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/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/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 ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  84 static struct file_system_type ext2_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext2",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 MODULE_ALIAS_FS("ext2");
  92 MODULE_ALIAS("ext2");
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 static struct file_system_type ext3_fs_type = {
 100         .owner          = THIS_MODULE,
 101         .name           = "ext3",
 102         .mount          = ext4_mount,
 103         .kill_sb        = kill_block_super,
 104         .fs_flags       = FS_REQUIRES_DEV,
 105 };
 106 MODULE_ALIAS_FS("ext3");
 107 MODULE_ALIAS("ext3");
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109
 110 static int ext4_verify_csum_type(struct super_block *sb,
 111                                  struct ext4_super_block *es)
 112 {
 113         if (!ext4_has_feature_metadata_csum(sb))
 114                 return 1;
 115
 116         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 117 }
 118
 119 static __le32 ext4_superblock_csum(struct super_block *sb,
 120                                    struct ext4_super_block *es)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         int offset = offsetof(struct ext4_super_block, s_checksum);
 124         __u32 csum;
 125
 126         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_superblock_csum_verify(struct super_block *sb,
 132                                        struct ext4_super_block *es)
 133 {
 134         if (!ext4_has_metadata_csum(sb))
 135                 return 1;
 136
 137         return es->s_checksum == ext4_superblock_csum(sb, es);
 138 }
 139
 140 void ext4_superblock_csum_set(struct super_block *sb)
 141 {
 142         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 143
 144         if (!ext4_has_metadata_csum(sb))
 145                 return;
 146
 147         es->s_checksum = ext4_superblock_csum(sb, es);
 148 }
 149
 150 void *ext4_kvmalloc(size_t size, gfp_t flags)
 151 {
 152         void *ret;
 153
 154         ret = kmalloc(size, flags | __GFP_NOWARN);
 155         if (!ret)
 156                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 157         return ret;
 158 }
 159
 160 void *ext4_kvzalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kzalloc(size, flags | __GFP_NOWARN);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 171                                struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 176 }
 177
 178 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_table_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 192 }
 193
 194 __u32 ext4_free_group_clusters(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_free_inodes_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_used_dirs_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_itable_unused_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_itable_unused_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 224 }
 225
 226 void ext4_block_bitmap_set(struct super_block *sb,
 227                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_inode_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_table_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_free_group_clusters_set(struct super_block *sb,
 251                                   struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_free_inodes_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_used_dirs_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_itable_unused_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282
 283 static void __save_error_info(struct super_block *sb, const char *func,
 284                             unsigned int line)
 285 {
 286         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 287
 288         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 289         if (bdev_read_only(sb->s_bdev))
 290                 return;
 291         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 292         es->s_last_error_time = cpu_to_le32(get_seconds());
 293         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 294         es->s_last_error_line = cpu_to_le32(line);
 295         if (!es->s_first_error_time) {
 296                 es->s_first_error_time = es->s_last_error_time;
 297                 strncpy(es->s_first_error_func, func,
 298                         sizeof(es->s_first_error_func));
 299                 es->s_first_error_line = cpu_to_le32(line);
 300                 es->s_first_error_ino = es->s_last_error_ino;
 301                 es->s_first_error_block = es->s_last_error_block;
 302         }
 303         /*
 304          * Start the daily error reporting function if it hasn't been
 305          * started already
 306          */
 307         if (!es->s_error_count)
 308                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 309         le32_add_cpu(&es->s_error_count, 1);
 310 }
 311
 312 static void save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         __save_error_info(sb, func, line);
 316         ext4_commit_super(sb, 1);
 317 }
 318
 319 /*
 320  * The del_gendisk() function uninitializes the disk-specific data
 321  * structures, including the bdi structure, without telling anyone
 322  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 323  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 324  * This is a kludge to prevent these oops until we can put in a proper
 325  * hook in del_gendisk() to inform the VFS and file system layers.
 326  */
 327 static int block_device_ejected(struct super_block *sb)
 328 {
 329         struct inode *bd_inode = sb->s_bdev->bd_inode;
 330         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 331
 332         return bdi->dev == NULL;
 333 }
 334
 335 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 336 {
 337         struct super_block              *sb = journal->j_private;
 338         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 339         int                             error = is_journal_aborted(journal);
 340         struct ext4_journal_cb_entry    *jce;
 341
 342         BUG_ON(txn->t_state == T_FINISHED);
 343         spin_lock(&sbi->s_md_lock);
 344         while (!list_empty(&txn->t_private_list)) {
 345                 jce = list_entry(txn->t_private_list.next,
 346                                  struct ext4_journal_cb_entry, jce_list);
 347                 list_del_init(&jce->jce_list);
 348                 spin_unlock(&sbi->s_md_lock);
 349                 jce->jce_func(sb, jce, error);
 350                 spin_lock(&sbi->s_md_lock);
 351         }
 352         spin_unlock(&sbi->s_md_lock);
 353 }
 354
 355 /* Deal with the reporting of failure conditions on a filesystem such as
 356  * inconsistencies detected or read IO failures.
 357  *
 358  * On ext2, we can store the error state of the filesystem in the
 359  * superblock.  That is not possible on ext4, because we may have other
 360  * write ordering constraints on the superblock which prevent us from
 361  * writing it out straight away; and given that the journal is about to
 362  * be aborted, we can't rely on the current, or future, transactions to
 363  * write out the superblock safely.
 364  *
 365  * We'll just use the jbd2_journal_abort() error code to record an error in
 366  * the journal instead.  On recovery, the journal will complain about
 367  * that error until we've noted it down and cleared it.
 368  */
 369
 370 static void ext4_handle_error(struct super_block *sb)
 371 {
 372         if (sb->s_flags & MS_RDONLY)
 373                 return;
 374
 375         if (!test_opt(sb, ERRORS_CONT)) {
 376                 journal_t *journal = EXT4_SB(sb)->s_journal;
 377
 378                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 379                 if (journal)
 380                         jbd2_journal_abort(journal, -EIO);
 381         }
 382         if (test_opt(sb, ERRORS_RO)) {
 383                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 384                 /*
 385                  * Make sure updated value of ->s_mount_flags will be visible
 386                  * before ->s_flags update
 387                  */
 388                 smp_wmb();
 389                 sb->s_flags |= MS_RDONLY;
 390         }
 391         if (test_opt(sb, ERRORS_PANIC)) {
 392                 if (EXT4_SB(sb)->s_journal &&
 393                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 394                         return;
 395                 panic("EXT4-fs (device %s): panic forced after error\n",
 396                         sb->s_id);
 397         }
 398 }
 399
 400 #define ext4_error_ratelimit(sb)                                        \
 401                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 402                              "EXT4-fs error")
 403
 404 void __ext4_error(struct super_block *sb, const char *function,
 405                   unsigned int line, const char *fmt, ...)
 406 {
 407         struct va_format vaf;
 408         va_list args;
 409
 410         if (ext4_error_ratelimit(sb)) {
 411                 va_start(args, fmt);
 412                 vaf.fmt = fmt;
 413                 vaf.va = &args;
 414                 printk(KERN_CRIT
 415                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 416                        sb->s_id, function, line, current->comm, &vaf);
 417                 va_end(args);
 418         }
 419         save_error_info(sb, function, line);
 420         ext4_handle_error(sb);
 421 }
 422
 423 void __ext4_error_inode(struct inode *inode, const char *function,
 424                         unsigned int line, ext4_fsblk_t block,
 425                         const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         if (ext4_error_ratelimit(inode->i_sb)) {
 434                 va_start(args, fmt);
 435                 vaf.fmt = fmt;
 436                 vaf.va = &args;
 437                 if (block)
 438                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 439                                "inode #%lu: block %llu: comm %s: %pV\n",
 440                                inode->i_sb->s_id, function, line, inode->i_ino,
 441                                block, current->comm, &vaf);
 442                 else
 443                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 444                                "inode #%lu: comm %s: %pV\n",
 445                                inode->i_sb->s_id, function, line, inode->i_ino,
 446                                current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(inode->i_sb, function, line);
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 void __ext4_error_file(struct file *file, const char *function,
 454                        unsigned int line, ext4_fsblk_t block,
 455                        const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es;
 460         struct inode *inode = file_inode(file);
 461         char pathname[80], *path;
 462
 463         es = EXT4_SB(inode->i_sb)->s_es;
 464         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 465         if (ext4_error_ratelimit(inode->i_sb)) {
 466                 path = file_path(file, pathname, sizeof(pathname));
 467                 if (IS_ERR(path))
 468                         path = "(unknown)";
 469                 va_start(args, fmt);
 470                 vaf.fmt = fmt;
 471                 vaf.va = &args;
 472                 if (block)
 473                         printk(KERN_CRIT
 474                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 475                                "block %llu: comm %s: path %s: %pV\n",
 476                                inode->i_sb->s_id, function, line, inode->i_ino,
 477                                block, current->comm, path, &vaf);
 478                 else
 479                         printk(KERN_CRIT
 480                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 481                                "comm %s: path %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                current->comm, path, &vaf);
 484                 va_end(args);
 485         }
 486         save_error_info(inode->i_sb, function, line);
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EFSCORRUPTED:
 497                 errstr = "Corrupt filesystem";
 498                 break;
 499         case -EFSBADCRC:
 500                 errstr = "Filesystem failed CRC";
 501                 break;
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         if (ext4_error_ratelimit(sb)) {
 547                 errstr = ext4_decode_error(sb, errno, nbuf);
 548                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 549                        sb->s_id, function, line, errstr);
 550         }
 551
 552         save_error_info(sb, function, line);
 553         ext4_handle_error(sb);
 554 }
 555
 556 /*
 557  * ext4_abort is a much stronger failure handler than ext4_error.  The
 558  * abort function may be used to deal with unrecoverable failures such
 559  * as journal IO errors or ENOMEM at a critical moment in log management.
 560  *
 561  * We unconditionally force the filesystem into an ABORT|READONLY state,
 562  * unless the error response on the fs has been set to panic in which
 563  * case we take the easy way out and panic immediately.
 564  */
 565
 566 void __ext4_abort(struct super_block *sb, const char *function,
 567                 unsigned int line, const char *fmt, ...)
 568 {
 569         va_list args;
 570
 571         save_error_info(sb, function, line);
 572         va_start(args, fmt);
 573         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 574                function, line);
 575         vprintk(fmt, args);
 576         printk("\n");
 577         va_end(args);
 578
 579         if ((sb->s_flags & MS_RDONLY) == 0) {
 580                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 /*
 583                  * Make sure updated value of ->s_mount_flags will be visible
 584                  * before ->s_flags update
 585                  */
 586                 smp_wmb();
 587                 sb->s_flags |= MS_RDONLY;
 588                 if (EXT4_SB(sb)->s_journal)
 589                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 590                 save_error_info(sb, function, line);
 591         }
 592         if (test_opt(sb, ERRORS_PANIC)) {
 593                 if (EXT4_SB(sb)->s_journal &&
 594                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 595                         return;
 596                 panic("EXT4-fs panic from previous error\n");
 597         }
 598 }
 599
 600 void __ext4_msg(struct super_block *sb,
 601                 const char *prefix, const char *fmt, ...)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605
 606         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 607                 return;
 608
 609         va_start(args, fmt);
 610         vaf.fmt = fmt;
 611         vaf.va = &args;
 612         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 613         va_end(args);
 614 }
 615
 616 #define ext4_warning_ratelimit(sb)                                      \
 617                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 618                              "EXT4-fs warning")
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!ext4_warning_ratelimit(sb))
 627                 return;
 628
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635 }
 636
 637 void __ext4_warning_inode(const struct inode *inode, const char *function,
 638                           unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (!ext4_warning_ratelimit(inode->i_sb))
 644                 return;
 645
 646         va_start(args, fmt);
 647         vaf.fmt = fmt;
 648         vaf.va = &args;
 649         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 650                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 651                function, line, inode->i_ino, current->comm, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_grp_locked_error(const char *function, unsigned int line,
 656                              struct super_block *sb, ext4_group_t grp,
 657                              unsigned long ino, ext4_fsblk_t block,
 658                              const char *fmt, ...)
 659 __releases(bitlock)
 660 __acquires(bitlock)
 661 {
 662         struct va_format vaf;
 663         va_list args;
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         es->s_last_error_ino = cpu_to_le32(ino);
 667         es->s_last_error_block = cpu_to_le64(block);
 668         __save_error_info(sb, function, line);
 669
 670         if (ext4_error_ratelimit(sb)) {
 671                 va_start(args, fmt);
 672                 vaf.fmt = fmt;
 673                 vaf.va = &args;
 674                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 675                        sb->s_id, function, line, grp);
 676                 if (ino)
 677                         printk(KERN_CONT "inode %lu: ", ino);
 678                 if (block)
 679                         printk(KERN_CONT "block %llu:",
 680                                (unsigned long long) block);
 681                 printk(KERN_CONT "%pV\n", &vaf);
 682                 va_end(args);
 683         }
 684
 685         if (test_opt(sb, ERRORS_CONT)) {
 686                 ext4_commit_super(sb, 0);
 687                 return;
 688         }
 689
 690         ext4_unlock_group(sb, grp);
 691         ext4_commit_super(sb, 1);
 692         ext4_handle_error(sb);
 693         /*
 694          * We only get here in the ERRORS_RO case; relocking the group
 695          * may be dangerous, but nothing bad will happen since the
 696          * filesystem will have already been marked read/only and the
 697          * journal has been aborted.  We return 1 as a hint to callers
 698          * who might what to use the return value from
 699          * ext4_grp_locked_error() to distinguish between the
 700          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 701          * aggressively from the ext4 function in question, with a
 702          * more appropriate error code.
 703          */
 704         ext4_lock_group(sb, grp);
 705         return;
 706 }
 707
 708 void ext4_update_dynamic_rev(struct super_block *sb)
 709 {
 710         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 711
 712         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 713                 return;
 714
 715         ext4_warning(sb,
 716                      "updating to rev %d because of new feature flag, "
 717                      "running e2fsck is recommended",
 718                      EXT4_DYNAMIC_REV);
 719
 720         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 721         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 722         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 723         /* leave es->s_feature_*compat flags alone */
 724         /* es->s_uuid will be set by e2fsck if empty */
 725
 726         /*
 727          * The rest of the superblock fields should be zero, and if not it
 728          * means they are likely already in use, so leave them alone.  We
 729          * can leave it up to e2fsck to clean up any inconsistencies there.
 730          */
 731 }
 732
 733 /*
 734  * Open the external journal device
 735  */
 736 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 737 {
 738         struct block_device *bdev;
 739         char b[BDEVNAME_SIZE];
 740
 741         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 742         if (IS_ERR(bdev))
 743                 goto fail;
 744         return bdev;
 745
 746 fail:
 747         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 748                         __bdevname(dev, b), PTR_ERR(bdev));
 749         return NULL;
 750 }
 751
 752 /*
 753  * Release the journal device
 754  */
 755 static void ext4_blkdev_put(struct block_device *bdev)
 756 {
 757         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 758 }
 759
 760 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 761 {
 762         struct block_device *bdev;
 763         bdev = sbi->journal_bdev;
 764         if (bdev) {
 765                 ext4_blkdev_put(bdev);
 766                 sbi->journal_bdev = NULL;
 767         }
 768 }
 769
 770 static inline struct inode *orphan_list_entry(struct list_head *l)
 771 {
 772         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 773 }
 774
 775 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 776 {
 777         struct list_head *l;
 778
 779         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 780                  le32_to_cpu(sbi->s_es->s_last_orphan));
 781
 782         printk(KERN_ERR "sb_info orphan list:\n");
 783         list_for_each(l, &sbi->s_orphan) {
 784                 struct inode *inode = orphan_list_entry(l);
 785                 printk(KERN_ERR "  "
 786                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 787                        inode->i_sb->s_id, inode->i_ino, inode,
 788                        inode->i_mode, inode->i_nlink,
 789                        NEXT_ORPHAN(inode));
 790         }
 791 }
 792
 793 static void ext4_put_super(struct super_block *sb)
 794 {
 795         struct ext4_sb_info *sbi = EXT4_SB(sb);
 796         struct ext4_super_block *es = sbi->s_es;
 797         int aborted = 0;
 798         int i, err;
 799
 800         ext4_unregister_li_request(sb);
 801         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 802
 803         flush_workqueue(sbi->rsv_conversion_wq);
 804         destroy_workqueue(sbi->rsv_conversion_wq);
 805
 806         if (sbi->s_journal) {
 807                 aborted = is_journal_aborted(sbi->s_journal);
 808                 err = jbd2_journal_destroy(sbi->s_journal);
 809                 sbi->s_journal = NULL;
 810                 if ((err < 0) && !aborted)
 811                         ext4_abort(sb, "Couldn't clean up the journal");
 812         }
 813
 814         ext4_unregister_sysfs(sb);
 815         ext4_es_unregister_shrinker(sbi);
 816         del_timer_sync(&sbi->s_err_report);
 817         ext4_release_system_zone(sb);
 818         ext4_mb_release(sb);
 819         ext4_ext_release(sb);
 820
 821         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 822                 ext4_clear_feature_journal_needs_recovery(sb);
 823                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 824         }
 825         if (!(sb->s_flags & MS_RDONLY))
 826                 ext4_commit_super(sb, 1);
 827
 828         for (i = 0; i < sbi->s_gdb_count; i++)
 829                 brelse(sbi->s_group_desc[i]);
 830         kvfree(sbi->s_group_desc);
 831         kvfree(sbi->s_flex_groups);
 832         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 833         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 834         percpu_counter_destroy(&sbi->s_dirs_counter);
 835         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 836         brelse(sbi->s_sbh);
 837 #ifdef CONFIG_QUOTA
 838         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 839                 kfree(sbi->s_qf_names[i]);
 840 #endif
 841
 842         /* Debugging code just in case the in-memory inode orphan list
 843          * isn't empty.  The on-disk one can be non-empty if we've
 844          * detected an error and taken the fs readonly, but the
 845          * in-memory list had better be clean by this point. */
 846         if (!list_empty(&sbi->s_orphan))
 847                 dump_orphan_list(sb, sbi);
 848         J_ASSERT(list_empty(&sbi->s_orphan));
 849
 850         sync_blockdev(sb->s_bdev);
 851         invalidate_bdev(sb->s_bdev);
 852         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 853                 /*
 854                  * Invalidate the journal device's buffers.  We don't want them
 855                  * floating about in memory - the physical journal device may
 856                  * hotswapped, and it breaks the `ro-after' testing code.
 857                  */
 858                 sync_blockdev(sbi->journal_bdev);
 859                 invalidate_bdev(sbi->journal_bdev);
 860                 ext4_blkdev_remove(sbi);
 861         }
 862         if (sbi->s_mb_cache) {
 863                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 864                 sbi->s_mb_cache = NULL;
 865         }
 866         if (sbi->s_mmp_tsk)
 867                 kthread_stop(sbi->s_mmp_tsk);
 868         sb->s_fs_info = NULL;
 869         /*
 870          * Now that we are completely done shutting down the
 871          * superblock, we need to actually destroy the kobject.
