, if present */
4104         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4105                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4106                                                      EXT4_GOOD_OLD_INODE_SIZE;
4107                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4108                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4109                         if (sbi->s_want_extra_isize <
4110                             le16_to_cpu(es->s_want_extra_isize))
4111                                 sbi->s_want_extra_isize =
4112                                         le16_to_cpu(es->s_want_extra_isize);
4113                         if (sbi->s_want_extra_isize <
4114                             le16_to_cpu(es->s_min_extra_isize))
4115                                 sbi->s_want_extra_isize =
4116                                         le16_to_cpu(es->s_min_extra_isize);
4117                 }
4118         }
4119         /* Check if enough inode space is available */
4120         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4121                                                         sbi->s_inode_size) {
4122                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4123                                                        EXT4_GOOD_OLD_INODE_SIZE;
4124                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4125                          "available");
4126         }
4127
4128         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4129         if (err) {
4130                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4131                          "reserved pool", ext4_calculate_resv_clusters(sb));
4132                 goto failed_mount4a;
4133         }
4134
4135         err = ext4_setup_system_zone(sb);
4136         if (err) {
4137                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4138                          "zone (%d)", err);
4139                 goto failed_mount4a;
4140         }
4141
4142         ext4_ext_init(sb);
4143         err = ext4_mb_init(sb);
4144         if (err) {
4145                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4146                          err);
4147                 goto failed_mount5;
4148         }
4149
4150         block = ext4_count_free_clusters(sb);
4151         ext4_free_blocks_count_set(sbi->s_es,
4152                                    EXT4_C2B(sbi, block));
4153         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4154                                   GFP_KERNEL);
4155         if (!err) {
4156                 unsigned long freei = ext4_count_free_inodes(sb);
4157                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4158                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4159                                           GFP_KERNEL);
4160         }
4161         if (!err)
4162                 err = percpu_counter_init(&sbi->s_dirs_counter,
4163                                           ext4_count_dirs(sb), GFP_KERNEL);
4164         if (!err)
4165                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4166                                           GFP_KERNEL);
4167         if (err) {
4168                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4169                 goto failed_mount6;
4170         }
4171
4172         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4173                 if (!ext4_fill_flex_info(sb)) {
4174                         ext4_msg(sb, KERN_ERR,
4175                                "unable to initialize "
4176                                "flex_bg meta info!");
4177                         goto failed_mount6;
4178                 }
4179
4180         err = ext4_register_li_request(sb, first_not_zeroed);
4181         if (err)
4182                 goto failed_mount6;
4183
4184         sbi->s_kobj.kset = ext4_kset;
4185         init_completion(&sbi->s_kobj_unregister);
4186         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4187                                    "%s", sb->s_id);
4188         if (err)
4189                 goto failed_mount7;
4190
4191 #ifdef CONFIG_QUOTA
4192         /* Enable quota usage during mount. */
4193         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4194             !(sb->s_flags & MS_RDONLY)) {
4195                 err = ext4_enable_quotas(sb);
4196                 if (err)
4197                         goto failed_mount8;
4198         }
4199 #endif  /* CONFIG_QUOTA */
4200
4201         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4202         ext4_orphan_cleanup(sb, es);
4203         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4204         if (needs_recovery) {
4205                 ext4_msg(sb, KERN_INFO, "recovery complete");
4206                 ext4_mark_recovery_complete(sb, es);
4207         }
4208         if (EXT4_SB(sb)->s_journal) {
4209                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4210                         descr = " journalled data mode";
4211                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4212                         descr = " ordered data mode";
4213                 else
4214                         descr = " writeback data mode";
4215         } else
4216                 descr = "out journal";
4217
4218         if (test_opt(sb, DISCARD)) {
4219                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4220                 if (!blk_queue_discard(q))
4221                         ext4_msg(sb, KERN_WARNING,
4222                                  "mounting with \"discard\" option, but "
4223                                  "the device does not support discard");
4224         }
4225
4226         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4227                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4228                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4229
4230         if (es->s_error_count)
4231                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4232
4233         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4234         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4235         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4236         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4237
4238         kfree(orig_data);
4239         return 0;
4240
4241 cantfind_ext4:
4242         if (!silent)
4243                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4244         goto failed_mount;
4245
4246 #ifdef CONFIG_QUOTA
4247 failed_mount8:
4248         kobject_del(&sbi->s_kobj);
4249 #endif
4250 failed_mount7:
4251         ext4_unregister_li_request(sb);
4252 failed_mount6:
4253         ext4_mb_release(sb);
4254         if (sbi->s_flex_groups)
4255                 kvfree(sbi->s_flex_groups);
4256         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4257         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4258         percpu_counter_destroy(&sbi->s_dirs_counter);
4259         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4260 failed_mount5:
4261         ext4_ext_release(sb);
4262         ext4_release_system_zone(sb);
4263 failed_mount4a:
4264         dput(sb->s_root);
4265         sb->s_root = NULL;
4266 failed_mount4:
4267         ext4_msg(sb, KERN_ERR, "mount failed");
4268         if (EXT4_SB(sb)->rsv_conversion_wq)
4269                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4270 failed_mount_wq:
4271         if (sbi->s_journal) {
4272                 jbd2_journal_destroy(sbi->s_journal);
4273                 sbi->s_journal = NULL;
4274         }
4275 failed_mount3a:
4276         ext4_es_unregister_shrinker(sbi);
4277 failed_mount3:
4278         del_timer_sync(&sbi->s_err_report);
4279         if (sbi->s_mmp_tsk)
4280                 kthread_stop(sbi->s_mmp_tsk);
4281 failed_mount2:
4282         for (i = 0; i < db_count; i++)
4283                 brelse(sbi->s_group_desc[i]);
4284         kvfree(sbi->s_group_desc);
4285 failed_mount:
4286         if (sbi->s_chksum_driver)
4287                 crypto_free_shash(sbi->s_chksum_driver);
4288         if (sbi->s_proc) {
4289                 remove_proc_entry("options", sbi->s_proc);
4290                 remove_proc_entry(sb->s_id, ext4_proc_root);
4291         }
4292 #ifdef CONFIG_QUOTA
4293         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4294                 kfree(sbi->s_qf_names[i]);
4295 #endif
4296         ext4_blkdev_remove(sbi);
4297         brelse(bh);
4298 out_fail:
4299         sb->s_fs_info = NULL;
4300         kfree(sbi->s_blockgroup_lock);
4301         kfree(sbi);
4302 out_free_orig:
4303         kfree(orig_data);
4304         return err ? err : ret;
4305 }
4306
4307 /*
4308  * Setup any per-fs journal parameters now.  We'll do this both on
4309  * initial mount, once the journal has been initialised but before we've
4310  * done any recovery; and again on any subsequent remount.
