kernel/fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/parser.h>
  28 #include <linux/buffer_head.h>
  29 #include <linux/exportfs.h>
  30 #include <linux/vfs.h>
  31 #include <linux/random.h>
  32 #include <linux/mount.h>
  33 #include <linux/namei.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct proc_dir_entry *ext4_proc_root;
  57 static struct kset *ext4_kset;
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ext4_features *ext4_feat;
  61 static int ext4_mballoc_ready;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static inline int ext2_feature_set_ok(struct super_block *sb);
  79 static inline int ext3_feature_set_ok(struct super_block *sb);
  80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  81 static void ext4_destroy_lazyinit_thread(void);
  82 static void ext4_unregister_li_request(struct super_block *sb);
  83 static void ext4_clear_request_list(void);
  84 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 MODULE_ALIAS_FS("ext2");
  95 MODULE_ALIAS("ext2");
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 MODULE_ALIAS_FS("ext3");
 111 MODULE_ALIAS("ext3");
 112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 113 #else
 114 #define IS_EXT3_SB(sb) (0)
 115 #endif
 116
 117 static int ext4_verify_csum_type(struct super_block *sb,
 118                                  struct ext4_super_block *es)
 119 {
 120         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 121                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 122                 return 1;
 123
 124         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 125 }
 126
 127 static __le32 ext4_superblock_csum(struct super_block *sb,
 128                                    struct ext4_super_block *es)
 129 {
 130         struct ext4_sb_info *sbi = EXT4_SB(sb);
 131         int offset = offsetof(struct ext4_super_block, s_checksum);
 132         __u32 csum;
 133
 134         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 135
 136         return cpu_to_le32(csum);
 137 }
 138
 139 static int ext4_superblock_csum_verify(struct super_block *sb,
 140                                        struct ext4_super_block *es)
 141 {
 142         if (!ext4_has_metadata_csum(sb))
 143                 return 1;
 144
 145         return es->s_checksum == ext4_superblock_csum(sb, es);
 146 }
 147
 148 void ext4_superblock_csum_set(struct super_block *sb)
 149 {
 150         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 151
 152         if (!ext4_has_metadata_csum(sb))
 153                 return;
 154
 155         es->s_checksum = ext4_superblock_csum(sb, es);
 156 }
 157
 158 void *ext4_kvmalloc(size_t size, gfp_t flags)
 159 {
 160         void *ret;
 161
 162         ret = kmalloc(size, flags | __GFP_NOWARN);
 163         if (!ret)
 164                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 165         return ret;
 166 }
 167
 168 void *ext4_kvzalloc(size_t size, gfp_t flags)
 169 {
 170         void *ret;
 171
 172         ret = kzalloc(size, flags | __GFP_NOWARN);
 173         if (!ret)
 174                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 175         return ret;
 176 }
 177
 178 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 187                                struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 192 }
 193
 194 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 195                               struct ext4_group_desc *bg)
 196 {
 197         return le32_to_cpu(bg->bg_inode_table_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 200 }
 201
 202 __u32 ext4_free_group_clusters(struct super_block *sb,
 203                                struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_free_inodes_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_used_dirs_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 224 }
 225
 226 __u32 ext4_itable_unused_count(struct super_block *sb,
 227                               struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_itable_unused_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 232 }
 233
 234 void ext4_block_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_bitmap_set(struct super_block *sb,
 243                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_inode_table_set(struct super_block *sb,
 251                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 252 {
 253         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 256 }
 257
 258 void ext4_free_group_clusters_set(struct super_block *sb,
 259                                   struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_free_inodes_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_used_dirs_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282 void ext4_itable_unused_set(struct super_block *sb,
 283                           struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290
 291 static void __save_error_info(struct super_block *sb, const char *func,
 292                             unsigned int line)
 293 {
 294         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 295
 296         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 297         if (bdev_read_only(sb->s_bdev))
 298                 return;
 299         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 300         es->s_last_error_time = cpu_to_le32(get_seconds());
 301         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 302         es->s_last_error_line = cpu_to_le32(line);
 303         if (!es->s_first_error_time) {
 304                 es->s_first_error_time = es->s_last_error_time;
 305                 strncpy(es->s_first_error_func, func,
 306                         sizeof(es->s_first_error_func));
 307                 es->s_first_error_line = cpu_to_le32(line);
 308                 es->s_first_error_ino = es->s_last_error_ino;
 309                 es->s_first_error_block = es->s_last_error_block;
 310         }
 311         /*
 312          * Start the daily error reporting function if it hasn't been
 313          * started already
 314          */
 315         if (!es->s_error_count)
 316                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 317         le32_add_cpu(&es->s_error_count, 1);
 318 }
 319
 320 static void save_error_info(struct super_block *sb, const char *func,
 321                             unsigned int line)
 322 {
 323         __save_error_info(sb, func, line);
 324         ext4_commit_super(sb, 1);
 325 }
 326
 327 /*
 328  * The del_gendisk() function uninitializes the disk-specific data
 329  * structures, including the bdi structure, without telling anyone
 330  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 331  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 332  * This is a kludge to prevent these oops until we can put in a proper
 333  * hook in del_gendisk() to inform the VFS and file system layers.
 334  */
 335 static int block_device_ejected(struct super_block *sb)
 336 {
 337         struct inode *bd_inode = sb->s_bdev->bd_inode;
 338         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 339
 340         return bdi->dev == NULL;
 341 }
 342
 343 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 344 {
 345         struct super_block              *sb = journal->j_private;
 346         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 347         int                             error = is_journal_aborted(journal);
 348         struct ext4_journal_cb_entry    *jce;
 349
 350         BUG_ON(txn->t_state == T_FINISHED);
 351         spin_lock(&sbi->s_md_lock);
 352         while (!list_empty(&txn->t_private_list)) {
 353                 jce = list_entry(txn->t_private_list.next,
 354                                  struct ext4_journal_cb_entry, jce_list);
 355                 list_del_init(&jce->jce_list);
 356                 spin_unlock(&sbi->s_md_lock);
 357                 jce->jce_func(sb, jce, error);
 358                 spin_lock(&sbi->s_md_lock);
 359         }
 360         spin_unlock(&sbi->s_md_lock);
 361 }
 362
 363 /* Deal with the reporting of failure conditions on a filesystem such as
 364  * inconsistencies detected or read IO failures.
 365  *
 366  * On ext2, we can store the error state of the filesystem in the
 367  * superblock.  That is not possible on ext4, because we may have other
 368  * write ordering constraints on the superblock which prevent us from
 369  * writing it out straight away; and given that the journal is about to
 370  * be aborted, we can't rely on the current, or future, transactions to
 371  * write out the superblock safely.
 372  *
 373  * We'll just use the jbd2_journal_abort() error code to record an error in
 374  * the journal instead.  On recovery, the journal will complain about
 375  * that error until we've noted it down and cleared it.
 376  */
 377
 378 static void ext4_handle_error(struct super_block *sb)
 379 {
 380         if (sb->s_flags & MS_RDONLY)
 381                 return;
 382
 383         if (!test_opt(sb, ERRORS_CONT)) {
 384                 journal_t *journal = EXT4_SB(sb)->s_journal;
 385
 386                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 387                 if (journal)
 388                         jbd2_journal_abort(journal, -EIO);
 389         }
 390         if (test_opt(sb, ERRORS_RO)) {
 391                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 392                 /*
 393                  * Make sure updated value of ->s_mount_flags will be visible
 394                  * before ->s_flags update
 395                  */
 396                 smp_wmb();
 397                 sb->s_flags |= MS_RDONLY;
 398         }
 399         if (test_opt(sb, ERRORS_PANIC))
 400                 panic("EXT4-fs (device %s): panic forced after error\n",
 401                         sb->s_id);
 402 }
 403
 404 #define ext4_error_ratelimit(sb)                                        \
 405                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 406                              "EXT4-fs error")
 407
 408 void __ext4_error(struct super_block *sb, const char *function,
 409                   unsigned int line, const char *fmt, ...)
 410 {
 411         struct va_format vaf;
 412         va_list args;
 413
 414         if (ext4_error_ratelimit(sb)) {
 415                 va_start(args, fmt);
 416                 vaf.fmt = fmt;
 417                 vaf.va = &args;
 418                 printk(KERN_CRIT
 419                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 420                        sb->s_id, function, line, current->comm, &vaf);
 421                 va_end(args);
 422         }
 423         save_error_info(sb, function, line);
 424         ext4_handle_error(sb);
 425 }
 426
 427 void __ext4_error_inode(struct inode *inode, const char *function,
 428                         unsigned int line, ext4_fsblk_t block,
 429                         const char *fmt, ...)
 430 {
 431         va_list args;
 432         struct va_format vaf;
 433         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 434
 435         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 436         es->s_last_error_block = cpu_to_le64(block);
 437         if (ext4_error_ratelimit(inode->i_sb)) {
 438                 va_start(args, fmt);
 439                 vaf.fmt = fmt;
 440                 vaf.va = &args;
 441                 if (block)
 442                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 443                                "inode #%lu: block %llu: comm %s: %pV\n",
 444                                inode->i_sb->s_id, function, line, inode->i_ino,
 445                                block, current->comm, &vaf);
 446                 else
 447                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 448                                "inode #%lu: comm %s: %pV\n",
 449                                inode->i_sb->s_id, function, line, inode->i_ino,
 450                                current->comm, &vaf);
 451                 va_end(args);
 452         }
 453         save_error_info(inode->i_sb, function, line);
 454         ext4_handle_error(inode->i_sb);
 455 }
 456
 457 void __ext4_error_file(struct file *file, const char *function,
 458                        unsigned int line, ext4_fsblk_t block,
 459                        const char *fmt, ...)
 460 {
 461         va_list args;
 462         struct va_format vaf;
 463         struct ext4_super_block *es;
 464         struct inode *inode = file_inode(file);
 465         char pathname[80], *path;
 466
 467         es = EXT4_SB(inode->i_sb)->s_es;
 468         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 469         if (ext4_error_ratelimit(inode->i_sb)) {
 470                 path = d_path(&(file->f_path), pathname, sizeof(pathname));
 471                 if (IS_ERR(path))
 472                         path = "(unknown)";
 473                 va_start(args, fmt);
 474                 vaf.fmt = fmt;
 475                 vaf.va = &args;
 476                 if (block)
 477                         printk(KERN_CRIT
 478                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 479                                "block %llu: comm %s: path %s: %pV\n",
 480                                inode->i_sb->s_id, function, line, inode->i_ino,
 481                                block, current->comm, path, &vaf);
 482                 else
 483                         printk(KERN_CRIT
 484                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 485                                "comm %s: path %s: %pV\n",
 486                                inode->i_sb->s_id, function, line, inode->i_ino,
 487                                current->comm, path, &vaf);
 488                 va_end(args);
 489         }
 490         save_error_info(inode->i_sb, function, line);
 491         ext4_handle_error(inode->i_sb);
 492 }
 493
 494 const char *ext4_decode_error(struct super_block *sb, int errno,
 495                               char nbuf[16])
 496 {
 497         char *errstr = NULL;
 498
 499         switch (errno) {
 500         case -EIO:
 501                 errstr = "IO failure";
 502                 break;
 503         case -ENOMEM:
 504                 errstr = "Out of memory";
 505                 break;
 506         case -EROFS:
 507                 if (!sb || (EXT4_SB(sb)->s_journal &&
 508                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 509                         errstr = "Journal has aborted";
 510                 else
 511                         errstr = "Readonly filesystem";
 512                 break;
 513         default:
 514                 /* If the caller passed in an extra buffer for unknown
 515                  * errors, textualise them now.  Else we just return
 516                  * NULL. */
 517                 if (nbuf) {
 518                         /* Check for truncated error codes... */
 519                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 520                                 errstr = nbuf;
 521                 }
 522                 break;
 523         }
 524
 525         return errstr;
 526 }
 527
 528 /* __ext4_std_error decodes expected errors from journaling functions
 529  * automatically and invokes the appropriate error response.  */
 530
 531 void __ext4_std_error(struct super_block *sb, const char *function,
 532                       unsigned int line, int errno)
 533 {
 534         char nbuf[16];
 535         const char *errstr;
 536
 537         /* Special case: if the error is EROFS, and we're not already
 538          * inside a transaction, then there's really no point in logging
 539          * an error. */
 540         if (errno == -EROFS && journal_current_handle() == NULL &&
 541             (sb->s_flags & MS_RDONLY))
 542                 return;
 543
 544         if (ext4_error_ratelimit(sb)) {
 545                 errstr = ext4_decode_error(sb, errno, nbuf);
 546                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 547                        sb->s_id, function, line, errstr);
 548         }
 549
 550         save_error_info(sb, function, line);
 551         ext4_handle_error(sb);
 552 }
 553
 554 /*
 555  * ext4_abort is a much stronger failure handler than ext4_error.  The
 556  * abort function may be used to deal with unrecoverable failures such
 557  * as journal IO errors or ENOMEM at a critical moment in log management.
 558  *
 559  * We unconditionally force the filesystem into an ABORT|READONLY state,
 560  * unless the error response on the fs has been set to panic in which
 561  * case we take the easy way out and panic immediately.
 562  */
 563
 564 void __ext4_abort(struct super_block *sb, const char *function,
 565                 unsigned int line, const char *fmt, ...)
 566 {
 567         va_list args;
 568
 569         save_error_info(sb, function, line);
 570         va_start(args, fmt);
 571         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 572                function, line);
 573         vprintk(fmt, args);
 574         printk("\n");
 575         va_end(args);
 576
 577         if ((sb->s_flags & MS_RDONLY) == 0) {
 578                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 579                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 580                 /*
 581                  * Make sure updated value of ->s_mount_flags will be visible
 582                  * before ->s_flags update
 583                  */
 584                 smp_wmb();
 585                 sb->s_flags |= MS_RDONLY;
 586                 if (EXT4_SB(sb)->s_journal)
 587                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 588                 save_error_info(sb, function, line);
 589         }
 590         if (test_opt(sb, ERRORS_PANIC))
 591                 panic("EXT4-fs panic from previous error\n");
 592 }
 593
 594 void __ext4_msg(struct super_block *sb,
 595                 const char *prefix, const char *fmt, ...)
 596 {
 597         struct va_format vaf;
 598         va_list args;
 599
 600         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 601                 return;
 602
 603         va_start(args, fmt);
 604         vaf.fmt = fmt;
 605         vaf.va = &args;
 606         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 607         va_end(args);
 608 }
 609
 610 void __ext4_warning(struct super_block *sb, const char *function,
 611                     unsigned int line, const char *fmt, ...)
 612 {
 613         struct va_format vaf;
 614         va_list args;
 615
 616         if (!___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
 617                           "EXT4-fs warning"))
 618                 return;
 619
 620         va_start(args, fmt);
 621         vaf.fmt = fmt;
 622         vaf.va = &args;
 623         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 624                sb->s_id, function, line, &vaf);
 625         va_end(args);
 626 }
 627
 628 void __ext4_grp_locked_error(const char *function, unsigned int line,
 629                              struct super_block *sb, ext4_group_t grp,
 630                              unsigned long ino, ext4_fsblk_t block,
 631                              const char *fmt, ...)
 632 __releases(bitlock)
 633 __acquires(bitlock)
 634 {
 635         struct va_format vaf;
 636         va_list args;
 637         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 638
 639         es->s_last_error_ino = cpu_to_le32(ino);
 640         es->s_last_error_block = cpu_to_le64(block);
 641         __save_error_info(sb, function, line);
 642
 643         if (ext4_error_ratelimit(sb)) {
 644                 va_start(args, fmt);
 645                 vaf.fmt = fmt;
 646                 vaf.va = &args;
 647                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 648                        sb->s_id, function, line, grp);
 649                 if (ino)
 650                         printk(KERN_CONT "inode %lu: ", ino);
 651                 if (block)
 652                         printk(KERN_CONT "block %llu:",
 653                                (unsigned long long) block);
 654                 printk(KERN_CONT "%pV\n", &vaf);
 655                 va_end(args);
 656         }
 657
 658         if (test_opt(sb, ERRORS_CONT)) {
 659                 ext4_commit_super(sb, 0);
 660                 return;
 661         }
 662
 663         ext4_unlock_group(sb, grp);
 664         ext4_handle_error(sb);
 665         /*
 666          * We only get here in the ERRORS_RO case; relocking the group
 667          * may be dangerous, but nothing bad will happen since the
 668          * filesystem will have already been marked read/only and the
 669          * journal has been aborted.  We return 1 as a hint to callers
 670          * who might what to use the return value from
 671          * ext4_grp_locked_error() to distinguish between the
 672          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 673          * aggressively from the ext4 function in question, with a
 674          * more appropriate error code.
 675          */
 676         ext4_lock_group(sb, grp);
 677         return;
 678 }
 679
 680 void ext4_update_dynamic_rev(struct super_block *sb)
 681 {
 682         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 683
 684         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 685                 return;
 686
 687         ext4_warning(sb,
 688                      "updating to rev %d because of new feature flag, "
 689                      "running e2fsck is recommended",
 690                      EXT4_DYNAMIC_REV);
 691
 692         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 693         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 694         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 695         /* leave es->s_feature_*compat flags alone */
 696         /* es->s_uuid will be set by e2fsck if empty */
 697
 698         /*
 699          * The rest of the superblock fields should be zero, and if not it
 700          * means they are likely already in use, so leave them alone.  We
 701          * can leave it up to e2fsck to clean up any inconsistencies there.
 702          */
 703 }
 704
 705 /*
 706  * Open the external journal device
 707  */
 708 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 709 {
 710         struct block_device *bdev;
 711         char b[BDEVNAME_SIZE];
 712
 713         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 714         if (IS_ERR(bdev))
 715                 goto fail;
 716         return bdev;
 717
 718 fail:
 719         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 720                         __bdevname(dev, b), PTR_ERR(bdev));
 721         return NULL;
 722 }
 723
 724 /*
 725  * Release the journal device
 726  */
 727 static void ext4_blkdev_put(struct block_device *bdev)
 728 {
 729         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 730 }
 731
 732 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 733 {
 734         struct block_device *bdev;
 735         bdev = sbi->journal_bdev;
 736         if (bdev) {
 737                 ext4_blkdev_put(bdev);
 738                 sbi->journal_bdev = NULL;
 739         }
 740 }
 741
 742 static inline struct inode *orphan_list_entry(struct list_head *l)
 743 {
 744         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 745 }
 746
 747 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 748 {
 749         struct list_head *l;
 750
 751         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 752                  le32_to_cpu(sbi->s_es->s_last_orphan));
 753
 754         printk(KERN_ERR "sb_info orphan list:\n");
 755         list_for_each(l, &sbi->s_orphan) {
 756                 struct inode *inode = orphan_list_entry(l);
 757                 printk(KERN_ERR "  "
 758                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 759                        inode->i_sb->s_id, inode->i_ino, inode,
 760                        inode->i_mode, inode->i_nlink,
 761                        NEXT_ORPHAN(inode));
 762         }
 763 }
 764
 765 static void ext4_put_super(struct super_block *sb)
 766 {
 767         struct ext4_sb_info *sbi = EXT4_SB(sb);
 768         struct ext4_super_block *es = sbi->s_es;
 769         int i, err;
 770
 771         ext4_unregister_li_request(sb);
 772         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 773
 774         flush_workqueue(sbi->rsv_conversion_wq);
 775         destroy_workqueue(sbi->rsv_conversion_wq);
 776
 777         if (sbi->s_journal) {
 778                 err = jbd2_journal_destroy(sbi->s_journal);
 779                 sbi->s_journal = NULL;
 780                 if (err < 0)
 781                         ext4_abort(sb, "Couldn't clean up the journal");
 782         }
 783
 784         ext4_es_unregister_shrinker(sbi);
 785         del_timer_sync(&sbi->s_err_report);
 786         ext4_release_system_zone(sb);
 787         ext4_mb_release(sb);
 788         ext4_ext_release(sb);
 789         ext4_xattr_put_super(sb);
 790
 791         if (!(sb->s_flags & MS_RDONLY)) {
 792                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 793                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 794         }
 795         if (!(sb->s_flags & MS_RDONLY))
 796                 ext4_commit_super(sb, 1);
 797
 798         if (sbi->s_proc) {
 799                 remove_proc_entry("options", sbi->s_proc);
 800                 remove_proc_entry(sb->s_id, ext4_proc_root);
 801         }
 802         kobject_del(&sbi->s_kobj);
 803
 804         for (i = 0; i < sbi->s_gdb_count; i++)
 805                 brelse(sbi->s_group_desc[i]);
 806         kvfree(sbi->s_group_desc);
 807         kvfree(sbi->s_flex_groups);
 808         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 809         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 810         percpu_counter_destroy(&sbi->s_dirs_counter);
 811         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 812         brelse(sbi->s_sbh);
 813 #ifdef CONFIG_QUOTA
 814         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 815                 kfree(sbi->s_qf_names[i]);
 816 #endif
 817
 818         /* Debugging code just in case the in-memory inode orphan list
 819          * isn't empty.  The on-disk one can be non-empty if we've
 820          * detected an error and taken the fs readonly, but the
 821          * in-memory list had better be clean by this point. */
 822         if (!list_empty(&sbi->s_orphan))
 823                 dump_orphan_list(sb, sbi);
 824         J_ASSERT(list_empty(&sbi->s_orphan));
 825
 826         sync_blockdev(sb->s_bdev);
 827         invalidate_bdev(sb->s_bdev);
 828         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 829                 /*
 830                  * Invalidate the journal device's buffers.  We don't want them
 831                  * floating about in memory - the physical journal device may
 832                  * hotswapped, and it breaks the `ro-after' testing code.