 872          */
 873         kobject_put(&sbi->s_kobj);
 874         wait_for_completion(&sbi->s_kobj_unregister);
 875         if (sbi->s_chksum_driver)
 876                 crypto_free_shash(sbi->s_chksum_driver);
 877         kfree(sbi->s_blockgroup_lock);
 878         kfree(sbi);
 879 }
 880
 881 static struct kmem_cache *ext4_inode_cachep;
 882
 883 /*
 884  * Called inside transaction, so use GFP_NOFS
 885  */
 886 static struct inode *ext4_alloc_inode(struct super_block *sb)
 887 {
 888         struct ext4_inode_info *ei;
 889
 890         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 891         if (!ei)
 892                 return NULL;
 893
 894         ei->vfs_inode.i_version = 1;
 895         spin_lock_init(&ei->i_raw_lock);
 896         INIT_LIST_HEAD(&ei->i_prealloc_list);
 897         spin_lock_init(&ei->i_prealloc_lock);
 898         ext4_es_init_tree(&ei->i_es_tree);
 899         rwlock_init(&ei->i_es_lock);
 900         INIT_LIST_HEAD(&ei->i_es_list);
 901         ei->i_es_all_nr = 0;
 902         ei->i_es_shk_nr = 0;
 903         ei->i_es_shrink_lblk = 0;
 904         ei->i_reserved_data_blocks = 0;
 905         ei->i_reserved_meta_blocks = 0;
 906         ei->i_allocated_meta_blocks = 0;
 907         ei->i_da_metadata_calc_len = 0;
 908         ei->i_da_metadata_calc_last_lblock = 0;
 909         spin_lock_init(&(ei->i_block_reservation_lock));
 910 #ifdef CONFIG_QUOTA
 911         ei->i_reserved_quota = 0;
 912         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 913 #endif
 914         ei->jinode = NULL;
 915         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 916         spin_lock_init(&ei->i_completed_io_lock);
 917         ei->i_sync_tid = 0;
 918         ei->i_datasync_tid = 0;
 919         atomic_set(&ei->i_ioend_count, 0);
 920         atomic_set(&ei->i_unwritten, 0);
 921         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 922 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 923         ei->i_crypt_info = NULL;
 924 #endif
 925         return &ei->vfs_inode;
 926 }
 927
 928 static int ext4_drop_inode(struct inode *inode)
 929 {
 930         int drop = generic_drop_inode(inode);
 931
 932         trace_ext4_drop_inode(inode, drop);
 933         return drop;
 934 }
 935
 936 static void ext4_i_callback(struct rcu_head *head)
 937 {
 938         struct inode *inode = container_of(head, struct inode, i_rcu);
 939         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 940 }
 941
 942 static void ext4_destroy_inode(struct inode *inode)
 943 {
 944         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 945                 ext4_msg(inode->i_sb, KERN_ERR,
 946                          "Inode %lu (%p): orphan list check failed!",
 947                          inode->i_ino, EXT4_I(inode));
 948                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 949                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 950                                 true);
 951                 dump_stack();
 952         }
 953         call_rcu(&inode->i_rcu, ext4_i_callback);
 954 }
 955
 956 static void init_once(void *foo)
 957 {
 958         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 959
 960         INIT_LIST_HEAD(&ei->i_orphan);
 961         init_rwsem(&ei->xattr_sem);
 962         init_rwsem(&ei->i_data_sem);
 963         init_rwsem(&ei->i_mmap_sem);
 964         inode_init_once(&ei->vfs_inode);
 965 }
 966
 967 static int __init init_inodecache(void)
 968 {
 969         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 970                                              sizeof(struct ext4_inode_info),
 971                                              0, (SLAB_RECLAIM_ACCOUNT|
 972                                                 SLAB_MEM_SPREAD),
 973                                              init_once);
 974         if (ext4_inode_cachep == NULL)
 975                 return -ENOMEM;
 976         return 0;
 977 }
 978
 979 static void destroy_inodecache(void)
 980 {
 981         /*
 982          * Make sure all delayed rcu free inodes are flushed before we
 983          * destroy cache.
 984          */
 985         rcu_barrier();
 986         kmem_cache_destroy(ext4_inode_cachep);
 987 }
 988
 989 void ext4_clear_inode(struct inode *inode)
 990 {
 991         invalidate_inode_buffers(inode);
 992         clear_inode(inode);
 993         dquot_drop(inode);
 994         ext4_discard_preallocations(inode);
 995         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 996         if (EXT4_I(inode)->jinode) {
 997                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 998                                                EXT4_I(inode)->jinode);
 999                 jbd2_free_inode(EXT4_I(inode)->jinode);
1000                 EXT4_I(inode)->jinode = NULL;
1001         }
1002 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1003         if (EXT4_I(inode)->i_crypt_info)
1004                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1005 #endif
1006 }
1007
1008 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1009                                         u64 ino, u32 generation)
1010 {
1011         struct inode *inode;
1012
1013         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1014                 return ERR_PTR(-ESTALE);
1015         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1016                 return ERR_PTR(-ESTALE);
1017
1018         /* iget isn't really right if the inode is currently unallocated!!
1019          *
1020          * ext4_read_inode will return a bad_inode if the inode had been
1021          * deleted, so we should be safe.
1022          *
1023          * Currently we don't know the generation for parent directory, so
1024          * a generation of 0 means "accept any"
1025          */
1026         inode = ext4_iget_normal(sb, ino);
1027         if (IS_ERR(inode))
1028                 return ERR_CAST(inode);
1029         if (generation && inode->i_generation != generation) {
1030                 iput(inode);
1031                 return ERR_PTR(-ESTALE);
1032         }
1033
1034         return inode;
1035 }
1036
1037 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1038                                         int fh_len, int fh_type)
1039 {
1040         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1041                                     ext4_nfs_get_inode);
1042 }
1043
1044 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1045                                         int fh_len, int fh_type)
1046 {
1047         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1048                                     ext4_nfs_get_inode);
1049 }
1050
1051 /*
1052  * Try to release metadata pages (indirect blocks, directories) which are
1053  * mapped via the block device.  Since these pages could have journal heads
1054  * which would prevent try_to_free_buffers() from freeing them, we must use
1055  * jbd2 layer's try_to_free_buffers() function to release them.
1056  */
1057 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1058                                  gfp_t wait)
1059 {
1060         journal_t *journal = EXT4_SB(sb)->s_journal;
1061
1062         WARN_ON(PageChecked(page));
1063         if (!page_has_buffers(page))
1064                 return 0;
1065         if (journal)
1066                 return jbd2_journal_try_to_free_buffers(journal, page,
1067                                                 wait & ~__GFP_DIRECT_RECLAIM);
1068         return try_to_free_buffers(page);
1069 }
1070
1071 #ifdef CONFIG_QUOTA
1072 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1073 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1074
1075 static int ext4_write_dquot(struct dquot *dquot);
1076 static int ext4_acquire_dquot(struct dquot *dquot);
1077 static int ext4_release_dquot(struct dquot *dquot);
1078 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1079 static int ext4_write_info(struct super_block *sb, int type);
1080 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1081                          struct path *path);
1082 static int ext4_quota_off(struct super_block *sb, int type);
1083 static int ext4_quota_on_mount(struct super_block *sb, int type);
1084 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1085                                size_t len, loff_t off);
1086 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1087                                 const char *data, size_t len, loff_t off);
1088 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1089                              unsigned int flags);
1090 static int ext4_enable_quotas(struct super_block *sb);
1091
1092 static struct dquot **ext4_get_dquots(struct inode *inode)
1093 {
1094         return EXT4_I(inode)->i_dquot;
1095 }
1096
1097 static const struct dquot_operations ext4_quota_operations = {
1098         .get_reserved_space = ext4_get_reserved_space,
1099         .write_dquot    = ext4_write_dquot,
1100         .acquire_dquot  = ext4_acquire_dquot,
1101         .release_dquot  = ext4_release_dquot,
1102         .mark_dirty     = ext4_mark_dquot_dirty,
1103         .write_info     = ext4_write_info,
1104         .alloc_dquot    = dquot_alloc,
1105         .destroy_dquot  = dquot_destroy,
1106 };
1107
1108 static const struct quotactl_ops ext4_qctl_operations = {
1109         .quota_on       = ext4_quota_on,
1110         .quota_off      = ext4_quota_off,
1111         .quota_sync     = dquot_quota_sync,
1112         .get_state      = dquot_get_state,
1113         .set_info       = dquot_set_dqinfo,
1114         .get_dqblk      = dquot_get_dqblk,
1115         .set_dqblk      = dquot_set_dqblk
1116 };
1117 #endif
1118
1119 static const struct super_operations ext4_sops = {
1120         .alloc_inode    = ext4_alloc_inode,
1121         .destroy_inode  = ext4_destroy_inode,
1122         .write_inode    = ext4_write_inode,
1123         .dirty_inode    = ext4_dirty_inode,
1124         .drop_inode     = ext4_drop_inode,
1125         .evict_inode    = ext4_evict_inode,
1126         .put_super      = ext4_put_super,
1127         .sync_fs        = ext4_sync_fs,
1128         .freeze_fs      = ext4_freeze,
1129         .unfreeze_fs    = ext4_unfreeze,
1130         .statfs         = ext4_statfs,
1131         .remount_fs     = ext4_remount,
1132         .show_options   = ext4_show_options,
1133 #ifdef CONFIG_QUOTA
1134         .quota_read     = ext4_quota_read,
1135         .quota_write    = ext4_quota_write,
1136         .get_dquots     = ext4_get_dquots,
1137 #endif
1138         .bdev_try_to_free_page = bdev_try_to_free_page,
1139 };
1140
1141 static const struct export_operations ext4_export_ops = {
1142         .fh_to_dentry = ext4_fh_to_dentry,
1143         .fh_to_parent = ext4_fh_to_parent,
1144         .get_parent = ext4_get_parent,
1145 };
1146
1147 enum {
1148         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1149         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1150         Opt_nouid32, Opt_debug, Opt_removed,
1151         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1152         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1153         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1154         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1155         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1156         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1157         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1158         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1159         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1160         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1161         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1162         Opt_lazytime, Opt_nolazytime,
1163         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1164         Opt_inode_readahead_blks, Opt_journal_ioprio,
1165         Opt_dioread_nolock, Opt_dioread_lock,
1166         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1167         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1168 };
1169
1170 static const match_table_t tokens = {
1171         {Opt_bsd_df, "bsddf"},
1172         {Opt_minix_df, "minixdf"},
1173         {Opt_grpid, "grpid"},
1174         {Opt_grpid, "bsdgroups"},
1175         {Opt_nogrpid, "nogrpid"},
1176         {Opt_nogrpid, "sysvgroups"},
1177         {Opt_resgid, "resgid=%u"},
1178         {Opt_resuid, "resuid=%u"},
1179         {Opt_sb, "sb=%u"},
1180         {Opt_err_cont, "errors=continue"},
1181         {Opt_err_panic, "errors=panic"},
1182         {Opt_err_ro, "errors=remount-ro"},
1183         {Opt_nouid32, "nouid32"},
1184         {Opt_debug, "debug"},
1185         {Opt_removed, "oldalloc"},
1186         {Opt_removed, "orlov"},
1187         {Opt_user_xattr, "user_xattr"},
1188         {Opt_nouser_xattr, "nouser_xattr"},
1189         {Opt_acl, "acl"},
1190         {Opt_noacl, "noacl"},
1191         {Opt_noload, "norecovery"},
1192         {Opt_noload, "noload"},
1193         {Opt_removed, "nobh"},
1194         {Opt_removed, "bh"},
1195         {Opt_commit, "commit=%u"},
1196         {Opt_min_batch_time, "min_batch_time=%u"},
1197         {Opt_max_batch_time, "max_batch_time=%u"},
1198         {Opt_journal_dev, "journal_dev=%u"},
1199         {Opt_journal_path, "journal_path=%s"},
1200         {Opt_journal_checksum, "journal_checksum"},
1201         {Opt_nojournal_checksum, "nojournal_checksum"},
1202         {Opt_journal_async_commit, "journal_async_commit"},
1203         {Opt_abort, "abort"},
1204         {Opt_data_journal, "data=journal"},
1205         {Opt_data_ordered, "data=ordered"},
1206         {Opt_data_writeback, "data=writeback"},
1207         {Opt_data_err_abort, "data_err=abort"},
1208         {Opt_data_err_ignore, "data_err=ignore"},
1209         {Opt_offusrjquota, "usrjquota="},
1210         {Opt_usrjquota, "usrjquota=%s"},
1211         {Opt_offgrpjquota, "grpjquota="},
1212         {Opt_grpjquota, "grpjquota=%s"},
1213         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1214         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1215         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1216         {Opt_grpquota, "grpquota"},
1217         {Opt_noquota, "noquota"},
1218         {Opt_quota, "quota"},
1219         {Opt_usrquota, "usrquota"},
1220         {Opt_barrier, "barrier=%u"},
1221         {Opt_barrier, "barrier"},
1222         {Opt_nobarrier, "nobarrier"},
1223         {Opt_i_version, "i_version"},
1224         {Opt_dax, "dax"},
1225         {Opt_stripe, "stripe=%u"},
1226         {Opt_delalloc, "delalloc"},
1227         {Opt_lazytime, "lazytime"},
1228         {Opt_nolazytime, "nolazytime"},
1229         {Opt_nodelalloc, "nodelalloc"},
1230         {Opt_removed, "mblk_io_submit"},
1231         {Opt_removed, "nomblk_io_submit"},
1232         {Opt_block_validity, "block_validity"},
1233         {Opt_noblock_validity, "noblock_validity"},
1234         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1235         {Opt_journal_ioprio, "journal_ioprio=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1237         {Opt_auto_da_alloc, "auto_da_alloc"},
1238         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1239         {Opt_dioread_nolock, "dioread_nolock"},
1240         {Opt_dioread_lock, "dioread_lock"},
1241         {Opt_discard, "discard"},
1242         {Opt_nodiscard, "nodiscard"},
1243         {Opt_init_itable, "init_itable=%u"},
1244         {Opt_init_itable, "init_itable"},
1245         {Opt_noinit_itable, "noinit_itable"},
1246         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1247         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1248         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1249         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1250         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1251         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1252         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1253         {Opt_err, NULL},
1254 };
1255
1256 static ext4_fsblk_t get_sb_block(void **data)
1257 {
1258         ext4_fsblk_t    sb_block;
1259         char            *options = (char *) *data;
1260
1261         if (!options || strncmp(options, "sb=", 3) != 0)
1262                 return 1;       /* Default location */
1263
1264         options += 3;
1265         /* TODO: use simple_strtoll with >32bit ext4 */
1266         sb_block = simple_strtoul(options, &options, 0);
1267         if (*options && *options != ',') {
1268                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1269                        (char *) *data);
1270                 return 1;
1271         }
1272         if (*options == ',')
1273                 options++;
1274         *data = (void *) options;
1275
1276         return sb_block;
1277 }
1278
1279 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1280 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1281         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1282
1283 #ifdef CONFIG_QUOTA
1284 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1285 {
1286         struct ext4_sb_info *sbi = EXT4_SB(sb);
1287         char *qname;
1288         int ret = -1;
1289
1290         if (sb_any_quota_loaded(sb) &&
1291                 !sbi->s_qf_names[qtype]) {
1292                 ext4_msg(sb, KERN_ERR,
1293                         "Cannot change journaled "
1294                         "quota options when quota turned on");
1295                 return -1;
1296         }
1297         if (ext4_has_feature_quota(sb)) {
1298                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1299                          "ignored when QUOTA feature is enabled");
1300                 return 1;
1301         }
1302         qname = match_strdup(args);
1303         if (!qname) {
1304                 ext4_msg(sb, KERN_ERR,
1305                         "Not enough memory for storing quotafile name");
1306                 return -1;
1307         }
1308         if (sbi->s_qf_names[qtype]) {
1309                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1310                         ret = 1;
1311                 else
1312                         ext4_msg(sb, KERN_ERR,
1313                                  "%s quota file already specified",
1314                                  QTYPE2NAME(qtype));
1315                 goto errout;
1316         }
1317         if (strchr(qname, '/')) {
1318                 ext4_msg(sb, KERN_ERR,
1319                         "quotafile must be on filesystem root");
1320                 goto errout;
1321         }
1322         sbi->s_qf_names[qtype] = qname;
1323         set_opt(sb, QUOTA);
1324         return 1;
1325 errout:
1326         kfree(qname);
1327         return ret;
1328 }
1329
1330 static int clear_qf_name(struct super_block *sb, int qtype)
1331 {
1332
1333         struct ext4_sb_info *sbi = EXT4_SB(sb);
1334
1335         if (sb_any_quota_loaded(sb) &&
1336                 sbi->s_qf_names[qtype]) {
1337                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1338                         " when quota turned on");
1339                 return -1;
1340         }
1341         kfree(sbi->s_qf_names[qtype]);
1342         sbi->s_qf_names[qtype] = NULL;
1343         return 1;
1344 }
1345 #endif
1346
1347 #define MOPT_SET        0x0001
1348 #define MOPT_CLEAR      0x0002
1349 #define MOPT_NOSUPPORT  0x0004
1350 #define MOPT_EXPLICIT   0x0008
1351 #define MOPT_CLEAR_ERR  0x0010
1352 #define MOPT_GTE0       0x0020
1353 #ifdef CONFIG_QUOTA
1354 #define MOPT_Q          0
1355 #define MOPT_QFMT       0x0040
1356 #else
1357 #define MOPT_Q          MOPT_NOSUPPORT
1358 #define MOPT_QFMT       MOPT_NOSUPPORT
1359 #endif
1360 #define MOPT_DATAJ      0x0080
1361 #define MOPT_NO_EXT2    0x0100
1362 #define MOPT_NO_EXT3    0x0200
1363 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1364 #define MOPT_STRING     0x0400
1365
1366 static const struct mount_opts {
1367         int     token;
1368         int     mount_opt;
1369         int     flags;
1370 } ext4_mount_opts[] = {
1371         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1372         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1373         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1374         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1375         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1376         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1377         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1378          MOPT_EXT4_ONLY | MOPT_SET},
1379         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1380          MOPT_EXT4_ONLY | MOPT_CLEAR},
1381         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1382         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1383         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1384          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1385         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1386          MOPT_EXT4_ONLY | MOPT_CLEAR},
1387         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1388          MOPT_EXT4_ONLY | MOPT_CLEAR},
1389         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1390          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1391         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1392                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1393          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1394         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1395         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1397         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1398         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1399          MOPT_NO_EXT2 | MOPT_SET},
1400         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1401          MOPT_NO_EXT2 | MOPT_CLEAR},
1402         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1403         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1404         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1405         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1406         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1407         {Opt_commit, 0, MOPT_GTE0},
1408         {Opt_max_batch_time, 0, MOPT_GTE0},
1409         {Opt_min_batch_time, 0, MOPT_GTE0},
1410         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1411         {Opt_init_itable, 0, MOPT_GTE0},
1412         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1413         {Opt_stripe, 0, MOPT_GTE0},
1414         {Opt_resuid, 0, MOPT_GTE0},
1415         {Opt_resgid, 0, MOPT_GTE0},
1416         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1417         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1418         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1419         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1420         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1421         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1422          MOPT_NO_EXT2 | MOPT_DATAJ},
1423         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1424         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1425 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1426         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1427         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1428 #else
1429         {Opt_acl, 0, MOPT_NOSUPPORT},
1430         {Opt_noacl, 0, MOPT_NOSUPPORT},
1431 #endif
1432         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1433         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1434         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1435         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1436                                                         MOPT_SET | MOPT_Q},
1437         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1438                                                         MOPT_SET | MOPT_Q},
1439         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1440                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1441         {Opt_usrjquota, 0, MOPT_Q},
1442         {Opt_grpjquota, 0, MOPT_Q},
1443         {Opt_offusrjquota, 0, MOPT_Q},
1444         {Opt_offgrpjquota, 0, MOPT_Q},
1445         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1446         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1447         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1448         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1449         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1450         {Opt_err, 0, 0}
1451 };
1452
1453 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1454                             substring_t *args, unsigned long *journal_devnum,
1455                             unsigned int *journal_ioprio, int is_remount)
1456 {
1457         struct ext4_sb_info *sbi = EXT4_SB(sb);
1458         const struct mount_opts *m;
1459         kuid_t uid;
1460         kgid_t gid;
1461         int arg = 0;
1462
1463 #ifdef CONFIG_QUOTA
1464         if (token == Opt_usrjquota)
1465                 return set_qf_name(sb, USRQUOTA, &args[0]);
1466         else if (token == Opt_grpjquota)
1467                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1468         else if (token == Opt_offusrjquota)
1469                 return clear_qf_name(sb, USRQUOTA);
1470         else if (token == Opt_offgrpjquota)
1471                 return clear_qf_name(sb, GRPQUOTA);
1472 #endif
1473         switch (token) {
1474         case Opt_noacl:
1475         case Opt_nouser_xattr:
1476                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1477                 break;
1478         case Opt_sb:
1479                 return 1;       /* handled by get_sb_block() */
1480         case Opt_removed:
1481                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1482                 return 1;
1483         case Opt_abort:
1484                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1485                 return 1;
1486         case Opt_i_version:
1487                 sb->s_flags |= MS_I_VERSION;
1488                 return 1;
1489         case Opt_lazytime:
1490                 sb->s_flags |= MS_LAZYTIME;
1491                 return 1;
1492         case Opt_nolazytime:
1493                 sb->s_flags &= ~MS_LAZYTIME;
1494                 return 1;
1495         }
1496
1497         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1498                 if (token == m->token)
1499                         break;
1500
1501         if (m->token == Opt_err) {
1502                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1503                          "or missing value", opt);
1504                 return -1;
1505         }
1506
1507         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1508                 ext4_msg(sb, KERN_ERR,
1509                          "Mount option \"%s\" incompatible with ext2", opt);
1510                 return -1;
1511         }
1512         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1513                 ext4_msg(sb, KERN_ERR,
1514                          "Mount option \"%s\" incompatible with ext3", opt);
1515                 return -1;
1516         }
1517
1518         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1519                 return -1;
1520         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1521                 return -1;
1522         if (m->flags & MOPT_EXPLICIT) {
1523                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1524                         set_opt2(sb, EXPLICIT_DELALLOC);
1525                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1526                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1527                 } else
1528                         return -1;
1529         }
1530         if (m->flags & MOPT_CLEAR_ERR)
1531                 clear_opt(sb, ERRORS_MASK);
1532         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1533                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1534                          "options when quota turned on");
1535                 return -1;
1536         }
1537
1538         if (m->flags & MOPT_NOSUPPORT) {
1539                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1540         } else if (token == Opt_commit) {
1541                 if (arg == 0)
1542                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1543                 sbi->s_commit_interval = HZ * arg;
1544         } else if (token == Opt_max_batch_time) {
1545                 sbi->s_max_batch_time = arg;
1546         } else if (token == Opt_min_batch_time) {
1547                 sbi->s_min_batch_time = arg;
1548         } else if (token == Opt_inode_readahead_blks) {
1549                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1550                         ext4_msg(sb, KERN_ERR,
1551                                  "EXT4-fs: inode_readahead_blks must be "
1552                                  "0 or a power of 2 smaller than 2^31");
1553                         return -1;
1554                 }
1555                 sbi->s_inode_readahead_blks = arg;
1556         } else if (token == Opt_init_itable) {
1557                 set_opt(sb, INIT_INODE_TABLE);
1558                 if (!args->from)
1559                         arg = EXT4_DEF_LI_WAIT_MULT;
1560                 sbi->s_li_wait_mult = arg;
1561         } else if (token == Opt_max_dir_size_kb) {
1562                 sbi->s_max_dir_size_kb = arg;
1563         } else if (token == Opt_stripe) {