4311  */
4312 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4313 {
4314         struct ext4_sb_info *sbi = EXT4_SB(sb);
4315
4316         journal->j_commit_interval = sbi->s_commit_interval;
4317         journal->j_min_batch_time = sbi->s_min_batch_time;
4318         journal->j_max_batch_time = sbi->s_max_batch_time;
4319
4320         write_lock(&journal->j_state_lock);
4321         if (test_opt(sb, BARRIER))
4322                 journal->j_flags |= JBD2_BARRIER;
4323         else
4324                 journal->j_flags &= ~JBD2_BARRIER;
4325         if (test_opt(sb, DATA_ERR_ABORT))
4326                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4327         else
4328                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4329         write_unlock(&journal->j_state_lock);
4330 }
4331
4332 static journal_t *ext4_get_journal(struct super_block *sb,
4333                                    unsigned int journal_inum)
4334 {
4335         struct inode *journal_inode;
4336         journal_t *journal;
4337
4338         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4339
4340         /* First, test for the existence of a valid inode on disk.  Bad
4341          * things happen if we iget() an unused inode, as the subsequent
4342          * iput() will try to delete it. */
4343
4344         journal_inode = ext4_iget(sb, journal_inum);
4345         if (IS_ERR(journal_inode)) {
4346                 ext4_msg(sb, KERN_ERR, "no journal found");
4347                 return NULL;
4348         }
4349         if (!journal_inode->i_nlink) {
4350                 make_bad_inode(journal_inode);
4351                 iput(journal_inode);
4352                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4353                 return NULL;
4354         }
4355
4356         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4357                   journal_inode, journal_inode->i_size);
4358         if (!S_ISREG(journal_inode->i_mode)) {
4359                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4360                 iput(journal_inode);
4361                 return NULL;
4362         }
4363
4364         journal = jbd2_journal_init_inode(journal_inode);
4365         if (!journal) {
4366                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4367                 iput(journal_inode);
4368                 return NULL;
4369         }
4370         journal->j_private = sb;
4371         ext4_init_journal_params(sb, journal);
4372         return journal;
4373 }
4374
4375 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4376                                        dev_t j_dev)
4377 {
4378         struct buffer_head *bh;
4379         journal_t *journal;
4380         ext4_fsblk_t start;
4381         ext4_fsblk_t len;
4382         int hblock, blocksize;
4383         ext4_fsblk_t sb_block;
4384         unsigned long offset;
4385         struct ext4_super_block *es;
4386         struct block_device *bdev;
4387
4388         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4389
4390         bdev = ext4_blkdev_get(j_dev, sb);
4391         if (bdev == NULL)
4392                 return NULL;
4393
4394         blocksize = sb->s_blocksize;
4395         hblock = bdev_logical_block_size(bdev);
4396         if (blocksize < hblock) {
4397                 ext4_msg(sb, KERN_ERR,
4398                         "blocksize too small for journal device");
4399                 goto out_bdev;
4400         }
4401
4402         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4403         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4404         set_blocksize(bdev, blocksize);
4405         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4406                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4407                        "external journal");
4408                 goto out_bdev;
4409         }
4410
4411         es = (struct ext4_super_block *) (bh->b_data + offset);
4412         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4413             !(le32_to_cpu(es->s_feature_incompat) &
4414               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4415                 ext4_msg(sb, KERN_ERR, "external journal has "
4416                                         "bad superblock");
4417                 brelse(bh);
4418                 goto out_bdev;
4419         }
4420
4421         if ((le32_to_cpu(es->s_feature_ro_compat) &
4422              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4423             es->s_checksum != ext4_superblock_csum(sb, es)) {
4424                 ext4_msg(sb, KERN_ERR, "external journal has "
4425                                        "corrupt superblock");
4426                 brelse(bh);
4427                 goto out_bdev;
4428         }
4429
4430         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4431                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4432                 brelse(bh);
4433                 goto out_bdev;
4434         }
4435
4436         len = ext4_blocks_count(es);
4437         start = sb_block + 1;
4438         brelse(bh);     /* we're done with the superblock */
4439
4440         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4441                                         start, len, blocksize);
4442         if (!journal) {
4443                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4444                 goto out_bdev;
4445         }
4446         journal->j_private = sb;
4447         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4448         wait_on_buffer(journal->j_sb_buffer);
4449         if (!buffer_uptodate(journal->j_sb_buffer)) {
4450                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4451                 goto out_journal;
4452         }
4453         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4454                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4455                                         "user (unsupported) - %d",
4456                         be32_to_cpu(journal->j_superblock->s_nr_users));
4457                 goto out_journal;
4458         }
4459         EXT4_SB(sb)->journal_bdev = bdev;
4460         ext4_init_journal_params(sb, journal);
4461         return journal;
4462
4463 out_journal:
4464         jbd2_journal_destroy(journal);
4465 out_bdev:
4466         ext4_blkdev_put(bdev);
4467         return NULL;
4468 }
4469
4470 static int ext4_load_journal(struct super_block *sb,
4471                              struct ext4_super_block *es,
4472                              unsigned long journal_devnum)
4473 {
4474         journal_t *journal;
4475         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4476         dev_t journal_dev;
4477         int err = 0;
4478         int really_read_only;
4479
4480         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4481
4482         if (journal_devnum &&
4483             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4484                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4485                         "numbers have changed");
4486                 journal_dev = new_decode_dev(journal_devnum);
4487         } else
4488                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4489
4490         really_read_only = bdev_read_only(sb->s_bdev);
4491
4492         /*
4493          * Are we loading a blank journal or performing recovery after a
4494          * crash?  For recovery, we need to check in advance whether we
4495          * can get read-write access to the device.