 833                  */
 834                 sync_blockdev(sbi->journal_bdev);
 835                 invalidate_bdev(sbi->journal_bdev);
 836                 ext4_blkdev_remove(sbi);
 837         }
 838         if (sbi->s_mb_cache) {
 839                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 840                 sbi->s_mb_cache = NULL;
 841         }
 842         if (sbi->s_mmp_tsk)
 843                 kthread_stop(sbi->s_mmp_tsk);
 844         sb->s_fs_info = NULL;
 845         /*
 846          * Now that we are completely done shutting down the
 847          * superblock, we need to actually destroy the kobject.
 848          */
 849         kobject_put(&sbi->s_kobj);
 850         wait_for_completion(&sbi->s_kobj_unregister);
 851         if (sbi->s_chksum_driver)
 852                 crypto_free_shash(sbi->s_chksum_driver);
 853         kfree(sbi->s_blockgroup_lock);
 854         kfree(sbi);
 855 }
 856
 857 static struct kmem_cache *ext4_inode_cachep;
 858
 859 /*
 860  * Called inside transaction, so use GFP_NOFS
 861  */
 862 static struct inode *ext4_alloc_inode(struct super_block *sb)
 863 {
 864         struct ext4_inode_info *ei;
 865
 866         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 867         if (!ei)
 868                 return NULL;
 869
 870         ei->vfs_inode.i_version = 1;
 871         spin_lock_init(&ei->i_raw_lock);
 872         INIT_LIST_HEAD(&ei->i_prealloc_list);
 873         spin_lock_init(&ei->i_prealloc_lock);
 874         ext4_es_init_tree(&ei->i_es_tree);
 875         rwlock_init(&ei->i_es_lock);
 876         INIT_LIST_HEAD(&ei->i_es_list);
 877         ei->i_es_all_nr = 0;
 878         ei->i_es_shk_nr = 0;
 879         ei->i_es_shrink_lblk = 0;
 880         ei->i_reserved_data_blocks = 0;
 881         ei->i_reserved_meta_blocks = 0;
 882         ei->i_allocated_meta_blocks = 0;
 883         ei->i_da_metadata_calc_len = 0;
 884         ei->i_da_metadata_calc_last_lblock = 0;
 885         spin_lock_init(&(ei->i_block_reservation_lock));
 886 #ifdef CONFIG_QUOTA
 887         ei->i_reserved_quota = 0;
 888         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 889 #endif
 890         ei->jinode = NULL;
 891         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 892         spin_lock_init(&ei->i_completed_io_lock);
 893         ei->i_sync_tid = 0;
 894         ei->i_datasync_tid = 0;
 895         atomic_set(&ei->i_ioend_count, 0);
 896         atomic_set(&ei->i_unwritten, 0);
 897         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 898 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 899         ei->i_encryption_key.mode = EXT4_ENCRYPTION_MODE_INVALID;
 900 #endif
 901
 902         return &ei->vfs_inode;
 903 }
 904
 905 static int ext4_drop_inode(struct inode *inode)
 906 {
 907         int drop = generic_drop_inode(inode);
 908
 909         trace_ext4_drop_inode(inode, drop);
 910         return drop;
 911 }
 912
 913 static void ext4_i_callback(struct rcu_head *head)
 914 {
 915         struct inode *inode = container_of(head, struct inode, i_rcu);
 916         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 917 }
 918
 919 static void ext4_destroy_inode(struct inode *inode)
 920 {
 921         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 922                 ext4_msg(inode->i_sb, KERN_ERR,
 923                          "Inode %lu (%p): orphan list check failed!",
 924                          inode->i_ino, EXT4_I(inode));
 925                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 926                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 927                                 true);
 928                 dump_stack();
 929         }
 930         call_rcu(&inode->i_rcu, ext4_i_callback);
 931 }
 932
 933 static void init_once(void *foo)
 934 {
 935         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 936
 937         INIT_LIST_HEAD(&ei->i_orphan);
 938         init_rwsem(&ei->xattr_sem);
 939         init_rwsem(&ei->i_data_sem);
 940         inode_init_once(&ei->vfs_inode);
 941 }
 942
 943 static int __init init_inodecache(void)
 944 {
 945         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 946                                              sizeof(struct ext4_inode_info),
 947                                              0, (SLAB_RECLAIM_ACCOUNT|
 948                                                 SLAB_MEM_SPREAD),
 949                                              init_once);
 950         if (ext4_inode_cachep == NULL)
 951                 return -ENOMEM;
 952         return 0;
 953 }
 954
 955 static void destroy_inodecache(void)
 956 {
 957         /*
 958          * Make sure all delayed rcu free inodes are flushed before we
 959          * destroy cache.
 960          */
 961         rcu_barrier();
 962         kmem_cache_destroy(ext4_inode_cachep);
 963 }
 964
 965 void ext4_clear_inode(struct inode *inode)
 966 {
 967         invalidate_inode_buffers(inode);
 968         clear_inode(inode);
 969         dquot_drop(inode);
 970         ext4_discard_preallocations(inode);
 971         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 972         if (EXT4_I(inode)->jinode) {
 973                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 974                                                EXT4_I(inode)->jinode);
 975                 jbd2_free_inode(EXT4_I(inode)->jinode);
 976                 EXT4_I(inode)->jinode = NULL;
 977         }
 978 }
 979
 980 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 981                                         u64 ino, u32 generation)
 982 {
 983         struct inode *inode;
 984
 985         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 986                 return ERR_PTR(-ESTALE);
 987         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 988                 return ERR_PTR(-ESTALE);
 989
 990         /* iget isn't really right if the inode is currently unallocated!!
 991          *
 992          * ext4_read_inode will return a bad_inode if the inode had been
 993          * deleted, so we should be safe.
 994          *
 995          * Currently we don't know the generation for parent directory, so
 996          * a generation of 0 means "accept any"
 997          */
 998         inode = ext4_iget_normal(sb, ino);
 999         if (IS_ERR(inode))
1000                 return ERR_CAST(inode);
1001         if (generation && inode->i_generation != generation) {
1002                 iput(inode);
1003                 return ERR_PTR(-ESTALE);
1004         }
1005
1006         return inode;
1007 }
1008
1009 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1010                                         int fh_len, int fh_type)
1011 {
1012         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1013                                     ext4_nfs_get_inode);
1014 }
1015
1016 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1017                                         int fh_len, int fh_type)
1018 {
1019         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1020                                     ext4_nfs_get_inode);
1021 }
1022
1023 /*
1024  * Try to release metadata pages (indirect blocks, directories) which are
1025  * mapped via the block device.  Since these pages could have journal heads
1026  * which would prevent try_to_free_buffers() from freeing them, we must use
1027  * jbd2 layer's try_to_free_buffers() function to release them.
1028  */
1029 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1030                                  gfp_t wait)
1031 {
1032         journal_t *journal = EXT4_SB(sb)->s_journal;
1033
1034         WARN_ON(PageChecked(page));
1035         if (!page_has_buffers(page))
1036                 return 0;
1037         if (journal)
1038                 return jbd2_journal_try_to_free_buffers(journal, page,
1039                                                         wait & ~__GFP_WAIT);
1040         return try_to_free_buffers(page);
1041 }
1042
1043 #ifdef CONFIG_QUOTA
1044 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1045 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1046
1047 static int ext4_write_dquot(struct dquot *dquot);
1048 static int ext4_acquire_dquot(struct dquot *dquot);
1049 static int ext4_release_dquot(struct dquot *dquot);
1050 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1051 static int ext4_write_info(struct super_block *sb, int type);
1052 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1053                          struct path *path);
1054 static int ext4_quota_off(struct super_block *sb, int type);
1055 static int ext4_quota_on_mount(struct super_block *sb, int type);
1056 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1057                                size_t len, loff_t off);
1058 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1059                                 const char *data, size_t len, loff_t off);
1060 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1061                              unsigned int flags);
1062 static int ext4_enable_quotas(struct super_block *sb);
1063
1064 static struct dquot **ext4_get_dquots(struct inode *inode)
1065 {
1066         return EXT4_I(inode)->i_dquot;
1067 }
1068
1069 static const struct dquot_operations ext4_quota_operations = {
1070         .get_reserved_space = ext4_get_reserved_space,
1071         .write_dquot    = ext4_write_dquot,
1072         .acquire_dquot  = ext4_acquire_dquot,
1073         .release_dquot  = ext4_release_dquot,
1074         .mark_dirty     = ext4_mark_dquot_dirty,
1075         .write_info     = ext4_write_info,
1076         .alloc_dquot    = dquot_alloc,
1077         .destroy_dquot  = dquot_destroy,
1078 };
1079
1080 static const struct quotactl_ops ext4_qctl_operations = {
1081         .quota_on       = ext4_quota_on,
1082         .quota_off      = ext4_quota_off,
1083         .quota_sync     = dquot_quota_sync,
1084         .get_state      = dquot_get_state,
1085         .set_info       = dquot_set_dqinfo,
1086         .get_dqblk      = dquot_get_dqblk,
1087         .set_dqblk      = dquot_set_dqblk
1088 };
1089 #endif
1090
1091 static const struct super_operations ext4_sops = {
1092         .alloc_inode    = ext4_alloc_inode,
1093         .destroy_inode  = ext4_destroy_inode,
1094         .write_inode    = ext4_write_inode,
1095         .dirty_inode    = ext4_dirty_inode,
1096         .drop_inode     = ext4_drop_inode,
1097         .evict_inode    = ext4_evict_inode,
1098         .put_super      = ext4_put_super,
1099         .sync_fs        = ext4_sync_fs,
1100         .freeze_fs      = ext4_freeze,
1101         .unfreeze_fs    = ext4_unfreeze,
1102         .statfs         = ext4_statfs,
1103         .remount_fs     = ext4_remount,
1104         .show_options   = ext4_show_options,
1105 #ifdef CONFIG_QUOTA
1106         .quota_read     = ext4_quota_read,
1107         .quota_write    = ext4_quota_write,
1108         .get_dquots     = ext4_get_dquots,
1109 #endif
1110         .bdev_try_to_free_page = bdev_try_to_free_page,
1111 };
1112
1113 static const struct export_operations ext4_export_ops = {
1114         .fh_to_dentry = ext4_fh_to_dentry,
1115         .fh_to_parent = ext4_fh_to_parent,
1116         .get_parent = ext4_get_parent,
1117 };
1118
1119 enum {
1120         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1121         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1122         Opt_nouid32, Opt_debug, Opt_removed,
1123         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1124         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1125         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1126         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1127         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1128         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1129         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1130         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1131         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1132         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1133         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1134         Opt_lazytime, Opt_nolazytime,
1135         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1136         Opt_inode_readahead_blks, Opt_journal_ioprio,
1137         Opt_dioread_nolock, Opt_dioread_lock,
1138         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1139         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1140 };
1141
1142 static const match_table_t tokens = {
1143         {Opt_bsd_df, "bsddf"},
1144         {Opt_minix_df, "minixdf"},
1145         {Opt_grpid, "grpid"},
1146         {Opt_grpid, "bsdgroups"},
1147         {Opt_nogrpid, "nogrpid"},
1148         {Opt_nogrpid, "sysvgroups"},
1149         {Opt_resgid, "resgid=%u"},
1150         {Opt_resuid, "resuid=%u"},
1151         {Opt_sb, "sb=%u"},
1152         {Opt_err_cont, "errors=continue"},
1153         {Opt_err_panic, "errors=panic"},
1154         {Opt_err_ro, "errors=remount-ro"},
1155         {Opt_nouid32, "nouid32"},
1156         {Opt_debug, "debug"},
1157         {Opt_removed, "oldalloc"},
1158         {Opt_removed, "orlov"},
1159         {Opt_user_xattr, "user_xattr"},
1160         {Opt_nouser_xattr, "nouser_xattr"},
1161         {Opt_acl, "acl"},
1162         {Opt_noacl, "noacl"},
1163         {Opt_noload, "norecovery"},
1164         {Opt_noload, "noload"},
1165         {Opt_removed, "nobh"},
1166         {Opt_removed, "bh"},
1167         {Opt_commit, "commit=%u"},
1168         {Opt_min_batch_time, "min_batch_time=%u"},
1169         {Opt_max_batch_time, "max_batch_time=%u"},
1170         {Opt_journal_dev, "journal_dev=%u"},
1171         {Opt_journal_path, "journal_path=%s"},
1172         {Opt_journal_checksum, "journal_checksum"},
1173         {Opt_nojournal_checksum, "nojournal_checksum"},
1174         {Opt_journal_async_commit, "journal_async_commit"},
1175         {Opt_abort, "abort"},
1176         {Opt_data_journal, "data=journal"},
1177         {Opt_data_ordered, "data=ordered"},
1178         {Opt_data_writeback, "data=writeback"},
1179         {Opt_data_err_abort, "data_err=abort"},
1180         {Opt_data_err_ignore, "data_err=ignore"},
1181         {Opt_offusrjquota, "usrjquota="},
1182         {Opt_usrjquota, "usrjquota=%s"},
1183         {Opt_offgrpjquota, "grpjquota="},
1184         {Opt_grpjquota, "grpjquota=%s"},
1185         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1186         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1187         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1188         {Opt_grpquota, "grpquota"},
1189         {Opt_noquota, "noquota"},
1190         {Opt_quota, "quota"},
1191         {Opt_usrquota, "usrquota"},
1192         {Opt_barrier, "barrier=%u"},
1193         {Opt_barrier, "barrier"},
1194         {Opt_nobarrier, "nobarrier"},
1195         {Opt_i_version, "i_version"},
1196         {Opt_dax, "dax"},
1197         {Opt_stripe, "stripe=%u"},
1198         {Opt_delalloc, "delalloc"},
1199         {Opt_lazytime, "lazytime"},
1200         {Opt_nolazytime, "nolazytime"},
1201         {Opt_nodelalloc, "nodelalloc"},
1202         {Opt_removed, "mblk_io_submit"},
1203         {Opt_removed, "nomblk_io_submit"},
1204         {Opt_block_validity, "block_validity"},
1205         {Opt_noblock_validity, "noblock_validity"},
1206         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1207         {Opt_journal_ioprio, "journal_ioprio=%u"},
1208         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1209         {Opt_auto_da_alloc, "auto_da_alloc"},
1210         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1211         {Opt_dioread_nolock, "dioread_nolock"},
1212         {Opt_dioread_lock, "dioread_lock"},
1213         {Opt_discard, "discard"},
1214         {Opt_nodiscard, "nodiscard"},
1215         {Opt_init_itable, "init_itable=%u"},
1216         {Opt_init_itable, "init_itable"},
1217         {Opt_noinit_itable, "noinit_itable"},
1218         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1219         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1220         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1221         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1222         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1223         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1224         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1225         {Opt_err, NULL},
1226 };
1227
1228 static ext4_fsblk_t get_sb_block(void **data)
1229 {
1230         ext4_fsblk_t    sb_block;
1231         char            *options = (char *) *data;
1232
1233         if (!options || strncmp(options, "sb=", 3) != 0)
1234                 return 1;       /* Default location */
1235
1236         options += 3;
1237         /* TODO: use simple_strtoll with >32bit ext4 */
1238         sb_block = simple_strtoul(options, &options, 0);
1239         if (*options && *options != ',') {
1240                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1241                        (char *) *data);
1242                 return 1;
1243         }
1244         if (*options == ',')
1245                 options++;
1246         *data = (void *) options;
1247
1248         return sb_block;
1249 }
1250
1251 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1252 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1253         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1254
1255 #ifdef CONFIG_QUOTA
1256 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1257 {
1258         struct ext4_sb_info *sbi = EXT4_SB(sb);
1259         char *qname;
1260         int ret = -1;
1261
1262         if (sb_any_quota_loaded(sb) &&
1263                 !sbi->s_qf_names[qtype]) {
1264                 ext4_msg(sb, KERN_ERR,
1265                         "Cannot change journaled "
1266                         "quota options when quota turned on");
1267                 return -1;
1268         }
1269         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1270                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1271                          "when QUOTA feature is enabled");
1272                 return -1;
1273         }
1274         qname = match_strdup(args);
1275         if (!qname) {
1276                 ext4_msg(sb, KERN_ERR,
1277                         "Not enough memory for storing quotafile name");
1278                 return -1;
1279         }
1280         if (sbi->s_qf_names[qtype]) {
1281                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1282                         ret = 1;
1283                 else
1284                         ext4_msg(sb, KERN_ERR,
1285                                  "%s quota file already specified",
1286                                  QTYPE2NAME(qtype));
1287                 goto errout;
1288         }
1289         if (strchr(qname, '/')) {
1290                 ext4_msg(sb, KERN_ERR,
1291                         "quotafile must be on filesystem root");
1292                 goto errout;
1293         }
1294         sbi->s_qf_names[qtype] = qname;
1295         set_opt(sb, QUOTA);
1296         return 1;
1297 errout:
1298         kfree(qname);
1299         return ret;
1300 }
1301
1302 static int clear_qf_name(struct super_block *sb, int qtype)
1303 {
1304
1305         struct ext4_sb_info *sbi = EXT4_SB(sb);
1306
1307         if (sb_any_quota_loaded(sb) &&
1308                 sbi->s_qf_names[qtype]) {
1309                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1310                         " when quota turned on");
1311                 return -1;
1312         }
1313         kfree(sbi->s_qf_names[qtype]);
1314         sbi->s_qf_names[qtype] = NULL;
1315         return 1;
1316 }
1317 #endif
1318
1319 #define MOPT_SET        0x0001
1320 #define MOPT_CLEAR      0x0002
1321 #define MOPT_NOSUPPORT  0x0004
1322 #define MOPT_EXPLICIT   0x0008
1323 #define MOPT_CLEAR_ERR  0x0010
1324 #define MOPT_GTE0       0x0020
1325 #ifdef CONFIG_QUOTA
1326 #define MOPT_Q          0
1327 #define MOPT_QFMT       0x0040
1328 #else
1329 #define MOPT_Q          MOPT_NOSUPPORT
1330 #define MOPT_QFMT       MOPT_NOSUPPORT
1331 #endif
1332 #define MOPT_DATAJ      0x0080
1333 #define MOPT_NO_EXT2    0x0100
1334 #define MOPT_NO_EXT3    0x0200
1335 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1336 #define MOPT_STRING     0x0400
1337
1338 static const struct mount_opts {
1339         int     token;
1340         int     mount_opt;
1341         int     flags;
1342 } ext4_mount_opts[] = {
1343         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1344         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1345         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1346         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1347         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1348         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1349         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1350          MOPT_EXT4_ONLY | MOPT_SET},
1351         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1352          MOPT_EXT4_ONLY | MOPT_CLEAR},
1353         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1354         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1355         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1356          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1357         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1358          MOPT_EXT4_ONLY | MOPT_CLEAR},
1359         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1360          MOPT_EXT4_ONLY | MOPT_CLEAR},
1361         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1362          MOPT_EXT4_ONLY | MOPT_SET},
1363         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1364                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1365          MOPT_EXT4_ONLY | MOPT_SET},
1366         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1367         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1368         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1369         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1370         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1371          MOPT_NO_EXT2 | MOPT_SET},
1372         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1373          MOPT_NO_EXT2 | MOPT_CLEAR},
1374         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1375         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1376         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1377         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1378         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1379         {Opt_commit, 0, MOPT_GTE0},
1380         {Opt_max_batch_time, 0, MOPT_GTE0},
1381         {Opt_min_batch_time, 0, MOPT_GTE0},
1382         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1383         {Opt_init_itable, 0, MOPT_GTE0},
1384         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1385         {Opt_stripe, 0, MOPT_GTE0},
1386         {Opt_resuid, 0, MOPT_GTE0},
1387         {Opt_resgid, 0, MOPT_GTE0},
1388         {Opt_journal_dev, 0, MOPT_GTE0},
1389         {Opt_journal_path, 0, MOPT_STRING},
1390         {Opt_journal_ioprio, 0, MOPT_GTE0},
1391         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1392         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1393         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1394          MOPT_NO_EXT2 | MOPT_DATAJ},
1395         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1396         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1397 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1398         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1399         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1400 #else
1401         {Opt_acl, 0, MOPT_NOSUPPORT},
1402         {Opt_noacl, 0, MOPT_NOSUPPORT},
1403 #endif
1404         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1405         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1406         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1407         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1408                                                         MOPT_SET | MOPT_Q},
1409         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1410                                                         MOPT_SET | MOPT_Q},
1411         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1412                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1413         {Opt_usrjquota, 0, MOPT_Q},
1414         {Opt_grpjquota, 0, MOPT_Q},
1415         {Opt_offusrjquota, 0, MOPT_Q},
1416         {Opt_offgrpjquota, 0, MOPT_Q},
1417         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1418         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1419         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1420         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1421         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1422         {Opt_err, 0, 0}
1423 };
1424
1425 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1426                             substring_t *args, unsigned long *journal_devnum,
1427                             unsigned int *journal_ioprio, int is_remount)
1428 {
1429         struct ext4_sb_info *sbi = EXT4_SB(sb);
1430         const struct mount_opts *m;
1431         kuid_t uid;
1432         kgid_t gid;
1433         int arg = 0;
1434
1435 #ifdef CONFIG_QUOTA
1436         if (token == Opt_usrjquota)
1437                 return set_qf_name(sb, USRQUOTA, &args[0]);
1438         else if (token == Opt_grpjquota)
1439                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1440         else if (token == Opt_offusrjquota)
1441                 return clear_qf_name(sb, USRQUOTA);
1442         else if (token == Opt_offgrpjquota)
1443                 return clear_qf_name(sb, GRPQUOTA);
1444 #endif
1445         switch (token) {
1446         case Opt_noacl:
1447         case Opt_nouser_xattr:
1448                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1449                 break;
1450         case Opt_sb:
1451                 return 1;       /* handled by get_sb_block() */
1452         case Opt_removed:
1453                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1454                 return 1;
1455         case Opt_abort:
1456                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1457                 return 1;
1458         case Opt_i_version:
1459                 sb->s_flags |= MS_I_VERSION;
1460                 return 1;
1461         case Opt_lazytime:
1462                 sb->s_flags |= MS_LAZYTIME;
1463                 return 1;
1464         case Opt_nolazytime:
1465                 sb->s_flags &= ~MS_LAZYTIME;
1466                 return 1;
1467         }
1468
1469         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1470                 if (token == m->token)
1471                         break;
1472
1473         if (m->token == Opt_err) {
1474                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1475                          "or missing value", opt);
1476                 return -1;
1477         }
1478
1479         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1480                 ext4_msg(sb, KERN_ERR,
1481                          "Mount option \"%s\" incompatible with ext2", opt);
1482                 return -1;
1483         }
1484         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1485                 ext4_msg(sb, KERN_ERR,
1486                          "Mount option \"%s\" incompatible with ext3", opt);
1487                 return -1;
1488         }
1489
1490         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1491                 return -1;
1492         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1493                 return -1;
1494         if (m->flags & MOPT_EXPLICIT)
1495                 set_opt2(sb, EXPLICIT_DELALLOC);
1496         if (m->flags & MOPT_CLEAR_ERR)
1497                 clear_opt(sb, ERRORS_MASK);
1498         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1499                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1500                          "options when quota turned on");
1501                 return -1;
1502         }
1503
1504         if (m->flags & MOPT_NOSUPPORT) {
1505                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1506         } else if (token == Opt_commit) {