1564                 sbi->s_stripe = arg;
1565         } else if (token == Opt_resuid) {
1566                 uid = make_kuid(current_user_ns(), arg);
1567                 if (!uid_valid(uid)) {
1568                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1569                         return -1;
1570                 }
1571                 sbi->s_resuid = uid;
1572         } else if (token == Opt_resgid) {
1573                 gid = make_kgid(current_user_ns(), arg);
1574                 if (!gid_valid(gid)) {
1575                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1576                         return -1;
1577                 }
1578                 sbi->s_resgid = gid;
1579         } else if (token == Opt_journal_dev) {
1580                 if (is_remount) {
1581                         ext4_msg(sb, KERN_ERR,
1582                                  "Cannot specify journal on remount");
1583                         return -1;
1584                 }
1585                 *journal_devnum = arg;
1586         } else if (token == Opt_journal_path) {
1587                 char *journal_path;
1588                 struct inode *journal_inode;
1589                 struct path path;
1590                 int error;
1591
1592                 if (is_remount) {
1593                         ext4_msg(sb, KERN_ERR,
1594                                  "Cannot specify journal on remount");
1595                         return -1;
1596                 }
1597                 journal_path = match_strdup(&args[0]);
1598                 if (!journal_path) {
1599                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1600                                 "journal device string");
1601                         return -1;
1602                 }
1603
1604                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1605                 if (error) {
1606                         ext4_msg(sb, KERN_ERR, "error: could not find "
1607                                 "journal device path: error %d", error);
1608                         kfree(journal_path);
1609                         return -1;
1610                 }
1611
1612                 journal_inode = d_inode(path.dentry);
1613                 if (!S_ISBLK(journal_inode->i_mode)) {
1614                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1615                                 "is not a block device", journal_path);
1616                         path_put(&path);
1617                         kfree(journal_path);
1618                         return -1;
1619                 }
1620
1621                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1622                 path_put(&path);
1623                 kfree(journal_path);
1624         } else if (token == Opt_journal_ioprio) {
1625                 if (arg > 7) {
1626                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1627                                  " (must be 0-7)");
1628                         return -1;
1629                 }
1630                 *journal_ioprio =
1631                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1632         } else if (token == Opt_test_dummy_encryption) {
1633 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1634                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1635                 ext4_msg(sb, KERN_WARNING,
1636                          "Test dummy encryption mode enabled");
1637 #else
1638                 ext4_msg(sb, KERN_WARNING,
1639                          "Test dummy encryption mount option ignored");
1640 #endif
1641         } else if (m->flags & MOPT_DATAJ) {
1642                 if (is_remount) {
1643                         if (!sbi->s_journal)
1644                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1645                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1646                                 ext4_msg(sb, KERN_ERR,
1647                                          "Cannot change data mode on remount");
1648                                 return -1;
1649                         }
1650                 } else {
1651                         clear_opt(sb, DATA_FLAGS);
1652                         sbi->s_mount_opt |= m->mount_opt;
1653                 }
1654 #ifdef CONFIG_QUOTA
1655         } else if (m->flags & MOPT_QFMT) {
1656                 if (sb_any_quota_loaded(sb) &&
1657                     sbi->s_jquota_fmt != m->mount_opt) {
1658                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1659                                  "quota options when quota turned on");
1660                         return -1;
1661                 }
1662                 if (ext4_has_feature_quota(sb)) {
1663                         ext4_msg(sb, KERN_INFO,
1664                                  "Quota format mount options ignored "
1665                                  "when QUOTA feature is enabled");
1666                         return 1;
1667                 }
1668                 sbi->s_jquota_fmt = m->mount_opt;
1669 #endif
1670         } else if (token == Opt_dax) {
1671 #ifdef CONFIG_FS_DAX
1672                 ext4_msg(sb, KERN_WARNING,
1673                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1674                         sbi->s_mount_opt |= m->mount_opt;
1675 #else
1676                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1677                 return -1;
1678 #endif
1679         } else {
1680                 if (!args->from)
1681                         arg = 1;
1682                 if (m->flags & MOPT_CLEAR)
1683                         arg = !arg;
1684                 else if (unlikely(!(m->flags & MOPT_SET))) {
1685                         ext4_msg(sb, KERN_WARNING,
1686                                  "buggy handling of option %s", opt);
1687                         WARN_ON(1);
1688                         return -1;
1689                 }
1690                 if (arg != 0)
1691                         sbi->s_mount_opt |= m->mount_opt;
1692                 else
1693                         sbi->s_mount_opt &= ~m->mount_opt;
1694         }
1695         return 1;
1696 }
1697
1698 static int parse_options(char *options, struct super_block *sb,
1699                          unsigned long *journal_devnum,
1700                          unsigned int *journal_ioprio,
1701                          int is_remount)
1702 {
1703         struct ext4_sb_info *sbi = EXT4_SB(sb);
1704         char *p;
1705         substring_t args[MAX_OPT_ARGS];
1706         int token;
1707
1708         if (!options)
1709                 return 1;
1710
1711         while ((p = strsep(&options, ",")) != NULL) {
1712                 if (!*p)
1713                         continue;
1714                 /*
1715                  * Initialize args struct so we know whether arg was
1716                  * found; some options take optional arguments.
1717                  */
1718                 args[0].to = args[0].from = NULL;
1719                 token = match_token(p, tokens, args);
1720                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1721                                      journal_ioprio, is_remount) < 0)
1722                         return 0;
1723         }
1724 #ifdef CONFIG_QUOTA
1725         if (ext4_has_feature_quota(sb) &&
1726             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1727                 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1728                          "mount options ignored.");
1729                 clear_opt(sb, USRQUOTA);
1730                 clear_opt(sb, GRPQUOTA);
1731         } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1732                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1733                         clear_opt(sb, USRQUOTA);
1734
1735                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1736                         clear_opt(sb, GRPQUOTA);
1737
1738                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1739                         ext4_msg(sb, KERN_ERR, "old and new quota "
1740                                         "format mixing");
1741                         return 0;
1742                 }
1743
1744                 if (!sbi->s_jquota_fmt) {
1745                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1746                                         "not specified");
1747                         return 0;
1748                 }
1749         }
1750 #endif
1751         if (test_opt(sb, DIOREAD_NOLOCK)) {
1752                 int blocksize =
1753                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1754
1755                 if (blocksize < PAGE_CACHE_SIZE) {
1756                         ext4_msg(sb, KERN_ERR, "can't mount with "
1757                                  "dioread_nolock if block size != PAGE_SIZE");
1758                         return 0;
1759                 }
1760         }
1761         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1762             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1763                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1764                          "in data=ordered mode");
1765                 return 0;
1766         }
1767         return 1;
1768 }
1769
1770 static inline void ext4_show_quota_options(struct seq_file *seq,
1771                                            struct super_block *sb)
1772 {
1773 #if defined(CONFIG_QUOTA)
1774         struct ext4_sb_info *sbi = EXT4_SB(sb);
1775
1776         if (sbi->s_jquota_fmt) {
1777                 char *fmtname = "";
1778
1779                 switch (sbi->s_jquota_fmt) {
1780                 case QFMT_VFS_OLD:
1781                         fmtname = "vfsold";
1782                         break;
1783                 case QFMT_VFS_V0:
1784                         fmtname = "vfsv0";
1785                         break;
1786                 case QFMT_VFS_V1:
1787                         fmtname = "vfsv1";
1788                         break;
1789                 }
1790                 seq_printf(seq, ",jqfmt=%s", fmtname);
1791         }
1792
1793         if (sbi->s_qf_names[USRQUOTA])
1794                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1795
1796         if (sbi->s_qf_names[GRPQUOTA])
1797                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1798 #endif
1799 }
1800
1801 static const char *token2str(int token)
1802 {
1803         const struct match_token *t;
1804
1805         for (t = tokens; t->token != Opt_err; t++)
1806                 if (t->token == token && !strchr(t->pattern, '='))
1807                         break;
1808         return t->pattern;
1809 }
1810
1811 /*
1812  * Show an option if
1813  *  - it's set to a non-default value OR
1814  *  - if the per-sb default is different from the global default
1815  */
1816 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1817                               int nodefs)
1818 {
1819         struct ext4_sb_info *sbi = EXT4_SB(sb);
1820         struct ext4_super_block *es = sbi->s_es;
1821         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1822         const struct mount_opts *m;
1823         char sep = nodefs ? '\n' : ',';
1824
1825 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1826 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1827
1828         if (sbi->s_sb_block != 1)
1829                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1830
1831         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1832                 int want_set = m->flags & MOPT_SET;
1833                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1834                     (m->flags & MOPT_CLEAR_ERR))
1835                         continue;
1836                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1837                         continue; /* skip if same as the default */
1838                 if ((want_set &&
1839                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1840                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1841                         continue; /* select Opt_noFoo vs Opt_Foo */
1842                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1843         }
1844
1845         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1846             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1847                 SEQ_OPTS_PRINT("resuid=%u",
1848                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1849         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1850             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1851                 SEQ_OPTS_PRINT("resgid=%u",
1852                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1853         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1854         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1855                 SEQ_OPTS_PUTS("errors=remount-ro");
1856         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1857                 SEQ_OPTS_PUTS("errors=continue");
1858         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1859                 SEQ_OPTS_PUTS("errors=panic");
1860         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1861                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1862         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1863                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1864         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1865                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1866         if (sb->s_flags & MS_I_VERSION)
1867                 SEQ_OPTS_PUTS("i_version");
1868         if (nodefs || sbi->s_stripe)
1869                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1870         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1871                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1872                         SEQ_OPTS_PUTS("data=journal");
1873                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1874                         SEQ_OPTS_PUTS("data=ordered");
1875                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1876                         SEQ_OPTS_PUTS("data=writeback");
1877         }
1878         if (nodefs ||
1879             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1880                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1881                                sbi->s_inode_readahead_blks);
1882
1883         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1884                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1885                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1886         if (nodefs || sbi->s_max_dir_size_kb)
1887                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1888
1889         ext4_show_quota_options(seq, sb);
1890         return 0;
1891 }
1892
1893 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1894 {
1895         return _ext4_show_options(seq, root->d_sb, 0);
1896 }
1897
1898 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1899 {
1900         struct super_block *sb = seq->private;
1901         int rc;
1902
1903         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1904         rc = _ext4_show_options(seq, sb, 1);
1905         seq_puts(seq, "\n");
1906         return rc;
1907 }
1908
1909 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1910                             int read_only)
1911 {
1912         struct ext4_sb_info *sbi = EXT4_SB(sb);
1913         int res = 0;
1914
1915         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1916                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1917                          "forcing read-only mode");
1918                 res = MS_RDONLY;
1919         }
1920         if (read_only)
1921                 goto done;
1922         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1923                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1924                          "running e2fsck is recommended");
1925         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1926                 ext4_msg(sb, KERN_WARNING,
1927                          "warning: mounting fs with errors, "
1928                          "running e2fsck is recommended");
1929         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1930                  le16_to_cpu(es->s_mnt_count) >=
1931                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1932                 ext4_msg(sb, KERN_WARNING,
1933                          "warning: maximal mount count reached, "
1934                          "running e2fsck is recommended");
1935         else if (le32_to_cpu(es->s_checkinterval) &&
1936                 (le32_to_cpu(es->s_lastcheck) +
1937                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1938                 ext4_msg(sb, KERN_WARNING,
1939                          "warning: checktime reached, "
1940                          "running e2fsck is recommended");
1941         if (!sbi->s_journal)
1942                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1943         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1944                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1945         le16_add_cpu(&es->s_mnt_count, 1);
1946         es->s_mtime = cpu_to_le32(get_seconds());
1947         ext4_update_dynamic_rev(sb);
1948         if (sbi->s_journal)
1949                 ext4_set_feature_journal_needs_recovery(sb);
1950
1951         ext4_commit_super(sb, 1);
1952 done:
1953         if (test_opt(sb, DEBUG))
1954                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1955                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1956                         sb->s_blocksize,
1957                         sbi->s_groups_count,
1958                         EXT4_BLOCKS_PER_GROUP(sb),
1959                         EXT4_INODES_PER_GROUP(sb),
1960                         sbi->s_mount_opt, sbi->s_mount_opt2);
1961
1962         cleancache_init_fs(sb);
1963         return res;
1964 }
1965
1966 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1967 {
1968         struct ext4_sb_info *sbi = EXT4_SB(sb);
1969         struct flex_groups *new_groups;
1970         int size;
1971
1972         if (!sbi->s_log_groups_per_flex)
1973                 return 0;
1974
1975         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1976         if (size <= sbi->s_flex_groups_allocated)
1977                 return 0;
1978
1979         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1980         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1981         if (!new_groups) {
1982                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1983                          size / (int) sizeof(struct flex_groups));
1984                 return -ENOMEM;
1985         }
1986
1987         if (sbi->s_flex_groups) {
1988                 memcpy(new_groups, sbi->s_flex_groups,
1989                        (sbi->s_flex_groups_allocated *
1990                         sizeof(struct flex_groups)));
1991                 kvfree(sbi->s_flex_groups);
1992         }
1993         sbi->s_flex_groups = new_groups;
1994         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1995         return 0;
1996 }
1997
1998 static int ext4_fill_flex_info(struct super_block *sb)
1999 {
2000         struct ext4_sb_info *sbi = EXT4_SB(sb);
2001         struct ext4_group_desc *gdp = NULL;
2002         ext4_group_t flex_group;
2003         int i, err;
2004
2005         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2006         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2007                 sbi->s_log_groups_per_flex = 0;
2008                 return 1;
2009         }
2010
2011         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2012         if (err)
2013                 goto failed;
2014
2015         for (i = 0; i < sbi->s_groups_count; i++) {
2016                 gdp = ext4_get_group_desc(sb, i, NULL);
2017
2018                 flex_group = ext4_flex_group(sbi, i);
2019                 atomic_add(ext4_free_inodes_count(sb, gdp),
2020                            &sbi->s_flex_groups[flex_group].free_inodes);
2021                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2022                              &sbi->s_flex_groups[flex_group].free_clusters);
2023                 atomic_add(ext4_used_dirs_count(sb, gdp),
2024                            &sbi->s_flex_groups[flex_group].used_dirs);
2025         }
2026
2027         return 1;
2028 failed:
2029         return 0;
2030 }
2031
2032 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2033                                    struct ext4_group_desc *gdp)
2034 {
2035         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2036         __u16 crc = 0;
2037         __le32 le_group = cpu_to_le32(block_group);
2038         struct ext4_sb_info *sbi = EXT4_SB(sb);
2039
2040         if (ext4_has_metadata_csum(sbi->s_sb)) {
2041                 /* Use new metadata_csum algorithm */
2042                 __u32 csum32;
2043                 __u16 dummy_csum = 0;
2044
2045                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2046                                      sizeof(le_group));
2047                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2048                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2049                                      sizeof(dummy_csum));
2050                 offset += sizeof(dummy_csum);
2051                 if (offset < sbi->s_desc_size)
2052                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2053                                              sbi->s_desc_size - offset);
2054
2055                 crc = csum32 & 0xFFFF;
2056                 goto out;
2057         }
2058
2059         /* old crc16 code */
2060         if (!ext4_has_feature_gdt_csum(sb))
2061                 return 0;
2062
2063         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2064         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2065         crc = crc16(crc, (__u8 *)gdp, offset);
2066         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2067         /* for checksum of struct ext4_group_desc do the rest...*/
2068         if (ext4_has_feature_64bit(sb) &&
2069             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2070                 crc = crc16(crc, (__u8 *)gdp + offset,
2071                             le16_to_cpu(sbi->s_es->s_desc_size) -
2072                                 offset);
2073
2074 out:
2075         return cpu_to_le16(crc);
2076 }
2077
2078 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2079                                 struct ext4_group_desc *gdp)
2080 {
2081         if (ext4_has_group_desc_csum(sb) &&
2082             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2083                 return 0;
2084
2085         return 1;
2086 }
2087
2088 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2089                               struct ext4_group_desc *gdp)
2090 {
2091         if (!ext4_has_group_desc_csum(sb))
2092                 return;
2093         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2094 }
2095
2096 /* Called at mount-time, super-block is locked */
2097 static int ext4_check_descriptors(struct super_block *sb,
2098                                   ext4_fsblk_t sb_block,
2099                                   ext4_group_t *first_not_zeroed)
2100 {
2101         struct ext4_sb_info *sbi = EXT4_SB(sb);
2102         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2103         ext4_fsblk_t last_block;
2104         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2105         ext4_fsblk_t block_bitmap;
2106         ext4_fsblk_t inode_bitmap;
2107         ext4_fsblk_t inode_table;
2108         int flexbg_flag = 0;
2109         ext4_group_t i, grp = sbi->s_groups_count;
2110
2111         if (ext4_has_feature_flex_bg(sb))
2112                 flexbg_flag = 1;
2113
2114         ext4_debug("Checking group descriptors");
2115
2116         for (i = 0; i < sbi->s_groups_count; i++) {
2117                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2118
2119                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2120                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2121                 else
2122                         last_block = first_block +
2123                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2124
2125                 if ((grp == sbi->s_groups_count) &&
2126                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2127                         grp = i;
2128
2129                 block_bitmap = ext4_block_bitmap(sb, gdp);
2130                 if (block_bitmap == sb_block) {
2131                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2132                                  "Block bitmap for group %u overlaps "
2133                                  "superblock", i);
2134                         if (!(sb->s_flags & MS_RDONLY))
2135                                 return 0;
2136                 }
2137                 if (block_bitmap >= sb_block + 1 &&
2138                     block_bitmap <= last_bg_block) {
2139                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2140                                  "Block bitmap for group %u overlaps "
2141                                  "block group descriptors", i);
2142                         if (!(sb->s_flags & MS_RDONLY))
2143                                 return 0;
2144                 }
2145                 if (block_bitmap < first_block || block_bitmap > last_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                "Block bitmap for group %u not in group "
2148                                "(block %llu)!", i, block_bitmap);
2149                         return 0;
2150                 }
2151                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2152                 if (inode_bitmap == sb_block) {
2153                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2154                                  "Inode bitmap for group %u overlaps "
2155                                  "superblock", i);
2156                         if (!(sb->s_flags & MS_RDONLY))
2157                                 return 0;
2158                 }
2159                 if (inode_bitmap >= sb_block + 1 &&
2160                     inode_bitmap <= last_bg_block) {
2161                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2162                                  "Inode bitmap for group %u overlaps "
2163                                  "block group descriptors", i);
2164                         if (!(sb->s_flags & MS_RDONLY))
2165                                 return 0;
2166                 }
2167                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2168                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2169                                "Inode bitmap for group %u not in group "
2170                                "(block %llu)!", i, inode_bitmap);
2171                         return 0;
2172                 }
2173                 inode_table = ext4_inode_table(sb, gdp);
2174                 if (inode_table == sb_block) {
2175                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2176                                  "Inode table for group %u overlaps "
2177                                  "superblock", i);
2178                         if (!(sb->s_flags & MS_RDONLY))
2179                                 return 0;
2180                 }
2181                 if (inode_table >= sb_block + 1 &&
2182                     inode_table <= last_bg_block) {
2183                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2184                                  "Inode table for group %u overlaps "
2185                                  "block group descriptors", i);
2186                         if (!(sb->s_flags & MS_RDONLY))
2187                                 return 0;
2188                 }
2189                 if (inode_table < first_block ||
2190                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2191                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2192                                "Inode table for group %u not in group "
2193                                "(block %llu)!", i, inode_table);
2194                         return 0;
2195                 }
2196                 ext4_lock_group(sb, i);
2197                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2198                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2199                                  "Checksum for group %u failed (%u!=%u)",
2200                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2201                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2202                         if (!(sb->s_flags & MS_RDONLY)) {
2203                                 ext4_unlock_group(sb, i);
2204                                 return 0;
2205                         }
2206                 }
2207                 ext4_unlock_group(sb, i);
2208                 if (!flexbg_flag)
2209                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2210         }
2211         if (NULL != first_not_zeroed)
2212                 *first_not_zeroed = grp;
2213         return 1;
2214 }
2215
2216 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2217  * the superblock) which were deleted from all directories, but held open by
2218  * a process at the time of a crash.  We walk the list and try to delete these
2219  * inodes at recovery time (only with a read-write filesystem).