4496          */
4497         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4498                 if (sb->s_flags & MS_RDONLY) {
4499                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4500                                         "required on readonly filesystem");
4501                         if (really_read_only) {
4502                                 ext4_msg(sb, KERN_ERR, "write access "
4503                                         "unavailable, cannot proceed");
4504                                 return -EROFS;
4505                         }
4506                         ext4_msg(sb, KERN_INFO, "write access will "
4507                                "be enabled during recovery");
4508                 }
4509         }
4510
4511         if (journal_inum && journal_dev) {
4512                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4513                        "and inode journals!");
4514                 return -EINVAL;
4515         }
4516
4517         if (journal_inum) {
4518                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4519                         return -EINVAL;
4520         } else {
4521                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4522                         return -EINVAL;
4523         }
4524
4525         if (!(journal->j_flags & JBD2_BARRIER))
4526                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4527
4528         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4529                 err = jbd2_journal_wipe(journal, !really_read_only);
4530         if (!err) {
4531                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4532                 if (save)
4533                         memcpy(save, ((char *) es) +
4534                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4535                 err = jbd2_journal_load(journal);
4536                 if (save)
4537                         memcpy(((char *) es) + EXT4_S_ERR_START,
4538                                save, EXT4_S_ERR_LEN);
4539                 kfree(save);
4540         }
4541
4542         if (err) {
4543                 ext4_msg(sb, KERN_ERR, "error loading journal");
4544                 jbd2_journal_destroy(journal);
4545                 return err;
4546         }
4547
4548         EXT4_SB(sb)->s_journal = journal;
4549         ext4_clear_journal_err(sb, es);
4550
4551         if (!really_read_only && journal_devnum &&
4552             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4553                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4554
4555                 /* Make sure we flush the recovery flag to disk. */
4556                 ext4_commit_super(sb, 1);
4557         }
4558
4559         return 0;
4560 }
4561
4562 static int ext4_commit_super(struct super_block *sb, int sync)
4563 {
4564         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4565         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4566         int error = 0;
4567
4568         if (!sbh || block_device_ejected(sb))
4569                 return error;
4570         if (buffer_write_io_error(sbh)) {
4571                 /*
4572                  * Oh, dear.  A previous attempt to write the
4573                  * superblock failed.  This could happen because the
4574                  * USB device was yanked out.  Or it could happen to
4575                  * be a transient write error and maybe the block will
4576                  * be remapped.  Nothing we can do but to retry the
4577                  * write and hope for the best.
4578                  */
4579                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4580                        "superblock detected");
4581                 clear_buffer_write_io_error(sbh);
4582                 set_buffer_uptodate(sbh);
4583         }
4584         /*
4585          * If the file system is mounted read-only, don't update the
4586          * superblock write time.  This avoids updating the superblock
4587          * write time when we are mounting the root file system
4588          * read/only but we need to replay the journal; at that point,
4589          * for people who are east of GMT and who make their clock
4590          * tick in localtime for Windows bug-for-bug compatibility,
4591          * the clock is set in the future, and this will cause e2fsck
4592          * to complain and force a full file system check.
4593          */
4594         if (!(sb->s_flags & MS_RDONLY))
4595                 es->s_wtime = cpu_to_le32(get_seconds());
4596         if (sb->s_bdev->bd_part)
4597                 es->s_kbytes_written =
4598                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4599                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4600                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4601         else
4602                 es->s_kbytes_written =
4603                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4604         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4605                 ext4_free_blocks_count_set(es,
4606                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4607                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4608         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4609                 es->s_free_inodes_count =
4610                         cpu_to_le32(percpu_counter_sum_positive(
4611                                 &EXT4_SB(sb)->s_freeinodes_counter));
4612         BUFFER_TRACE(sbh, "marking dirty");
4613         ext4_superblock_csum_set(sb);
4614         mark_buffer_dirty(sbh);
4615         if (sync) {
4616                 error = sync_dirty_buffer(sbh);
4617                 if (error)
4618                         return error;
4619
4620                 error = buffer_write_io_error(sbh);
4621                 if (error) {
4622                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4623                                "superblock");
4624                         clear_buffer_write_io_error(sbh);
4625                         set_buffer_uptodate(sbh);
4626                 }
4627         }
4628         return error;
4629 }
4630
4631 /*
4632  * Have we just finished recovery?  If so, and if we are mounting (or
4633  * remounting) the filesystem readonly, then we will end up with a
4634  * consistent fs on disk.  Record that fact.
4635  */
4636 static void ext4_mark_recovery_complete(struct super_block *sb,
4637                                         struct ext4_super_block *es)
4638 {
4639         journal_t *journal = EXT4_SB(sb)->s_journal;
4640
4641         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4642                 BUG_ON(journal != NULL);
4643                 return;
4644         }
4645         jbd2_journal_lock_updates(journal);
4646         if (jbd2_journal_flush(journal) < 0)
4647                 goto out;
4648
4649         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4650             sb->s_flags & MS_RDONLY) {
4651                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4652                 ext4_commit_super(sb, 1);
4653         }
4654
4655 out:
4656         jbd2_journal_unlock_updates(journal);
4657 }
4658
4659 /*
4660  * If we are mounting (or read-write remounting) a filesystem whose journal
4661  * has recorded an error from a previous lifetime, move that error to the
4662  * main filesystem now.