1507                 if (arg == 0)
1508                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1509                 sbi->s_commit_interval = HZ * arg;
1510         } else if (token == Opt_max_batch_time) {
1511                 sbi->s_max_batch_time = arg;
1512         } else if (token == Opt_min_batch_time) {
1513                 sbi->s_min_batch_time = arg;
1514         } else if (token == Opt_inode_readahead_blks) {
1515                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1516                         ext4_msg(sb, KERN_ERR,
1517                                  "EXT4-fs: inode_readahead_blks must be "
1518                                  "0 or a power of 2 smaller than 2^31");
1519                         return -1;
1520                 }
1521                 sbi->s_inode_readahead_blks = arg;
1522         } else if (token == Opt_init_itable) {
1523                 set_opt(sb, INIT_INODE_TABLE);
1524                 if (!args->from)
1525                         arg = EXT4_DEF_LI_WAIT_MULT;
1526                 sbi->s_li_wait_mult = arg;
1527         } else if (token == Opt_max_dir_size_kb) {
1528                 sbi->s_max_dir_size_kb = arg;
1529         } else if (token == Opt_stripe) {
1530                 sbi->s_stripe = arg;
1531         } else if (token == Opt_resuid) {
1532                 uid = make_kuid(current_user_ns(), arg);
1533                 if (!uid_valid(uid)) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1535                         return -1;
1536                 }
1537                 sbi->s_resuid = uid;
1538         } else if (token == Opt_resgid) {
1539                 gid = make_kgid(current_user_ns(), arg);
1540                 if (!gid_valid(gid)) {
1541                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1542                         return -1;
1543                 }
1544                 sbi->s_resgid = gid;
1545         } else if (token == Opt_journal_dev) {
1546                 if (is_remount) {
1547                         ext4_msg(sb, KERN_ERR,
1548                                  "Cannot specify journal on remount");
1549                         return -1;
1550                 }
1551                 *journal_devnum = arg;
1552         } else if (token == Opt_journal_path) {
1553                 char *journal_path;
1554                 struct inode *journal_inode;
1555                 struct path path;
1556                 int error;
1557
1558                 if (is_remount) {
1559                         ext4_msg(sb, KERN_ERR,
1560                                  "Cannot specify journal on remount");
1561                         return -1;
1562                 }
1563                 journal_path = match_strdup(&args[0]);
1564                 if (!journal_path) {
1565                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1566                                 "journal device string");
1567                         return -1;
1568                 }
1569
1570                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1571                 if (error) {
1572                         ext4_msg(sb, KERN_ERR, "error: could not find "
1573                                 "journal device path: error %d", error);
1574                         kfree(journal_path);
1575                         return -1;
1576                 }
1577
1578                 journal_inode = d_inode(path.dentry);
1579                 if (!S_ISBLK(journal_inode->i_mode)) {
1580                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1581                                 "is not a block device", journal_path);
1582                         path_put(&path);
1583                         kfree(journal_path);
1584                         return -1;
1585                 }
1586
1587                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1588                 path_put(&path);
1589                 kfree(journal_path);
1590         } else if (token == Opt_journal_ioprio) {
1591                 if (arg > 7) {
1592                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1593                                  " (must be 0-7)");
1594                         return -1;
1595                 }
1596                 *journal_ioprio =
1597                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1598         } else if (token == Opt_test_dummy_encryption) {
1599 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1600                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1601                 ext4_msg(sb, KERN_WARNING,
1602                          "Test dummy encryption mode enabled");
1603 #else
1604                 ext4_msg(sb, KERN_WARNING,
1605                          "Test dummy encryption mount option ignored");
1606 #endif
1607         } else if (m->flags & MOPT_DATAJ) {
1608                 if (is_remount) {
1609                         if (!sbi->s_journal)
1610                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1611                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1612                                 ext4_msg(sb, KERN_ERR,
1613                                          "Cannot change data mode on remount");
1614                                 return -1;
1615                         }
1616                 } else {
1617                         clear_opt(sb, DATA_FLAGS);
1618                         sbi->s_mount_opt |= m->mount_opt;
1619                 }
1620 #ifdef CONFIG_QUOTA
1621         } else if (m->flags & MOPT_QFMT) {
1622                 if (sb_any_quota_loaded(sb) &&
1623                     sbi->s_jquota_fmt != m->mount_opt) {
1624                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1625                                  "quota options when quota turned on");
1626                         return -1;
1627                 }
1628                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1629                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1630                         ext4_msg(sb, KERN_ERR,
1631                                  "Cannot set journaled quota options "
1632                                  "when QUOTA feature is enabled");
1633                         return -1;
1634                 }
1635                 sbi->s_jquota_fmt = m->mount_opt;
1636 #endif
1637 #ifndef CONFIG_FS_DAX
1638         } else if (token == Opt_dax) {
1639                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1640                 return -1;
1641 #endif
1642         } else {
1643                 if (!args->from)
1644                         arg = 1;
1645                 if (m->flags & MOPT_CLEAR)
1646                         arg = !arg;
1647                 else if (unlikely(!(m->flags & MOPT_SET))) {
1648                         ext4_msg(sb, KERN_WARNING,
1649                                  "buggy handling of option %s", opt);
1650                         WARN_ON(1);
1651                         return -1;
1652                 }
1653                 if (arg != 0)
1654                         sbi->s_mount_opt |= m->mount_opt;
1655                 else
1656                         sbi->s_mount_opt &= ~m->mount_opt;
1657         }
1658         return 1;
1659 }
1660
1661 static int parse_options(char *options, struct super_block *sb,
1662                          unsigned long *journal_devnum,
1663                          unsigned int *journal_ioprio,
1664                          int is_remount)
1665 {
1666         struct ext4_sb_info *sbi = EXT4_SB(sb);
1667         char *p;
1668         substring_t args[MAX_OPT_ARGS];
1669         int token;
1670
1671         if (!options)
1672                 return 1;
1673
1674         while ((p = strsep(&options, ",")) != NULL) {
1675                 if (!*p)
1676                         continue;
1677                 /*
1678                  * Initialize args struct so we know whether arg was
1679                  * found; some options take optional arguments.
1680                  */
1681                 args[0].to = args[0].from = NULL;
1682                 token = match_token(p, tokens, args);
1683                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1684                                      journal_ioprio, is_remount) < 0)
1685                         return 0;
1686         }
1687 #ifdef CONFIG_QUOTA
1688         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1689             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1690                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1691                          "feature is enabled");
1692                 return 0;
1693         }
1694         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1695                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1696                         clear_opt(sb, USRQUOTA);
1697
1698                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1699                         clear_opt(sb, GRPQUOTA);
1700
1701                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1702                         ext4_msg(sb, KERN_ERR, "old and new quota "
1703                                         "format mixing");
1704                         return 0;
1705                 }
1706
1707                 if (!sbi->s_jquota_fmt) {
1708                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1709                                         "not specified");
1710                         return 0;
1711                 }
1712         }
1713 #endif
1714         if (test_opt(sb, DIOREAD_NOLOCK)) {
1715                 int blocksize =
1716                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1717
1718                 if (blocksize < PAGE_CACHE_SIZE) {
1719                         ext4_msg(sb, KERN_ERR, "can't mount with "
1720                                  "dioread_nolock if block size != PAGE_SIZE");
1721                         return 0;
1722                 }
1723         }
1724         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1725             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1726                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1727                          "in data=ordered mode");
1728                 return 0;
1729         }
1730         return 1;
1731 }
1732
1733 static inline void ext4_show_quota_options(struct seq_file *seq,
1734                                            struct super_block *sb)
1735 {
1736 #if defined(CONFIG_QUOTA)
1737         struct ext4_sb_info *sbi = EXT4_SB(sb);
1738
1739         if (sbi->s_jquota_fmt) {
1740                 char *fmtname = "";
1741
1742                 switch (sbi->s_jquota_fmt) {
1743                 case QFMT_VFS_OLD:
1744                         fmtname = "vfsold";
1745                         break;
1746                 case QFMT_VFS_V0:
1747                         fmtname = "vfsv0";
1748                         break;
1749                 case QFMT_VFS_V1:
1750                         fmtname = "vfsv1";
1751                         break;
1752                 }
1753                 seq_printf(seq, ",jqfmt=%s", fmtname);
1754         }
1755
1756         if (sbi->s_qf_names[USRQUOTA])
1757                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1758
1759         if (sbi->s_qf_names[GRPQUOTA])
1760                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1761 #endif
1762 }
1763
1764 static const char *token2str(int token)
1765 {
1766         const struct match_token *t;
1767
1768         for (t = tokens; t->token != Opt_err; t++)
1769                 if (t->token == token && !strchr(t->pattern, '='))
1770                         break;
1771         return t->pattern;
1772 }
1773
1774 /*
1775  * Show an option if
1776  *  - it's set to a non-default value OR
1777  *  - if the per-sb default is different from the global default
1778  */
1779 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1780                               int nodefs)
1781 {
1782         struct ext4_sb_info *sbi = EXT4_SB(sb);
1783         struct ext4_super_block *es = sbi->s_es;
1784         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1785         const struct mount_opts *m;
1786         char sep = nodefs ? '\n' : ',';
1787
1788 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1789 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1790
1791         if (sbi->s_sb_block != 1)
1792                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1793
1794         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1795                 int want_set = m->flags & MOPT_SET;
1796                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1797                     (m->flags & MOPT_CLEAR_ERR))
1798                         continue;
1799                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1800                         continue; /* skip if same as the default */
1801                 if ((want_set &&
1802                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1803                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1804                         continue; /* select Opt_noFoo vs Opt_Foo */
1805                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1806         }
1807
1808         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1809             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1810                 SEQ_OPTS_PRINT("resuid=%u",
1811                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1812         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1813             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1814                 SEQ_OPTS_PRINT("resgid=%u",
1815                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1816         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1817         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1818                 SEQ_OPTS_PUTS("errors=remount-ro");
1819         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1820                 SEQ_OPTS_PUTS("errors=continue");
1821         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1822                 SEQ_OPTS_PUTS("errors=panic");
1823         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1824                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1825         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1826                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1827         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1828                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1829         if (sb->s_flags & MS_I_VERSION)
1830                 SEQ_OPTS_PUTS("i_version");
1831         if (nodefs || sbi->s_stripe)
1832                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1833         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1834                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1835                         SEQ_OPTS_PUTS("data=journal");
1836                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1837                         SEQ_OPTS_PUTS("data=ordered");
1838                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1839                         SEQ_OPTS_PUTS("data=writeback");
1840         }
1841         if (nodefs ||
1842             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1843                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1844                                sbi->s_inode_readahead_blks);
1845
1846         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1847                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1848                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1849         if (nodefs || sbi->s_max_dir_size_kb)
1850                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1851
1852         ext4_show_quota_options(seq, sb);
1853         return 0;
1854 }
1855
1856 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1857 {
1858         return _ext4_show_options(seq, root->d_sb, 0);
1859 }
1860
1861 static int options_seq_show(struct seq_file *seq, void *offset)
1862 {
1863         struct super_block *sb = seq->private;
1864         int rc;
1865
1866         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1867         rc = _ext4_show_options(seq, sb, 1);
1868         seq_puts(seq, "\n");
1869         return rc;
1870 }
1871
1872 static int options_open_fs(struct inode *inode, struct file *file)
1873 {
1874         return single_open(file, options_seq_show, PDE_DATA(inode));
1875 }
1876
1877 static const struct file_operations ext4_seq_options_fops = {
1878         .owner = THIS_MODULE,
1879         .open = options_open_fs,
1880         .read = seq_read,
1881         .llseek = seq_lseek,
1882         .release = single_release,
1883 };
1884
1885 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1886                             int read_only)
1887 {
1888         struct ext4_sb_info *sbi = EXT4_SB(sb);
1889         int res = 0;
1890
1891         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1892                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1893                          "forcing read-only mode");
1894                 res = MS_RDONLY;
1895         }
1896         if (read_only)
1897                 goto done;
1898         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1899                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1900                          "running e2fsck is recommended");
1901         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1902                 ext4_msg(sb, KERN_WARNING,
1903                          "warning: mounting fs with errors, "
1904                          "running e2fsck is recommended");
1905         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1906                  le16_to_cpu(es->s_mnt_count) >=
1907                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1908                 ext4_msg(sb, KERN_WARNING,
1909                          "warning: maximal mount count reached, "
1910                          "running e2fsck is recommended");
1911         else if (le32_to_cpu(es->s_checkinterval) &&
1912                 (le32_to_cpu(es->s_lastcheck) +
1913                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1914                 ext4_msg(sb, KERN_WARNING,
1915                          "warning: checktime reached, "
1916                          "running e2fsck is recommended");
1917         if (!sbi->s_journal)
1918                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1919         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1920                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1921         le16_add_cpu(&es->s_mnt_count, 1);
1922         es->s_mtime = cpu_to_le32(get_seconds());
1923         ext4_update_dynamic_rev(sb);
1924         if (sbi->s_journal)
1925                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1926
1927         ext4_commit_super(sb, 1);
1928 done:
1929         if (test_opt(sb, DEBUG))
1930                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1931                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1932                         sb->s_blocksize,
1933                         sbi->s_groups_count,
1934                         EXT4_BLOCKS_PER_GROUP(sb),
1935                         EXT4_INODES_PER_GROUP(sb),
1936                         sbi->s_mount_opt, sbi->s_mount_opt2);
1937
1938         cleancache_init_fs(sb);
1939         return res;
1940 }
1941
1942 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1943 {
1944         struct ext4_sb_info *sbi = EXT4_SB(sb);
1945         struct flex_groups *new_groups;
1946         int size;
1947
1948         if (!sbi->s_log_groups_per_flex)
1949                 return 0;
1950
1951         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1952         if (size <= sbi->s_flex_groups_allocated)
1953                 return 0;
1954
1955         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1956         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1957         if (!new_groups) {
1958                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1959                          size / (int) sizeof(struct flex_groups));
1960                 return -ENOMEM;
1961         }
1962
1963         if (sbi->s_flex_groups) {
1964                 memcpy(new_groups, sbi->s_flex_groups,
1965                        (sbi->s_flex_groups_allocated *
1966                         sizeof(struct flex_groups)));
1967                 kvfree(sbi->s_flex_groups);
1968         }
1969         sbi->s_flex_groups = new_groups;
1970         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1971         return 0;
1972 }
1973
1974 static int ext4_fill_flex_info(struct super_block *sb)
1975 {
1976         struct ext4_sb_info *sbi = EXT4_SB(sb);
1977         struct ext4_group_desc *gdp = NULL;
1978         ext4_group_t flex_group;
1979         int i, err;
1980
1981         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1982         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1983                 sbi->s_log_groups_per_flex = 0;
1984                 return 1;
1985         }
1986
1987         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1988         if (err)
1989                 goto failed;
1990
1991         for (i = 0; i < sbi->s_groups_count; i++) {
1992                 gdp = ext4_get_group_desc(sb, i, NULL);
1993
1994                 flex_group = ext4_flex_group(sbi, i);
1995                 atomic_add(ext4_free_inodes_count(sb, gdp),
1996                            &sbi->s_flex_groups[flex_group].free_inodes);
1997                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1998                              &sbi->s_flex_groups[flex_group].free_clusters);
1999                 atomic_add(ext4_used_dirs_count(sb, gdp),
2000                            &sbi->s_flex_groups[flex_group].used_dirs);
2001         }
2002
2003         return 1;
2004 failed:
2005         return 0;
2006 }
2007
2008 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2009                                    struct ext4_group_desc *gdp)
2010 {
2011         int offset;
2012         __u16 crc = 0;
2013         __le32 le_group = cpu_to_le32(block_group);
2014
2015         if (ext4_has_metadata_csum(sbi->s_sb)) {
2016                 /* Use new metadata_csum algorithm */
2017                 __le16 save_csum;
2018                 __u32 csum32;
2019
2020                 save_csum = gdp->bg_checksum;
2021                 gdp->bg_checksum = 0;
2022                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2023                                      sizeof(le_group));
2024                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2025                                      sbi->s_desc_size);
2026                 gdp->bg_checksum = save_csum;
2027
2028                 crc = csum32 & 0xFFFF;
2029                 goto out;
2030         }
2031
2032         /* old crc16 code */
2033         if (!(sbi->s_es->s_feature_ro_compat &
2034               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2035                 return 0;
2036
2037         offset = offsetof(struct ext4_group_desc, bg_checksum);
2038
2039         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2040         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2041         crc = crc16(crc, (__u8 *)gdp, offset);
2042         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2043         /* for checksum of struct ext4_group_desc do the rest...*/
2044         if ((sbi->s_es->s_feature_incompat &
2045              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2046             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2047                 crc = crc16(crc, (__u8 *)gdp + offset,
2048                             le16_to_cpu(sbi->s_es->s_desc_size) -
2049                                 offset);
2050
2051 out:
2052         return cpu_to_le16(crc);
2053 }
2054
2055 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2056                                 struct ext4_group_desc *gdp)
2057 {
2058         if (ext4_has_group_desc_csum(sb) &&
2059             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2060                                                       block_group, gdp)))
2061                 return 0;
2062
2063         return 1;
2064 }
2065
2066 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2067                               struct ext4_group_desc *gdp)
2068 {
2069         if (!ext4_has_group_desc_csum(sb))
2070                 return;
2071         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2072 }
2073
2074 /* Called at mount-time, super-block is locked */
2075 static int ext4_check_descriptors(struct super_block *sb,
2076                                   ext4_group_t *first_not_zeroed)
2077 {
2078         struct ext4_sb_info *sbi = EXT4_SB(sb);
2079         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2080         ext4_fsblk_t last_block;
2081         ext4_fsblk_t block_bitmap;
2082         ext4_fsblk_t inode_bitmap;
2083         ext4_fsblk_t inode_table;
2084         int flexbg_flag = 0;
2085         ext4_group_t i, grp = sbi->s_groups_count;
2086
2087         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2088                 flexbg_flag = 1;
2089
2090         ext4_debug("Checking group descriptors");
2091
2092         for (i = 0; i < sbi->s_groups_count; i++) {
2093                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2094
2095                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2096                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2097                 else
2098                         last_block = first_block +
2099                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2100
2101                 if ((grp == sbi->s_groups_count) &&
2102                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2103                         grp = i;
2104
2105                 block_bitmap = ext4_block_bitmap(sb, gdp);
2106                 if (block_bitmap < first_block || block_bitmap > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Block bitmap for group %u not in group "
2109                                "(block %llu)!", i, block_bitmap);
2110                         return 0;
2111                 }
2112                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2113                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2114                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2115                                "Inode bitmap for group %u not in group "
2116                                "(block %llu)!", i, inode_bitmap);
2117                         return 0;
2118                 }
2119                 inode_table = ext4_inode_table(sb, gdp);
2120                 if (inode_table < first_block ||
2121                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2122                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2123                                "Inode table for group %u not in group "
2124                                "(block %llu)!", i, inode_table);
2125                         return 0;
2126                 }
2127                 ext4_lock_group(sb, i);
2128                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2129                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2130                                  "Checksum for group %u failed (%u!=%u)",
2131                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2132                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2133                         if (!(sb->s_flags & MS_RDONLY)) {
2134                                 ext4_unlock_group(sb, i);
2135                                 return 0;
2136                         }
2137                 }
2138                 ext4_unlock_group(sb, i);
2139                 if (!flexbg_flag)
2140                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2141         }
2142         if (NULL != first_not_zeroed)
2143                 *first_not_zeroed = grp;
2144         return 1;
2145 }
2146
2147 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2148  * the superblock) which were deleted from all directories, but held open by
2149  * a process at the time of a crash.  We walk the list and try to delete these
2150  * inodes at recovery time (only with a read-write filesystem).