2220  *
2221  * In order to keep the orphan inode chain consistent during traversal (in
2222  * case of crash during recovery), we link each inode into the superblock
2223  * orphan list_head and handle it the same way as an inode deletion during
2224  * normal operation (which journals the operations for us).
2225  *
2226  * We only do an iget() and an iput() on each inode, which is very safe if we
2227  * accidentally point at an in-use or already deleted inode.  The worst that
2228  * can happen in this case is that we get a "bit already cleared" message from
2229  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2230  * e2fsck was run on this filesystem, and it must have already done the orphan
2231  * inode cleanup for us, so we can safely abort without any further action.
2232  */
2233 static void ext4_orphan_cleanup(struct super_block *sb,
2234                                 struct ext4_super_block *es)
2235 {
2236         unsigned int s_flags = sb->s_flags;
2237         int nr_orphans = 0, nr_truncates = 0;
2238 #ifdef CONFIG_QUOTA
2239         int quota_update = 0;
2240         int i;
2241 #endif
2242         if (!es->s_last_orphan) {
2243                 jbd_debug(4, "no orphan inodes to clean up\n");
2244                 return;
2245         }
2246
2247         if (bdev_read_only(sb->s_bdev)) {
2248                 ext4_msg(sb, KERN_ERR, "write access "
2249                         "unavailable, skipping orphan cleanup");
2250                 return;
2251         }
2252
2253         /* Check if feature set would not allow a r/w mount */
2254         if (!ext4_feature_set_ok(sb, 0)) {
2255                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2256                          "unknown ROCOMPAT features");
2257                 return;
2258         }
2259
2260         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2261                 /* don't clear list on RO mount w/ errors */
2262                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2263                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2264                                   "clearing orphan list.\n");
2265                         es->s_last_orphan = 0;
2266                 }
2267                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2268                 return;
2269         }
2270
2271         if (s_flags & MS_RDONLY) {
2272                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2273                 sb->s_flags &= ~MS_RDONLY;
2274         }
2275 #ifdef CONFIG_QUOTA
2276         /* Needed for iput() to work correctly and not trash data */
2277         sb->s_flags |= MS_ACTIVE;
2278
2279         /*
2280          * Turn on quotas which were not enabled for read-only mounts if
2281          * filesystem has quota feature, so that they are updated correctly.
2282          */
2283         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2284                 int ret = ext4_enable_quotas(sb);
2285
2286                 if (!ret)
2287                         quota_update = 1;
2288                 else
2289                         ext4_msg(sb, KERN_ERR,
2290                                 "Cannot turn on quotas: error %d", ret);
2291         }
2292
2293         /* Turn on journaled quotas used for old sytle */
2294         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2295                 if (EXT4_SB(sb)->s_qf_names[i]) {
2296                         int ret = ext4_quota_on_mount(sb, i);
2297
2298                         if (!ret)
2299                                 quota_update = 1;
2300                         else
2301                                 ext4_msg(sb, KERN_ERR,
2302                                         "Cannot turn on journaled "
2303                                         "quota: type %d: error %d", i, ret);
2304                 }
2305         }
2306 #endif
2307
2308         while (es->s_last_orphan) {
2309                 struct inode *inode;
2310
2311                 /*
2312                  * We may have encountered an error during cleanup; if
2313                  * so, skip the rest.
2314                  */
2315                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2316                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2317                         es->s_last_orphan = 0;
2318                         break;
2319                 }
2320
2321                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2322                 if (IS_ERR(inode)) {
2323                         es->s_last_orphan = 0;
2324                         break;
2325                 }
2326
2327                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2328                 dquot_initialize(inode);
2329                 if (inode->i_nlink) {
2330                         if (test_opt(sb, DEBUG))
2331                                 ext4_msg(sb, KERN_DEBUG,
2332                                         "%s: truncating inode %lu to %lld bytes",
2333                                         __func__, inode->i_ino, inode->i_size);
2334                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2335                                   inode->i_ino, inode->i_size);
2336                         mutex_lock(&inode->i_mutex);
2337                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2338                         ext4_truncate(inode);
2339                         mutex_unlock(&inode->i_mutex);
2340                         nr_truncates++;
2341                 } else {
2342                         if (test_opt(sb, DEBUG))
2343                                 ext4_msg(sb, KERN_DEBUG,
2344                                         "%s: deleting unreferenced inode %lu",
2345                                         __func__, inode->i_ino);
2346                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2347                                   inode->i_ino);
2348                         nr_orphans++;
2349                 }
2350                 iput(inode);  /* The delete magic happens here! */
2351         }
2352
2353 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2354
2355         if (nr_orphans)
2356                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2357                        PLURAL(nr_orphans));
2358         if (nr_truncates)
2359                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2360                        PLURAL(nr_truncates));
2361 #ifdef CONFIG_QUOTA
2362         /* Turn off quotas if they were enabled for orphan cleanup */
2363         if (quota_update) {
2364                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2365                         if (sb_dqopt(sb)->files[i])
2366                                 dquot_quota_off(sb, i);
2367                 }
2368         }
2369 #endif
2370         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2371 }
2372
2373 /*
2374  * Maximal extent format file size.
2375  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2376  * extent format containers, within a sector_t, and within i_blocks
2377  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2378  * so that won't be a limiting factor.
2379  *
2380  * However there is other limiting factor. We do store extents in the form
2381  * of starting block and length, hence the resulting length of the extent
2382  * covering maximum file size must fit into on-disk format containers as
2383  * well. Given that length is always by 1 unit bigger than max unit (because
2384  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2385  *
2386  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2387  */
2388 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2389 {
2390         loff_t res;
2391         loff_t upper_limit = MAX_LFS_FILESIZE;
2392
2393         /* small i_blocks in vfs inode? */
2394         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2395                 /*
2396                  * CONFIG_LBDAF is not enabled implies the inode
2397                  * i_block represent total blocks in 512 bytes
2398                  * 32 == size of vfs inode i_blocks * 8
2399                  */
2400                 upper_limit = (1LL << 32) - 1;
2401
2402                 /* total blocks in file system block size */
2403                 upper_limit >>= (blkbits - 9);
2404                 upper_limit <<= blkbits;
2405         }
2406
2407         /*
2408          * 32-bit extent-start container, ee_block. We lower the maxbytes
2409          * by one fs block, so ee_len can cover the extent of maximum file
2410          * size
2411          */
2412         res = (1LL << 32) - 1;
2413         res <<= blkbits;
2414
2415         /* Sanity check against vm- & vfs- imposed limits */
2416         if (res > upper_limit)
2417                 res = upper_limit;
2418
2419         return res;
2420 }
2421
2422 /*
2423  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2424  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2425  * We need to be 1 filesystem block less than the 2^48 sector limit.
2426  */
2427 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2428 {
2429         loff_t res = EXT4_NDIR_BLOCKS;
2430         int meta_blocks;
2431         loff_t upper_limit;
2432         /* This is calculated to be the largest file size for a dense, block
2433          * mapped file such that the file's total number of 512-byte sectors,
2434          * including data and all indirect blocks, does not exceed (2^48 - 1).
2435          *
2436          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2437          * number of 512-byte sectors of the file.
2438          */
2439
2440         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2441                 /*
2442                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2443                  * the inode i_block field represents total file blocks in
2444                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2445                  */
2446                 upper_limit = (1LL << 32) - 1;
2447
2448                 /* total blocks in file system block size */
2449                 upper_limit >>= (bits - 9);
2450
2451         } else {
2452                 /*
2453                  * We use 48 bit ext4_inode i_blocks
2454                  * With EXT4_HUGE_FILE_FL set the i_blocks
2455                  * represent total number of blocks in
2456                  * file system block size
2457                  */
2458                 upper_limit = (1LL << 48) - 1;
2459
2460         }
2461
2462         /* indirect blocks */
2463         meta_blocks = 1;
2464         /* double indirect blocks */
2465         meta_blocks += 1 + (1LL << (bits-2));
2466         /* tripple indirect blocks */
2467         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2468
2469         upper_limit -= meta_blocks;
2470         upper_limit <<= bits;
2471
2472         res += 1LL << (bits-2);
2473         res += 1LL << (2*(bits-2));
2474         res += 1LL << (3*(bits-2));
2475         res <<= bits;
2476         if (res > upper_limit)
2477                 res = upper_limit;
2478
2479         if (res > MAX_LFS_FILESIZE)
2480                 res = MAX_LFS_FILESIZE;
2481
2482         return res;
2483 }
2484
2485 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2486                                    ext4_fsblk_t logical_sb_block, int nr)
2487 {
2488         struct ext4_sb_info *sbi = EXT4_SB(sb);
2489         ext4_group_t bg, first_meta_bg;
2490         int has_super = 0;
2491
2492         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2493
2494         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2495                 return logical_sb_block + nr + 1;
2496         bg = sbi->s_desc_per_block * nr;
2497         if (ext4_bg_has_super(sb, bg))
2498                 has_super = 1;
2499
2500         /*
2501          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2502          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2503          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2504          * compensate.
2505          */
2506         if (sb->s_blocksize == 1024 && nr == 0 &&
2507             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2508                 has_super++;
2509
2510         return (has_super + ext4_group_first_block_no(sb, bg));
2511 }
2512
2513 /**
2514  * ext4_get_stripe_size: Get the stripe size.
2515  * @sbi: In memory super block info
2516  *
2517  * If we have specified it via mount option, then
2518  * use the mount option value. If the value specified at mount time is
2519  * greater than the blocks per group use the super block value.
2520  * If the super block value is greater than blocks per group return 0.
2521  * Allocator needs it be less than blocks per group.
2522  *
2523  */
2524 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2525 {
2526         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2527         unsigned long stripe_width =
2528                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2529         int ret;
2530
2531         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2532                 ret = sbi->s_stripe;
2533         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2534                 ret = stripe_width;
2535         else if (stride && stride <= sbi->s_blocks_per_group)
2536                 ret = stride;
2537         else
2538                 ret = 0;
2539
2540         /*
2541          * If the stripe width is 1, this makes no sense and
2542          * we set it to 0 to turn off stripe handling code.
2543          */
2544         if (ret <= 1)
2545                 ret = 0;
2546
2547         return ret;
2548 }
2549
2550 /*
2551  * Check whether this filesystem can be mounted based on
2552  * the features present and the RDONLY/RDWR mount requested.
2553  * Returns 1 if this filesystem can be mounted as requested,
2554  * 0 if it cannot be.
2555  */
2556 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2557 {
2558         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2559                 ext4_msg(sb, KERN_ERR,
2560                         "Couldn't mount because of "
2561                         "unsupported optional features (%x)",
2562                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2563                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2564                 return 0;
2565         }
2566
2567         if (readonly)
2568                 return 1;
2569
2570         if (ext4_has_feature_readonly(sb)) {
2571                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2572                 sb->s_flags |= MS_RDONLY;
2573                 return 1;
2574         }
2575
2576         /* Check that feature set is OK for a read-write mount */
2577         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2578                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2579                          "unsupported optional features (%x)",
2580                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2581                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2582                 return 0;
2583         }
2584         /*
2585          * Large file size enabled file system can only be mounted
2586          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2587          */
2588         if (ext4_has_feature_huge_file(sb)) {
2589                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2590                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2591                                  "cannot be mounted RDWR without "
2592                                  "CONFIG_LBDAF");
2593                         return 0;
2594                 }
2595         }
2596         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2597                 ext4_msg(sb, KERN_ERR,
2598                          "Can't support bigalloc feature without "
2599                          "extents feature\n");
2600                 return 0;
2601         }
2602
2603 #ifndef CONFIG_QUOTA
2604         if (ext4_has_feature_quota(sb) && !readonly) {
2605                 ext4_msg(sb, KERN_ERR,
2606                          "Filesystem with quota feature cannot be mounted RDWR "
2607                          "without CONFIG_QUOTA");
2608                 return 0;
2609         }
2610 #endif  /* CONFIG_QUOTA */
2611         return 1;
2612 }
2613
2614 /*
2615  * This function is called once a day if we have errors logged
2616  * on the file system
2617  */
2618 static void print_daily_error_info(unsigned long arg)
2619 {
2620         struct super_block *sb = (struct super_block *) arg;
2621         struct ext4_sb_info *sbi;
2622         struct ext4_super_block *es;
2623
2624         sbi = EXT4_SB(sb);
2625         es = sbi->s_es;
2626
2627         if (es->s_error_count)
2628                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2629                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2630                          le32_to_cpu(es->s_error_count));
2631         if (es->s_first_error_time) {
2632                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2633                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2634                        (int) sizeof(es->s_first_error_func),
2635                        es->s_first_error_func,
2636                        le32_to_cpu(es->s_first_error_line));
2637                 if (es->s_first_error_ino)
2638                         printk(": inode %u",
2639                                le32_to_cpu(es->s_first_error_ino));
2640                 if (es->s_first_error_block)
2641                         printk(": block %llu", (unsigned long long)
2642                                le64_to_cpu(es->s_first_error_block));
2643                 printk("\n");
2644         }
2645         if (es->s_last_error_time) {
2646                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2647                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2648                        (int) sizeof(es->s_last_error_func),
2649                        es->s_last_error_func,
2650                        le32_to_cpu(es->s_last_error_line));
2651                 if (es->s_last_error_ino)
2652                         printk(": inode %u",
2653                                le32_to_cpu(es->s_last_error_ino));
2654                 if (es->s_last_error_block)
2655                         printk(": block %llu", (unsigned long long)
2656                                le64_to_cpu(es->s_last_error_block));
2657                 printk("\n");
2658         }
2659         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2660 }
2661
2662 /* Find next suitable group and run ext4_init_inode_table */
2663 static int ext4_run_li_request(struct ext4_li_request *elr)
2664 {
2665         struct ext4_group_desc *gdp = NULL;
2666         ext4_group_t group, ngroups;
2667         struct super_block *sb;
2668         unsigned long timeout = 0;
2669         int ret = 0;
2670
2671         sb = elr->lr_super;
2672         ngroups = EXT4_SB(sb)->s_groups_count;
2673
2674         sb_start_write(sb);
2675         for (group = elr->lr_next_group; group < ngroups; group++) {
2676                 gdp = ext4_get_group_desc(sb, group, NULL);
2677                 if (!gdp) {
2678                         ret = 1;
2679                         break;
2680                 }
2681
2682                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2683                         break;
2684         }
2685
2686         if (group >= ngroups)
2687                 ret = 1;
2688
2689         if (!ret) {
2690                 timeout = jiffies;
2691                 ret = ext4_init_inode_table(sb, group,
2692                                             elr->lr_timeout ? 0 : 1);
2693                 if (elr->lr_timeout == 0) {
2694                         timeout = (jiffies - timeout) *
2695                                   elr->lr_sbi->s_li_wait_mult;
2696                         elr->lr_timeout = timeout;
2697                 }
2698                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2699                 elr->lr_next_group = group + 1;
2700         }
2701         sb_end_write(sb);
2702
2703         return ret;
2704 }
2705
2706 /*
2707  * Remove lr_request from the list_request and free the
2708  * request structure. Should be called with li_list_mtx held
2709  */
2710 static void ext4_remove_li_request(struct ext4_li_request *elr)
2711 {
2712         struct ext4_sb_info *sbi;
2713
2714         if (!elr)
2715                 return;
2716
2717         sbi = elr->lr_sbi;
2718
2719         list_del(&elr->lr_request);
2720         sbi->s_li_request = NULL;
2721         kfree(elr);
2722 }
2723
2724 static void ext4_unregister_li_request(struct super_block *sb)
2725 {
2726         mutex_lock(&ext4_li_mtx);
2727         if (!ext4_li_info) {
2728                 mutex_unlock(&ext4_li_mtx);
2729                 return;
2730         }
2731
2732         mutex_lock(&ext4_li_info->li_list_mtx);
2733         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2734         mutex_unlock(&ext4_li_info->li_list_mtx);
2735         mutex_unlock(&ext4_li_mtx);
2736 }
2737
2738 static struct task_struct *ext4_lazyinit_task;
2739
2740 /*
2741  * This is the function where ext4lazyinit thread lives. It walks
2742  * through the request list searching for next scheduled filesystem.
2743  * When such a fs is found, run the lazy initialization request
2744  * (ext4_rn_li_request) and keep track of the time spend in this
2745  * function. Based on that time we compute next schedule time of
2746  * the request. When walking through the list is complete, compute
2747  * next waking time and put itself into sleep.
2748  */
2749 static int ext4_lazyinit_thread(void *arg)
2750 {
2751         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2752         struct list_head *pos, *n;
2753         struct ext4_li_request *elr;
2754         unsigned long next_wakeup, cur;
2755
2756         BUG_ON(NULL == eli);
2757
2758 cont_thread:
2759         while (true) {
2760                 next_wakeup = MAX_JIFFY_OFFSET;
2761
2762                 mutex_lock(&eli->li_list_mtx);
2763                 if (list_empty(&eli->li_request_list)) {
2764                         mutex_unlock(&eli->li_list_mtx);
2765                         goto exit_thread;
2766                 }
2767
2768                 list_for_each_safe(pos, n, &eli->li_request_list) {
2769                         elr = list_entry(pos, struct ext4_li_request,
2770                                          lr_request);
2771
2772                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2773                                 if (ext4_run_li_request(elr) != 0) {
2774                                         /* error, remove the lazy_init job */
2775                                         ext4_remove_li_request(elr);
2776                                         continue;
2777                                 }
2778                         }
2779
2780                         if (time_before(elr->lr_next_sched, next_wakeup))
2781                                 next_wakeup = elr->lr_next_sched;
2782                 }
2783                 mutex_unlock(&eli->li_list_mtx);
2784
2785                 try_to_freeze();
2786
2787                 cur = jiffies;
2788                 if ((time_after_eq(cur, next_wakeup)) ||
2789                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2790                         cond_resched();
2791                         continue;
2792                 }
2793
2794                 schedule_timeout_interruptible(next_wakeup - cur);
2795
2796                 if (kthread_should_stop()) {
2797                         ext4_clear_request_list();
2798                         goto exit_thread;
2799                 }
2800         }
2801
2802 exit_thread:
2803         /*
2804          * It looks like the request list is empty, but we need
2805          * to check it under the li_list_mtx lock, to prevent any
2806          * additions into it, and of course we should lock ext4_li_mtx
2807          * to atomically free the list and ext4_li_info, because at
2808          * this point another ext4 filesystem could be registering
2809          * new one.
2810          */
2811         mutex_lock(&ext4_li_mtx);
2812         mutex_lock(&eli->li_list_mtx);
2813         if (!list_empty(&eli->li_request_list)) {
2814                 mutex_unlock(&eli->li_list_mtx);
2815                 mutex_unlock(&ext4_li_mtx);
2816                 goto cont_thread;
2817         }
2818         mutex_unlock(&eli->li_list_mtx);
2819         kfree(ext4_li_info);
2820         ext4_li_info = NULL;
2821         mutex_unlock(&ext4_li_mtx);
2822
2823         return 0;
2824 }
2825
2826 static void ext4_clear_request_list(void)
2827 {
2828         struct list_head *pos, *n;
2829         struct ext4_li_request *elr;
2830
2831         mutex_lock(&ext4_li_info->li_list_mtx);
2832         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2833                 elr = list_entry(pos, struct ext4_li_request,
2834                                  lr_request);
2835                 ext4_remove_li_request(elr);
2836         }
2837         mutex_unlock(&ext4_li_info->li_list_mtx);
2838 }
2839
2840 static int ext4_run_lazyinit_thread(void)
2841 {
2842         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2843                                          ext4_li_info, "ext4lazyinit");
2844         if (IS_ERR(ext4_lazyinit_task)) {
2845                 int err = PTR_ERR(ext4_lazyinit_task);
2846                 ext4_clear_request_list();
2847                 kfree(ext4_li_info);
2848                 ext4_li_info = NULL;
2849                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2850                                  "initialization thread\n",
2851                                  err);
2852                 return err;
2853         }
2854         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2855         return 0;
2856 }
2857
2858 /*
2859  * Check whether it make sense to run itable init. thread or not.