4663  */
4664 static void ext4_clear_journal_err(struct super_block *sb,
4665                                    struct ext4_super_block *es)
4666 {
4667         journal_t *journal;
4668         int j_errno;
4669         const char *errstr;
4670
4671         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4672
4673         journal = EXT4_SB(sb)->s_journal;
4674
4675         /*
4676          * Now check for any error status which may have been recorded in the
4677          * journal by a prior ext4_error() or ext4_abort()
4678          */
4679
4680         j_errno = jbd2_journal_errno(journal);
4681         if (j_errno) {
4682                 char nbuf[16];
4683
4684                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4685                 ext4_warning(sb, "Filesystem error recorded "
4686                              "from previous mount: %s", errstr);
4687                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4688
4689                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4690                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4691                 ext4_commit_super(sb, 1);
4692
4693                 jbd2_journal_clear_err(journal);
4694                 jbd2_journal_update_sb_errno(journal);
4695         }
4696 }
4697
4698 /*
4699  * Force the running and committing transactions to commit,
4700  * and wait on the commit.
4701  */
4702 int ext4_force_commit(struct super_block *sb)
4703 {
4704         journal_t *journal;
4705
4706         if (sb->s_flags & MS_RDONLY)
4707                 return 0;
4708
4709         journal = EXT4_SB(sb)->s_journal;
4710         return ext4_journal_force_commit(journal);
4711 }
4712
4713 static int ext4_sync_fs(struct super_block *sb, int wait)
4714 {
4715         int ret = 0;
4716         tid_t target;
4717         bool needs_barrier = false;
4718         struct ext4_sb_info *sbi = EXT4_SB(sb);
4719
4720         trace_ext4_sync_fs(sb, wait);
4721         flush_workqueue(sbi->rsv_conversion_wq);
4722         /*
4723          * Writeback quota in non-journalled quota case - journalled quota has
4724          * no dirty dquots
4725          */
4726         dquot_writeback_dquots(sb, -1);
4727         /*
4728          * Data writeback is possible w/o journal transaction, so barrier must
4729          * being sent at the end of the function. But we can skip it if
4730          * transaction_commit will do it for us.
4731          */
4732         if (sbi->s_journal) {
4733                 target = jbd2_get_latest_transaction(sbi->s_journal);
4734                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4735                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4736                         needs_barrier = true;
4737
4738                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4739                         if (wait)
4740                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4741                                                            target);
4742                 }
4743         } else if (wait && test_opt(sb, BARRIER))
4744                 needs_barrier = true;
4745         if (needs_barrier) {
4746                 int err;
4747                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4748                 if (!ret)
4749                         ret = err;
4750         }
4751
4752         return ret;
4753 }
4754
4755 /*
4756  * LVM calls this function before a (read-only) snapshot is created.  This
4757  * gives us a chance to flush the journal completely and mark the fs clean.
4758  *
4759  * Note that only this function cannot bring a filesystem to be in a clean
4760  * state independently. It relies on upper layer to stop all data & metadata
4761  * modifications.
4762  */
4763 static int ext4_freeze(struct super_block *sb)
4764 {
4765         int error = 0;
4766         journal_t *journal;
4767
4768         if (sb->s_flags & MS_RDONLY)
4769                 return 0;
4770
4771         journal = EXT4_SB(sb)->s_journal;
4772
4773         if (journal) {
4774                 /* Now we set up the journal barrier. */
4775                 jbd2_journal_lock_updates(journal);
4776
4777                 /*
4778                  * Don't clear the needs_recovery flag if we failed to
4779                  * flush the journal.
4780                  */
4781                 error = jbd2_journal_flush(journal);
4782                 if (error < 0)
4783                         goto out;
4784         }
4785
4786         /* Journal blocked and flushed, clear needs_recovery flag. */
4787         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4788         error = ext4_commit_super(sb, 1);
4789 out:
4790         if (journal)
4791                 /* we rely on upper layer to stop further updates */
4792                 jbd2_journal_unlock_updates(journal);
4793         return error;
4794 }
4795
4796 /*
4797  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4798  * flag here, even though the filesystem is not technically dirty yet.
4799  */
4800 static int ext4_unfreeze(struct super_block *sb)
4801 {
4802         if (sb->s_flags & MS_RDONLY)
4803                 return 0;
4804
4805         /* Reset the needs_recovery flag before the fs is unlocked. */
4806         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4807         ext4_commit_super(sb, 1);
4808         return 0;
4809 }
4810
4811 /*
4812  * Structure to save mount options for ext4_remount's benefit
4813  */
4814 struct ext4_mount_options {
4815         unsigned long s_mount_opt;
4816         unsigned long s_mount_opt2;
4817         kuid_t s_resuid;
4818         kgid_t s_resgid;
4819         unsigned long s_commit_interval;
4820         u32 s_min_batch_time, s_max_batch_time;
4821 #ifdef CONFIG_QUOTA
4822         int s_jquota_fmt;
4823         char *s_qf_names[EXT4_MAXQUOTAS];
4824 #endif
4825 };
4826
4827 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4828 {
4829         struct ext4_super_block *es;
4830         struct ext4_sb_info *sbi = EXT4_SB(sb);
4831         unsigned long old_sb_flags;
4832         struct ext4_mount_options old_opts;
4833         int enable_quota = 0;
4834         ext4_group_t g;
4835         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4836         int err = 0;
4837 #ifdef CONFIG_QUOTA
4838         int i, j;
4839 #endif
4840         char *orig_data = kstrdup(data, GFP_KERNEL);
4841
4842         /* Store the original options */
4843         old_sb_flags = sb->s_flags;
4844         old_opts.s_mount_opt = sbi->s_mount_opt;
4845         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4846         old_opts.s_resuid = sbi->s_resuid;
4847         old_opts.s_resgid = sbi->s_resgid;
4848         old_opts.s_commit_interval = sbi->s_commit_interval;
4849         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4850         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4851 #ifdef CONFIG_QUOTA
4852         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4853         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4854                 if (sbi->s_qf_names[i]) {
4855                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4856                                                          GFP_KERNEL);
4857                         if (!old_opts.s_qf_names[i]) {
4858                                 for (j = 0; j < i; j++)
4859                                         kfree(old_opts.s_qf_names[j]);
4860                                 kfree(orig_data);
4861                                 return -ENOMEM;
4862                         }
4863                 } else
4864                         old_opts.s_qf_names[i] = NULL;
4865 #endif
4866         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4867                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4868
4869         /*
4870          * Allow the "check" option to be passed as a remount option.