2151  *
2152  * In order to keep the orphan inode chain consistent during traversal (in
2153  * case of crash during recovery), we link each inode into the superblock
2154  * orphan list_head and handle it the same way as an inode deletion during
2155  * normal operation (which journals the operations for us).
2156  *
2157  * We only do an iget() and an iput() on each inode, which is very safe if we
2158  * accidentally point at an in-use or already deleted inode.  The worst that
2159  * can happen in this case is that we get a "bit already cleared" message from
2160  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2161  * e2fsck was run on this filesystem, and it must have already done the orphan
2162  * inode cleanup for us, so we can safely abort without any further action.
2163  */
2164 static void ext4_orphan_cleanup(struct super_block *sb,
2165                                 struct ext4_super_block *es)
2166 {
2167         unsigned int s_flags = sb->s_flags;
2168         int nr_orphans = 0, nr_truncates = 0;
2169 #ifdef CONFIG_QUOTA
2170         int i;
2171 #endif
2172         if (!es->s_last_orphan) {
2173                 jbd_debug(4, "no orphan inodes to clean up\n");
2174                 return;
2175         }
2176
2177         if (bdev_read_only(sb->s_bdev)) {
2178                 ext4_msg(sb, KERN_ERR, "write access "
2179                         "unavailable, skipping orphan cleanup");
2180                 return;
2181         }
2182
2183         /* Check if feature set would not allow a r/w mount */
2184         if (!ext4_feature_set_ok(sb, 0)) {
2185                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2186                          "unknown ROCOMPAT features");
2187                 return;
2188         }
2189
2190         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2191                 /* don't clear list on RO mount w/ errors */
2192                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2193                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2194                                   "clearing orphan list.\n");
2195                         es->s_last_orphan = 0;
2196                 }
2197                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2198                 return;
2199         }
2200
2201         if (s_flags & MS_RDONLY) {
2202                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2203                 sb->s_flags &= ~MS_RDONLY;
2204         }
2205 #ifdef CONFIG_QUOTA
2206         /* Needed for iput() to work correctly and not trash data */
2207         sb->s_flags |= MS_ACTIVE;
2208         /* Turn on quotas so that they are updated correctly */
2209         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2210                 if (EXT4_SB(sb)->s_qf_names[i]) {
2211                         int ret = ext4_quota_on_mount(sb, i);
2212                         if (ret < 0)
2213                                 ext4_msg(sb, KERN_ERR,
2214                                         "Cannot turn on journaled "
2215                                         "quota: error %d", ret);
2216                 }
2217         }
2218 #endif
2219
2220         while (es->s_last_orphan) {
2221                 struct inode *inode;
2222
2223                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2224                 if (IS_ERR(inode)) {
2225                         es->s_last_orphan = 0;
2226                         break;
2227                 }
2228
2229                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2230                 dquot_initialize(inode);
2231                 if (inode->i_nlink) {
2232                         if (test_opt(sb, DEBUG))
2233                                 ext4_msg(sb, KERN_DEBUG,
2234                                         "%s: truncating inode %lu to %lld bytes",
2235                                         __func__, inode->i_ino, inode->i_size);
2236                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2237                                   inode->i_ino, inode->i_size);
2238                         mutex_lock(&inode->i_mutex);
2239                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2240                         ext4_truncate(inode);
2241                         mutex_unlock(&inode->i_mutex);
2242                         nr_truncates++;
2243                 } else {
2244                         if (test_opt(sb, DEBUG))
2245                                 ext4_msg(sb, KERN_DEBUG,
2246                                         "%s: deleting unreferenced inode %lu",
2247                                         __func__, inode->i_ino);
2248                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2249                                   inode->i_ino);
2250                         nr_orphans++;
2251                 }
2252                 iput(inode);  /* The delete magic happens here! */
2253         }
2254
2255 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2256
2257         if (nr_orphans)
2258                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2259                        PLURAL(nr_orphans));
2260         if (nr_truncates)
2261                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2262                        PLURAL(nr_truncates));
2263 #ifdef CONFIG_QUOTA
2264         /* Turn quotas off */
2265         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2266                 if (sb_dqopt(sb)->files[i])
2267                         dquot_quota_off(sb, i);
2268         }
2269 #endif
2270         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2271 }
2272
2273 /*
2274  * Maximal extent format file size.
2275  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2276  * extent format containers, within a sector_t, and within i_blocks
2277  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2278  * so that won't be a limiting factor.
2279  *
2280  * However there is other limiting factor. We do store extents in the form
2281  * of starting block and length, hence the resulting length of the extent
2282  * covering maximum file size must fit into on-disk format containers as
2283  * well. Given that length is always by 1 unit bigger than max unit (because
2284  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2285  *
2286  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2287  */
2288 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2289 {
2290         loff_t res;
2291         loff_t upper_limit = MAX_LFS_FILESIZE;
2292
2293         /* small i_blocks in vfs inode? */
2294         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2295                 /*
2296                  * CONFIG_LBDAF is not enabled implies the inode
2297                  * i_block represent total blocks in 512 bytes
2298                  * 32 == size of vfs inode i_blocks * 8
2299                  */
2300                 upper_limit = (1LL << 32) - 1;
2301
2302                 /* total blocks in file system block size */
2303                 upper_limit >>= (blkbits - 9);
2304                 upper_limit <<= blkbits;
2305         }
2306
2307         /*
2308          * 32-bit extent-start container, ee_block. We lower the maxbytes
2309          * by one fs block, so ee_len can cover the extent of maximum file
2310          * size
2311          */
2312         res = (1LL << 32) - 1;
2313         res <<= blkbits;
2314
2315         /* Sanity check against vm- & vfs- imposed limits */
2316         if (res > upper_limit)
2317                 res = upper_limit;
2318
2319         return res;
2320 }
2321
2322 /*
2323  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2324  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2325  * We need to be 1 filesystem block less than the 2^48 sector limit.
2326  */
2327 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2328 {
2329         loff_t res = EXT4_NDIR_BLOCKS;
2330         int meta_blocks;
2331         loff_t upper_limit;
2332         /* This is calculated to be the largest file size for a dense, block
2333          * mapped file such that the file's total number of 512-byte sectors,
2334          * including data and all indirect blocks, does not exceed (2^48 - 1).
2335          *
2336          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2337          * number of 512-byte sectors of the file.
2338          */
2339
2340         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2341                 /*
2342                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2343                  * the inode i_block field represents total file blocks in
2344                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2345                  */
2346                 upper_limit = (1LL << 32) - 1;
2347
2348                 /* total blocks in file system block size */
2349                 upper_limit >>= (bits - 9);
2350
2351         } else {
2352                 /*
2353                  * We use 48 bit ext4_inode i_blocks
2354                  * With EXT4_HUGE_FILE_FL set the i_blocks
2355                  * represent total number of blocks in
2356                  * file system block size
2357                  */
2358                 upper_limit = (1LL << 48) - 1;
2359
2360         }
2361
2362         /* indirect blocks */
2363         meta_blocks = 1;
2364         /* double indirect blocks */
2365         meta_blocks += 1 + (1LL << (bits-2));
2366         /* tripple indirect blocks */
2367         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2368
2369         upper_limit -= meta_blocks;
2370         upper_limit <<= bits;
2371
2372         res += 1LL << (bits-2);
2373         res += 1LL << (2*(bits-2));
2374         res += 1LL << (3*(bits-2));
2375         res <<= bits;
2376         if (res > upper_limit)
2377                 res = upper_limit;
2378
2379         if (res > MAX_LFS_FILESIZE)
2380                 res = MAX_LFS_FILESIZE;
2381
2382         return res;
2383 }
2384
2385 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2386                                    ext4_fsblk_t logical_sb_block, int nr)
2387 {
2388         struct ext4_sb_info *sbi = EXT4_SB(sb);
2389         ext4_group_t bg, first_meta_bg;
2390         int has_super = 0;
2391
2392         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2393
2394         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2395             nr < first_meta_bg)
2396                 return logical_sb_block + nr + 1;
2397         bg = sbi->s_desc_per_block * nr;
2398         if (ext4_bg_has_super(sb, bg))
2399                 has_super = 1;
2400
2401         /*
2402          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2403          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2404          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2405          * compensate.
2406          */
2407         if (sb->s_blocksize == 1024 && nr == 0 &&
2408             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2409                 has_super++;
2410
2411         return (has_super + ext4_group_first_block_no(sb, bg));
2412 }
2413
2414 /**
2415  * ext4_get_stripe_size: Get the stripe size.
2416  * @sbi: In memory super block info
2417  *
2418  * If we have specified it via mount option, then
2419  * use the mount option value. If the value specified at mount time is
2420  * greater than the blocks per group use the super block value.
2421  * If the super block value is greater than blocks per group return 0.
2422  * Allocator needs it be less than blocks per group.
2423  *
2424  */
2425 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2426 {
2427         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2428         unsigned long stripe_width =
2429                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2430         int ret;
2431
2432         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2433                 ret = sbi->s_stripe;
2434         else if (stripe_width <= sbi->s_blocks_per_group)
2435                 ret = stripe_width;
2436         else if (stride <= sbi->s_blocks_per_group)
2437                 ret = stride;
2438         else
2439                 ret = 0;
2440
2441         /*
2442          * If the stripe width is 1, this makes no sense and
2443          * we set it to 0 to turn off stripe handling code.
2444          */
2445         if (ret <= 1)
2446                 ret = 0;
2447
2448         return ret;
2449 }
2450
2451 /* sysfs supprt */
2452
2453 struct ext4_attr {
2454         struct attribute attr;
2455         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2456         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2457                          const char *, size_t);
2458         union {
2459                 int offset;
2460                 int deprecated_val;
2461         } u;
2462 };
2463
2464 static int parse_strtoull(const char *buf,
2465                 unsigned long long max, unsigned long long *value)
2466 {
2467         int ret;
2468
2469         ret = kstrtoull(skip_spaces(buf), 0, value);
2470         if (!ret && *value > max)
2471                 ret = -EINVAL;
2472         return ret;
2473 }
2474
2475 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2476                                               struct ext4_sb_info *sbi,
2477                                               char *buf)
2478 {
2479         return snprintf(buf, PAGE_SIZE, "%llu\n",
2480                 (s64) EXT4_C2B(sbi,
2481                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2482 }
2483
2484 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2485                                          struct ext4_sb_info *sbi, char *buf)
2486 {
2487         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2488
2489         if (!sb->s_bdev->bd_part)
2490                 return snprintf(buf, PAGE_SIZE, "0\n");
2491         return snprintf(buf, PAGE_SIZE, "%lu\n",
2492                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2493                          sbi->s_sectors_written_start) >> 1);
2494 }
2495
2496 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2497                                           struct ext4_sb_info *sbi, char *buf)
2498 {
2499         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2500
2501         if (!sb->s_bdev->bd_part)
2502                 return snprintf(buf, PAGE_SIZE, "0\n");
2503         return snprintf(buf, PAGE_SIZE, "%llu\n",
2504                         (unsigned long long)(sbi->s_kbytes_written +
2505                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2506                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2507 }
2508
2509 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2510                                           struct ext4_sb_info *sbi,
2511                                           const char *buf, size_t count)
2512 {
2513         unsigned long t;
2514         int ret;
2515
2516         ret = kstrtoul(skip_spaces(buf), 0, &t);
2517         if (ret)
2518                 return ret;
2519
2520         if (t && (!is_power_of_2(t) || t > 0x40000000))
2521                 return -EINVAL;
2522
2523         sbi->s_inode_readahead_blks = t;
2524         return count;
2525 }
2526
2527 static ssize_t sbi_ui_show(struct ext4_attr *a,
2528                            struct ext4_sb_info *sbi, char *buf)
2529 {
2530         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2531
2532         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2533 }
2534
2535 static ssize_t sbi_ui_store(struct ext4_attr *a,
2536                             struct ext4_sb_info *sbi,
2537                             const char *buf, size_t count)
2538 {
2539         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2540         unsigned long t;
2541         int ret;
2542
2543         ret = kstrtoul(skip_spaces(buf), 0, &t);
2544         if (ret)
2545                 return ret;
2546         *ui = t;
2547         return count;
2548 }
2549
2550 static ssize_t es_ui_show(struct ext4_attr *a,
2551                            struct ext4_sb_info *sbi, char *buf)
2552 {
2553
2554         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2555                            a->u.offset);
2556
2557         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2558 }
2559
2560 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2561                                   struct ext4_sb_info *sbi, char *buf)
2562 {
2563         return snprintf(buf, PAGE_SIZE, "%llu\n",
2564                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2565 }
2566
2567 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2568                                    struct ext4_sb_info *sbi,
2569                                    const char *buf, size_t count)
2570 {
2571         unsigned long long val;
2572         int ret;
2573
2574         if (parse_strtoull(buf, -1ULL, &val))
2575                 return -EINVAL;
2576         ret = ext4_reserve_clusters(sbi, val);
2577
2578         return ret ? ret : count;
2579 }
2580
2581 static ssize_t trigger_test_error(struct ext4_attr *a,
2582                                   struct ext4_sb_info *sbi,
2583                                   const char *buf, size_t count)
2584 {
2585         int len = count;
2586
2587         if (!capable(CAP_SYS_ADMIN))
2588                 return -EPERM;
2589
2590         if (len && buf[len-1] == '\n')
2591                 len--;
2592
2593         if (len)
2594                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2595         return count;
2596 }
2597
2598 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2599                                    struct ext4_sb_info *sbi, char *buf)
2600 {
2601         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2602 }
2603
2604 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2605 static struct ext4_attr ext4_attr_##_name = {                   \
2606         .attr = {.name = __stringify(_name), .mode = _mode },   \
2607         .show   = _show,                                        \
2608         .store  = _store,                                       \
2609         .u = {                                                  \
2610                 .offset = offsetof(struct ext4_sb_info, _elname),\
2611         },                                                      \
2612 }
2613
2614 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2615 static struct ext4_attr ext4_attr_##_name = {                           \
2616         .attr = {.name = __stringify(_name), .mode = _mode },           \
2617         .show   = _show,                                                \
2618         .store  = _store,                                               \
2619         .u = {                                                          \
2620                 .offset = offsetof(struct ext4_super_block, _elname),   \
2621         },                                                              \
2622 }
2623
2624 #define EXT4_ATTR(name, mode, show, store) \
2625 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2626
2627 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2628 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2629 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2630
2631 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2632         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2633 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2634         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2635
2636 #define ATTR_LIST(name) &ext4_attr_##name.attr
2637 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2638 static struct ext4_attr ext4_attr_##_name = {                   \
2639         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2640         .show   = sbi_deprecated_show,                          \
2641         .u = {                                                  \
2642                 .deprecated_val = _val,                         \
2643         },                                                      \
2644 }
2645
2646 EXT4_RO_ATTR(delayed_allocation_blocks);
2647 EXT4_RO_ATTR(session_write_kbytes);
2648 EXT4_RO_ATTR(lifetime_write_kbytes);
2649 EXT4_RW_ATTR(reserved_clusters);
2650 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2651                  inode_readahead_blks_store, s_inode_readahead_blks);
2652 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2653 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2654 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2655 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2656 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2657 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2658 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2659 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2660 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2661 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2662 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2663 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2664 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2665 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2666 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2667 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2668 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2669 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2670 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2671
2672 static struct attribute *ext4_attrs[] = {
2673         ATTR_LIST(delayed_allocation_blocks),
2674         ATTR_LIST(session_write_kbytes),
2675         ATTR_LIST(lifetime_write_kbytes),
2676         ATTR_LIST(reserved_clusters),
2677         ATTR_LIST(inode_readahead_blks),
2678         ATTR_LIST(inode_goal),
2679         ATTR_LIST(mb_stats),
2680         ATTR_LIST(mb_max_to_scan),
2681         ATTR_LIST(mb_min_to_scan),
2682         ATTR_LIST(mb_order2_req),
2683         ATTR_LIST(mb_stream_req),
2684         ATTR_LIST(mb_group_prealloc),
2685         ATTR_LIST(max_writeback_mb_bump),
2686         ATTR_LIST(extent_max_zeroout_kb),
2687         ATTR_LIST(trigger_fs_error),
2688         ATTR_LIST(err_ratelimit_interval_ms),
2689         ATTR_LIST(err_ratelimit_burst),
2690         ATTR_LIST(warning_ratelimit_interval_ms),
2691         ATTR_LIST(warning_ratelimit_burst),
2692         ATTR_LIST(msg_ratelimit_interval_ms),
2693         ATTR_LIST(msg_ratelimit_burst),
2694         ATTR_LIST(errors_count),
2695         ATTR_LIST(first_error_time),
2696         ATTR_LIST(last_error_time),
2697         NULL,
2698 };
2699
2700 /* Features this copy of ext4 supports */
2701 EXT4_INFO_ATTR(lazy_itable_init);
2702 EXT4_INFO_ATTR(batched_discard);
2703 EXT4_INFO_ATTR(meta_bg_resize);
2704 EXT4_INFO_ATTR(encryption);
2705
2706 static struct attribute *ext4_feat_attrs[] = {
2707         ATTR_LIST(lazy_itable_init),
2708         ATTR_LIST(batched_discard),
2709         ATTR_LIST(meta_bg_resize),
2710         ATTR_LIST(encryption),
2711         NULL,
2712 };
2713
2714 static ssize_t ext4_attr_show(struct kobject *kobj,
2715                               struct attribute *attr, char *buf)
2716 {
2717         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2718                                                 s_kobj);
2719         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2720
2721         return a->show ? a->show(a, sbi, buf) : 0;
2722 }
2723
2724 static ssize_t ext4_attr_store(struct kobject *kobj,
2725                                struct attribute *attr,
2726                                const char *buf, size_t len)
2727 {
2728         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2729                                                 s_kobj);
2730         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2731
2732         return a->store ? a->store(a, sbi, buf, len) : 0;
2733 }
2734
2735 static void ext4_sb_release(struct kobject *kobj)
2736 {
2737         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2738                                                 s_kobj);
2739         complete(&sbi->s_kobj_unregister);
2740 }
2741
2742 static const struct sysfs_ops ext4_attr_ops = {
2743         .show   = ext4_attr_show,
2744         .store  = ext4_attr_store,
2745 };
2746
2747 static struct kobj_type ext4_ktype = {
2748         .default_attrs  = ext4_attrs,
2749         .sysfs_ops      = &ext4_attr_ops,
2750         .release        = ext4_sb_release,
2751 };
2752
2753 static void ext4_feat_release(struct kobject *kobj)
2754 {
2755         complete(&ext4_feat->f_kobj_unregister);
2756 }
2757
2758 static ssize_t ext4_feat_show(struct kobject *kobj,
2759                               struct attribute *attr, char *buf)
2760 {
2761         return snprintf(buf, PAGE_SIZE, "supported\n");
2762 }
2763
2764 /*
2765  * We can not use ext4_attr_show/store because it relies on the kobject
2766  * being embedded in the ext4_sb_info structure which is definitely not
2767  * true in this case.
2768  */
2769 static const struct sysfs_ops ext4_feat_ops = {
2770         .show   = ext4_feat_show,
2771         .store  = NULL,
2772 };
2773
2774 static struct kobj_type ext4_feat_ktype = {
2775         .default_attrs  = ext4_feat_attrs,
2776         .sysfs_ops      = &ext4_feat_ops,
2777         .release        = ext4_feat_release,
2778 };
2779
2780 /*
2781  * Check whether this filesystem can be mounted based on
2782  * the features present and the RDONLY/RDWR mount requested.
2783  * Returns 1 if this filesystem can be mounted as requested,
2784  * 0 if it cannot be.