2860  * If there is at least one uninitialized inode table, return
2861  * corresponding group number, else the loop goes through all
2862  * groups and return total number of groups.
2863  */
2864 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2865 {
2866         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2867         struct ext4_group_desc *gdp = NULL;
2868
2869         if (!ext4_has_group_desc_csum(sb))
2870                 return ngroups;
2871
2872         for (group = 0; group < ngroups; group++) {
2873                 gdp = ext4_get_group_desc(sb, group, NULL);
2874                 if (!gdp)
2875                         continue;
2876
2877                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
2878                         continue;
2879                 if (group != 0)
2880                         break;
2881                 ext4_error(sb, "Inode table for bg 0 marked as "
2882                            "needing zeroing");
2883                 if (sb->s_flags & MS_RDONLY)
2884                         return ngroups;
2885         }
2886
2887         return group;
2888 }
2889
2890 static int ext4_li_info_new(void)
2891 {
2892         struct ext4_lazy_init *eli = NULL;
2893
2894         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2895         if (!eli)
2896                 return -ENOMEM;
2897
2898         INIT_LIST_HEAD(&eli->li_request_list);
2899         mutex_init(&eli->li_list_mtx);
2900
2901         eli->li_state |= EXT4_LAZYINIT_QUIT;
2902
2903         ext4_li_info = eli;
2904
2905         return 0;
2906 }
2907
2908 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2909                                             ext4_group_t start)
2910 {
2911         struct ext4_sb_info *sbi = EXT4_SB(sb);
2912         struct ext4_li_request *elr;
2913
2914         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2915         if (!elr)
2916                 return NULL;
2917
2918         elr->lr_super = sb;
2919         elr->lr_sbi = sbi;
2920         elr->lr_next_group = start;
2921
2922         /*
2923          * Randomize first schedule time of the request to
2924          * spread the inode table initialization requests
2925          * better.
2926          */
2927         elr->lr_next_sched = jiffies + (prandom_u32() %
2928                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2929         return elr;
2930 }
2931
2932 int ext4_register_li_request(struct super_block *sb,
2933                              ext4_group_t first_not_zeroed)
2934 {
2935         struct ext4_sb_info *sbi = EXT4_SB(sb);
2936         struct ext4_li_request *elr = NULL;
2937         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2938         int ret = 0;
2939
2940         mutex_lock(&ext4_li_mtx);
2941         if (sbi->s_li_request != NULL) {
2942                 /*
2943                  * Reset timeout so it can be computed again, because
2944                  * s_li_wait_mult might have changed.
2945                  */
2946                 sbi->s_li_request->lr_timeout = 0;
2947                 goto out;
2948         }
2949
2950         if (first_not_zeroed == ngroups ||
2951             (sb->s_flags & MS_RDONLY) ||
2952             !test_opt(sb, INIT_INODE_TABLE))
2953                 goto out;
2954
2955         elr = ext4_li_request_new(sb, first_not_zeroed);
2956         if (!elr) {
2957                 ret = -ENOMEM;
2958                 goto out;
2959         }
2960
2961         if (NULL == ext4_li_info) {
2962                 ret = ext4_li_info_new();
2963                 if (ret)
2964                         goto out;
2965         }
2966
2967         mutex_lock(&ext4_li_info->li_list_mtx);
2968         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2969         mutex_unlock(&ext4_li_info->li_list_mtx);
2970
2971         sbi->s_li_request = elr;
2972         /*
2973          * set elr to NULL here since it has been inserted to
2974          * the request_list and the removal and free of it is
2975          * handled by ext4_clear_request_list from now on.
2976          */
2977         elr = NULL;
2978
2979         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2980                 ret = ext4_run_lazyinit_thread();
2981                 if (ret)
2982                         goto out;
2983         }
2984 out:
2985         mutex_unlock(&ext4_li_mtx);
2986         if (ret)
2987                 kfree(elr);
2988         return ret;
2989 }
2990
2991 /*
2992  * We do not need to lock anything since this is called on
2993  * module unload.
2994  */
2995 static void ext4_destroy_lazyinit_thread(void)
2996 {
2997         /*
2998          * If thread exited earlier
2999          * there's nothing to be done.
3000          */
3001         if (!ext4_li_info || !ext4_lazyinit_task)
3002                 return;
3003
3004         kthread_stop(ext4_lazyinit_task);
3005 }
3006
3007 static int set_journal_csum_feature_set(struct super_block *sb)
3008 {
3009         int ret = 1;
3010         int compat, incompat;
3011         struct ext4_sb_info *sbi = EXT4_SB(sb);
3012
3013         if (ext4_has_metadata_csum(sb)) {
3014                 /* journal checksum v3 */
3015                 compat = 0;
3016                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3017         } else {
3018                 /* journal checksum v1 */
3019                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3020                 incompat = 0;
3021         }
3022
3023         jbd2_journal_clear_features(sbi->s_journal,
3024                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3025                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3026                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3027         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3028                 ret = jbd2_journal_set_features(sbi->s_journal,
3029                                 compat, 0,
3030                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3031                                 incompat);
3032         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3033                 ret = jbd2_journal_set_features(sbi->s_journal,
3034                                 compat, 0,
3035                                 incompat);
3036                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3037                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3038         } else {
3039                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3040                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3041         }
3042
3043         return ret;
3044 }
3045
3046 /*
3047  * Note: calculating the overhead so we can be compatible with
3048  * historical BSD practice is quite difficult in the face of
3049  * clusters/bigalloc.  This is because multiple metadata blocks from
3050  * different block group can end up in the same allocation cluster.
3051  * Calculating the exact overhead in the face of clustered allocation
3052  * requires either O(all block bitmaps) in memory or O(number of block
3053  * groups**2) in time.  We will still calculate the superblock for
3054  * older file systems --- and if we come across with a bigalloc file
3055  * system with zero in s_overhead_clusters the estimate will be close to
3056  * correct especially for very large cluster sizes --- but for newer
3057  * file systems, it's better to calculate this figure once at mkfs
3058  * time, and store it in the superblock.  If the superblock value is
3059  * present (even for non-bigalloc file systems), we will use it.
3060  */
3061 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3062                           char *buf)
3063 {
3064         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3065         struct ext4_group_desc  *gdp;
3066         ext4_fsblk_t            first_block, last_block, b;
3067         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3068         int                     s, j, count = 0;
3069
3070         if (!ext4_has_feature_bigalloc(sb))
3071                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3072                         sbi->s_itb_per_group + 2);
3073
3074         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3075                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3076         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3077         for (i = 0; i < ngroups; i++) {
3078                 gdp = ext4_get_group_desc(sb, i, NULL);
3079                 b = ext4_block_bitmap(sb, gdp);
3080                 if (b >= first_block && b <= last_block) {
3081                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3082                         count++;
3083                 }
3084                 b = ext4_inode_bitmap(sb, gdp);
3085                 if (b >= first_block && b <= last_block) {
3086                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3087                         count++;
3088                 }
3089                 b = ext4_inode_table(sb, gdp);
3090                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3091                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3092                                 int c = EXT4_B2C(sbi, b - first_block);
3093                                 ext4_set_bit(c, buf);
3094                                 count++;
3095                         }
3096                 if (i != grp)
3097                         continue;
3098                 s = 0;
3099                 if (ext4_bg_has_super(sb, grp)) {
3100                         ext4_set_bit(s++, buf);
3101                         count++;
3102                 }
3103                 j = ext4_bg_num_gdb(sb, grp);
3104                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3105                         ext4_error(sb, "Invalid number of block group "
3106                                    "descriptor blocks: %d", j);
3107                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3108                 }
3109                 count += j;
3110                 for (; j > 0; j--)
3111                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3112         }
3113         if (!count)
3114                 return 0;
3115         return EXT4_CLUSTERS_PER_GROUP(sb) -
3116                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3117 }
3118
3119 /*
3120  * Compute the overhead and stash it in sbi->s_overhead
3121  */
3122 int ext4_calculate_overhead(struct super_block *sb)
3123 {
3124         struct ext4_sb_info *sbi = EXT4_SB(sb);
3125         struct ext4_super_block *es = sbi->s_es;
3126         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3127         ext4_fsblk_t overhead = 0;
3128         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3129
3130         if (!buf)
3131                 return -ENOMEM;
3132
3133         /*
3134          * Compute the overhead (FS structures).  This is constant
3135          * for a given filesystem unless the number of block groups
3136          * changes so we cache the previous value until it does.
3137          */
3138
3139         /*
3140          * All of the blocks before first_data_block are overhead
3141          */
3142         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3143
3144         /*
3145          * Add the overhead found in each block group
3146          */
3147         for (i = 0; i < ngroups; i++) {
3148                 int blks;
3149
3150                 blks = count_overhead(sb, i, buf);
3151                 overhead += blks;
3152                 if (blks)
3153                         memset(buf, 0, PAGE_SIZE);
3154                 cond_resched();
3155         }
3156         /* Add the internal journal blocks as well */
3157         if (sbi->s_journal && !sbi->journal_bdev)
3158                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3159
3160         sbi->s_overhead = overhead;
3161         smp_wmb();
3162         free_page((unsigned long) buf);
3163         return 0;
3164 }
3165
3166 static void ext4_set_resv_clusters(struct super_block *sb)
3167 {
3168         ext4_fsblk_t resv_clusters;
3169         struct ext4_sb_info *sbi = EXT4_SB(sb);
3170
3171         /*
3172          * There's no need to reserve anything when we aren't using extents.
3173          * The space estimates are exact, there are no unwritten extents,
3174          * hole punching doesn't need new metadata... This is needed especially
3175          * to keep ext2/3 backward compatibility.
3176          */
3177         if (!ext4_has_feature_extents(sb))
3178                 return;
3179         /*
3180          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3181          * This should cover the situations where we can not afford to run
3182          * out of space like for example punch hole, or converting
3183          * unwritten extents in delalloc path. In most cases such
3184          * allocation would require 1, or 2 blocks, higher numbers are
3185          * very rare.
3186          */
3187         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3188                          sbi->s_cluster_bits);
3189
3190         do_div(resv_clusters, 50);
3191         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3192
3193         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3194 }
3195
3196 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3197 {
3198         char *orig_data = kstrdup(data, GFP_KERNEL);
3199         struct buffer_head *bh;
3200         struct ext4_super_block *es = NULL;
3201         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3202         ext4_fsblk_t block;
3203         ext4_fsblk_t sb_block = get_sb_block(&data);
3204         ext4_fsblk_t logical_sb_block;
3205         unsigned long offset = 0;
3206         unsigned long journal_devnum = 0;
3207         unsigned long def_mount_opts;
3208         struct inode *root;
3209         const char *descr;
3210         int ret = -ENOMEM;
3211         int blocksize, clustersize;
3212         unsigned int db_count;
3213         unsigned int i;
3214         int needs_recovery, has_huge_files, has_bigalloc;
3215         __u64 blocks_count;
3216         int err = 0;
3217         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3218         ext4_group_t first_not_zeroed;
3219
3220         if ((data && !orig_data) || !sbi)
3221                 goto out_free_base;
3222
3223         sbi->s_blockgroup_lock =
3224                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3225         if (!sbi->s_blockgroup_lock)
3226                 goto out_free_base;
3227
3228         sb->s_fs_info = sbi;
3229         sbi->s_sb = sb;
3230         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3231         sbi->s_sb_block = sb_block;
3232         if (sb->s_bdev->bd_part)
3233                 sbi->s_sectors_written_start =
3234                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3235
3236         /* Cleanup superblock name */
3237         strreplace(sb->s_id, '/', '!');
3238
3239         /* -EINVAL is default */
3240         ret = -EINVAL;
3241         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3242         if (!blocksize) {
3243                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3244                 goto out_fail;
3245         }
3246
3247         /*
3248          * The ext4 superblock will not be buffer aligned for other than 1kB
3249          * block sizes.  We need to calculate the offset from buffer start.
3250          */
3251         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3252                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3253                 offset = do_div(logical_sb_block, blocksize);
3254         } else {
3255                 logical_sb_block = sb_block;
3256         }
3257
3258         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3259                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3260                 goto out_fail;
3261         }
3262         /*
3263          * Note: s_es must be initialized as soon as possible because
3264          *       some ext4 macro-instructions depend on its value
3265          */
3266         es = (struct ext4_super_block *) (bh->b_data + offset);
3267         sbi->s_es = es;
3268         sb->s_magic = le16_to_cpu(es->s_magic);
3269         if (sb->s_magic != EXT4_SUPER_MAGIC)
3270                 goto cantfind_ext4;
3271         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3272
3273         /* Warn if metadata_csum and gdt_csum are both set. */
3274         if (ext4_has_feature_metadata_csum(sb) &&
3275             ext4_has_feature_gdt_csum(sb))
3276                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3277                              "redundant flags; please run fsck.");
3278
3279         /* Check for a known checksum algorithm */
3280         if (!ext4_verify_csum_type(sb, es)) {
3281                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3282                          "unknown checksum algorithm.");
3283                 silent = 1;
3284                 goto cantfind_ext4;
3285         }
3286
3287         /* Load the checksum driver */
3288         if (ext4_has_feature_metadata_csum(sb)) {
3289                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3290                 if (IS_ERR(sbi->s_chksum_driver)) {
3291                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3292                         ret = PTR_ERR(sbi->s_chksum_driver);
3293                         sbi->s_chksum_driver = NULL;
3294                         goto failed_mount;
3295                 }
3296         }
3297
3298         /* Check superblock checksum */
3299         if (!ext4_superblock_csum_verify(sb, es)) {
3300                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3301                          "invalid superblock checksum.  Run e2fsck?");
3302                 silent = 1;
3303                 ret = -EFSBADCRC;
3304                 goto cantfind_ext4;
3305         }
3306
3307         /* Precompute checksum seed for all metadata */
3308         if (ext4_has_feature_csum_seed(sb))
3309                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3310         else if (ext4_has_metadata_csum(sb))
3311                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3312                                                sizeof(es->s_uuid));
3313
3314         /* Set defaults before we parse the mount options */
3315         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3316         set_opt(sb, INIT_INODE_TABLE);
3317         if (def_mount_opts & EXT4_DEFM_DEBUG)
3318                 set_opt(sb, DEBUG);
3319         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3320                 set_opt(sb, GRPID);
3321         if (def_mount_opts & EXT4_DEFM_UID16)
3322                 set_opt(sb, NO_UID32);
3323         /* xattr user namespace & acls are now defaulted on */
3324         set_opt(sb, XATTR_USER);
3325 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3326         set_opt(sb, POSIX_ACL);
3327 #endif
3328         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3329         if (ext4_has_metadata_csum(sb))
3330                 set_opt(sb, JOURNAL_CHECKSUM);
3331
3332         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3333                 set_opt(sb, JOURNAL_DATA);
3334         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3335                 set_opt(sb, ORDERED_DATA);
3336         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3337                 set_opt(sb, WRITEBACK_DATA);
3338
3339         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3340                 set_opt(sb, ERRORS_PANIC);
3341         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3342                 set_opt(sb, ERRORS_CONT);
3343         else
3344                 set_opt(sb, ERRORS_RO);
3345         /* block_validity enabled by default; disable with noblock_validity */
3346         set_opt(sb, BLOCK_VALIDITY);
3347         if (def_mount_opts & EXT4_DEFM_DISCARD)
3348                 set_opt(sb, DISCARD);
3349
3350         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3351         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3352         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3353         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3354         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3355
3356         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3357                 set_opt(sb, BARRIER);
3358
3359         /*
3360          * enable delayed allocation by default
3361          * Use -o nodelalloc to turn it off
3362          */
3363         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3364             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3365                 set_opt(sb, DELALLOC);
3366
3367         /*
3368          * set default s_li_wait_mult for lazyinit, for the case there is
3369          * no mount option specified.
3370          */
3371         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3372
3373         if (sbi->s_es->s_mount_opts[0]) {
3374                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3375                                               sizeof(sbi->s_es->s_mount_opts),
3376                                               GFP_KERNEL);
3377                 if (!s_mount_opts)
3378                         goto failed_mount;
3379                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3380                                    &journal_ioprio, 0)) {
3381                         ext4_msg(sb, KERN_WARNING,
3382                                  "failed to parse options in superblock: %s",
3383                                  s_mount_opts);
3384                 }
3385                 kfree(s_mount_opts);
3386         }
3387         sbi->s_def_mount_opt = sbi->s_mount_opt;
3388         if (!parse_options((char *) data, sb, &journal_devnum,
3389                            &journal_ioprio, 0))
3390                 goto failed_mount;
3391
3392         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3393                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3394                             "with data=journal disables delayed "
3395                             "allocation and O_DIRECT support!\n");
3396                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3397                         ext4_msg(sb, KERN_ERR, "can't mount with "
3398                                  "both data=journal and delalloc");
3399                         goto failed_mount;
3400                 }
3401                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3402                         ext4_msg(sb, KERN_ERR, "can't mount with "
3403                                  "both data=journal and dioread_nolock");
3404                         goto failed_mount;
3405                 }
3406                 if (test_opt(sb, DAX)) {
3407                         ext4_msg(sb, KERN_ERR, "can't mount with "
3408                                  "both data=journal and dax");
3409                         goto failed_mount;
3410                 }
3411                 if (ext4_has_feature_encrypt(sb)) {
3412                         ext4_msg(sb, KERN_WARNING,
3413                                  "encrypted files will use data=ordered "
3414                                  "instead of data journaling mode");
3415                 }
3416                 if (test_opt(sb, DELALLOC))
3417                         clear_opt(sb, DELALLOC);
3418         } else {
3419                 sb->s_iflags |= SB_I_CGROUPWB;
3420         }
3421
3422         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3423                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3424
3425         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3426             (ext4_has_compat_features(sb) ||
3427              ext4_has_ro_compat_features(sb) ||
3428              ext4_has_incompat_features(sb)))
3429                 ext4_msg(sb, KERN_WARNING,
3430                        "feature flags set on rev 0 fs, "
3431                        "running e2fsck is recommended");
3432
3433         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3434                 set_opt2(sb, HURD_COMPAT);
3435                 if (ext4_has_feature_64bit(sb)) {
3436                         ext4_msg(sb, KERN_ERR,
3437                                  "The Hurd can't support 64-bit file systems");
3438                         goto failed_mount;
3439                 }
3440         }
3441
3442         if (IS_EXT2_SB(sb)) {
3443                 if (ext2_feature_set_ok(sb))
3444                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3445                                  "using the ext4 subsystem");
3446                 else {
3447                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3448                                  "to feature incompatibilities");
3449                         goto failed_mount;
3450                 }
3451         }
3452
3453         if (IS_EXT3_SB(sb)) {
3454                 if (ext3_feature_set_ok(sb))
3455                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3456                                  "using the ext4 subsystem");
3457                 else {
3458                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3459                                  "to feature incompatibilities");
3460                         goto failed_mount;
3461                 }
3462         }
3463
3464         /*
3465          * Check feature flags regardless of the revision level, since we
3466          * previously didn't change the revision level when setting the flags,
3467          * so there is a chance incompat flags are set on a rev 0 filesystem.