4871          */
4872         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4873                 err = -EINVAL;
4874                 goto restore_opts;
4875         }
4876
4877         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4878             test_opt(sb, JOURNAL_CHECKSUM)) {
4879                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4880                          "during remount not supported");
4881                 err = -EINVAL;
4882                 goto restore_opts;
4883         }
4884
4885         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4886             test_opt(sb, JOURNAL_CHECKSUM)) {
4887                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4888                          "during remount not supported");
4889                 err = -EINVAL;
4890                 goto restore_opts;
4891         }
4892
4893         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4894                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4895                         ext4_msg(sb, KERN_ERR, "can't mount with "
4896                                  "both data=journal and delalloc");
4897                         err = -EINVAL;
4898                         goto restore_opts;
4899                 }
4900                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4901                         ext4_msg(sb, KERN_ERR, "can't mount with "
4902                                  "both data=journal and dioread_nolock");
4903                         err = -EINVAL;
4904                         goto restore_opts;
4905                 }
4906                 if (test_opt(sb, DAX)) {
4907                         ext4_msg(sb, KERN_ERR, "can't mount with "
4908                                  "both data=journal and dax");
4909                         err = -EINVAL;
4910                         goto restore_opts;
4911                 }
4912         }
4913
4914         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4915                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4916                         "dax flag with busy inodes while remounting");
4917                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4918         }
4919
4920         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4921                 ext4_abort(sb, "Abort forced by user");
4922
4923         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4924                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4925
4926         es = sbi->s_es;
4927
4928         if (sbi->s_journal) {
4929                 ext4_init_journal_params(sb, sbi->s_journal);
4930                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4931         }
4932
4933         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4934                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4935                         err = -EROFS;
4936                         goto restore_opts;
4937                 }
4938
4939                 if (*flags & MS_RDONLY) {
4940                         err = sync_filesystem(sb);
4941                         if (err < 0)
4942                                 goto restore_opts;
4943                         err = dquot_suspend(sb, -1);
4944                         if (err < 0)
4945                                 goto restore_opts;
4946
4947                         /*
4948                          * First of all, the unconditional stuff we have to do
4949                          * to disable replay of the journal when we next remount
4950                          */
4951                         sb->s_flags |= MS_RDONLY;
4952
4953                         /*
4954                          * OK, test if we are remounting a valid rw partition
4955                          * readonly, and if so set the rdonly flag and then
4956                          * mark the partition as valid again.
4957                          */
4958                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4959                             (sbi->s_mount_state & EXT4_VALID_FS))
4960                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4961
4962                         if (sbi->s_journal)
4963                                 ext4_mark_recovery_complete(sb, es);
4964                 } else {
4965                         /* Make sure we can mount this feature set readwrite */
4966                         if (!ext4_feature_set_ok(sb, 0)) {
4967                                 err = -EROFS;
4968                                 goto restore_opts;
4969                         }
4970                         /*
4971                          * Make sure the group descriptor checksums
4972                          * are sane.  If they aren't, refuse to remount r/w.
4973                          */
4974                         for (g = 0; g < sbi->s_groups_count; g++) {
4975                                 struct ext4_group_desc *gdp =
4976                                         ext4_get_group_desc(sb, g, NULL);
4977
4978                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4979                                         ext4_msg(sb, KERN_ERR,
4980                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4981                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4982                                                le16_to_cpu(gdp->bg_checksum));
4983                                         err = -EINVAL;
4984                                         goto restore_opts;
4985                                 }
4986                         }
4987
4988                         /*
4989                          * If we have an unprocessed orphan list hanging
4990                          * around from a previously readonly bdev mount,
4991                          * require a full umount/remount for now.
4992                          */
4993                         if (es->s_last_orphan) {
4994                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4995                                        "remount RDWR because of unprocessed "
4996                                        "orphan inode list.  Please "
4997                                        "umount/remount instead");
4998                                 err = -EINVAL;
4999                                 goto restore_opts;
5000                         }
5001
5002                         /*
5003                          * Mounting a RDONLY partition read-write, so reread
5004                          * and store the current valid flag.  (It may have
5005                          * been changed by e2fsck since we originally mounted
5006                          * the partition.)