2785  */
2786 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2787 {
2788         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2789                 ext4_msg(sb, KERN_ERR,
2790                         "Couldn't mount because of "
2791                         "unsupported optional features (%x)",
2792                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2793                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2794                 return 0;
2795         }
2796
2797         if (readonly)
2798                 return 1;
2799
2800         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_READONLY)) {
2801                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2802                 sb->s_flags |= MS_RDONLY;
2803                 return 1;
2804         }
2805
2806         /* Check that feature set is OK for a read-write mount */
2807         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2808                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2809                          "unsupported optional features (%x)",
2810                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2811                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2812                 return 0;
2813         }
2814         /*
2815          * Large file size enabled file system can only be mounted
2816          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2817          */
2818         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2819                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2820                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2821                                  "cannot be mounted RDWR without "
2822                                  "CONFIG_LBDAF");
2823                         return 0;
2824                 }
2825         }
2826         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2827             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2828                 ext4_msg(sb, KERN_ERR,
2829                          "Can't support bigalloc feature without "
2830                          "extents feature\n");
2831                 return 0;
2832         }
2833
2834 #ifndef CONFIG_QUOTA
2835         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2836             !readonly) {
2837                 ext4_msg(sb, KERN_ERR,
2838                          "Filesystem with quota feature cannot be mounted RDWR "
2839                          "without CONFIG_QUOTA");
2840                 return 0;
2841         }
2842 #endif  /* CONFIG_QUOTA */
2843         return 1;
2844 }
2845
2846 /*
2847  * This function is called once a day if we have errors logged
2848  * on the file system
2849  */
2850 static void print_daily_error_info(unsigned long arg)
2851 {
2852         struct super_block *sb = (struct super_block *) arg;
2853         struct ext4_sb_info *sbi;
2854         struct ext4_super_block *es;
2855
2856         sbi = EXT4_SB(sb);
2857         es = sbi->s_es;
2858
2859         if (es->s_error_count)
2860                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2861                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2862                          le32_to_cpu(es->s_error_count));
2863         if (es->s_first_error_time) {
2864                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2865                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2866                        (int) sizeof(es->s_first_error_func),
2867                        es->s_first_error_func,
2868                        le32_to_cpu(es->s_first_error_line));
2869                 if (es->s_first_error_ino)
2870                         printk(": inode %u",
2871                                le32_to_cpu(es->s_first_error_ino));
2872                 if (es->s_first_error_block)
2873                         printk(": block %llu", (unsigned long long)
2874                                le64_to_cpu(es->s_first_error_block));
2875                 printk("\n");
2876         }
2877         if (es->s_last_error_time) {
2878                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2879                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2880                        (int) sizeof(es->s_last_error_func),
2881                        es->s_last_error_func,
2882                        le32_to_cpu(es->s_last_error_line));
2883                 if (es->s_last_error_ino)
2884                         printk(": inode %u",
2885                                le32_to_cpu(es->s_last_error_ino));
2886                 if (es->s_last_error_block)
2887                         printk(": block %llu", (unsigned long long)
2888                                le64_to_cpu(es->s_last_error_block));
2889                 printk("\n");
2890         }
2891         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2892 }
2893
2894 /* Find next suitable group and run ext4_init_inode_table */
2895 static int ext4_run_li_request(struct ext4_li_request *elr)
2896 {
2897         struct ext4_group_desc *gdp = NULL;
2898         ext4_group_t group, ngroups;
2899         struct super_block *sb;
2900         unsigned long timeout = 0;
2901         int ret = 0;
2902
2903         sb = elr->lr_super;
2904         ngroups = EXT4_SB(sb)->s_groups_count;
2905
2906         sb_start_write(sb);
2907         for (group = elr->lr_next_group; group < ngroups; group++) {
2908                 gdp = ext4_get_group_desc(sb, group, NULL);
2909                 if (!gdp) {
2910                         ret = 1;
2911                         break;
2912                 }
2913
2914                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2915                         break;
2916         }
2917
2918         if (group >= ngroups)
2919                 ret = 1;
2920
2921         if (!ret) {
2922                 timeout = jiffies;
2923                 ret = ext4_init_inode_table(sb, group,
2924                                             elr->lr_timeout ? 0 : 1);
2925                 if (elr->lr_timeout == 0) {
2926                         timeout = (jiffies - timeout) *
2927                                   elr->lr_sbi->s_li_wait_mult;
2928                         elr->lr_timeout = timeout;
2929                 }
2930                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2931                 elr->lr_next_group = group + 1;
2932         }
2933         sb_end_write(sb);
2934
2935         return ret;
2936 }
2937
2938 /*
2939  * Remove lr_request from the list_request and free the
2940  * request structure. Should be called with li_list_mtx held
2941  */
2942 static void ext4_remove_li_request(struct ext4_li_request *elr)
2943 {
2944         struct ext4_sb_info *sbi;
2945
2946         if (!elr)
2947                 return;
2948
2949         sbi = elr->lr_sbi;
2950
2951         list_del(&elr->lr_request);
2952         sbi->s_li_request = NULL;
2953         kfree(elr);
2954 }
2955
2956 static void ext4_unregister_li_request(struct super_block *sb)
2957 {
2958         mutex_lock(&ext4_li_mtx);
2959         if (!ext4_li_info) {
2960                 mutex_unlock(&ext4_li_mtx);
2961                 return;
2962         }
2963
2964         mutex_lock(&ext4_li_info->li_list_mtx);
2965         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2966         mutex_unlock(&ext4_li_info->li_list_mtx);
2967         mutex_unlock(&ext4_li_mtx);
2968 }
2969
2970 static struct task_struct *ext4_lazyinit_task;
2971
2972 /*
2973  * This is the function where ext4lazyinit thread lives. It walks
2974  * through the request list searching for next scheduled filesystem.
2975  * When such a fs is found, run the lazy initialization request
2976  * (ext4_rn_li_request) and keep track of the time spend in this
2977  * function. Based on that time we compute next schedule time of
2978  * the request. When walking through the list is complete, compute
2979  * next waking time and put itself into sleep.
2980  */
2981 static int ext4_lazyinit_thread(void *arg)
2982 {
2983         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2984         struct list_head *pos, *n;
2985         struct ext4_li_request *elr;
2986         unsigned long next_wakeup, cur;
2987
2988         BUG_ON(NULL == eli);
2989
2990 cont_thread:
2991         while (true) {
2992                 next_wakeup = MAX_JIFFY_OFFSET;
2993
2994                 mutex_lock(&eli->li_list_mtx);
2995                 if (list_empty(&eli->li_request_list)) {
2996                         mutex_unlock(&eli->li_list_mtx);
2997                         goto exit_thread;
2998                 }
2999
3000                 list_for_each_safe(pos, n, &eli->li_request_list) {
3001                         elr = list_entry(pos, struct ext4_li_request,
3002                                          lr_request);
3003
3004                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
3005                                 if (ext4_run_li_request(elr) != 0) {
3006                                         /* error, remove the lazy_init job */
3007                                         ext4_remove_li_request(elr);
3008                                         continue;
3009                                 }
3010                         }
3011
3012                         if (time_before(elr->lr_next_sched, next_wakeup))
3013                                 next_wakeup = elr->lr_next_sched;
3014                 }
3015                 mutex_unlock(&eli->li_list_mtx);
3016
3017                 try_to_freeze();
3018
3019                 cur = jiffies;
3020                 if ((time_after_eq(cur, next_wakeup)) ||
3021                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3022                         cond_resched();
3023                         continue;
3024                 }
3025
3026                 schedule_timeout_interruptible(next_wakeup - cur);
3027
3028                 if (kthread_should_stop()) {
3029                         ext4_clear_request_list();
3030                         goto exit_thread;
3031                 }
3032         }
3033
3034 exit_thread:
3035         /*
3036          * It looks like the request list is empty, but we need
3037          * to check it under the li_list_mtx lock, to prevent any
3038          * additions into it, and of course we should lock ext4_li_mtx
3039          * to atomically free the list and ext4_li_info, because at
3040          * this point another ext4 filesystem could be registering
3041          * new one.
3042          */
3043         mutex_lock(&ext4_li_mtx);
3044         mutex_lock(&eli->li_list_mtx);
3045         if (!list_empty(&eli->li_request_list)) {
3046                 mutex_unlock(&eli->li_list_mtx);
3047                 mutex_unlock(&ext4_li_mtx);
3048                 goto cont_thread;
3049         }
3050         mutex_unlock(&eli->li_list_mtx);
3051         kfree(ext4_li_info);
3052         ext4_li_info = NULL;
3053         mutex_unlock(&ext4_li_mtx);
3054
3055         return 0;
3056 }
3057
3058 static void ext4_clear_request_list(void)
3059 {
3060         struct list_head *pos, *n;
3061         struct ext4_li_request *elr;
3062
3063         mutex_lock(&ext4_li_info->li_list_mtx);
3064         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3065                 elr = list_entry(pos, struct ext4_li_request,
3066                                  lr_request);
3067                 ext4_remove_li_request(elr);
3068         }
3069         mutex_unlock(&ext4_li_info->li_list_mtx);
3070 }
3071
3072 static int ext4_run_lazyinit_thread(void)
3073 {
3074         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3075                                          ext4_li_info, "ext4lazyinit");
3076         if (IS_ERR(ext4_lazyinit_task)) {
3077                 int err = PTR_ERR(ext4_lazyinit_task);
3078                 ext4_clear_request_list();
3079                 kfree(ext4_li_info);
3080                 ext4_li_info = NULL;
3081                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3082                                  "initialization thread\n",
3083                                  err);
3084                 return err;
3085         }
3086         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3087         return 0;
3088 }
3089
3090 /*
3091  * Check whether it make sense to run itable init. thread or not.
3092  * If there is at least one uninitialized inode table, return
3093  * corresponding group number, else the loop goes through all
3094  * groups and return total number of groups.
3095  */
3096 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3097 {
3098         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3099         struct ext4_group_desc *gdp = NULL;
3100
3101         for (group = 0; group < ngroups; group++) {
3102                 gdp = ext4_get_group_desc(sb, group, NULL);
3103                 if (!gdp)
3104                         continue;
3105
3106                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3107                         break;
3108         }
3109
3110         return group;
3111 }
3112
3113 static int ext4_li_info_new(void)
3114 {
3115         struct ext4_lazy_init *eli = NULL;
3116
3117         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3118         if (!eli)
3119                 return -ENOMEM;
3120
3121         INIT_LIST_HEAD(&eli->li_request_list);
3122         mutex_init(&eli->li_list_mtx);
3123
3124         eli->li_state |= EXT4_LAZYINIT_QUIT;
3125
3126         ext4_li_info = eli;
3127
3128         return 0;
3129 }
3130
3131 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3132                                             ext4_group_t start)
3133 {
3134         struct ext4_sb_info *sbi = EXT4_SB(sb);
3135         struct ext4_li_request *elr;
3136
3137         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3138         if (!elr)
3139                 return NULL;
3140
3141         elr->lr_super = sb;
3142         elr->lr_sbi = sbi;
3143         elr->lr_next_group = start;
3144
3145         /*
3146          * Randomize first schedule time of the request to
3147          * spread the inode table initialization requests
3148          * better.
3149          */
3150         elr->lr_next_sched = jiffies + (prandom_u32() %
3151                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3152         return elr;
3153 }
3154
3155 int ext4_register_li_request(struct super_block *sb,
3156                              ext4_group_t first_not_zeroed)
3157 {
3158         struct ext4_sb_info *sbi = EXT4_SB(sb);
3159         struct ext4_li_request *elr = NULL;
3160         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3161         int ret = 0;
3162
3163         mutex_lock(&ext4_li_mtx);
3164         if (sbi->s_li_request != NULL) {
3165                 /*
3166                  * Reset timeout so it can be computed again, because
3167                  * s_li_wait_mult might have changed.
3168                  */
3169                 sbi->s_li_request->lr_timeout = 0;
3170                 goto out;
3171         }
3172
3173         if (first_not_zeroed == ngroups ||
3174             (sb->s_flags & MS_RDONLY) ||
3175             !test_opt(sb, INIT_INODE_TABLE))
3176                 goto out;
3177
3178         elr = ext4_li_request_new(sb, first_not_zeroed);
3179         if (!elr) {
3180                 ret = -ENOMEM;
3181                 goto out;
3182         }
3183
3184         if (NULL == ext4_li_info) {
3185                 ret = ext4_li_info_new();
3186                 if (ret)
3187                         goto out;
3188         }
3189
3190         mutex_lock(&ext4_li_info->li_list_mtx);
3191         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3192         mutex_unlock(&ext4_li_info->li_list_mtx);
3193
3194         sbi->s_li_request = elr;
3195         /*
3196          * set elr to NULL here since it has been inserted to
3197          * the request_list and the removal and free of it is
3198          * handled by ext4_clear_request_list from now on.
3199          */
3200         elr = NULL;
3201
3202         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3203                 ret = ext4_run_lazyinit_thread();
3204                 if (ret)
3205                         goto out;
3206         }
3207 out:
3208         mutex_unlock(&ext4_li_mtx);
3209         if (ret)
3210                 kfree(elr);
3211         return ret;
3212 }
3213
3214 /*
3215  * We do not need to lock anything since this is called on
3216  * module unload.
3217  */
3218 static void ext4_destroy_lazyinit_thread(void)
3219 {
3220         /*
3221          * If thread exited earlier
3222          * there's nothing to be done.
3223          */
3224         if (!ext4_li_info || !ext4_lazyinit_task)
3225                 return;
3226
3227         kthread_stop(ext4_lazyinit_task);
3228 }
3229
3230 static int set_journal_csum_feature_set(struct super_block *sb)
3231 {
3232         int ret = 1;
3233         int compat, incompat;
3234         struct ext4_sb_info *sbi = EXT4_SB(sb);
3235
3236         if (ext4_has_metadata_csum(sb)) {
3237                 /* journal checksum v3 */
3238                 compat = 0;
3239                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3240         } else {
3241                 /* journal checksum v1 */
3242                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3243                 incompat = 0;
3244         }
3245
3246         jbd2_journal_clear_features(sbi->s_journal,
3247                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3248                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3249                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3250         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3251                 ret = jbd2_journal_set_features(sbi->s_journal,
3252                                 compat, 0,
3253                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3254                                 incompat);
3255         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3256                 ret = jbd2_journal_set_features(sbi->s_journal,
3257                                 compat, 0,
3258                                 incompat);
3259                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3260                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3261         } else {
3262                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3263                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3264         }
3265
3266         return ret;
3267 }
3268
3269 /*
3270  * Note: calculating the overhead so we can be compatible with
3271  * historical BSD practice is quite difficult in the face of
3272  * clusters/bigalloc.  This is because multiple metadata blocks from
3273  * different block group can end up in the same allocation cluster.
3274  * Calculating the exact overhead in the face of clustered allocation
3275  * requires either O(all block bitmaps) in memory or O(number of block
3276  * groups**2) in time.  We will still calculate the superblock for
3277  * older file systems --- and if we come across with a bigalloc file
3278  * system with zero in s_overhead_clusters the estimate will be close to
3279  * correct especially for very large cluster sizes --- but for newer
3280  * file systems, it's better to calculate this figure once at mkfs
3281  * time, and store it in the superblock.  If the superblock value is
3282  * present (even for non-bigalloc file systems), we will use it.
3283  */
3284 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3285                           char *buf)
3286 {
3287         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3288         struct ext4_group_desc  *gdp;
3289         ext4_fsblk_t            first_block, last_block, b;
3290         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3291         int                     s, j, count = 0;
3292
3293         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3294                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3295                         sbi->s_itb_per_group + 2);
3296
3297         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3298                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3299         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3300         for (i = 0; i < ngroups; i++) {
3301                 gdp = ext4_get_group_desc(sb, i, NULL);
3302                 b = ext4_block_bitmap(sb, gdp);
3303                 if (b >= first_block && b <= last_block) {
3304                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3305                         count++;
3306                 }
3307                 b = ext4_inode_bitmap(sb, gdp);
3308                 if (b >= first_block && b <= last_block) {
3309                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3310                         count++;
3311                 }
3312                 b = ext4_inode_table(sb, gdp);
3313                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3314                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3315                                 int c = EXT4_B2C(sbi, b - first_block);
3316                                 ext4_set_bit(c, buf);
3317                                 count++;
3318                         }
3319                 if (i != grp)
3320                         continue;
3321                 s = 0;
3322                 if (ext4_bg_has_super(sb, grp)) {
3323                         ext4_set_bit(s++, buf);
3324                         count++;
3325                 }
3326                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3327                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3328                         count++;
3329                 }
3330         }
3331         if (!count)
3332                 return 0;
3333         return EXT4_CLUSTERS_PER_GROUP(sb) -
3334                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3335 }
3336
3337 /*
3338  * Compute the overhead and stash it in sbi->s_overhead
3339  */
3340 int ext4_calculate_overhead(struct super_block *sb)
3341 {
3342         struct ext4_sb_info *sbi = EXT4_SB(sb);
3343         struct ext4_super_block *es = sbi->s_es;
3344         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3345         ext4_fsblk_t overhead = 0;
3346         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3347
3348         if (!buf)
3349                 return -ENOMEM;
3350
3351         /*
3352          * Compute the overhead (FS structures).  This is constant
3353          * for a given filesystem unless the number of block groups
3354          * changes so we cache the previous value until it does.
3355          */
3356
3357         /*
3358          * All of the blocks before first_data_block are overhead
3359          */
3360         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3361
3362         /*
3363          * Add the overhead found in each block group
3364          */
3365         for (i = 0; i < ngroups; i++) {
3366                 int blks;
3367
3368                 blks = count_overhead(sb, i, buf);
3369                 overhead += blks;
3370                 if (blks)
3371                         memset(buf, 0, PAGE_SIZE);
3372                 cond_resched();
3373         }
3374         /* Add the internal journal blocks as well */
3375         if (sbi->s_journal && !sbi->journal_bdev)
3376                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3377
3378         sbi->s_overhead = overhead;
3379         smp_wmb();
3380         free_page((unsigned long) buf);
3381         return 0;
3382 }
3383
3384
3385 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3386 {
3387         ext4_fsblk_t resv_clusters;
3388
3389         /*
3390          * There's no need to reserve anything when we aren't using extents.
3391          * The space estimates are exact, there are no unwritten extents,
3392          * hole punching doesn't need new metadata... This is needed especially
3393          * to keep ext2/3 backward compatibility.
3394          */
3395         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3396                 return 0;
3397         /*
3398          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3399          * This should cover the situations where we can not afford to run
3400          * out of space like for example punch hole, or converting
3401          * unwritten extents in delalloc path. In most cases such
3402          * allocation would require 1, or 2 blocks, higher numbers are
3403          * very rare.
3404          */
3405         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3406                         EXT4_SB(sb)->s_cluster_bits;
3407
3408         do_div(resv_clusters, 50);
3409         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3410
3411         return resv_clusters;
3412 }
3413
3414
3415 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3416 {
3417         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3418                                 sbi->s_cluster_bits;
3419
3420         if (count >= clusters)
3421                 return -EINVAL;
3422
3423         atomic64_set(&sbi->s_resv_clusters, count);
3424         return 0;
3425 }
3426
3427 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3428 {
3429         char *orig_data = kstrdup(data, GFP_KERNEL);
3430         struct buffer_head *bh;
3431         struct ext4_super_block *es = NULL;
3432         struct ext4_sb_info *sbi;
3433         ext4_fsblk_t block;
3434         ext4_fsblk_t sb_block = get_sb_block(&data);
3435         ext4_fsblk_t logical_sb_block;
3436         unsigned long offset = 0;
3437         unsigned long journal_devnum = 0;
3438         unsigned long def_mount_opts;
3439         struct inode *root;
3440         char *cp;
3441         const char *descr;
3442         int ret = -ENOMEM;
3443         int blocksize, clustersize;
3444         unsigned int db_count;
3445         unsigned int i;
3446         int needs_recovery, has_huge_files, has_bigalloc;
3447         __u64 blocks_count;
3448         int err = 0;
3449         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3450         ext4_group_t first_not_zeroed;
3451
3452         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3453         if (!sbi)
3454                 goto out_free_orig;
3455
3456         sbi->s_blockgroup_lock =
3457                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3458         if (!sbi->s_blockgroup_lock) {
3459                 kfree(sbi);
3460                 goto out_free_orig;
3461         }
3462         sb->s_fs_info = sbi;
3463         sbi->s_sb = sb;
3464         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3465         sbi->s_sb_block = sb_block;
3466         if (sb->s_bdev->bd_part)
3467                 sbi->s_sectors_written_start =
3468                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3469 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3470         /* Modes of operations for file and directory encryption. */
3471         sbi->s_file_encryption_mode = EXT4_ENCRYPTION_MODE_AES_256_XTS;
3472         sbi->s_dir_encryption_mode = EXT4_ENCRYPTION_MODE_INVALID;
3473 #endif
3474
3475         /* Cleanup superblock name */
3476         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3477                 *cp = '!';
3478
3479         /* -EINVAL is default */
3480         ret = -EINVAL;
3481         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3482         if (!blocksize) {
3483                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3484                 goto out_fail;
3485         }
3486
3487         /*
3488          * The ext4 superblock will not be buffer aligned for other than 1kB
3489          * block sizes.  We need to calculate the offset from buffer start.
3490          */
3491         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3492                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3493                 offset = do_div(logical_sb_block, blocksize);
3494         } else {
3495                 logical_sb_block = sb_block;
3496         }
3497
3498         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3499                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3500                 goto out_fail;
3501         }
3502         /*
3503          * Note: s_es must be initialized as soon as possible because
3504          *       some ext4 macro-instructions depend on its value
3505          */
3506         es = (struct ext4_super_block *) (bh->b_data + offset);
3507         sbi->s_es = es;
3508         sb->s_magic = le16_to_cpu(es->s_magic);
3509         if (sb->s_magic != EXT4_SUPER_MAGIC)
3510                 goto cantfind_ext4;
3511         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3512
3513         /* Warn if metadata_csum and gdt_csum are both set. */
3514         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3515                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3516             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3517                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3518                              "redundant flags; please run fsck.");
3519
3520         /* Check for a known checksum algorithm */
3521         if (!ext4_verify_csum_type(sb, es)) {
3522                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3523                          "unknown checksum algorithm.");
3524                 silent = 1;
3525                 goto cantfind_ext4;
3526         }
3527
3528         /* Load the checksum driver */
3529         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3530                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3531                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3532                 if (IS_ERR(sbi->s_chksum_driver)) {
3533                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3534                         ret = PTR_ERR(sbi->s_chksum_driver);
3535                         sbi->s_chksum_driver = NULL;
3536                         goto failed_mount;
3537                 }
3538         }
3539
3540         /* Check superblock checksum */
3541         if (!ext4_superblock_csum_verify(sb, es)) {
3542                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3543                          "invalid superblock checksum.  Run e2fsck?");
3544                 silent = 1;
3545                 goto cantfind_ext4;
3546         }
3547
3548         /* Precompute checksum seed for all metadata */
3549         if (ext4_has_metadata_csum(sb))
3550                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3551                                                sizeof(es->s_uuid));
3552
3553         /* Set defaults before we parse the mount options */
3554         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3555         set_opt(sb, INIT_INODE_TABLE);
3556         if (def_mount_opts & EXT4_DEFM_DEBUG)
3557                 set_opt(sb, DEBUG);
3558         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3559                 set_opt(sb, GRPID);
3560         if (def_mount_opts & EXT4_DEFM_UID16)
3561                 set_opt(sb, NO_UID32);
3562         /* xattr user namespace & acls are now defaulted on */
3563         set_opt(sb, XATTR_USER);
3564 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3565         set_opt(sb, POSIX_ACL);
3566 #endif
3567         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3568         if (ext4_has_metadata_csum(sb))
3569                 set_opt(sb, JOURNAL_CHECKSUM);
3570
3571         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3572                 set_opt(sb, JOURNAL_DATA);
3573         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3574                 set_opt(sb, ORDERED_DATA);
3575         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3576                 set_opt(sb, WRITEBACK_DATA);
3577
3578         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3579                 set_opt(sb, ERRORS_PANIC);
3580         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3581                 set_opt(sb, ERRORS_CONT);
3582         else
3583                 set_opt(sb, ERRORS_RO);
3584         /* block_validity enabled by default; disable with noblock_validity */
3585         set_opt(sb, BLOCK_VALIDITY);
3586         if (def_mount_opts & EXT4_DEFM_DISCARD)
3587                 set_opt(sb, DISCARD);
3588
3589         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3590         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3591         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3592         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3593         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3594
3595         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3596                 set_opt(sb, BARRIER);
3597
3598         /*
3599          * enable delayed allocation by default
3600          * Use -o nodelalloc to turn it off
3601          */
3602         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3603             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3604                 set_opt(sb, DELALLOC);
3605
3606         /*
3607          * set default s_li_wait_mult for lazyinit, for the case there is
3608          * no mount option specified.