3468          */
3469         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3470                 goto failed_mount;
3471
3472         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3473         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3474             blocksize > EXT4_MAX_BLOCK_SIZE) {
3475                 ext4_msg(sb, KERN_ERR,
3476                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3477                          blocksize, le32_to_cpu(es->s_log_block_size));
3478                 goto failed_mount;
3479         }
3480         if (le32_to_cpu(es->s_log_block_size) >
3481             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3482                 ext4_msg(sb, KERN_ERR,
3483                          "Invalid log block size: %u",
3484                          le32_to_cpu(es->s_log_block_size));
3485                 goto failed_mount;
3486         }
3487         if (le32_to_cpu(es->s_log_cluster_size) >
3488             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3489                 ext4_msg(sb, KERN_ERR,
3490                          "Invalid log cluster size: %u",
3491                          le32_to_cpu(es->s_log_cluster_size));
3492                 goto failed_mount;
3493         }
3494
3495         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3496                 ext4_msg(sb, KERN_ERR,
3497                          "Number of reserved GDT blocks insanely large: %d",
3498                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3499                 goto failed_mount;
3500         }
3501
3502         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3503                 if (blocksize != PAGE_SIZE) {
3504                         ext4_msg(sb, KERN_ERR,
3505                                         "error: unsupported blocksize for dax");
3506                         goto failed_mount;
3507                 }
3508                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3509                         ext4_msg(sb, KERN_ERR,
3510                                         "error: device does not support dax");
3511                         goto failed_mount;
3512                 }
3513         }
3514
3515         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3516                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3517                          es->s_encryption_level);
3518                 goto failed_mount;
3519         }
3520
3521         if (sb->s_blocksize != blocksize) {
3522                 /* Validate the filesystem blocksize */
3523                 if (!sb_set_blocksize(sb, blocksize)) {
3524                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3525                                         blocksize);
3526                         goto failed_mount;
3527                 }
3528
3529                 brelse(bh);
3530                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3531                 offset = do_div(logical_sb_block, blocksize);
3532                 bh = sb_bread_unmovable(sb, logical_sb_block);
3533                 if (!bh) {
3534                         ext4_msg(sb, KERN_ERR,
3535                                "Can't read superblock on 2nd try");
3536                         goto failed_mount;
3537                 }
3538                 es = (struct ext4_super_block *)(bh->b_data + offset);
3539                 sbi->s_es = es;
3540                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3541                         ext4_msg(sb, KERN_ERR,
3542                                "Magic mismatch, very weird!");
3543                         goto failed_mount;
3544                 }
3545         }
3546
3547         has_huge_files = ext4_has_feature_huge_file(sb);
3548         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3549                                                       has_huge_files);
3550         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3551
3552         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3553                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3554                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3555         } else {
3556                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3557                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3558                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3559                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3560                                  sbi->s_first_ino);
3561                         goto failed_mount;
3562                 }
3563                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3564                     (!is_power_of_2(sbi->s_inode_size)) ||
3565                     (sbi->s_inode_size > blocksize)) {
3566                         ext4_msg(sb, KERN_ERR,
3567                                "unsupported inode size: %d",
3568                                sbi->s_inode_size);
3569                         goto failed_mount;
3570                 }
3571                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3572                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3573         }
3574
3575         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3576         if (ext4_has_feature_64bit(sb)) {
3577                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3578                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3579                     !is_power_of_2(sbi->s_desc_size)) {
3580                         ext4_msg(sb, KERN_ERR,
3581                                "unsupported descriptor size %lu",
3582                                sbi->s_desc_size);
3583                         goto failed_mount;
3584                 }
3585         } else
3586                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3587
3588         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3589         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3590
3591         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3592         if (sbi->s_inodes_per_block == 0)
3593                 goto cantfind_ext4;
3594         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3595             sbi->s_inodes_per_group > blocksize * 8) {
3596                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3597                          sbi->s_blocks_per_group);
3598                 goto failed_mount;
3599         }
3600         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3601                                         sbi->s_inodes_per_block;
3602         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3603         sbi->s_sbh = bh;
3604         sbi->s_mount_state = le16_to_cpu(es->s_state);
3605         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3606         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3607
3608         for (i = 0; i < 4; i++)
3609                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3610         sbi->s_def_hash_version = es->s_def_hash_version;
3611         if (ext4_has_feature_dir_index(sb)) {
3612                 i = le32_to_cpu(es->s_flags);
3613                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3614                         sbi->s_hash_unsigned = 3;
3615                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3616 #ifdef __CHAR_UNSIGNED__
3617                         if (!(sb->s_flags & MS_RDONLY))
3618                                 es->s_flags |=
3619                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3620                         sbi->s_hash_unsigned = 3;
3621 #else
3622                         if (!(sb->s_flags & MS_RDONLY))
3623                                 es->s_flags |=
3624                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3625 #endif
3626                 }
3627         }
3628
3629         /* Handle clustersize */
3630         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3631         has_bigalloc = ext4_has_feature_bigalloc(sb);
3632         if (has_bigalloc) {
3633                 if (clustersize < blocksize) {
3634                         ext4_msg(sb, KERN_ERR,
3635                                  "cluster size (%d) smaller than "
3636                                  "block size (%d)", clustersize, blocksize);
3637                         goto failed_mount;
3638                 }
3639                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3640                         le32_to_cpu(es->s_log_block_size);
3641                 sbi->s_clusters_per_group =
3642                         le32_to_cpu(es->s_clusters_per_group);
3643                 if (sbi->s_clusters_per_group > blocksize * 8) {
3644                         ext4_msg(sb, KERN_ERR,
3645                                  "#clusters per group too big: %lu",
3646                                  sbi->s_clusters_per_group);
3647                         goto failed_mount;
3648                 }
3649                 if (sbi->s_blocks_per_group !=
3650                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3651                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3652                                  "clusters per group (%lu) inconsistent",
3653                                  sbi->s_blocks_per_group,
3654                                  sbi->s_clusters_per_group);
3655                         goto failed_mount;
3656                 }
3657         } else {
3658                 if (clustersize != blocksize) {
3659                         ext4_msg(sb, KERN_ERR,
3660                                  "fragment/cluster size (%d) != "
3661                                  "block size (%d)", clustersize, blocksize);
3662                         goto failed_mount;
3663                 }
3664                 if (sbi->s_blocks_per_group > blocksize * 8) {
3665                         ext4_msg(sb, KERN_ERR,
3666                                  "#blocks per group too big: %lu",
3667                                  sbi->s_blocks_per_group);
3668                         goto failed_mount;
3669                 }
3670                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3671                 sbi->s_cluster_bits = 0;
3672         }
3673         sbi->s_cluster_ratio = clustersize / blocksize;
3674
3675         /* Do we have standard group size of clustersize * 8 blocks ? */
3676         if (sbi->s_blocks_per_group == clustersize << 3)
3677                 set_opt2(sb, STD_GROUP_SIZE);
3678
3679         /*
3680          * Test whether we have more sectors than will fit in sector_t,
3681          * and whether the max offset is addressable by the page cache.
3682          */
3683         err = generic_check_addressable(sb->s_blocksize_bits,
3684                                         ext4_blocks_count(es));
3685         if (err) {
3686                 ext4_msg(sb, KERN_ERR, "filesystem"
3687                          " too large to mount safely on this system");
3688                 if (sizeof(sector_t) < 8)
3689                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3690                 goto failed_mount;
3691         }
3692
3693         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3694                 goto cantfind_ext4;
3695
3696         /* check blocks count against device size */
3697         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3698         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3699                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3700                        "exceeds size of device (%llu blocks)",
3701                        ext4_blocks_count(es), blocks_count);
3702                 goto failed_mount;
3703         }
3704
3705         /*
3706          * It makes no sense for the first data block to be beyond the end
3707          * of the filesystem.
3708          */
3709         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3710                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3711                          "block %u is beyond end of filesystem (%llu)",
3712                          le32_to_cpu(es->s_first_data_block),
3713                          ext4_blocks_count(es));
3714                 goto failed_mount;
3715         }
3716         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
3717             (sbi->s_cluster_ratio == 1)) {
3718                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3719                          "block is 0 with a 1k block and cluster size");
3720                 goto failed_mount;
3721         }
3722
3723         blocks_count = (ext4_blocks_count(es) -
3724                         le32_to_cpu(es->s_first_data_block) +
3725                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3726         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3727         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3728                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3729                        "(block count %llu, first data block %u, "
3730                        "blocks per group %lu)", sbi->s_groups_count,
3731                        ext4_blocks_count(es),
3732                        le32_to_cpu(es->s_first_data_block),
3733                        EXT4_BLOCKS_PER_GROUP(sb));
3734                 goto failed_mount;
3735         }
3736         sbi->s_groups_count = blocks_count;
3737         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3738                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3739         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3740                    EXT4_DESC_PER_BLOCK(sb);
3741         if (ext4_has_feature_meta_bg(sb)) {
3742                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3743                         ext4_msg(sb, KERN_WARNING,
3744                                  "first meta block group too large: %u "
3745                                  "(group descriptor block count %u)",
3746                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3747                         goto failed_mount;
3748                 }
3749         }
3750         sbi->s_group_desc = ext4_kvmalloc(db_count *
3751                                           sizeof(struct buffer_head *),
3752                                           GFP_KERNEL);
3753         if (sbi->s_group_desc == NULL) {
3754                 ext4_msg(sb, KERN_ERR, "not enough memory");
3755                 ret = -ENOMEM;
3756                 goto failed_mount;
3757         }
3758         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
3759             le32_to_cpu(es->s_inodes_count)) {
3760                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
3761                          le32_to_cpu(es->s_inodes_count),
3762                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
3763                 ret = -EINVAL;
3764                 goto failed_mount;
3765         }
3766
3767         bgl_lock_init(sbi->s_blockgroup_lock);
3768
3769         for (i = 0; i < db_count; i++) {
3770                 block = descriptor_loc(sb, logical_sb_block, i);
3771                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3772                 if (!sbi->s_group_desc[i]) {
3773                         ext4_msg(sb, KERN_ERR,
3774                                "can't read group descriptor %d", i);
3775                         db_count = i;
3776                         goto failed_mount2;
3777                 }
3778         }
3779         sbi->s_gdb_count = db_count;
3780         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3781                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3782                 ret = -EFSCORRUPTED;
3783                 goto failed_mount2;
3784         }
3785
3786         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3787         spin_lock_init(&sbi->s_next_gen_lock);
3788
3789         setup_timer(&sbi->s_err_report, print_daily_error_info,
3790                 (unsigned long) sb);
3791
3792         /* Register extent status tree shrinker */
3793         if (ext4_es_register_shrinker(sbi))
3794                 goto failed_mount3;
3795
3796         sbi->s_stripe = ext4_get_stripe_size(sbi);
3797         sbi->s_extent_max_zeroout_kb = 32;
3798
3799         /*
3800          * set up enough so that it can read an inode
3801          */
3802         sb->s_op = &ext4_sops;
3803         sb->s_export_op = &ext4_export_ops;
3804         sb->s_xattr = ext4_xattr_handlers;
3805 #ifdef CONFIG_QUOTA
3806         sb->dq_op = &ext4_quota_operations;
3807         if (ext4_has_feature_quota(sb))
3808                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3809         else
3810                 sb->s_qcop = &ext4_qctl_operations;
3811         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3812 #endif
3813         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3814
3815         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3816         mutex_init(&sbi->s_orphan_lock);
3817
3818         sb->s_root = NULL;
3819
3820         needs_recovery = (es->s_last_orphan != 0 ||
3821                           ext4_has_feature_journal_needs_recovery(sb));
3822
3823         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3824                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3825                         goto failed_mount3a;
3826
3827         /*
3828          * The first inode we look at is the journal inode.  Don't try
3829          * root first: it may be modified in the journal!
3830          */
3831         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3832                 err = ext4_load_journal(sb, es, journal_devnum);
3833                 if (err)
3834                         goto failed_mount3a;
3835         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3836                    ext4_has_feature_journal_needs_recovery(sb)) {
3837                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3838                        "suppressed and not mounted read-only");
3839                 goto failed_mount_wq;
3840         } else {
3841                 /* Nojournal mode, all journal mount options are illegal */
3842                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3843                         ext4_msg(sb, KERN_ERR, "can't mount with "
3844                                  "journal_checksum, fs mounted w/o journal");
3845                         goto failed_mount_wq;
3846                 }
3847                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3848                         ext4_msg(sb, KERN_ERR, "can't mount with "
3849                                  "journal_async_commit, fs mounted w/o journal");
3850                         goto failed_mount_wq;
3851                 }
3852                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3853                         ext4_msg(sb, KERN_ERR, "can't mount with "
3854                                  "commit=%lu, fs mounted w/o journal",
3855                                  sbi->s_commit_interval / HZ);
3856                         goto failed_mount_wq;
3857                 }
3858                 if (EXT4_MOUNT_DATA_FLAGS &
3859                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3860                         ext4_msg(sb, KERN_ERR, "can't mount with "
3861                                  "data=, fs mounted w/o journal");
3862                         goto failed_mount_wq;
3863                 }
3864                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3865                 clear_opt(sb, JOURNAL_CHECKSUM);
3866                 clear_opt(sb, DATA_FLAGS);
3867                 sbi->s_journal = NULL;
3868                 needs_recovery = 0;
3869                 goto no_journal;
3870         }
3871
3872         if (ext4_has_feature_64bit(sb) &&
3873             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3874                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3875                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3876                 goto failed_mount_wq;
3877         }
3878
3879         if (!set_journal_csum_feature_set(sb)) {
3880                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3881                          "feature set");
3882                 goto failed_mount_wq;
3883         }
3884
3885         /* We have now updated the journal if required, so we can
3886          * validate the data journaling mode. */
3887         switch (test_opt(sb, DATA_FLAGS)) {
3888         case 0:
3889                 /* No mode set, assume a default based on the journal
3890                  * capabilities: ORDERED_DATA if the journal can
3891                  * cope, else JOURNAL_DATA
3892                  */
3893                 if (jbd2_journal_check_available_features
3894                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3895                         set_opt(sb, ORDERED_DATA);
3896                 else
3897                         set_opt(sb, JOURNAL_DATA);
3898                 break;
3899
3900         case EXT4_MOUNT_ORDERED_DATA:
3901         case EXT4_MOUNT_WRITEBACK_DATA:
3902                 if (!jbd2_journal_check_available_features
3903                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3904                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3905                                "requested data journaling mode");
3906                         goto failed_mount_wq;
3907                 }
3908         default:
3909                 break;
3910         }
3911         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3912
3913         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3914
3915 no_journal:
3916         if (ext4_mballoc_ready) {
3917                 sbi->s_mb_cache = ext4_xattr_create_cache();
3918                 if (!sbi->s_mb_cache) {
3919                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3920                         goto failed_mount_wq;
3921                 }
3922         }
3923
3924         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3925             (blocksize != PAGE_CACHE_SIZE)) {
3926                 ext4_msg(sb, KERN_ERR,
3927                          "Unsupported blocksize for fs encryption");
3928                 goto failed_mount_wq;
3929         }
3930
3931         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3932             !ext4_has_feature_encrypt(sb)) {
3933                 ext4_set_feature_encrypt(sb);
3934                 ext4_commit_super(sb, 1);
3935         }
3936
3937         /*
3938          * Get the # of file system overhead blocks from the
3939          * superblock if present.
3940          */
3941         if (es->s_overhead_clusters)
3942                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3943         else {
3944                 err = ext4_calculate_overhead(sb);
3945                 if (err)
3946                         goto failed_mount_wq;
3947         }
3948
3949         /*
3950          * The maximum number of concurrent works can be high and
3951          * concurrency isn't really necessary.  Limit it to 1.
3952          */
3953         EXT4_SB(sb)->rsv_conversion_wq =
3954                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3955         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3956                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3957                 ret = -ENOMEM;
3958                 goto failed_mount4;
3959         }
3960
3961         /*
3962          * The jbd2_journal_load will have done any necessary log recovery,
3963          * so we can safely mount the rest of the filesystem now.
3964          */
3965
3966         root = ext4_iget(sb, EXT4_ROOT_INO);
3967         if (IS_ERR(root)) {
3968                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3969                 ret = PTR_ERR(root);
3970                 root = NULL;
3971                 goto failed_mount4;
3972         }
3973         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3974                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3975                 iput(root);
3976                 goto failed_mount4;
3977         }
3978         sb->s_root = d_make_root(root);
3979         if (!sb->s_root) {
3980                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3981                 ret = -ENOMEM;
3982                 goto failed_mount4;
3983         }
3984
3985         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3986                 sb->s_flags |= MS_RDONLY;
3987
3988         /* determine the minimum size of new large inodes, if present */
3989         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3990                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3991                                                      EXT4_GOOD_OLD_INODE_SIZE;
3992                 if (ext4_has_feature_extra_isize(sb)) {
3993                         if (sbi->s_want_extra_isize <
3994                             le16_to_cpu(es->s_want_extra_isize))
3995                                 sbi->s_want_extra_isize =
3996                                         le16_to_cpu(es->s_want_extra_isize);
3997                         if (sbi->s_want_extra_isize <
3998                             le16_to_cpu(es->s_min_extra_isize))
3999                                 sbi->s_want_extra_isize =
4000                                         le16_to_cpu(es->s_min_extra_isize);
4001                 }
4002         }
4003         /* Check if enough inode space is available */
4004         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4005                                                         sbi->s_inode_size) {
4006                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4007                                                        EXT4_GOOD_OLD_INODE_SIZE;
4008                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4009                          "available");
4010         }
4011
4012         ext4_set_resv_clusters(sb);
4013
4014         err = ext4_setup_system_zone(sb);
4015         if (err) {
4016                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4017                          "zone (%d)", err);
4018                 goto failed_mount4a;
4019         }
4020
4021         ext4_ext_init(sb);
4022         err = ext4_mb_init(sb);
4023         if (err) {
4024                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4025                          err);
4026                 goto failed_mount5;
4027         }
4028
4029         block = ext4_count_free_clusters(sb);
4030         ext4_free_blocks_count_set(sbi->s_es,
4031                                    EXT4_C2B(sbi, block));
4032         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4033                                   GFP_KERNEL);
4034         if (!err) {
4035                 unsigned long freei = ext4_count_free_inodes(sb);
4036                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4037                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4038                                           GFP_KERNEL);
4039         }
4040         if (!err)
4041                 err = percpu_counter_init(&sbi->s_dirs_counter,
4042                                           ext4_count_dirs(sb), GFP_KERNEL);
4043         if (!err)
4044                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4045                                           GFP_KERNEL);
4046         if (err) {
4047                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4048                 goto failed_mount6;
4049         }
4050
4051         if (ext4_has_feature_flex_bg(sb))
4052                 if (!ext4_fill_flex_info(sb)) {
4053                         ext4_msg(sb, KERN_ERR,
4054                                "unable to initialize "
4055                                "flex_bg meta info!");
4056                         goto failed_mount6;
4057                 }
4058
4059         err = ext4_register_li_request(sb, first_not_zeroed);
4060         if (err)
4061                 goto failed_mount6;
4062
4063         err = ext4_register_sysfs(sb);
4064         if (err)
4065                 goto failed_mount7;
4066
4067 #ifdef CONFIG_QUOTA
4068         /* Enable quota usage during mount. */
4069         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4070                 err = ext4_enable_quotas(sb);
4071                 if (err)
4072                         goto failed_mount8;
4073         }
4074 #endif  /* CONFIG_QUOTA */
4075
4076         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4077         ext4_orphan_cleanup(sb, es);
4078         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4079         if (needs_recovery) {
4080                 ext4_msg(sb, KERN_INFO, "recovery complete");
4081                 ext4_mark_recovery_complete(sb, es);
4082         }
4083         if (EXT4_SB(sb)->s_journal) {
4084                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4085                         descr = " journalled data mode";
4086                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4087                         descr = " ordered data mode";
4088                 else
4089                         descr = " writeback data mode";
4090         } else
4091                 descr = "out journal";
4092
4093         if (test_opt(sb, DISCARD)) {
4094                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4095                 if (!blk_queue_discard(q))
4096                         ext4_msg(sb, KERN_WARNING,
4097                                  "mounting with \"discard\" option, but "
4098                                  "the device does not support discard");
4099         }
4100
4101         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4102                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4103                          "Opts: %.*s%s%s", descr,
4104                          (int) sizeof(sbi->s_es->s_mount_opts),
4105                          sbi->s_es->s_mount_opts,
4106                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4107
4108         if (es->s_error_count)
4109                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4110
4111         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4112         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4113         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4114         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4115
4116         kfree(orig_data);
4117         return 0;
4118
4119 cantfind_ext4:
4120         if (!silent)
4121                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4122         goto failed_mount;
4123
4124 #ifdef CONFIG_QUOTA
4125 failed_mount8:
4126         ext4_unregister_sysfs(sb);
4127 #endif
4128 failed_mount7:
4129         ext4_unregister_li_request(sb);
4130 failed_mount6:
4131         ext4_mb_release(sb);
4132         if (sbi->s_flex_groups)
4133                 kvfree(sbi->s_flex_groups);
4134         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4135         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4136         percpu_counter_destroy(&sbi->s_dirs_counter);
4137         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4138 failed_mount5:
4139         ext4_ext_release(sb);
4140         ext4_release_system_zone(sb);
4141 failed_mount4a:
4142         dput(sb->s_root);
4143         sb->s_root = NULL;
4144 failed_mount4:
4145         ext4_msg(sb, KERN_ERR, "mount failed");
4146         if (EXT4_SB(sb)->rsv_conversion_wq)
4147                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4148 failed_mount_wq:
4149         if (sbi->s_mb_cache) {
4150                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4151                 sbi->s_mb_cache = NULL;
4152         }
4153         if (sbi->s_journal) {
4154                 jbd2_journal_destroy(sbi->s_journal);
4155                 sbi->s_journal = NULL;
4156         }
4157 failed_mount3a:
4158         ext4_es_unregister_shrinker(sbi);
4159 failed_mount3:
4160         del_timer_sync(&sbi->s_err_report);
4161         if (sbi->s_mmp_tsk)
4162                 kthread_stop(sbi->s_mmp_tsk);
4163 failed_mount2:
4164         for (i = 0; i < db_count; i++)
4165                 brelse(sbi->s_group_desc[i]);
4166         kvfree(sbi->s_group_desc);
4167 failed_mount:
4168         if (sbi->s_chksum_driver)
4169                 crypto_free_shash(sbi->s_chksum_driver);
4170 #ifdef CONFIG_QUOTA
4171         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4172                 kfree(sbi->s_qf_names[i]);
4173 #endif
4174         ext4_blkdev_remove(sbi);
4175         brelse(bh);
4176 out_fail:
4177         sb->s_fs_info = NULL;
4178         kfree(sbi->s_blockgroup_lock);
4179 out_free_base:
4180         kfree(sbi);
4181         kfree(orig_data);
4182         return err ? err : ret;
4183 }
4184
4185 /*
4186  * Setup any per-fs journal parameters now.  We'll do this both on
4187  * initial mount, once the journal has been initialised but before we've
4188  * done any recovery; and again on any subsequent remount.