5007                          */
5008                         if (sbi->s_journal)
5009                                 ext4_clear_journal_err(sb, es);
5010                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5011                         if (!ext4_setup_super(sb, es, 0))
5012                                 sb->s_flags &= ~MS_RDONLY;
5013                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
5014                                                      EXT4_FEATURE_INCOMPAT_MMP))
5015                                 if (ext4_multi_mount_protect(sb,
5016                                                 le64_to_cpu(es->s_mmp_block))) {
5017                                         err = -EROFS;
5018                                         goto restore_opts;
5019                                 }
5020                         enable_quota = 1;
5021                 }
5022         }
5023
5024         /*
5025          * Reinitialize lazy itable initialization thread based on
5026          * current settings
5027          */
5028         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5029                 ext4_unregister_li_request(sb);
5030         else {
5031                 ext4_group_t first_not_zeroed;
5032                 first_not_zeroed = ext4_has_uninit_itable(sb);
5033                 ext4_register_li_request(sb, first_not_zeroed);
5034         }
5035
5036         ext4_setup_system_zone(sb);
5037         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5038                 ext4_commit_super(sb, 1);
5039
5040 #ifdef CONFIG_QUOTA
5041         /* Release old quota file names */
5042         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5043                 kfree(old_opts.s_qf_names[i]);
5044         if (enable_quota) {
5045                 if (sb_any_quota_suspended(sb))
5046                         dquot_resume(sb, -1);
5047                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5048                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5049                         err = ext4_enable_quotas(sb);
5050                         if (err)
5051                                 goto restore_opts;
5052                 }
5053         }
5054 #endif
5055
5056         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5057         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5058         kfree(orig_data);
5059         return 0;
5060
5061 restore_opts:
5062         sb->s_flags = old_sb_flags;
5063         sbi->s_mount_opt = old_opts.s_mount_opt;
5064         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5065         sbi->s_resuid = old_opts.s_resuid;
5066         sbi->s_resgid = old_opts.s_resgid;
5067         sbi->s_commit_interval = old_opts.s_commit_interval;
5068         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5069         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5070 #ifdef CONFIG_QUOTA
5071         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5072         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5073                 kfree(sbi->s_qf_names[i]);
5074                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5075         }
5076 #endif
5077         kfree(orig_data);
5078         return err;
5079 }
5080
5081 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5082 {
5083         struct super_block *sb = dentry->d_sb;
5084         struct ext4_sb_info *sbi = EXT4_SB(sb);
5085         struct ext4_super_block *es = sbi->s_es;
5086         ext4_fsblk_t overhead = 0, resv_blocks;
5087         u64 fsid;
5088         s64 bfree;
5089         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5090
5091         if (!test_opt(sb, MINIX_DF))
5092                 overhead = sbi->s_overhead;
5093
5094         buf->f_type = EXT4_SUPER_MAGIC;
5095         buf->f_bsize = sb->s_blocksize;
5096         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5097         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5098                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5099         /* prevent underflow in case that few free space is available */
5100         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5101         buf->f_bavail = buf->f_bfree -
5102                         (ext4_r_blocks_count(es) + resv_blocks);
5103         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5104                 buf->f_bavail = 0;
5105         buf->f_files = le32_to_cpu(es->s_inodes_count);
5106         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5107         buf->f_namelen = EXT4_NAME_LEN;
5108         fsid = le64_to_cpup((void *)es->s_uuid) ^
5109                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5110         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5111         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5112
5113         return 0;
5114 }
5115
5116 /* Helper function for writing quotas on sync - we need to start transaction
5117  * before quota file is locked for write. Otherwise the are possible deadlocks:
5118  * Process 1                         Process 2
5119  * ext4_create()                     quota_sync()
5120  *   jbd2_journal_start()                  write_dquot()
5121  *   dquot_initialize()                         down(dqio_mutex)
5122  *     down(dqio_mutex)                    jbd2_journal_start()
5123  *
5124  */
5125
5126 #ifdef CONFIG_QUOTA
5127
5128 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5129 {
5130         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5131 }
5132
5133 static int ext4_write_dquot(struct dquot *dquot)
5134 {
5135         int ret, err;
5136         handle_t *handle;
5137         struct inode *inode;
5138
5139         inode = dquot_to_inode(dquot);
5140         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5141                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5142         if (IS_ERR(handle))
5143                 return PTR_ERR(handle);
5144         ret = dquot_commit(dquot);
5145         err = ext4_journal_stop(handle);
5146         if (!ret)
5147                 ret = err;
5148         return ret;
5149 }
5150
5151 static int ext4_acquire_dquot(struct dquot *dquot)
5152 {
5153         int ret, err;
5154         handle_t *handle;
5155
5156         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5157                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5158         if (IS_ERR(handle))
5159                 return PTR_ERR(handle);
5160         ret = dquot_acquire(dquot);
5161         err = ext4_journal_stop(handle);
5162         if (!ret)
5163                 ret = err;
5164         return ret;
5165 }
5166
5167 static int ext4_release_dquot(struct dquot *dquot)
5168 {
5169         int ret, err;
5170         handle_t *handle;
5171
5172         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5173                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5174         if (IS_ERR(handle)) {
5175                 /* Release dquot anyway to avoid endless cycle in dqput() */
5176                 dquot_release(dquot);
5177                 return PTR_ERR(handle);
5178         }
5179         ret = dquot_release(dquot);
5180         err = ext4_journal_stop(handle);
5181         if (!ret)
5182                 ret = err;
5183         return ret;
5184 }
5185
5186 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5187 {
5188         struct super_block *sb = dquot->dq_sb;
5189         struct ext4_sb_info *sbi = EXT4_SB(sb);
5190
5191         /* Are we journaling quotas? */
5192         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5193             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5194                 dquot_mark_dquot_dirty(dquot);
5195                 return ext4_write_dquot(dquot);
5196         } else {
5197                 return dquot_mark_dquot_dirty(dquot);
5198         }
5199 }
5200
5201 static int ext4_write_info(struct super_block *sb, int type)
5202 {
5203         int ret, err;
5204         handle_t *handle;
5205
5206         /* Data block + inode block */
5207         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5208         if (IS_ERR(handle))
5209                 return PTR_ERR(handle);
5210         ret = dquot_commit_info(sb, type);
5211         err = ext4_journal_stop(handle);
5212         if (!ret)
5213                 ret = err;
5214         return ret;
5215 }
5216
5217 /*
5218  * Turn on quotas during mount time - we need to find
5219  * the quota file and such...
5220  */
5221 static int ext4_quota_on_mount(struct super_block *sb, int type)
5222 {
5223         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5224                                         EXT4_SB(sb)->s_jquota_fmt, type);
5225 }
5226
5227 /*
5228  * Standard function to be called on quota_on
5229  */
5230 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5231                          struct path *path)
5232 {
5233         int err;
5234
5235         if (!test_opt(sb, QUOTA))
5236                 return -EINVAL;
5237
5238         /* Quotafile not on the same filesystem? */
5239         if (path->dentry->d_sb != sb)
5240                 return -EXDEV;
5241         /* Journaling quota? */
5242         if (EXT4_SB(sb)->s_qf_names[type]) {
5243                 /* Quotafile not in fs root? */
5244                 if (path->dentry->d_parent != sb->s_root)
5245                         ext4_msg(sb, KERN_WARNING,
5246                                 "Quota file not on filesystem root. "
5247                                 "Journaled quota will not work");
5248         }
5249
5250         /*
5251          * When we journal data on quota file, we have to flush journal to see
5252          * all updates to the file when we bypass pagecache...