3609          */
3610         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3611
3612         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3613                            &journal_devnum, &journal_ioprio, 0)) {
3614                 ext4_msg(sb, KERN_WARNING,
3615                          "failed to parse options in superblock: %s",
3616                          sbi->s_es->s_mount_opts);
3617         }
3618         sbi->s_def_mount_opt = sbi->s_mount_opt;
3619         if (!parse_options((char *) data, sb, &journal_devnum,
3620                            &journal_ioprio, 0))
3621                 goto failed_mount;
3622
3623         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3624                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3625                             "with data=journal disables delayed "
3626                             "allocation and O_DIRECT support!\n");
3627                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3628                         ext4_msg(sb, KERN_ERR, "can't mount with "
3629                                  "both data=journal and delalloc");
3630                         goto failed_mount;
3631                 }
3632                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3633                         ext4_msg(sb, KERN_ERR, "can't mount with "
3634                                  "both data=journal and dioread_nolock");
3635                         goto failed_mount;
3636                 }
3637                 if (test_opt(sb, DAX)) {
3638                         ext4_msg(sb, KERN_ERR, "can't mount with "
3639                                  "both data=journal and dax");
3640                         goto failed_mount;
3641                 }
3642                 if (test_opt(sb, DELALLOC))
3643                         clear_opt(sb, DELALLOC);
3644         }
3645
3646         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3647                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3648
3649         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3650             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3651              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3652              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3653                 ext4_msg(sb, KERN_WARNING,
3654                        "feature flags set on rev 0 fs, "
3655                        "running e2fsck is recommended");
3656
3657         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3658                 set_opt2(sb, HURD_COMPAT);
3659                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3660                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3661                         ext4_msg(sb, KERN_ERR,
3662                                  "The Hurd can't support 64-bit file systems");
3663                         goto failed_mount;
3664                 }
3665         }
3666
3667         if (IS_EXT2_SB(sb)) {
3668                 if (ext2_feature_set_ok(sb))
3669                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3670                                  "using the ext4 subsystem");
3671                 else {
3672                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3673                                  "to feature incompatibilities");
3674                         goto failed_mount;
3675                 }
3676         }
3677
3678         if (IS_EXT3_SB(sb)) {
3679                 if (ext3_feature_set_ok(sb))
3680                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3681                                  "using the ext4 subsystem");
3682                 else {
3683                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3684                                  "to feature incompatibilities");
3685                         goto failed_mount;
3686                 }
3687         }
3688
3689         /*
3690          * Check feature flags regardless of the revision level, since we
3691          * previously didn't change the revision level when setting the flags,
3692          * so there is a chance incompat flags are set on a rev 0 filesystem.
3693          */
3694         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3695                 goto failed_mount;
3696
3697         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3698         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3699             blocksize > EXT4_MAX_BLOCK_SIZE) {
3700                 ext4_msg(sb, KERN_ERR,
3701                        "Unsupported filesystem blocksize %d", blocksize);
3702                 goto failed_mount;
3703         }
3704
3705         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3706                 if (blocksize != PAGE_SIZE) {
3707                         ext4_msg(sb, KERN_ERR,
3708                                         "error: unsupported blocksize for dax");
3709                         goto failed_mount;
3710                 }
3711                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                         "error: device does not support dax");
3714                         goto failed_mount;
3715                 }
3716         }
3717
3718         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT) &&
3719             es->s_encryption_level) {
3720                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3721                          es->s_encryption_level);
3722                 goto failed_mount;
3723         }
3724
3725         if (sb->s_blocksize != blocksize) {
3726                 /* Validate the filesystem blocksize */
3727                 if (!sb_set_blocksize(sb, blocksize)) {
3728                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3729                                         blocksize);
3730                         goto failed_mount;
3731                 }
3732
3733                 brelse(bh);
3734                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3735                 offset = do_div(logical_sb_block, blocksize);
3736                 bh = sb_bread_unmovable(sb, logical_sb_block);
3737                 if (!bh) {
3738                         ext4_msg(sb, KERN_ERR,
3739                                "Can't read superblock on 2nd try");
3740                         goto failed_mount;
3741                 }
3742                 es = (struct ext4_super_block *)(bh->b_data + offset);
3743                 sbi->s_es = es;
3744                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3745                         ext4_msg(sb, KERN_ERR,
3746                                "Magic mismatch, very weird!");
3747                         goto failed_mount;
3748                 }
3749         }
3750
3751         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3752                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3753         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3754                                                       has_huge_files);
3755         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3756
3757         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3758                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3759                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3760         } else {
3761                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3762                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3763                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3764                     (!is_power_of_2(sbi->s_inode_size)) ||
3765                     (sbi->s_inode_size > blocksize)) {
3766                         ext4_msg(sb, KERN_ERR,
3767                                "unsupported inode size: %d",
3768                                sbi->s_inode_size);
3769                         goto failed_mount;
3770                 }
3771                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3772                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3773         }
3774
3775         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3776         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3777                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3778                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3779                     !is_power_of_2(sbi->s_desc_size)) {
3780                         ext4_msg(sb, KERN_ERR,
3781                                "unsupported descriptor size %lu",
3782                                sbi->s_desc_size);
3783                         goto failed_mount;
3784                 }
3785         } else
3786                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3787
3788         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3789         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3790         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3791                 goto cantfind_ext4;
3792
3793         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3794         if (sbi->s_inodes_per_block == 0)
3795                 goto cantfind_ext4;
3796         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3797                                         sbi->s_inodes_per_block;
3798         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3799         sbi->s_sbh = bh;
3800         sbi->s_mount_state = le16_to_cpu(es->s_state);
3801         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3802         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3803
3804         for (i = 0; i < 4; i++)
3805                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3806         sbi->s_def_hash_version = es->s_def_hash_version;
3807         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3808                 i = le32_to_cpu(es->s_flags);
3809                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3810                         sbi->s_hash_unsigned = 3;
3811                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3812 #ifdef __CHAR_UNSIGNED__
3813                         if (!(sb->s_flags & MS_RDONLY))
3814                                 es->s_flags |=
3815                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3816                         sbi->s_hash_unsigned = 3;
3817 #else
3818                         if (!(sb->s_flags & MS_RDONLY))
3819                                 es->s_flags |=
3820                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3821 #endif
3822                 }
3823         }
3824
3825         /* Handle clustersize */
3826         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3827         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3828                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3829         if (has_bigalloc) {
3830                 if (clustersize < blocksize) {
3831                         ext4_msg(sb, KERN_ERR,
3832                                  "cluster size (%d) smaller than "
3833                                  "block size (%d)", clustersize, blocksize);
3834                         goto failed_mount;
3835                 }
3836                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3837                         le32_to_cpu(es->s_log_block_size);
3838                 sbi->s_clusters_per_group =
3839                         le32_to_cpu(es->s_clusters_per_group);
3840                 if (sbi->s_clusters_per_group > blocksize * 8) {
3841                         ext4_msg(sb, KERN_ERR,
3842                                  "#clusters per group too big: %lu",
3843                                  sbi->s_clusters_per_group);
3844                         goto failed_mount;
3845                 }
3846                 if (sbi->s_blocks_per_group !=
3847                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3848                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3849                                  "clusters per group (%lu) inconsistent",
3850                                  sbi->s_blocks_per_group,
3851                                  sbi->s_clusters_per_group);
3852                         goto failed_mount;
3853                 }
3854         } else {
3855                 if (clustersize != blocksize) {
3856                         ext4_warning(sb, "fragment/cluster size (%d) != "
3857                                      "block size (%d)", clustersize,
3858                                      blocksize);
3859                         clustersize = blocksize;
3860                 }
3861                 if (sbi->s_blocks_per_group > blocksize * 8) {
3862                         ext4_msg(sb, KERN_ERR,
3863                                  "#blocks per group too big: %lu",
3864                                  sbi->s_blocks_per_group);
3865                         goto failed_mount;
3866                 }
3867                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3868                 sbi->s_cluster_bits = 0;
3869         }
3870         sbi->s_cluster_ratio = clustersize / blocksize;
3871
3872         if (sbi->s_inodes_per_group > blocksize * 8) {
3873                 ext4_msg(sb, KERN_ERR,
3874                        "#inodes per group too big: %lu",
3875                        sbi->s_inodes_per_group);
3876                 goto failed_mount;
3877         }
3878
3879         /* Do we have standard group size of clustersize * 8 blocks ? */
3880         if (sbi->s_blocks_per_group == clustersize << 3)
3881                 set_opt2(sb, STD_GROUP_SIZE);
3882
3883         /*
3884          * Test whether we have more sectors than will fit in sector_t,
3885          * and whether the max offset is addressable by the page cache.
3886          */
3887         err = generic_check_addressable(sb->s_blocksize_bits,
3888                                         ext4_blocks_count(es));
3889         if (err) {
3890                 ext4_msg(sb, KERN_ERR, "filesystem"
3891                          " too large to mount safely on this system");
3892                 if (sizeof(sector_t) < 8)
3893                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3894                 goto failed_mount;
3895         }
3896
3897         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3898                 goto cantfind_ext4;
3899
3900         /* check blocks count against device size */
3901         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3902         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3903                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3904                        "exceeds size of device (%llu blocks)",
3905                        ext4_blocks_count(es), blocks_count);
3906                 goto failed_mount;
3907         }
3908
3909         /*
3910          * It makes no sense for the first data block to be beyond the end
3911          * of the filesystem.
3912          */
3913         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3914                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3915                          "block %u is beyond end of filesystem (%llu)",
3916                          le32_to_cpu(es->s_first_data_block),
3917                          ext4_blocks_count(es));
3918                 goto failed_mount;
3919         }
3920         blocks_count = (ext4_blocks_count(es) -
3921                         le32_to_cpu(es->s_first_data_block) +
3922                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3923         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3924         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3925                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3926                        "(block count %llu, first data block %u, "
3927                        "blocks per group %lu)", sbi->s_groups_count,
3928                        ext4_blocks_count(es),
3929                        le32_to_cpu(es->s_first_data_block),
3930                        EXT4_BLOCKS_PER_GROUP(sb));
3931                 goto failed_mount;
3932         }
3933         sbi->s_groups_count = blocks_count;
3934         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3935                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3936         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3937                    EXT4_DESC_PER_BLOCK(sb);
3938         sbi->s_group_desc = ext4_kvmalloc(db_count *
3939                                           sizeof(struct buffer_head *),
3940                                           GFP_KERNEL);
3941         if (sbi->s_group_desc == NULL) {
3942                 ext4_msg(sb, KERN_ERR, "not enough memory");
3943                 ret = -ENOMEM;
3944                 goto failed_mount;
3945         }
3946
3947         if (ext4_proc_root)
3948                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3949
3950         if (sbi->s_proc)
3951                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3952                                  &ext4_seq_options_fops, sb);
3953
3954         bgl_lock_init(sbi->s_blockgroup_lock);
3955
3956         for (i = 0; i < db_count; i++) {
3957                 block = descriptor_loc(sb, logical_sb_block, i);
3958                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3959                 if (!sbi->s_group_desc[i]) {
3960                         ext4_msg(sb, KERN_ERR,
3961                                "can't read group descriptor %d", i);
3962                         db_count = i;
3963                         goto failed_mount2;
3964                 }
3965         }
3966         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3967                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3968                 goto failed_mount2;
3969         }
3970
3971         sbi->s_gdb_count = db_count;
3972         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3973         spin_lock_init(&sbi->s_next_gen_lock);
3974
3975         setup_timer(&sbi->s_err_report, print_daily_error_info,
3976                 (unsigned long) sb);
3977
3978         /* Register extent status tree shrinker */
3979         if (ext4_es_register_shrinker(sbi))
3980                 goto failed_mount3;
3981
3982         sbi->s_stripe = ext4_get_stripe_size(sbi);
3983         sbi->s_extent_max_zeroout_kb = 32;
3984
3985         /*
3986          * set up enough so that it can read an inode
3987          */
3988         sb->s_op = &ext4_sops;
3989         sb->s_export_op = &ext4_export_ops;
3990         sb->s_xattr = ext4_xattr_handlers;
3991 #ifdef CONFIG_QUOTA
3992         sb->dq_op = &ext4_quota_operations;
3993         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3994                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3995         else
3996                 sb->s_qcop = &ext4_qctl_operations;
3997         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3998 #endif
3999         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4000
4001         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4002         mutex_init(&sbi->s_orphan_lock);
4003
4004         sb->s_root = NULL;
4005
4006         needs_recovery = (es->s_last_orphan != 0 ||
4007                           EXT4_HAS_INCOMPAT_FEATURE(sb,
4008                                     EXT4_FEATURE_INCOMPAT_RECOVER));
4009
4010         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
4011             !(sb->s_flags & MS_RDONLY))
4012                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4013                         goto failed_mount3a;
4014
4015         /*
4016          * The first inode we look at is the journal inode.  Don't try
4017          * root first: it may be modified in the journal!
4018          */
4019         if (!test_opt(sb, NOLOAD) &&
4020             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4021                 if (ext4_load_journal(sb, es, journal_devnum))
4022                         goto failed_mount3a;
4023         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
4024               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4025                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4026                        "suppressed and not mounted read-only");
4027                 goto failed_mount_wq;
4028         } else {
4029                 clear_opt(sb, DATA_FLAGS);
4030                 sbi->s_journal = NULL;
4031                 needs_recovery = 0;
4032                 goto no_journal;
4033         }
4034
4035         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
4036             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4037                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4038                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4039                 goto failed_mount_wq;
4040         }
4041
4042         if (!set_journal_csum_feature_set(sb)) {
4043                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4044                          "feature set");
4045                 goto failed_mount_wq;
4046         }
4047
4048         /* We have now updated the journal if required, so we can
4049          * validate the data journaling mode. */
4050         switch (test_opt(sb, DATA_FLAGS)) {
4051         case 0:
4052                 /* No mode set, assume a default based on the journal
4053                  * capabilities: ORDERED_DATA if the journal can
4054                  * cope, else JOURNAL_DATA
4055                  */
4056                 if (jbd2_journal_check_available_features
4057                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4058                         set_opt(sb, ORDERED_DATA);
4059                 else
4060                         set_opt(sb, JOURNAL_DATA);
4061                 break;
4062
4063         case EXT4_MOUNT_ORDERED_DATA:
4064         case EXT4_MOUNT_WRITEBACK_DATA:
4065                 if (!jbd2_journal_check_available_features
4066                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4067                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4068                                "requested data journaling mode");
4069                         goto failed_mount_wq;
4070                 }
4071         default:
4072                 break;
4073         }
4074         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4075
4076         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4077
4078 no_journal:
4079         if (ext4_mballoc_ready) {
4080                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4081                 if (!sbi->s_mb_cache) {
4082                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4083                         goto failed_mount_wq;
4084                 }
4085         }
4086
4087         if (unlikely(sbi->s_mount_flags & EXT4_MF_TEST_DUMMY_ENCRYPTION) &&
4088             !(sb->s_flags & MS_RDONLY) &&
4089             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT)) {
4090                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT);
4091                 ext4_commit_super(sb, 1);
4092         }
4093
4094         /*
4095          * Get the # of file system overhead blocks from the
4096          * superblock if present.
4097          */
4098         if (es->s_overhead_clusters)
4099                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4100         else {
4101                 err = ext4_calculate_overhead(sb);
4102                 if (err)
4103                         goto failed_mount_wq;
4104         }
4105
4106         /*
4107          * The maximum number of concurrent works can be high and
4108          * concurrency isn't really necessary.  Limit it to 1.
4109          */
4110         EXT4_SB(sb)->rsv_conversion_wq =
4111                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4112         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4113                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4114                 ret = -ENOMEM;
4115                 goto failed_mount4;
4116         }
4117
4118         /*
4119          * The jbd2_journal_load will have done any necessary log recovery,
4120          * so we can safely mount the rest of the filesystem now.
4121          */
4122
4123         root = ext4_iget(sb, EXT4_ROOT_INO);
4124         if (IS_ERR(root)) {
4125                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4126                 ret = PTR_ERR(root);
4127                 root = NULL;
4128                 goto failed_mount4;
4129         }
4130         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4131                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4132                 iput(root);
4133                 goto failed_mount4;
4134         }
4135         sb->s_root = d_make_root(root);
4136         if (!sb->s_root) {
4137                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4138                 ret = -ENOMEM;
4139                 goto failed_mount4;
4140         }
4141
4142         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4143                 sb->s_flags |= MS_RDONLY;
4144
4145         /* determine the minimum size of new large inodes, if present */
4146         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4147                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4148                                                      EXT4_GOOD_OLD_INODE_SIZE;
4149                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4150                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4151                         if (sbi->s_want_extra_isize <
4152                             le16_to_cpu(es->s_want_extra_isize))
4153                                 sbi->s_want_extra_isize =
4154                                         le16_to_cpu(es->s_want_extra_isize);
4155                         if (sbi->s_want_extra_isize <
4156                             le16_to_cpu(es->s_min_extra_isize))
4157                                 sbi->s_want_extra_isize =
4158                                         le16_to_cpu(es->s_min_extra_isize);
4159                 }
4160         }
4161         /* Check if enough inode space is available */
4162         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4163                                                         sbi->s_inode_size) {
4164                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4165                                                        EXT4_GOOD_OLD_INODE_SIZE;
4166                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4167                          "available");
4168         }
4169
4170         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4171         if (err) {
4172                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4173                          "reserved pool", ext4_calculate_resv_clusters(sb));
4174                 goto failed_mount4a;
4175         }
4176
4177         err = ext4_setup_system_zone(sb);
4178         if (err) {
4179                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4180                          "zone (%d)", err);
4181                 goto failed_mount4a;
4182         }
4183
4184         ext4_ext_init(sb);
4185         err = ext4_mb_init(sb);
4186         if (err) {
4187                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4188                          err);
4189                 goto failed_mount5;
4190         }
4191
4192         block = ext4_count_free_clusters(sb);
4193         ext4_free_blocks_count_set(sbi->s_es,
4194                                    EXT4_C2B(sbi, block));
4195         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4196                                   GFP_KERNEL);
4197         if (!err) {
4198                 unsigned long freei = ext4_count_free_inodes(sb);
4199                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4200                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4201                                           GFP_KERNEL);
4202         }
4203         if (!err)
4204                 err = percpu_counter_init(&sbi->s_dirs_counter,
4205                                           ext4_count_dirs(sb), GFP_KERNEL);
4206         if (!err)
4207                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4208                                           GFP_KERNEL);
4209         if (err) {
4210                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4211                 goto failed_mount6;
4212         }
4213
4214         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4215                 if (!ext4_fill_flex_info(sb)) {
4216                         ext4_msg(sb, KERN_ERR,
4217                                "unable to initialize "
4218                                "flex_bg meta info!");
4219                         goto failed_mount6;
4220                 }
4221
4222         err = ext4_register_li_request(sb, first_not_zeroed);
4223         if (err)
4224                 goto failed_mount6;
4225
4226         sbi->s_kobj.kset = ext4_kset;
4227         init_completion(&sbi->s_kobj_unregister);
4228         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4229                                    "%s", sb->s_id);
4230         if (err)
4231                 goto failed_mount7;
4232
4233 #ifdef CONFIG_QUOTA
4234         /* Enable quota usage during mount. */
4235         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4236             !(sb->s_flags & MS_RDONLY)) {
4237                 err = ext4_enable_quotas(sb);
4238                 if (err)
4239                         goto failed_mount8;
4240         }
4241 #endif  /* CONFIG_QUOTA */
4242
4243         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4244         ext4_orphan_cleanup(sb, es);
4245         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4246         if (needs_recovery) {
4247                 ext4_msg(sb, KERN_INFO, "recovery complete");
4248                 ext4_mark_recovery_complete(sb, es);
4249         }
4250         if (EXT4_SB(sb)->s_journal) {
4251                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4252                         descr = " journalled data mode";
4253                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4254                         descr = " ordered data mode";
4255                 else
4256                         descr = " writeback data mode";
4257         } else
4258                 descr = "out journal";
4259
4260         if (test_opt(sb, DISCARD)) {
4261                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4262                 if (!blk_queue_discard(q))
4263                         ext4_msg(sb, KERN_WARNING,
4264                                  "mounting with \"discard\" option, but "
4265                                  "the device does not support discard");
4266         }
4267
4268         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4269                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4270                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4271
4272         if (es->s_error_count)
4273                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4274
4275         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4276         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4277         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4278         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4279
4280         kfree(orig_data);
4281         return 0;
4282
4283 cantfind_ext4:
4284         if (!silent)
4285                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4286         goto failed_mount;
4287
4288 #ifdef CONFIG_QUOTA
4289 failed_mount8:
4290         kobject_del(&sbi->s_kobj);
4291 #endif
4292 failed_mount7:
4293         ext4_unregister_li_request(sb);
4294 failed_mount6:
4295         ext4_mb_release(sb);
4296         if (sbi->s_flex_groups)
4297                 kvfree(sbi->s_flex_groups);
4298         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4299         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4300         percpu_counter_destroy(&sbi->s_dirs_counter);
4301         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4302 failed_mount5:
4303         ext4_ext_release(sb);
4304         ext4_release_system_zone(sb);
4305 failed_mount4a:
4306         dput(sb->s_root);
4307         sb->s_root = NULL;
4308 failed_mount4:
4309         ext4_msg(sb, KERN_ERR, "mount failed");
4310         if (EXT4_SB(sb)->rsv_conversion_wq)
4311                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4312 failed_mount_wq:
4313         if (sbi->s_journal) {
4314                 jbd2_journal_destroy(sbi->s_journal);
4315                 sbi->s_journal = NULL;
4316         }
4317 failed_mount3a:
4318         ext4_es_unregister_shrinker(sbi);
4319 failed_mount3:
4320         del_timer_sync(&sbi->s_err_report);
4321         if (sbi->s_mmp_tsk)
4322                 kthread_stop(sbi->s_mmp_tsk);
4323 failed_mount2:
4324         for (i = 0; i < db_count; i++)
4325                 brelse(sbi->s_group_desc[i]);
4326         kvfree(sbi->s_group_desc);
4327 failed_mount:
4328         if (sbi->s_chksum_driver)
4329                 crypto_free_shash(sbi->s_chksum_driver);
4330         if (sbi->s_proc) {
4331                 remove_proc_entry("options", sbi->s_proc);
4332                 remove_proc_entry(sb->s_id, ext4_proc_root);
4333         }
4334 #ifdef CONFIG_QUOTA
4335         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4336                 kfree(sbi->s_qf_names[i]);
4337 #endif
4338         ext4_blkdev_remove(sbi);
4339         brelse(bh);
4340 out_fail:
4341         sb->s_fs_info = NULL;
4342         kfree(sbi->s_blockgroup_lock);
4343         kfree(sbi);
4344 out_free_orig:
4345         kfree(orig_data);
4346         return err ? err : ret;
4347 }
4348
4349 /*
4350  * Setup any per-fs journal parameters now.  We'll do this both on
4351  * initial mount, once the journal has been initialised but before we've
4352  * done any recovery; and again on any subsequent remount.