4189  */
4190 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4191 {
4192         struct ext4_sb_info *sbi = EXT4_SB(sb);
4193
4194         journal->j_commit_interval = sbi->s_commit_interval;
4195         journal->j_min_batch_time = sbi->s_min_batch_time;
4196         journal->j_max_batch_time = sbi->s_max_batch_time;
4197
4198         write_lock(&journal->j_state_lock);
4199         if (test_opt(sb, BARRIER))
4200                 journal->j_flags |= JBD2_BARRIER;
4201         else
4202                 journal->j_flags &= ~JBD2_BARRIER;
4203         if (test_opt(sb, DATA_ERR_ABORT))
4204                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4205         else
4206                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4207         write_unlock(&journal->j_state_lock);
4208 }
4209
4210 static journal_t *ext4_get_journal(struct super_block *sb,
4211                                    unsigned int journal_inum)
4212 {
4213         struct inode *journal_inode;
4214         journal_t *journal;
4215
4216         BUG_ON(!ext4_has_feature_journal(sb));
4217
4218         /* First, test for the existence of a valid inode on disk.  Bad
4219          * things happen if we iget() an unused inode, as the subsequent
4220          * iput() will try to delete it. */
4221
4222         journal_inode = ext4_iget(sb, journal_inum);
4223         if (IS_ERR(journal_inode)) {
4224                 ext4_msg(sb, KERN_ERR, "no journal found");
4225                 return NULL;
4226         }
4227         if (!journal_inode->i_nlink) {
4228                 make_bad_inode(journal_inode);
4229                 iput(journal_inode);
4230                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4231                 return NULL;
4232         }
4233
4234         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4235                   journal_inode, journal_inode->i_size);
4236         if (!S_ISREG(journal_inode->i_mode)) {
4237                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4238                 iput(journal_inode);
4239                 return NULL;
4240         }
4241
4242         journal = jbd2_journal_init_inode(journal_inode);
4243         if (!journal) {
4244                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4245                 iput(journal_inode);
4246                 return NULL;
4247         }
4248         journal->j_private = sb;
4249         ext4_init_journal_params(sb, journal);
4250         return journal;
4251 }
4252
4253 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4254                                        dev_t j_dev)
4255 {
4256         struct buffer_head *bh;
4257         journal_t *journal;
4258         ext4_fsblk_t start;
4259         ext4_fsblk_t len;
4260         int hblock, blocksize;
4261         ext4_fsblk_t sb_block;
4262         unsigned long offset;
4263         struct ext4_super_block *es;
4264         struct block_device *bdev;
4265
4266         BUG_ON(!ext4_has_feature_journal(sb));
4267
4268         bdev = ext4_blkdev_get(j_dev, sb);
4269         if (bdev == NULL)
4270                 return NULL;
4271
4272         blocksize = sb->s_blocksize;
4273         hblock = bdev_logical_block_size(bdev);
4274         if (blocksize < hblock) {
4275                 ext4_msg(sb, KERN_ERR,
4276                         "blocksize too small for journal device");
4277                 goto out_bdev;
4278         }
4279
4280         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4281         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4282         set_blocksize(bdev, blocksize);
4283         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4284                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4285                        "external journal");
4286                 goto out_bdev;
4287         }
4288
4289         es = (struct ext4_super_block *) (bh->b_data + offset);
4290         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4291             !(le32_to_cpu(es->s_feature_incompat) &
4292               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4293                 ext4_msg(sb, KERN_ERR, "external journal has "
4294                                         "bad superblock");
4295                 brelse(bh);
4296                 goto out_bdev;
4297         }
4298
4299         if ((le32_to_cpu(es->s_feature_ro_compat) &
4300              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4301             es->s_checksum != ext4_superblock_csum(sb, es)) {
4302                 ext4_msg(sb, KERN_ERR, "external journal has "
4303                                        "corrupt superblock");
4304                 brelse(bh);
4305                 goto out_bdev;
4306         }
4307
4308         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4309                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4310                 brelse(bh);
4311                 goto out_bdev;
4312         }
4313
4314         len = ext4_blocks_count(es);
4315         start = sb_block + 1;
4316         brelse(bh);     /* we're done with the superblock */
4317
4318         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4319                                         start, len, blocksize);
4320         if (!journal) {
4321                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4322                 goto out_bdev;
4323         }
4324         journal->j_private = sb;
4325         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4326         wait_on_buffer(journal->j_sb_buffer);
4327         if (!buffer_uptodate(journal->j_sb_buffer)) {
4328                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4329                 goto out_journal;
4330         }
4331         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4332                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4333                                         "user (unsupported) - %d",
4334                         be32_to_cpu(journal->j_superblock->s_nr_users));
4335                 goto out_journal;
4336         }
4337         EXT4_SB(sb)->journal_bdev = bdev;
4338         ext4_init_journal_params(sb, journal);
4339         return journal;
4340
4341 out_journal:
4342         jbd2_journal_destroy(journal);
4343 out_bdev:
4344         ext4_blkdev_put(bdev);
4345         return NULL;
4346 }
4347
4348 static int ext4_load_journal(struct super_block *sb,
4349                              struct ext4_super_block *es,
4350                              unsigned long journal_devnum)
4351 {
4352         journal_t *journal;
4353         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4354         dev_t journal_dev;
4355         int err = 0;
4356         int really_read_only;
4357
4358         BUG_ON(!ext4_has_feature_journal(sb));
4359
4360         if (journal_devnum &&
4361             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4362                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4363                         "numbers have changed");
4364                 journal_dev = new_decode_dev(journal_devnum);
4365         } else
4366                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4367
4368         really_read_only = bdev_read_only(sb->s_bdev);
4369
4370         /*
4371          * Are we loading a blank journal or performing recovery after a
4372          * crash?  For recovery, we need to check in advance whether we
4373          * can get read-write access to the device.
4374          */
4375         if (ext4_has_feature_journal_needs_recovery(sb)) {
4376                 if (sb->s_flags & MS_RDONLY) {
4377                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4378                                         "required on readonly filesystem");
4379                         if (really_read_only) {
4380                                 ext4_msg(sb, KERN_ERR, "write access "
4381                                         "unavailable, cannot proceed");
4382                                 return -EROFS;
4383                         }
4384                         ext4_msg(sb, KERN_INFO, "write access will "
4385                                "be enabled during recovery");
4386                 }
4387         }
4388
4389         if (journal_inum && journal_dev) {
4390                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4391                        "and inode journals!");
4392                 return -EINVAL;
4393         }
4394
4395         if (journal_inum) {
4396                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4397                         return -EINVAL;
4398         } else {
4399                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4400                         return -EINVAL;
4401         }
4402
4403         if (!(journal->j_flags & JBD2_BARRIER))
4404                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4405
4406         if (!ext4_has_feature_journal_needs_recovery(sb))
4407                 err = jbd2_journal_wipe(journal, !really_read_only);
4408         if (!err) {
4409                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4410                 if (save)
4411                         memcpy(save, ((char *) es) +
4412                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4413                 err = jbd2_journal_load(journal);
4414                 if (save)
4415                         memcpy(((char *) es) + EXT4_S_ERR_START,
4416                                save, EXT4_S_ERR_LEN);
4417                 kfree(save);
4418         }
4419
4420         if (err) {
4421                 ext4_msg(sb, KERN_ERR, "error loading journal");
4422                 jbd2_journal_destroy(journal);
4423                 return err;
4424         }
4425
4426         EXT4_SB(sb)->s_journal = journal;
4427         ext4_clear_journal_err(sb, es);
4428
4429         if (!really_read_only && journal_devnum &&
4430             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4431                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4432
4433                 /* Make sure we flush the recovery flag to disk. */
4434                 ext4_commit_super(sb, 1);
4435         }
4436
4437         return 0;
4438 }
4439
4440 static int ext4_commit_super(struct super_block *sb, int sync)
4441 {
4442         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4443         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4444         int error = 0;
4445
4446         if (!sbh || block_device_ejected(sb))
4447                 return error;
4448
4449         /*
4450          * The superblock bh should be mapped, but it might not be if the
4451          * device was hot-removed. Not much we can do but fail the I/O.
4452          */
4453         if (!buffer_mapped(sbh))
4454                 return error;
4455
4456         if (buffer_write_io_error(sbh)) {
4457                 /*
4458                  * Oh, dear.  A previous attempt to write the
4459                  * superblock failed.  This could happen because the
4460                  * USB device was yanked out.  Or it could happen to
4461                  * be a transient write error and maybe the block will
4462                  * be remapped.  Nothing we can do but to retry the
4463                  * write and hope for the best.
4464                  */
4465                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4466                        "superblock detected");
4467                 clear_buffer_write_io_error(sbh);
4468                 set_buffer_uptodate(sbh);
4469         }
4470         /*
4471          * If the file system is mounted read-only, don't update the
4472          * superblock write time.  This avoids updating the superblock
4473          * write time when we are mounting the root file system
4474          * read/only but we need to replay the journal; at that point,
4475          * for people who are east of GMT and who make their clock
4476          * tick in localtime for Windows bug-for-bug compatibility,
4477          * the clock is set in the future, and this will cause e2fsck
4478          * to complain and force a full file system check.
4479          */
4480         if (!(sb->s_flags & MS_RDONLY))
4481                 es->s_wtime = cpu_to_le32(get_seconds());
4482         if (sb->s_bdev->bd_part)
4483                 es->s_kbytes_written =
4484                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4485                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4486                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4487         else
4488                 es->s_kbytes_written =
4489                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4490         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4491                 ext4_free_blocks_count_set(es,
4492                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4493                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4494         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4495                 es->s_free_inodes_count =
4496                         cpu_to_le32(percpu_counter_sum_positive(
4497                                 &EXT4_SB(sb)->s_freeinodes_counter));
4498         BUFFER_TRACE(sbh, "marking dirty");
4499         ext4_superblock_csum_set(sb);
4500         mark_buffer_dirty(sbh);
4501         if (sync) {
4502                 error = __sync_dirty_buffer(sbh,
4503                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4504                 if (error)
4505                         return error;
4506
4507                 error = buffer_write_io_error(sbh);
4508                 if (error) {
4509                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4510                                "superblock");
4511                         clear_buffer_write_io_error(sbh);
4512                         set_buffer_uptodate(sbh);
4513                 }
4514         }
4515         return error;
4516 }
4517
4518 /*
4519  * Have we just finished recovery?  If so, and if we are mounting (or
4520  * remounting) the filesystem readonly, then we will end up with a
4521  * consistent fs on disk.  Record that fact.
4522  */
4523 static void ext4_mark_recovery_complete(struct super_block *sb,
4524                                         struct ext4_super_block *es)
4525 {
4526         journal_t *journal = EXT4_SB(sb)->s_journal;
4527
4528         if (!ext4_has_feature_journal(sb)) {
4529                 BUG_ON(journal != NULL);
4530                 return;
4531         }
4532         jbd2_journal_lock_updates(journal);
4533         if (jbd2_journal_flush(journal) < 0)
4534                 goto out;
4535
4536         if (ext4_has_feature_journal_needs_recovery(sb) &&
4537             sb->s_flags & MS_RDONLY) {
4538                 ext4_clear_feature_journal_needs_recovery(sb);
4539                 ext4_commit_super(sb, 1);
4540         }
4541
4542 out:
4543         jbd2_journal_unlock_updates(journal);
4544 }
4545
4546 /*
4547  * If we are mounting (or read-write remounting) a filesystem whose journal
4548  * has recorded an error from a previous lifetime, move that error to the
4549  * main filesystem now.
4550  */
4551 static void ext4_clear_journal_err(struct super_block *sb,
4552                                    struct ext4_super_block *es)
4553 {
4554         journal_t *journal;
4555         int j_errno;
4556         const char *errstr;
4557
4558         BUG_ON(!ext4_has_feature_journal(sb));
4559
4560         journal = EXT4_SB(sb)->s_journal;
4561
4562         /*
4563          * Now check for any error status which may have been recorded in the
4564          * journal by a prior ext4_error() or ext4_abort()
4565          */
4566
4567         j_errno = jbd2_journal_errno(journal);
4568         if (j_errno) {
4569                 char nbuf[16];
4570
4571                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4572                 ext4_warning(sb, "Filesystem error recorded "
4573                              "from previous mount: %s", errstr);
4574                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4575
4576                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4577                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4578                 ext4_commit_super(sb, 1);
4579
4580                 jbd2_journal_clear_err(journal);
4581                 jbd2_journal_update_sb_errno(journal);
4582         }
4583 }
4584
4585 /*
4586  * Force the running and committing transactions to commit,
4587  * and wait on the commit.
4588  */
4589 int ext4_force_commit(struct super_block *sb)
4590 {
4591         journal_t *journal;
4592
4593         if (sb->s_flags & MS_RDONLY)
4594                 return 0;
4595
4596         journal = EXT4_SB(sb)->s_journal;
4597         return ext4_journal_force_commit(journal);
4598 }
4599
4600 static int ext4_sync_fs(struct super_block *sb, int wait)
4601 {
4602         int ret = 0;
4603         tid_t target;
4604         bool needs_barrier = false;
4605         struct ext4_sb_info *sbi = EXT4_SB(sb);
4606
4607         trace_ext4_sync_fs(sb, wait);
4608         flush_workqueue(sbi->rsv_conversion_wq);
4609         /*
4610          * Writeback quota in non-journalled quota case - journalled quota has
4611          * no dirty dquots
4612          */
4613         dquot_writeback_dquots(sb, -1);
4614         /*
4615          * Data writeback is possible w/o journal transaction, so barrier must
4616          * being sent at the end of the function. But we can skip it if
4617          * transaction_commit will do it for us.
4618          */
4619         if (sbi->s_journal) {
4620                 target = jbd2_get_latest_transaction(sbi->s_journal);
4621                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4622                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4623                         needs_barrier = true;
4624
4625                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4626                         if (wait)
4627                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4628                                                            target);
4629                 }
4630         } else if (wait && test_opt(sb, BARRIER))
4631                 needs_barrier = true;
4632         if (needs_barrier) {
4633                 int err;
4634                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4635                 if (!ret)
4636                         ret = err;
4637         }
4638
4639         return ret;
4640 }
4641
4642 /*
4643  * LVM calls this function before a (read-only) snapshot is created.  This
4644  * gives us a chance to flush the journal completely and mark the fs clean.
4645  *
4646  * Note that only this function cannot bring a filesystem to be in a clean
4647  * state independently. It relies on upper layer to stop all data & metadata
4648  * modifications.
4649  */
4650 static int ext4_freeze(struct super_block *sb)
4651 {
4652         int error = 0;
4653         journal_t *journal;
4654
4655         if (sb->s_flags & MS_RDONLY)
4656                 return 0;
4657
4658         journal = EXT4_SB(sb)->s_journal;
4659
4660         if (journal) {
4661                 /* Now we set up the journal barrier. */
4662                 jbd2_journal_lock_updates(journal);
4663
4664                 /*
4665                  * Don't clear the needs_recovery flag if we failed to
4666                  * flush the journal.
4667                  */
4668                 error = jbd2_journal_flush(journal);
4669                 if (error < 0)
4670                         goto out;
4671
4672                 /* Journal blocked and flushed, clear needs_recovery flag. */
4673                 ext4_clear_feature_journal_needs_recovery(sb);
4674         }
4675
4676         error = ext4_commit_super(sb, 1);
4677 out:
4678         if (journal)
4679                 /* we rely on upper layer to stop further updates */
4680                 jbd2_journal_unlock_updates(journal);
4681         return error;
4682 }
4683
4684 /*
4685  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4686  * flag here, even though the filesystem is not technically dirty yet.
4687  */
4688 static int ext4_unfreeze(struct super_block *sb)
4689 {
4690         if (sb->s_flags & MS_RDONLY)
4691                 return 0;
4692
4693         if (EXT4_SB(sb)->s_journal) {
4694                 /* Reset the needs_recovery flag before the fs is unlocked. */
4695                 ext4_set_feature_journal_needs_recovery(sb);
4696         }
4697
4698         ext4_commit_super(sb, 1);
4699         return 0;
4700 }
4701
4702 /*
4703  * Structure to save mount options for ext4_remount's benefit
4704  */
4705 struct ext4_mount_options {
4706         unsigned long s_mount_opt;
4707         unsigned long s_mount_opt2;
4708         kuid_t s_resuid;
4709         kgid_t s_resgid;
4710         unsigned long s_commit_interval;
4711         u32 s_min_batch_time, s_max_batch_time;
4712 #ifdef CONFIG_QUOTA
4713         int s_jquota_fmt;
4714         char *s_qf_names[EXT4_MAXQUOTAS];
4715 #endif
4716 };
4717
4718 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4719 {
4720         struct ext4_super_block *es;
4721         struct ext4_sb_info *sbi = EXT4_SB(sb);
4722         unsigned long old_sb_flags;
4723         struct ext4_mount_options old_opts;
4724         int enable_quota = 0;
4725         ext4_group_t g;
4726         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4727         int err = 0;
4728 #ifdef CONFIG_QUOTA
4729         int i, j;
4730 #endif
4731         char *orig_data = kstrdup(data, GFP_KERNEL);
4732
4733         /* Store the original options */
4734         old_sb_flags = sb->s_flags;
4735         old_opts.s_mount_opt = sbi->s_mount_opt;
4736         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4737         old_opts.s_resuid = sbi->s_resuid;
4738         old_opts.s_resgid = sbi->s_resgid;
4739         old_opts.s_commit_interval = sbi->s_commit_interval;
4740         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4741         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4742 #ifdef CONFIG_QUOTA
4743         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4744         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4745                 if (sbi->s_qf_names[i]) {
4746                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4747                                                          GFP_KERNEL);
4748                         if (!old_opts.s_qf_names[i]) {
4749                                 for (j = 0; j < i; j++)
4750                                         kfree(old_opts.s_qf_names[j]);
4751                                 kfree(orig_data);
4752                                 return -ENOMEM;
4753                         }
4754                 } else
4755                         old_opts.s_qf_names[i] = NULL;
4756 #endif
4757         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4758                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4759
4760         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4761                 err = -EINVAL;
4762                 goto restore_opts;
4763         }
4764
4765         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4766             test_opt(sb, JOURNAL_CHECKSUM)) {
4767                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4768                          "during remount not supported; ignoring");
4769                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4770         }
4771
4772         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4773                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4774                         ext4_msg(sb, KERN_ERR, "can't mount with "
4775                                  "both data=journal and delalloc");
4776                         err = -EINVAL;
4777                         goto restore_opts;
4778                 }
4779                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4780                         ext4_msg(sb, KERN_ERR, "can't mount with "
4781                                  "both data=journal and dioread_nolock");
4782                         err = -EINVAL;
4783                         goto restore_opts;
4784                 }
4785                 if (test_opt(sb, DAX)) {
4786                         ext4_msg(sb, KERN_ERR, "can't mount with "
4787                                  "both data=journal and dax");
4788                         err = -EINVAL;
4789                         goto restore_opts;
4790                 }
4791         }
4792
4793         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4794                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4795                         "dax flag with busy inodes while remounting");
4796                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4797         }
4798
4799         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4800                 ext4_abort(sb, "Abort forced by user");
4801
4802         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4803                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4804
4805         es = sbi->s_es;
4806
4807         if (sbi->s_journal) {
4808                 ext4_init_journal_params(sb, sbi->s_journal);
4809                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4810         }
4811
4812         if (*flags & MS_LAZYTIME)
4813                 sb->s_flags |= MS_LAZYTIME;
4814
4815         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4816                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4817                         err = -EROFS;
4818                         goto restore_opts;
4819                 }
4820
4821                 if (*flags & MS_RDONLY) {
4822                         err = sync_filesystem(sb);
4823                         if (err < 0)
4824                                 goto restore_opts;
4825                         err = dquot_suspend(sb, -1);
4826                         if (err < 0)
4827                                 goto restore_opts;
4828
4829                         /*
4830                          * First of all, the unconditional stuff we have to do
4831                          * to disable replay of the journal when we next remount
4832                          */
4833                         sb->s_flags |= MS_RDONLY;
4834
4835                         /*
4836                          * OK, test if we are remounting a valid rw partition
4837                          * readonly, and if so set the rdonly flag and then
4838                          * mark the partition as valid again.