5253          */
5254         if (EXT4_SB(sb)->s_journal &&
5255             ext4_should_journal_data(path->dentry->d_inode)) {
5256                 /*
5257                  * We don't need to lock updates but journal_flush() could
5258                  * otherwise be livelocked...
5259                  */
5260                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5261                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5262                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5263                 if (err)
5264                         return err;
5265         }
5266
5267         return dquot_quota_on(sb, type, format_id, path);
5268 }
5269
5270 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5271                              unsigned int flags)
5272 {
5273         int err;
5274         struct inode *qf_inode;
5275         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5276                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5277                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5278         };
5279
5280         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5281
5282         if (!qf_inums[type])
5283                 return -EPERM;
5284
5285         qf_inode = ext4_iget(sb, qf_inums[type]);
5286         if (IS_ERR(qf_inode)) {
5287                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5288                 return PTR_ERR(qf_inode);
5289         }
5290
5291         /* Don't account quota for quota files to avoid recursion */
5292         qf_inode->i_flags |= S_NOQUOTA;
5293         err = dquot_enable(qf_inode, type, format_id, flags);
5294         iput(qf_inode);
5295
5296         return err;
5297 }
5298
5299 /* Enable usage tracking for all quota types. */
5300 static int ext4_enable_quotas(struct super_block *sb)
5301 {
5302         int type, err = 0;
5303         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5304                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5305                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5306         };
5307
5308         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5309         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5310                 if (qf_inums[type]) {
5311                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5312                                                 DQUOT_USAGE_ENABLED);
5313                         if (err) {
5314                                 ext4_warning(sb,
5315                                         "Failed to enable quota tracking "
5316                                         "(type=%d, err=%d). Please run "
5317                                         "e2fsck to fix.", type, err);
5318                                 return err;
5319                         }
5320                 }
5321         }
5322         return 0;
5323 }
5324
5325 static int ext4_quota_off(struct super_block *sb, int type)
5326 {
5327         struct inode *inode = sb_dqopt(sb)->files[type];
5328         handle_t *handle;
5329
5330         /* Force all delayed allocation blocks to be allocated.
5331          * Caller already holds s_umount sem */
5332         if (test_opt(sb, DELALLOC))
5333                 sync_filesystem(sb);
5334
5335         if (!inode)
5336                 goto out;
5337
5338         /* Update modification times of quota files when userspace can
5339          * start looking at them */
5340         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5341         if (IS_ERR(handle))
5342                 goto out;
5343         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5344         ext4_mark_inode_dirty(handle, inode);
5345         ext4_journal_stop(handle);
5346
5347 out:
5348         return dquot_quota_off(sb, type);
5349 }
5350
5351 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5352  * acquiring the locks... As quota files are never truncated and quota code
5353  * itself serializes the operations (and no one else should touch the files)
5354  * we don't have to be afraid of races */
5355 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5356                                size_t len, loff_t off)
5357 {
5358         struct inode *inode = sb_dqopt(sb)->files[type];
5359         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5360         int offset = off & (sb->s_blocksize - 1);
5361         int tocopy;
5362         size_t toread;
5363         struct buffer_head *bh;
5364         loff_t i_size = i_size_read(inode);
5365
5366         if (off > i_size)
5367                 return 0;
5368         if (off+len > i_size)
5369                 len = i_size-off;
5370         toread = len;
5371         while (toread > 0) {
5372                 tocopy = sb->s_blocksize - offset < toread ?
5373                                 sb->s_blocksize - offset : toread;
5374                 bh = ext4_bread(NULL, inode, blk, 0);
5375                 if (IS_ERR(bh))
5376                         return PTR_ERR(bh);
5377                 if (!bh)        /* A hole? */
5378                         memset(data, 0, tocopy);
5379                 else
5380                         memcpy(data, bh->b_data+offset, tocopy);
5381                 brelse(bh);
5382                 offset = 0;
5383                 toread -= tocopy;
5384                 data += tocopy;
5385                 blk++;
5386         }
5387         return len;
5388 }
5389
5390 /* Write to quotafile (we know the transaction is already started and has
5391  * enough credits) */
5392 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5393                                 const char *data, size_t len, loff_t off)
5394 {
5395         struct inode *inode = sb_dqopt(sb)->files[type];
5396         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5397         int err, offset = off & (sb->s_blocksize - 1);
5398         struct buffer_head *bh;
5399         handle_t *handle = journal_current_handle();
5400
5401         if (EXT4_SB(sb)->s_journal && !handle) {
5402                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5403                         " cancelled because transaction is not started",
5404                         (unsigned long long)off, (unsigned long long)len);
5405                 return -EIO;
5406         }
5407         /*
5408          * Since we account only one data block in transaction credits,
5409          * then it is impossible to cross a block boundary.