4353  */
4354 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4355 {
4356         struct ext4_sb_info *sbi = EXT4_SB(sb);
4357
4358         journal->j_commit_interval = sbi->s_commit_interval;
4359         journal->j_min_batch_time = sbi->s_min_batch_time;
4360         journal->j_max_batch_time = sbi->s_max_batch_time;
4361
4362         write_lock(&journal->j_state_lock);
4363         if (test_opt(sb, BARRIER))
4364                 journal->j_flags |= JBD2_BARRIER;
4365         else
4366                 journal->j_flags &= ~JBD2_BARRIER;
4367         if (test_opt(sb, DATA_ERR_ABORT))
4368                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4369         else
4370                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4371         write_unlock(&journal->j_state_lock);
4372 }
4373
4374 static journal_t *ext4_get_journal(struct super_block *sb,
4375                                    unsigned int journal_inum)
4376 {
4377         struct inode *journal_inode;
4378         journal_t *journal;
4379
4380         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4381
4382         /* First, test for the existence of a valid inode on disk.  Bad
4383          * things happen if we iget() an unused inode, as the subsequent
4384          * iput() will try to delete it. */
4385
4386         journal_inode = ext4_iget(sb, journal_inum);
4387         if (IS_ERR(journal_inode)) {
4388                 ext4_msg(sb, KERN_ERR, "no journal found");
4389                 return NULL;
4390         }
4391         if (!journal_inode->i_nlink) {
4392                 make_bad_inode(journal_inode);
4393                 iput(journal_inode);
4394                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4395                 return NULL;
4396         }
4397
4398         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4399                   journal_inode, journal_inode->i_size);
4400         if (!S_ISREG(journal_inode->i_mode)) {
4401                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4402                 iput(journal_inode);
4403                 return NULL;
4404         }
4405
4406         journal = jbd2_journal_init_inode(journal_inode);
4407         if (!journal) {
4408                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4409                 iput(journal_inode);
4410                 return NULL;
4411         }
4412         journal->j_private = sb;
4413         ext4_init_journal_params(sb, journal);
4414         return journal;
4415 }
4416
4417 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4418                                        dev_t j_dev)
4419 {
4420         struct buffer_head *bh;
4421         journal_t *journal;
4422         ext4_fsblk_t start;
4423         ext4_fsblk_t len;
4424         int hblock, blocksize;
4425         ext4_fsblk_t sb_block;
4426         unsigned long offset;
4427         struct ext4_super_block *es;
4428         struct block_device *bdev;
4429
4430         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4431
4432         bdev = ext4_blkdev_get(j_dev, sb);
4433         if (bdev == NULL)
4434                 return NULL;
4435
4436         blocksize = sb->s_blocksize;
4437         hblock = bdev_logical_block_size(bdev);
4438         if (blocksize < hblock) {
4439                 ext4_msg(sb, KERN_ERR,
4440                         "blocksize too small for journal device");
4441                 goto out_bdev;
4442         }
4443
4444         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4445         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4446         set_blocksize(bdev, blocksize);
4447         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4448                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4449                        "external journal");
4450                 goto out_bdev;
4451         }
4452
4453         es = (struct ext4_super_block *) (bh->b_data + offset);
4454         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4455             !(le32_to_cpu(es->s_feature_incompat) &
4456               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4457                 ext4_msg(sb, KERN_ERR, "external journal has "
4458                                         "bad superblock");
4459                 brelse(bh);
4460                 goto out_bdev;
4461         }
4462
4463         if ((le32_to_cpu(es->s_feature_ro_compat) &
4464              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4465             es->s_checksum != ext4_superblock_csum(sb, es)) {
4466                 ext4_msg(sb, KERN_ERR, "external journal has "
4467                                        "corrupt superblock");
4468                 brelse(bh);
4469                 goto out_bdev;
4470         }
4471
4472         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4473                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4474                 brelse(bh);
4475                 goto out_bdev;
4476         }
4477
4478         len = ext4_blocks_count(es);
4479         start = sb_block + 1;
4480         brelse(bh);     /* we're done with the superblock */
4481
4482         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4483                                         start, len, blocksize);
4484         if (!journal) {
4485                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4486                 goto out_bdev;
4487         }
4488         journal->j_private = sb;
4489         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4490         wait_on_buffer(journal->j_sb_buffer);
4491         if (!buffer_uptodate(journal->j_sb_buffer)) {
4492                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4493                 goto out_journal;
4494         }
4495         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4496                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4497                                         "user (unsupported) - %d",
4498                         be32_to_cpu(journal->j_superblock->s_nr_users));
4499                 goto out_journal;
4500         }
4501         EXT4_SB(sb)->journal_bdev = bdev;
4502         ext4_init_journal_params(sb, journal);
4503         return journal;
4504
4505 out_journal:
4506         jbd2_journal_destroy(journal);
4507 out_bdev:
4508         ext4_blkdev_put(bdev);
4509         return NULL;
4510 }
4511
4512 static int ext4_load_journal(struct super_block *sb,
4513                              struct ext4_super_block *es,
4514                              unsigned long journal_devnum)
4515 {
4516         journal_t *journal;
4517         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4518         dev_t journal_dev;
4519         int err = 0;
4520         int really_read_only;
4521
4522         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4523
4524         if (journal_devnum &&
4525             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4526                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4527                         "numbers have changed");
4528                 journal_dev = new_decode_dev(journal_devnum);
4529         } else
4530                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4531
4532         really_read_only = bdev_read_only(sb->s_bdev);
4533
4534         /*
4535          * Are we loading a blank journal or performing recovery after a
4536          * crash?  For recovery, we need to check in advance whether we
4537          * can get read-write access to the device.
4538          */
4539         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4540                 if (sb->s_flags & MS_RDONLY) {
4541                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4542                                         "required on readonly filesystem");
4543                         if (really_read_only) {
4544                                 ext4_msg(sb, KERN_ERR, "write access "
4545                                         "unavailable, cannot proceed");
4546                                 return -EROFS;
4547                         }
4548                         ext4_msg(sb, KERN_INFO, "write access will "
4549                                "be enabled during recovery");
4550                 }
4551         }
4552
4553         if (journal_inum && journal_dev) {
4554                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4555                        "and inode journals!");
4556                 return -EINVAL;
4557         }
4558
4559         if (journal_inum) {
4560                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4561                         return -EINVAL;
4562         } else {
4563                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4564                         return -EINVAL;
4565         }
4566
4567         if (!(journal->j_flags & JBD2_BARRIER))
4568                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4569
4570         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4571                 err = jbd2_journal_wipe(journal, !really_read_only);
4572         if (!err) {
4573                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4574                 if (save)
4575                         memcpy(save, ((char *) es) +
4576                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4577                 err = jbd2_journal_load(journal);
4578                 if (save)
4579                         memcpy(((char *) es) + EXT4_S_ERR_START,
4580                                save, EXT4_S_ERR_LEN);
4581                 kfree(save);
4582         }
4583
4584         if (err) {
4585                 ext4_msg(sb, KERN_ERR, "error loading journal");
4586                 jbd2_journal_destroy(journal);
4587                 return err;
4588         }
4589
4590         EXT4_SB(sb)->s_journal = journal;
4591         ext4_clear_journal_err(sb, es);
4592
4593         if (!really_read_only && journal_devnum &&
4594             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4595                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4596
4597                 /* Make sure we flush the recovery flag to disk. */
4598                 ext4_commit_super(sb, 1);
4599         }
4600
4601         return 0;
4602 }
4603
4604 static int ext4_commit_super(struct super_block *sb, int sync)
4605 {
4606         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4607         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4608         int error = 0;
4609
4610         if (!sbh || block_device_ejected(sb))
4611                 return error;
4612         if (buffer_write_io_error(sbh)) {
4613                 /*
4614                  * Oh, dear.  A previous attempt to write the
4615                  * superblock failed.  This could happen because the
4616                  * USB device was yanked out.  Or it could happen to
4617                  * be a transient write error and maybe the block will
4618                  * be remapped.  Nothing we can do but to retry the
4619                  * write and hope for the best.
4620                  */
4621                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4622                        "superblock detected");
4623                 clear_buffer_write_io_error(sbh);
4624                 set_buffer_uptodate(sbh);
4625         }
4626         /*
4627          * If the file system is mounted read-only, don't update the
4628          * superblock write time.  This avoids updating the superblock
4629          * write time when we are mounting the root file system
4630          * read/only but we need to replay the journal; at that point,
4631          * for people who are east of GMT and who make their clock
4632          * tick in localtime for Windows bug-for-bug compatibility,
4633          * the clock is set in the future, and this will cause e2fsck
4634          * to complain and force a full file system check.
4635          */
4636         if (!(sb->s_flags & MS_RDONLY))
4637                 es->s_wtime = cpu_to_le32(get_seconds());
4638         if (sb->s_bdev->bd_part)
4639                 es->s_kbytes_written =
4640                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4641                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4642                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4643         else
4644                 es->s_kbytes_written =
4645                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4646         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4647                 ext4_free_blocks_count_set(es,
4648                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4649                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4650         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4651                 es->s_free_inodes_count =
4652                         cpu_to_le32(percpu_counter_sum_positive(
4653                                 &EXT4_SB(sb)->s_freeinodes_counter));
4654         BUFFER_TRACE(sbh, "marking dirty");
4655         ext4_superblock_csum_set(sb);
4656         mark_buffer_dirty(sbh);
4657         if (sync) {
4658                 error = sync_dirty_buffer(sbh);
4659                 if (error)
4660                         return error;
4661
4662                 error = buffer_write_io_error(sbh);
4663                 if (error) {
4664                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4665                                "superblock");
4666                         clear_buffer_write_io_error(sbh);
4667                         set_buffer_uptodate(sbh);
4668                 }
4669         }
4670         return error;
4671 }
4672
4673 /*
4674  * Have we just finished recovery?  If so, and if we are mounting (or
4675  * remounting) the filesystem readonly, then we will end up with a
4676  * consistent fs on disk.  Record that fact.
4677  */
4678 static void ext4_mark_recovery_complete(struct super_block *sb,
4679                                         struct ext4_super_block *es)
4680 {
4681         journal_t *journal = EXT4_SB(sb)->s_journal;
4682
4683         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4684                 BUG_ON(journal != NULL);
4685                 return;
4686         }
4687         jbd2_journal_lock_updates(journal);
4688         if (jbd2_journal_flush(journal) < 0)
4689                 goto out;
4690
4691         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4692             sb->s_flags & MS_RDONLY) {
4693                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4694                 ext4_commit_super(sb, 1);
4695         }
4696
4697 out:
4698         jbd2_journal_unlock_updates(journal);
4699 }
4700
4701 /*
4702  * If we are mounting (or read-write remounting) a filesystem whose journal
4703  * has recorded an error from a previous lifetime, move that error to the
4704  * main filesystem now.
4705  */
4706 static void ext4_clear_journal_err(struct super_block *sb,
4707                                    struct ext4_super_block *es)
4708 {
4709         journal_t *journal;
4710         int j_errno;
4711         const char *errstr;
4712
4713         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4714
4715         journal = EXT4_SB(sb)->s_journal;
4716
4717         /*
4718          * Now check for any error status which may have been recorded in the
4719          * journal by a prior ext4_error() or ext4_abort()
4720          */
4721
4722         j_errno = jbd2_journal_errno(journal);
4723         if (j_errno) {
4724                 char nbuf[16];
4725
4726                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4727                 ext4_warning(sb, "Filesystem error recorded "
4728                              "from previous mount: %s", errstr);
4729                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4730
4731                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4732                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4733                 ext4_commit_super(sb, 1);
4734
4735                 jbd2_journal_clear_err(journal);
4736                 jbd2_journal_update_sb_errno(journal);
4737         }
4738 }
4739
4740 /*
4741  * Force the running and committing transactions to commit,
4742  * and wait on the commit.
4743  */
4744 int ext4_force_commit(struct super_block *sb)
4745 {
4746         journal_t *journal;
4747
4748         if (sb->s_flags & MS_RDONLY)
4749                 return 0;
4750
4751         journal = EXT4_SB(sb)->s_journal;
4752         return ext4_journal_force_commit(journal);
4753 }
4754
4755 static int ext4_sync_fs(struct super_block *sb, int wait)
4756 {
4757         int ret = 0;
4758         tid_t target;
4759         bool needs_barrier = false;
4760         struct ext4_sb_info *sbi = EXT4_SB(sb);
4761
4762         trace_ext4_sync_fs(sb, wait);
4763         flush_workqueue(sbi->rsv_conversion_wq);
4764         /*
4765          * Writeback quota in non-journalled quota case - journalled quota has
4766          * no dirty dquots
4767          */
4768         dquot_writeback_dquots(sb, -1);
4769         /*
4770          * Data writeback is possible w/o journal transaction, so barrier must
4771          * being sent at the end of the function. But we can skip it if
4772          * transaction_commit will do it for us.
4773          */
4774         if (sbi->s_journal) {
4775                 target = jbd2_get_latest_transaction(sbi->s_journal);
4776                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4777                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4778                         needs_barrier = true;
4779
4780                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4781                         if (wait)
4782                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4783                                                            target);
4784                 }
4785         } else if (wait && test_opt(sb, BARRIER))
4786                 needs_barrier = true;
4787         if (needs_barrier) {
4788                 int err;
4789                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4790                 if (!ret)
4791                         ret = err;
4792         }
4793
4794         return ret;
4795 }
4796
4797 /*
4798  * LVM calls this function before a (read-only) snapshot is created.  This
4799  * gives us a chance to flush the journal completely and mark the fs clean.
4800  *
4801  * Note that only this function cannot bring a filesystem to be in a clean
4802  * state independently. It relies on upper layer to stop all data & metadata
4803  * modifications.
4804  */
4805 static int ext4_freeze(struct super_block *sb)
4806 {
4807         int error = 0;
4808         journal_t *journal;
4809
4810         if (sb->s_flags & MS_RDONLY)
4811                 return 0;
4812
4813         journal = EXT4_SB(sb)->s_journal;
4814
4815         if (journal) {
4816                 /* Now we set up the journal barrier. */
4817                 jbd2_journal_lock_updates(journal);
4818
4819                 /*
4820                  * Don't clear the needs_recovery flag if we failed to
4821                  * flush the journal.
4822                  */
4823                 error = jbd2_journal_flush(journal);
4824                 if (error < 0)
4825                         goto out;
4826
4827                 /* Journal blocked and flushed, clear needs_recovery flag. */
4828                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4829         }
4830
4831         error = ext4_commit_super(sb, 1);
4832 out:
4833         if (journal)
4834                 /* we rely on upper layer to stop further updates */
4835                 jbd2_journal_unlock_updates(journal);
4836         return error;
4837 }
4838
4839 /*
4840  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4841  * flag here, even though the filesystem is not technically dirty yet.
4842  */
4843 static int ext4_unfreeze(struct super_block *sb)
4844 {
4845         if (sb->s_flags & MS_RDONLY)
4846                 return 0;
4847
4848         if (EXT4_SB(sb)->s_journal) {
4849                 /* Reset the needs_recovery flag before the fs is unlocked. */
4850                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4851         }
4852
4853         ext4_commit_super(sb, 1);
4854         return 0;
4855 }
4856
4857 /*
4858  * Structure to save mount options for ext4_remount's benefit
4859  */
4860 struct ext4_mount_options {
4861         unsigned long s_mount_opt;
4862         unsigned long s_mount_opt2;
4863         kuid_t s_resuid;
4864         kgid_t s_resgid;
4865         unsigned long s_commit_interval;
4866         u32 s_min_batch_time, s_max_batch_time;
4867 #ifdef CONFIG_QUOTA
4868         int s_jquota_fmt;
4869         char *s_qf_names[EXT4_MAXQUOTAS];
4870 #endif
4871 };
4872
4873 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4874 {
4875         struct ext4_super_block *es;
4876         struct ext4_sb_info *sbi = EXT4_SB(sb);
4877         unsigned long old_sb_flags;
4878         struct ext4_mount_options old_opts;
4879         int enable_quota = 0;
4880         ext4_group_t g;
4881         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4882         int err = 0;
4883 #ifdef CONFIG_QUOTA
4884         int i, j;
4885 #endif
4886         char *orig_data = kstrdup(data, GFP_KERNEL);
4887
4888         /* Store the original options */
4889         old_sb_flags = sb->s_flags;
4890         old_opts.s_mount_opt = sbi->s_mount_opt;
4891         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4892         old_opts.s_resuid = sbi->s_resuid;
4893         old_opts.s_resgid = sbi->s_resgid;
4894         old_opts.s_commit_interval = sbi->s_commit_interval;
4895         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4896         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4897 #ifdef CONFIG_QUOTA
4898         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4899         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4900                 if (sbi->s_qf_names[i]) {
4901                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4902                                                          GFP_KERNEL);
4903                         if (!old_opts.s_qf_names[i]) {
4904                                 for (j = 0; j < i; j++)
4905                                         kfree(old_opts.s_qf_names[j]);
4906                                 kfree(orig_data);
4907                                 return -ENOMEM;
4908                         }
4909                 } else
4910                         old_opts.s_qf_names[i] = NULL;
4911 #endif
4912         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4913                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4914
4915         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4916                 err = -EINVAL;
4917                 goto restore_opts;
4918         }
4919
4920         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4921             test_opt(sb, JOURNAL_CHECKSUM)) {
4922                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4923                          "during remount not supported; ignoring");
4924                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4925         }
4926
4927         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4928                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4929                         ext4_msg(sb, KERN_ERR, "can't mount with "
4930                                  "both data=journal and delalloc");
4931                         err = -EINVAL;
4932                         goto restore_opts;
4933                 }
4934                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4935                         ext4_msg(sb, KERN_ERR, "can't mount with "
4936                                  "both data=journal and dioread_nolock");
4937                         err = -EINVAL;
4938                         goto restore_opts;
4939                 }
4940                 if (test_opt(sb, DAX)) {
4941                         ext4_msg(sb, KERN_ERR, "can't mount with "
4942                                  "both data=journal and dax");
4943                         err = -EINVAL;
4944                         goto restore_opts;
4945                 }
4946         }
4947
4948         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4949                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4950                         "dax flag with busy inodes while remounting");
4951                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4952         }
4953
4954         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4955                 ext4_abort(sb, "Abort forced by user");
4956
4957         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4958                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4959
4960         es = sbi->s_es;
4961
4962         if (sbi->s_journal) {
4963                 ext4_init_journal_params(sb, sbi->s_journal);
4964                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4965         }
4966
4967         if (*flags & MS_LAZYTIME)
4968                 sb->s_flags |= MS_LAZYTIME;
4969
4970         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4971                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4972                         err = -EROFS;
4973                         goto restore_opts;
4974                 }
4975
4976                 if (*flags & MS_RDONLY) {
4977                         err = sync_filesystem(sb);
4978                         if (err < 0)
4979                                 goto restore_opts;
4980                         err = dquot_suspend(sb, -1);
4981                         if (err < 0)
4982                                 goto restore_opts;
4983
4984                         /*
4985                          * First of all, the unconditional stuff we have to do
4986                          * to disable replay of the journal when we next remount
4987                          */
4988                         sb->s_flags |= MS_RDONLY;
4989
4990                         /*
4991                          * OK, test if we are remounting a valid rw partition
4992                          * readonly, and if so set the rdonly flag and then
4993                          * mark the partition as valid again.