4839                          */
4840                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4841                             (sbi->s_mount_state & EXT4_VALID_FS))
4842                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4843
4844                         if (sbi->s_journal)
4845                                 ext4_mark_recovery_complete(sb, es);
4846                 } else {
4847                         /* Make sure we can mount this feature set readwrite */
4848                         if (ext4_has_feature_readonly(sb) ||
4849                             !ext4_feature_set_ok(sb, 0)) {
4850                                 err = -EROFS;
4851                                 goto restore_opts;
4852                         }
4853                         /*
4854                          * Make sure the group descriptor checksums
4855                          * are sane.  If they aren't, refuse to remount r/w.
4856                          */
4857                         for (g = 0; g < sbi->s_groups_count; g++) {
4858                                 struct ext4_group_desc *gdp =
4859                                         ext4_get_group_desc(sb, g, NULL);
4860
4861                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4862                                         ext4_msg(sb, KERN_ERR,
4863                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4864                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4865                                                le16_to_cpu(gdp->bg_checksum));
4866                                         err = -EFSBADCRC;
4867                                         goto restore_opts;
4868                                 }
4869                         }
4870
4871                         /*
4872                          * If we have an unprocessed orphan list hanging
4873                          * around from a previously readonly bdev mount,
4874                          * require a full umount/remount for now.
4875                          */
4876                         if (es->s_last_orphan) {
4877                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4878                                        "remount RDWR because of unprocessed "
4879                                        "orphan inode list.  Please "
4880                                        "umount/remount instead");
4881                                 err = -EINVAL;
4882                                 goto restore_opts;
4883                         }
4884
4885                         /*
4886                          * Mounting a RDONLY partition read-write, so reread
4887                          * and store the current valid flag.  (It may have
4888                          * been changed by e2fsck since we originally mounted
4889                          * the partition.)
4890                          */
4891                         if (sbi->s_journal)
4892                                 ext4_clear_journal_err(sb, es);
4893                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4894                         if (!ext4_setup_super(sb, es, 0))
4895                                 sb->s_flags &= ~MS_RDONLY;
4896                         if (ext4_has_feature_mmp(sb))
4897                                 if (ext4_multi_mount_protect(sb,
4898                                                 le64_to_cpu(es->s_mmp_block))) {
4899                                         err = -EROFS;
4900                                         goto restore_opts;
4901                                 }
4902                         enable_quota = 1;
4903                 }
4904         }
4905
4906         /*
4907          * Reinitialize lazy itable initialization thread based on
4908          * current settings
4909          */
4910         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4911                 ext4_unregister_li_request(sb);
4912         else {
4913                 ext4_group_t first_not_zeroed;
4914                 first_not_zeroed = ext4_has_uninit_itable(sb);
4915                 ext4_register_li_request(sb, first_not_zeroed);
4916         }
4917
4918         ext4_setup_system_zone(sb);
4919         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4920                 ext4_commit_super(sb, 1);
4921
4922 #ifdef CONFIG_QUOTA
4923         /* Release old quota file names */
4924         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4925                 kfree(old_opts.s_qf_names[i]);
4926         if (enable_quota) {
4927                 if (sb_any_quota_suspended(sb))
4928                         dquot_resume(sb, -1);
4929                 else if (ext4_has_feature_quota(sb)) {
4930                         err = ext4_enable_quotas(sb);
4931                         if (err)
4932                                 goto restore_opts;
4933                 }
4934         }
4935 #endif
4936
4937         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4938         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4939         kfree(orig_data);
4940         return 0;
4941
4942 restore_opts:
4943         sb->s_flags = old_sb_flags;
4944         sbi->s_mount_opt = old_opts.s_mount_opt;
4945         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4946         sbi->s_resuid = old_opts.s_resuid;
4947         sbi->s_resgid = old_opts.s_resgid;
4948         sbi->s_commit_interval = old_opts.s_commit_interval;
4949         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4950         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4951 #ifdef CONFIG_QUOTA
4952         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4953         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4954                 kfree(sbi->s_qf_names[i]);
4955                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4956         }
4957 #endif
4958         kfree(orig_data);
4959         return err;
4960 }
4961
4962 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4963 {
4964         struct super_block *sb = dentry->d_sb;
4965         struct ext4_sb_info *sbi = EXT4_SB(sb);
4966         struct ext4_super_block *es = sbi->s_es;
4967         ext4_fsblk_t overhead = 0, resv_blocks;
4968         u64 fsid;
4969         s64 bfree;
4970         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4971
4972         if (!test_opt(sb, MINIX_DF))
4973                 overhead = sbi->s_overhead;
4974
4975         buf->f_type = EXT4_SUPER_MAGIC;
4976         buf->f_bsize = sb->s_blocksize;
4977         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4978         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4979                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4980         /* prevent underflow in case that few free space is available */
4981         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4982         buf->f_bavail = buf->f_bfree -
4983                         (ext4_r_blocks_count(es) + resv_blocks);
4984         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4985                 buf->f_bavail = 0;
4986         buf->f_files = le32_to_cpu(es->s_inodes_count);
4987         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4988         buf->f_namelen = EXT4_NAME_LEN;
4989         fsid = le64_to_cpup((void *)es->s_uuid) ^
4990                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4991         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4992         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4993
4994         return 0;
4995 }
4996
4997 /* Helper function for writing quotas on sync - we need to start transaction
4998  * before quota file is locked for write. Otherwise the are possible deadlocks:
4999  * Process 1                         Process 2
5000  * ext4_create()                     quota_sync()
5001  *   jbd2_journal_start()                  write_dquot()
5002  *   dquot_initialize()                         down(dqio_mutex)
5003  *     down(dqio_mutex)                    jbd2_journal_start()
5004  *
5005  */
5006
5007 #ifdef CONFIG_QUOTA
5008
5009 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5010 {
5011         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5012 }
5013
5014 static int ext4_write_dquot(struct dquot *dquot)
5015 {
5016         int ret, err;
5017         handle_t *handle;
5018         struct inode *inode;
5019
5020         inode = dquot_to_inode(dquot);
5021         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5022                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5023         if (IS_ERR(handle))
5024                 return PTR_ERR(handle);
5025         ret = dquot_commit(dquot);
5026         err = ext4_journal_stop(handle);
5027         if (!ret)
5028                 ret = err;
5029         return ret;
5030 }
5031
5032 static int ext4_acquire_dquot(struct dquot *dquot)
5033 {
5034         int ret, err;
5035         handle_t *handle;
5036
5037         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5038                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5039         if (IS_ERR(handle))
5040                 return PTR_ERR(handle);
5041         ret = dquot_acquire(dquot);
5042         err = ext4_journal_stop(handle);
5043         if (!ret)
5044                 ret = err;
5045         return ret;
5046 }
5047
5048 static int ext4_release_dquot(struct dquot *dquot)
5049 {
5050         int ret, err;
5051         handle_t *handle;
5052
5053         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5054                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5055         if (IS_ERR(handle)) {
5056                 /* Release dquot anyway to avoid endless cycle in dqput() */
5057                 dquot_release(dquot);
5058                 return PTR_ERR(handle);
5059         }
5060         ret = dquot_release(dquot);
5061         err = ext4_journal_stop(handle);
5062         if (!ret)
5063                 ret = err;
5064         return ret;
5065 }
5066
5067 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5068 {
5069         struct super_block *sb = dquot->dq_sb;
5070         struct ext4_sb_info *sbi = EXT4_SB(sb);
5071
5072         /* Are we journaling quotas? */
5073         if (ext4_has_feature_quota(sb) ||
5074             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5075                 dquot_mark_dquot_dirty(dquot);
5076                 return ext4_write_dquot(dquot);
5077         } else {
5078                 return dquot_mark_dquot_dirty(dquot);
5079         }
5080 }
5081
5082 static int ext4_write_info(struct super_block *sb, int type)
5083 {
5084         int ret, err;
5085         handle_t *handle;
5086
5087         /* Data block + inode block */
5088         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5089         if (IS_ERR(handle))
5090                 return PTR_ERR(handle);
5091         ret = dquot_commit_info(sb, type);
5092         err = ext4_journal_stop(handle);
5093         if (!ret)
5094                 ret = err;
5095         return ret;
5096 }
5097
5098 /*
5099  * Turn on quotas during mount time - we need to find
5100  * the quota file and such...
5101  */
5102 static int ext4_quota_on_mount(struct super_block *sb, int type)
5103 {
5104         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5105                                         EXT4_SB(sb)->s_jquota_fmt, type);
5106 }
5107
5108 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5109 {
5110         struct ext4_inode_info *ei = EXT4_I(inode);
5111
5112         /* The first argument of lockdep_set_subclass has to be
5113          * *exactly* the same as the argument to init_rwsem() --- in
5114          * this case, in init_once() --- or lockdep gets unhappy
5115          * because the name of the lock is set using the
5116          * stringification of the argument to init_rwsem().
5117          */
5118         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5119         lockdep_set_subclass(&ei->i_data_sem, subclass);
5120 }
5121
5122 /*
5123  * Standard function to be called on quota_on
5124  */
5125 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5126                          struct path *path)
5127 {
5128         int err;
5129
5130         if (!test_opt(sb, QUOTA))
5131                 return -EINVAL;
5132
5133         /* Quotafile not on the same filesystem? */
5134         if (path->dentry->d_sb != sb)
5135                 return -EXDEV;
5136         /* Journaling quota? */
5137         if (EXT4_SB(sb)->s_qf_names[type]) {
5138                 /* Quotafile not in fs root? */
5139                 if (path->dentry->d_parent != sb->s_root)
5140                         ext4_msg(sb, KERN_WARNING,
5141                                 "Quota file not on filesystem root. "
5142                                 "Journaled quota will not work");
5143         }
5144
5145         /*
5146          * When we journal data on quota file, we have to flush journal to see
5147          * all updates to the file when we bypass pagecache...
5148          */
5149         if (EXT4_SB(sb)->s_journal &&
5150             ext4_should_journal_data(d_inode(path->dentry))) {
5151                 /*
5152                  * We don't need to lock updates but journal_flush() could
5153                  * otherwise be livelocked...
5154                  */
5155                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5156                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5157                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5158                 if (err)
5159                         return err;
5160         }
5161         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5162         err = dquot_quota_on(sb, type, format_id, path);
5163         if (err)
5164                 lockdep_set_quota_inode(path->dentry->d_inode,
5165                                              I_DATA_SEM_NORMAL);
5166         return err;
5167 }
5168
5169 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5170                              unsigned int flags)
5171 {
5172         int err;
5173         struct inode *qf_inode;
5174         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5175                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5176                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5177         };
5178
5179         BUG_ON(!ext4_has_feature_quota(sb));
5180
5181         if (!qf_inums[type])
5182                 return -EPERM;
5183
5184         qf_inode = ext4_iget(sb, qf_inums[type]);
5185         if (IS_ERR(qf_inode)) {
5186                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5187                 return PTR_ERR(qf_inode);
5188         }
5189
5190         /* Don't account quota for quota files to avoid recursion */
5191         qf_inode->i_flags |= S_NOQUOTA;
5192         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5193         err = dquot_enable(qf_inode, type, format_id, flags);
5194         iput(qf_inode);
5195         if (err)
5196                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5197
5198         return err;
5199 }
5200
5201 /* Enable usage tracking for all quota types. */
5202 static int ext4_enable_quotas(struct super_block *sb)
5203 {
5204         int type, err = 0;
5205         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5206                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5207                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5208         };
5209
5210         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5211         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5212                 if (qf_inums[type]) {
5213                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5214                                                 DQUOT_USAGE_ENABLED);
5215                         if (err) {
5216                                 for (type--; type >= 0; type--)
5217                                         dquot_quota_off(sb, type);
5218
5219                                 ext4_warning(sb,
5220                                         "Failed to enable quota tracking "
5221                                         "(type=%d, err=%d). Please run "
5222                                         "e2fsck to fix.", type, err);
5223                                 return err;
5224                         }
5225                 }
5226         }
5227         return 0;
5228 }
5229
5230 static int ext4_quota_off(struct super_block *sb, int type)
5231 {
5232         struct inode *inode = sb_dqopt(sb)->files[type];
5233         handle_t *handle;
5234
5235         /* Force all delayed allocation blocks to be allocated.
5236          * Caller already holds s_umount sem */
5237         if (test_opt(sb, DELALLOC))
5238                 sync_filesystem(sb);
5239
5240         if (!inode)
5241                 goto out;
5242
5243         /* Update modification times of quota files when userspace can
5244          * start looking at them */
5245         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5246         if (IS_ERR(handle))
5247                 goto out;
5248         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5249         ext4_mark_inode_dirty(handle, inode);
5250         ext4_journal_stop(handle);
5251
5252 out:
5253         return dquot_quota_off(sb, type);
5254 }
5255
5256 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5257  * acquiring the locks... As quota files are never truncated and quota code
5258  * itself serializes the operations (and no one else should touch the files)
5259  * we don't have to be afraid of races */
5260 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5261                                size_t len, loff_t off)
5262 {
5263         struct inode *inode = sb_dqopt(sb)->files[type];
5264         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5265         int offset = off & (sb->s_blocksize - 1);
5266         int tocopy;
5267         size_t toread;
5268         struct buffer_head *bh;
5269         loff_t i_size = i_size_read(inode);
5270
5271         if (off > i_size)
5272                 return 0;
5273         if (off+len > i_size)
5274                 len = i_size-off;
5275         toread = len;
5276         while (toread > 0) {
5277                 tocopy = sb->s_blocksize - offset < toread ?
5278                                 sb->s_blocksize - offset : toread;
5279                 bh = ext4_bread(NULL, inode, blk, 0);
5280                 if (IS_ERR(bh))
5281                         return PTR_ERR(bh);
5282                 if (!bh)        /* A hole? */
5283                         memset(data, 0, tocopy);
5284                 else
5285                         memcpy(data, bh->b_data+offset, tocopy);
5286                 brelse(bh);
5287                 offset = 0;
5288                 toread -= tocopy;
5289                 data += tocopy;
5290                 blk++;
5291         }
5292         return len;
5293 }
5294
5295 /* Write to quotafile (we know the transaction is already started and has
5296  * enough credits) */
5297 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5298                                 const char *data, size_t len, loff_t off)
5299 {
5300         struct inode *inode = sb_dqopt(sb)->files[type];
5301         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5302         int err, offset = off & (sb->s_blocksize - 1);
5303         int retries = 0;
5304         struct buffer_head *bh;
5305         handle_t *handle = journal_current_handle();
5306
5307         if (EXT4_SB(sb)->s_journal && !handle) {
5308                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5309                         " cancelled because transaction is not started",
5310                         (unsigned long long)off, (unsigned long long)len);
5311                 return -EIO;
5312         }
5313         /*
5314          * Since we account only one data block in transaction credits,
5315          * then it is impossible to cross a block boundary.
5316          */
5317         if (sb->s_blocksize - offset < len) {
5318                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5319                         " cancelled because not block aligned",
5320                         (unsigned long long)off, (unsigned long long)len);
5321                 return -EIO;
5322         }
5323
5324         do {
5325                 bh = ext4_bread(handle, inode, blk,
5326                                 EXT4_GET_BLOCKS_CREATE |
5327                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5328         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5329                  ext4_should_retry_alloc(inode->i_sb, &retries));
5330         if (IS_ERR(bh))
5331                 return PTR_ERR(bh);
5332         if (!bh)
5333                 goto out;
5334         BUFFER_TRACE(bh, "get write access");
5335         err = ext4_journal_get_write_access(handle, bh);
5336         if (err) {
5337                 brelse(bh);
5338                 return err;
5339         }
5340         lock_buffer(bh);
5341         memcpy(bh->b_data+offset, data, len);
5342         flush_dcache_page(bh->b_page);
5343         unlock_buffer(bh);
5344         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5345         brelse(bh);
5346 out:
5347         if (inode->i_size < off + len) {
5348                 i_size_write(inode, off + len);
5349                 EXT4_I(inode)->i_disksize = inode->i_size;
5350                 ext4_mark_inode_dirty(handle, inode);
5351         }
5352         return len;
5353 }
5354
5355 #endif
5356
5357 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5358                        const char *dev_name, void *data)
5359 {
5360         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5361 }
5362
5363 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5364 static inline void register_as_ext2(void)
5365 {
5366         int err = register_filesystem(&ext2_fs_type);
5367         if (err)
5368                 printk(KERN_WARNING
5369                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5370 }
5371
5372 static inline void unregister_as_ext2(void)
5373 {
5374         unregister_filesystem(&ext2_fs_type);
5375 }
5376
5377 static inline int ext2_feature_set_ok(struct super_block *sb)
5378 {
5379         if (ext4_has_unknown_ext2_incompat_features(sb))
5380                 return 0;
5381         if (sb->s_flags & MS_RDONLY)
5382                 return 1;
5383         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5384                 return 0;
5385         return 1;
5386 }
5387 #else
5388 static inline void register_as_ext2(void) { }
5389 static inline void unregister_as_ext2(void) { }
5390 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5391 #endif
5392
5393 static inline void register_as_ext3(void)
5394 {
5395         int err = register_filesystem(&ext3_fs_type);
5396         if (err)
5397                 printk(KERN_WARNING
5398                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5399 }
5400
5401 static inline void unregister_as_ext3(void)
5402 {
5403         unregister_filesystem(&ext3_fs_type);
5404 }
5405
5406 static inline int ext3_feature_set_ok(struct super_block *sb)
5407 {
5408         if (ext4_has_unknown_ext3_incompat_features(sb))
5409                 return 0;
5410         if (!ext4_has_feature_journal(sb))
5411                 return 0;
5412         if (sb->s_flags & MS_RDONLY)
5413                 return 1;
5414         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5415                 return 0;
5416         return 1;
5417 }
5418
5419 static struct file_system_type ext4_fs_type = {
5420         .owner          = THIS_MODULE,
5421         .name           = "ext4",
5422         .mount          = ext4_mount,
5423         .kill_sb        = kill_block_super,
5424         .fs_flags       = FS_REQUIRES_DEV,
5425 };
5426 MODULE_ALIAS_FS("ext4");
5427
5428 /* Shared across all ext4 file systems */
5429 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5430 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5431
5432 static int __init ext4_init_fs(void)
5433 {
5434         int i, err;
5435
5436         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5437         ext4_li_info = NULL;
5438         mutex_init(&ext4_li_mtx);
5439
5440         /* Build-time check for flags consistency */
5441         ext4_check_flag_values();
5442
5443         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5444                 mutex_init(&ext4__aio_mutex[i]);
5445                 init_waitqueue_head(&ext4__ioend_wq[i]);
5446         }
5447
5448         err = ext4_init_es();
5449         if (err)
5450                 return err;
5451
5452         err = ext4_init_pageio();
5453         if (err)
5454                 goto out5;
5455
5456         err = ext4_init_system_zone();
5457         if (err)
5458                 goto out4;
5459
5460         err = ext4_init_sysfs();
5461         if (err)
5462                 goto out3;
5463
5464         err = ext4_init_mballoc();
5465         if (err)
5466                 goto out2;
5467         else
5468                 ext4_mballoc_ready = 1;
5469         err = init_inodecache();
5470         if (err)
5471                 goto out1;
5472         register_as_ext3();
5473         register_as_ext2();
5474         err = register_filesystem(&ext4_fs_type);
5475         if (err)
5476                 goto out;
5477
5478         return 0;
5479 out:
5480         unregister_as_ext2();
5481         unregister_as_ext3();
5482         destroy_inodecache();
5483 out1:
5484         ext4_mballoc_ready = 0;
5485         ext4_exit_mballoc();
5486 out2:
5487         ext4_exit_sysfs();
5488 out3:
5489         ext4_exit_system_zone();
5490 out4:
5491         ext4_exit_pageio();
5492 out5:
5493         ext4_exit_es();
5494
5495         return err;
5496 }
5497
5498 static void __exit ext4_exit_fs(void)
5499 {
5500         ext4_exit_crypto();
5501         ext4_destroy_lazyinit_thread();
5502         unregister_as_ext2();
5503         unregister_as_ext3();
5504         unregister_filesystem(&ext4_fs_type);
5505         destroy_inodecache();
5506         ext4_exit_mballoc();
5507         ext4_exit_sysfs();
5508         ext4_exit_system_zone();
5509         ext4_exit_pageio();
5510         ext4_exit_es();
5511 }
5512
5513 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5514 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5515 MODULE_LICENSE("GPL");
5516 module_init(ext4_init_fs)
5517 module_exit(ext4_exit_fs)