5410          */
5411         if (sb->s_blocksize - offset < len) {
5412                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5413                         " cancelled because not block aligned",
5414                         (unsigned long long)off, (unsigned long long)len);
5415                 return -EIO;
5416         }
5417
5418         bh = ext4_bread(handle, inode, blk, 1);
5419         if (IS_ERR(bh))
5420                 return PTR_ERR(bh);
5421         if (!bh)
5422                 goto out;
5423         BUFFER_TRACE(bh, "get write access");
5424         err = ext4_journal_get_write_access(handle, bh);
5425         if (err) {
5426                 brelse(bh);
5427                 return err;
5428         }
5429         lock_buffer(bh);
5430         memcpy(bh->b_data+offset, data, len);
5431         flush_dcache_page(bh->b_page);
5432         unlock_buffer(bh);
5433         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5434         brelse(bh);
5435 out:
5436         if (inode->i_size < off + len) {
5437                 i_size_write(inode, off + len);
5438                 EXT4_I(inode)->i_disksize = inode->i_size;
5439                 ext4_mark_inode_dirty(handle, inode);
5440         }
5441         return len;
5442 }
5443
5444 #endif
5445
5446 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5447                        const char *dev_name, void *data)
5448 {
5449         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5450 }
5451
5452 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5453 static inline void register_as_ext2(void)
5454 {
5455         int err = register_filesystem(&ext2_fs_type);
5456         if (err)
5457                 printk(KERN_WARNING
5458                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5459 }
5460
5461 static inline void unregister_as_ext2(void)
5462 {
5463         unregister_filesystem(&ext2_fs_type);
5464 }
5465
5466 static inline int ext2_feature_set_ok(struct super_block *sb)
5467 {
5468         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5469                 return 0;
5470         if (sb->s_flags & MS_RDONLY)
5471                 return 1;
5472         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5473                 return 0;
5474         return 1;
5475 }
5476 #else
5477 static inline void register_as_ext2(void) { }
5478 static inline void unregister_as_ext2(void) { }
5479 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5480 #endif
5481
5482 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5483 static inline void register_as_ext3(void)
5484 {
5485         int err = register_filesystem(&ext3_fs_type);
5486         if (err)
5487                 printk(KERN_WARNING
5488                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5489 }
5490
5491 static inline void unregister_as_ext3(void)
5492 {
5493         unregister_filesystem(&ext3_fs_type);
5494 }
5495
5496 static inline int ext3_feature_set_ok(struct super_block *sb)
5497 {
5498         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5499                 return 0;
5500         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5501                 return 0;
5502         if (sb->s_flags & MS_RDONLY)
5503                 return 1;
5504         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5505                 return 0;
5506         return 1;
5507 }
5508 #else
5509 static inline void register_as_ext3(void) { }
5510 static inline void unregister_as_ext3(void) { }
5511 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5512 #endif
5513
5514 static struct file_system_type ext4_fs_type = {
5515         .owner          = THIS_MODULE,
5516         .name           = "ext4",
5517         .mount          = ext4_mount,
5518         .kill_sb        = kill_block_super,
5519         .fs_flags       = FS_REQUIRES_DEV,
5520 };
5521 MODULE_ALIAS_FS("ext4");
5522
5523 static int __init ext4_init_feat_adverts(void)
5524 {
5525         struct ext4_features *ef;
5526         int ret = -ENOMEM;
5527
5528         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5529         if (!ef)
5530                 goto out;
5531
5532         ef->f_kobj.kset = ext4_kset;
5533         init_completion(&ef->f_kobj_unregister);
5534         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5535                                    "features");
5536         if (ret) {
5537                 kfree(ef);
5538                 goto out;
5539         }
5540
5541         ext4_feat = ef;
5542         ret = 0;
5543 out:
5544         return ret;
5545 }
5546
5547 static void ext4_exit_feat_adverts(void)
5548 {
5549         kobject_put(&ext4_feat->f_kobj);
5550         wait_for_completion(&ext4_feat->f_kobj_unregister);
5551         kfree(ext4_feat);
5552 }
5553
5554 /* Shared across all ext4 file systems */
5555 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5556 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5557
5558 static int __init ext4_init_fs(void)
5559 {
5560         int i, err;
5561
5562         ext4_li_info = NULL;
5563         mutex_init(&ext4_li_mtx);
5564
5565         /* Build-time check for flags consistency */
5566         ext4_check_flag_values();
5567
5568         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5569                 mutex_init(&ext4__aio_mutex[i]);
5570                 init_waitqueue_head(&ext4__ioend_wq[i]);
5571         }
5572
5573         err = ext4_init_es();
5574         if (err)
5575                 return err;
5576
5577         err = ext4_init_pageio();
5578         if (err)
5579                 goto out7;
5580
5581         err = ext4_init_system_zone();
5582         if (err)
5583                 goto out6;
5584         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5585         if (!ext4_kset) {
5586                 err = -ENOMEM;
5587                 goto out5;
5588         }
5589         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5590
5591         err = ext4_init_feat_adverts();
5592         if (err)
5593                 goto out4;
5594
5595         err = ext4_init_mballoc();
5596         if (err)
5597                 goto out2;
5598         else
5599                 ext4_mballoc_ready = 1;
5600         err = init_inodecache();
5601         if (err)
5602                 goto out1;
5603         register_as_ext3();
5604         register_as_ext2();
5605         err = register_filesystem(&ext4_fs_type);
5606         if (err)
5607                 goto out;
5608
5609         return 0;
5610 out:
5611         unregister_as_ext2();
5612         unregister_as_ext3();
5613         destroy_inodecache();
5614 out1:
5615         ext4_mballoc_ready = 0;
5616         ext4_exit_mballoc();
5617 out2:
5618         ext4_exit_feat_adverts();
5619 out4:
5620         if (ext4_proc_root)
5621                 remove_proc_entry("fs/ext4", NULL);
5622         kset_unregister(ext4_kset);
5623 out5:
5624         ext4_exit_system_zone();
5625 out6:
5626         ext4_exit_pageio();
5627 out7:
5628         ext4_exit_es();
5629
5630         return err;
5631 }
5632
5633 static void __exit ext4_exit_fs(void)
5634 {
5635         ext4_destroy_lazyinit_thread();
5636         unregister_as_ext2();
5637         unregister_as_ext3();
5638         unregister_filesystem(&ext4_fs_type);
5639         destroy_inodecache();
5640         ext4_exit_mballoc();
5641         ext4_exit_feat_adverts();
5642         remove_proc_entry("fs/ext4", NULL);
5643         kset_unregister(ext4_kset);
5644         ext4_exit_system_zone();
5645         ext4_exit_pageio();
5646         ext4_exit_es();
5647 }
5648
5649 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5650 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5651 MODULE_LICENSE("GPL");
5652 module_init(ext4_init_fs)
5653 module_exit(ext4_exit_fs)