4994                          */
4995                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4996                             (sbi->s_mount_state & EXT4_VALID_FS))
4997                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4998
4999                         if (sbi->s_journal)
5000                                 ext4_mark_recovery_complete(sb, es);
5001                 } else {
5002                         /* Make sure we can mount this feature set readwrite */
5003                         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5004                                         EXT4_FEATURE_RO_COMPAT_READONLY) ||
5005                             !ext4_feature_set_ok(sb, 0)) {
5006                                 err = -EROFS;
5007                                 goto restore_opts;
5008                         }
5009                         /*
5010                          * Make sure the group descriptor checksums
5011                          * are sane.  If they aren't, refuse to remount r/w.
5012                          */
5013                         for (g = 0; g < sbi->s_groups_count; g++) {
5014                                 struct ext4_group_desc *gdp =
5015                                         ext4_get_group_desc(sb, g, NULL);
5016
5017                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5018                                         ext4_msg(sb, KERN_ERR,
5019                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5020                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
5021                                                le16_to_cpu(gdp->bg_checksum));
5022                                         err = -EINVAL;
5023                                         goto restore_opts;
5024                                 }
5025                         }
5026
5027                         /*
5028                          * If we have an unprocessed orphan list hanging
5029                          * around from a previously readonly bdev mount,
5030                          * require a full umount/remount for now.
5031                          */
5032                         if (es->s_last_orphan) {
5033                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5034                                        "remount RDWR because of unprocessed "
5035                                        "orphan inode list.  Please "
5036                                        "umount/remount instead");
5037                                 err = -EINVAL;
5038                                 goto restore_opts;
5039                         }
5040
5041                         /*
5042                          * Mounting a RDONLY partition read-write, so reread
5043                          * and store the current valid flag.  (It may have
5044                          * been changed by e2fsck since we originally mounted
5045                          * the partition.)
5046                          */
5047                         if (sbi->s_journal)
5048                                 ext4_clear_journal_err(sb, es);
5049                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5050                         if (!ext4_setup_super(sb, es, 0))
5051                                 sb->s_flags &= ~MS_RDONLY;
5052                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
5053                                                      EXT4_FEATURE_INCOMPAT_MMP))
5054                                 if (ext4_multi_mount_protect(sb,
5055                                                 le64_to_cpu(es->s_mmp_block))) {
5056                                         err = -EROFS;
5057                                         goto restore_opts;
5058                                 }
5059                         enable_quota = 1;
5060                 }
5061         }
5062
5063         /*
5064          * Reinitialize lazy itable initialization thread based on
5065          * current settings
5066          */
5067         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5068                 ext4_unregister_li_request(sb);
5069         else {
5070                 ext4_group_t first_not_zeroed;
5071                 first_not_zeroed = ext4_has_uninit_itable(sb);
5072                 ext4_register_li_request(sb, first_not_zeroed);
5073         }
5074
5075         ext4_setup_system_zone(sb);
5076         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5077                 ext4_commit_super(sb, 1);
5078
5079 #ifdef CONFIG_QUOTA
5080         /* Release old quota file names */
5081         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5082                 kfree(old_opts.s_qf_names[i]);
5083         if (enable_quota) {
5084                 if (sb_any_quota_suspended(sb))
5085                         dquot_resume(sb, -1);
5086                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5087                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5088                         err = ext4_enable_quotas(sb);
5089                         if (err)
5090                                 goto restore_opts;
5091                 }
5092         }
5093 #endif
5094
5095         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5096         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5097         kfree(orig_data);
5098         return 0;
5099
5100 restore_opts:
5101         sb->s_flags = old_sb_flags;
5102         sbi->s_mount_opt = old_opts.s_mount_opt;
5103         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5104         sbi->s_resuid = old_opts.s_resuid;
5105         sbi->s_resgid = old_opts.s_resgid;
5106         sbi->s_commit_interval = old_opts.s_commit_interval;
5107         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5108         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5109 #ifdef CONFIG_QUOTA
5110         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5111         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5112                 kfree(sbi->s_qf_names[i]);
5113                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5114         }
5115 #endif
5116         kfree(orig_data);
5117         return err;
5118 }
5119
5120 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5121 {
5122         struct super_block *sb = dentry->d_sb;
5123         struct ext4_sb_info *sbi = EXT4_SB(sb);
5124         struct ext4_super_block *es = sbi->s_es;
5125         ext4_fsblk_t overhead = 0, resv_blocks;
5126         u64 fsid;
5127         s64 bfree;
5128         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5129
5130         if (!test_opt(sb, MINIX_DF))
5131                 overhead = sbi->s_overhead;
5132
5133         buf->f_type = EXT4_SUPER_MAGIC;
5134         buf->f_bsize = sb->s_blocksize;
5135         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5136         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5137                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5138         /* prevent underflow in case that few free space is available */
5139         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5140         buf->f_bavail = buf->f_bfree -
5141                         (ext4_r_blocks_count(es) + resv_blocks);
5142         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5143                 buf->f_bavail = 0;
5144         buf->f_files = le32_to_cpu(es->s_inodes_count);
5145         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5146         buf->f_namelen = EXT4_NAME_LEN;
5147         fsid = le64_to_cpup((void *)es->s_uuid) ^
5148                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5149         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5150         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5151
5152         return 0;
5153 }
5154
5155 /* Helper function for writing quotas on sync - we need to start transaction
5156  * before quota file is locked for write. Otherwise the are possible deadlocks:
5157  * Process 1                         Process 2
5158  * ext4_create()                     quota_sync()
5159  *   jbd2_journal_start()                  write_dquot()
5160  *   dquot_initialize()                         down(dqio_mutex)
5161  *     down(dqio_mutex)                    jbd2_journal_start()
5162  *
5163  */
5164
5165 #ifdef CONFIG_QUOTA
5166
5167 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5168 {
5169         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5170 }
5171
5172 static int ext4_write_dquot(struct dquot *dquot)
5173 {
5174         int ret, err;
5175         handle_t *handle;
5176         struct inode *inode;
5177
5178         inode = dquot_to_inode(dquot);
5179         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5180                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5181         if (IS_ERR(handle))
5182                 return PTR_ERR(handle);
5183         ret = dquot_commit(dquot);
5184         err = ext4_journal_stop(handle);
5185         if (!ret)
5186                 ret = err;
5187         return ret;
5188 }
5189
5190 static int ext4_acquire_dquot(struct dquot *dquot)
5191 {
5192         int ret, err;
5193         handle_t *handle;
5194
5195         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5196                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5197         if (IS_ERR(handle))
5198                 return PTR_ERR(handle);
5199         ret = dquot_acquire(dquot);
5200         err = ext4_journal_stop(handle);
5201         if (!ret)
5202                 ret = err;
5203         return ret;
5204 }
5205
5206 static int ext4_release_dquot(struct dquot *dquot)
5207 {
5208         int ret, err;
5209         handle_t *handle;
5210
5211         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5212                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5213         if (IS_ERR(handle)) {
5214                 /* Release dquot anyway to avoid endless cycle in dqput() */
5215                 dquot_release(dquot);
5216                 return PTR_ERR(handle);
5217         }
5218         ret = dquot_release(dquot);
5219         err = ext4_journal_stop(handle);
5220         if (!ret)
5221                 ret = err;
5222         return ret;
5223 }
5224
5225 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5226 {
5227         struct super_block *sb = dquot->dq_sb;
5228         struct ext4_sb_info *sbi = EXT4_SB(sb);
5229
5230         /* Are we journaling quotas? */
5231         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5232             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5233                 dquot_mark_dquot_dirty(dquot);
5234                 return ext4_write_dquot(dquot);
5235         } else {
5236                 return dquot_mark_dquot_dirty(dquot);
5237         }
5238 }
5239
5240 static int ext4_write_info(struct super_block *sb, int type)
5241 {
5242         int ret, err;
5243         handle_t *handle;
5244
5245         /* Data block + inode block */
5246         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5247         if (IS_ERR(handle))
5248                 return PTR_ERR(handle);
5249         ret = dquot_commit_info(sb, type);
5250         err = ext4_journal_stop(handle);
5251         if (!ret)
5252                 ret = err;
5253         return ret;
5254 }
5255
5256 /*
5257  * Turn on quotas during mount time - we need to find
5258  * the quota file and such...
5259  */
5260 static int ext4_quota_on_mount(struct super_block *sb, int type)
5261 {
5262         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5263                                         EXT4_SB(sb)->s_jquota_fmt, type);
5264 }
5265
5266 /*
5267  * Standard function to be called on quota_on
5268  */
5269 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5270                          struct path *path)
5271 {
5272         int err;
5273
5274         if (!test_opt(sb, QUOTA))
5275                 return -EINVAL;
5276
5277         /* Quotafile not on the same filesystem? */
5278         if (path->dentry->d_sb != sb)
5279                 return -EXDEV;
5280         /* Journaling quota? */
5281         if (EXT4_SB(sb)->s_qf_names[type]) {
5282                 /* Quotafile not in fs root? */
5283                 if (path->dentry->d_parent != sb->s_root)
5284                         ext4_msg(sb, KERN_WARNING,
5285                                 "Quota file not on filesystem root. "
5286                                 "Journaled quota will not work");
5287         }
5288
5289         /*
5290          * When we journal data on quota file, we have to flush journal to see
5291          * all updates to the file when we bypass pagecache...
5292          */
5293         if (EXT4_SB(sb)->s_journal &&
5294             ext4_should_journal_data(d_inode(path->dentry))) {
5295                 /*
5296                  * We don't need to lock updates but journal_flush() could
5297                  * otherwise be livelocked...
5298                  */
5299                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5300                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5301                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5302                 if (err)
5303                         return err;
5304         }
5305
5306         return dquot_quota_on(sb, type, format_id, path);
5307 }
5308
5309 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5310                              unsigned int flags)
5311 {
5312         int err;
5313         struct inode *qf_inode;
5314         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5315                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5316                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5317         };
5318
5319         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5320
5321         if (!qf_inums[type])
5322                 return -EPERM;
5323
5324         qf_inode = ext4_iget(sb, qf_inums[type]);
5325         if (IS_ERR(qf_inode)) {
5326                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5327                 return PTR_ERR(qf_inode);
5328         }
5329
5330         /* Don't account quota for quota files to avoid recursion */
5331         qf_inode->i_flags |= S_NOQUOTA;
5332         err = dquot_enable(qf_inode, type, format_id, flags);
5333         iput(qf_inode);
5334
5335         return err;
5336 }
5337
5338 /* Enable usage tracking for all quota types. */
5339 static int ext4_enable_quotas(struct super_block *sb)
5340 {
5341         int type, err = 0;
5342         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5343                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5344                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5345         };
5346
5347         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5348         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5349                 if (qf_inums[type]) {
5350                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5351                                                 DQUOT_USAGE_ENABLED);
5352                         if (err) {
5353                                 ext4_warning(sb,
5354                                         "Failed to enable quota tracking "
5355                                         "(type=%d, err=%d). Please run "
5356                                         "e2fsck to fix.", type, err);
5357                                 return err;
5358                         }
5359                 }
5360         }
5361         return 0;
5362 }
5363
5364 static int ext4_quota_off(struct super_block *sb, int type)
5365 {
5366         struct inode *inode = sb_dqopt(sb)->files[type];
5367         handle_t *handle;
5368
5369         /* Force all delayed allocation blocks to be allocated.
5370          * Caller already holds s_umount sem */
5371         if (test_opt(sb, DELALLOC))
5372                 sync_filesystem(sb);
5373
5374         if (!inode)
5375                 goto out;
5376
5377         /* Update modification times of quota files when userspace can
5378          * start looking at them */
5379         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5380         if (IS_ERR(handle))
5381                 goto out;
5382         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5383         ext4_mark_inode_dirty(handle, inode);
5384         ext4_journal_stop(handle);
5385
5386 out:
5387         return dquot_quota_off(sb, type);
5388 }
5389
5390 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5391  * acquiring the locks... As quota files are never truncated and quota code
5392  * itself serializes the operations (and no one else should touch the files)
5393  * we don't have to be afraid of races */
5394 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5395                                size_t len, loff_t off)
5396 {
5397         struct inode *inode = sb_dqopt(sb)->files[type];
5398         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5399         int offset = off & (sb->s_blocksize - 1);
5400         int tocopy;
5401         size_t toread;
5402         struct buffer_head *bh;
5403         loff_t i_size = i_size_read(inode);
5404
5405         if (off > i_size)
5406                 return 0;
5407         if (off+len > i_size)
5408                 len = i_size-off;
5409         toread = len;
5410         while (toread > 0) {
5411                 tocopy = sb->s_blocksize - offset < toread ?
5412                                 sb->s_blocksize - offset : toread;
5413                 bh = ext4_bread(NULL, inode, blk, 0);
5414                 if (IS_ERR(bh))
5415                         return PTR_ERR(bh);
5416                 if (!bh)        /* A hole? */
5417                         memset(data, 0, tocopy);
5418                 else
5419                         memcpy(data, bh->b_data+offset, tocopy);
5420                 brelse(bh);
5421                 offset = 0;
5422                 toread -= tocopy;
5423                 data += tocopy;
5424                 blk++;
5425         }
5426         return len;
5427 }
5428
5429 /* Write to quotafile (we know the transaction is already started and has
5430  * enough credits) */
5431 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5432                                 const char *data, size_t len, loff_t off)
5433 {
5434         struct inode *inode = sb_dqopt(sb)->files[type];
5435         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5436         int err, offset = off & (sb->s_blocksize - 1);
5437         struct buffer_head *bh;
5438         handle_t *handle = journal_current_handle();
5439
5440         if (EXT4_SB(sb)->s_journal && !handle) {
5441                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5442                         " cancelled because transaction is not started",
5443                         (unsigned long long)off, (unsigned long long)len);
5444                 return -EIO;
5445         }
5446         /*
5447          * Since we account only one data block in transaction credits,
5448          * then it is impossible to cross a block boundary.
5449          */
5450         if (sb->s_blocksize - offset < len) {
5451                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5452                         " cancelled because not block aligned",
5453                         (unsigned long long)off, (unsigned long long)len);
5454                 return -EIO;
5455         }
5456
5457         bh = ext4_bread(handle, inode, blk, 1);
5458         if (IS_ERR(bh))
5459                 return PTR_ERR(bh);
5460         if (!bh)
5461                 goto out;
5462         BUFFER_TRACE(bh, "get write access");
5463         err = ext4_journal_get_write_access(handle, bh);
5464         if (err) {
5465                 brelse(bh);
5466                 return err;
5467         }
5468         lock_buffer(bh);
5469         memcpy(bh->b_data+offset, data, len);
5470         flush_dcache_page(bh->b_page);
5471         unlock_buffer(bh);
5472         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5473         brelse(bh);
5474 out:
5475         if (inode->i_size < off + len) {
5476                 i_size_write(inode, off + len);
5477                 EXT4_I(inode)->i_disksize = inode->i_size;
5478                 ext4_mark_inode_dirty(handle, inode);
5479         }
5480         return len;
5481 }
5482
5483 #endif
5484
5485 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5486                        const char *dev_name, void *data)
5487 {
5488         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5489 }
5490
5491 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5492 static inline void register_as_ext2(void)
5493 {
5494         int err = register_filesystem(&ext2_fs_type);
5495         if (err)
5496                 printk(KERN_WARNING
5497                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5498 }
5499
5500 static inline void unregister_as_ext2(void)
5501 {
5502         unregister_filesystem(&ext2_fs_type);
5503 }
5504
5505 static inline int ext2_feature_set_ok(struct super_block *sb)
5506 {
5507         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5508                 return 0;
5509         if (sb->s_flags & MS_RDONLY)
5510                 return 1;
5511         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5512                 return 0;
5513         return 1;
5514 }
5515 #else
5516 static inline void register_as_ext2(void) { }
5517 static inline void unregister_as_ext2(void) { }
5518 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5519 #endif
5520
5521 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5522 static inline void register_as_ext3(void)
5523 {
5524         int err = register_filesystem(&ext3_fs_type);
5525         if (err)
5526                 printk(KERN_WARNING
5527                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5528 }
5529
5530 static inline void unregister_as_ext3(void)
5531 {
5532         unregister_filesystem(&ext3_fs_type);
5533 }
5534
5535 static inline int ext3_feature_set_ok(struct super_block *sb)
5536 {
5537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5538                 return 0;
5539         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5540                 return 0;
5541         if (sb->s_flags & MS_RDONLY)
5542                 return 1;
5543         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5544                 return 0;
5545         return 1;
5546 }
5547 #else
5548 static inline void register_as_ext3(void) { }
5549 static inline void unregister_as_ext3(void) { }
5550 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5551 #endif
5552
5553 static struct file_system_type ext4_fs_type = {
5554         .owner          = THIS_MODULE,
5555         .name           = "ext4",
5556         .mount          = ext4_mount,
5557         .kill_sb        = kill_block_super,
5558         .fs_flags       = FS_REQUIRES_DEV,
5559 };
5560 MODULE_ALIAS_FS("ext4");
5561
5562 static int __init ext4_init_feat_adverts(void)
5563 {
5564         struct ext4_features *ef;
5565         int ret = -ENOMEM;
5566
5567         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5568         if (!ef)
5569                 goto out;
5570
5571         ef->f_kobj.kset = ext4_kset;
5572         init_completion(&ef->f_kobj_unregister);
5573         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5574                                    "features");
5575         if (ret) {
5576                 kfree(ef);
5577                 goto out;
5578         }
5579
5580         ext4_feat = ef;
5581         ret = 0;
5582 out:
5583         return ret;
5584 }
5585
5586 static void ext4_exit_feat_adverts(void)
5587 {
5588         kobject_put(&ext4_feat->f_kobj);
5589         wait_for_completion(&ext4_feat->f_kobj_unregister);
5590         kfree(ext4_feat);
5591 }
5592
5593 /* Shared across all ext4 file systems */
5594 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5595 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5596
5597 static int __init ext4_init_fs(void)
5598 {
5599         int i, err;
5600
5601         ext4_li_info = NULL;
5602         mutex_init(&ext4_li_mtx);
5603
5604         /* Build-time check for flags consistency */
5605         ext4_check_flag_values();
5606
5607         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5608                 mutex_init(&ext4__aio_mutex[i]);
5609                 init_waitqueue_head(&ext4__ioend_wq[i]);
5610         }
5611
5612         err = ext4_init_es();
5613         if (err)
5614                 return err;
5615
5616         err = ext4_init_pageio();
5617         if (err)
5618                 goto out7;
5619
5620         err = ext4_init_system_zone();
5621         if (err)
5622                 goto out6;
5623         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5624         if (!ext4_kset) {
5625                 err = -ENOMEM;
5626                 goto out5;
5627         }
5628         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5629
5630         err = ext4_init_feat_adverts();
5631         if (err)
5632                 goto out4;
5633
5634         err = ext4_init_mballoc();
5635         if (err)
5636                 goto out2;
5637         else
5638                 ext4_mballoc_ready = 1;
5639         err = init_inodecache();
5640         if (err)
5641                 goto out1;
5642         register_as_ext3();
5643         register_as_ext2();
5644         err = register_filesystem(&ext4_fs_type);
5645         if (err)
5646                 goto out;
5647
5648         return 0;
5649 out:
5650         unregister_as_ext2();
5651         unregister_as_ext3();
5652         destroy_inodecache();
5653 out1:
5654         ext4_mballoc_ready = 0;
5655         ext4_exit_mballoc();
5656 out2:
5657         ext4_exit_feat_adverts();
5658 out4:
5659         if (ext4_proc_root)
5660                 remove_proc_entry("fs/ext4", NULL);
5661         kset_unregister(ext4_kset);
5662 out5:
5663         ext4_exit_system_zone();
5664 out6:
5665         ext4_exit_pageio();
5666 out7:
5667         ext4_exit_es();
5668
5669         return err;
5670 }
5671
5672 static void __exit ext4_exit_fs(void)
5673 {
5674         ext4_destroy_lazyinit_thread();
5675         unregister_as_ext2();
5676         unregister_as_ext3();
5677         unregister_filesystem(&ext4_fs_type);
5678         destroy_inodecache();
5679         ext4_exit_mballoc();
5680         ext4_exit_feat_adverts();
5681         remove_proc_entry("fs/ext4", NULL);
5682         kset_unregister(ext4_kset);
5683         ext4_exit_system_zone();
5684         ext4_exit_pageio();
5685         ext4_exit_es();
5686 }
5687
5688 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5689 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5690 MODULE_LICENSE("GPL");
5691 module_init(ext4_init_fs)
5692 module_exit(ext4_exit_fs)